xref: /openbmc/qemu/block/qcow2.c (revision d0c832f616cadd5bace44986824c0c9142913795)
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 
1730     /* Repair image if dirty */
1731     if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1732         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1733         BdrvCheckResult result = {0};
1734 
1735         ret = qcow2_co_check_locked(bs, &result,
1736                                     BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1737         if (ret < 0 || result.check_errors) {
1738             if (ret >= 0) {
1739                 ret = -EIO;
1740             }
1741             error_setg_errno(errp, -ret, "Could not repair dirty image");
1742             goto fail;
1743         }
1744     }
1745 
1746 #ifdef DEBUG_ALLOC
1747     {
1748         BdrvCheckResult result = {0};
1749         qcow2_check_refcounts(bs, &result, 0);
1750     }
1751 #endif
1752 
1753     qemu_co_queue_init(&s->thread_task_queue);
1754 
1755     return ret;
1756 
1757  fail:
1758     g_free(s->image_data_file);
1759     if (has_data_file(bs)) {
1760         bdrv_unref_child(bs, s->data_file);
1761     }
1762     g_free(s->unknown_header_fields);
1763     cleanup_unknown_header_ext(bs);
1764     qcow2_free_snapshots(bs);
1765     qcow2_refcount_close(bs);
1766     qemu_vfree(s->l1_table);
1767     /* else pre-write overlap checks in cache_destroy may crash */
1768     s->l1_table = NULL;
1769     cache_clean_timer_del(bs);
1770     if (s->l2_table_cache) {
1771         qcow2_cache_destroy(s->l2_table_cache);
1772     }
1773     if (s->refcount_block_cache) {
1774         qcow2_cache_destroy(s->refcount_block_cache);
1775     }
1776     qcrypto_block_free(s->crypto);
1777     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1778     return ret;
1779 }
1780 
1781 typedef struct QCow2OpenCo {
1782     BlockDriverState *bs;
1783     QDict *options;
1784     int flags;
1785     Error **errp;
1786     int ret;
1787 } QCow2OpenCo;
1788 
1789 static void coroutine_fn qcow2_open_entry(void *opaque)
1790 {
1791     QCow2OpenCo *qoc = opaque;
1792     BDRVQcow2State *s = qoc->bs->opaque;
1793 
1794     qemu_co_mutex_lock(&s->lock);
1795     qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp);
1796     qemu_co_mutex_unlock(&s->lock);
1797 }
1798 
1799 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1800                       Error **errp)
1801 {
1802     BDRVQcow2State *s = bs->opaque;
1803     QCow2OpenCo qoc = {
1804         .bs = bs,
1805         .options = options,
1806         .flags = flags,
1807         .errp = errp,
1808         .ret = -EINPROGRESS
1809     };
1810 
1811     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
1812                                false, errp);
1813     if (!bs->file) {
1814         return -EINVAL;
1815     }
1816 
1817     /* Initialise locks */
1818     qemu_co_mutex_init(&s->lock);
1819 
1820     if (qemu_in_coroutine()) {
1821         /* From bdrv_co_create.  */
1822         qcow2_open_entry(&qoc);
1823     } else {
1824         assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1825         qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc));
1826         BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
1827     }
1828     return qoc.ret;
1829 }
1830 
1831 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1832 {
1833     BDRVQcow2State *s = bs->opaque;
1834 
1835     if (bs->encrypted) {
1836         /* Encryption works on a sector granularity */
1837         bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto);
1838     }
1839     bs->bl.pwrite_zeroes_alignment = s->cluster_size;
1840     bs->bl.pdiscard_alignment = s->cluster_size;
1841 }
1842 
1843 static int qcow2_reopen_prepare(BDRVReopenState *state,
1844                                 BlockReopenQueue *queue, Error **errp)
1845 {
1846     Qcow2ReopenState *r;
1847     int ret;
1848 
1849     r = g_new0(Qcow2ReopenState, 1);
1850     state->opaque = r;
1851 
1852     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1853                                        state->flags, errp);
1854     if (ret < 0) {
1855         goto fail;
1856     }
1857 
1858     /* We need to write out any unwritten data if we reopen read-only. */
1859     if ((state->flags & BDRV_O_RDWR) == 0) {
1860         ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1861         if (ret < 0) {
1862             goto fail;
1863         }
1864 
1865         ret = bdrv_flush(state->bs);
1866         if (ret < 0) {
1867             goto fail;
1868         }
1869 
1870         ret = qcow2_mark_clean(state->bs);
1871         if (ret < 0) {
1872             goto fail;
1873         }
1874     }
1875 
1876     return 0;
1877 
1878 fail:
1879     qcow2_update_options_abort(state->bs, r);
1880     g_free(r);
1881     return ret;
1882 }
1883 
1884 static void qcow2_reopen_commit(BDRVReopenState *state)
1885 {
1886     qcow2_update_options_commit(state->bs, state->opaque);
1887     g_free(state->opaque);
1888 }
1889 
1890 static void qcow2_reopen_commit_post(BDRVReopenState *state)
1891 {
1892     if (state->flags & BDRV_O_RDWR) {
1893         Error *local_err = NULL;
1894 
1895         if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) {
1896             /*
1897              * This is not fatal, bitmaps just left read-only, so all following
1898              * writes will fail. User can remove read-only bitmaps to unblock
1899              * writes or retry reopen.
1900              */
1901             error_reportf_err(local_err,
1902                               "%s: Failed to make dirty bitmaps writable: ",
1903                               bdrv_get_node_name(state->bs));
1904         }
1905     }
1906 }
1907 
1908 static void qcow2_reopen_abort(BDRVReopenState *state)
1909 {
1910     qcow2_update_options_abort(state->bs, state->opaque);
1911     g_free(state->opaque);
1912 }
1913 
1914 static void qcow2_join_options(QDict *options, QDict *old_options)
1915 {
1916     bool has_new_overlap_template =
1917         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1918         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1919     bool has_new_total_cache_size =
1920         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1921     bool has_all_cache_options;
1922 
1923     /* New overlap template overrides all old overlap options */
1924     if (has_new_overlap_template) {
1925         qdict_del(old_options, QCOW2_OPT_OVERLAP);
1926         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1927         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1928         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1929         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1930         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1931         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1932         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1933         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1934         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1935     }
1936 
1937     /* New total cache size overrides all old options */
1938     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1939         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1940         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1941     }
1942 
1943     qdict_join(options, old_options, false);
1944 
1945     /*
1946      * If after merging all cache size options are set, an old total size is
1947      * overwritten. Do keep all options, however, if all three are new. The
1948      * resulting error message is what we want to happen.
1949      */
1950     has_all_cache_options =
1951         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1952         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1953         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1954 
1955     if (has_all_cache_options && !has_new_total_cache_size) {
1956         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1957     }
1958 }
1959 
1960 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
1961                                               bool want_zero,
1962                                               int64_t offset, int64_t count,
1963                                               int64_t *pnum, int64_t *map,
1964                                               BlockDriverState **file)
1965 {
1966     BDRVQcow2State *s = bs->opaque;
1967     uint64_t cluster_offset;
1968     unsigned int bytes;
1969     int ret, status = 0;
1970 
1971     qemu_co_mutex_lock(&s->lock);
1972 
1973     if (!s->metadata_preallocation_checked) {
1974         ret = qcow2_detect_metadata_preallocation(bs);
1975         s->metadata_preallocation = (ret == 1);
1976         s->metadata_preallocation_checked = true;
1977     }
1978 
1979     bytes = MIN(INT_MAX, count);
1980     ret = qcow2_get_cluster_offset(bs, offset, &bytes, &cluster_offset);
1981     qemu_co_mutex_unlock(&s->lock);
1982     if (ret < 0) {
1983         return ret;
1984     }
1985 
1986     *pnum = bytes;
1987 
1988     if ((ret == QCOW2_CLUSTER_NORMAL || ret == QCOW2_CLUSTER_ZERO_ALLOC) &&
1989         !s->crypto) {
1990         *map = cluster_offset | offset_into_cluster(s, offset);
1991         *file = s->data_file->bs;
1992         status |= BDRV_BLOCK_OFFSET_VALID;
1993     }
1994     if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) {
1995         status |= BDRV_BLOCK_ZERO;
1996     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1997         status |= BDRV_BLOCK_DATA;
1998     }
1999     if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) &&
2000         (status & BDRV_BLOCK_OFFSET_VALID))
2001     {
2002         status |= BDRV_BLOCK_RECURSE;
2003     }
2004     return status;
2005 }
2006 
2007 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs,
2008                                             QCowL2Meta **pl2meta,
2009                                             bool link_l2)
2010 {
2011     int ret = 0;
2012     QCowL2Meta *l2meta = *pl2meta;
2013 
2014     while (l2meta != NULL) {
2015         QCowL2Meta *next;
2016 
2017         if (link_l2) {
2018             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
2019             if (ret) {
2020                 goto out;
2021             }
2022         } else {
2023             qcow2_alloc_cluster_abort(bs, l2meta);
2024         }
2025 
2026         /* Take the request off the list of running requests */
2027         if (l2meta->nb_clusters != 0) {
2028             QLIST_REMOVE(l2meta, next_in_flight);
2029         }
2030 
2031         qemu_co_queue_restart_all(&l2meta->dependent_requests);
2032 
2033         next = l2meta->next;
2034         g_free(l2meta);
2035         l2meta = next;
2036     }
2037 out:
2038     *pl2meta = l2meta;
2039     return ret;
2040 }
2041 
2042 static coroutine_fn int
2043 qcow2_co_preadv_encrypted(BlockDriverState *bs,
2044                            uint64_t file_cluster_offset,
2045                            uint64_t offset,
2046                            uint64_t bytes,
2047                            QEMUIOVector *qiov,
2048                            uint64_t qiov_offset)
2049 {
2050     int ret;
2051     BDRVQcow2State *s = bs->opaque;
2052     uint8_t *buf;
2053 
2054     assert(bs->encrypted && s->crypto);
2055     assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2056 
2057     /*
2058      * For encrypted images, read everything into a temporary
2059      * contiguous buffer on which the AES functions can work.
2060      * Also, decryption in a separate buffer is better as it
2061      * prevents the guest from learning information about the
2062      * encrypted nature of the virtual disk.
2063      */
2064 
2065     buf = qemu_try_blockalign(s->data_file->bs, bytes);
2066     if (buf == NULL) {
2067         return -ENOMEM;
2068     }
2069 
2070     BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2071     ret = bdrv_co_pread(s->data_file,
2072                         file_cluster_offset + offset_into_cluster(s, offset),
2073                         bytes, buf, 0);
2074     if (ret < 0) {
2075         goto fail;
2076     }
2077 
2078     if (qcow2_co_decrypt(bs,
2079                          file_cluster_offset + offset_into_cluster(s, offset),
2080                          offset, buf, bytes) < 0)
2081     {
2082         ret = -EIO;
2083         goto fail;
2084     }
2085     qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes);
2086 
2087 fail:
2088     qemu_vfree(buf);
2089 
2090     return ret;
2091 }
2092 
2093 typedef struct Qcow2AioTask {
2094     AioTask task;
2095 
2096     BlockDriverState *bs;
2097     QCow2ClusterType cluster_type; /* only for read */
2098     uint64_t file_cluster_offset;
2099     uint64_t offset;
2100     uint64_t bytes;
2101     QEMUIOVector *qiov;
2102     uint64_t qiov_offset;
2103     QCowL2Meta *l2meta; /* only for write */
2104 } Qcow2AioTask;
2105 
2106 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task);
2107 static coroutine_fn int qcow2_add_task(BlockDriverState *bs,
2108                                        AioTaskPool *pool,
2109                                        AioTaskFunc func,
2110                                        QCow2ClusterType cluster_type,
2111                                        uint64_t file_cluster_offset,
2112                                        uint64_t offset,
2113                                        uint64_t bytes,
2114                                        QEMUIOVector *qiov,
2115                                        size_t qiov_offset,
2116                                        QCowL2Meta *l2meta)
2117 {
2118     Qcow2AioTask local_task;
2119     Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task;
2120 
2121     *task = (Qcow2AioTask) {
2122         .task.func = func,
2123         .bs = bs,
2124         .cluster_type = cluster_type,
2125         .qiov = qiov,
2126         .file_cluster_offset = file_cluster_offset,
2127         .offset = offset,
2128         .bytes = bytes,
2129         .qiov_offset = qiov_offset,
2130         .l2meta = l2meta,
2131     };
2132 
2133     trace_qcow2_add_task(qemu_coroutine_self(), bs, pool,
2134                          func == qcow2_co_preadv_task_entry ? "read" : "write",
2135                          cluster_type, file_cluster_offset, offset, bytes,
2136                          qiov, qiov_offset);
2137 
2138     if (!pool) {
2139         return func(&task->task);
2140     }
2141 
2142     aio_task_pool_start_task(pool, &task->task);
2143 
2144     return 0;
2145 }
2146 
2147 static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs,
2148                                              QCow2ClusterType cluster_type,
2149                                              uint64_t file_cluster_offset,
2150                                              uint64_t offset, uint64_t bytes,
2151                                              QEMUIOVector *qiov,
2152                                              size_t qiov_offset)
2153 {
2154     BDRVQcow2State *s = bs->opaque;
2155     int offset_in_cluster = offset_into_cluster(s, offset);
2156 
2157     switch (cluster_type) {
2158     case QCOW2_CLUSTER_ZERO_PLAIN:
2159     case QCOW2_CLUSTER_ZERO_ALLOC:
2160         /* Both zero types are handled in qcow2_co_preadv_part */
2161         g_assert_not_reached();
2162 
2163     case QCOW2_CLUSTER_UNALLOCATED:
2164         assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */
2165 
2166         BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
2167         return bdrv_co_preadv_part(bs->backing, offset, bytes,
2168                                    qiov, qiov_offset, 0);
2169 
2170     case QCOW2_CLUSTER_COMPRESSED:
2171         return qcow2_co_preadv_compressed(bs, file_cluster_offset,
2172                                           offset, bytes, qiov, qiov_offset);
2173 
2174     case QCOW2_CLUSTER_NORMAL:
2175         assert(offset_into_cluster(s, file_cluster_offset) == 0);
2176         if (bs->encrypted) {
2177             return qcow2_co_preadv_encrypted(bs, file_cluster_offset,
2178                                              offset, bytes, qiov, qiov_offset);
2179         }
2180 
2181         BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2182         return bdrv_co_preadv_part(s->data_file,
2183                                    file_cluster_offset + offset_in_cluster,
2184                                    bytes, qiov, qiov_offset, 0);
2185 
2186     default:
2187         g_assert_not_reached();
2188     }
2189 
2190     g_assert_not_reached();
2191 }
2192 
2193 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task)
2194 {
2195     Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2196 
2197     assert(!t->l2meta);
2198 
2199     return qcow2_co_preadv_task(t->bs, t->cluster_type, t->file_cluster_offset,
2200                                 t->offset, t->bytes, t->qiov, t->qiov_offset);
2201 }
2202 
2203 static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs,
2204                                              uint64_t offset, uint64_t bytes,
2205                                              QEMUIOVector *qiov,
2206                                              size_t qiov_offset, int flags)
2207 {
2208     BDRVQcow2State *s = bs->opaque;
2209     int ret = 0;
2210     unsigned int cur_bytes; /* number of bytes in current iteration */
2211     uint64_t cluster_offset = 0;
2212     AioTaskPool *aio = NULL;
2213 
2214     while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2215         /* prepare next request */
2216         cur_bytes = MIN(bytes, INT_MAX);
2217         if (s->crypto) {
2218             cur_bytes = MIN(cur_bytes,
2219                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2220         }
2221 
2222         qemu_co_mutex_lock(&s->lock);
2223         ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset);
2224         qemu_co_mutex_unlock(&s->lock);
2225         if (ret < 0) {
2226             goto out;
2227         }
2228 
2229         if (ret == QCOW2_CLUSTER_ZERO_PLAIN ||
2230             ret == QCOW2_CLUSTER_ZERO_ALLOC ||
2231             (ret == QCOW2_CLUSTER_UNALLOCATED && !bs->backing))
2232         {
2233             qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes);
2234         } else {
2235             if (!aio && cur_bytes != bytes) {
2236                 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2237             }
2238             ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, ret,
2239                                  cluster_offset, offset, cur_bytes,
2240                                  qiov, qiov_offset, NULL);
2241             if (ret < 0) {
2242                 goto out;
2243             }
2244         }
2245 
2246         bytes -= cur_bytes;
2247         offset += cur_bytes;
2248         qiov_offset += cur_bytes;
2249     }
2250 
2251 out:
2252     if (aio) {
2253         aio_task_pool_wait_all(aio);
2254         if (ret == 0) {
2255             ret = aio_task_pool_status(aio);
2256         }
2257         g_free(aio);
2258     }
2259 
2260     return ret;
2261 }
2262 
2263 /* Check if it's possible to merge a write request with the writing of
2264  * the data from the COW regions */
2265 static bool merge_cow(uint64_t offset, unsigned bytes,
2266                       QEMUIOVector *qiov, size_t qiov_offset,
2267                       QCowL2Meta *l2meta)
2268 {
2269     QCowL2Meta *m;
2270 
2271     for (m = l2meta; m != NULL; m = m->next) {
2272         /* If both COW regions are empty then there's nothing to merge */
2273         if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
2274             continue;
2275         }
2276 
2277         /* If COW regions are handled already, skip this too */
2278         if (m->skip_cow) {
2279             continue;
2280         }
2281 
2282         /* The data (middle) region must be immediately after the
2283          * start region */
2284         if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
2285             continue;
2286         }
2287 
2288         /* The end region must be immediately after the data (middle)
2289          * region */
2290         if (m->offset + m->cow_end.offset != offset + bytes) {
2291             continue;
2292         }
2293 
2294         /* Make sure that adding both COW regions to the QEMUIOVector
2295          * does not exceed IOV_MAX */
2296         if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) {
2297             continue;
2298         }
2299 
2300         m->data_qiov = qiov;
2301         m->data_qiov_offset = qiov_offset;
2302         return true;
2303     }
2304 
2305     return false;
2306 }
2307 
2308 static bool is_unallocated(BlockDriverState *bs, int64_t offset, int64_t bytes)
2309 {
2310     int64_t nr;
2311     return !bytes ||
2312         (!bdrv_is_allocated_above(bs, NULL, false, offset, bytes, &nr) &&
2313          nr == bytes);
2314 }
2315 
2316 static bool is_zero_cow(BlockDriverState *bs, QCowL2Meta *m)
2317 {
2318     /*
2319      * This check is designed for optimization shortcut so it must be
2320      * efficient.
