xref: /openbmc/qemu/block/qcow2.c (revision e3a99063)
1 /*
2  * Block driver for the QCOW version 2 format
3  *
4  * Copyright (c) 2004-2006 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 
25 #include "qemu/osdep.h"
26 
27 #include "block/qdict.h"
28 #include "sysemu/block-backend.h"
29 #include "qemu/main-loop.h"
30 #include "qemu/module.h"
31 #include "qcow2.h"
32 #include "qemu/error-report.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-events-block-core.h"
35 #include "qapi/qmp/qdict.h"
36 #include "qapi/qmp/qstring.h"
37 #include "trace.h"
38 #include "qemu/option_int.h"
39 #include "qemu/cutils.h"
40 #include "qemu/bswap.h"
41 #include "qapi/qobject-input-visitor.h"
42 #include "qapi/qapi-visit-block-core.h"
43 #include "crypto.h"
44 #include "block/aio_task.h"
45 
46 /*
47   Differences with QCOW:
48 
49   - Support for multiple incremental snapshots.
50   - Memory management by reference counts.
51   - Clusters which have a reference count of one have the bit
52     QCOW_OFLAG_COPIED to optimize write performance.
53   - Size of compressed clusters is stored in sectors to reduce bit usage
54     in the cluster offsets.
55   - Support for storing additional data (such as the VM state) in the
56     snapshots.
57   - If a backing store is used, the cluster size is not constrained
58     (could be backported to QCOW).
59   - L2 tables have always a size of one cluster.
60 */
61 
62 
63 typedef struct {
64     uint32_t magic;
65     uint32_t len;
66 } QEMU_PACKED QCowExtension;
67 
68 #define  QCOW2_EXT_MAGIC_END 0
69 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
70 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
71 #define  QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
72 #define  QCOW2_EXT_MAGIC_BITMAPS 0x23852875
73 #define  QCOW2_EXT_MAGIC_DATA_FILE 0x44415441
74 
75 static int coroutine_fn
76 qcow2_co_preadv_compressed(BlockDriverState *bs,
77                            uint64_t file_cluster_offset,
78                            uint64_t offset,
79                            uint64_t bytes,
80                            QEMUIOVector *qiov,
81                            size_t qiov_offset);
82 
83 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
84 {
85     const QCowHeader *cow_header = (const void *)buf;
86 
87     if (buf_size >= sizeof(QCowHeader) &&
88         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
89         be32_to_cpu(cow_header->version) >= 2)
90         return 100;
91     else
92         return 0;
93 }
94 
95 
96 static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset,
97                                           uint8_t *buf, size_t buflen,
98                                           void *opaque, Error **errp)
99 {
100     BlockDriverState *bs = opaque;
101     BDRVQcow2State *s = bs->opaque;
102     ssize_t ret;
103 
104     if ((offset + buflen) > s->crypto_header.length) {
105         error_setg(errp, "Request for data outside of extension header");
106         return -1;
107     }
108 
109     ret = bdrv_pread(bs->file,
110                      s->crypto_header.offset + offset, buf, buflen);
111     if (ret < 0) {
112         error_setg_errno(errp, -ret, "Could not read encryption header");
113         return -1;
114     }
115     return ret;
116 }
117 
118 
119 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen,
120                                           void *opaque, Error **errp)
121 {
122     BlockDriverState *bs = opaque;
123     BDRVQcow2State *s = bs->opaque;
124     int64_t ret;
125     int64_t clusterlen;
126 
127     ret = qcow2_alloc_clusters(bs, headerlen);
128     if (ret < 0) {
129         error_setg_errno(errp, -ret,
130                          "Cannot allocate cluster for LUKS header size %zu",
131                          headerlen);
132         return -1;
133     }
134 
135     s->crypto_header.length = headerlen;
136     s->crypto_header.offset = ret;
137 
138     /*
139      * Zero fill all space in cluster so it has predictable
140      * content, as we may not initialize some regions of the
141      * header (eg only 1 out of 8 key slots will be initialized)
142      */
143     clusterlen = size_to_clusters(s, headerlen) * s->cluster_size;
144     assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen, false) == 0);
145     ret = bdrv_pwrite_zeroes(bs->file,
146                              ret,
147                              clusterlen, 0);
148     if (ret < 0) {
149         error_setg_errno(errp, -ret, "Could not zero fill encryption header");
150         return -1;
151     }
152 
153     return ret;
154 }
155 
156 
157 static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset,
158                                            const uint8_t *buf, size_t buflen,
159                                            void *opaque, Error **errp)
160 {
161     BlockDriverState *bs = opaque;
162     BDRVQcow2State *s = bs->opaque;
163     ssize_t ret;
164 
165     if ((offset + buflen) > s->crypto_header.length) {
166         error_setg(errp, "Request for data outside of extension header");
167         return -1;
168     }
169 
170     ret = bdrv_pwrite(bs->file,
171                       s->crypto_header.offset + offset, buf, buflen);
172     if (ret < 0) {
173         error_setg_errno(errp, -ret, "Could not read encryption header");
174         return -1;
175     }
176     return ret;
177 }
178 
179 
180 /*
181  * read qcow2 extension and fill bs
182  * start reading from start_offset
183  * finish reading upon magic of value 0 or when end_offset reached
184  * unknown magic is skipped (future extension this version knows nothing about)
185  * return 0 upon success, non-0 otherwise
186  */
187 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
188                                  uint64_t end_offset, void **p_feature_table,
189                                  int flags, bool *need_update_header,
190                                  Error **errp)
191 {
192     BDRVQcow2State *s = bs->opaque;
193     QCowExtension ext;
194     uint64_t offset;
195     int ret;
196     Qcow2BitmapHeaderExt bitmaps_ext;
197 
198     if (need_update_header != NULL) {
199         *need_update_header = false;
200     }
201 
202 #ifdef DEBUG_EXT
203     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
204 #endif
205     offset = start_offset;
206     while (offset < end_offset) {
207 
208 #ifdef DEBUG_EXT
209         /* Sanity check */
210         if (offset > s->cluster_size)
211             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
212 
213         printf("attempting to read extended header in offset %lu\n", offset);
214 #endif
215 
216         ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
217         if (ret < 0) {
218             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
219                              "pread fail from offset %" PRIu64, offset);
220             return 1;
221         }
222         ext.magic = be32_to_cpu(ext.magic);
223         ext.len = be32_to_cpu(ext.len);
224         offset += sizeof(ext);
225 #ifdef DEBUG_EXT
226         printf("ext.magic = 0x%x\n", ext.magic);
227 #endif
228         if (offset > end_offset || ext.len > end_offset - offset) {
229             error_setg(errp, "Header extension too large");
230             return -EINVAL;
231         }
232 
233         switch (ext.magic) {
234         case QCOW2_EXT_MAGIC_END:
235             return 0;
236 
237         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
238             if (ext.len >= sizeof(bs->backing_format)) {
239                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
240                            " too large (>=%zu)", ext.len,
241                            sizeof(bs->backing_format));
242                 return 2;
243             }
244             ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len);
245             if (ret < 0) {
246                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
247                                  "Could not read format name");
248                 return 3;
249             }
250             bs->backing_format[ext.len] = '\0';
251             s->image_backing_format = g_strdup(bs->backing_format);
252 #ifdef DEBUG_EXT
253             printf("Qcow2: Got format extension %s\n", bs->backing_format);
254 #endif
255             break;
256 
257         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
258             if (p_feature_table != NULL) {
259                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
260                 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
261                 if (ret < 0) {
262                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
263                                      "Could not read table");
264                     return ret;
265                 }
266 
267                 *p_feature_table = feature_table;
268             }
269             break;
270 
271         case QCOW2_EXT_MAGIC_CRYPTO_HEADER: {
272             unsigned int cflags = 0;
273             if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
274                 error_setg(errp, "CRYPTO header extension only "
275                            "expected with LUKS encryption method");
276                 return -EINVAL;
277             }
278             if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) {
279                 error_setg(errp, "CRYPTO header extension size %u, "
280                            "but expected size %zu", ext.len,
281                            sizeof(Qcow2CryptoHeaderExtension));
282                 return -EINVAL;
283             }
284 
285             ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len);
286             if (ret < 0) {
287                 error_setg_errno(errp, -ret,
288                                  "Unable to read CRYPTO header extension");
289                 return ret;
290             }
291             s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
292             s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
293 
294             if ((s->crypto_header.offset % s->cluster_size) != 0) {
295                 error_setg(errp, "Encryption header offset '%" PRIu64 "' is "
296                            "not a multiple of cluster size '%u'",
297                            s->crypto_header.offset, s->cluster_size);
298                 return -EINVAL;
299             }
300 
301             if (flags & BDRV_O_NO_IO) {
302                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
303             }
304             s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
305                                            qcow2_crypto_hdr_read_func,
306                                            bs, cflags, QCOW2_MAX_THREADS, errp);
307             if (!s->crypto) {
308                 return -EINVAL;
309             }
310         }   break;
311 
312         case QCOW2_EXT_MAGIC_BITMAPS:
313             if (ext.len != sizeof(bitmaps_ext)) {
314                 error_setg_errno(errp, -ret, "bitmaps_ext: "
315                                  "Invalid extension length");
316                 return -EINVAL;
317             }
318 
319             if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) {
320                 if (s->qcow_version < 3) {
321                     /* Let's be a bit more specific */
322                     warn_report("This qcow2 v2 image contains bitmaps, but "
323                                 "they may have been modified by a program "
324                                 "without persistent bitmap support; so now "
325                                 "they must all be considered inconsistent");
326                 } else {
327                     warn_report("a program lacking bitmap support "
328                                 "modified this file, so all bitmaps are now "
329                                 "considered inconsistent");
330                 }
331                 error_printf("Some clusters may be leaked, "
332                              "run 'qemu-img check -r' on the image "
333                              "file to fix.");
334                 if (need_update_header != NULL) {
335                     /* Updating is needed to drop invalid bitmap extension. */
336                     *need_update_header = true;
337                 }
338                 break;
339             }
340 
341             ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len);
342             if (ret < 0) {
343                 error_setg_errno(errp, -ret, "bitmaps_ext: "
344                                  "Could not read ext header");
345                 return ret;
346             }
347 
348             if (bitmaps_ext.reserved32 != 0) {
349                 error_setg_errno(errp, -ret, "bitmaps_ext: "
350                                  "Reserved field is not zero");
351                 return -EINVAL;
352             }
353 
354             bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps);
355             bitmaps_ext.bitmap_directory_size =
356                 be64_to_cpu(bitmaps_ext.bitmap_directory_size);
357             bitmaps_ext.bitmap_directory_offset =
358                 be64_to_cpu(bitmaps_ext.bitmap_directory_offset);
359 
360             if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) {
361                 error_setg(errp,
362                            "bitmaps_ext: Image has %" PRIu32 " bitmaps, "
363                            "exceeding the QEMU supported maximum of %d",
364                            bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS);
365                 return -EINVAL;
366             }
367 
368             if (bitmaps_ext.nb_bitmaps == 0) {
369                 error_setg(errp, "found bitmaps extension with zero bitmaps");
370                 return -EINVAL;
371             }
372 
373             if (offset_into_cluster(s, bitmaps_ext.bitmap_directory_offset)) {
374                 error_setg(errp, "bitmaps_ext: "
375                                  "invalid bitmap directory offset");
376                 return -EINVAL;
377             }
378 
379             if (bitmaps_ext.bitmap_directory_size >
380                 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) {
381                 error_setg(errp, "bitmaps_ext: "
382                                  "bitmap directory size (%" PRIu64 ") exceeds "
383                                  "the maximum supported size (%d)",
384                                  bitmaps_ext.bitmap_directory_size,
385                                  QCOW2_MAX_BITMAP_DIRECTORY_SIZE);
386                 return -EINVAL;
387             }
388 
389             s->nb_bitmaps = bitmaps_ext.nb_bitmaps;
390             s->bitmap_directory_offset =
391                     bitmaps_ext.bitmap_directory_offset;
392             s->bitmap_directory_size =
393                     bitmaps_ext.bitmap_directory_size;
394 
395 #ifdef DEBUG_EXT
396             printf("Qcow2: Got bitmaps extension: "
397                    "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n",
398                    s->bitmap_directory_offset, s->nb_bitmaps);
399 #endif
400             break;
401 
402         case QCOW2_EXT_MAGIC_DATA_FILE:
403         {
404             s->image_data_file = g_malloc0(ext.len + 1);
405             ret = bdrv_pread(bs->file, offset, s->image_data_file, ext.len);
406             if (ret < 0) {
407                 error_setg_errno(errp, -ret,
408                                  "ERROR: Could not read data file name");
409                 return ret;
410             }
411 #ifdef DEBUG_EXT
412             printf("Qcow2: Got external data file %s\n", s->image_data_file);
413 #endif
414             break;
415         }
416 
417         default:
418             /* unknown magic - save it in case we need to rewrite the header */
419             /* If you add a new feature, make sure to also update the fast
420              * path of qcow2_make_empty() to deal with it. */
421             {
422                 Qcow2UnknownHeaderExtension *uext;
423 
424                 uext = g_malloc0(sizeof(*uext)  + ext.len);
425                 uext->magic = ext.magic;
426                 uext->len = ext.len;
427                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
428 
429                 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
430                 if (ret < 0) {
431                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
432                                      "Could not read data");
433                     return ret;
434                 }
435             }
436             break;
437         }
438 
439         offset += ((ext.len + 7) & ~7);
440     }
441 
442     return 0;
443 }
444 
445 static void cleanup_unknown_header_ext(BlockDriverState *bs)
446 {
447     BDRVQcow2State *s = bs->opaque;
448     Qcow2UnknownHeaderExtension *uext, *next;
449 
450     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
451         QLIST_REMOVE(uext, next);
452         g_free(uext);
453     }
454 }
455 
456 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
457                                        uint64_t mask)
458 {
459     g_autoptr(GString) features = g_string_sized_new(60);
460 
461     while (table && table->name[0] != '\0') {
462         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
463             if (mask & (1ULL << table->bit)) {
464                 if (features->len > 0) {
465                     g_string_append(features, ", ");
466                 }
467                 g_string_append_printf(features, "%.46s", table->name);
468                 mask &= ~(1ULL << table->bit);
469             }
470         }
471         table++;
472     }
473 
474     if (mask) {
475         if (features->len > 0) {
476             g_string_append(features, ", ");
477         }
478         g_string_append_printf(features,
479                                "Unknown incompatible feature: %" PRIx64, mask);
480     }
481 
482     error_setg(errp, "Unsupported qcow2 feature(s): %s", features->str);
483 }
484 
485 /*
486  * Sets the dirty bit and flushes afterwards if necessary.
487  *
488  * The incompatible_features bit is only set if the image file header was
489  * updated successfully.  Therefore it is not required to check the return
490  * value of this function.
491  */
492 int qcow2_mark_dirty(BlockDriverState *bs)
493 {
494     BDRVQcow2State *s = bs->opaque;
495     uint64_t val;
496     int ret;
497 
498     assert(s->qcow_version >= 3);
499 
500     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
501         return 0; /* already dirty */
502     }
503 
504     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
505     ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
506                       &val, sizeof(val));
507     if (ret < 0) {
508         return ret;
509     }
510     ret = bdrv_flush(bs->file->bs);
511     if (ret < 0) {
512         return ret;
513     }
514 
515     /* Only treat image as dirty if the header was updated successfully */
516     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
517     return 0;
518 }
519 
520 /*
521  * Clears the dirty bit and flushes before if necessary.  Only call this
522  * function when there are no pending requests, it does not guard against
523  * concurrent requests dirtying the image.
524  */
525 static int qcow2_mark_clean(BlockDriverState *bs)
526 {
527     BDRVQcow2State *s = bs->opaque;
528 
529     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
530         int ret;
531 
532         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
533 
534         ret = qcow2_flush_caches(bs);
535         if (ret < 0) {
536             return ret;
537         }
538 
539         return qcow2_update_header(bs);
540     }
541     return 0;
542 }
543 
544 /*
545  * Marks the image as corrupt.
546  */
547 int qcow2_mark_corrupt(BlockDriverState *bs)
548 {
549     BDRVQcow2State *s = bs->opaque;
550 
551     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
552     return qcow2_update_header(bs);
553 }
554 
555 /*
556  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
557  * before if necessary.
558  */
559 int qcow2_mark_consistent(BlockDriverState *bs)
560 {
561     BDRVQcow2State *s = bs->opaque;
562 
563     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
564         int ret = qcow2_flush_caches(bs);
565         if (ret < 0) {
566             return ret;
567         }
568 
569         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
570         return qcow2_update_header(bs);
571     }
572     return 0;
573 }
574 
575 static void qcow2_add_check_result(BdrvCheckResult *out,
576                                    const BdrvCheckResult *src,
577                                    bool set_allocation_info)
578 {
579     out->corruptions += src->corruptions;
580     out->leaks += src->leaks;
581     out->check_errors += src->check_errors;
582     out->corruptions_fixed += src->corruptions_fixed;
583     out->leaks_fixed += src->leaks_fixed;
584 
585     if (set_allocation_info) {
586         out->image_end_offset = src->image_end_offset;
587         out->bfi = src->bfi;
588     }
589 }
590 
591 static int coroutine_fn qcow2_co_check_locked(BlockDriverState *bs,
592                                               BdrvCheckResult *result,
593                                               BdrvCheckMode fix)
594 {
595     BdrvCheckResult snapshot_res = {};
596     BdrvCheckResult refcount_res = {};
597     int ret;
598 
599     memset(result, 0, sizeof(*result));
600 
601     ret = qcow2_check_read_snapshot_table(bs, &snapshot_res, fix);
602     if (ret < 0) {
603         qcow2_add_check_result(result, &snapshot_res, false);
604         return ret;
605     }
606 
607     ret = qcow2_check_refcounts(bs, &refcount_res, fix);
608     qcow2_add_check_result(result, &refcount_res, true);
609     if (ret < 0) {
610         qcow2_add_check_result(result, &snapshot_res, false);
611         return ret;
612     }
613 
614     ret = qcow2_check_fix_snapshot_table(bs, &snapshot_res, fix);
615     qcow2_add_check_result(result, &snapshot_res, false);
616     if (ret < 0) {
617         return ret;
618     }
619 
620     if (fix && result->check_errors == 0 && result->corruptions == 0) {
621         ret = qcow2_mark_clean(bs);
622         if (ret < 0) {
623             return ret;
624         }
625         return qcow2_mark_consistent(bs);
626     }
627     return ret;
628 }
629 
630 static int coroutine_fn qcow2_co_check(BlockDriverState *bs,
631                                        BdrvCheckResult *result,
632                                        BdrvCheckMode fix)
633 {
634     BDRVQcow2State *s = bs->opaque;
635     int ret;
636 
637     qemu_co_mutex_lock(&s->lock);
638     ret = qcow2_co_check_locked(bs, result, fix);
639     qemu_co_mutex_unlock(&s->lock);
640     return ret;
641 }
642 
643 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset,
644                          uint64_t entries, size_t entry_len,
645                          int64_t max_size_bytes, const char *table_name,
646                          Error **errp)
647 {
648     BDRVQcow2State *s = bs->opaque;
649 
650     if (entries > max_size_bytes / entry_len) {
651         error_setg(errp, "%s too large", table_name);
652         return -EFBIG;
653     }
654 
655     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
656      * because values will be passed to qemu functions taking int64_t. */
657     if ((INT64_MAX - entries * entry_len < offset) ||
658         (offset_into_cluster(s, offset) != 0)) {
659         error_setg(errp, "%s offset invalid", table_name);
660         return -EINVAL;
661     }
662 
663     return 0;
664 }
665 
666 static const char *const mutable_opts[] = {
667     QCOW2_OPT_LAZY_REFCOUNTS,
668     QCOW2_OPT_DISCARD_REQUEST,
669     QCOW2_OPT_DISCARD_SNAPSHOT,
670     QCOW2_OPT_DISCARD_OTHER,
671     QCOW2_OPT_OVERLAP,
672     QCOW2_OPT_OVERLAP_TEMPLATE,
673     QCOW2_OPT_OVERLAP_MAIN_HEADER,
674     QCOW2_OPT_OVERLAP_ACTIVE_L1,
675     QCOW2_OPT_OVERLAP_ACTIVE_L2,
676     QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
677     QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
678     QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
679     QCOW2_OPT_OVERLAP_INACTIVE_L1,
680     QCOW2_OPT_OVERLAP_INACTIVE_L2,
681     QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
682     QCOW2_OPT_CACHE_SIZE,
683     QCOW2_OPT_L2_CACHE_SIZE,
684     QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
685     QCOW2_OPT_REFCOUNT_CACHE_SIZE,
686     QCOW2_OPT_CACHE_CLEAN_INTERVAL,
687     NULL
688 };
689 
690 static QemuOptsList qcow2_runtime_opts = {
691     .name = "qcow2",
692     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
693     .desc = {
694         {
695             .name = QCOW2_OPT_LAZY_REFCOUNTS,
696             .type = QEMU_OPT_BOOL,
697             .help = "Postpone refcount updates",
698         },
699         {
700             .name = QCOW2_OPT_DISCARD_REQUEST,
701             .type = QEMU_OPT_BOOL,
702             .help = "Pass guest discard requests to the layer below",
703         },
704         {
705             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
706             .type = QEMU_OPT_BOOL,
707             .help = "Generate discard requests when snapshot related space "
708                     "is freed",
709         },
710         {
711             .name = QCOW2_OPT_DISCARD_OTHER,
712             .type = QEMU_OPT_BOOL,
713             .help = "Generate discard requests when other clusters are freed",
714         },
715         {
716             .name = QCOW2_OPT_OVERLAP,
717             .type = QEMU_OPT_STRING,
718             .help = "Selects which overlap checks to perform from a range of "
719                     "templates (none, constant, cached, all)",
720         },
721         {
722             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
723             .type = QEMU_OPT_STRING,
724             .help = "Selects which overlap checks to perform from a range of "
725                     "templates (none, constant, cached, all)",
726         },
727         {
728             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
729             .type = QEMU_OPT_BOOL,
730             .help = "Check for unintended writes into the main qcow2 header",
731         },
732         {
733             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
734             .type = QEMU_OPT_BOOL,
735             .help = "Check for unintended writes into the active L1 table",
736         },
737         {
738             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
739             .type = QEMU_OPT_BOOL,
740             .help = "Check for unintended writes into an active L2 table",
741         },
742         {
743             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
744             .type = QEMU_OPT_BOOL,
745             .help = "Check for unintended writes into the refcount table",
746         },
747         {
748             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
749             .type = QEMU_OPT_BOOL,
750             .help = "Check for unintended writes into a refcount block",
751         },
752         {
753             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
754             .type = QEMU_OPT_BOOL,
755             .help = "Check for unintended writes into the snapshot table",
756         },
757         {
758             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
759             .type = QEMU_OPT_BOOL,
760             .help = "Check for unintended writes into an inactive L1 table",
761         },
762         {
763             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
764             .type = QEMU_OPT_BOOL,
765             .help = "Check for unintended writes into an inactive L2 table",
766         },
767         {
768             .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
769             .type = QEMU_OPT_BOOL,
770             .help = "Check for unintended writes into the bitmap directory",
771         },
772         {
773             .name = QCOW2_OPT_CACHE_SIZE,
774             .type = QEMU_OPT_SIZE,
775             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
776                     "cache size",
777         },
778         {
779             .name = QCOW2_OPT_L2_CACHE_SIZE,
780             .type = QEMU_OPT_SIZE,
781             .help = "Maximum L2 table cache size",
782         },
783         {
784             .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
785             .type = QEMU_OPT_SIZE,
786             .help = "Size of each entry in the L2 cache",
787         },
788         {
789             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
790             .type = QEMU_OPT_SIZE,
791             .help = "Maximum refcount block cache size",
792         },
793         {
794             .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
795             .type = QEMU_OPT_NUMBER,
796             .help = "Clean unused cache entries after this time (in seconds)",
797         },
798         BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
799             "ID of secret providing qcow2 AES key or LUKS passphrase"),
800         { /* end of list */ }
801     },
802 };
803 
804 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
805     [QCOW2_OL_MAIN_HEADER_BITNR]      = QCOW2_OPT_OVERLAP_MAIN_HEADER,
806     [QCOW2_OL_ACTIVE_L1_BITNR]        = QCOW2_OPT_OVERLAP_ACTIVE_L1,
807     [QCOW2_OL_ACTIVE_L2_BITNR]        = QCOW2_OPT_OVERLAP_ACTIVE_L2,
808     [QCOW2_OL_REFCOUNT_TABLE_BITNR]   = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
809     [QCOW2_OL_REFCOUNT_BLOCK_BITNR]   = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
810     [QCOW2_OL_SNAPSHOT_TABLE_BITNR]   = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
811     [QCOW2_OL_INACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_INACTIVE_L1,
812     [QCOW2_OL_INACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_INACTIVE_L2,
813     [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
814 };
815 
816 static void cache_clean_timer_cb(void *opaque)
817 {
818     BlockDriverState *bs = opaque;
819     BDRVQcow2State *s = bs->opaque;
820     qcow2_cache_clean_unused(s->l2_table_cache);
821     qcow2_cache_clean_unused(s->refcount_block_cache);
822     timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
823               (int64_t) s->cache_clean_interval * 1000);
824 }
825 
826 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
827 {
828     BDRVQcow2State *s = bs->opaque;
829     if (s->cache_clean_interval > 0) {
830         s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
831                                              SCALE_MS, cache_clean_timer_cb,
832                                              bs);
833         timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
834                   (int64_t) s->cache_clean_interval * 1000);
835     }
836 }
837 
838 static void cache_clean_timer_del(BlockDriverState *bs)
839 {
840     BDRVQcow2State *s = bs->opaque;
841     if (s->cache_clean_timer) {
842         timer_del(s->cache_clean_timer);
843         timer_free(s->cache_clean_timer);
844         s->cache_clean_timer = NULL;
845     }
846 }
847 
848 static void qcow2_detach_aio_context(BlockDriverState *bs)
849 {
850     cache_clean_timer_del(bs);
851 }
852 
853 static void qcow2_attach_aio_context(BlockDriverState *bs,
854                                      AioContext *new_context)
855 {
856     cache_clean_timer_init(bs, new_context);
857 }
858 
859 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
860                              uint64_t *l2_cache_size,
861                              uint64_t *l2_cache_entry_size,
862                              uint64_t *refcount_cache_size, Error **errp)
863 {
864     BDRVQcow2State *s = bs->opaque;
865     uint64_t combined_cache_size, l2_cache_max_setting;
866     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
867     bool l2_cache_entry_size_set;
868     int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size;
869     uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
870     uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size);
871     /* An L2 table is always one cluster in size so the max cache size
872      * should be a multiple of the cluster size. */
873     uint64_t max_l2_cache = ROUND_UP(max_l2_entries * sizeof(uint64_t),
874                                      s->cluster_size);
875 
876     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
877     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
878     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
879     l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE);
880 
881     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
882     l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE,
883                                              DEFAULT_L2_CACHE_MAX_SIZE);
884     *refcount_cache_size = qemu_opt_get_size(opts,
885                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
886 
887     *l2_cache_entry_size = qemu_opt_get_size(
888         opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size);
889 
890     *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting);
891 
892     if (combined_cache_size_set) {
893         if (l2_cache_size_set && refcount_cache_size_set) {
894             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
895                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
896                        "at the same time");
897             return;
898         } else if (l2_cache_size_set &&
899                    (l2_cache_max_setting > combined_cache_size)) {
900             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
901                        QCOW2_OPT_CACHE_SIZE);
902             return;
903         } else if (*refcount_cache_size > combined_cache_size) {
904             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
905                        QCOW2_OPT_CACHE_SIZE);
906             return;
907         }
908 
909         if (l2_cache_size_set) {
910             *refcount_cache_size = combined_cache_size - *l2_cache_size;
911         } else if (refcount_cache_size_set) {
912             *l2_cache_size = combined_cache_size - *refcount_cache_size;
913         } else {
914             /* Assign as much memory as possible to the L2 cache, and
915              * use the remainder for the refcount cache */
916             if (combined_cache_size >= max_l2_cache + min_refcount_cache) {
917                 *l2_cache_size = max_l2_cache;
918                 *refcount_cache_size = combined_cache_size - *l2_cache_size;
919             } else {
920                 *refcount_cache_size =
921                     MIN(combined_cache_size, min_refcount_cache);
922                 *l2_cache_size = combined_cache_size - *refcount_cache_size;
923             }
924         }
925     }
926 
927     /*
928      * If the L2 cache is not enough to cover the whole disk then
929      * default to 4KB entries. Smaller entries reduce the cost of
930      * loads and evictions and increase I/O performance.