2321      * Instead of is_zero(), use is_unallocated() as it is faster (but not
2322      * as accurate and can result in false negatives).
2323      */
2324     return is_unallocated(bs, m->offset + m->cow_start.offset,
2325                           m->cow_start.nb_bytes) &&
2326            is_unallocated(bs, m->offset + m->cow_end.offset,
2327                           m->cow_end.nb_bytes);
2328 }
2329 
2330 static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
2331 {
2332     BDRVQcow2State *s = bs->opaque;
2333     QCowL2Meta *m;
2334 
2335     if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) {
2336         return 0;
2337     }
2338 
2339     if (bs->encrypted) {
2340         return 0;
2341     }
2342 
2343     for (m = l2meta; m != NULL; m = m->next) {
2344         int ret;
2345 
2346         if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
2347             continue;
2348         }
2349 
2350         if (!is_zero_cow(bs, m)) {
2351             continue;
2352         }
2353 
2354         /*
2355          * instead of writing zero COW buffers,
2356          * efficiently zero out the whole clusters
2357          */
2358 
2359         ret = qcow2_pre_write_overlap_check(bs, 0, m->alloc_offset,
2360                                             m->nb_clusters * s->cluster_size,
2361                                             true);
2362         if (ret < 0) {
2363             return ret;
2364         }
2365 
2366         BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE);
2367         ret = bdrv_co_pwrite_zeroes(s->data_file, m->alloc_offset,
2368                                     m->nb_clusters * s->cluster_size,
2369                                     BDRV_REQ_NO_FALLBACK);
2370         if (ret < 0) {
2371             if (ret != -ENOTSUP && ret != -EAGAIN) {
2372                 return ret;
2373             }
2374             continue;
2375         }
2376 
2377         trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters);
2378         m->skip_cow = true;
2379     }
2380     return 0;
2381 }
2382 
2383 /*
2384  * qcow2_co_pwritev_task
2385  * Called with s->lock unlocked
2386  * l2meta  - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2387  *           not use it somehow after qcow2_co_pwritev_task() call
2388  */
2389 static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs,
2390                                               uint64_t file_cluster_offset,
2391                                               uint64_t offset, uint64_t bytes,
2392                                               QEMUIOVector *qiov,
2393                                               uint64_t qiov_offset,
2394                                               QCowL2Meta *l2meta)
2395 {
2396     int ret;
2397     BDRVQcow2State *s = bs->opaque;
2398     void *crypt_buf = NULL;
2399     int offset_in_cluster = offset_into_cluster(s, offset);
2400     QEMUIOVector encrypted_qiov;
2401 
2402     if (bs->encrypted) {
2403         assert(s->crypto);
2404         assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2405         crypt_buf = qemu_try_blockalign(bs->file->bs, bytes);
2406         if (crypt_buf == NULL) {
2407             ret = -ENOMEM;
2408             goto out_unlocked;
2409         }
2410         qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes);
2411 
2412         if (qcow2_co_encrypt(bs, file_cluster_offset + offset_in_cluster,
2413                              offset, crypt_buf, bytes) < 0)
2414         {
2415             ret = -EIO;
2416             goto out_unlocked;
2417         }
2418 
2419         qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes);
2420         qiov = &encrypted_qiov;
2421         qiov_offset = 0;
2422     }
2423 
2424     /* Try to efficiently initialize the physical space with zeroes */
2425     ret = handle_alloc_space(bs, l2meta);
2426     if (ret < 0) {
2427         goto out_unlocked;
2428     }
2429 
2430     /*
2431      * If we need to do COW, check if it's possible to merge the
2432      * writing of the guest data together with that of the COW regions.
2433      * If it's not possible (or not necessary) then write the
2434      * guest data now.
2435      */
2436     if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) {
2437         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
2438         trace_qcow2_writev_data(qemu_coroutine_self(),
2439                                 file_cluster_offset + offset_in_cluster);
2440         ret = bdrv_co_pwritev_part(s->data_file,
2441                                    file_cluster_offset + offset_in_cluster,
2442                                    bytes, qiov, qiov_offset, 0);
2443         if (ret < 0) {
2444             goto out_unlocked;
2445         }
2446     }
2447 
2448     qemu_co_mutex_lock(&s->lock);
2449 
2450     ret = qcow2_handle_l2meta(bs, &l2meta, true);
2451     goto out_locked;
2452 
2453 out_unlocked:
2454     qemu_co_mutex_lock(&s->lock);
2455 
2456 out_locked:
2457     qcow2_handle_l2meta(bs, &l2meta, false);
2458     qemu_co_mutex_unlock(&s->lock);
2459 
2460     qemu_vfree(crypt_buf);
2461 
2462     return ret;
2463 }
2464 
2465 static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task)
2466 {
2467     Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2468 
2469     assert(!t->cluster_type);
2470 
2471     return qcow2_co_pwritev_task(t->bs, t->file_cluster_offset,
2472                                  t->offset, t->bytes, t->qiov, t->qiov_offset,
2473                                  t->l2meta);
2474 }
2475 
2476 static coroutine_fn int qcow2_co_pwritev_part(
2477         BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2478         QEMUIOVector *qiov, size_t qiov_offset, int flags)
2479 {
2480     BDRVQcow2State *s = bs->opaque;
2481     int offset_in_cluster;
2482     int ret;
2483     unsigned int cur_bytes; /* number of sectors in current iteration */
2484     uint64_t cluster_offset;
2485     QCowL2Meta *l2meta = NULL;
2486     AioTaskPool *aio = NULL;
2487 
2488     trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
2489 
2490     while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2491 
2492         l2meta = NULL;
2493 
2494         trace_qcow2_writev_start_part(qemu_coroutine_self());
2495         offset_in_cluster = offset_into_cluster(s, offset);
2496         cur_bytes = MIN(bytes, INT_MAX);
2497         if (bs->encrypted) {
2498             cur_bytes = MIN(cur_bytes,
2499                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
2500                             - offset_in_cluster);
2501         }
2502 
2503         qemu_co_mutex_lock(&s->lock);
2504 
2505         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2506                                          &cluster_offset, &l2meta);
2507         if (ret < 0) {
2508             goto out_locked;
2509         }
2510 
2511         assert(offset_into_cluster(s, cluster_offset) == 0);
2512 
2513         ret = qcow2_pre_write_overlap_check(bs, 0,
2514                                             cluster_offset + offset_in_cluster,
2515                                             cur_bytes, true);
2516         if (ret < 0) {
2517             goto out_locked;
2518         }
2519 
2520         qemu_co_mutex_unlock(&s->lock);
2521 
2522         if (!aio && cur_bytes != bytes) {
2523             aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2524         }
2525         ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0,
2526                              cluster_offset, offset, cur_bytes,
2527                              qiov, qiov_offset, l2meta);
2528         l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */
2529         if (ret < 0) {
2530             goto fail_nometa;
2531         }
2532 
2533         bytes -= cur_bytes;
2534         offset += cur_bytes;
2535         qiov_offset += cur_bytes;
2536         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2537     }
2538     ret = 0;
2539 
2540     qemu_co_mutex_lock(&s->lock);
2541 
2542 out_locked:
2543     qcow2_handle_l2meta(bs, &l2meta, false);
2544 
2545     qemu_co_mutex_unlock(&s->lock);
2546 
2547 fail_nometa:
2548     if (aio) {
2549         aio_task_pool_wait_all(aio);
2550         if (ret == 0) {
2551             ret = aio_task_pool_status(aio);
2552         }
2553         g_free(aio);
2554     }
2555 
2556     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2557 
2558     return ret;
2559 }
2560 
2561 static int qcow2_inactivate(BlockDriverState *bs)
2562 {
2563     BDRVQcow2State *s = bs->opaque;
2564     int ret, result = 0;
2565     Error *local_err = NULL;
2566 
2567     qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err);
2568     if (local_err != NULL) {
2569         result = -EINVAL;
2570         error_reportf_err(local_err, "Lost persistent bitmaps during "
2571                           "inactivation of node '%s': ",
2572                           bdrv_get_device_or_node_name(bs));
2573     }
2574 
2575     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2576     if (ret) {
2577         result = ret;
2578         error_report("Failed to flush the L2 table cache: %s",
2579                      strerror(-ret));
2580     }
2581 
2582     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2583     if (ret) {
2584         result = ret;
2585         error_report("Failed to flush the refcount block cache: %s",
2586                      strerror(-ret));
2587     }
2588 
2589     if (result == 0) {
2590         qcow2_mark_clean(bs);
2591     }
2592 
2593     return result;
2594 }
2595 
2596 static void qcow2_close(BlockDriverState *bs)
2597 {
2598     BDRVQcow2State *s = bs->opaque;
2599     qemu_vfree(s->l1_table);
2600     /* else pre-write overlap checks in cache_destroy may crash */
2601     s->l1_table = NULL;
2602 
2603     if (!(s->flags & BDRV_O_INACTIVE)) {
2604         qcow2_inactivate(bs);
2605     }
2606 
2607     cache_clean_timer_del(bs);
2608     qcow2_cache_destroy(s->l2_table_cache);
2609     qcow2_cache_destroy(s->refcount_block_cache);
2610 
2611     qcrypto_block_free(s->crypto);
2612     s->crypto = NULL;
2613 
2614     g_free(s->unknown_header_fields);
2615     cleanup_unknown_header_ext(bs);
2616 
2617     g_free(s->image_data_file);
2618     g_free(s->image_backing_file);
2619     g_free(s->image_backing_format);
2620 
2621     if (has_data_file(bs)) {
2622         bdrv_unref_child(bs, s->data_file);
2623     }
2624 
2625     qcow2_refcount_close(bs);
2626     qcow2_free_snapshots(bs);
2627 }
2628 
2629 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
2630                                                    Error **errp)
2631 {
2632     BDRVQcow2State *s = bs->opaque;
2633     int flags = s->flags;
2634     QCryptoBlock *crypto = NULL;
2635     QDict *options;
2636     Error *local_err = NULL;
2637     int ret;
2638 
2639     /*
2640      * Backing files are read-only which makes all of their metadata immutable,
2641      * that means we don't have to worry about reopening them here.
2642      */
2643 
2644     crypto = s->crypto;
2645     s->crypto = NULL;
2646 
2647     qcow2_close(bs);
2648 
2649     memset(s, 0, sizeof(BDRVQcow2State));
2650     options = qdict_clone_shallow(bs->options);
2651 
2652     flags &= ~BDRV_O_INACTIVE;
2653     qemu_co_mutex_lock(&s->lock);
2654     ret = qcow2_do_open(bs, options, flags, &local_err);
2655     qemu_co_mutex_unlock(&s->lock);
2656     qobject_unref(options);
2657     if (local_err) {
2658         error_propagate_prepend(errp, local_err,
2659                                 "Could not reopen qcow2 layer: ");
2660         bs->drv = NULL;
2661         return;
2662     } else if (ret < 0) {
2663         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
2664         bs->drv = NULL;
2665         return;
2666     }
2667 
2668     s->crypto = crypto;
2669 }
2670 
2671 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2672     size_t len, size_t buflen)
2673 {
2674     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2675     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2676 
2677     if (buflen < ext_len) {
2678         return -ENOSPC;
2679     }
2680 
2681     *ext_backing_fmt = (QCowExtension) {
2682         .magic  = cpu_to_be32(magic),
2683         .len    = cpu_to_be32(len),
2684     };
2685 
2686     if (len) {
2687         memcpy(buf + sizeof(QCowExtension), s, len);
2688     }
2689 
2690     return ext_len;
2691 }
2692 
2693 /*
2694  * Updates the qcow2 header, including the variable length parts of it, i.e.
2695  * the backing file name and all extensions. qcow2 was not designed to allow
2696  * such changes, so if we run out of space (we can only use the first cluster)
2697  * this function may fail.
2698  *
2699  * Returns 0 on success, -errno in error cases.
2700  */
2701 int qcow2_update_header(BlockDriverState *bs)
2702 {
2703     BDRVQcow2State *s = bs->opaque;
2704     QCowHeader *header;
2705     char *buf;
2706     size_t buflen = s->cluster_size;
2707     int ret;
2708     uint64_t total_size;
2709     uint32_t refcount_table_clusters;
2710     size_t header_length;
2711     Qcow2UnknownHeaderExtension *uext;
2712 
2713     buf = qemu_blockalign(bs, buflen);
2714 
2715     /* Header structure */
2716     header = (QCowHeader*) buf;
2717 
2718     if (buflen < sizeof(*header)) {
2719         ret = -ENOSPC;
2720         goto fail;
2721     }
2722 
2723     header_length = sizeof(*header) + s->unknown_header_fields_size;
2724     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2725     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2726 
2727     *header = (QCowHeader) {
2728         /* Version 2 fields */
2729         .magic                  = cpu_to_be32(QCOW_MAGIC),
2730         .version                = cpu_to_be32(s->qcow_version),
2731         .backing_file_offset    = 0,
2732         .backing_file_size      = 0,
2733         .cluster_bits           = cpu_to_be32(s->cluster_bits),
2734         .size                   = cpu_to_be64(total_size),
2735         .crypt_method           = cpu_to_be32(s->crypt_method_header),
2736         .l1_size                = cpu_to_be32(s->l1_size),
2737         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
2738         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
2739         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2740         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
2741         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
2742 
2743         /* Version 3 fields */
2744         .incompatible_features  = cpu_to_be64(s->incompatible_features),
2745         .compatible_features    = cpu_to_be64(s->compatible_features),
2746         .autoclear_features     = cpu_to_be64(s->autoclear_features),
2747         .refcount_order         = cpu_to_be32(s->refcount_order),
2748         .header_length          = cpu_to_be32(header_length),
2749     };
2750 
2751     /* For older versions, write a shorter header */
2752     switch (s->qcow_version) {
2753     case 2:
2754         ret = offsetof(QCowHeader, incompatible_features);
2755         break;
2756     case 3:
2757         ret = sizeof(*header);
2758         break;
2759     default:
2760         ret = -EINVAL;
2761         goto fail;
2762     }
2763 
2764     buf += ret;
2765     buflen -= ret;
2766     memset(buf, 0, buflen);
2767 
2768     /* Preserve any unknown field in the header */
2769     if (s->unknown_header_fields_size) {
2770         if (buflen < s->unknown_header_fields_size) {
2771             ret = -ENOSPC;
2772             goto fail;
2773         }
2774 
2775         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2776         buf += s->unknown_header_fields_size;
2777         buflen -= s->unknown_header_fields_size;
2778     }
2779 
2780     /* Backing file format header extension */
2781     if (s->image_backing_format) {
2782         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
2783                              s->image_backing_format,
2784                              strlen(s->image_backing_format),
2785                              buflen);
2786         if (ret < 0) {
2787             goto fail;
2788         }
2789 
2790         buf += ret;
2791         buflen -= ret;
2792     }
2793 
2794     /* External data file header extension */
2795     if (has_data_file(bs) && s->image_data_file) {
2796         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE,
2797                              s->image_data_file, strlen(s->image_data_file),
2798                              buflen);
2799         if (ret < 0) {
2800             goto fail;
2801         }
2802 
2803         buf += ret;
2804         buflen -= ret;
2805     }
2806 
2807     /* Full disk encryption header pointer extension */
2808     if (s->crypto_header.offset != 0) {
2809         s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset);
2810         s->crypto_header.length = cpu_to_be64(s->crypto_header.length);
2811         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
2812                              &s->crypto_header, sizeof(s->crypto_header),
2813                              buflen);
2814         s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
2815         s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
2816         if (ret < 0) {
2817             goto fail;
2818         }
2819         buf += ret;
2820         buflen -= ret;
2821     }
2822 
2823     /* Feature table */
2824     if (s->qcow_version >= 3) {
2825         Qcow2Feature features[] = {
2826             {
2827                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2828                 .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
2829                 .name = "dirty bit",
2830             },
2831             {
2832                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2833                 .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
2834                 .name = "corrupt bit",
2835             },
2836             {
2837                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2838                 .bit  = QCOW2_INCOMPAT_DATA_FILE_BITNR,
2839                 .name = "external data file",
2840             },
2841             {
2842                 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
2843                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
2844                 .name = "lazy refcounts",
2845             },
2846         };
2847 
2848         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
2849                              features, sizeof(features), buflen);
2850         if (ret < 0) {
2851             goto fail;
2852         }
2853         buf += ret;
2854         buflen -= ret;
2855     }
2856 
2857     /* Bitmap extension */
2858     if (s->nb_bitmaps > 0) {
2859         Qcow2BitmapHeaderExt bitmaps_header = {
2860             .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
2861             .bitmap_directory_size =
2862                     cpu_to_be64(s->bitmap_directory_size),
2863             .bitmap_directory_offset =
2864                     cpu_to_be64(s->bitmap_directory_offset)
2865         };
2866         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
2867                              &bitmaps_header, sizeof(bitmaps_header),
2868                              buflen);
2869         if (ret < 0) {
2870             goto fail;
2871         }
2872         buf += ret;
2873         buflen -= ret;
2874     }
2875 
2876     /* Keep unknown header extensions */
2877     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
2878         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
2879         if (ret < 0) {
2880             goto fail;
2881         }
2882 
2883         buf += ret;
2884         buflen -= ret;
2885     }
2886 
2887     /* End of header extensions */
2888     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
2889     if (ret < 0) {
2890         goto fail;
2891     }
2892 
2893     buf += ret;
2894     buflen -= ret;
2895 
2896     /* Backing file name */
2897     if (s->image_backing_file) {
2898         size_t backing_file_len = strlen(s->image_backing_file);
2899 
2900         if (buflen < backing_file_len) {
2901             ret = -ENOSPC;
2902             goto fail;
2903         }
2904 
2905         /* Using strncpy is ok here, since buf is not NUL-terminated. */
2906         strncpy(buf, s->image_backing_file, buflen);
2907 
2908         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
2909         header->backing_file_size   = cpu_to_be32(backing_file_len);
2910     }
2911 
2912     /* Write the new header */
2913     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
2914     if (ret < 0) {
2915         goto fail;
2916     }
2917 
2918     ret = 0;
2919 fail:
2920     qemu_vfree(header);
2921     return ret;
2922 }
2923 
2924 static int qcow2_change_backing_file(BlockDriverState *bs,
2925     const char *backing_file, const char *backing_fmt)
2926 {
2927     BDRVQcow2State *s = bs->opaque;
2928 
2929     /* Adding a backing file means that the external data file alone won't be
2930      * enough to make sense of the content */
2931     if (backing_file && data_file_is_raw(bs)) {
2932         return -EINVAL;
2933     }
2934 
2935     if (backing_file && strlen(backing_file) > 1023) {
2936         return -EINVAL;
2937     }
2938 
2939     pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2940             backing_file ?: "");
2941     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2942     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2943 
2944     g_free(s->image_backing_file);
2945     g_free(s->image_backing_format);
2946 
2947     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
2948     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
2949 
2950     return qcow2_update_header(bs);
2951 }
2952 
2953 static int qcow2_crypt_method_from_format(const char *encryptfmt)
2954 {
2955     if (g_str_equal(encryptfmt, "luks")) {
2956         return QCOW_CRYPT_LUKS;
2957     } else if (g_str_equal(encryptfmt, "aes")) {
2958         return QCOW_CRYPT_AES;
2959     } else {
2960         return -EINVAL;
2961     }
2962 }
2963 
2964 static int qcow2_set_up_encryption(BlockDriverState *bs,
2965                                    QCryptoBlockCreateOptions *cryptoopts,
2966                                    Error **errp)
2967 {
2968     BDRVQcow2State *s = bs->opaque;
2969     QCryptoBlock *crypto = NULL;
2970     int fmt, ret;
2971 
2972     switch (cryptoopts->format) {
2973     case Q_CRYPTO_BLOCK_FORMAT_LUKS:
2974         fmt = QCOW_CRYPT_LUKS;
2975         break;
2976     case Q_CRYPTO_BLOCK_FORMAT_QCOW:
2977         fmt = QCOW_CRYPT_AES;
2978         break;
2979     default:
2980         error_setg(errp, "Crypto format not supported in qcow2");
2981         return -EINVAL;
2982     }
2983 
2984     s->crypt_method_header = fmt;
2985 
2986     crypto = qcrypto_block_create(cryptoopts, "encrypt.",
2987                                   qcow2_crypto_hdr_init_func,
2988                                   qcow2_crypto_hdr_write_func,
2989                                   bs, errp);
2990     if (!crypto) {
2991         return -EINVAL;
2992     }
2993 
2994     ret = qcow2_update_header(bs);
2995     if (ret < 0) {
2996         error_setg_errno(errp, -ret, "Could not write encryption header");
2997         goto out;
2998     }
2999 
3000     ret = 0;
3001  out:
3002     qcrypto_block_free(crypto);
3003     return ret;
3004 }
3005 
3006 /**
3007  * Preallocates metadata structures for data clusters between @offset (in the
3008  * guest disk) and @new_length (which is thus generally the new guest disk
3009  * size).
3010  *
3011  * Returns: 0 on success, -errno on failure.
3012  */
3013 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset,
3014                                        uint64_t new_length, PreallocMode mode,
3015                                        Error **errp)
3016 {
3017     BDRVQcow2State *s = bs->opaque;
3018     uint64_t bytes;
3019     uint64_t host_offset = 0;
3020     int64_t file_length;
3021     unsigned int cur_bytes;
3022     int ret;
3023     QCowL2Meta *meta;
3024 
3025     assert(offset <= new_length);
3026     bytes = new_length - offset;
3027 
3028     while (bytes) {
3029         cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size));
3030         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
3031                                          &host_offset, &meta);
3032         if (ret < 0) {
3033             error_setg_errno(errp, -ret, "Allocating clusters failed");
3034             return ret;
3035         }
3036 
3037         while (meta) {
3038             QCowL2Meta *next = meta->next;
3039 
3040             ret = qcow2_alloc_cluster_link_l2(bs, meta);
3041             if (ret < 0) {
3042                 error_setg_errno(errp, -ret, "Mapping clusters failed");
3043                 qcow2_free_any_clusters(bs, meta->alloc_offset,
3044                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
3045                 return ret;
3046             }
3047 
3048             /* There are no dependent requests, but we need to remove our
3049              * request from the list of in-flight requests */
3050             QLIST_REMOVE(meta, next_in_flight);
3051 
3052             g_free(meta);
3053             meta = next;
3054         }
3055 
3056         /* TODO Preallocate data if requested */
3057 
3058         bytes -= cur_bytes;
3059         offset += cur_bytes;
3060     }
3061 
3062     /*
3063      * It is expected that the image file is large enough to actually contain
3064      * all of the allocated clusters (otherwise we get failing reads after
3065      * EOF). Extend the image to the last allocated sector.
3066      */
3067     file_length = bdrv_getlength(s->data_file->bs);
3068     if (file_length < 0) {
3069         error_setg_errno(errp, -file_length, "Could not get file size");
3070         return file_length;
3071     }
3072 
3073     if (host_offset + cur_bytes > file_length) {
3074         if (mode == PREALLOC_MODE_METADATA) {
3075             mode = PREALLOC_MODE_OFF;
3076         }
3077         ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false,
3078                                mode, errp);
3079         if (ret < 0) {
3080             return ret;
3081         }
3082     }
3083 
3084     return 0;
3085 }
3086 
3087 /* qcow2_refcount_metadata_size:
3088  * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3089  * @cluster_size: size of a cluster, in bytes
3090  * @refcount_order: refcount bits power-of-2 exponent
3091  * @generous_increase: allow for the refcount table to be 1.5x as large as it
3092  *                     needs to be
3093  *
3094  * Returns: Number of bytes required for refcount blocks and table metadata.
3095  */
3096 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
3097                                      int refcount_order, bool generous_increase,
3098                                      uint64_t *refblock_count)
3099 {
3100     /*
3101      * Every host cluster is reference-counted, including metadata (even
3102      * refcount metadata is recursively included).
3103      *
3104      * An accurate formula for the size of refcount metadata size is difficult
3105      * to derive.  An easier method of calculation is finding the fixed point
3106      * where no further refcount blocks or table clusters are required to
3107      * reference count every cluster.
3108      */
3109     int64_t blocks_per_table_cluster = cluster_size / sizeof(uint64_t);
3110     int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
3111     int64_t table = 0;  /* number of refcount table clusters */
3112     int64_t blocks = 0; /* number of refcount block clusters */
3113     int64_t last;
3114     int64_t n = 0;
3115 
3116     do {
3117         last = n;
3118         blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
3119         table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
3120         n = clusters + blocks + table;
3121 
3122         if (n == last && generous_increase) {
3123             clusters += DIV_ROUND_UP(table, 2);
3124             n = 0; /* force another loop */
3125             generous_increase = false;
3126         }
3127     } while (n != last);
3128 
3129     if (refblock_count) {
3130         *refblock_count = blocks;
3131     }
3132 
3133     return (blocks + table) * cluster_size;
3134 }
3135 
3136 /**
3137  * qcow2_calc_prealloc_size:
3138  * @total_size: virtual disk size in bytes
3139  * @cluster_size: cluster size in bytes
3140  * @refcount_order: refcount bits power-of-2 exponent
3141  *
3142  * Returns: Total number of bytes required for the fully allocated image
3143  * (including metadata).
3144  */
3145 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
3146                                         size_t cluster_size,
3147                                         int refcount_order)
3148 {
3149     int64_t meta_size = 0;
3150     uint64_t nl1e, nl2e;
3151     int64_t aligned_total_size = ROUND_UP(total_size, cluster_size);
3152 
3153     /* header: 1 cluster */
3154     meta_size += cluster_size;
3155 
3156     /* total size of L2 tables */
3157     nl2e = aligned_total_size / cluster_size;
3158     nl2e = ROUND_UP(nl2e, cluster_size / sizeof(uint64_t));
3159     meta_size += nl2e * sizeof(uint64_t);
3160 
3161     /* total size of L1 tables */
3162     nl1e = nl2e * sizeof(uint64_t) / cluster_size;
3163     nl1e = ROUND_UP(nl1e, cluster_size / sizeof(uint64_t));
3164     meta_size += nl1e * sizeof(uint64_t);
3165 
3166     /* total size of refcount table and blocks */
3167     meta_size += qcow2_refcount_metadata_size(
3168             (meta_size + aligned_total_size) / cluster_size,
3169             cluster_size, refcount_order, false, NULL);
3170 
3171     return meta_size + aligned_total_size;
3172 }
3173 
3174 static bool validate_cluster_size(size_t cluster_size, Error **errp)
3175 {
3176     int cluster_bits = ctz32(cluster_size);
3177     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
3178         (1 << cluster_bits) != cluster_size)
3179     {
3180         error_setg(errp, "Cluster size must be a power of two between %d and "
3181                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
3182         return false;
3183     }
3184     return true;
3185 }
3186 
3187 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, Error **errp)
3188 {
3189     size_t cluster_size;
3190 
3191     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
3192                                          DEFAULT_CLUSTER_SIZE);
3193     if (!validate_cluster_size(cluster_size, errp)) {
3194         return 0;
3195     }
3196     return cluster_size;
3197 }
3198 
3199 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
3200 {
3201     char *buf;
3202     int ret;
3203 
3204     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
3205     if (!buf) {
3206         ret = 3; /* default */
3207     } else if (!strcmp(buf, "0.10")) {
3208         ret = 2;
3209     } else if (!strcmp(buf, "1.1")) {
3210         ret = 3;
3211     } else {
3212         error_setg(errp, "Invalid compatibility level: '%s'", buf);
3213         ret = -EINVAL;
3214     }
3215     g_free(buf);
3216     return ret;
3217 }
3218 
3219 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
3220                                                 Error **errp)
3221 {
3222     uint64_t refcount_bits;
3223 
3224     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
3225     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
3226         error_setg(errp, "Refcount width must be a power of two and may not "
3227                    "exceed 64 bits");
3228         return 0;
3229     }
3230 
3231     if (version < 3 && refcount_bits != 16) {
3232         error_setg(errp, "Different refcount widths than 16 bits require "
3233                    "compatibility level 1.1 or above (use compat=1.1 or "
3234                    "greater)");
3235         return 0;
3236     }
3237 
3238     return refcount_bits;
3239 }
3240 
3241 static int coroutine_fn
3242 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
3243 {
3244     BlockdevCreateOptionsQcow2 *qcow2_opts;
3245     QDict *options;
3246 
3247     /*
3248      * Open the image file and write a minimal qcow2 header.
3249      *
3250      * We keep things simple and start with a zero-sized image. We also
3251      * do without refcount blocks or a L1 table for now. We'll fix the
3252      * inconsistency later.
3253      *
3254      * We do need a refcount table because growing the refcount table means
3255      * allocating two new refcount blocks - the seconds of which would be at
3256      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3257      * size for any qcow2 image.