931      */
932     if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) {
933         *l2_cache_entry_size = MIN(s->cluster_size, 4096);
934     }
935 
936     /* l2_cache_size and refcount_cache_size are ensured to have at least
937      * their minimum values in qcow2_update_options_prepare() */
938 
939     if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) ||
940         *l2_cache_entry_size > s->cluster_size ||
941         !is_power_of_2(*l2_cache_entry_size)) {
942         error_setg(errp, "L2 cache entry size must be a power of two "
943                    "between %d and the cluster size (%d)",
944                    1 << MIN_CLUSTER_BITS, s->cluster_size);
945         return;
946     }
947 }
948 
949 typedef struct Qcow2ReopenState {
950     Qcow2Cache *l2_table_cache;
951     Qcow2Cache *refcount_block_cache;
952     int l2_slice_size; /* Number of entries in a slice of the L2 table */
953     bool use_lazy_refcounts;
954     int overlap_check;
955     bool discard_passthrough[QCOW2_DISCARD_MAX];
956     uint64_t cache_clean_interval;
957     QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
958 } Qcow2ReopenState;
959 
960 static int qcow2_update_options_prepare(BlockDriverState *bs,
961                                         Qcow2ReopenState *r,
962                                         QDict *options, int flags,
963                                         Error **errp)
964 {
965     BDRVQcow2State *s = bs->opaque;
966     QemuOpts *opts = NULL;
967     const char *opt_overlap_check, *opt_overlap_check_template;
968     int overlap_check_template = 0;
969     uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size;
970     int i;
971     const char *encryptfmt;
972     QDict *encryptopts = NULL;
973     Error *local_err = NULL;
974     int ret;
975 
976     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
977     encryptfmt = qdict_get_try_str(encryptopts, "format");
978 
979     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
980     qemu_opts_absorb_qdict(opts, options, &local_err);
981     if (local_err) {
982         error_propagate(errp, local_err);
983         ret = -EINVAL;
984         goto fail;
985     }
986 
987     /* get L2 table/refcount block cache size from command line options */
988     read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size,
989                      &refcount_cache_size, &local_err);
990     if (local_err) {
991         error_propagate(errp, local_err);
992         ret = -EINVAL;
993         goto fail;
994     }
995 
996     l2_cache_size /= l2_cache_entry_size;
997     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
998         l2_cache_size = MIN_L2_CACHE_SIZE;
999     }
1000     if (l2_cache_size > INT_MAX) {
1001         error_setg(errp, "L2 cache size too big");
1002         ret = -EINVAL;
1003         goto fail;
1004     }
1005 
1006     refcount_cache_size /= s->cluster_size;
1007     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
1008         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
1009     }
1010     if (refcount_cache_size > INT_MAX) {
1011         error_setg(errp, "Refcount cache size too big");
1012         ret = -EINVAL;
1013         goto fail;
1014     }
1015 
1016     /* alloc new L2 table/refcount block cache, flush old one */
1017     if (s->l2_table_cache) {
1018         ret = qcow2_cache_flush(bs, s->l2_table_cache);
1019         if (ret) {
1020             error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
1021             goto fail;
1022         }
1023     }
1024 
1025     if (s->refcount_block_cache) {
1026         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1027         if (ret) {
1028             error_setg_errno(errp, -ret,
1029                              "Failed to flush the refcount block cache");
1030             goto fail;
1031         }
1032     }
1033 
1034     r->l2_slice_size = l2_cache_entry_size / sizeof(uint64_t);
1035     r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size,
1036                                            l2_cache_entry_size);
1037     r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size,
1038                                                  s->cluster_size);
1039     if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
1040         error_setg(errp, "Could not allocate metadata caches");
1041         ret = -ENOMEM;
1042         goto fail;
1043     }
1044 
1045     /* New interval for cache cleanup timer */
1046     r->cache_clean_interval =
1047         qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
1048                             DEFAULT_CACHE_CLEAN_INTERVAL);
1049 #ifndef CONFIG_LINUX
1050     if (r->cache_clean_interval != 0) {
1051         error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL
1052                    " not supported on this host");
1053         ret = -EINVAL;
1054         goto fail;
1055     }
1056 #endif
1057     if (r->cache_clean_interval > UINT_MAX) {
1058         error_setg(errp, "Cache clean interval too big");
1059         ret = -EINVAL;
1060         goto fail;
1061     }
1062 
1063     /* lazy-refcounts; flush if going from enabled to disabled */
1064     r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
1065         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
1066     if (r->use_lazy_refcounts && s->qcow_version < 3) {
1067         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
1068                    "qemu 1.1 compatibility level");
1069         ret = -EINVAL;
1070         goto fail;
1071     }
1072 
1073     if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
1074         ret = qcow2_mark_clean(bs);
1075         if (ret < 0) {
1076             error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
1077             goto fail;
1078         }
1079     }
1080 
1081     /* Overlap check options */
1082     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
1083     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
1084     if (opt_overlap_check_template && opt_overlap_check &&
1085         strcmp(opt_overlap_check_template, opt_overlap_check))
1086     {
1087         error_setg(errp, "Conflicting values for qcow2 options '"
1088                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1089                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
1090         ret = -EINVAL;
1091         goto fail;
1092     }
1093     if (!opt_overlap_check) {
1094         opt_overlap_check = opt_overlap_check_template ?: "cached";
1095     }
1096 
1097     if (!strcmp(opt_overlap_check, "none")) {
1098         overlap_check_template = 0;
1099     } else if (!strcmp(opt_overlap_check, "constant")) {
1100         overlap_check_template = QCOW2_OL_CONSTANT;
1101     } else if (!strcmp(opt_overlap_check, "cached")) {
1102         overlap_check_template = QCOW2_OL_CACHED;
1103     } else if (!strcmp(opt_overlap_check, "all")) {
1104         overlap_check_template = QCOW2_OL_ALL;
1105     } else {
1106         error_setg(errp, "Unsupported value '%s' for qcow2 option "
1107                    "'overlap-check'. Allowed are any of the following: "
1108                    "none, constant, cached, all", opt_overlap_check);
1109         ret = -EINVAL;
1110         goto fail;
1111     }
1112 
1113     r->overlap_check = 0;
1114     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
1115         /* overlap-check defines a template bitmask, but every flag may be
1116          * overwritten through the associated boolean option */
1117         r->overlap_check |=
1118             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
1119                               overlap_check_template & (1 << i)) << i;
1120     }
1121 
1122     r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
1123     r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
1124     r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
1125         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
1126                           flags & BDRV_O_UNMAP);
1127     r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
1128         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
1129     r->discard_passthrough[QCOW2_DISCARD_OTHER] =
1130         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
1131 
1132     switch (s->crypt_method_header) {
1133     case QCOW_CRYPT_NONE:
1134         if (encryptfmt) {
1135             error_setg(errp, "No encryption in image header, but options "
1136                        "specified format '%s'", encryptfmt);
1137             ret = -EINVAL;
1138             goto fail;
1139         }
1140         break;
1141 
1142     case QCOW_CRYPT_AES:
1143         if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
1144             error_setg(errp,
1145                        "Header reported 'aes' encryption format but "
1146                        "options specify '%s'", encryptfmt);
1147             ret = -EINVAL;
1148             goto fail;
1149         }
1150         qdict_put_str(encryptopts, "format", "qcow");
1151         r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1152         break;
1153 
1154     case QCOW_CRYPT_LUKS:
1155         if (encryptfmt && !g_str_equal(encryptfmt, "luks")) {
1156             error_setg(errp,
1157                        "Header reported 'luks' encryption format but "
1158                        "options specify '%s'", encryptfmt);
1159             ret = -EINVAL;
1160             goto fail;
1161         }
1162         qdict_put_str(encryptopts, "format", "luks");
1163         r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1164         break;
1165 
1166     default:
1167         error_setg(errp, "Unsupported encryption method %d",
1168                    s->crypt_method_header);
1169         break;
1170     }
1171     if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) {
1172         ret = -EINVAL;
1173         goto fail;
1174     }
1175 
1176     ret = 0;
1177 fail:
1178     qobject_unref(encryptopts);
1179     qemu_opts_del(opts);
1180     opts = NULL;
1181     return ret;
1182 }
1183 
1184 static void qcow2_update_options_commit(BlockDriverState *bs,
1185                                         Qcow2ReopenState *r)
1186 {
1187     BDRVQcow2State *s = bs->opaque;
1188     int i;
1189 
1190     if (s->l2_table_cache) {
1191         qcow2_cache_destroy(s->l2_table_cache);
1192     }
1193     if (s->refcount_block_cache) {
1194         qcow2_cache_destroy(s->refcount_block_cache);
1195     }
1196     s->l2_table_cache = r->l2_table_cache;
1197     s->refcount_block_cache = r->refcount_block_cache;
1198     s->l2_slice_size = r->l2_slice_size;
1199 
1200     s->overlap_check = r->overlap_check;
1201     s->use_lazy_refcounts = r->use_lazy_refcounts;
1202 
1203     for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
1204         s->discard_passthrough[i] = r->discard_passthrough[i];
1205     }
1206 
1207     if (s->cache_clean_interval != r->cache_clean_interval) {
1208         cache_clean_timer_del(bs);
1209         s->cache_clean_interval = r->cache_clean_interval;
1210         cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
1211     }
1212 
1213     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1214     s->crypto_opts = r->crypto_opts;
1215 }
1216 
1217 static void qcow2_update_options_abort(BlockDriverState *bs,
1218                                        Qcow2ReopenState *r)
1219 {
1220     if (r->l2_table_cache) {
1221         qcow2_cache_destroy(r->l2_table_cache);
1222     }
1223     if (r->refcount_block_cache) {
1224         qcow2_cache_destroy(r->refcount_block_cache);
1225     }
1226     qapi_free_QCryptoBlockOpenOptions(r->crypto_opts);
1227 }
1228 
1229 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
1230                                 int flags, Error **errp)
1231 {
1232     Qcow2ReopenState r = {};
1233     int ret;
1234 
1235     ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
1236     if (ret >= 0) {
1237         qcow2_update_options_commit(bs, &r);
1238     } else {
1239         qcow2_update_options_abort(bs, &r);
1240     }
1241 
1242     return ret;
1243 }
1244 
1245 /* Called with s->lock held.  */
1246 static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
1247                                       int flags, Error **errp)
1248 {
1249     BDRVQcow2State *s = bs->opaque;
1250     unsigned int len, i;
1251     int ret = 0;
1252     QCowHeader header;
1253     Error *local_err = NULL;
1254     uint64_t ext_end;
1255     uint64_t l1_vm_state_index;
1256     bool update_header = false;
1257 
1258     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
1259     if (ret < 0) {
1260         error_setg_errno(errp, -ret, "Could not read qcow2 header");
1261         goto fail;
1262     }
1263     header.magic = be32_to_cpu(header.magic);
1264     header.version = be32_to_cpu(header.version);
1265     header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
1266     header.backing_file_size = be32_to_cpu(header.backing_file_size);
1267     header.size = be64_to_cpu(header.size);
1268     header.cluster_bits = be32_to_cpu(header.cluster_bits);
1269     header.crypt_method = be32_to_cpu(header.crypt_method);
1270     header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
1271     header.l1_size = be32_to_cpu(header.l1_size);
1272     header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset);
1273     header.refcount_table_clusters =
1274         be32_to_cpu(header.refcount_table_clusters);
1275     header.snapshots_offset = be64_to_cpu(header.snapshots_offset);
1276     header.nb_snapshots = be32_to_cpu(header.nb_snapshots);
1277 
1278     if (header.magic != QCOW_MAGIC) {
1279         error_setg(errp, "Image is not in qcow2 format");
1280         ret = -EINVAL;
1281         goto fail;
1282     }
1283     if (header.version < 2 || header.version > 3) {
1284         error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
1285         ret = -ENOTSUP;
1286         goto fail;
1287     }
1288 
1289     s->qcow_version = header.version;
1290 
1291     /* Initialise cluster size */
1292     if (header.cluster_bits < MIN_CLUSTER_BITS ||
1293         header.cluster_bits > MAX_CLUSTER_BITS) {
1294         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
1295                    header.cluster_bits);
1296         ret = -EINVAL;
1297         goto fail;
1298     }
1299 
1300     s->cluster_bits = header.cluster_bits;
1301     s->cluster_size = 1 << s->cluster_bits;
1302 
1303     /* Initialise version 3 header fields */
1304     if (header.version == 2) {
1305         header.incompatible_features    = 0;
1306         header.compatible_features      = 0;
1307         header.autoclear_features       = 0;
1308         header.refcount_order           = 4;
1309         header.header_length            = 72;
1310     } else {
1311         header.incompatible_features =
1312             be64_to_cpu(header.incompatible_features);
1313         header.compatible_features = be64_to_cpu(header.compatible_features);
1314         header.autoclear_features = be64_to_cpu(header.autoclear_features);
1315         header.refcount_order = be32_to_cpu(header.refcount_order);
1316         header.header_length = be32_to_cpu(header.header_length);
1317 
1318         if (header.header_length < 104) {
1319             error_setg(errp, "qcow2 header too short");
1320             ret = -EINVAL;
1321             goto fail;
1322         }
1323     }
1324 
1325     if (header.header_length > s->cluster_size) {
1326         error_setg(errp, "qcow2 header exceeds cluster size");
1327         ret = -EINVAL;
1328         goto fail;
1329     }
1330 
1331     if (header.header_length > sizeof(header)) {
1332         s->unknown_header_fields_size = header.header_length - sizeof(header);
1333         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
1334         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
1335                          s->unknown_header_fields_size);
1336         if (ret < 0) {
1337             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
1338                              "fields");
1339             goto fail;
1340         }
1341     }
1342 
1343     if (header.backing_file_offset > s->cluster_size) {
1344         error_setg(errp, "Invalid backing file offset");
1345         ret = -EINVAL;
1346         goto fail;
1347     }
1348 
1349     if (header.backing_file_offset) {
1350         ext_end = header.backing_file_offset;
1351     } else {
1352         ext_end = 1 << header.cluster_bits;
1353     }
1354 
1355     /* Handle feature bits */
1356     s->incompatible_features    = header.incompatible_features;
1357     s->compatible_features      = header.compatible_features;
1358     s->autoclear_features       = header.autoclear_features;
1359 
1360     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
1361         void *feature_table = NULL;
1362         qcow2_read_extensions(bs, header.header_length, ext_end,
1363                               &feature_table, flags, NULL, NULL);
1364         report_unsupported_feature(errp, feature_table,
1365                                    s->incompatible_features &
1366                                    ~QCOW2_INCOMPAT_MASK);
1367         ret = -ENOTSUP;
1368         g_free(feature_table);
1369         goto fail;
1370     }
1371 
1372     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
1373         /* Corrupt images may not be written to unless they are being repaired
1374          */
1375         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
1376             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
1377                        "read/write");
1378             ret = -EACCES;
1379             goto fail;
1380         }
1381     }
1382 
1383     /* Check support for various header values */
1384     if (header.refcount_order > 6) {
1385         error_setg(errp, "Reference count entry width too large; may not "
1386                    "exceed 64 bits");
1387         ret = -EINVAL;
1388         goto fail;
1389     }
1390     s->refcount_order = header.refcount_order;
1391     s->refcount_bits = 1 << s->refcount_order;
1392     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1393     s->refcount_max += s->refcount_max - 1;
1394 
1395     s->crypt_method_header = header.crypt_method;
1396     if (s->crypt_method_header) {
1397         if (bdrv_uses_whitelist() &&
1398             s->crypt_method_header == QCOW_CRYPT_AES) {
1399             error_setg(errp,
1400                        "Use of AES-CBC encrypted qcow2 images is no longer "
1401                        "supported in system emulators");
1402             error_append_hint(errp,
1403                               "You can use 'qemu-img convert' to convert your "
1404                               "image to an alternative supported format, such "
1405                               "as unencrypted qcow2, or raw with the LUKS "
1406                               "format instead.\n");
1407             ret = -ENOSYS;
1408             goto fail;
1409         }
1410 
1411         if (s->crypt_method_header == QCOW_CRYPT_AES) {
1412             s->crypt_physical_offset = false;
1413         } else {
1414             /* Assuming LUKS and any future crypt methods we
1415              * add will all use physical offsets, due to the
1416              * fact that the alternative is insecure...  */
1417             s->crypt_physical_offset = true;
1418         }
1419 
1420         bs->encrypted = true;
1421     }
1422 
1423     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
1424     s->l2_size = 1 << s->l2_bits;
1425     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1426     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1427     s->refcount_block_size = 1 << s->refcount_block_bits;
1428     bs->total_sectors = header.size / BDRV_SECTOR_SIZE;
1429     s->csize_shift = (62 - (s->cluster_bits - 8));
1430     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1431     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1432 
1433     s->refcount_table_offset = header.refcount_table_offset;
1434     s->refcount_table_size =
1435         header.refcount_table_clusters << (s->cluster_bits - 3);
1436 
1437     if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) {
1438         error_setg(errp, "Image does not contain a reference count table");
1439         ret = -EINVAL;
1440         goto fail;
1441     }
1442 
1443     ret = qcow2_validate_table(bs, s->refcount_table_offset,
1444                                header.refcount_table_clusters,
1445                                s->cluster_size, QCOW_MAX_REFTABLE_SIZE,
1446                                "Reference count table", errp);
1447     if (ret < 0) {
1448         goto fail;
1449     }
1450 
1451     if (!(flags & BDRV_O_CHECK)) {
1452         /*
1453          * The total size in bytes of the snapshot table is checked in
1454          * qcow2_read_snapshots() because the size of each snapshot is
1455          * variable and we don't know it yet.