3258      */
3259     BlockBackend *blk = NULL;
3260     BlockDriverState *bs = NULL;
3261     BlockDriverState *data_bs = NULL;
3262     QCowHeader *header;
3263     size_t cluster_size;
3264     int version;
3265     int refcount_order;
3266     uint64_t* refcount_table;
3267     Error *local_err = NULL;
3268     int ret;
3269 
3270     assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2);
3271     qcow2_opts = &create_options->u.qcow2;
3272 
3273     bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp);
3274     if (bs == NULL) {
3275         return -EIO;
3276     }
3277 
3278     /* Validate options and set default values */
3279     if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) {
3280         error_setg(errp, "Image size must be a multiple of %u bytes",
3281                    (unsigned) BDRV_SECTOR_SIZE);
3282         ret = -EINVAL;
3283         goto out;
3284     }
3285 
3286     if (qcow2_opts->has_version) {
3287         switch (qcow2_opts->version) {
3288         case BLOCKDEV_QCOW2_VERSION_V2:
3289             version = 2;
3290             break;
3291         case BLOCKDEV_QCOW2_VERSION_V3:
3292             version = 3;
3293             break;
3294         default:
3295             g_assert_not_reached();
3296         }
3297     } else {
3298         version = 3;
3299     }
3300 
3301     if (qcow2_opts->has_cluster_size) {
3302         cluster_size = qcow2_opts->cluster_size;
3303     } else {
3304         cluster_size = DEFAULT_CLUSTER_SIZE;
3305     }
3306 
3307     if (!validate_cluster_size(cluster_size, errp)) {
3308         ret = -EINVAL;
3309         goto out;
3310     }
3311 
3312     if (!qcow2_opts->has_preallocation) {
3313         qcow2_opts->preallocation = PREALLOC_MODE_OFF;
3314     }
3315     if (qcow2_opts->has_backing_file &&
3316         qcow2_opts->preallocation != PREALLOC_MODE_OFF)
3317     {
3318         error_setg(errp, "Backing file and preallocation cannot be used at "
3319                    "the same time");
3320         ret = -EINVAL;
3321         goto out;
3322     }
3323     if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) {
3324         error_setg(errp, "Backing format cannot be used without backing file");
3325         ret = -EINVAL;
3326         goto out;
3327     }
3328 
3329     if (!qcow2_opts->has_lazy_refcounts) {
3330         qcow2_opts->lazy_refcounts = false;
3331     }
3332     if (version < 3 && qcow2_opts->lazy_refcounts) {
3333         error_setg(errp, "Lazy refcounts only supported with compatibility "
3334                    "level 1.1 and above (use version=v3 or greater)");
3335         ret = -EINVAL;
3336         goto out;
3337     }
3338 
3339     if (!qcow2_opts->has_refcount_bits) {
3340         qcow2_opts->refcount_bits = 16;
3341     }
3342     if (qcow2_opts->refcount_bits > 64 ||
3343         !is_power_of_2(qcow2_opts->refcount_bits))
3344     {
3345         error_setg(errp, "Refcount width must be a power of two and may not "
3346                    "exceed 64 bits");
3347         ret = -EINVAL;
3348         goto out;
3349     }
3350     if (version < 3 && qcow2_opts->refcount_bits != 16) {
3351         error_setg(errp, "Different refcount widths than 16 bits require "
3352                    "compatibility level 1.1 or above (use version=v3 or "
3353                    "greater)");
3354         ret = -EINVAL;
3355         goto out;
3356     }
3357     refcount_order = ctz32(qcow2_opts->refcount_bits);
3358 
3359     if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) {
3360         error_setg(errp, "data-file-raw requires data-file");
3361         ret = -EINVAL;
3362         goto out;
3363     }
3364     if (qcow2_opts->data_file_raw && qcow2_opts->has_backing_file) {
3365         error_setg(errp, "Backing file and data-file-raw cannot be used at "
3366                    "the same time");
3367         ret = -EINVAL;
3368         goto out;
3369     }
3370 
3371     if (qcow2_opts->data_file) {
3372         if (version < 3) {
3373             error_setg(errp, "External data files are only supported with "
3374                        "compatibility level 1.1 and above (use version=v3 or "
3375                        "greater)");
3376             ret = -EINVAL;
3377             goto out;
3378         }
3379         data_bs = bdrv_open_blockdev_ref(qcow2_opts->data_file, errp);
3380         if (data_bs == NULL) {
3381             ret = -EIO;
3382             goto out;
3383         }
3384     }
3385 
3386     /* Create BlockBackend to write to the image */
3387     blk = blk_new(bdrv_get_aio_context(bs),
3388                   BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
3389     ret = blk_insert_bs(blk, bs, errp);
3390     if (ret < 0) {
3391         goto out;
3392     }
3393     blk_set_allow_write_beyond_eof(blk, true);
3394 
3395     /* Write the header */
3396     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
3397     header = g_malloc0(cluster_size);
3398     *header = (QCowHeader) {
3399         .magic                      = cpu_to_be32(QCOW_MAGIC),
3400         .version                    = cpu_to_be32(version),
3401         .cluster_bits               = cpu_to_be32(ctz32(cluster_size)),
3402         .size                       = cpu_to_be64(0),
3403         .l1_table_offset            = cpu_to_be64(0),
3404         .l1_size                    = cpu_to_be32(0),
3405         .refcount_table_offset      = cpu_to_be64(cluster_size),
3406         .refcount_table_clusters    = cpu_to_be32(1),
3407         .refcount_order             = cpu_to_be32(refcount_order),
3408         .header_length              = cpu_to_be32(sizeof(*header)),
3409     };
3410 
3411     /* We'll update this to correct value later */
3412     header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
3413 
3414     if (qcow2_opts->lazy_refcounts) {
3415         header->compatible_features |=
3416             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
3417     }
3418     if (data_bs) {
3419         header->incompatible_features |=
3420             cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE);
3421     }
3422     if (qcow2_opts->data_file_raw) {
3423         header->autoclear_features |=
3424             cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW);
3425     }
3426 
3427     ret = blk_pwrite(blk, 0, header, cluster_size, 0);
3428     g_free(header);
3429     if (ret < 0) {
3430         error_setg_errno(errp, -ret, "Could not write qcow2 header");
3431         goto out;
3432     }
3433 
3434     /* Write a refcount table with one refcount block */
3435     refcount_table = g_malloc0(2 * cluster_size);
3436     refcount_table[0] = cpu_to_be64(2 * cluster_size);
3437     ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
3438     g_free(refcount_table);
3439 
3440     if (ret < 0) {
3441         error_setg_errno(errp, -ret, "Could not write refcount table");
3442         goto out;
3443     }
3444 
3445     blk_unref(blk);
3446     blk = NULL;
3447 
3448     /*
3449      * And now open the image and make it consistent first (i.e. increase the
3450      * refcount of the cluster that is occupied by the header and the refcount
3451      * table)
3452      */
3453     options = qdict_new();
3454     qdict_put_str(options, "driver", "qcow2");
3455     qdict_put_str(options, "file", bs->node_name);
3456     if (data_bs) {
3457         qdict_put_str(options, "data-file", data_bs->node_name);
3458     }
3459     blk = blk_new_open(NULL, NULL, options,
3460                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
3461                        &local_err);
3462     if (blk == NULL) {
3463         error_propagate(errp, local_err);
3464         ret = -EIO;
3465         goto out;
3466     }
3467 
3468     ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
3469     if (ret < 0) {
3470         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
3471                          "header and refcount table");
3472         goto out;
3473 
3474     } else if (ret != 0) {
3475         error_report("Huh, first cluster in empty image is already in use?");
3476         abort();
3477     }
3478 
3479     /* Set the external data file if necessary */
3480     if (data_bs) {
3481         BDRVQcow2State *s = blk_bs(blk)->opaque;
3482         s->image_data_file = g_strdup(data_bs->filename);
3483     }
3484 
3485     /* Create a full header (including things like feature table) */
3486     ret = qcow2_update_header(blk_bs(blk));
3487     if (ret < 0) {
3488         error_setg_errno(errp, -ret, "Could not update qcow2 header");
3489         goto out;
3490     }
3491 
3492     /* Okay, now that we have a valid image, let's give it the right size */
3493     ret = blk_truncate(blk, qcow2_opts->size, false, qcow2_opts->preallocation,
3494                        errp);
3495     if (ret < 0) {
3496         error_prepend(errp, "Could not resize image: ");
3497         goto out;
3498     }
3499 
3500     /* Want a backing file? There you go.*/
3501     if (qcow2_opts->has_backing_file) {
3502         const char *backing_format = NULL;
3503 
3504         if (qcow2_opts->has_backing_fmt) {
3505             backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt);
3506         }
3507 
3508         ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file,
3509                                        backing_format);
3510         if (ret < 0) {
3511             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
3512                              "with format '%s'", qcow2_opts->backing_file,
3513                              backing_format);
3514             goto out;
3515         }
3516     }
3517 
3518     /* Want encryption? There you go. */
3519     if (qcow2_opts->has_encrypt) {
3520         ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp);
3521         if (ret < 0) {
3522             goto out;
3523         }
3524     }
3525 
3526     blk_unref(blk);
3527     blk = NULL;
3528 
3529     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3530      * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3531      * have to setup decryption context. We're not doing any I/O on the top
3532      * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3533      * not have effect.
3534      */
3535     options = qdict_new();
3536     qdict_put_str(options, "driver", "qcow2");
3537     qdict_put_str(options, "file", bs->node_name);
3538     if (data_bs) {
3539         qdict_put_str(options, "data-file", data_bs->node_name);
3540     }
3541     blk = blk_new_open(NULL, NULL, options,
3542                        BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO,
3543                        &local_err);
3544     if (blk == NULL) {
3545         error_propagate(errp, local_err);
3546         ret = -EIO;
3547         goto out;
3548     }
3549 
3550     ret = 0;
3551 out:
3552     blk_unref(blk);
3553     bdrv_unref(bs);
3554     bdrv_unref(data_bs);
3555     return ret;
3556 }
3557 
3558 static int coroutine_fn qcow2_co_create_opts(const char *filename, QemuOpts *opts,
3559                                              Error **errp)
3560 {
3561     BlockdevCreateOptions *create_options = NULL;
3562     QDict *qdict;
3563     Visitor *v;
3564     BlockDriverState *bs = NULL;
3565     BlockDriverState *data_bs = NULL;
3566     Error *local_err = NULL;
3567     const char *val;
3568     int ret;
3569 
3570     /* Only the keyval visitor supports the dotted syntax needed for
3571      * encryption, so go through a QDict before getting a QAPI type. Ignore
3572      * options meant for the protocol layer so that the visitor doesn't
3573      * complain. */
3574     qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts,
3575                                         true);
3576 
3577     /* Handle encryption options */
3578     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
3579     if (val && !strcmp(val, "on")) {
3580         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
3581     } else if (val && !strcmp(val, "off")) {
3582         qdict_del(qdict, BLOCK_OPT_ENCRYPT);
3583     }
3584 
3585     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
3586     if (val && !strcmp(val, "aes")) {
3587         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
3588     }
3589 
3590     /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3591      * version=v2/v3 below. */
3592     val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL);
3593     if (val && !strcmp(val, "0.10")) {
3594         qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2");
3595     } else if (val && !strcmp(val, "1.1")) {
3596         qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3");
3597     }
3598 
3599     /* Change legacy command line options into QMP ones */
3600     static const QDictRenames opt_renames[] = {
3601         { BLOCK_OPT_BACKING_FILE,       "backing-file" },
3602         { BLOCK_OPT_BACKING_FMT,        "backing-fmt" },
3603         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
3604         { BLOCK_OPT_LAZY_REFCOUNTS,     "lazy-refcounts" },
3605         { BLOCK_OPT_REFCOUNT_BITS,      "refcount-bits" },
3606         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
3607         { BLOCK_OPT_COMPAT_LEVEL,       "version" },
3608         { BLOCK_OPT_DATA_FILE_RAW,      "data-file-raw" },
3609         { NULL, NULL },
3610     };
3611 
3612     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
3613         ret = -EINVAL;
3614         goto finish;
3615     }
3616 
3617     /* Create and open the file (protocol layer) */
3618     ret = bdrv_create_file(filename, opts, errp);
3619     if (ret < 0) {
3620         goto finish;
3621     }
3622 
3623     bs = bdrv_open(filename, NULL, NULL,
3624                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
3625     if (bs == NULL) {
3626         ret = -EIO;
3627         goto finish;
3628     }
3629 
3630     /* Create and open an external data file (protocol layer) */
3631     val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE);
3632     if (val) {
3633         ret = bdrv_create_file(val, opts, errp);
3634         if (ret < 0) {
3635             goto finish;
3636         }
3637 
3638         data_bs = bdrv_open(val, NULL, NULL,
3639                             BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
3640                             errp);
3641         if (data_bs == NULL) {
3642             ret = -EIO;
3643             goto finish;
3644         }
3645 
3646         qdict_del(qdict, BLOCK_OPT_DATA_FILE);
3647         qdict_put_str(qdict, "data-file", data_bs->node_name);
3648     }
3649 
3650     /* Set 'driver' and 'node' options */
3651     qdict_put_str(qdict, "driver", "qcow2");
3652     qdict_put_str(qdict, "file", bs->node_name);
3653 
3654     /* Now get the QAPI type BlockdevCreateOptions */
3655     v = qobject_input_visitor_new_flat_confused(qdict, errp);
3656     if (!v) {
3657         ret = -EINVAL;
3658         goto finish;
3659     }
3660 
3661     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
3662     visit_free(v);
3663 
3664     if (local_err) {
3665         error_propagate(errp, local_err);
3666         ret = -EINVAL;
3667         goto finish;
3668     }
3669 
3670     /* Silently round up size */
3671     create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size,
3672                                             BDRV_SECTOR_SIZE);
3673 
3674     /* Create the qcow2 image (format layer) */
3675     ret = qcow2_co_create(create_options, errp);
3676     if (ret < 0) {
3677         goto finish;
3678     }
3679 
3680     ret = 0;
3681 finish:
3682     qobject_unref(qdict);
3683     bdrv_unref(bs);