1456          * Here we only check the offset and number of snapshots.
1457          */
1458         ret = qcow2_validate_table(bs, header.snapshots_offset,
1459                                    header.nb_snapshots,
1460                                    sizeof(QCowSnapshotHeader),
1461                                    sizeof(QCowSnapshotHeader) *
1462                                        QCOW_MAX_SNAPSHOTS,
1463                                    "Snapshot table", errp);
1464         if (ret < 0) {
1465             goto fail;
1466         }
1467     }
1468 
1469     /* read the level 1 table */
1470     ret = qcow2_validate_table(bs, header.l1_table_offset,
1471                                header.l1_size, sizeof(uint64_t),
1472                                QCOW_MAX_L1_SIZE, "Active L1 table", errp);
1473     if (ret < 0) {
1474         goto fail;
1475     }
1476     s->l1_size = header.l1_size;
1477     s->l1_table_offset = header.l1_table_offset;
1478 
1479     l1_vm_state_index = size_to_l1(s, header.size);
1480     if (l1_vm_state_index > INT_MAX) {
1481         error_setg(errp, "Image is too big");
1482         ret = -EFBIG;
1483         goto fail;
1484     }
1485     s->l1_vm_state_index = l1_vm_state_index;
1486 
1487     /* the L1 table must contain at least enough entries to put
1488        header.size bytes */
1489     if (s->l1_size < s->l1_vm_state_index) {
1490         error_setg(errp, "L1 table is too small");
1491         ret = -EINVAL;
1492         goto fail;
1493     }
1494 
1495     if (s->l1_size > 0) {
1496         s->l1_table = qemu_try_blockalign(bs->file->bs,
1497                                           s->l1_size * sizeof(uint64_t));
1498         if (s->l1_table == NULL) {
1499             error_setg(errp, "Could not allocate L1 table");
1500             ret = -ENOMEM;
1501             goto fail;
1502         }
1503         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
1504                          s->l1_size * sizeof(uint64_t));
1505         if (ret < 0) {
1506             error_setg_errno(errp, -ret, "Could not read L1 table");
1507             goto fail;
1508         }
1509         for(i = 0;i < s->l1_size; i++) {
1510             s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
1511         }
1512     }
1513 
1514     /* Parse driver-specific options */
1515     ret = qcow2_update_options(bs, options, flags, errp);
1516     if (ret < 0) {
1517         goto fail;
1518     }
1519 
1520     s->flags = flags;
1521 
1522     ret = qcow2_refcount_init(bs);
1523     if (ret != 0) {
1524         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1525         goto fail;
1526     }
1527 
1528     QLIST_INIT(&s->cluster_allocs);
1529     QTAILQ_INIT(&s->discards);
1530 
1531     /* read qcow2 extensions */
1532     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1533                               flags, &update_header, &local_err)) {
1534         error_propagate(errp, local_err);
1535         ret = -EINVAL;
1536         goto fail;
1537     }
1538 
1539     /* Open external data file */
1540     s->data_file = bdrv_open_child(NULL, options, "data-file", bs, &child_file,
1541                                    true, &local_err);
1542     if (local_err) {
1543         error_propagate(errp, local_err);
1544         ret = -EINVAL;
1545         goto fail;
1546     }
1547 
1548     if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) {
1549         if (!s->data_file && s->image_data_file) {
1550             s->data_file = bdrv_open_child(s->image_data_file, options,
1551                                            "data-file", bs, &child_file,
1552                                            false, errp);
1553             if (!s->data_file) {
1554                 ret = -EINVAL;
1555                 goto fail;
1556             }
1557         }
1558         if (!s->data_file) {
1559             error_setg(errp, "'data-file' is required for this image");
1560             ret = -EINVAL;
1561             goto fail;
1562         }
1563     } else {
1564         if (s->data_file) {
1565             error_setg(errp, "'data-file' can only be set for images with an "
1566                              "external data file");
1567             ret = -EINVAL;
1568             goto fail;
1569         }
1570 
1571         s->data_file = bs->file;
1572 
1573         if (data_file_is_raw(bs)) {
1574             error_setg(errp, "data-file-raw requires a data file");
1575             ret = -EINVAL;
1576             goto fail;
1577         }
1578     }
1579 
1580     /* qcow2_read_extension may have set up the crypto context
1581      * if the crypt method needs a header region, some methods
1582      * don't need header extensions, so must check here
1583      */
1584     if (s->crypt_method_header && !s->crypto) {
1585         if (s->crypt_method_header == QCOW_CRYPT_AES) {
1586             unsigned int cflags = 0;
1587             if (flags & BDRV_O_NO_IO) {
1588                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1589             }
1590             s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
1591                                            NULL, NULL, cflags,
1592                                            QCOW2_MAX_THREADS, errp);
1593             if (!s->crypto) {
1594                 ret = -EINVAL;
1595                 goto fail;
1596             }
1597         } else if (!(flags & BDRV_O_NO_IO)) {
1598             error_setg(errp, "Missing CRYPTO header for crypt method %d",
1599                        s->crypt_method_header);
1600             ret = -EINVAL;
1601             goto fail;
1602         }
1603     }
1604 
1605     /* read the backing file name */
1606     if (header.backing_file_offset != 0) {
1607         len = header.backing_file_size;
1608         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1609             len >= sizeof(bs->backing_file)) {
1610             error_setg(errp, "Backing file name too long");
1611             ret = -EINVAL;
1612             goto fail;
1613         }
1614         ret = bdrv_pread(bs->file, header.backing_file_offset,
1615                          bs->auto_backing_file, len);
1616         if (ret < 0) {
1617             error_setg_errno(errp, -ret, "Could not read backing file name");
1618             goto fail;
1619         }
1620         bs->auto_backing_file[len] = '\0';
1621         pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1622                 bs->auto_backing_file);
1623         s->image_backing_file = g_strdup(bs->auto_backing_file);
1624     }
1625 
1626     /*
1627      * Internal snapshots; skip reading them in check mode, because
1628      * we do not need them then, and we do not want to abort because
1629      * of a broken table.
1630      */
1631     if (!(flags & BDRV_O_CHECK)) {
1632         s->snapshots_offset = header.snapshots_offset;
1633         s->nb_snapshots = header.nb_snapshots;
1634 
1635         ret = qcow2_read_snapshots(bs, errp);
1636         if (ret < 0) {
1637             goto fail;
1638         }
1639     }
1640 
1641     /* Clear unknown autoclear feature bits */
1642     update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
1643     update_header =
1644         update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE);
1645     if (update_header) {
1646         s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
1647     }
1648 
1649     /* == Handle persistent dirty bitmaps ==
1650      *
1651      * We want load dirty bitmaps in three cases:
1652      *
1653      * 1. Normal open of the disk in active mode, not related to invalidation
1654      *    after migration.
1655      *
1656      * 2. Invalidation of the target vm after pre-copy phase of migration, if
1657      *    bitmaps are _not_ migrating through migration channel, i.e.
1658      *    'dirty-bitmaps' capability is disabled.
1659      *
1660      * 3. Invalidation of source vm after failed or canceled migration.
1661      *    This is a very interesting case. There are two possible types of
1662      *    bitmaps:
1663      *
1664      *    A. Stored on inactivation and removed. They should be loaded from the
1665      *       image.
1666      *
1667      *    B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1668      *       the migration channel (with dirty-bitmaps capability).
1669      *
1670      *    On the other hand, there are two possible sub-cases:
1671      *
1672      *    3.1 disk was changed by somebody else while were inactive. In this
1673      *        case all in-RAM dirty bitmaps (both persistent and not) are
1674      *        definitely invalid. And we don't have any method to determine
1675      *        this.
1676      *
1677      *        Simple and safe thing is to just drop all the bitmaps of type B on
1678      *        inactivation. But in this case we lose bitmaps in valid 4.2 case.
1679      *
1680      *        On the other hand, resuming source vm, if disk was already changed
1681      *        is a bad thing anyway: not only bitmaps, the whole vm state is
1682      *        out of sync with disk.
1683      *
1684      *        This means, that user or management tool, who for some reason
1685      *        decided to resume source vm, after disk was already changed by
1686      *        target vm, should at least drop all dirty bitmaps by hand.
1687      *
1688      *        So, we can ignore this case for now, but TODO: "generation"
1689      *        extension for qcow2, to determine, that image was changed after
1690      *        last inactivation. And if it is changed, we will drop (or at least
1691      *        mark as 'invalid' all the bitmaps of type B, both persistent
1692      *        and not).
1693      *
1694      *    3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1695      *        to disk ('dirty-bitmaps' capability disabled), or not saved
1696      *        ('dirty-bitmaps' capability enabled), but we don't need to care
1697      *        of: let's load bitmaps as always: stored bitmaps will be loaded,
1698      *        and not stored has flag IN_USE=1 in the image and will be skipped
1699      *        on loading.
1700      *
1701      * One remaining possible case when we don't want load bitmaps:
1702      *
1703      * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1704      *    will be loaded on invalidation, no needs try loading them before)
1705      */
1706 
1707     if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) {
1708         /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1709         bool header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
1710         if (local_err != NULL) {
1711             error_propagate(errp, local_err);
1712             ret = -EINVAL;
1713             goto fail;
1714         }
1715 
1716         update_header = update_header && !header_updated;
1717     }
1718 
1719     if (update_header) {
1720         ret = qcow2_update_header(bs);
1721         if (ret < 0) {
1722             error_setg_errno(errp, -ret, "Could not update qcow2 header");
1723             goto fail;
1724         }
1725     }
1726 
1727     bs->supported_zero_flags = header.version >= 3 ?
1728                                BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0;
1729     bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
1730 
1731     /* Repair image if dirty */
1732     if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1733         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1734         BdrvCheckResult result = {0};
1735 
1736         ret = qcow2_co_check_locked(bs, &result,
1737                                     BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1738         if (ret < 0 || result.check_errors) {
1739             if (ret >= 0) {
1740                 ret = -EIO;
1741             }
1742             error_setg_errno(errp, -ret, "Could not repair dirty image");
1743             goto fail;
1744         }
1745     }
1746 
1747 #ifdef DEBUG_ALLOC
1748     {
1749         BdrvCheckResult result = {0};
1750         qcow2_check_refcounts(bs, &result, 0);
1751     }
1752 #endif
1753 
1754     qemu_co_queue_init(&s->thread_task_queue);
1755 
1756     return ret;
1757 
1758  fail:
1759     g_free(s->image_data_file);
1760     if (has_data_file(bs)) {
1761         bdrv_unref_child(bs, s->data_file);
1762         s->data_file = NULL;
1763     }
1764     g_free(s->unknown_header_fields);
1765     cleanup_unknown_header_ext(bs);
1766     qcow2_free_snapshots(bs);
1767     qcow2_refcount_close(bs);
1768     qemu_vfree(s->l1_table);
1769     /* else pre-write overlap checks in cache_destroy may crash */
1770     s->l1_table = NULL;
1771     cache_clean_timer_del(bs);
1772     if (s->l2_table_cache) {
1773         qcow2_cache_destroy(s->l2_table_cache);
1774     }
1775     if (s->refcount_block_cache) {
1776         qcow2_cache_destroy(s->refcount_block_cache);
1777     }
1778     qcrypto_block_free(s->crypto);
1779     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1780     return ret;
1781 }
1782 
1783 typedef struct QCow2OpenCo {
1784     BlockDriverState *bs;
1785     QDict *options;
1786     int flags;
1787     Error **errp;
1788     int ret;
1789 } QCow2OpenCo;
1790 
1791 static void coroutine_fn qcow2_open_entry(void *opaque)
1792 {
1793     QCow2OpenCo *qoc = opaque;
1794     BDRVQcow2State *s = qoc->bs->opaque;
1795 
1796     qemu_co_mutex_lock(&s->lock);
1797     qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp);
1798     qemu_co_mutex_unlock(&s->lock);
1799 }
1800 
1801 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1802                       Error **errp)
1803 {
1804     BDRVQcow2State *s = bs->opaque;
1805     QCow2OpenCo qoc = {
1806         .bs = bs,
1807         .options = options,
1808         .flags = flags,
1809         .errp = errp,
1810         .ret = -EINPROGRESS
1811     };
1812 
1813     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
1814                                false, errp);
1815     if (!bs->file) {
1816         return -EINVAL;
1817     }
1818 
1819     /* Initialise locks */
1820     qemu_co_mutex_init(&s->lock);
1821 
1822     if (qemu_in_coroutine()) {
1823         /* From bdrv_co_create.  */
1824         qcow2_open_entry(&qoc);
1825     } else {
1826         assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1827         qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc));
1828         BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
1829     }
1830     return qoc.ret;
1831 }
1832 
1833 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1834 {
1835     BDRVQcow2State *s = bs->opaque;
1836 
1837     if (bs->encrypted) {
1838         /* Encryption works on a sector granularity */
1839         bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto);
1840     }
1841     bs->bl.pwrite_zeroes_alignment = s->cluster_size;
1842     bs->bl.pdiscard_alignment = s->cluster_size;
1843 }
1844 
1845 static int qcow2_reopen_prepare(BDRVReopenState *state,
1846                                 BlockReopenQueue *queue, Error **errp)
1847 {
1848     Qcow2ReopenState *r;
1849     int ret;
1850 
1851     r = g_new0(Qcow2ReopenState, 1);
1852     state->opaque = r;
1853 
1854     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1855                                        state->flags, errp);
1856     if (ret < 0) {
1857         goto fail;
1858     }
1859 
1860     /* We need to write out any unwritten data if we reopen read-only. */
1861     if ((state->flags & BDRV_O_RDWR) == 0) {
1862         ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1863         if (ret < 0) {
1864             goto fail;
1865         }
1866 
1867         ret = bdrv_flush(state->bs);
1868         if (ret < 0) {
1869             goto fail;
1870         }
1871 
1872         ret = qcow2_mark_clean(state->bs);
1873         if (ret < 0) {
1874             goto fail;
1875         }
1876     }
1877 
1878     return 0;
1879 
1880 fail:
1881     qcow2_update_options_abort(state->bs, r);
1882     g_free(r);
1883     return ret;
1884 }
1885 
1886 static void qcow2_reopen_commit(BDRVReopenState *state)
1887 {
1888     qcow2_update_options_commit(state->bs, state->opaque);
1889     g_free(state->opaque);
1890 }
1891 
1892 static void qcow2_reopen_commit_post(BDRVReopenState *state)
1893 {
1894     if (state->flags & BDRV_O_RDWR) {
1895         Error *local_err = NULL;
1896 
1897         if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) {
1898             /*
1899              * This is not fatal, bitmaps just left read-only, so all following
1900              * writes will fail. User can remove read-only bitmaps to unblock
1901              * writes or retry reopen.
1902              */
1903             error_reportf_err(local_err,
1904                               "%s: Failed to make dirty bitmaps writable: ",
1905                               bdrv_get_node_name(state->bs));
1906         }
1907     }
1908 }
1909 
1910 static void qcow2_reopen_abort(BDRVReopenState *state)
1911 {
1912     qcow2_update_options_abort(state->bs, state->opaque);
1913     g_free(state->opaque);
1914 }
1915 
1916 static void qcow2_join_options(QDict *options, QDict *old_options)
1917 {
1918     bool has_new_overlap_template =
1919         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1920         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1921     bool has_new_total_cache_size =
1922         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1923     bool has_all_cache_options;
1924 
1925     /* New overlap template overrides all old overlap options */
1926     if (has_new_overlap_template) {
1927         qdict_del(old_options, QCOW2_OPT_OVERLAP);
1928         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1929         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1930         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1931         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1932         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1933         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1934         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1935         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1936         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1937     }
1938 
1939     /* New total cache size overrides all old options */
1940     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1941         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1942         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1943     }
1944 
1945     qdict_join(options, old_options, false);
1946 
1947     /*
1948      * If after merging all cache size options are set, an old total size is
1949      * overwritten. Do keep all options, however, if all three are new. The
1950      * resulting error message is what we want to happen.
1951      */
1952     has_all_cache_options =
1953         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1954         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1955         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1956 
1957     if (has_all_cache_options && !has_new_total_cache_size) {
1958         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1959     }
1960 }
1961 
1962 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
1963                                               bool want_zero,
1964                                               int64_t offset, int64_t count,
1965                                               int64_t *pnum, int64_t *map,
1966                                               BlockDriverState **file)
1967 {
1968     BDRVQcow2State *s = bs->opaque;
1969     uint64_t cluster_offset;
1970     unsigned int bytes;
1971     int ret, status = 0;
1972 
1973     qemu_co_mutex_lock(&s->lock);
1974 
1975     if (!s->metadata_preallocation_checked) {
1976         ret = qcow2_detect_metadata_preallocation(bs);
1977         s->metadata_preallocation = (ret == 1);
1978         s->metadata_preallocation_checked = true;
1979     }
1980 
1981     bytes = MIN(INT_MAX, count);
1982     ret = qcow2_get_cluster_offset(bs, offset, &bytes, &cluster_offset);
1983     qemu_co_mutex_unlock(&s->lock);
1984     if (ret < 0) {
1985         return ret;
1986     }
1987 
1988     *pnum = bytes;
1989 
1990     if ((ret == QCOW2_CLUSTER_NORMAL || ret == QCOW2_CLUSTER_ZERO_ALLOC) &&
1991         !s->crypto) {
1992         *map = cluster_offset | offset_into_cluster(s, offset);
1993         *file = s->data_file->bs;
1994         status |= BDRV_BLOCK_OFFSET_VALID;
1995     }
1996     if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) {
1997         status |= BDRV_BLOCK_ZERO;
1998     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1999         status |= BDRV_BLOCK_DATA;
2000     }
2001     if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) &&
2002         (status & BDRV_BLOCK_OFFSET_VALID))
2003     {
2004         status |= BDRV_BLOCK_RECURSE;
2005     }
2006     return status;
2007 }
2008 
2009 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs,
2010                                             QCowL2Meta **pl2meta,
2011                                             bool link_l2)
2012 {
2013     int ret = 0;
2014     QCowL2Meta *l2meta = *pl2meta;
2015 
2016     while (l2meta != NULL) {
2017         QCowL2Meta *next;
2018 
2019         if (link_l2) {
2020             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
2021             if (ret) {
2022                 goto out;
2023             }
2024         } else {
2025             qcow2_alloc_cluster_abort(bs, l2meta);
2026         }
2027 
2028         /* Take the request off the list of running requests */
2029         if (l2meta->nb_clusters != 0) {
2030             QLIST_REMOVE(l2meta, next_in_flight);
2031         }
2032 
2033         qemu_co_queue_restart_all(&l2meta->dependent_requests);
2034 
2035         next = l2meta->next;
2036         g_free(l2meta);
2037         l2meta = next;
2038     }
2039 out:
2040     *pl2meta = l2meta;
2041     return ret;
2042 }
2043 
2044 static coroutine_fn int
2045 qcow2_co_preadv_encrypted(BlockDriverState *bs,
2046                            uint64_t file_cluster_offset,
2047                            uint64_t offset,
2048                            uint64_t bytes,
2049                            QEMUIOVector *qiov,
2050                            uint64_t qiov_offset)
2051 {
2052     int ret;
2053     BDRVQcow2State *s = bs->opaque;
2054     uint8_t *buf;
2055 
2056     assert(bs->encrypted && s->crypto);
2057     assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2058 
2059     /*
2060      * For encrypted images, read everything into a temporary
2061      * contiguous buffer on which the AES functions can work.
2062      * Also, decryption in a separate buffer is better as it
2063      * prevents the guest from learning information about the
2064      * encrypted nature of the virtual disk.
2065      */
2066 
2067     buf = qemu_try_blockalign(s->data_file->bs, bytes);
2068     if (buf == NULL) {
2069         return -ENOMEM;
2070     }
2071 
2072     BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2073     ret = bdrv_co_pread(s->data_file,
2074                         file_cluster_offset + offset_into_cluster(s, offset),
2075                         bytes, buf, 0);
2076     if (ret < 0) {
2077         goto fail;
2078     }
2079 
2080     if (qcow2_co_decrypt(bs,
2081                          file_cluster_offset + offset_into_cluster(s, offset),
2082                          offset, buf, bytes) < 0)
2083     {
2084         ret = -EIO;
2085         goto fail;
2086     }
2087     qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes);
2088 
2089 fail:
2090     qemu_vfree(buf);
2091 
2092     return ret;
2093 }
2094 
2095 typedef struct Qcow2AioTask {
2096     AioTask task;
2097 
2098     BlockDriverState *bs;
2099     QCow2ClusterType cluster_type; /* only for read */
2100     uint64_t file_cluster_offset;
2101     uint64_t offset;
2102     uint64_t bytes;
2103     QEMUIOVector *qiov;
2104     uint64_t qiov_offset;
2105     QCowL2Meta *l2meta; /* only for write */
2106 } Qcow2AioTask;
2107 
2108 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task);
2109 static coroutine_fn int qcow2_add_task(BlockDriverState *bs,
2110                                        AioTaskPool *pool,
2111                                        AioTaskFunc func,
2112                                        QCow2ClusterType cluster_type,
2113                                        uint64_t file_cluster_offset,
2114                                        uint64_t offset,
2115                                        uint64_t bytes,
2116                                        QEMUIOVector *qiov,
2117                                        size_t qiov_offset,
2118                                        QCowL2Meta *l2meta)
2119 {
2120     Qcow2AioTask local_task;
2121     Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task;
2122 
2123     *task = (Qcow2AioTask) {
2124         .task.func = func,
2125         .bs = bs,
2126         .cluster_type = cluster_type,
2127         .qiov = qiov,
2128         .file_cluster_offset = file_cluster_offset,
2129         .offset = offset,
2130         .bytes = bytes,
2131         .qiov_offset = qiov_offset,
2132         .l2meta = l2meta,
2133     };
2134 
2135     trace_qcow2_add_task(qemu_coroutine_self(), bs, pool,
2136                          func == qcow2_co_preadv_task_entry ? "read" : "write",
2137                          cluster_type, file_cluster_offset, offset, bytes,
2138                          qiov, qiov_offset);
2139 
2140     if (!pool) {
2141         return func(&task->task);
2142     }
2143 
2144     aio_task_pool_start_task(pool, &task->task);
2145 
2146     return 0;
2147 }
2148 
2149 static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs,
2150                                              QCow2ClusterType cluster_type,
2151                                              uint64_t file_cluster_offset,
2152                                              uint64_t offset, uint64_t bytes,
2153                                              QEMUIOVector *qiov,
2154                                              size_t qiov_offset)
2155 {
2156     BDRVQcow2State *s = bs->opaque;
2157     int offset_in_cluster = offset_into_cluster(s, offset);
2158 
2159     switch (cluster_type) {
2160     case QCOW2_CLUSTER_ZERO_PLAIN:
2161     case QCOW2_CLUSTER_ZERO_ALLOC:
2162         /* Both zero types are handled in qcow2_co_preadv_part */
2163         g_assert_not_reached();
2164 
2165     case QCOW2_CLUSTER_UNALLOCATED:
2166         assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */
2167 
2168         BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
2169         return bdrv_co_preadv_part(bs->backing, offset, bytes,
2170                                    qiov, qiov_offset, 0);
2171 
2172     case QCOW2_CLUSTER_COMPRESSED:
2173         return qcow2_co_preadv_compressed(bs, file_cluster_offset,
2174                                           offset, bytes, qiov, qiov_offset);
2175 
2176     case QCOW2_CLUSTER_NORMAL:
2177         assert(offset_into_cluster(s, file_cluster_offset) == 0);
2178         if (bs->encrypted) {
2179             return qcow2_co_preadv_encrypted(bs, file_cluster_offset,
2180                                              offset, bytes, qiov, qiov_offset);
2181         }
2182 
2183         BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2184         return bdrv_co_preadv_part(s->data_file,
2185                                    file_cluster_offset + offset_in_cluster,
2186                                    bytes, qiov, qiov_offset, 0);
2187 
2188     default:
2189         g_assert_not_reached();
2190     }
2191 
2192     g_assert_not_reached();
2193 }
2194 
2195 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task)
2196 {
2197     Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2198 
2199     assert(!t->l2meta);
2200 
2201     return qcow2_co_preadv_task(t->bs, t->cluster_type, t->file_cluster_offset,
2202                                 t->offset, t->bytes, t->qiov, t->qiov_offset);
2203 }
2204 
2205 static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs,
2206                                              uint64_t offset, uint64_t bytes,
2207                                              QEMUIOVector *qiov,
2208                                              size_t qiov_offset, int flags)
2209 {
2210     BDRVQcow2State *s = bs->opaque;
2211     int ret = 0;
2212     unsigned int cur_bytes; /* number of bytes in current iteration */
2213     uint64_t cluster_offset = 0;
2214     AioTaskPool *aio = NULL;
2215 
2216     while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2217         /* prepare next request */
2218         cur_bytes = MIN(bytes, INT_MAX);
2219         if (s->crypto) {
2220             cur_bytes = MIN(cur_bytes,
2221                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2222         }
2223 
2224         qemu_co_mutex_lock(&s->lock);
2225         ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset);
2226         qemu_co_mutex_unlock(&s->lock);
2227         if (ret < 0) {
2228             goto out;
2229         }
2230 
2231         if (ret == QCOW2_CLUSTER_ZERO_PLAIN ||
2232             ret == QCOW2_CLUSTER_ZERO_ALLOC ||
2233             (ret == QCOW2_CLUSTER_UNALLOCATED && !bs->backing))
2234         {
2235             qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes);
2236         } else {
2237             if (!aio && cur_bytes != bytes) {
2238                 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2239             }
2240             ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, ret,
2241                                  cluster_offset, offset, cur_bytes,
2242                                  qiov, qiov_offset, NULL);
2243             if (ret < 0) {
2244                 goto out;
2245             }
2246         }
2247 
2248         bytes -= cur_bytes;
2249         offset += cur_bytes;
2250         qiov_offset += cur_bytes;
2251     }
2252 
2253 out:
2254     if (aio) {
2255         aio_task_pool_wait_all(aio);
2256         if (ret == 0) {
2257             ret = aio_task_pool_status(aio);
2258         }
2259         g_free(aio);
2260     }
2261 
2262     return ret;
2263 }
2264 
2265 /* Check if it's possible to merge a write request with the writing of
2266  * the data from the COW regions */
2267 static bool merge_cow(uint64_t offset, unsigned bytes,
2268                       QEMUIOVector *qiov, size_t qiov_offset,
2269                       QCowL2Meta *l2meta)
2270 {
2271     QCowL2Meta *m;
2272 
2273     for (m = l2meta; m != NULL; m = m->next) {
2274         /* If both COW regions are empty then there's nothing to merge */
2275         if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
2276             continue;
2277         }
2278 
2279         /* If COW regions are handled already, skip this too */
2280         if (m->skip_cow) {
2281             continue;
2282         }
2283 
2284         /* The data (middle) region must be immediately after the
2285          * start region */
2286         if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
2287             continue;
2288         }
2289 
2290         /* The end region must be immediately after the data (middle)
2291          * region */
2292         if (m->offset + m->cow_end.offset != offset + bytes) {
2293             continue;
2294         }
2295 
2296         /* Make sure that adding both COW regions to the QEMUIOVector
2297          * does not exceed IOV_MAX */
2298         if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) {
2299             continue;
2300         }
2301 
2302         m->data_qiov = qiov;
2303         m->data_qiov_offset = qiov_offset;
2304         return true;
2305     }
2306 
2307     return false;
2308 }
2309 
2310 static bool is_unallocated(BlockDriverState *bs, int64_t offset, int64_t bytes)
2311 {
2312     int64_t nr;
2313     return !bytes ||
2314         (!bdrv_is_allocated_above(bs, NULL, false, offset, bytes, &nr) &&
2315          nr == bytes);
2316 }
2317 
2318 static bool is_zero_cow(BlockDriverState *bs, QCowL2Meta *m)
2319 {
2320     /*
2321      * This check is designed for optimization shortcut so it must be
2322      * efficient.
2323      * Instead of is_zero(), use is_unallocated() as it is faster (but not
2324      * as accurate and can result in false negatives).
2325      */
2326     return is_unallocated(bs, m->offset + m->cow_start.offset,
2327                           m->cow_start.nb_bytes) &&
2328            is_unallocated(bs, m->offset + m->cow_end.offset,
2329                           m->cow_end.nb_bytes);
2330 }
2331 
2332 static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
2333 {
2334     BDRVQcow2State *s = bs->opaque;
2335     QCowL2Meta *m;
2336 
2337     if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) {
2338         return 0;
2339     }
2340 
2341     if (bs->encrypted) {
2342         return 0;
2343     }
2344 
2345     for (m = l2meta; m != NULL; m = m->next) {
2346         int ret;
2347 
2348         if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
2349             continue;
2350         }
2351 
2352         if (!is_zero_cow(bs, m)) {
2353             continue;
2354         }
2355 
2356         /*
2357          * instead of writing zero COW buffers,
2358          * efficiently zero out the whole clusters
2359          */
2360 
2361         ret = qcow2_pre_write_overlap_check(bs, 0, m->alloc_offset,
2362                                             m->nb_clusters * s->cluster_size,
2363                                             true);
2364         if (ret < 0) {
2365             return ret;
2366         }
2367 
2368         BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE);
2369         ret = bdrv_co_pwrite_zeroes(s->data_file, m->alloc_offset,
2370                                     m->nb_clusters * s->cluster_size,
2371                                     BDRV_REQ_NO_FALLBACK);
2372         if (ret < 0) {
2373             if (ret != -ENOTSUP && ret != -EAGAIN) {
2374                 return ret;
2375             }
2376             continue;
2377         }
2378 
2379         trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters);
2380         m->skip_cow = true;
2381     }
2382     return 0;
2383 }
2384 
2385 /*
2386  * qcow2_co_pwritev_task
2387  * Called with s->lock unlocked
2388  * l2meta  - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2389  *           not use it somehow after qcow2_co_pwritev_task() call
2390  */
2391 static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs,
2392                                               uint64_t file_cluster_offset,
2393                                               uint64_t offset, uint64_t bytes,
2394                                               QEMUIOVector *qiov,
2395                                               uint64_t qiov_offset,
2396                                               QCowL2Meta *l2meta)
2397 {
2398     int ret;
2399     BDRVQcow2State *s = bs->opaque;
2400     void *crypt_buf = NULL;
2401     int offset_in_cluster = offset_into_cluster(s, offset);
2402     QEMUIOVector encrypted_qiov;
2403 
2404     if (bs->encrypted) {
2405         assert(s->crypto);
2406         assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2407         crypt_buf = qemu_try_blockalign(bs->file->bs, bytes);
2408         if (crypt_buf == NULL) {
2409             ret = -ENOMEM;
2410             goto out_unlocked;
2411         }
2412         qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes);
2413 
2414         if (qcow2_co_encrypt(bs, file_cluster_offset + offset_in_cluster,
2415                              offset, crypt_buf, bytes) < 0)
2416         {
2417             ret = -EIO;
2418             goto out_unlocked;
2419         }
2420 
2421         qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes);
2422         qiov = &encrypted_qiov;
2423         qiov_offset = 0;
2424     }
2425 
2426     /* Try to efficiently initialize the physical space with zeroes */
2427     ret = handle_alloc_space(bs, l2meta);
2428     if (ret < 0) {
2429         goto out_unlocked;
2430     }
2431 
2432     /*
2433      * If we need to do COW, check if it's possible to merge the
2434      * writing of the guest data together with that of the COW regions.
2435      * If it's not possible (or not necessary) then write the
2436      * guest data now.
2437      */
2438     if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) {
2439         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
2440         trace_qcow2_writev_data(qemu_coroutine_self(),
2441                                 file_cluster_offset + offset_in_cluster);
2442         ret = bdrv_co_pwritev_part(s->data_file,
2443                                    file_cluster_offset + offset_in_cluster,
2444                                    bytes, qiov, qiov_offset, 0);
2445         if (ret < 0) {
2446             goto out_unlocked;
2447         }
2448     }
2449 
2450     qemu_co_mutex_lock(&s->lock);
2451 
2452     ret = qcow2_handle_l2meta(bs, &l2meta, true);
2453     goto out_locked;
2454 
2455 out_unlocked:
2456     qemu_co_mutex_lock(&s->lock);
2457 
2458 out_locked:
2459     qcow2_handle_l2meta(bs, &l2meta, false);
2460     qemu_co_mutex_unlock(&s->lock);
2461 
2462     qemu_vfree(crypt_buf);
2463 
2464     return ret;
2465 }
2466 
2467 static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task)
2468 {
2469     Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2470 
2471     assert(!t->cluster_type);
2472 
2473     return qcow2_co_pwritev_task(t->bs, t->file_cluster_offset,
2474                                  t->offset, t->bytes, t->qiov, t->qiov_offset,
2475                                  t->l2meta);
2476 }
2477 
2478 static coroutine_fn int qcow2_co_pwritev_part(
2479         BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2480         QEMUIOVector *qiov, size_t qiov_offset, int flags)
2481 {
2482     BDRVQcow2State *s = bs->opaque;
2483     int offset_in_cluster;
2484     int ret;
2485     unsigned int cur_bytes; /* number of sectors in current iteration */
2486     uint64_t cluster_offset;
2487     QCowL2Meta *l2meta = NULL;
2488     AioTaskPool *aio = NULL;
2489 
2490     trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
2491 
2492     while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2493 
2494         l2meta = NULL;
2495 
2496         trace_qcow2_writev_start_part(qemu_coroutine_self());
2497         offset_in_cluster = offset_into_cluster(s, offset);
2498         cur_bytes = MIN(bytes, INT_MAX);
2499         if (bs->encrypted) {
2500             cur_bytes = MIN(cur_bytes,
2501                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
2502                             - offset_in_cluster);
2503         }
2504 
2505         qemu_co_mutex_lock(&s->lock);
2506 
2507         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2508                                          &cluster_offset, &l2meta);
2509         if (ret < 0) {
2510             goto out_locked;
2511         }
2512 
2513         assert(offset_into_cluster(s, cluster_offset) == 0);
2514 
2515         ret = qcow2_pre_write_overlap_check(bs, 0,
2516                                             cluster_offset + offset_in_cluster,
2517                                             cur_bytes, true);
2518         if (ret < 0) {
2519             goto out_locked;
2520         }
2521 
2522         qemu_co_mutex_unlock(&s->lock);
2523 
2524         if (!aio && cur_bytes != bytes) {
2525             aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2526         }
2527         ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0,
2528                              cluster_offset, offset, cur_bytes,
2529                              qiov, qiov_offset, l2meta);
2530         l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */
2531         if (ret < 0) {
2532             goto fail_nometa;
2533         }
2534 
2535         bytes -= cur_bytes;
2536         offset += cur_bytes;
2537         qiov_offset += cur_bytes;
2538         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2539     }
2540     ret = 0;
2541 
2542     qemu_co_mutex_lock(&s->lock);
2543 
2544 out_locked:
2545     qcow2_handle_l2meta(bs, &l2meta, false);
2546 
2547     qemu_co_mutex_unlock(&s->lock);
2548 
2549 fail_nometa:
2550     if (aio) {
2551         aio_task_pool_wait_all(aio);
2552         if (ret == 0) {
2553             ret = aio_task_pool_status(aio);
2554         }
2555         g_free(aio);
2556     }
2557 
2558     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2559 
2560     return ret;
2561 }
2562 
2563 static int qcow2_inactivate(BlockDriverState *bs)
2564 {
2565     BDRVQcow2State *s = bs->opaque;
2566     int ret, result = 0;
2567     Error *local_err = NULL;
2568 
2569     qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err);
2570     if (local_err != NULL) {
2571         result = -EINVAL;
2572         error_reportf_err(local_err, "Lost persistent bitmaps during "
2573                           "inactivation of node '%s': ",
2574                           bdrv_get_device_or_node_name(bs));
2575     }
2576 
2577     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2578     if (ret) {
2579         result = ret;
2580         error_report("Failed to flush the L2 table cache: %s",
2581                      strerror(-ret));
2582     }
2583 
2584     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2585     if (ret) {
2586         result = ret;
2587         error_report("Failed to flush the refcount block cache: %s",
2588                      strerror(-ret));
2589     }
2590 
2591     if (result == 0) {
2592         qcow2_mark_clean(bs);
2593     }
2594 
2595     return result;
2596 }
2597 
2598 static void qcow2_close(BlockDriverState *bs)
2599 {
2600     BDRVQcow2State *s = bs->opaque;
2601     qemu_vfree(s->l1_table);
2602     /* else pre-write overlap checks in cache_destroy may crash */
2603     s->l1_table = NULL;
2604 
2605     if (!(s->flags & BDRV_O_INACTIVE)) {
2606         qcow2_inactivate(bs);
2607     }
2608 
2609     cache_clean_timer_del(bs);
2610     qcow2_cache_destroy(s->l2_table_cache);
2611     qcow2_cache_destroy(s->refcount_block_cache);
2612 
2613     qcrypto_block_free(s->crypto);
2614     s->crypto = NULL;
2615     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
2616 
2617     g_free(s->unknown_header_fields);
2618     cleanup_unknown_header_ext(bs);
2619 
2620     g_free(s->image_data_file);
2621     g_free(s->image_backing_file);
2622     g_free(s->image_backing_format);
2623 
2624     if (has_data_file(bs)) {
2625         bdrv_unref_child(bs, s->data_file);
2626         s->data_file = NULL;
2627     }
2628 
2629     qcow2_refcount_close(bs);
2630     qcow2_free_snapshots(bs);
2631 }
2632 
2633 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
2634                                                    Error **errp)
2635 {
2636     BDRVQcow2State *s = bs->opaque;
2637     int flags = s->flags;
2638     QCryptoBlock *crypto = NULL;
2639     QDict *options;
2640     Error *local_err = NULL;
2641     int ret;
2642 
2643     /*
2644      * Backing files are read-only which makes all of their metadata immutable,
2645      * that means we don't have to worry about reopening them here.
2646      */
2647 
2648     crypto = s->crypto;
2649     s->crypto = NULL;
2650 
2651     qcow2_close(bs);
2652 
2653     memset(s, 0, sizeof(BDRVQcow2State));
2654     options = qdict_clone_shallow(bs->options);
2655 
2656     flags &= ~BDRV_O_INACTIVE;
2657     qemu_co_mutex_lock(&s->lock);
2658     ret = qcow2_do_open(bs, options, flags, &local_err);
2659     qemu_co_mutex_unlock(&s->lock);
2660     qobject_unref(options);
2661     if (local_err) {
2662         error_propagate_prepend(errp, local_err,
2663                                 "Could not reopen qcow2 layer: ");
2664         bs->drv = NULL;
2665         return;
2666     } else if (ret < 0) {
2667         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
2668         bs->drv = NULL;
2669         return;
2670     }
2671 
2672     s->crypto = crypto;
2673 }
2674 
2675 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2676     size_t len, size_t buflen)
2677 {
2678     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2679     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2680 
2681     if (buflen < ext_len) {
2682         return -ENOSPC;
2683     }
2684 
2685     *ext_backing_fmt = (QCowExtension) {
2686         .magic  = cpu_to_be32(magic),
2687         .len    = cpu_to_be32(len),
2688     };
2689 
2690     if (len) {
2691         memcpy(buf + sizeof(QCowExtension), s, len);
2692     }
2693 
2694     return ext_len;
2695 }
2696 
2697 /*
2698  * Updates the qcow2 header, including the variable length parts of it, i.e.
2699  * the backing file name and all extensions. qcow2 was not designed to allow
2700  * such changes, so if we run out of space (we can only use the first cluster)
2701  * this function may fail.
2702  *
2703  * Returns 0 on success, -errno in error cases.
2704  */
2705 int qcow2_update_header(BlockDriverState *bs)
2706 {
2707     BDRVQcow2State *s = bs->opaque;
2708     QCowHeader *header;
2709     char *buf;
2710     size_t buflen = s->cluster_size;
2711     int ret;
2712     uint64_t total_size;
2713     uint32_t refcount_table_clusters;
2714     size_t header_length;
2715     Qcow2UnknownHeaderExtension *uext;
2716 
2717     buf = qemu_blockalign(bs, buflen);
2718 
2719     /* Header structure */
2720     header = (QCowHeader*) buf;
2721 
2722     if (buflen < sizeof(*header)) {
2723         ret = -ENOSPC;
2724         goto fail;
2725     }
2726 
2727     header_length = sizeof(*header) + s->unknown_header_fields_size;
2728     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2729     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2730 
2731     *header = (QCowHeader) {
2732         /* Version 2 fields */
2733         .magic                  = cpu_to_be32(QCOW_MAGIC),
2734         .version                = cpu_to_be32(s->qcow_version),
2735         .backing_file_offset    = 0,
2736         .backing_file_size      = 0,
2737         .cluster_bits           = cpu_to_be32(s->cluster_bits),
2738         .size                   = cpu_to_be64(total_size),
2739         .crypt_method           = cpu_to_be32(s->crypt_method_header),
2740         .l1_size                = cpu_to_be32(s->l1_size),
2741         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
2742         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
2743         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2744         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
2745         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
2746 
2747         /* Version 3 fields */
2748         .incompatible_features  = cpu_to_be64(s->incompatible_features),
2749         .compatible_features    = cpu_to_be64(s->compatible_features),
2750         .autoclear_features     = cpu_to_be64(s->autoclear_features),
2751         .refcount_order         = cpu_to_be32(s->refcount_order),
2752         .header_length          = cpu_to_be32(header_length),
2753     };
2754 
2755     /* For older versions, write a shorter header */
2756     switch (s->qcow_version) {
2757     case 2:
2758         ret = offsetof(QCowHeader, incompatible_features);
2759         break;
2760     case 3:
2761         ret = sizeof(*header);
2762         break;
2763     default:
2764         ret = -EINVAL;
2765         goto fail;
2766     }
2767 
2768     buf += ret;
2769     buflen -= ret;
2770     memset(buf, 0, buflen);
2771 
2772     /* Preserve any unknown field in the header */
2773     if (s->unknown_header_fields_size) {
2774         if (buflen < s->unknown_header_fields_size) {
2775             ret = -ENOSPC;
2776             goto fail;
2777         }
2778 
2779         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2780         buf += s->unknown_header_fields_size;
2781         buflen -= s->unknown_header_fields_size;
2782     }
2783 
2784     /* Backing file format header extension */
2785     if (s->image_backing_format) {
2786         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
2787                              s->image_backing_format,
2788                              strlen(s->image_backing_format),
2789                              buflen);
2790         if (ret < 0) {
2791             goto fail;
2792         }
2793 
2794         buf += ret;
2795         buflen -= ret;
2796     }
2797 
2798     /* External data file header extension */
2799     if (has_data_file(bs) && s->image_data_file) {
2800         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE,
2801                              s->image_data_file, strlen(s->image_data_file),
2802                              buflen);
2803         if (ret < 0) {
2804             goto fail;
2805         }
2806 
2807         buf += ret;
2808         buflen -= ret;
2809     }
2810 
2811     /* Full disk encryption header pointer extension */
2812     if (s->crypto_header.offset != 0) {
2813         s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset);
2814         s->crypto_header.length = cpu_to_be64(s->crypto_header.length);
2815         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
2816                              &s->crypto_header, sizeof(s->crypto_header),
2817                              buflen);
2818         s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
2819         s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
2820         if (ret < 0) {
2821             goto fail;
2822         }
2823         buf += ret;
2824         buflen -= ret;
2825     }
2826 
2827     /*
2828      * Feature table.  A mere 8 feature names occupies 392 bytes, and
2829      * when coupled with the v3 minimum header of 104 bytes plus the
2830      * 8-byte end-of-extension marker, that would leave only 8 bytes
2831      * for a backing file name in an image with 512-byte clusters.
2832      * Thus, we choose to omit this header for cluster sizes 4k and
2833      * smaller.