3684     bdrv_unref(data_bs);
3685     qapi_free_BlockdevCreateOptions(create_options);
3686     return ret;
3687 }
3688 
3689 
3690 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes)
3691 {
3692     int64_t nr;
3693     int res;
3694 
3695     /* Clamp to image length, before checking status of underlying sectors */
3696     if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
3697         bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset;
3698     }
3699 
3700     if (!bytes) {
3701         return true;
3702     }
3703     res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL);
3704     return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == bytes;
3705 }
3706 
3707 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
3708     int64_t offset, int bytes, BdrvRequestFlags flags)
3709 {
3710     int ret;
3711     BDRVQcow2State *s = bs->opaque;
3712 
3713     uint32_t head = offset % s->cluster_size;
3714     uint32_t tail = (offset + bytes) % s->cluster_size;
3715 
3716     trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes);
3717     if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) {
3718         tail = 0;
3719     }
3720 
3721     if (head || tail) {
3722         uint64_t off;
3723         unsigned int nr;
3724 
3725         assert(head + bytes <= s->cluster_size);
3726 
3727         /* check whether remainder of cluster already reads as zero */
3728         if (!(is_zero(bs, offset - head, head) &&
3729               is_zero(bs, offset + bytes,
3730                       tail ? s->cluster_size - tail : 0))) {
3731             return -ENOTSUP;
3732         }
3733 
3734         qemu_co_mutex_lock(&s->lock);
3735         /* We can have new write after previous check */
3736         offset = QEMU_ALIGN_DOWN(offset, s->cluster_size);
3737         bytes = s->cluster_size;
3738         nr = s->cluster_size;
3739         ret = qcow2_get_cluster_offset(bs, offset, &nr, &off);
3740         if (ret != QCOW2_CLUSTER_UNALLOCATED &&
3741             ret != QCOW2_CLUSTER_ZERO_PLAIN &&
3742             ret != QCOW2_CLUSTER_ZERO_ALLOC) {
3743             qemu_co_mutex_unlock(&s->lock);
3744             return -ENOTSUP;
3745         }
3746     } else {
3747         qemu_co_mutex_lock(&s->lock);
3748     }
3749 
3750     trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes);
3751 
3752     /* Whatever is left can use real zero clusters */
3753     ret = qcow2_cluster_zeroize(bs, offset, bytes, flags);
3754     qemu_co_mutex_unlock(&s->lock);
3755 
3756     return ret;
3757 }
3758 
3759 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
3760                                           int64_t offset, int bytes)
3761 {
3762     int ret;
3763     BDRVQcow2State *s = bs->opaque;
3764 
3765     if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
3766         assert(bytes < s->cluster_size);
3767         /* Ignore partial clusters, except for the special case of the
3768          * complete partial cluster at the end of an unaligned file */
3769         if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
3770             offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) {
3771             return -ENOTSUP;
3772         }
3773     }
3774 
3775     qemu_co_mutex_lock(&s->lock);
3776     ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST,
3777                                 false);
3778     qemu_co_mutex_unlock(&s->lock);
3779     return ret;
3780 }
3781 
3782 static int coroutine_fn
3783 qcow2_co_copy_range_from(BlockDriverState *bs,
3784                          BdrvChild *src, uint64_t src_offset,
3785                          BdrvChild *dst, uint64_t dst_offset,
3786                          uint64_t bytes, BdrvRequestFlags read_flags,
3787                          BdrvRequestFlags write_flags)
3788 {
3789     BDRVQcow2State *s = bs->opaque;
3790     int ret;
3791     unsigned int cur_bytes; /* number of bytes in current iteration */
3792     BdrvChild *child = NULL;
3793     BdrvRequestFlags cur_write_flags;
3794 
3795     assert(!bs->encrypted);
3796     qemu_co_mutex_lock(&s->lock);
3797 
3798     while (bytes != 0) {
3799         uint64_t copy_offset = 0;
3800         /* prepare next request */
3801         cur_bytes = MIN(bytes, INT_MAX);
3802         cur_write_flags = write_flags;
3803 
3804         ret = qcow2_get_cluster_offset(bs, src_offset, &cur_bytes, &copy_offset);
3805         if (ret < 0) {
3806             goto out;
3807         }
3808 
3809         switch (ret) {
3810         case QCOW2_CLUSTER_UNALLOCATED:
3811             if (bs->backing && bs->backing->bs) {
3812                 int64_t backing_length = bdrv_getlength(bs->backing->bs);
3813                 if (src_offset >= backing_length) {
3814                     cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3815                 } else {
3816                     child = bs->backing;
3817                     cur_bytes = MIN(cur_bytes, backing_length - src_offset);
3818                     copy_offset = src_offset;
3819                 }
3820             } else {
3821                 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3822             }
3823             break;
3824 
3825         case QCOW2_CLUSTER_ZERO_PLAIN:
3826         case QCOW2_CLUSTER_ZERO_ALLOC:
3827             cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3828             break;
3829 
3830         case QCOW2_CLUSTER_COMPRESSED:
3831             ret = -ENOTSUP;
3832             goto out;
3833 
3834         case QCOW2_CLUSTER_NORMAL:
3835             child = s->data_file;
3836             copy_offset += offset_into_cluster(s, src_offset);
3837             break;
3838 
3839         default:
3840             abort();
3841         }
3842         qemu_co_mutex_unlock(&s->lock);
3843         ret = bdrv_co_copy_range_from(child,
3844                                       copy_offset,
3845                                       dst, dst_offset,
3846                                       cur_bytes, read_flags, cur_write_flags);
3847         qemu_co_mutex_lock(&s->lock);
3848         if (ret < 0) {
3849             goto out;
3850         }
3851 
3852         bytes -= cur_bytes;
3853         src_offset += cur_bytes;
3854         dst_offset += cur_bytes;
3855     }
3856     ret = 0;
3857 
3858 out:
3859     qemu_co_mutex_unlock(&s->lock);
3860     return ret;
3861 }
3862 
3863 static int coroutine_fn
3864 qcow2_co_copy_range_to(BlockDriverState *bs,
3865                        BdrvChild *src, uint64_t src_offset,
3866                        BdrvChild *dst, uint64_t dst_offset,
3867                        uint64_t bytes, BdrvRequestFlags read_flags,
3868                        BdrvRequestFlags write_flags)
3869 {
3870     BDRVQcow2State *s = bs->opaque;
3871     int offset_in_cluster;
3872     int ret;
3873     unsigned int cur_bytes; /* number of sectors in current iteration */
3874     uint64_t cluster_offset;
3875     QCowL2Meta *l2meta = NULL;
3876 
3877     assert(!bs->encrypted);
3878 
3879     qemu_co_mutex_lock(&s->lock);
3880 
3881     while (bytes != 0) {
3882 
3883         l2meta = NULL;
3884 
3885         offset_in_cluster = offset_into_cluster(s, dst_offset);
3886         cur_bytes = MIN(bytes, INT_MAX);
3887 
3888         /* TODO:
3889          * If src->bs == dst->bs, we could simply copy by incrementing
3890          * the refcnt, without copying user data.
3891          * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
3892         ret = qcow2_alloc_cluster_offset(bs, dst_offset, &cur_bytes,
3893                                          &cluster_offset, &l2meta);
3894         if (ret < 0) {
3895             goto fail;
3896         }
3897 
3898         assert(offset_into_cluster(s, cluster_offset) == 0);
3899 
3900         ret = qcow2_pre_write_overlap_check(bs, 0,
3901                 cluster_offset + offset_in_cluster, cur_bytes, true);
3902         if (ret < 0) {
3903             goto fail;
3904         }
3905 
3906         qemu_co_mutex_unlock(&s->lock);
3907         ret = bdrv_co_copy_range_to(src, src_offset,
3908                                     s->data_file,
3909                                     cluster_offset + offset_in_cluster,
3910                                     cur_bytes, read_flags, write_flags);
3911         qemu_co_mutex_lock(&s->lock);
3912         if (ret < 0) {
3913             goto fail;
3914         }
3915 
3916         ret = qcow2_handle_l2meta(bs, &l2meta, true);
3917         if (ret) {
3918             goto fail;
3919         }
3920 
3921         bytes -= cur_bytes;
3922         src_offset += cur_bytes;
3923         dst_offset += cur_bytes;
3924     }
3925     ret = 0;
3926 
3927 fail:
3928     qcow2_handle_l2meta(bs, &l2meta, false);
3929 
3930     qemu_co_mutex_unlock(&s->lock);
3931 
3932     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
3933 
3934     return ret;
3935 }
3936 
3937 static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,
3938                                           bool exact, PreallocMode prealloc,
3939                                           Error **errp)
3940 {
3941     BDRVQcow2State *s = bs->opaque;
3942     uint64_t old_length;
3943     int64_t new_l1_size;
3944     int ret;
3945     QDict *options;
3946 
3947     if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA &&
3948         prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL)
3949     {
3950         error_setg(errp, "Unsupported preallocation mode '%s'",
3951                    PreallocMode_str(prealloc));
3952         return -ENOTSUP;
3953     }
3954 
3955     if (!QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)) {
3956         error_setg(errp, "The new size must be a multiple of %u",
3957                    (unsigned) BDRV_SECTOR_SIZE);
3958         return -EINVAL;
3959     }
3960 
3961     qemu_co_mutex_lock(&s->lock);
3962 
3963     /* cannot proceed if image has snapshots */
3964     if (s->nb_snapshots) {
3965         error_setg(errp, "Can't resize an image which has snapshots");
3966         ret = -ENOTSUP;
3967         goto fail;
3968     }
3969 
3970     /* cannot proceed if image has bitmaps */
3971     if (qcow2_truncate_bitmaps_check(bs, errp)) {
3972         ret = -ENOTSUP;
3973         goto fail;
3974     }
3975 
3976     old_length = bs->total_sectors * BDRV_SECTOR_SIZE;
3977     new_l1_size = size_to_l1(s, offset);
3978 
3979     if (offset < old_length) {
3980         int64_t last_cluster, old_file_size;
3981         if (prealloc != PREALLOC_MODE_OFF) {
3982             error_setg(errp,
3983                        "Preallocation can't be used for shrinking an image");
3984             ret = -EINVAL;
3985             goto fail;
3986         }
3987 
3988         ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size),
3989                                     old_length - ROUND_UP(offset,
3990                                                           s->cluster_size),
3991                                     QCOW2_DISCARD_ALWAYS, true);
3992         if (ret < 0) {
3993             error_setg_errno(errp, -ret, "Failed to discard cropped clusters");
3994             goto fail;
3995         }
3996 
3997         ret = qcow2_shrink_l1_table(bs, new_l1_size);
3998         if (ret < 0) {
3999             error_setg_errno(errp, -ret,
4000                              "Failed to reduce the number of L2 tables");
4001             goto fail;
4002         }
4003 
4004         ret = qcow2_shrink_reftable(bs);
4005         if (ret < 0) {
4006             error_setg_errno(errp, -ret,
4007                              "Failed to discard unused refblocks");
4008             goto fail;
4009         }
4010 
4011         old_file_size = bdrv_getlength(bs->file->bs);
4012         if (old_file_size < 0) {
4013             error_setg_errno(errp, -old_file_size,
4014                              "Failed to inquire current file length");
4015             ret = old_file_size;
4016             goto fail;
4017         }
4018         last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4019         if (last_cluster < 0) {
4020             error_setg_errno(errp, -last_cluster,
4021                              "Failed to find the last cluster");
4022             ret = last_cluster;
4023             goto fail;
4024         }
4025         if ((last_cluster + 1) * s->cluster_size < old_file_size) {
4026             Error *local_err = NULL;
4027 
4028             /*
4029              * Do not pass @exact here: It will not help the user if
4030              * we get an error here just because they wanted to shrink
4031              * their qcow2 image (on a block device) with qemu-img.
4032              * (And on the qcow2 layer, the @exact requirement is
4033              * always fulfilled, so there is no need to pass it on.)
4034              */
4035             bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size,
4036                              false, PREALLOC_MODE_OFF, &local_err);
4037             if (local_err) {
4038                 warn_reportf_err(local_err,
4039                                  "Failed to truncate the tail of the image: ");
4040             }
4041         }
4042     } else {
4043         ret = qcow2_grow_l1_table(bs, new_l1_size, true);
4044         if (ret < 0) {
4045             error_setg_errno(errp, -ret, "Failed to grow the L1 table");
4046             goto fail;
4047         }
4048     }
4049 
4050     switch (prealloc) {
4051     case PREALLOC_MODE_OFF:
4052         if (has_data_file(bs)) {
4053             /*
4054              * If the caller wants an exact resize, the external data
4055              * file should be resized to the exact target size, too,
4056              * so we pass @exact here.
4057              */
4058             ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, errp);
4059             if (ret < 0) {
4060                 goto fail;
4061             }
4062         }
4063         break;
4064 
4065     case PREALLOC_MODE_METADATA:
4066         ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4067         if (ret < 0) {
4068             goto fail;
4069         }
4070         break;
4071 
4072     case PREALLOC_MODE_FALLOC:
4073     case PREALLOC_MODE_FULL:
4074     {
4075         int64_t allocation_start, host_offset, guest_offset;
4076         int64_t clusters_allocated;
4077         int64_t old_file_size, new_file_size;
4078         uint64_t nb_new_data_clusters, nb_new_l2_tables;
4079 
4080         /* With a data file, preallocation means just allocating the metadata
4081          * and forwarding the truncate request to the data file */
4082         if (has_data_file(bs)) {
4083             ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4084             if (ret < 0) {
4085                 goto fail;
4086             }
4087             break;
4088         }
4089 
4090         old_file_size = bdrv_getlength(bs->file->bs);
4091         if (old_file_size < 0) {
4092             error_setg_errno(errp, -old_file_size,
4093                              "Failed to inquire current file length");
4094             ret = old_file_size;
4095             goto fail;
4096         }
4097         old_file_size = ROUND_UP(old_file_size, s->cluster_size);
4098 
4099         nb_new_data_clusters = DIV_ROUND_UP(offset - old_length,
4100                                             s->cluster_size);
4101 
4102         /* This is an overestimation; we will not actually allocate space for
4103          * these in the file but just make sure the new refcount structures are
4104          * able to cover them so we will not have to allocate new refblocks
4105          * while entering the data blocks in the potentially new L2 tables.
4106          * (We do not actually care where the L2 tables are placed. Maybe they
4107          *  are already allocated or they can be placed somewhere before
4108          *  @old_file_size. It does not matter because they will be fully
4109          *  allocated automatically, so they do not need to be covered by the
4110          *  preallocation. All that matters is that we will not have to allocate
4111          *  new refcount structures for them.) */
4112         nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters,
4113                                         s->cluster_size / sizeof(uint64_t));
4114         /* The cluster range may not be aligned to L2 boundaries, so add one L2
4115          * table for a potential head/tail */
4116         nb_new_l2_tables++;
4117 
4118         allocation_start = qcow2_refcount_area(bs, old_file_size,
4119                                                nb_new_data_clusters +
4120                                                nb_new_l2_tables,
4121                                                true, 0, 0);
4122         if (allocation_start < 0) {
4123             error_setg_errno(errp, -allocation_start,
4124                              "Failed to resize refcount structures");
4125             ret = allocation_start;
4126             goto fail;
4127         }
4128 
4129         clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start,
4130                                                      nb_new_data_clusters);
4131         if (clusters_allocated < 0) {
4132             error_setg_errno(errp, -clusters_allocated,
4133                              "Failed to allocate data clusters");
4134             ret = clusters_allocated;
4135             goto fail;
4136         }
4137 
4138         assert(clusters_allocated == nb_new_data_clusters);
4139 
4140         /* Allocate the data area */
4141         new_file_size = allocation_start +
4142                         nb_new_data_clusters * s->cluster_size;
4143         /* Image file grows, so @exact does not matter */
4144         ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, errp);
4145         if (ret < 0) {
4146             error_prepend(errp, "Failed to resize underlying file: ");
4147             qcow2_free_clusters(bs, allocation_start,
4148                                 nb_new_data_clusters * s->cluster_size,
4149                                 QCOW2_DISCARD_OTHER);
4150             goto fail;
4151         }
4152 
4153         /* Create the necessary L2 entries */
4154         host_offset = allocation_start;
4155         guest_offset = old_length;
4156         while (nb_new_data_clusters) {
4157             int64_t nb_clusters = MIN(
4158                 nb_new_data_clusters,
4159                 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset));
4160             QCowL2Meta allocation = {
4161                 .offset       = guest_offset,
4162                 .alloc_offset = host_offset,
4163                 .nb_clusters  = nb_clusters,
4164             };
4165             qemu_co_queue_init(&allocation.dependent_requests);
4166 
4167             ret = qcow2_alloc_cluster_link_l2(bs, &allocation);
4168             if (ret < 0) {
4169                 error_setg_errno(errp, -ret, "Failed to update L2 tables");
4170                 qcow2_free_clusters(bs, host_offset,
4171                                     nb_new_data_clusters * s->cluster_size,
4172                                     QCOW2_DISCARD_OTHER);
4173                 goto fail;
4174             }
4175 
4176             guest_offset += nb_clusters * s->cluster_size;
4177             host_offset += nb_clusters * s->cluster_size;
4178             nb_new_data_clusters -= nb_clusters;
4179         }
4180         break;
4181     }
4182 
4183     default:
4184         g_assert_not_reached();
4185     }
4186 
4187     if (prealloc != PREALLOC_MODE_OFF) {
4188         /* Flush metadata before actually changing the image size */
4189         ret = qcow2_write_caches(bs);
4190         if (ret < 0) {
4191             error_setg_errno(errp, -ret,
4192                              "Failed to flush the preallocated area to disk");
4193             goto fail;
4194         }
4195     }
4196 
4197     bs->total_sectors = offset / BDRV_SECTOR_SIZE;
4198 
4199     /* write updated header.size */
4200     offset = cpu_to_be64(offset);
4201     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
4202                            &offset, sizeof(uint64_t));
4203     if (ret < 0) {
4204         error_setg_errno(errp, -ret, "Failed to update the image size");
4205         goto fail;
4206     }
4207 
4208     s->l1_vm_state_index = new_l1_size;
4209 
4210     /* Update cache sizes */
4211     options = qdict_clone_shallow(bs->options);
4212     ret = qcow2_update_options(bs, options, s->flags, errp);
4213     qobject_unref(options);
4214     if (ret < 0) {
4215         goto fail;
4216     }
4217     ret = 0;
4218 fail:
4219     qemu_co_mutex_unlock(&s->lock);
4220     return ret;
4221 }
4222 
4223 static coroutine_fn int
4224 qcow2_co_pwritev_compressed_task(BlockDriverState *bs,
4225                                  uint64_t offset, uint64_t bytes,
4226                                  QEMUIOVector *qiov, size_t qiov_offset)
4227 {
4228     BDRVQcow2State *s = bs->opaque;
4229     int ret;
4230     ssize_t out_len;
4231     uint8_t *buf, *out_buf;
4232     uint64_t cluster_offset;
4233 
4234     assert(bytes == s->cluster_size || (bytes < s->cluster_size &&
4235            (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS)));
4236 
4237     buf = qemu_blockalign(bs, s->cluster_size);
4238     if (bytes < s->cluster_size) {
4239         /* Zero-pad last write if image size is not cluster aligned */
4240         memset(buf + bytes, 0, s->cluster_size - bytes);
4241     }
4242     qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes);
4243 
4244     out_buf = g_malloc(s->cluster_size);
4245 
4246     out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1,
4247                                 buf, s->cluster_size);
4248     if (out_len == -ENOMEM) {
4249         /* could not compress: write normal cluster */
4250         ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0);
4251         if (ret < 0) {
4252             goto fail;
4253         }
4254         goto success;
4255     } else if (out_len < 0) {
4256         ret = -EINVAL;
4257         goto fail;
4258     }
4259 
4260     qemu_co_mutex_lock(&s->lock);
4261     ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len,
4262                                                 &cluster_offset);
4263     if (ret < 0) {
4264         qemu_co_mutex_unlock(&s->lock);
4265         goto fail;
4266     }
4267 
4268     ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true);
4269     qemu_co_mutex_unlock(&s->lock);
4270     if (ret < 0) {
4271         goto fail;
4272     }
4273 
4274     BLKDBG_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED);
4275     ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0);
4276     if (ret < 0) {
4277         goto fail;
4278     }
4279 success:
4280     ret = 0;
4281 fail:
4282     qemu_vfree(buf);
4283     g_free(out_buf);
4284     return ret;
4285 }
4286 
4287 static coroutine_fn int qcow2_co_pwritev_compressed_task_entry(AioTask *task)
4288 {
4289     Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
4290 
4291     assert(!t->cluster_type && !t->l2meta);
4292 
4293     return qcow2_co_pwritev_compressed_task(t->bs, t->offset, t->bytes, t->qiov,
4294                                             t->qiov_offset);
4295 }
4296 
4297 /*
4298  * XXX: put compressed sectors first, then all the cluster aligned
4299  * tables to avoid losing bytes in alignment
4300  */
4301 static coroutine_fn int
4302 qcow2_co_pwritev_compressed_part(BlockDriverState *bs,
4303                                  uint64_t offset, uint64_t bytes,
4304                                  QEMUIOVector *qiov, size_t qiov_offset)
4305 {
4306     BDRVQcow2State *s = bs->opaque;
4307     AioTaskPool *aio = NULL;
4308     int ret = 0;
4309 
4310     if (has_data_file(bs)) {
4311         return -ENOTSUP;
4312     }
4313 
4314     if (bytes == 0) {
4315         /*
4316          * align end of file to a sector boundary to ease reading with
4317          * sector based I/Os
4318          */
4319         int64_t len = bdrv_getlength(bs->file->bs);
4320         if (len < 0) {
4321             return len;
4322         }
4323         return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, NULL);
4324     }
4325 
4326     if (offset_into_cluster(s, offset)) {
4327         return -EINVAL;
4328     }
4329 
4330     while (bytes && aio_task_pool_status(aio) == 0) {
4331         uint64_t chunk_size = MIN(bytes, s->cluster_size);
4332 
4333         if (!aio && chunk_size != bytes) {
4334             aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
4335         }
4336 
4337         ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_compressed_task_entry,
4338                              0, 0, offset, chunk_size, qiov, qiov_offset, NULL);
4339         if (ret < 0) {
4340             break;
4341         }
4342         qiov_offset += chunk_size;
4343         offset += chunk_size;
4344         bytes -= chunk_size;
4345     }
4346 
4347     if (aio) {
4348         aio_task_pool_wait_all(aio);
4349         if (ret == 0) {
4350             ret = aio_task_pool_status(aio);
4351         }
4352         g_free(aio);
4353     }
4354 
4355     return ret;
4356 }
4357 
4358 static int coroutine_fn
4359 qcow2_co_preadv_compressed(BlockDriverState *bs,
4360                            uint64_t file_cluster_offset,
4361                            uint64_t offset,
4362                            uint64_t bytes,
4363                            QEMUIOVector *qiov,
4364                            size_t qiov_offset)
4365 {
4366     BDRVQcow2State *s = bs->opaque;
4367     int ret = 0, csize, nb_csectors;
4368     uint64_t coffset;
4369     uint8_t *buf, *out_buf;
4370     int offset_in_cluster = offset_into_cluster(s, offset);
4371 
4372     coffset = file_cluster_offset & s->cluster_offset_mask;
4373     nb_csectors = ((file_cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
4374     csize = nb_csectors * QCOW2_COMPRESSED_SECTOR_SIZE -
4375         (coffset & ~QCOW2_COMPRESSED_SECTOR_MASK);
4376 
4377     buf = g_try_malloc(csize);
4378     if (!buf) {
4379         return -ENOMEM;
4380     }
4381 
4382     out_buf = qemu_blockalign(bs, s->cluster_size);
4383 
4384     BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
4385     ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0);
4386     if (ret < 0) {
4387         goto fail;
4388     }
4389 
4390     if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) {
4391         ret = -EIO;
4392         goto fail;
4393     }
4394 
4395     qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes);
4396 
4397 fail:
4398     qemu_vfree(out_buf);
4399     g_free(buf);
4400 
4401     return ret;
4402 }
4403 
4404 static int make_completely_empty(BlockDriverState *bs)
4405 {
4406     BDRVQcow2State *s = bs->opaque;
4407     Error *local_err = NULL;
4408     int ret, l1_clusters;
4409     int64_t offset;
4410     uint64_t *new_reftable = NULL;
4411     uint64_t rt_entry, l1_size2;
4412     struct {
4413         uint64_t l1_offset;
4414         uint64_t reftable_offset;
4415         uint32_t reftable_clusters;
4416     } QEMU_PACKED l1_ofs_rt_ofs_cls;
4417 
4418     ret = qcow2_cache_empty(bs, s->l2_table_cache);
4419     if (ret < 0) {
4420         goto fail;
4421     }
4422 
4423     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
4424     if (ret < 0) {
4425         goto fail;
4426     }
4427 
4428     /* Refcounts will be broken utterly */
4429     ret = qcow2_mark_dirty(bs);
4430     if (ret < 0) {
4431         goto fail;
4432     }
4433 
4434     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4435 
4436     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
4437     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
4438 
4439     /* After this call, neither the in-memory nor the on-disk refcount
4440      * information accurately describe the actual references */
4441 
4442     ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
4443                              l1_clusters * s->cluster_size, 0);
4444     if (ret < 0) {
4445         goto fail_broken_refcounts;
4446     }
4447     memset(s->l1_table, 0, l1_size2);
4448 
4449     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
4450 
4451     /* Overwrite enough clusters at the beginning of the sectors to place
4452      * the refcount table, a refcount block and the L1 table in; this may
4453      * overwrite parts of the existing refcount and L1 table, which is not
4454      * an issue because the dirty flag is set, complete data loss is in fact
4455      * desired and partial data loss is consequently fine as well */
4456     ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
4457                              (2 + l1_clusters) * s->cluster_size, 0);
4458     /* This call (even if it failed overall) may have overwritten on-disk
4459      * refcount structures; in that case, the in-memory refcount information
4460      * will probably differ from the on-disk information which makes the BDS
4461      * unusable */
4462     if (ret < 0) {
4463         goto fail_broken_refcounts;
4464     }
4465 
4466     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4467     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
4468 
4469     /* "Create" an empty reftable (one cluster) directly after the image
4470      * header and an empty L1 table three clusters after the image header;
4471      * the cluster between those two will be used as the first refblock */
4472     l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
4473     l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
4474     l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
4475     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
4476                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
4477     if (ret < 0) {
4478         goto fail_broken_refcounts;
4479     }
4480 
4481     s->l1_table_offset = 3 * s->cluster_size;
4482 
4483     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
4484     if (!new_reftable) {
4485         ret = -ENOMEM;
4486         goto fail_broken_refcounts;
4487     }
4488 
4489     s->refcount_table_offset = s->cluster_size;
4490     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
4491     s->max_refcount_table_index = 0;
4492 
4493     g_free(s->refcount_table);
4494     s->refcount_table = new_reftable;
4495     new_reftable = NULL;
4496 
4497     /* Now the in-memory refcount information again corresponds to the on-disk
4498      * information (reftable is empty and no refblocks (the refblock cache is
4499      * empty)); however, this means some clusters (e.g. the image header) are
4500      * referenced, but not refcounted, but the normal qcow2 code assumes that
4501      * the in-memory information is always correct */
4502 
4503     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
4504 
4505     /* Enter the first refblock into the reftable */
4506     rt_entry = cpu_to_be64(2 * s->cluster_size);
4507     ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
4508                            &rt_entry, sizeof(rt_entry));
4509     if (ret < 0) {
4510         goto fail_broken_refcounts;
4511     }
4512     s->refcount_table[0] = 2 * s->cluster_size;
4513 
4514     s->free_cluster_index = 0;
4515     assert(3 + l1_clusters <= s->refcount_block_size);
4516     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
4517     if (offset < 0) {
4518         ret = offset;
4519         goto fail_broken_refcounts;
4520     } else if (offset > 0) {
4521         error_report("First cluster in emptied image is in use");
4522         abort();
4523     }
4524 
4525     /* Now finally the in-memory information corresponds to the on-disk
4526      * structures and is correct */
4527     ret = qcow2_mark_clean(bs);
4528     if (ret < 0) {
4529         goto fail;
4530     }
4531 
4532     ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false,
4533                         PREALLOC_MODE_OFF, &local_err);
4534     if (ret < 0) {
4535         error_report_err(local_err);
4536         goto fail;
4537     }
4538 
4539     return 0;
4540 
4541 fail_broken_refcounts:
4542     /* The BDS is unusable at this point. If we wanted to make it usable, we
4543      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4544      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4545      * again. However, because the functions which could have caused this error
4546      * path to be taken are used by those functions as well, it's very likely
4547      * that that sequence will fail as well. Therefore, just eject the BDS. */
4548     bs->drv = NULL;
4549 
4550 fail:
4551     g_free(new_reftable);
4552     return ret;
4553 }
4554 
4555 static int qcow2_make_empty(BlockDriverState *bs)
4556 {
4557     BDRVQcow2State *s = bs->opaque;
4558     uint64_t offset, end_offset;
4559     int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
4560     int l1_clusters, ret = 0;
4561 
4562     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
4563 
4564     if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps &&
4565         3 + l1_clusters <= s->refcount_block_size &&
4566         s->crypt_method_header != QCOW_CRYPT_LUKS &&
4567         !has_data_file(bs)) {
4568         /* The following function only works for qcow2 v3 images (it
4569          * requires the dirty flag) and only as long as there are no
4570          * features that reserve extra clusters (such as snapshots,
4571          * LUKS header, or persistent bitmaps), because it completely
4572          * empties the image.  Furthermore, the L1 table and three
4573          * additional clusters (image header, refcount table, one
4574          * refcount block) have to fit inside one refcount block. It
4575          * only resets the image file, i.e. does not work with an
4576          * external data file. */
4577         return make_completely_empty(bs);
4578     }
4579 
4580     /* This fallback code simply discards every active cluster; this is slow,
4581      * but works in all cases */
4582     end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
4583     for (offset = 0; offset < end_offset; offset += step) {
4584         /* As this function is generally used after committing an external
4585          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4586          * default action for this kind of discard is to pass the discard,
4587          * which will ideally result in an actually smaller image file, as
4588          * is probably desired. */
4589         ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
4590                                     QCOW2_DISCARD_SNAPSHOT, true);
4591         if (ret < 0) {
4592             break;
4593         }
4594     }
4595 
4596     return ret;
4597 }
4598 
4599 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
4600 {
4601     BDRVQcow2State *s = bs->opaque;
4602     int ret;
4603 
4604     qemu_co_mutex_lock(&s->lock);
4605     ret = qcow2_write_caches(bs);
4606     qemu_co_mutex_unlock(&s->lock);
4607 
4608     return ret;
4609 }
4610 
4611 static ssize_t qcow2_measure_crypto_hdr_init_func(QCryptoBlock *block,
4612         size_t headerlen, void *opaque, Error **errp)
4613 {
4614     size_t *headerlenp = opaque;
4615 
4616     /* Stash away the payload size */
4617     *headerlenp = headerlen;
4618     return 0;
4619 }
4620 
4621 static ssize_t qcow2_measure_crypto_hdr_write_func(QCryptoBlock *block,
4622         size_t offset, const uint8_t *buf, size_t buflen,
4623         void *opaque, Error **errp)
4624 {
4625     /* Discard the bytes, we're not actually writing to an image */
4626     return buflen;
4627 }
4628 
4629 /* Determine the number of bytes for the LUKS payload */
4630 static bool qcow2_measure_luks_headerlen(QemuOpts *opts, size_t *len,
4631                                          Error **errp)
4632 {
4633     QDict *opts_qdict;
4634     QDict *cryptoopts_qdict;
4635     QCryptoBlockCreateOptions *cryptoopts;
4636     QCryptoBlock *crypto;
4637 
4638     /* Extract "encrypt." options into a qdict */
4639     opts_qdict = qemu_opts_to_qdict(opts, NULL);
4640     qdict_extract_subqdict(opts_qdict, &cryptoopts_qdict, "encrypt.");
4641     qobject_unref(opts_qdict);
4642 
4643     /* Build QCryptoBlockCreateOptions object from qdict */
4644     qdict_put_str(cryptoopts_qdict, "format", "luks");
4645     cryptoopts = block_crypto_create_opts_init(cryptoopts_qdict, errp);
4646     qobject_unref(cryptoopts_qdict);
4647     if (!cryptoopts) {
4648         return false;
4649     }
4650 
4651     /* Fake LUKS creation in order to determine the payload size */
4652     crypto = qcrypto_block_create(cryptoopts, "encrypt.",
4653                                   qcow2_measure_crypto_hdr_init_func,
4654                                   qcow2_measure_crypto_hdr_write_func,
4655                                   len, errp);
4656     qapi_free_QCryptoBlockCreateOptions(cryptoopts);
4657     if (!crypto) {
4658         return false;
4659     }
4660 
4661     qcrypto_block_free(crypto);
4662     return true;
4663 }
4664 
4665 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs,
4666                                        Error **errp)
4667 {
4668     Error *local_err = NULL;
4669     BlockMeasureInfo *info;
4670     uint64_t required = 0; /* bytes that contribute to required size */
4671     uint64_t virtual_size; /* disk size as seen by guest */
4672     uint64_t refcount_bits;
4673     uint64_t l2_tables;
4674     uint64_t luks_payload_size = 0;
4675     size_t cluster_size;
4676     int version;
4677     char *optstr;
4678     PreallocMode prealloc;
4679     bool has_backing_file;
4680     bool has_luks;
4681 
4682     /* Parse image creation options */
4683     cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err);
4684     if (local_err) {
4685         goto err;
4686     }
4687 
4688     version = qcow2_opt_get_version_del(opts, &local_err);
4689     if (local_err) {
4690         goto err;
4691     }
4692 
4693     refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
4694     if (local_err) {
4695         goto err;
4696     }
4697 
4698     optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
4699     prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr,
4700                                PREALLOC_MODE_OFF, &local_err);
4701     g_free(optstr);
4702     if (local_err) {
4703         goto err;
4704     }
4705 
4706     optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
4707     has_backing_file = !!optstr;
4708     g_free(optstr);
4709 
4710     optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
4711     has_luks = optstr && strcmp(optstr, "luks") == 0;
4712     g_free(optstr);
4713 
4714     if (has_luks) {
4715         size_t headerlen;
4716 
4717         if (!qcow2_measure_luks_headerlen(opts, &headerlen, &local_err)) {
4718             goto err;
4719         }
4720 
4721         luks_payload_size = ROUND_UP(headerlen, cluster_size);
4722     }
4723 
4724     virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
4725     virtual_size = ROUND_UP(virtual_size, cluster_size);
4726 
4727     /* Check that virtual disk size is valid */
4728     l2_tables = DIV_ROUND_UP(virtual_size / cluster_size,
4729                              cluster_size / sizeof(uint64_t));
4730     if (l2_tables * sizeof(uint64_t) > QCOW_MAX_L1_SIZE) {
4731         error_setg(&local_err, "The image size is too large "
4732                                "(try using a larger cluster size)");
4733         goto err;
4734     }
4735 
4736     /* Account for input image */
4737     if (in_bs) {
4738         int64_t ssize = bdrv_getlength(in_bs);
4739         if (ssize < 0) {
4740             error_setg_errno(&local_err, -ssize,
4741                              "Unable to get image virtual_size");
4742             goto err;
4743         }
4744 
4745         virtual_size = ROUND_UP(ssize, cluster_size);
4746 
4747         if (has_backing_file) {
4748             /* We don't how much of the backing chain is shared by the input
4749              * image and the new image file.  In the worst case the new image's
4750              * backing file has nothing in common with the input image.  Be
4751              * conservative and assume all clusters need to be written.