2834      */
2835     if (s->qcow_version >= 3 && s->cluster_size > 4096) {
2836         static const Qcow2Feature features[] = {
2837             {
2838                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2839                 .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
2840                 .name = "dirty bit",
2841             },
2842             {
2843                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2844                 .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
2845                 .name = "corrupt bit",
2846             },
2847             {
2848                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2849                 .bit  = QCOW2_INCOMPAT_DATA_FILE_BITNR,
2850                 .name = "external data file",
2851             },
2852             {
2853                 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
2854                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
2855                 .name = "lazy refcounts",
2856             },
2857             {
2858                 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
2859                 .bit  = QCOW2_AUTOCLEAR_BITMAPS_BITNR,
2860                 .name = "bitmaps",
2861             },
2862             {
2863                 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
2864                 .bit  = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR,
2865                 .name = "raw external data",
2866             },
2867         };
2868 
2869         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
2870                              features, sizeof(features), buflen);
2871         if (ret < 0) {
2872             goto fail;
2873         }
2874         buf += ret;
2875         buflen -= ret;
2876     }
2877 
2878     /* Bitmap extension */
2879     if (s->nb_bitmaps > 0) {
2880         Qcow2BitmapHeaderExt bitmaps_header = {
2881             .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
2882             .bitmap_directory_size =
2883                     cpu_to_be64(s->bitmap_directory_size),
2884             .bitmap_directory_offset =
2885                     cpu_to_be64(s->bitmap_directory_offset)
2886         };
2887         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
2888                              &bitmaps_header, sizeof(bitmaps_header),
2889                              buflen);
2890         if (ret < 0) {
2891             goto fail;
2892         }
2893         buf += ret;
2894         buflen -= ret;
2895     }
2896 
2897     /* Keep unknown header extensions */
2898     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
2899         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
2900         if (ret < 0) {
2901             goto fail;
2902         }
2903 
2904         buf += ret;
2905         buflen -= ret;
2906     }
2907 
2908     /* End of header extensions */
2909     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
2910     if (ret < 0) {
2911         goto fail;
2912     }
2913 
2914     buf += ret;
2915     buflen -= ret;
2916 
2917     /* Backing file name */
2918     if (s->image_backing_file) {
2919         size_t backing_file_len = strlen(s->image_backing_file);
2920 
2921         if (buflen < backing_file_len) {
2922             ret = -ENOSPC;
2923             goto fail;
2924         }
2925 
2926         /* Using strncpy is ok here, since buf is not NUL-terminated. */
2927         strncpy(buf, s->image_backing_file, buflen);
2928 
2929         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
2930         header->backing_file_size   = cpu_to_be32(backing_file_len);
2931     }
2932 
2933     /* Write the new header */
2934     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
2935     if (ret < 0) {
2936         goto fail;
2937     }
2938 
2939     ret = 0;
2940 fail:
2941     qemu_vfree(header);
2942     return ret;
2943 }
2944 
2945 static int qcow2_change_backing_file(BlockDriverState *bs,
2946     const char *backing_file, const char *backing_fmt)
2947 {
2948     BDRVQcow2State *s = bs->opaque;
2949 
2950     /* Adding a backing file means that the external data file alone won't be
2951      * enough to make sense of the content */
2952     if (backing_file && data_file_is_raw(bs)) {
2953         return -EINVAL;
2954     }
2955 
2956     if (backing_file && strlen(backing_file) > 1023) {
2957         return -EINVAL;
2958     }
2959 
2960     pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2961             backing_file ?: "");
2962     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2963     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2964 
2965     g_free(s->image_backing_file);
2966     g_free(s->image_backing_format);
2967 
2968     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
2969     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
2970 
2971     return qcow2_update_header(bs);
2972 }
2973 
2974 static int qcow2_crypt_method_from_format(const char *encryptfmt)
2975 {
2976     if (g_str_equal(encryptfmt, "luks")) {
2977         return QCOW_CRYPT_LUKS;
2978     } else if (g_str_equal(encryptfmt, "aes")) {
2979         return QCOW_CRYPT_AES;
2980     } else {
2981         return -EINVAL;
2982     }
2983 }
2984 
2985 static int qcow2_set_up_encryption(BlockDriverState *bs,
2986                                    QCryptoBlockCreateOptions *cryptoopts,
2987                                    Error **errp)
2988 {
2989     BDRVQcow2State *s = bs->opaque;
2990     QCryptoBlock *crypto = NULL;
2991     int fmt, ret;
2992 
2993     switch (cryptoopts->format) {
2994     case Q_CRYPTO_BLOCK_FORMAT_LUKS:
2995         fmt = QCOW_CRYPT_LUKS;
2996         break;
2997     case Q_CRYPTO_BLOCK_FORMAT_QCOW:
2998         fmt = QCOW_CRYPT_AES;
2999         break;
3000     default:
3001         error_setg(errp, "Crypto format not supported in qcow2");
3002         return -EINVAL;
3003     }
3004 
3005     s->crypt_method_header = fmt;
3006 
3007     crypto = qcrypto_block_create(cryptoopts, "encrypt.",
3008                                   qcow2_crypto_hdr_init_func,
3009                                   qcow2_crypto_hdr_write_func,
3010                                   bs, errp);
3011     if (!crypto) {
3012         return -EINVAL;
3013     }
3014 
3015     ret = qcow2_update_header(bs);
3016     if (ret < 0) {
3017         error_setg_errno(errp, -ret, "Could not write encryption header");
3018         goto out;
3019     }
3020 
3021     ret = 0;
3022  out:
3023     qcrypto_block_free(crypto);
3024     return ret;
3025 }
3026 
3027 /**
3028  * Preallocates metadata structures for data clusters between @offset (in the
3029  * guest disk) and @new_length (which is thus generally the new guest disk
3030  * size).
3031  *
3032  * Returns: 0 on success, -errno on failure.
3033  */
3034 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset,
3035                                        uint64_t new_length, PreallocMode mode,
3036                                        Error **errp)
3037 {
3038     BDRVQcow2State *s = bs->opaque;
3039     uint64_t bytes;
3040     uint64_t host_offset = 0;
3041     int64_t file_length;
3042     unsigned int cur_bytes;
3043     int ret;
3044     QCowL2Meta *meta;
3045 
3046     assert(offset <= new_length);
3047     bytes = new_length - offset;
3048 
3049     while (bytes) {
3050         cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size));
3051         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
3052                                          &host_offset, &meta);
3053         if (ret < 0) {
3054             error_setg_errno(errp, -ret, "Allocating clusters failed");
3055             return ret;
3056         }
3057 
3058         while (meta) {
3059             QCowL2Meta *next = meta->next;
3060 
3061             ret = qcow2_alloc_cluster_link_l2(bs, meta);
3062             if (ret < 0) {
3063                 error_setg_errno(errp, -ret, "Mapping clusters failed");
3064                 qcow2_free_any_clusters(bs, meta->alloc_offset,
3065                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
3066                 return ret;
3067             }
3068 
3069             /* There are no dependent requests, but we need to remove our
3070              * request from the list of in-flight requests */
3071             QLIST_REMOVE(meta, next_in_flight);
3072 
3073             g_free(meta);
3074             meta = next;
3075         }
3076 
3077         /* TODO Preallocate data if requested */
3078 
3079         bytes -= cur_bytes;
3080         offset += cur_bytes;
3081     }
3082 
3083     /*
3084      * It is expected that the image file is large enough to actually contain
3085      * all of the allocated clusters (otherwise we get failing reads after
3086      * EOF). Extend the image to the last allocated sector.
3087      */
3088     file_length = bdrv_getlength(s->data_file->bs);
3089     if (file_length < 0) {
3090         error_setg_errno(errp, -file_length, "Could not get file size");
3091         return file_length;
3092     }
3093 
3094     if (host_offset + cur_bytes > file_length) {
3095         if (mode == PREALLOC_MODE_METADATA) {
3096             mode = PREALLOC_MODE_OFF;
3097         }
3098         ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false,
3099                                mode, 0, errp);
3100         if (ret < 0) {
3101             return ret;
3102         }
3103     }
3104 
3105     return 0;
3106 }
3107 
3108 /* qcow2_refcount_metadata_size:
3109  * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3110  * @cluster_size: size of a cluster, in bytes
3111  * @refcount_order: refcount bits power-of-2 exponent
3112  * @generous_increase: allow for the refcount table to be 1.5x as large as it
3113  *                     needs to be
3114  *
3115  * Returns: Number of bytes required for refcount blocks and table metadata.
3116  */
3117 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
3118                                      int refcount_order, bool generous_increase,
3119                                      uint64_t *refblock_count)
3120 {
3121     /*
3122      * Every host cluster is reference-counted, including metadata (even
3123      * refcount metadata is recursively included).
3124      *
3125      * An accurate formula for the size of refcount metadata size is difficult
3126      * to derive.  An easier method of calculation is finding the fixed point
3127      * where no further refcount blocks or table clusters are required to
3128      * reference count every cluster.
3129      */
3130     int64_t blocks_per_table_cluster = cluster_size / sizeof(uint64_t);
3131     int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
3132     int64_t table = 0;  /* number of refcount table clusters */
3133     int64_t blocks = 0; /* number of refcount block clusters */
3134     int64_t last;
3135     int64_t n = 0;
3136 
3137     do {
3138         last = n;
3139         blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
3140         table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
3141         n = clusters + blocks + table;
3142 
3143         if (n == last && generous_increase) {
3144             clusters += DIV_ROUND_UP(table, 2);
3145             n = 0; /* force another loop */
3146             generous_increase = false;
3147         }
3148     } while (n != last);
3149 
3150     if (refblock_count) {
3151         *refblock_count = blocks;
3152     }
3153 
3154     return (blocks + table) * cluster_size;
3155 }
3156 
3157 /**
3158  * qcow2_calc_prealloc_size:
3159  * @total_size: virtual disk size in bytes
3160  * @cluster_size: cluster size in bytes
3161  * @refcount_order: refcount bits power-of-2 exponent
3162  *
3163  * Returns: Total number of bytes required for the fully allocated image
3164  * (including metadata).
3165  */
3166 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
3167                                         size_t cluster_size,
3168                                         int refcount_order)
3169 {
3170     int64_t meta_size = 0;
3171     uint64_t nl1e, nl2e;
3172     int64_t aligned_total_size = ROUND_UP(total_size, cluster_size);
3173 
3174     /* header: 1 cluster */
3175     meta_size += cluster_size;
3176 
3177     /* total size of L2 tables */
3178     nl2e = aligned_total_size / cluster_size;
3179     nl2e = ROUND_UP(nl2e, cluster_size / sizeof(uint64_t));
3180     meta_size += nl2e * sizeof(uint64_t);
3181 
3182     /* total size of L1 tables */
3183     nl1e = nl2e * sizeof(uint64_t) / cluster_size;
3184     nl1e = ROUND_UP(nl1e, cluster_size / sizeof(uint64_t));
3185     meta_size += nl1e * sizeof(uint64_t);
3186 
3187     /* total size of refcount table and blocks */
3188     meta_size += qcow2_refcount_metadata_size(
3189             (meta_size + aligned_total_size) / cluster_size,
3190             cluster_size, refcount_order, false, NULL);
3191 
3192     return meta_size + aligned_total_size;
3193 }
3194 
3195 static bool validate_cluster_size(size_t cluster_size, Error **errp)
3196 {
3197     int cluster_bits = ctz32(cluster_size);
3198     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
3199         (1 << cluster_bits) != cluster_size)
3200     {
3201         error_setg(errp, "Cluster size must be a power of two between %d and "
3202                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
3203         return false;
3204     }
3205     return true;
3206 }
3207 
3208 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, Error **errp)
3209 {
3210     size_t cluster_size;
3211 
3212     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
3213                                          DEFAULT_CLUSTER_SIZE);
3214     if (!validate_cluster_size(cluster_size, errp)) {
3215         return 0;
3216     }
3217     return cluster_size;
3218 }
3219 
3220 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
3221 {
3222     char *buf;
3223     int ret;
3224 
3225     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
3226     if (!buf) {
3227         ret = 3; /* default */
3228     } else if (!strcmp(buf, "0.10")) {
3229         ret = 2;
3230     } else if (!strcmp(buf, "1.1")) {
3231         ret = 3;
3232     } else {
3233         error_setg(errp, "Invalid compatibility level: '%s'", buf);
3234         ret = -EINVAL;
3235     }
3236     g_free(buf);
3237     return ret;
3238 }
3239 
3240 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
3241                                                 Error **errp)
3242 {
3243     uint64_t refcount_bits;
3244 
3245     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
3246     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
3247         error_setg(errp, "Refcount width must be a power of two and may not "
3248                    "exceed 64 bits");
3249         return 0;
3250     }
3251 
3252     if (version < 3 && refcount_bits != 16) {
3253         error_setg(errp, "Different refcount widths than 16 bits require "
3254                    "compatibility level 1.1 or above (use compat=1.1 or "
3255                    "greater)");
3256         return 0;
3257     }
3258 
3259     return refcount_bits;
3260 }
3261 
3262 static int coroutine_fn
3263 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
3264 {
3265     BlockdevCreateOptionsQcow2 *qcow2_opts;
3266     QDict *options;
3267 
3268     /*
3269      * Open the image file and write a minimal qcow2 header.
3270      *
3271      * We keep things simple and start with a zero-sized image. We also
3272      * do without refcount blocks or a L1 table for now. We'll fix the
3273      * inconsistency later.
3274      *
3275      * We do need a refcount table because growing the refcount table means
3276      * allocating two new refcount blocks - the second of which would be at
3277      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3278      * size for any qcow2 image.
3279      */
3280     BlockBackend *blk = NULL;
3281     BlockDriverState *bs = NULL;
3282     BlockDriverState *data_bs = NULL;
3283     QCowHeader *header;
3284     size_t cluster_size;
3285     int version;
3286     int refcount_order;
3287     uint64_t* refcount_table;
3288     Error *local_err = NULL;
3289     int ret;
3290 
3291     assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2);
3292     qcow2_opts = &create_options->u.qcow2;
3293 
3294     bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp);
3295     if (bs == NULL) {
3296         return -EIO;
3297     }
3298 
3299     /* Validate options and set default values */
3300     if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) {
3301         error_setg(errp, "Image size must be a multiple of %u bytes",
3302                    (unsigned) BDRV_SECTOR_SIZE);
3303         ret = -EINVAL;
3304         goto out;
3305     }
3306 
3307     if (qcow2_opts->has_version) {
3308         switch (qcow2_opts->version) {
3309         case BLOCKDEV_QCOW2_VERSION_V2:
3310             version = 2;
3311             break;
3312         case BLOCKDEV_QCOW2_VERSION_V3:
3313             version = 3;
3314             break;
3315         default:
3316             g_assert_not_reached();
3317         }
3318     } else {
3319         version = 3;
3320     }
3321 
3322     if (qcow2_opts->has_cluster_size) {
3323         cluster_size = qcow2_opts->cluster_size;
3324     } else {
3325         cluster_size = DEFAULT_CLUSTER_SIZE;
3326     }
3327 
3328     if (!validate_cluster_size(cluster_size, errp)) {
3329         ret = -EINVAL;
3330         goto out;
3331     }
3332 
3333     if (!qcow2_opts->has_preallocation) {
3334         qcow2_opts->preallocation = PREALLOC_MODE_OFF;
3335     }
3336     if (qcow2_opts->has_backing_file &&
3337         qcow2_opts->preallocation != PREALLOC_MODE_OFF)
3338     {
3339         error_setg(errp, "Backing file and preallocation cannot be used at "
3340                    "the same time");
3341         ret = -EINVAL;
3342         goto out;
3343     }
3344     if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) {
3345         error_setg(errp, "Backing format cannot be used without backing file");
3346         ret = -EINVAL;
3347         goto out;
3348     }
3349 
3350     if (!qcow2_opts->has_lazy_refcounts) {
3351         qcow2_opts->lazy_refcounts = false;
3352     }
3353     if (version < 3 && qcow2_opts->lazy_refcounts) {
3354         error_setg(errp, "Lazy refcounts only supported with compatibility "
3355                    "level 1.1 and above (use version=v3 or greater)");
3356         ret = -EINVAL;
3357         goto out;
3358     }
3359 
3360     if (!qcow2_opts->has_refcount_bits) {
3361         qcow2_opts->refcount_bits = 16;
3362     }
3363     if (qcow2_opts->refcount_bits > 64 ||
3364         !is_power_of_2(qcow2_opts->refcount_bits))
3365     {
3366         error_setg(errp, "Refcount width must be a power of two and may not "
3367                    "exceed 64 bits");
3368         ret = -EINVAL;
3369         goto out;
3370     }
3371     if (version < 3 && qcow2_opts->refcount_bits != 16) {
3372         error_setg(errp, "Different refcount widths than 16 bits require "
3373                    "compatibility level 1.1 or above (use version=v3 or "
3374                    "greater)");
3375         ret = -EINVAL;
3376         goto out;
3377     }
3378     refcount_order = ctz32(qcow2_opts->refcount_bits);
3379 
3380     if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) {
3381         error_setg(errp, "data-file-raw requires data-file");
3382         ret = -EINVAL;
3383         goto out;
3384     }
3385     if (qcow2_opts->data_file_raw && qcow2_opts->has_backing_file) {
3386         error_setg(errp, "Backing file and data-file-raw cannot be used at "
3387                    "the same time");
3388         ret = -EINVAL;
3389         goto out;
3390     }
3391 
3392     if (qcow2_opts->data_file) {
3393         if (version < 3) {
3394             error_setg(errp, "External data files are only supported with "
3395                        "compatibility level 1.1 and above (use version=v3 or "
3396                        "greater)");
3397             ret = -EINVAL;
3398             goto out;
3399         }
3400         data_bs = bdrv_open_blockdev_ref(qcow2_opts->data_file, errp);
3401         if (data_bs == NULL) {
3402             ret = -EIO;
3403             goto out;
3404         }
3405     }
3406 
3407     /* Create BlockBackend to write to the image */
3408     blk = blk_new(bdrv_get_aio_context(bs),
3409                   BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
3410     ret = blk_insert_bs(blk, bs, errp);
3411     if (ret < 0) {
3412         goto out;
3413     }
3414     blk_set_allow_write_beyond_eof(blk, true);
3415 
3416     /* Write the header */
3417     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
3418     header = g_malloc0(cluster_size);
3419     *header = (QCowHeader) {
3420         .magic                      = cpu_to_be32(QCOW_MAGIC),
3421         .version                    = cpu_to_be32(version),
3422         .cluster_bits               = cpu_to_be32(ctz32(cluster_size)),
3423         .size                       = cpu_to_be64(0),
3424         .l1_table_offset            = cpu_to_be64(0),
3425         .l1_size                    = cpu_to_be32(0),
3426         .refcount_table_offset      = cpu_to_be64(cluster_size),
3427         .refcount_table_clusters    = cpu_to_be32(1),
3428         .refcount_order             = cpu_to_be32(refcount_order),
3429         .header_length              = cpu_to_be32(sizeof(*header)),
3430     };
3431 
3432     /* We'll update this to correct value later */
3433     header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
3434 
3435     if (qcow2_opts->lazy_refcounts) {
3436         header->compatible_features |=
3437             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
3438     }
3439     if (data_bs) {
3440         header->incompatible_features |=
3441             cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE);
3442     }
3443     if (qcow2_opts->data_file_raw) {
3444         header->autoclear_features |=
3445             cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW);
3446     }
3447 
3448     ret = blk_pwrite(blk, 0, header, cluster_size, 0);
3449     g_free(header);
3450     if (ret < 0) {
3451         error_setg_errno(errp, -ret, "Could not write qcow2 header");
3452         goto out;
3453     }
3454 
3455     /* Write a refcount table with one refcount block */
3456     refcount_table = g_malloc0(2 * cluster_size);
3457     refcount_table[0] = cpu_to_be64(2 * cluster_size);
3458     ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
3459     g_free(refcount_table);
3460 
3461     if (ret < 0) {
3462         error_setg_errno(errp, -ret, "Could not write refcount table");
3463         goto out;
3464     }
3465 
3466     blk_unref(blk);
3467     blk = NULL;
3468 
3469     /*
3470      * And now open the image and make it consistent first (i.e. increase the
3471      * refcount of the cluster that is occupied by the header and the refcount
3472      * table)
3473      */
3474     options = qdict_new();
3475     qdict_put_str(options, "driver", "qcow2");
3476     qdict_put_str(options, "file", bs->node_name);
3477     if (data_bs) {
3478         qdict_put_str(options, "data-file", data_bs->node_name);
3479     }
3480     blk = blk_new_open(NULL, NULL, options,
3481                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
3482                        &local_err);
3483     if (blk == NULL) {
3484         error_propagate(errp, local_err);
3485         ret = -EIO;
3486         goto out;
3487     }
3488 
3489     ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
3490     if (ret < 0) {
3491         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
3492                          "header and refcount table");
3493         goto out;
3494 
3495     } else if (ret != 0) {
3496         error_report("Huh, first cluster in empty image is already in use?");
3497         abort();
3498     }
3499 
3500     /* Set the external data file if necessary */
3501     if (data_bs) {
3502         BDRVQcow2State *s = blk_bs(blk)->opaque;
3503         s->image_data_file = g_strdup(data_bs->filename);
3504     }
3505 
3506     /* Create a full header (including things like feature table) */
3507     ret = qcow2_update_header(blk_bs(blk));
3508     if (ret < 0) {
3509         error_setg_errno(errp, -ret, "Could not update qcow2 header");
3510         goto out;
3511     }
3512 
3513     /* Okay, now that we have a valid image, let's give it the right size */
3514     ret = blk_truncate(blk, qcow2_opts->size, false, qcow2_opts->preallocation,
3515                        0, errp);
3516     if (ret < 0) {
3517         error_prepend(errp, "Could not resize image: ");
3518         goto out;
3519     }
3520 
3521     /* Want a backing file? There you go. */
3522     if (qcow2_opts->has_backing_file) {
3523         const char *backing_format = NULL;
3524 
3525         if (qcow2_opts->has_backing_fmt) {
3526             backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt);
3527         }
3528 
3529         ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file,
3530                                        backing_format);
3531         if (ret < 0) {
3532             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
3533                              "with format '%s'", qcow2_opts->backing_file,
3534                              backing_format);
3535             goto out;
3536         }
3537     }
3538 
3539     /* Want encryption? There you go. */
3540     if (qcow2_opts->has_encrypt) {
3541         ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp);
3542         if (ret < 0) {
3543             goto out;
3544         }
3545     }
3546 
3547     blk_unref(blk);
3548     blk = NULL;
3549 
3550     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3551      * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3552      * have to setup decryption context. We're not doing any I/O on the top
3553      * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3554      * not have effect.