4752              */
4753             required = virtual_size;
4754         } else {
4755             int64_t offset;
4756             int64_t pnum = 0;
4757 
4758             for (offset = 0; offset < ssize; offset += pnum) {
4759                 int ret;
4760 
4761                 ret = bdrv_block_status_above(in_bs, NULL, offset,
4762                                               ssize - offset, &pnum, NULL,
4763                                               NULL);
4764                 if (ret < 0) {
4765                     error_setg_errno(&local_err, -ret,
4766                                      "Unable to get block status");
4767                     goto err;
4768                 }
4769 
4770                 if (ret & BDRV_BLOCK_ZERO) {
4771                     /* Skip zero regions (safe with no backing file) */
4772                 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
4773                            (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
4774                     /* Extend pnum to end of cluster for next iteration */
4775                     pnum = ROUND_UP(offset + pnum, cluster_size) - offset;
4776 
4777                     /* Count clusters we've seen */
4778                     required += offset % cluster_size + pnum;
4779                 }
4780             }
4781         }
4782     }
4783 
4784     /* Take into account preallocation.  Nothing special is needed for
4785      * PREALLOC_MODE_METADATA since metadata is always counted.
4786      */
4787     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
4788         required = virtual_size;
4789     }
4790 
4791     info = g_new(BlockMeasureInfo, 1);
4792     info->fully_allocated =
4793         qcow2_calc_prealloc_size(virtual_size, cluster_size,
4794                                  ctz32(refcount_bits)) + luks_payload_size;
4795 
4796     /* Remove data clusters that are not required.  This overestimates the
4797      * required size because metadata needed for the fully allocated file is
4798      * still counted.
4799      */
4800     info->required = info->fully_allocated - virtual_size + required;
4801     return info;
4802 
4803 err:
4804     error_propagate(errp, local_err);
4805     return NULL;
4806 }
4807 
4808 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4809 {
4810     BDRVQcow2State *s = bs->opaque;
4811     bdi->unallocated_blocks_are_zero = true;
4812     bdi->cluster_size = s->cluster_size;
4813     bdi->vm_state_offset = qcow2_vm_state_offset(s);
4814     return 0;
4815 }
4816 
4817 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs,
4818                                                   Error **errp)
4819 {
4820     BDRVQcow2State *s = bs->opaque;
4821     ImageInfoSpecific *spec_info;
4822     QCryptoBlockInfo *encrypt_info = NULL;
4823     Error *local_err = NULL;
4824 
4825     if (s->crypto != NULL) {
4826         encrypt_info = qcrypto_block_get_info(s->crypto, &local_err);
4827         if (local_err) {
4828             error_propagate(errp, local_err);
4829             return NULL;
4830         }
4831     }
4832 
4833     spec_info = g_new(ImageInfoSpecific, 1);
4834     *spec_info = (ImageInfoSpecific){
4835         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
4836         .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1),
4837     };
4838     if (s->qcow_version == 2) {
4839         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
4840             .compat             = g_strdup("0.10"),
4841             .refcount_bits      = s->refcount_bits,
4842         };
4843     } else if (s->qcow_version == 3) {
4844         Qcow2BitmapInfoList *bitmaps;
4845         bitmaps = qcow2_get_bitmap_info_list(bs, &local_err);
4846         if (local_err) {
4847             error_propagate(errp, local_err);
4848             qapi_free_ImageInfoSpecific(spec_info);
4849             return NULL;
4850         }
4851         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
4852             .compat             = g_strdup("1.1"),
4853             .lazy_refcounts     = s->compatible_features &
4854                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
4855             .has_lazy_refcounts = true,
4856             .corrupt            = s->incompatible_features &
4857                                   QCOW2_INCOMPAT_CORRUPT,
4858             .has_corrupt        = true,
4859             .refcount_bits      = s->refcount_bits,
4860             .has_bitmaps        = !!bitmaps,
4861             .bitmaps            = bitmaps,
4862             .has_data_file      = !!s->image_data_file,
4863             .data_file          = g_strdup(s->image_data_file),
4864             .has_data_file_raw  = has_data_file(bs),
4865             .data_file_raw      = data_file_is_raw(bs),
4866         };
4867     } else {
4868         /* if this assertion fails, this probably means a new version was
4869          * added without having it covered here */
4870         assert(false);
4871     }
4872 
4873     if (encrypt_info) {
4874         ImageInfoSpecificQCow2Encryption *qencrypt =
4875             g_new(ImageInfoSpecificQCow2Encryption, 1);
4876         switch (encrypt_info->format) {
4877         case Q_CRYPTO_BLOCK_FORMAT_QCOW:
4878             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES;
4879             break;
4880         case Q_CRYPTO_BLOCK_FORMAT_LUKS:
4881             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS;
4882             qencrypt->u.luks = encrypt_info->u.luks;
4883             break;
4884         default:
4885             abort();
4886         }
4887         /* Since we did shallow copy above, erase any pointers
4888          * in the original info */
4889         memset(&encrypt_info->u, 0, sizeof(encrypt_info->u));
4890         qapi_free_QCryptoBlockInfo(encrypt_info);
4891 
4892         spec_info->u.qcow2.data->has_encrypt = true;
4893         spec_info->u.qcow2.data->encrypt = qencrypt;
4894     }
4895 
4896     return spec_info;
4897 }
4898 
4899 static int qcow2_has_zero_init(BlockDriverState *bs)
4900 {
4901     BDRVQcow2State *s = bs->opaque;
4902     bool preallocated;
4903 
4904     if (qemu_in_coroutine()) {
4905         qemu_co_mutex_lock(&s->lock);
4906     }
4907     /*
4908      * Check preallocation status: Preallocated images have all L2
4909      * tables allocated, nonpreallocated images have none.  It is
4910      * therefore enough to check the first one.
4911      */
4912     preallocated = s->l1_size > 0 && s->l1_table[0] != 0;
4913     if (qemu_in_coroutine()) {
4914         qemu_co_mutex_unlock(&s->lock);
4915     }
4916 
4917     if (!preallocated) {
4918         return 1;
4919     } else if (bs->encrypted) {
4920         return 0;
4921     } else {
4922         return bdrv_has_zero_init(s->data_file->bs);
4923     }
4924 }
4925 
4926 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
4927                               int64_t pos)
4928 {
4929     BDRVQcow2State *s = bs->opaque;
4930 
4931     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
4932     return bs->drv->bdrv_co_pwritev_part(bs, qcow2_vm_state_offset(s) + pos,
4933                                          qiov->size, qiov, 0, 0);
4934 }
4935 
4936 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
4937                               int64_t pos)
4938 {
4939     BDRVQcow2State *s = bs->opaque;
4940 
4941     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
4942     return bs->drv->bdrv_co_preadv_part(bs, qcow2_vm_state_offset(s) + pos,
4943                                         qiov->size, qiov, 0, 0);
4944 }
4945 
4946 /*
4947  * Downgrades an image's version. To achieve this, any incompatible features
4948  * have to be removed.
4949  */
4950 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
4951                            BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
4952                            Error **errp)
4953 {
4954     BDRVQcow2State *s = bs->opaque;
4955     int current_version = s->qcow_version;
4956     int ret;
4957 
4958     /* This is qcow2_downgrade(), not qcow2_upgrade() */
4959     assert(target_version < current_version);
4960 
4961     /* There are no other versions (now) that you can downgrade to */
4962     assert(target_version == 2);
4963 
4964     if (s->refcount_order != 4) {
4965         error_setg(errp, "compat=0.10 requires refcount_bits=16");
4966         return -ENOTSUP;
4967     }
4968 
4969     if (has_data_file(bs)) {
4970         error_setg(errp, "Cannot downgrade an image with a data file");
4971         return -ENOTSUP;
4972     }
4973 
4974     /* clear incompatible features */
4975     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
4976         ret = qcow2_mark_clean(bs);
4977         if (ret < 0) {
4978             error_setg_errno(errp, -ret, "Failed to make the image clean");
4979             return ret;
4980         }
4981     }
4982 
4983     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
4984      * the first place; if that happens nonetheless, returning -ENOTSUP is the
4985      * best thing to do anyway */
4986 
4987     if (s->incompatible_features) {
4988         error_setg(errp, "Cannot downgrade an image with incompatible features "
4989                    "%#" PRIx64 " set", s->incompatible_features);
4990         return -ENOTSUP;
4991     }
4992 
4993     /* since we can ignore compatible features, we can set them to 0 as well */
4994     s->compatible_features = 0;
4995     /* if lazy refcounts have been used, they have already been fixed through
4996      * clearing the dirty flag */
4997 
4998     /* clearing autoclear features is trivial */
4999     s->autoclear_features = 0;
5000 
5001     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
5002     if (ret < 0) {
5003         error_setg_errno(errp, -ret, "Failed to turn zero into data clusters");
5004         return ret;
5005     }
5006 
5007     s->qcow_version = target_version;
5008     ret = qcow2_update_header(bs);
5009     if (ret < 0) {
5010         s->qcow_version = current_version;
5011         error_setg_errno(errp, -ret, "Failed to update the image header");
5012         return ret;
5013     }
5014     return 0;
5015 }
5016 
5017 /*
5018  * Upgrades an image's version.  While newer versions encompass all
5019  * features of older versions, some things may have to be presented
5020  * differently.
5021  */
5022 static int qcow2_upgrade(BlockDriverState *bs, int target_version,
5023                          BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5024                          Error **errp)
5025 {
5026     BDRVQcow2State *s = bs->opaque;
5027     bool need_snapshot_update;
5028     int current_version = s->qcow_version;
5029     int i;
5030     int ret;
5031 
5032     /* This is qcow2_upgrade(), not qcow2_downgrade() */
5033     assert(target_version > current_version);
5034 
5035     /* There are no other versions (yet) that you can upgrade to */
5036     assert(target_version == 3);
5037 
5038     status_cb(bs, 0, 2, cb_opaque);
5039 
5040     /*
5041      * In v2, snapshots do not need to have extra data.  v3 requires
5042      * the 64-bit VM state size and the virtual disk size to be
5043      * present.
5044      * qcow2_write_snapshots() will always write the list in the
5045      * v3-compliant format.