3555      */
3556     options = qdict_new();
3557     qdict_put_str(options, "driver", "qcow2");
3558     qdict_put_str(options, "file", bs->node_name);
3559     if (data_bs) {
3560         qdict_put_str(options, "data-file", data_bs->node_name);
3561     }
3562     blk = blk_new_open(NULL, NULL, options,
3563                        BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO,
3564                        &local_err);
3565     if (blk == NULL) {
3566         error_propagate(errp, local_err);
3567         ret = -EIO;
3568         goto out;
3569     }
3570 
3571     ret = 0;
3572 out:
3573     blk_unref(blk);
3574     bdrv_unref(bs);
3575     bdrv_unref(data_bs);
3576     return ret;
3577 }
3578 
3579 static int coroutine_fn qcow2_co_create_opts(BlockDriver *drv,
3580                                              const char *filename,
3581                                              QemuOpts *opts,
3582                                              Error **errp)
3583 {
3584     BlockdevCreateOptions *create_options = NULL;
3585     QDict *qdict;
3586     Visitor *v;
3587     BlockDriverState *bs = NULL;
3588     BlockDriverState *data_bs = NULL;
3589     Error *local_err = NULL;
3590     const char *val;
3591     int ret;
3592 
3593     /* Only the keyval visitor supports the dotted syntax needed for
3594      * encryption, so go through a QDict before getting a QAPI type. Ignore
3595      * options meant for the protocol layer so that the visitor doesn't
3596      * complain. */
3597     qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts,
3598                                         true);
3599 
3600     /* Handle encryption options */
3601     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
3602     if (val && !strcmp(val, "on")) {
3603         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
3604     } else if (val && !strcmp(val, "off")) {
3605         qdict_del(qdict, BLOCK_OPT_ENCRYPT);
3606     }
3607 
3608     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
3609     if (val && !strcmp(val, "aes")) {
3610         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
3611     }
3612 
3613     /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3614      * version=v2/v3 below. */
3615     val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL);
3616     if (val && !strcmp(val, "0.10")) {
3617         qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2");
3618     } else if (val && !strcmp(val, "1.1")) {
3619         qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3");
3620     }
3621 
3622     /* Change legacy command line options into QMP ones */
3623     static const QDictRenames opt_renames[] = {
3624         { BLOCK_OPT_BACKING_FILE,       "backing-file" },
3625         { BLOCK_OPT_BACKING_FMT,        "backing-fmt" },
3626         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
3627         { BLOCK_OPT_LAZY_REFCOUNTS,     "lazy-refcounts" },
3628         { BLOCK_OPT_REFCOUNT_BITS,      "refcount-bits" },
3629         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
3630         { BLOCK_OPT_COMPAT_LEVEL,       "version" },
3631         { BLOCK_OPT_DATA_FILE_RAW,      "data-file-raw" },
3632         { NULL, NULL },
3633     };
3634 
3635     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
3636         ret = -EINVAL;
3637         goto finish;
3638     }
3639 
3640     /* Create and open the file (protocol layer) */
3641     ret = bdrv_create_file(filename, opts, errp);
3642     if (ret < 0) {
3643         goto finish;
3644     }
3645 
3646     bs = bdrv_open(filename, NULL, NULL,
3647                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
3648     if (bs == NULL) {
3649         ret = -EIO;
3650         goto finish;
3651     }
3652 
3653     /* Create and open an external data file (protocol layer) */
3654     val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE);
3655     if (val) {
3656         ret = bdrv_create_file(val, opts, errp);
3657         if (ret < 0) {
3658             goto finish;
3659         }
3660 
3661         data_bs = bdrv_open(val, NULL, NULL,
3662                             BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
3663                             errp);
3664         if (data_bs == NULL) {
3665             ret = -EIO;
3666             goto finish;
3667         }
3668 
3669         qdict_del(qdict, BLOCK_OPT_DATA_FILE);
3670         qdict_put_str(qdict, "data-file", data_bs->node_name);
3671     }
3672 
3673     /* Set 'driver' and 'node' options */
3674     qdict_put_str(qdict, "driver", "qcow2");
3675     qdict_put_str(qdict, "file", bs->node_name);
3676 
3677     /* Now get the QAPI type BlockdevCreateOptions */
3678     v = qobject_input_visitor_new_flat_confused(qdict, errp);
3679     if (!v) {
3680         ret = -EINVAL;
3681         goto finish;
3682     }
3683 
3684     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
3685     visit_free(v);
3686 
3687     if (local_err) {
3688         error_propagate(errp, local_err);
3689         ret = -EINVAL;
3690         goto finish;
3691     }
3692 
3693     /* Silently round up size */
3694     create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size,
3695                                             BDRV_SECTOR_SIZE);
3696 
3697     /* Create the qcow2 image (format layer) */
3698     ret = qcow2_co_create(create_options, errp);
3699     if (ret < 0) {
3700         goto finish;
3701     }
3702 
3703     ret = 0;
3704 finish:
3705     qobject_unref(qdict);
3706     bdrv_unref(bs);
3707     bdrv_unref(data_bs);
3708     qapi_free_BlockdevCreateOptions(create_options);
3709     return ret;
3710 }
3711 
3712 
3713 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes)
3714 {
3715     int64_t nr;
3716     int res;
3717 
3718     /* Clamp to image length, before checking status of underlying sectors */
3719     if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
3720         bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset;
3721     }
3722 
3723     if (!bytes) {
3724         return true;
3725     }
3726     res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL);
3727     return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == bytes;
3728 }
3729 
3730 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
3731     int64_t offset, int bytes, BdrvRequestFlags flags)
3732 {
3733     int ret;
3734     BDRVQcow2State *s = bs->opaque;
3735 
3736     uint32_t head = offset % s->cluster_size;
3737     uint32_t tail = (offset + bytes) % s->cluster_size;
3738 
3739     trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes);
3740     if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) {
3741         tail = 0;
3742     }
3743 
3744     if (head || tail) {
3745         uint64_t off;
3746         unsigned int nr;
3747 
3748         assert(head + bytes <= s->cluster_size);
3749 
3750         /* check whether remainder of cluster already reads as zero */
3751         if (!(is_zero(bs, offset - head, head) &&
3752               is_zero(bs, offset + bytes,
3753                       tail ? s->cluster_size - tail : 0))) {
3754             return -ENOTSUP;
3755         }
3756 
3757         qemu_co_mutex_lock(&s->lock);
3758         /* We can have new write after previous check */
3759         offset = QEMU_ALIGN_DOWN(offset, s->cluster_size);
3760         bytes = s->cluster_size;
3761         nr = s->cluster_size;
3762         ret = qcow2_get_cluster_offset(bs, offset, &nr, &off);
3763         if (ret != QCOW2_CLUSTER_UNALLOCATED &&
3764             ret != QCOW2_CLUSTER_ZERO_PLAIN &&
3765             ret != QCOW2_CLUSTER_ZERO_ALLOC) {
3766             qemu_co_mutex_unlock(&s->lock);
3767             return -ENOTSUP;
3768         }
3769     } else {
3770         qemu_co_mutex_lock(&s->lock);
3771     }
3772 
3773     trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes);
3774 
3775     /* Whatever is left can use real zero clusters */
3776     ret = qcow2_cluster_zeroize(bs, offset, bytes, flags);
3777     qemu_co_mutex_unlock(&s->lock);
3778 
3779     return ret;
3780 }
3781 
3782 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
3783                                           int64_t offset, int bytes)
3784 {
3785     int ret;
3786     BDRVQcow2State *s = bs->opaque;
3787 
3788     /* If the image does not support QCOW_OFLAG_ZERO then discarding
3789      * clusters could expose stale data from the backing file. */
3790     if (s->qcow_version < 3 && bs->backing) {
3791         return -ENOTSUP;
3792     }
3793 
3794     if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
3795         assert(bytes < s->cluster_size);
3796         /* Ignore partial clusters, except for the special case of the
3797          * complete partial cluster at the end of an unaligned file */
3798         if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
3799             offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) {
3800             return -ENOTSUP;
3801         }
3802     }
3803 
3804     qemu_co_mutex_lock(&s->lock);
3805     ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST,
3806                                 false);
3807     qemu_co_mutex_unlock(&s->lock);
3808     return ret;
3809 }
3810 
3811 static int coroutine_fn
3812 qcow2_co_copy_range_from(BlockDriverState *bs,
3813                          BdrvChild *src, uint64_t src_offset,
3814                          BdrvChild *dst, uint64_t dst_offset,
3815                          uint64_t bytes, BdrvRequestFlags read_flags,
3816                          BdrvRequestFlags write_flags)
3817 {
3818     BDRVQcow2State *s = bs->opaque;
3819     int ret;
3820     unsigned int cur_bytes; /* number of bytes in current iteration */
3821     BdrvChild *child = NULL;
3822     BdrvRequestFlags cur_write_flags;
3823 
3824     assert(!bs->encrypted);
3825     qemu_co_mutex_lock(&s->lock);
3826 
3827     while (bytes != 0) {
3828         uint64_t copy_offset = 0;
3829         /* prepare next request */
3830         cur_bytes = MIN(bytes, INT_MAX);
3831         cur_write_flags = write_flags;
3832 
3833         ret = qcow2_get_cluster_offset(bs, src_offset, &cur_bytes, &copy_offset);
3834         if (ret < 0) {
3835             goto out;
3836         }
3837 
3838         switch (ret) {
3839         case QCOW2_CLUSTER_UNALLOCATED:
3840             if (bs->backing && bs->backing->bs) {
3841                 int64_t backing_length = bdrv_getlength(bs->backing->bs);
3842                 if (src_offset >= backing_length) {
3843                     cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3844                 } else {
3845                     child = bs->backing;
3846                     cur_bytes = MIN(cur_bytes, backing_length - src_offset);
3847                     copy_offset = src_offset;
3848                 }
3849             } else {
3850                 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3851             }
3852             break;
3853 
3854         case QCOW2_CLUSTER_ZERO_PLAIN:
3855         case QCOW2_CLUSTER_ZERO_ALLOC:
3856             cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3857             break;
3858 
3859         case QCOW2_CLUSTER_COMPRESSED:
3860             ret = -ENOTSUP;
3861             goto out;
3862 
3863         case QCOW2_CLUSTER_NORMAL:
3864             child = s->data_file;
3865             copy_offset += offset_into_cluster(s, src_offset);
3866             break;
3867 
3868         default:
3869             abort();
3870         }
3871         qemu_co_mutex_unlock(&s->lock);
3872         ret = bdrv_co_copy_range_from(child,
3873                                       copy_offset,
3874                                       dst, dst_offset,
3875                                       cur_bytes, read_flags, cur_write_flags);
3876         qemu_co_mutex_lock(&s->lock);
3877         if (ret < 0) {
3878             goto out;
3879         }
3880 
3881         bytes -= cur_bytes;
3882         src_offset += cur_bytes;
3883         dst_offset += cur_bytes;
3884     }
3885     ret = 0;
3886 
3887 out:
3888     qemu_co_mutex_unlock(&s->lock);
3889     return ret;
3890 }
3891 
3892 static int coroutine_fn
3893 qcow2_co_copy_range_to(BlockDriverState *bs,
3894                        BdrvChild *src, uint64_t src_offset,
3895                        BdrvChild *dst, uint64_t dst_offset,
3896                        uint64_t bytes, BdrvRequestFlags read_flags,
3897                        BdrvRequestFlags write_flags)
3898 {
3899     BDRVQcow2State *s = bs->opaque;
3900     int offset_in_cluster;
3901     int ret;
3902     unsigned int cur_bytes; /* number of sectors in current iteration */
3903     uint64_t cluster_offset;
3904     QCowL2Meta *l2meta = NULL;
3905 
3906     assert(!bs->encrypted);
3907 
3908     qemu_co_mutex_lock(&s->lock);
3909 
3910     while (bytes != 0) {
3911 
3912         l2meta = NULL;
3913 
3914         offset_in_cluster = offset_into_cluster(s, dst_offset);
3915         cur_bytes = MIN(bytes, INT_MAX);
3916 
3917         /* TODO:
3918          * If src->bs == dst->bs, we could simply copy by incrementing
3919          * the refcnt, without copying user data.
3920          * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
3921         ret = qcow2_alloc_cluster_offset(bs, dst_offset, &cur_bytes,
3922                                          &cluster_offset, &l2meta);
3923         if (ret < 0) {
3924             goto fail;
3925         }
3926 
3927         assert(offset_into_cluster(s, cluster_offset) == 0);
3928 
3929         ret = qcow2_pre_write_overlap_check(bs, 0,
3930                 cluster_offset + offset_in_cluster, cur_bytes, true);
3931         if (ret < 0) {
3932             goto fail;
3933         }
3934 
3935         qemu_co_mutex_unlock(&s->lock);
3936         ret = bdrv_co_copy_range_to(src, src_offset,
3937                                     s->data_file,
3938                                     cluster_offset + offset_in_cluster,
3939                                     cur_bytes, read_flags, write_flags);
3940         qemu_co_mutex_lock(&s->lock);
3941         if (ret < 0) {
3942             goto fail;
3943         }
3944 
3945         ret = qcow2_handle_l2meta(bs, &l2meta, true);
3946         if (ret) {
3947             goto fail;
3948         }
3949 
3950         bytes -= cur_bytes;
3951         src_offset += cur_bytes;
3952         dst_offset += cur_bytes;
3953     }
3954     ret = 0;
3955 
3956 fail:
3957     qcow2_handle_l2meta(bs, &l2meta, false);
3958 
3959     qemu_co_mutex_unlock(&s->lock);
3960 
3961     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
3962 
3963     return ret;
3964 }
3965 
3966 static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,
3967                                           bool exact, PreallocMode prealloc,
3968                                           BdrvRequestFlags flags, Error **errp)
3969 {
3970     BDRVQcow2State *s = bs->opaque;
3971     uint64_t old_length;
3972     int64_t new_l1_size;
3973     int ret;
3974     QDict *options;
3975 
3976     if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA &&
3977         prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL)
3978     {
3979         error_setg(errp, "Unsupported preallocation mode '%s'",
3980                    PreallocMode_str(prealloc));
3981         return -ENOTSUP;
3982     }
3983 
3984     if (!QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)) {
3985         error_setg(errp, "The new size must be a multiple of %u",
3986                    (unsigned) BDRV_SECTOR_SIZE);
3987         return -EINVAL;
3988     }
3989 
3990     qemu_co_mutex_lock(&s->lock);
3991 
3992     /* cannot proceed if image has snapshots */
3993     if (s->nb_snapshots) {
3994         error_setg(errp, "Can't resize an image which has snapshots");
3995         ret = -ENOTSUP;
3996         goto fail;
3997     }
3998 
3999     /* cannot proceed if image has bitmaps */
4000     if (qcow2_truncate_bitmaps_check(bs, errp)) {
4001         ret = -ENOTSUP;
4002         goto fail;
4003     }
4004 
4005     old_length = bs->total_sectors * BDRV_SECTOR_SIZE;
4006     new_l1_size = size_to_l1(s, offset);
4007 
4008     if (offset < old_length) {
4009         int64_t last_cluster, old_file_size;
4010         if (prealloc != PREALLOC_MODE_OFF) {
4011             error_setg(errp,
4012                        "Preallocation can't be used for shrinking an image");
4013             ret = -EINVAL;
4014             goto fail;
4015         }
4016 
4017         ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size),
4018                                     old_length - ROUND_UP(offset,
4019                                                           s->cluster_size),
4020                                     QCOW2_DISCARD_ALWAYS, true);
4021         if (ret < 0) {
4022             error_setg_errno(errp, -ret, "Failed to discard cropped clusters");
4023             goto fail;
4024         }
4025 
4026         ret = qcow2_shrink_l1_table(bs, new_l1_size);
4027         if (ret < 0) {
4028             error_setg_errno(errp, -ret,
4029                              "Failed to reduce the number of L2 tables");
4030             goto fail;
4031         }
4032 
4033         ret = qcow2_shrink_reftable(bs);
4034         if (ret < 0) {
4035             error_setg_errno(errp, -ret,
4036                              "Failed to discard unused refblocks");
4037             goto fail;
4038         }
4039 
4040         old_file_size = bdrv_getlength(bs->file->bs);
4041         if (old_file_size < 0) {
4042             error_setg_errno(errp, -old_file_size,
4043                              "Failed to inquire current file length");
4044             ret = old_file_size;
4045             goto fail;
4046         }
4047         last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4048         if (last_cluster < 0) {
4049             error_setg_errno(errp, -last_cluster,
4050                              "Failed to find the last cluster");
4051             ret = last_cluster;
4052             goto fail;
4053         }
4054         if ((last_cluster + 1) * s->cluster_size < old_file_size) {
4055             Error *local_err = NULL;
4056 
4057             /*
4058              * Do not pass @exact here: It will not help the user if
4059              * we get an error here just because they wanted to shrink
4060              * their qcow2 image (on a block device) with qemu-img.
4061              * (And on the qcow2 layer, the @exact requirement is
4062              * always fulfilled, so there is no need to pass it on.)
4063              */
4064             bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size,
4065                              false, PREALLOC_MODE_OFF, 0, &local_err);
4066             if (local_err) {
4067                 warn_reportf_err(local_err,
4068                                  "Failed to truncate the tail of the image: ");
4069             }
4070         }
4071     } else {
4072         ret = qcow2_grow_l1_table(bs, new_l1_size, true);
4073         if (ret < 0) {
4074             error_setg_errno(errp, -ret, "Failed to grow the L1 table");
4075             goto fail;
4076         }
4077     }
4078 
4079     switch (prealloc) {
4080     case PREALLOC_MODE_OFF:
4081         if (has_data_file(bs)) {
4082             /*
4083              * If the caller wants an exact resize, the external data
4084              * file should be resized to the exact target size, too,
4085              * so we pass @exact here.
4086              */
4087             ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, 0,
4088                                    errp);
4089             if (ret < 0) {
4090                 goto fail;
4091             }
4092         }
4093         break;
4094 
4095     case PREALLOC_MODE_METADATA:
4096         ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4097         if (ret < 0) {
4098             goto fail;
4099         }
4100         break;
4101 
4102     case PREALLOC_MODE_FALLOC:
4103     case PREALLOC_MODE_FULL:
4104     {
4105         int64_t allocation_start, host_offset, guest_offset;
4106         int64_t clusters_allocated;
4107         int64_t old_file_size, new_file_size;
4108         uint64_t nb_new_data_clusters, nb_new_l2_tables;
4109 
4110         /* With a data file, preallocation means just allocating the metadata
4111          * and forwarding the truncate request to the data file */
4112         if (has_data_file(bs)) {
4113             ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4114             if (ret < 0) {
4115                 goto fail;
4116             }
4117             break;
4118         }
4119 
4120         old_file_size = bdrv_getlength(bs->file->bs);
4121         if (old_file_size < 0) {
4122             error_setg_errno(errp, -old_file_size,
4123                              "Failed to inquire current file length");
4124             ret = old_file_size;
4125             goto fail;
4126         }
4127         old_file_size = ROUND_UP(old_file_size, s->cluster_size);
4128 
4129         nb_new_data_clusters = DIV_ROUND_UP(offset - old_length,
4130                                             s->cluster_size);
4131 
4132         /* This is an overestimation; we will not actually allocate space for
4133          * these in the file but just make sure the new refcount structures are
4134          * able to cover them so we will not have to allocate new refblocks
4135          * while entering the data blocks in the potentially new L2 tables.
4136          * (We do not actually care where the L2 tables are placed. Maybe they
4137          *  are already allocated or they can be placed somewhere before
4138          *  @old_file_size. It does not matter because they will be fully
4139          *  allocated automatically, so they do not need to be covered by the
4140          *  preallocation. All that matters is that we will not have to allocate
4141          *  new refcount structures for them.) */
4142         nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters,
4143                                         s->cluster_size / sizeof(uint64_t));
4144         /* The cluster range may not be aligned to L2 boundaries, so add one L2
4145          * table for a potential head/tail */
4146         nb_new_l2_tables++;
4147 
4148         allocation_start = qcow2_refcount_area(bs, old_file_size,
4149                                                nb_new_data_clusters +
4150                                                nb_new_l2_tables,
4151                                                true, 0, 0);
4152         if (allocation_start < 0) {
4153             error_setg_errno(errp, -allocation_start,
4154                              "Failed to resize refcount structures");
4155             ret = allocation_start;
4156             goto fail;
4157         }
4158 
4159         clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start,
4160                                                      nb_new_data_clusters);
4161         if (clusters_allocated < 0) {
4162             error_setg_errno(errp, -clusters_allocated,
4163                              "Failed to allocate data clusters");
4164             ret = clusters_allocated;
4165             goto fail;
4166         }
4167 
4168         assert(clusters_allocated == nb_new_data_clusters);
4169 
4170         /* Allocate the data area */
4171         new_file_size = allocation_start +
4172                         nb_new_data_clusters * s->cluster_size;
4173         /*
4174          * Image file grows, so @exact does not matter.
4175          *
4176          * If we need to zero out the new area, try first whether the protocol
4177          * driver can already take care of this.
4178          */
4179         if (flags & BDRV_REQ_ZERO_WRITE) {
4180             ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc,
4181                                    BDRV_REQ_ZERO_WRITE, NULL);
4182             if (ret >= 0) {
4183                 flags &= ~BDRV_REQ_ZERO_WRITE;
4184             }
4185         } else {
4186             ret = -1;
4187         }
4188         if (ret < 0) {
4189             ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0,
4190                                    errp);
4191         }
4192         if (ret < 0) {
4193             error_prepend(errp, "Failed to resize underlying file: ");
4194             qcow2_free_clusters(bs, allocation_start,
4195                                 nb_new_data_clusters * s->cluster_size,
4196                                 QCOW2_DISCARD_OTHER);
4197             goto fail;
4198         }
4199 
4200         /* Create the necessary L2 entries */
4201         host_offset = allocation_start;
4202         guest_offset = old_length;
4203         while (nb_new_data_clusters) {
4204             int64_t nb_clusters = MIN(
4205                 nb_new_data_clusters,
4206                 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset));
4207             QCowL2Meta allocation = {
4208                 .offset       = guest_offset,
4209                 .alloc_offset = host_offset,
4210                 .nb_clusters  = nb_clusters,
4211             };
4212             qemu_co_queue_init(&allocation.dependent_requests);
4213 
4214             ret = qcow2_alloc_cluster_link_l2(bs, &allocation);
4215             if (ret < 0) {
4216                 error_setg_errno(errp, -ret, "Failed to update L2 tables");
4217                 qcow2_free_clusters(bs, host_offset,
4218                                     nb_new_data_clusters * s->cluster_size,
4219                                     QCOW2_DISCARD_OTHER);
4220                 goto fail;
4221             }
4222 
4223             guest_offset += nb_clusters * s->cluster_size;
4224             host_offset += nb_clusters * s->cluster_size;
4225             nb_new_data_clusters -= nb_clusters;
4226         }
4227         break;
4228     }
4229 
4230     default:
4231         g_assert_not_reached();
4232     }
4233 
4234     if ((flags & BDRV_REQ_ZERO_WRITE) && offset > old_length) {
4235         uint64_t zero_start = QEMU_ALIGN_UP(old_length, s->cluster_size);
4236 
4237         /*
4238          * Use zero clusters as much as we can. qcow2_cluster_zeroize()
4239          * requires a cluster-aligned start. The end may be unaligned if it is
4240          * at the end of the image (which it is here).