5046      */
5047     need_snapshot_update = false;
5048     for (i = 0; i < s->nb_snapshots; i++) {
5049         if (s->snapshots[i].extra_data_size <
5050             sizeof_field(QCowSnapshotExtraData, vm_state_size_large) +
5051             sizeof_field(QCowSnapshotExtraData, disk_size))
5052         {
5053             need_snapshot_update = true;
5054             break;
5055         }
5056     }
5057     if (need_snapshot_update) {
5058         ret = qcow2_write_snapshots(bs);
5059         if (ret < 0) {
5060             error_setg_errno(errp, -ret, "Failed to update the snapshot table");
5061             return ret;
5062         }
5063     }
5064     status_cb(bs, 1, 2, cb_opaque);
5065 
5066     s->qcow_version = target_version;
5067     ret = qcow2_update_header(bs);
5068     if (ret < 0) {
5069         s->qcow_version = current_version;
5070         error_setg_errno(errp, -ret, "Failed to update the image header");
5071         return ret;
5072     }
5073     status_cb(bs, 2, 2, cb_opaque);
5074 
5075     return 0;
5076 }
5077 
5078 typedef enum Qcow2AmendOperation {
5079     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5080      * statically initialized to so that the helper CB can discern the first
5081      * invocation from an operation change */
5082     QCOW2_NO_OPERATION = 0,
5083 
5084     QCOW2_UPGRADING,
5085     QCOW2_CHANGING_REFCOUNT_ORDER,
5086     QCOW2_DOWNGRADING,
5087 } Qcow2AmendOperation;
5088 
5089 typedef struct Qcow2AmendHelperCBInfo {
5090     /* The code coordinating the amend operations should only modify
5091      * these four fields; the rest will be managed by the CB */
5092     BlockDriverAmendStatusCB *original_status_cb;
5093     void *original_cb_opaque;
5094 
5095     Qcow2AmendOperation current_operation;
5096 
5097     /* Total number of operations to perform (only set once) */
5098     int total_operations;
5099 
5100     /* The following fields are managed by the CB */
5101 
5102     /* Number of operations completed */
5103     int operations_completed;
5104 
5105     /* Cumulative offset of all completed operations */
5106     int64_t offset_completed;
5107 
5108     Qcow2AmendOperation last_operation;
5109     int64_t last_work_size;
5110 } Qcow2AmendHelperCBInfo;
5111 
5112 static void qcow2_amend_helper_cb(BlockDriverState *bs,
5113                                   int64_t operation_offset,
5114                                   int64_t operation_work_size, void *opaque)
5115 {
5116     Qcow2AmendHelperCBInfo *info = opaque;
5117     int64_t current_work_size;
5118     int64_t projected_work_size;
5119 
5120     if (info->current_operation != info->last_operation) {
5121         if (info->last_operation != QCOW2_NO_OPERATION) {
5122             info->offset_completed += info->last_work_size;
5123             info->operations_completed++;
5124         }
5125 
5126         info->last_operation = info->current_operation;
5127     }
5128 
5129     assert(info->total_operations > 0);
5130     assert(info->operations_completed < info->total_operations);
5131 
5132     info->last_work_size = operation_work_size;
5133 
5134     current_work_size = info->offset_completed + operation_work_size;
5135 
5136     /* current_work_size is the total work size for (operations_completed + 1)
5137      * operations (which includes this one), so multiply it by the number of
5138      * operations not covered and divide it by the number of operations
5139      * covered to get a projection for the operations not covered */
5140     projected_work_size = current_work_size * (info->total_operations -
5141                                                info->operations_completed - 1)
5142                                             / (info->operations_completed + 1);
5143 
5144     info->original_status_cb(bs, info->offset_completed + operation_offset,
5145                              current_work_size + projected_work_size,
5146                              info->original_cb_opaque);
5147 }
5148 
5149 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
5150                                BlockDriverAmendStatusCB *status_cb,
5151                                void *cb_opaque,
5152                                Error **errp)
5153 {
5154     BDRVQcow2State *s = bs->opaque;
5155     int old_version = s->qcow_version, new_version = old_version;
5156     uint64_t new_size = 0;
5157     const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL;
5158     bool lazy_refcounts = s->use_lazy_refcounts;
5159     bool data_file_raw = data_file_is_raw(bs);
5160     const char *compat = NULL;
5161     uint64_t cluster_size = s->cluster_size;
5162     bool encrypt;
5163     int encformat;
5164     int refcount_bits = s->refcount_bits;
5165     int ret;
5166     QemuOptDesc *desc = opts->list->desc;
5167     Qcow2AmendHelperCBInfo helper_cb_info;
5168 
5169     while (desc && desc->name) {
5170         if (!qemu_opt_find(opts, desc->name)) {
5171             /* only change explicitly defined options */
5172             desc++;
5173             continue;
5174         }
5175 
5176         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
5177             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
5178             if (!compat) {
5179                 /* preserve default */
5180             } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) {
5181                 new_version = 2;
5182             } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) {
5183                 new_version = 3;
5184             } else {
5185                 error_setg(errp, "Unknown compatibility level %s", compat);
5186                 return -EINVAL;
5187             }
5188         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
5189             error_setg(errp, "Cannot change preallocation mode");
5190             return -ENOTSUP;
5191         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
5192             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
5193         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
5194             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5195         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
5196             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5197         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
5198             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
5199                                         !!s->crypto);
5200 
5201             if (encrypt != !!s->crypto) {
5202                 error_setg(errp,
5203                            "Changing the encryption flag is not supported");
5204                 return -ENOTSUP;
5205             }
5206         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT_FORMAT)) {
5207             encformat = qcow2_crypt_method_from_format(
5208                 qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT));
5209 
5210             if (encformat != s->crypt_method_header) {
5211                 error_setg(errp,
5212                            "Changing the encryption format is not supported");
5213                 return -ENOTSUP;
5214             }
5215         } else if (g_str_has_prefix(desc->name, "encrypt.")) {
5216             error_setg(errp,
5217                        "Changing the encryption parameters is not supported");
5218             return -ENOTSUP;
5219         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
5220             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
5221                                              cluster_size);
5222             if (cluster_size != s->cluster_size) {
5223                 error_setg(errp, "Changing the cluster size is not supported");
5224                 return -ENOTSUP;
5225             }
5226         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
5227             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
5228                                                lazy_refcounts);
5229         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
5230             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
5231                                                 refcount_bits);
5232 
5233             if (refcount_bits <= 0 || refcount_bits > 64 ||
5234                 !is_power_of_2(refcount_bits))
5235             {
5236                 error_setg(errp, "Refcount width must be a power of two and "
5237                            "may not exceed 64 bits");
5238                 return -EINVAL;
5239             }
5240         } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) {
5241             data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE);
5242             if (data_file && !has_data_file(bs)) {
5243                 error_setg(errp, "data-file can only be set for images that "
5244                                  "use an external data file");
5245                 return -EINVAL;
5246             }
5247         } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) {
5248             data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW,
5249                                               data_file_raw);
5250             if (data_file_raw && !data_file_is_raw(bs)) {
5251                 error_setg(errp, "data-file-raw cannot be set on existing "
5252                                  "images");
5253                 return -EINVAL;
5254             }
5255         } else {
5256             /* if this point is reached, this probably means a new option was
5257              * added without having it covered here */
5258             abort();
5259         }
5260 
5261         desc++;
5262     }
5263 
5264     helper_cb_info = (Qcow2AmendHelperCBInfo){
5265         .original_status_cb = status_cb,
5266         .original_cb_opaque = cb_opaque,
5267         .total_operations = (new_version != old_version)
5268                           + (s->refcount_bits != refcount_bits)
5269     };
5270 
5271     /* Upgrade first (some features may require compat=1.1) */
5272     if (new_version > old_version) {
5273         helper_cb_info.current_operation = QCOW2_UPGRADING;
5274         ret = qcow2_upgrade(bs, new_version, &qcow2_amend_helper_cb,
5275                             &helper_cb_info, errp);
5276         if (ret < 0) {
5277             return ret;
5278         }
5279     }
5280 
5281     if (s->refcount_bits != refcount_bits) {
5282         int refcount_order = ctz32(refcount_bits);
5283 
5284         if (new_version < 3 && refcount_bits != 16) {
5285             error_setg(errp, "Refcount widths other than 16 bits require "
5286                        "compatibility level 1.1 or above (use compat=1.1 or "
5287                        "greater)");
5288             return -EINVAL;
5289         }
5290 
5291         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
5292         ret = qcow2_change_refcount_order(bs, refcount_order,
5293                                           &qcow2_amend_helper_cb,
5294                                           &helper_cb_info, errp);
5295         if (ret < 0) {
5296             return ret;
5297         }
5298     }
5299 
5300     /* data-file-raw blocks backing files, so clear it first if requested */
5301     if (data_file_raw) {
5302         s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5303     } else {
5304         s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5305     }
5306 
5307     if (data_file) {
5308         g_free(s->image_data_file);
5309         s->image_data_file = *data_file ? g_strdup(data_file) : NULL;
5310     }
5311 
5312     ret = qcow2_update_header(bs);
5313     if (ret < 0) {
5314         error_setg_errno(errp, -ret, "Failed to update the image header");
5315         return ret;
5316     }
5317 
5318     if (backing_file || backing_format) {
5319         ret = qcow2_change_backing_file(bs,
5320                     backing_file ?: s->image_backing_file,
5321                     backing_format ?: s->image_backing_format);
5322         if (ret < 0) {
5323             error_setg_errno(errp, -ret, "Failed to change the backing file");
5324             return ret;
5325         }
5326     }
5327 
5328     if (s->use_lazy_refcounts != lazy_refcounts) {
5329         if (lazy_refcounts) {
5330             if (new_version < 3) {
5331                 error_setg(errp, "Lazy refcounts only supported with "
5332                            "compatibility level 1.1 and above (use compat=1.1 "
5333                            "or greater)");
5334                 return -EINVAL;
5335             }
5336             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5337             ret = qcow2_update_header(bs);
5338             if (ret < 0) {
5339                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5340                 error_setg_errno(errp, -ret, "Failed to update the image header");
5341                 return ret;
5342             }
5343             s->use_lazy_refcounts = true;
5344         } else {
5345             /* make image clean first */
5346             ret = qcow2_mark_clean(bs);
5347             if (ret < 0) {
5348                 error_setg_errno(errp, -ret, "Failed to make the image clean");
5349                 return ret;
5350             }
5351             /* now disallow lazy refcounts */
5352             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5353             ret = qcow2_update_header(bs);
5354             if (ret < 0) {
5355                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5356                 error_setg_errno(errp, -ret, "Failed to update the image header");
5357                 return ret;
5358             }
5359             s->use_lazy_refcounts = false;
5360         }
5361     }
5362 
5363     if (new_size) {
5364         BlockBackend *blk = blk_new(bdrv_get_aio_context(bs),
5365                                     BLK_PERM_RESIZE, BLK_PERM_ALL);
5366         ret = blk_insert_bs(blk, bs, errp);
5367         if (ret < 0) {
5368             blk_unref(blk);
5369             return ret;
5370         }
5371 
5372         /*
5373          * Amending image options should ensure that the image has
5374          * exactly the given new values, so pass exact=true here.
5375          */
5376         ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, errp);
5377         blk_unref(blk);
5378         if (ret < 0) {
5379             return ret;
5380         }
5381     }
5382 
5383     /* Downgrade last (so unsupported features can be removed before) */
5384     if (new_version < old_version) {
5385         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
5386         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
5387                               &helper_cb_info, errp);
5388         if (ret < 0) {
5389             return ret;
5390         }
5391     }
5392 
5393     return 0;
5394 }
5395 
5396 /*
5397  * If offset or size are negative, respectively, they will not be included in
5398  * the BLOCK_IMAGE_CORRUPTED event emitted.
5399  * fatal will be ignored for read-only BDS; corruptions found there will always
5400  * be considered non-fatal.
5401  */
5402 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
5403                              int64_t size, const char *message_format, ...)
5404 {
5405     BDRVQcow2State *s = bs->opaque;
5406     const char *node_name;
5407     char *message;
5408     va_list ap;
5409 
5410     fatal = fatal && bdrv_is_writable(bs);
5411 
5412     if (s->signaled_corruption &&
5413         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
5414     {
5415         return;
5416     }
5417 
5418     va_start(ap, message_format);
5419     message = g_strdup_vprintf(message_format, ap);
5420     va_end(ap);
5421 
5422     if (fatal) {
5423         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
5424                 "corruption events will be suppressed\n", message);
5425     } else {
5426         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
5427                 "corruption events will be suppressed\n", message);
5428     }
5429 
5430     node_name = bdrv_get_node_name(bs);
5431     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
5432                                           *node_name != '\0', node_name,
5433                                           message, offset >= 0, offset,
5434                                           size >= 0, size,
5435                                           fatal);
5436     g_free(message);
5437 
5438     if (fatal) {
5439         qcow2_mark_corrupt(bs);
5440         bs->drv = NULL; /* make BDS unusable */
5441     }
5442 
5443     s->signaled_corruption = true;
5444 }
5445 
5446 static QemuOptsList qcow2_create_opts = {
5447     .name = "qcow2-create-opts",
5448     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
5449     .desc = {
5450         {
5451             .name = BLOCK_OPT_SIZE,
5452             .type = QEMU_OPT_SIZE,
5453             .help = "Virtual disk size"
5454         },
5455         {
5456             .name = BLOCK_OPT_COMPAT_LEVEL,
5457             .type = QEMU_OPT_STRING,
5458             .help = "Compatibility level (v2 [0.10] or v3 [1.1])"
5459         },
5460         {
5461             .name = BLOCK_OPT_BACKING_FILE,
5462             .type = QEMU_OPT_STRING,
5463             .help = "File name of a base image"
5464         },
5465         {
5466             .name = BLOCK_OPT_BACKING_FMT,
5467             .type = QEMU_OPT_STRING,
5468             .help = "Image format of the base image"
5469         },
5470         {
5471             .name = BLOCK_OPT_DATA_FILE,
5472             .type = QEMU_OPT_STRING,
5473             .help = "File name of an external data file"
5474         },
5475         {
5476             .name = BLOCK_OPT_DATA_FILE_RAW,
5477             .type = QEMU_OPT_BOOL,
5478             .help = "The external data file must stay valid as a raw image"
5479         },
5480         {
5481             .name = BLOCK_OPT_ENCRYPT,
5482             .type = QEMU_OPT_BOOL,
5483             .help = "Encrypt the image with format 'aes'. (Deprecated "
5484                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
5485         },
5486         {
5487             .name = BLOCK_OPT_ENCRYPT_FORMAT,
5488             .type = QEMU_OPT_STRING,
5489             .help = "Encrypt the image, format choices: 'aes', 'luks'",
5490         },
5491         BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
5492             "ID of secret providing qcow AES key or LUKS passphrase"),
5493         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
5494         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
5495         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
5496         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
5497         BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
5498         BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
5499         {
5500             .name = BLOCK_OPT_CLUSTER_SIZE,
5501             .type = QEMU_OPT_SIZE,
5502             .help = "qcow2 cluster size",
5503             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
5504         },
5505         {
5506             .name = BLOCK_OPT_PREALLOC,
5507             .type = QEMU_OPT_STRING,
5508             .help = "Preallocation mode (allowed values: off, metadata, "
5509                     "falloc, full)"
5510         },
5511         {
5512             .name = BLOCK_OPT_LAZY_REFCOUNTS,
5513             .type = QEMU_OPT_BOOL,
5514             .help = "Postpone refcount updates",
5515             .def_value_str = "off"
5516         },
5517         {
5518             .name = BLOCK_OPT_REFCOUNT_BITS,
5519             .type = QEMU_OPT_NUMBER,
5520             .help = "Width of a reference count entry in bits",
5521             .def_value_str = "16"
5522         },
5523         { /* end of list */ }
5524     }
5525 };
5526 
5527 static const char *const qcow2_strong_runtime_opts[] = {
5528     "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
5529 
5530     NULL
5531 };
5532 
5533 BlockDriver bdrv_qcow2 = {
5534     .format_name        = "qcow2",
5535     .instance_size      = sizeof(BDRVQcow2State),
5536     .bdrv_probe         = qcow2_probe,
5537     .bdrv_open          = qcow2_open,
5538     .bdrv_close         = qcow2_close,
5539     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
5540     .bdrv_reopen_commit   = qcow2_reopen_commit,
5541     .bdrv_reopen_commit_post = qcow2_reopen_commit_post,
5542     .bdrv_reopen_abort    = qcow2_reopen_abort,
5543     .bdrv_join_options    = qcow2_join_options,
5544     .bdrv_child_perm      = bdrv_format_default_perms,
5545     .bdrv_co_create_opts  = qcow2_co_create_opts,
5546     .bdrv_co_create       = qcow2_co_create,
5547     .bdrv_has_zero_init   = qcow2_has_zero_init,
5548     .bdrv_has_zero_init_truncate = bdrv_has_zero_init_1,
5549     .bdrv_co_block_status = qcow2_co_block_status,
5550 
5551     .bdrv_co_preadv_part    = qcow2_co_preadv_part,
5552     .bdrv_co_pwritev_part   = qcow2_co_pwritev_part,
5553     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
5554 
5555     .bdrv_co_pwrite_zeroes  = qcow2_co_pwrite_zeroes,
5556     .bdrv_co_pdiscard       = qcow2_co_pdiscard,
5557     .bdrv_co_copy_range_from = qcow2_co_copy_range_from,
5558     .bdrv_co_copy_range_to  = qcow2_co_copy_range_to,
5559     .bdrv_co_truncate       = qcow2_co_truncate,
5560     .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part,
5561     .bdrv_make_empty        = qcow2_make_empty,
5562 
5563     .bdrv_snapshot_create   = qcow2_snapshot_create,
5564     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
5565     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
5566     .bdrv_snapshot_list     = qcow2_snapshot_list,
5567     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
5568     .bdrv_measure           = qcow2_measure,
5569     .bdrv_get_info          = qcow2_get_info,
5570     .bdrv_get_specific_info = qcow2_get_specific_info,
5571 
5572     .bdrv_save_vmstate    = qcow2_save_vmstate,
5573     .bdrv_load_vmstate    = qcow2_load_vmstate,
5574 
5575     .supports_backing           = true,
5576     .bdrv_change_backing_file   = qcow2_change_backing_file,
5577 
5578     .bdrv_refresh_limits        = qcow2_refresh_limits,
5579     .bdrv_co_invalidate_cache   = qcow2_co_invalidate_cache,
5580     .bdrv_inactivate            = qcow2_inactivate,
5581 
5582     .create_opts         = &qcow2_create_opts,
5583     .strong_runtime_opts = qcow2_strong_runtime_opts,
5584     .mutable_opts        = mutable_opts,
5585     .bdrv_co_check       = qcow2_co_check,
5586     .bdrv_amend_options  = qcow2_amend_options,
5587 
5588     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
5589     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
5590 
5591     .bdrv_co_can_store_new_dirty_bitmap = qcow2_co_can_store_new_dirty_bitmap,
5592     .bdrv_co_remove_persistent_dirty_bitmap =
5593             qcow2_co_remove_persistent_dirty_bitmap,
5594 };
5595 
5596 static void bdrv_qcow2_init(void)
5597 {
5598     bdrv_register(&bdrv_qcow2);
5599 }
5600 
5601 block_init(bdrv_qcow2_init);
5602