4241          */
4242         ret = qcow2_cluster_zeroize(bs, zero_start, offset - zero_start, 0);
4243         if (ret < 0) {
4244             error_setg_errno(errp, -ret, "Failed to zero out new clusters");
4245             goto fail;
4246         }
4247 
4248         /* Write explicit zeros for the unaligned head */
4249         if (zero_start > old_length) {
4250             uint64_t len = zero_start - old_length;
4251             uint8_t *buf = qemu_blockalign0(bs, len);
4252             QEMUIOVector qiov;
4253             qemu_iovec_init_buf(&qiov, buf, len);
4254 
4255             qemu_co_mutex_unlock(&s->lock);
4256             ret = qcow2_co_pwritev_part(bs, old_length, len, &qiov, 0, 0);
4257             qemu_co_mutex_lock(&s->lock);
4258 
4259             qemu_vfree(buf);
4260             if (ret < 0) {
4261                 error_setg_errno(errp, -ret, "Failed to zero out the new area");
4262                 goto fail;
4263             }
4264         }
4265     }
4266 
4267     if (prealloc != PREALLOC_MODE_OFF) {
4268         /* Flush metadata before actually changing the image size */
4269         ret = qcow2_write_caches(bs);
4270         if (ret < 0) {
4271             error_setg_errno(errp, -ret,
4272                              "Failed to flush the preallocated area to disk");
4273             goto fail;
4274         }
4275     }
4276 
4277     bs->total_sectors = offset / BDRV_SECTOR_SIZE;
4278 
4279     /* write updated header.size */
4280     offset = cpu_to_be64(offset);
4281     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
4282                            &offset, sizeof(uint64_t));
4283     if (ret < 0) {
4284         error_setg_errno(errp, -ret, "Failed to update the image size");
4285         goto fail;
4286     }
4287 
4288     s->l1_vm_state_index = new_l1_size;
4289 
4290     /* Update cache sizes */
4291     options = qdict_clone_shallow(bs->options);
4292     ret = qcow2_update_options(bs, options, s->flags, errp);
4293     qobject_unref(options);
4294     if (ret < 0) {
4295         goto fail;
4296     }
4297     ret = 0;
4298 fail:
4299     qemu_co_mutex_unlock(&s->lock);
4300     return ret;
4301 }
4302 
4303 static coroutine_fn int
4304 qcow2_co_pwritev_compressed_task(BlockDriverState *bs,
4305                                  uint64_t offset, uint64_t bytes,
4306                                  QEMUIOVector *qiov, size_t qiov_offset)
4307 {
4308     BDRVQcow2State *s = bs->opaque;
4309     int ret;
4310     ssize_t out_len;
4311     uint8_t *buf, *out_buf;
4312     uint64_t cluster_offset;
4313 
4314     assert(bytes == s->cluster_size || (bytes < s->cluster_size &&
4315            (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS)));
4316 
4317     buf = qemu_blockalign(bs, s->cluster_size);
4318     if (bytes < s->cluster_size) {
4319         /* Zero-pad last write if image size is not cluster aligned */
4320         memset(buf + bytes, 0, s->cluster_size - bytes);
4321     }
4322     qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes);
4323 
4324     out_buf = g_malloc(s->cluster_size);
4325 
4326     out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1,
4327                                 buf, s->cluster_size);
4328     if (out_len == -ENOMEM) {
4329         /* could not compress: write normal cluster */
4330         ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0);
4331         if (ret < 0) {
4332             goto fail;
4333         }
4334         goto success;
4335     } else if (out_len < 0) {
4336         ret = -EINVAL;
4337         goto fail;
4338     }
4339 
4340     qemu_co_mutex_lock(&s->lock);
4341     ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len,
4342                                                 &cluster_offset);
4343     if (ret < 0) {
4344         qemu_co_mutex_unlock(&s->lock);
4345         goto fail;
4346     }
4347 
4348     ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true);
4349     qemu_co_mutex_unlock(&s->lock);
4350     if (ret < 0) {
4351         goto fail;
4352     }
4353 
4354     BLKDBG_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED);
4355     ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0);
4356     if (ret < 0) {
4357         goto fail;
4358     }
4359 success:
4360     ret = 0;
4361 fail:
4362     qemu_vfree(buf);
4363     g_free(out_buf);
4364     return ret;
4365 }
4366 
4367 static coroutine_fn int qcow2_co_pwritev_compressed_task_entry(AioTask *task)
4368 {
4369     Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
4370 
4371     assert(!t->cluster_type && !t->l2meta);
4372 
4373     return qcow2_co_pwritev_compressed_task(t->bs, t->offset, t->bytes, t->qiov,
4374                                             t->qiov_offset);
4375 }
4376 
4377 /*
4378  * XXX: put compressed sectors first, then all the cluster aligned
4379  * tables to avoid losing bytes in alignment
4380  */
4381 static coroutine_fn int
4382 qcow2_co_pwritev_compressed_part(BlockDriverState *bs,
4383                                  uint64_t offset, uint64_t bytes,
4384                                  QEMUIOVector *qiov, size_t qiov_offset)
4385 {
4386     BDRVQcow2State *s = bs->opaque;
4387     AioTaskPool *aio = NULL;
4388     int ret = 0;
4389 
4390     if (has_data_file(bs)) {
4391         return -ENOTSUP;
4392     }
4393 
4394     if (bytes == 0) {
4395         /*
4396          * align end of file to a sector boundary to ease reading with
4397          * sector based I/Os
4398          */
4399         int64_t len = bdrv_getlength(bs->file->bs);
4400         if (len < 0) {
4401             return len;
4402         }
4403         return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, 0,
4404                                 NULL);
4405     }
4406 
4407     if (offset_into_cluster(s, offset)) {
4408         return -EINVAL;
4409     }
4410 
4411     if (offset_into_cluster(s, bytes) &&
4412         (offset + bytes) != (bs->total_sectors << BDRV_SECTOR_BITS)) {
4413         return -EINVAL;
4414     }
4415 
4416     while (bytes && aio_task_pool_status(aio) == 0) {
4417         uint64_t chunk_size = MIN(bytes, s->cluster_size);
4418 
4419         if (!aio && chunk_size != bytes) {
4420             aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
4421         }
4422 
4423         ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_compressed_task_entry,
4424                              0, 0, offset, chunk_size, qiov, qiov_offset, NULL);
4425         if (ret < 0) {
4426             break;
4427         }
4428         qiov_offset += chunk_size;
4429         offset += chunk_size;
4430         bytes -= chunk_size;
4431     }
4432 
4433     if (aio) {
4434         aio_task_pool_wait_all(aio);
4435         if (ret == 0) {
4436             ret = aio_task_pool_status(aio);
4437         }
4438         g_free(aio);
4439     }
4440 
4441     return ret;
4442 }
4443 
4444 static int coroutine_fn
4445 qcow2_co_preadv_compressed(BlockDriverState *bs,
4446                            uint64_t file_cluster_offset,
4447                            uint64_t offset,
4448                            uint64_t bytes,
4449                            QEMUIOVector *qiov,
4450                            size_t qiov_offset)
4451 {
4452     BDRVQcow2State *s = bs->opaque;
4453     int ret = 0, csize, nb_csectors;
4454     uint64_t coffset;
4455     uint8_t *buf, *out_buf;
4456     int offset_in_cluster = offset_into_cluster(s, offset);
4457 
4458     coffset = file_cluster_offset & s->cluster_offset_mask;
4459     nb_csectors = ((file_cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
4460     csize = nb_csectors * QCOW2_COMPRESSED_SECTOR_SIZE -
4461         (coffset & ~QCOW2_COMPRESSED_SECTOR_MASK);
4462 
4463     buf = g_try_malloc(csize);
4464     if (!buf) {
4465         return -ENOMEM;
4466     }
4467 
4468     out_buf = qemu_blockalign(bs, s->cluster_size);
4469 
4470     BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
4471     ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0);
4472     if (ret < 0) {
4473         goto fail;
4474     }
4475 
4476     if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) {
4477         ret = -EIO;
4478         goto fail;
4479     }
4480 
4481     qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes);
4482 
4483 fail:
4484     qemu_vfree(out_buf);
4485     g_free(buf);
4486 
4487     return ret;
4488 }
4489 
4490 static int make_completely_empty(BlockDriverState *bs)
4491 {
4492     BDRVQcow2State *s = bs->opaque;
4493     Error *local_err = NULL;
4494     int ret, l1_clusters;
4495     int64_t offset;
4496     uint64_t *new_reftable = NULL;
4497     uint64_t rt_entry, l1_size2;
4498     struct {
4499         uint64_t l1_offset;
4500         uint64_t reftable_offset;
4501         uint32_t reftable_clusters;
4502     } QEMU_PACKED l1_ofs_rt_ofs_cls;
4503 
4504     ret = qcow2_cache_empty(bs, s->l2_table_cache);
4505     if (ret < 0) {
4506         goto fail;
4507     }
4508 
4509     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
4510     if (ret < 0) {
4511         goto fail;
4512     }
4513 
4514     /* Refcounts will be broken utterly */
4515     ret = qcow2_mark_dirty(bs);
4516     if (ret < 0) {
4517         goto fail;
4518     }
4519 
4520     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4521 
4522     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
4523     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
4524 
4525     /* After this call, neither the in-memory nor the on-disk refcount
4526      * information accurately describe the actual references */
4527 
4528     ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
4529                              l1_clusters * s->cluster_size, 0);
4530     if (ret < 0) {
4531         goto fail_broken_refcounts;
4532     }
4533     memset(s->l1_table, 0, l1_size2);
4534 
4535     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
4536 
4537     /* Overwrite enough clusters at the beginning of the sectors to place
4538      * the refcount table, a refcount block and the L1 table in; this may
4539      * overwrite parts of the existing refcount and L1 table, which is not
4540      * an issue because the dirty flag is set, complete data loss is in fact
4541      * desired and partial data loss is consequently fine as well */
4542     ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
4543                              (2 + l1_clusters) * s->cluster_size, 0);
4544     /* This call (even if it failed overall) may have overwritten on-disk
4545      * refcount structures; in that case, the in-memory refcount information
4546      * will probably differ from the on-disk information which makes the BDS
4547      * unusable */
4548     if (ret < 0) {
4549         goto fail_broken_refcounts;
4550     }
4551 
4552     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4553     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
4554 
4555     /* "Create" an empty reftable (one cluster) directly after the image
4556      * header and an empty L1 table three clusters after the image header;
4557      * the cluster between those two will be used as the first refblock */
4558     l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
4559     l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
4560     l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
4561     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
4562                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
4563     if (ret < 0) {
4564         goto fail_broken_refcounts;
4565     }
4566 
4567     s->l1_table_offset = 3 * s->cluster_size;
4568 
4569     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
4570     if (!new_reftable) {
4571         ret = -ENOMEM;
4572         goto fail_broken_refcounts;
4573     }
4574 
4575     s->refcount_table_offset = s->cluster_size;
4576     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
4577     s->max_refcount_table_index = 0;
4578 
4579     g_free(s->refcount_table);
4580     s->refcount_table = new_reftable;
4581     new_reftable = NULL;
4582 
4583     /* Now the in-memory refcount information again corresponds to the on-disk
4584      * information (reftable is empty and no refblocks (the refblock cache is
4585      * empty)); however, this means some clusters (e.g. the image header) are
4586      * referenced, but not refcounted, but the normal qcow2 code assumes that
4587      * the in-memory information is always correct */
4588 
4589     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
4590 
4591     /* Enter the first refblock into the reftable */
4592     rt_entry = cpu_to_be64(2 * s->cluster_size);
4593     ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
4594                            &rt_entry, sizeof(rt_entry));
4595     if (ret < 0) {
4596         goto fail_broken_refcounts;
4597     }
4598     s->refcount_table[0] = 2 * s->cluster_size;
4599 
4600     s->free_cluster_index = 0;
4601     assert(3 + l1_clusters <= s->refcount_block_size);
4602     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
4603     if (offset < 0) {
4604         ret = offset;
4605         goto fail_broken_refcounts;
4606     } else if (offset > 0) {
4607         error_report("First cluster in emptied image is in use");
4608         abort();
4609     }
4610 
4611     /* Now finally the in-memory information corresponds to the on-disk
4612      * structures and is correct */
4613     ret = qcow2_mark_clean(bs);
4614     if (ret < 0) {
4615         goto fail;
4616     }
4617 
4618     ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false,
4619                         PREALLOC_MODE_OFF, 0, &local_err);
4620     if (ret < 0) {
4621         error_report_err(local_err);
4622         goto fail;
4623     }
4624 
4625     return 0;
4626 
4627 fail_broken_refcounts:
4628     /* The BDS is unusable at this point. If we wanted to make it usable, we
4629      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4630      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4631      * again. However, because the functions which could have caused this error
4632      * path to be taken are used by those functions as well, it's very likely
4633      * that that sequence will fail as well. Therefore, just eject the BDS. */
4634     bs->drv = NULL;
4635 
4636 fail:
4637     g_free(new_reftable);
4638     return ret;
4639 }
4640 
4641 static int qcow2_make_empty(BlockDriverState *bs)
4642 {
4643     BDRVQcow2State *s = bs->opaque;
4644     uint64_t offset, end_offset;
4645     int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
4646     int l1_clusters, ret = 0;
4647 
4648     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
4649 
4650     if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps &&
4651         3 + l1_clusters <= s->refcount_block_size &&
4652         s->crypt_method_header != QCOW_CRYPT_LUKS &&
4653         !has_data_file(bs)) {
4654         /* The following function only works for qcow2 v3 images (it
4655          * requires the dirty flag) and only as long as there are no
4656          * features that reserve extra clusters (such as snapshots,
4657          * LUKS header, or persistent bitmaps), because it completely
4658          * empties the image.  Furthermore, the L1 table and three
4659          * additional clusters (image header, refcount table, one
4660          * refcount block) have to fit inside one refcount block. It
4661          * only resets the image file, i.e. does not work with an
4662          * external data file. */
4663         return make_completely_empty(bs);
4664     }
4665 
4666     /* This fallback code simply discards every active cluster; this is slow,
4667      * but works in all cases */
4668     end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
4669     for (offset = 0; offset < end_offset; offset += step) {
4670         /* As this function is generally used after committing an external
4671          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4672          * default action for this kind of discard is to pass the discard,
4673          * which will ideally result in an actually smaller image file, as
4674          * is probably desired. */
4675         ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
4676                                     QCOW2_DISCARD_SNAPSHOT, true);
4677         if (ret < 0) {
4678             break;
4679         }
4680     }
4681 
4682     return ret;
4683 }
4684 
4685 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
4686 {
4687     BDRVQcow2State *s = bs->opaque;
4688     int ret;
4689 
4690     qemu_co_mutex_lock(&s->lock);
4691     ret = qcow2_write_caches(bs);
4692     qemu_co_mutex_unlock(&s->lock);
4693 
4694     return ret;
4695 }
4696 
4697 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs,
4698                                        Error **errp)
4699 {
4700     Error *local_err = NULL;
4701     BlockMeasureInfo *info;
4702     uint64_t required = 0; /* bytes that contribute to required size */
4703     uint64_t virtual_size; /* disk size as seen by guest */
4704     uint64_t refcount_bits;
4705     uint64_t l2_tables;
4706     uint64_t luks_payload_size = 0;
4707     size_t cluster_size;
4708     int version;
4709     char *optstr;
4710     PreallocMode prealloc;
4711     bool has_backing_file;
4712     bool has_luks;
4713 
4714     /* Parse image creation options */
4715     cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err);
4716     if (local_err) {
4717         goto err;
4718     }
4719 
4720     version = qcow2_opt_get_version_del(opts, &local_err);
4721     if (local_err) {
4722         goto err;
4723     }
4724 
4725     refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
4726     if (local_err) {
4727         goto err;
4728     }
4729 
4730     optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
4731     prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr,
4732                                PREALLOC_MODE_OFF, &local_err);
4733     g_free(optstr);
4734     if (local_err) {
4735         goto err;
4736     }
4737 
4738     optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
4739     has_backing_file = !!optstr;
4740     g_free(optstr);
4741 
4742     optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
4743     has_luks = optstr && strcmp(optstr, "luks") == 0;
4744     g_free(optstr);
4745 
4746     if (has_luks) {
4747         g_autoptr(QCryptoBlockCreateOptions) create_opts = NULL;
4748         QDict *opts_qdict;
4749         QDict *cryptoopts;
4750         size_t headerlen;
4751 
4752         opts_qdict = qemu_opts_to_qdict(opts, NULL);
4753         qdict_extract_subqdict(opts_qdict, &cryptoopts, "encrypt.");
4754         qobject_unref(opts_qdict);
4755 
4756         qdict_put_str(cryptoopts, "format", "luks");
4757 
4758         create_opts = block_crypto_create_opts_init(cryptoopts, errp);
4759         qobject_unref(cryptoopts);
4760         if (!create_opts) {
4761             goto err;
4762         }
4763 
4764         if (!qcrypto_block_calculate_payload_offset(create_opts,
4765                                                     "encrypt.",
4766                                                     &headerlen,
4767                                                     &local_err)) {
4768             goto err;
4769         }
4770 
4771         luks_payload_size = ROUND_UP(headerlen, cluster_size);
4772     }
4773 
4774     virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
4775     virtual_size = ROUND_UP(virtual_size, cluster_size);
4776 
4777     /* Check that virtual disk size is valid */
4778     l2_tables = DIV_ROUND_UP(virtual_size / cluster_size,
4779                              cluster_size / sizeof(uint64_t));
4780     if (l2_tables * sizeof(uint64_t) > QCOW_MAX_L1_SIZE) {
4781         error_setg(&local_err, "The image size is too large "
4782                                "(try using a larger cluster size)");
4783         goto err;
4784     }
4785 
4786     /* Account for input image */
4787     if (in_bs) {
4788         int64_t ssize = bdrv_getlength(in_bs);
4789         if (ssize < 0) {
4790             error_setg_errno(&local_err, -ssize,
4791                              "Unable to get image virtual_size");
4792             goto err;
4793         }
4794 
4795         virtual_size = ROUND_UP(ssize, cluster_size);
4796 
4797         if (has_backing_file) {
4798             /* We don't how much of the backing chain is shared by the input
4799              * image and the new image file.  In the worst case the new image's
4800              * backing file has nothing in common with the input image.  Be
4801              * conservative and assume all clusters need to be written.
4802              */
4803             required = virtual_size;
4804         } else {
4805             int64_t offset;
4806             int64_t pnum = 0;
4807 
4808             for (offset = 0; offset < ssize; offset += pnum) {
4809                 int ret;
4810 
4811                 ret = bdrv_block_status_above(in_bs, NULL, offset,
4812                                               ssize - offset, &pnum, NULL,
4813                                               NULL);
4814                 if (ret < 0) {
4815                     error_setg_errno(&local_err, -ret,
4816                                      "Unable to get block status");
4817                     goto err;
4818                 }
4819 
4820                 if (ret & BDRV_BLOCK_ZERO) {
4821                     /* Skip zero regions (safe with no backing file) */
4822                 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
4823                            (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
4824                     /* Extend pnum to end of cluster for next iteration */
4825                     pnum = ROUND_UP(offset + pnum, cluster_size) - offset;
4826 
4827                     /* Count clusters we've seen */
4828                     required += offset % cluster_size + pnum;
4829                 }
4830             }
4831         }
4832     }
4833 
4834     /* Take into account preallocation.  Nothing special is needed for
4835      * PREALLOC_MODE_METADATA since metadata is always counted.
4836      */
4837     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
4838         required = virtual_size;
4839     }
4840 
4841     info = g_new(BlockMeasureInfo, 1);
4842     info->fully_allocated =
4843         qcow2_calc_prealloc_size(virtual_size, cluster_size,
4844                                  ctz32(refcount_bits)) + luks_payload_size;
4845 
4846     /* Remove data clusters that are not required.  This overestimates the
4847      * required size because metadata needed for the fully allocated file is
4848      * still counted.
4849      */
4850     info->required = info->fully_allocated - virtual_size + required;
4851     return info;
4852 
4853 err:
4854     error_propagate(errp, local_err);
4855     return NULL;
4856 }
4857 
4858 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4859 {
4860     BDRVQcow2State *s = bs->opaque;
4861     bdi->unallocated_blocks_are_zero = true;
4862     bdi->cluster_size = s->cluster_size;
4863     bdi->vm_state_offset = qcow2_vm_state_offset(s);
4864     return 0;
4865 }
4866 
4867 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs,
4868                                                   Error **errp)
4869 {
4870     BDRVQcow2State *s = bs->opaque;
4871     ImageInfoSpecific *spec_info;
4872     QCryptoBlockInfo *encrypt_info = NULL;
4873     Error *local_err = NULL;
4874 
4875     if (s->crypto != NULL) {
4876         encrypt_info = qcrypto_block_get_info(s->crypto, &local_err);
4877         if (local_err) {
4878             error_propagate(errp, local_err);
4879             return NULL;
4880         }
4881     }
4882 
4883     spec_info = g_new(ImageInfoSpecific, 1);
4884     *spec_info = (ImageInfoSpecific){
4885         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
4886         .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1),
4887     };
4888     if (s->qcow_version == 2) {
4889         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
4890             .compat             = g_strdup("0.10"),
4891             .refcount_bits      = s->refcount_bits,
4892         };
4893     } else if (s->qcow_version == 3) {
4894         Qcow2BitmapInfoList *bitmaps;
4895         bitmaps = qcow2_get_bitmap_info_list(bs, &local_err);
4896         if (local_err) {
4897             error_propagate(errp, local_err);
4898             qapi_free_ImageInfoSpecific(spec_info);
4899             qapi_free_QCryptoBlockInfo(encrypt_info);
4900             return NULL;
4901         }
4902         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
4903             .compat             = g_strdup("1.1"),
4904             .lazy_refcounts     = s->compatible_features &
4905                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
4906             .has_lazy_refcounts = true,
4907             .corrupt            = s->incompatible_features &
4908                                   QCOW2_INCOMPAT_CORRUPT,
4909             .has_corrupt        = true,
4910             .refcount_bits      = s->refcount_bits,
4911             .has_bitmaps        = !!bitmaps,
4912             .bitmaps            = bitmaps,
4913             .has_data_file      = !!s->image_data_file,
4914             .data_file          = g_strdup(s->image_data_file),
4915             .has_data_file_raw  = has_data_file(bs),
4916             .data_file_raw      = data_file_is_raw(bs),
4917         };
4918     } else {
4919         /* if this assertion fails, this probably means a new version was
4920          * added without having it covered here */
4921         assert(false);
4922     }
4923 
4924     if (encrypt_info) {
4925         ImageInfoSpecificQCow2Encryption *qencrypt =
4926             g_new(ImageInfoSpecificQCow2Encryption, 1);
4927         switch (encrypt_info->format) {
4928         case Q_CRYPTO_BLOCK_FORMAT_QCOW:
4929             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES;
4930             break;
4931         case Q_CRYPTO_BLOCK_FORMAT_LUKS:
4932             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS;
4933             qencrypt->u.luks = encrypt_info->u.luks;
4934             break;
4935         default:
4936             abort();
4937         }
4938         /* Since we did shallow copy above, erase any pointers
4939          * in the original info */
4940         memset(&encrypt_info->u, 0, sizeof(encrypt_info->u));
4941         qapi_free_QCryptoBlockInfo(encrypt_info);
4942 
4943         spec_info->u.qcow2.data->has_encrypt = true;
4944         spec_info->u.qcow2.data->encrypt = qencrypt;
4945     }
4946 
4947     return spec_info;
4948 }
4949 
4950 static int qcow2_has_zero_init(BlockDriverState *bs)
4951 {
4952     BDRVQcow2State *s = bs->opaque;
4953     bool preallocated;
4954 
4955     if (qemu_in_coroutine()) {
4956         qemu_co_mutex_lock(&s->lock);
4957     }
4958     /*
4959      * Check preallocation status: Preallocated images have all L2
4960      * tables allocated, nonpreallocated images have none.  It is
4961      * therefore enough to check the first one.
4962      */
4963     preallocated = s->l1_size > 0 && s->l1_table[0] != 0;
4964     if (qemu_in_coroutine()) {
4965         qemu_co_mutex_unlock(&s->lock);
4966     }
4967 
4968     if (!preallocated) {
4969         return 1;
4970     } else if (bs->encrypted) {
4971         return 0;
4972     } else {
4973         return bdrv_has_zero_init(s->data_file->bs);
4974     }
4975 }
4976 
4977 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
4978                               int64_t pos)
4979 {
4980     BDRVQcow2State *s = bs->opaque;
4981 
4982     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
4983     return bs->drv->bdrv_co_pwritev_part(bs, qcow2_vm_state_offset(s) + pos,
4984                                          qiov->size, qiov, 0, 0);
4985 }
4986 
4987 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
4988                               int64_t pos)
4989 {
4990     BDRVQcow2State *s = bs->opaque;
4991 
4992     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
4993     return bs->drv->bdrv_co_preadv_part(bs, qcow2_vm_state_offset(s) + pos,
4994                                         qiov->size, qiov, 0, 0);
4995 }
4996 
4997 /*
4998  * Downgrades an image's version. To achieve this, any incompatible features
4999  * have to be removed.
5000  */
5001 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
5002                            BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5003                            Error **errp)
5004 {
5005     BDRVQcow2State *s = bs->opaque;
5006     int current_version = s->qcow_version;
5007     int ret;
5008 
5009     /* This is qcow2_downgrade(), not qcow2_upgrade() */
5010     assert(target_version < current_version);
5011 
5012     /* There are no other versions (now) that you can downgrade to */
5013     assert(target_version == 2);
5014 
5015     if (s->refcount_order != 4) {
5016         error_setg(errp, "compat=0.10 requires refcount_bits=16");
5017         return -ENOTSUP;
5018     }
5019 
5020     if (has_data_file(bs)) {
5021         error_setg(errp, "Cannot downgrade an image with a data file");
5022         return -ENOTSUP;
5023     }
5024 
5025     /* clear incompatible features */
5026     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
5027         ret = qcow2_mark_clean(bs);
5028         if (ret < 0) {
5029             error_setg_errno(errp, -ret, "Failed to make the image clean");
5030             return ret;
5031         }
5032     }
5033 
5034     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
5035      * the first place; if that happens nonetheless, returning -ENOTSUP is the
5036      * best thing to do anyway */
5037 
5038     if (s->incompatible_features) {
5039         error_setg(errp, "Cannot downgrade an image with incompatible features "
5040                    "%#" PRIx64 " set", s->incompatible_features);
5041         return -ENOTSUP;
5042     }
5043 
5044     /* since we can ignore compatible features, we can set them to 0 as well */
5045     s->compatible_features = 0;
5046     /* if lazy refcounts have been used, they have already been fixed through
5047      * clearing the dirty flag */
5048 
5049     /* clearing autoclear features is trivial */
5050     s->autoclear_features = 0;
5051 
5052     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
5053     if (ret < 0) {
5054         error_setg_errno(errp, -ret, "Failed to turn zero into data clusters");
5055         return ret;
5056     }
5057 
5058     s->qcow_version = target_version;
5059     ret = qcow2_update_header(bs);
5060     if (ret < 0) {
5061         s->qcow_version = current_version;
5062         error_setg_errno(errp, -ret, "Failed to update the image header");
5063         return ret;
5064     }
5065     return 0;
5066 }
5067 
5068 /*
5069  * Upgrades an image's version.  While newer versions encompass all
5070  * features of older versions, some things may have to be presented
5071  * differently.
5072  */
5073 static int qcow2_upgrade(BlockDriverState *bs, int target_version,
5074                          BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5075                          Error **errp)
5076 {
5077     BDRVQcow2State *s = bs->opaque;
5078     bool need_snapshot_update;
5079     int current_version = s->qcow_version;
5080     int i;
5081     int ret;
5082 
5083     /* This is qcow2_upgrade(), not qcow2_downgrade() */
5084     assert(target_version > current_version);
5085 
5086     /* There are no other versions (yet) that you can upgrade to */
5087     assert(target_version == 3);
5088 
5089     status_cb(bs, 0, 2, cb_opaque);
5090 
5091     /*
5092      * In v2, snapshots do not need to have extra data.  v3 requires
5093      * the 64-bit VM state size and the virtual disk size to be
5094      * present.
5095      * qcow2_write_snapshots() will always write the list in the
5096      * v3-compliant format.
5097      */
5098     need_snapshot_update = false;
5099     for (i = 0; i < s->nb_snapshots; i++) {
5100         if (s->snapshots[i].extra_data_size <
5101             sizeof_field(QCowSnapshotExtraData, vm_state_size_large) +
5102             sizeof_field(QCowSnapshotExtraData, disk_size))
5103         {
5104             need_snapshot_update = true;
5105             break;
5106         }
5107     }
5108     if (need_snapshot_update) {
5109         ret = qcow2_write_snapshots(bs);
5110         if (ret < 0) {
5111             error_setg_errno(errp, -ret, "Failed to update the snapshot table");
5112             return ret;
5113         }
5114     }
5115     status_cb(bs, 1, 2, cb_opaque);
5116 
5117     s->qcow_version = target_version;
5118     ret = qcow2_update_header(bs);
5119     if (ret < 0) {
5120         s->qcow_version = current_version;
5121         error_setg_errno(errp, -ret, "Failed to update the image header");
5122         return ret;
5123     }
5124     status_cb(bs, 2, 2, cb_opaque);
5125 
5126     return 0;
5127 }
5128 
5129 typedef enum Qcow2AmendOperation {
5130     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5131      * statically initialized to so that the helper CB can discern the first
5132      * invocation from an operation change */
5133     QCOW2_NO_OPERATION = 0,
5134 
5135     QCOW2_UPGRADING,
5136     QCOW2_CHANGING_REFCOUNT_ORDER,
5137     QCOW2_DOWNGRADING,
5138 } Qcow2AmendOperation;
5139 
5140 typedef struct Qcow2AmendHelperCBInfo {
5141     /* The code coordinating the amend operations should only modify
5142      * these four fields; the rest will be managed by the CB */
5143     BlockDriverAmendStatusCB *original_status_cb;
5144     void *original_cb_opaque;
5145 
5146     Qcow2AmendOperation current_operation;
5147 
5148     /* Total number of operations to perform (only set once) */
5149     int total_operations;
5150 
5151     /* The following fields are managed by the CB */
5152 
5153     /* Number of operations completed */
5154     int operations_completed;
5155 
5156     /* Cumulative offset of all completed operations */
5157     int64_t offset_completed;
5158 
5159     Qcow2AmendOperation last_operation;
5160     int64_t last_work_size;
5161 } Qcow2AmendHelperCBInfo;
5162 
5163 static void qcow2_amend_helper_cb(BlockDriverState *bs,
5164                                   int64_t operation_offset,
5165                                   int64_t operation_work_size, void *opaque)
5166 {
5167     Qcow2AmendHelperCBInfo *info = opaque;
5168     int64_t current_work_size;
5169     int64_t projected_work_size;
5170 
5171     if (info->current_operation != info->last_operation) {
5172         if (info->last_operation != QCOW2_NO_OPERATION) {
5173             info->offset_completed += info->last_work_size;
5174             info->operations_completed++;
5175         }
5176 
5177         info->last_operation = info->current_operation;
5178     }
5179 
5180     assert(info->total_operations > 0);
5181     assert(info->operations_completed < info->total_operations);
5182 
5183     info->last_work_size = operation_work_size;
5184 
5185     current_work_size = info->offset_completed + operation_work_size;
5186 
5187     /* current_work_size is the total work size for (operations_completed + 1)
5188      * operations (which includes this one), so multiply it by the number of
5189      * operations not covered and divide it by the number of operations
5190      * covered to get a projection for the operations not covered */
5191     projected_work_size = current_work_size * (info->total_operations -
5192                                                info->operations_completed - 1)
5193                                             / (info->operations_completed + 1);
5194 
5195     info->original_status_cb(bs, info->offset_completed + operation_offset,
5196                              current_work_size + projected_work_size,
5197                              info->original_cb_opaque);
5198 }
5199 
5200 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
5201                                BlockDriverAmendStatusCB *status_cb,
5202                                void *cb_opaque,
5203                                Error **errp)
5204 {
5205     BDRVQcow2State *s = bs->opaque;
5206     int old_version = s->qcow_version, new_version = old_version;
5207     uint64_t new_size = 0;
5208     const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL;
5209     bool lazy_refcounts = s->use_lazy_refcounts;
5210     bool data_file_raw = data_file_is_raw(bs);
5211     const char *compat = NULL;
5212     uint64_t cluster_size = s->cluster_size;
5213     bool encrypt;
5214     int encformat;
5215     int refcount_bits = s->refcount_bits;
5216     int ret;
5217     QemuOptDesc *desc = opts->list->desc;
5218     Qcow2AmendHelperCBInfo helper_cb_info;
5219 
5220     while (desc && desc->name) {
5221         if (!qemu_opt_find(opts, desc->name)) {
5222             /* only change explicitly defined options */
5223             desc++;
5224             continue;
5225         }
5226 
5227         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
5228             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
5229             if (!compat) {
5230                 /* preserve default */
5231             } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) {
5232                 new_version = 2;
5233             } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) {
5234                 new_version = 3;
5235             } else {
5236                 error_setg(errp, "Unknown compatibility level %s", compat);
5237                 return -EINVAL;
5238             }
5239         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
5240             error_setg(errp, "Cannot change preallocation mode");
5241             return -ENOTSUP;
5242         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
5243             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
5244         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
5245             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5246         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
5247             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5248         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
5249             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
5250                                         !!s->crypto);
5251 
5252             if (encrypt != !!s->crypto) {
5253                 error_setg(errp,
5254                            "Changing the encryption flag is not supported");
5255                 return -ENOTSUP;
5256             }
5257         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT_FORMAT)) {
5258             encformat = qcow2_crypt_method_from_format(
5259                 qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT));
5260 
5261             if (encformat != s->crypt_method_header) {
5262                 error_setg(errp,
5263                            "Changing the encryption format is not supported");
5264                 return -ENOTSUP;
5265             }
5266         } else if (g_str_has_prefix(desc->name, "encrypt.")) {
5267             error_setg(errp,
5268                        "Changing the encryption parameters is not supported");
5269             return -ENOTSUP;
5270         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
5271             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
5272                                              cluster_size);
5273             if (cluster_size != s->cluster_size) {
5274                 error_setg(errp, "Changing the cluster size is not supported");
5275                 return -ENOTSUP;
5276             }
5277         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
5278             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
5279                                                lazy_refcounts);
5280         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
5281             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
5282                                                 refcount_bits);
5283 
5284             if (refcount_bits <= 0 || refcount_bits > 64 ||
5285                 !is_power_of_2(refcount_bits))
5286             {
5287                 error_setg(errp, "Refcount width must be a power of two and "
5288                            "may not exceed 64 bits");
5289                 return -EINVAL;
5290             }
5291         } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) {
5292             data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE);
5293             if (data_file && !has_data_file(bs)) {
5294                 error_setg(errp, "data-file can only be set for images that "
5295                                  "use an external data file");
5296                 return -EINVAL;
5297             }
5298         } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) {
5299             data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW,
5300                                               data_file_raw);
5301             if (data_file_raw && !data_file_is_raw(bs)) {
5302                 error_setg(errp, "data-file-raw cannot be set on existing "
5303                                  "images");
5304                 return -EINVAL;
5305             }
5306         } else {
5307             /* if this point is reached, this probably means a new option was
5308              * added without having it covered here */
5309             abort();
5310         }
5311 
5312         desc++;
5313     }
5314 
5315     helper_cb_info = (Qcow2AmendHelperCBInfo){
5316         .original_status_cb = status_cb,
5317         .original_cb_opaque = cb_opaque,
5318         .total_operations = (new_version != old_version)
5319                           + (s->refcount_bits != refcount_bits)
5320     };
5321 
5322     /* Upgrade first (some features may require compat=1.1) */
5323     if (new_version > old_version) {
5324         helper_cb_info.current_operation = QCOW2_UPGRADING;
5325         ret = qcow2_upgrade(bs, new_version, &qcow2_amend_helper_cb,
5326                             &helper_cb_info, errp);
5327         if (ret < 0) {
5328             return ret;
5329         }
5330     }
5331 
5332     if (s->refcount_bits != refcount_bits) {
5333         int refcount_order = ctz32(refcount_bits);
5334 
5335         if (new_version < 3 && refcount_bits != 16) {
5336             error_setg(errp, "Refcount widths other than 16 bits require "
5337                        "compatibility level 1.1 or above (use compat=1.1 or "
5338                        "greater)");
5339             return -EINVAL;
5340         }
5341 
5342         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
5343         ret = qcow2_change_refcount_order(bs, refcount_order,
5344                                           &qcow2_amend_helper_cb,
5345                                           &helper_cb_info, errp);
5346         if (ret < 0) {
5347             return ret;
5348         }
5349     }
5350 
5351     /* data-file-raw blocks backing files, so clear it first if requested */
5352     if (data_file_raw) {
5353         s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5354     } else {
5355         s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5356     }
5357 
5358     if (data_file) {
5359         g_free(s->image_data_file);
5360         s->image_data_file = *data_file ? g_strdup(data_file) : NULL;
5361     }
5362 
5363     ret = qcow2_update_header(bs);
5364     if (ret < 0) {
5365         error_setg_errno(errp, -ret, "Failed to update the image header");
5366         return ret;
5367     }
5368 
5369     if (backing_file || backing_format) {
5370         ret = qcow2_change_backing_file(bs,
5371                     backing_file ?: s->image_backing_file,
5372                     backing_format ?: s->image_backing_format);
5373         if (ret < 0) {
5374             error_setg_errno(errp, -ret, "Failed to change the backing file");
5375             return ret;
5376         }
5377     }
5378 
5379     if (s->use_lazy_refcounts != lazy_refcounts) {
5380         if (lazy_refcounts) {
5381             if (new_version < 3) {
5382                 error_setg(errp, "Lazy refcounts only supported with "
5383                            "compatibility level 1.1 and above (use compat=1.1 "
5384                            "or greater)");
5385                 return -EINVAL;
5386             }
5387             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5388             ret = qcow2_update_header(bs);
5389             if (ret < 0) {
5390                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5391                 error_setg_errno(errp, -ret, "Failed to update the image header");
5392                 return ret;
5393             }
5394             s->use_lazy_refcounts = true;
5395         } else {
5396             /* make image clean first */
5397             ret = qcow2_mark_clean(bs);
5398             if (ret < 0) {
5399                 error_setg_errno(errp, -ret, "Failed to make the image clean");
5400                 return ret;
5401             }
5402             /* now disallow lazy refcounts */
5403             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5404             ret = qcow2_update_header(bs);
5405             if (ret < 0) {
5406                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5407                 error_setg_errno(errp, -ret, "Failed to update the image header");
5408                 return ret;
5409             }
5410             s->use_lazy_refcounts = false;
5411         }
5412     }
5413 
5414     if (new_size) {
5415         BlockBackend *blk = blk_new(bdrv_get_aio_context(bs),
5416                                     BLK_PERM_RESIZE, BLK_PERM_ALL);
5417         ret = blk_insert_bs(blk, bs, errp);
5418         if (ret < 0) {
5419             blk_unref(blk);
5420             return ret;
5421         }
5422 
5423         /*
5424          * Amending image options should ensure that the image has
5425          * exactly the given new values, so pass exact=true here.
5426          */
5427         ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, 0, errp);
5428         blk_unref(blk);
5429         if (ret < 0) {
5430             return ret;
5431         }
5432     }
5433 
5434     /* Downgrade last (so unsupported features can be removed before) */
5435     if (new_version < old_version) {
5436         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
5437         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
5438                               &helper_cb_info, errp);
5439         if (ret < 0) {
5440             return ret;
5441         }
5442     }
5443 
5444     return 0;
5445 }
5446 
5447 /*
5448  * If offset or size are negative, respectively, they will not be included in
5449  * the BLOCK_IMAGE_CORRUPTED event emitted.
5450  * fatal will be ignored for read-only BDS; corruptions found there will always
5451  * be considered non-fatal.
5452  */
5453 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
5454                              int64_t size, const char *message_format, ...)
5455 {
5456     BDRVQcow2State *s = bs->opaque;
5457     const char *node_name;
5458     char *message;
5459     va_list ap;
5460 
5461     fatal = fatal && bdrv_is_writable(bs);
5462 
5463     if (s->signaled_corruption &&
5464         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
5465     {
5466         return;
5467     }
5468 
5469     va_start(ap, message_format);
5470     message = g_strdup_vprintf(message_format, ap);
5471     va_end(ap);
5472 
5473     if (fatal) {
5474         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
5475                 "corruption events will be suppressed\n", message);
5476     } else {
5477         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
5478                 "corruption events will be suppressed\n", message);
5479     }
5480 
5481     node_name = bdrv_get_node_name(bs);
5482     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
5483                                           *node_name != '\0', node_name,
5484                                           message, offset >= 0, offset,
5485                                           size >= 0, size,
5486                                           fatal);
5487     g_free(message);
5488 
5489     if (fatal) {
5490         qcow2_mark_corrupt(bs);
5491         bs->drv = NULL; /* make BDS unusable */
5492     }
5493 
5494     s->signaled_corruption = true;
5495 }
5496 
5497 static QemuOptsList qcow2_create_opts = {
5498     .name = "qcow2-create-opts",
5499     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
5500     .desc = {
5501         {
5502             .name = BLOCK_OPT_SIZE,
5503             .type = QEMU_OPT_SIZE,
5504             .help = "Virtual disk size"
5505         },
5506         {
5507             .name = BLOCK_OPT_COMPAT_LEVEL,
5508             .type = QEMU_OPT_STRING,
5509             .help = "Compatibility level (v2 [0.10] or v3 [1.1])"
5510         },
5511         {
5512             .name = BLOCK_OPT_BACKING_FILE,
5513             .type = QEMU_OPT_STRING,
5514             .help = "File name of a base image"
5515         },
5516         {
5517             .name = BLOCK_OPT_BACKING_FMT,
5518             .type = QEMU_OPT_STRING,
5519             .help = "Image format of the base image"
5520         },
5521         {
5522             .name = BLOCK_OPT_DATA_FILE,
5523             .type = QEMU_OPT_STRING,
5524             .help = "File name of an external data file"
5525         },
5526         {
5527             .name = BLOCK_OPT_DATA_FILE_RAW,
5528             .type = QEMU_OPT_BOOL,
5529             .help = "The external data file must stay valid as a raw image"
5530         },
5531         {
5532             .name = BLOCK_OPT_ENCRYPT,
5533             .type = QEMU_OPT_BOOL,
5534             .help = "Encrypt the image with format 'aes'. (Deprecated "
5535                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
5536         },
5537         {
5538             .name = BLOCK_OPT_ENCRYPT_FORMAT,
5539             .type = QEMU_OPT_STRING,
5540             .help = "Encrypt the image, format choices: 'aes', 'luks'",
5541         },
5542         BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
5543             "ID of secret providing qcow AES key or LUKS passphrase"),
5544         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
5545         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
5546         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
5547         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
5548         BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
5549         BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
5550         {
5551             .name = BLOCK_OPT_CLUSTER_SIZE,
5552             .type = QEMU_OPT_SIZE,
5553             .help = "qcow2 cluster size",
5554             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
5555         },
5556         {
5557             .name = BLOCK_OPT_PREALLOC,
5558             .type = QEMU_OPT_STRING,
5559             .help = "Preallocation mode (allowed values: off, metadata, "
5560                     "falloc, full)"
5561         },
5562         {
5563             .name = BLOCK_OPT_LAZY_REFCOUNTS,
5564             .type = QEMU_OPT_BOOL,
5565             .help = "Postpone refcount updates",
5566             .def_value_str = "off"
5567         },
5568         {
5569             .name = BLOCK_OPT_REFCOUNT_BITS,
5570             .type = QEMU_OPT_NUMBER,
5571             .help = "Width of a reference count entry in bits",
5572             .def_value_str = "16"
5573         },
5574         { /* end of list */ }
5575     }
5576 };
5577 
5578 static const char *const qcow2_strong_runtime_opts[] = {
5579     "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
5580 
5581     NULL
5582 };
5583 
5584 BlockDriver bdrv_qcow2 = {
5585     .format_name        = "qcow2",
5586     .instance_size      = sizeof(BDRVQcow2State),
5587     .bdrv_probe         = qcow2_probe,
5588     .bdrv_open          = qcow2_open,
5589     .bdrv_close         = qcow2_close,
5590     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
5591     .bdrv_reopen_commit   = qcow2_reopen_commit,
5592     .bdrv_reopen_commit_post = qcow2_reopen_commit_post,
5593     .bdrv_reopen_abort    = qcow2_reopen_abort,
5594     .bdrv_join_options    = qcow2_join_options,
5595     .bdrv_child_perm      = bdrv_format_default_perms,
5596     .bdrv_co_create_opts  = qcow2_co_create_opts,
5597     .bdrv_co_create       = qcow2_co_create,
5598     .bdrv_has_zero_init   = qcow2_has_zero_init,
5599     .bdrv_has_zero_init_truncate = bdrv_has_zero_init_1,
5600     .bdrv_co_block_status = qcow2_co_block_status,
5601 
5602     .bdrv_co_preadv_part    = qcow2_co_preadv_part,
5603     .bdrv_co_pwritev_part   = qcow2_co_pwritev_part,
5604     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
5605 
5606     .bdrv_co_pwrite_zeroes  = qcow2_co_pwrite_zeroes,
5607     .bdrv_co_pdiscard       = qcow2_co_pdiscard,
5608     .bdrv_co_copy_range_from = qcow2_co_copy_range_from,
5609     .bdrv_co_copy_range_to  = qcow2_co_copy_range_to,
5610     .bdrv_co_truncate       = qcow2_co_truncate,
5611     .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part,
5612     .bdrv_make_empty        = qcow2_make_empty,
5613 
5614     .bdrv_snapshot_create   = qcow2_snapshot_create,
5615     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
5616     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
5617     .bdrv_snapshot_list     = qcow2_snapshot_list,
5618     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
5619     .bdrv_measure           = qcow2_measure,
5620     .bdrv_get_info          = qcow2_get_info,
5621     .bdrv_get_specific_info = qcow2_get_specific_info,
5622 
5623     .bdrv_save_vmstate    = qcow2_save_vmstate,
5624     .bdrv_load_vmstate    = qcow2_load_vmstate,
5625 
5626     .supports_backing           = true,
5627     .bdrv_change_backing_file   = qcow2_change_backing_file,
5628 
5629     .bdrv_refresh_limits        = qcow2_refresh_limits,
5630     .bdrv_co_invalidate_cache   = qcow2_co_invalidate_cache,
5631     .bdrv_inactivate            = qcow2_inactivate,
5632 
5633     .create_opts         = &qcow2_create_opts,
5634     .strong_runtime_opts = qcow2_strong_runtime_opts,
5635     .mutable_opts        = mutable_opts,
5636     .bdrv_co_check       = qcow2_co_check,
5637     .bdrv_amend_options  = qcow2_amend_options,
5638 
5639     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
5640     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
5641 
5642     .bdrv_co_can_store_new_dirty_bitmap = qcow2_co_can_store_new_dirty_bitmap,
5643     .bdrv_co_remove_persistent_dirty_bitmap =
5644             qcow2_co_remove_persistent_dirty_bitmap,
5645 };
5646 
5647 static void bdrv_qcow2_init(void)
5648 {
5649     bdrv_register(&bdrv_qcow2);
5650 }
5651 
5652 block_init(bdrv_qcow2_init);
5653