xref: /openbmc/qemu/block/qcow2.c (revision 259ebed4)
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 "qemu/memalign.h"
42 #include "qapi/qobject-input-visitor.h"
43 #include "qapi/qapi-visit-block-core.h"
44 #include "crypto.h"
45 #include "block/aio_task.h"
46 #include "block/dirty-bitmap.h"
47 
48 /*
49   Differences with QCOW:
50 
51   - Support for multiple incremental snapshots.
52   - Memory management by reference counts.
53   - Clusters which have a reference count of one have the bit
54     QCOW_OFLAG_COPIED to optimize write performance.
55   - Size of compressed clusters is stored in sectors to reduce bit usage
56     in the cluster offsets.
57   - Support for storing additional data (such as the VM state) in the
58     snapshots.
59   - If a backing store is used, the cluster size is not constrained
60     (could be backported to QCOW).
61   - L2 tables have always a size of one cluster.
62 */
63 
64 
65 typedef struct {
66     uint32_t magic;
67     uint32_t len;
68 } QEMU_PACKED QCowExtension;
69 
70 #define  QCOW2_EXT_MAGIC_END 0
71 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xe2792aca
72 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
73 #define  QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
74 #define  QCOW2_EXT_MAGIC_BITMAPS 0x23852875
75 #define  QCOW2_EXT_MAGIC_DATA_FILE 0x44415441
76 
77 static int coroutine_fn
78 qcow2_co_preadv_compressed(BlockDriverState *bs,
79                            uint64_t l2_entry,
80                            uint64_t offset,
81                            uint64_t bytes,
82                            QEMUIOVector *qiov,
83                            size_t qiov_offset);
84 
85 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
86 {
87     const QCowHeader *cow_header = (const void *)buf;
88 
89     if (buf_size >= sizeof(QCowHeader) &&
90         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
91         be32_to_cpu(cow_header->version) >= 2)
92         return 100;
93     else
94         return 0;
95 }
96 
97 
98 static int qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset,
99                                       uint8_t *buf, size_t buflen,
100                                       void *opaque, Error **errp)
101 {
102     BlockDriverState *bs = opaque;
103     BDRVQcow2State *s = bs->opaque;
104     ssize_t ret;
105 
106     if ((offset + buflen) > s->crypto_header.length) {
107         error_setg(errp, "Request for data outside of extension header");
108         return -1;
109     }
110 
111     ret = bdrv_pread(bs->file, s->crypto_header.offset + offset, buflen, buf,
112                      0);
113     if (ret < 0) {
114         error_setg_errno(errp, -ret, "Could not read encryption header");
115         return -1;
116     }
117     return 0;
118 }
119 
120 
121 static int coroutine_fn GRAPH_RDLOCK
122 qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen, void *opaque,
123                            Error **errp)
124 {
125     BlockDriverState *bs = opaque;
126     BDRVQcow2State *s = bs->opaque;
127     int64_t ret;
128     int64_t clusterlen;
129 
130     ret = qcow2_alloc_clusters(bs, headerlen);
131     if (ret < 0) {
132         error_setg_errno(errp, -ret,
133                          "Cannot allocate cluster for LUKS header size %zu",
134                          headerlen);
135         return -1;
136     }
137 
138     s->crypto_header.length = headerlen;
139     s->crypto_header.offset = ret;
140 
141     /*
142      * Zero fill all space in cluster so it has predictable
143      * content, as we may not initialize some regions of the
144      * header (eg only 1 out of 8 key slots will be initialized)
145      */
146     clusterlen = size_to_clusters(s, headerlen) * s->cluster_size;
147     assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen, false) == 0);
148     ret = bdrv_co_pwrite_zeroes(bs->file, ret, clusterlen, 0);
149     if (ret < 0) {
150         error_setg_errno(errp, -ret, "Could not zero fill encryption header");
151         return -1;
152     }
153 
154     return 0;
155 }
156 
157 
158 /* The graph lock must be held when called in coroutine context */
159 static int coroutine_mixed_fn
160 qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset,
161                             const uint8_t *buf, size_t buflen,
162                             void *opaque, Error **errp)
163 {
164     BlockDriverState *bs = opaque;
165     BDRVQcow2State *s = bs->opaque;
166     ssize_t ret;
167 
168     if ((offset + buflen) > s->crypto_header.length) {
169         error_setg(errp, "Request for data outside of extension header");
170         return -1;
171     }
172 
173     ret = bdrv_pwrite(bs->file, s->crypto_header.offset + offset, buflen, buf,
174                       0);
175     if (ret < 0) {
176         error_setg_errno(errp, -ret, "Could not read encryption header");
177         return -1;
178     }
179     return 0;
180 }
181 
182 static QDict*
183 qcow2_extract_crypto_opts(QemuOpts *opts, const char *fmt, Error **errp)
184 {
185     QDict *cryptoopts_qdict;
186     QDict *opts_qdict;
187 
188     /* Extract "encrypt." options into a qdict */
189     opts_qdict = qemu_opts_to_qdict(opts, NULL);
190     qdict_extract_subqdict(opts_qdict, &cryptoopts_qdict, "encrypt.");
191     qobject_unref(opts_qdict);
192     qdict_put_str(cryptoopts_qdict, "format", fmt);
193     return cryptoopts_qdict;
194 }
195 
196 /*
197  * read qcow2 extension and fill bs
198  * start reading from start_offset
199  * finish reading upon magic of value 0 or when end_offset reached
200  * unknown magic is skipped (future extension this version knows nothing about)
201  * return 0 upon success, non-0 otherwise
202  */
203 static int coroutine_fn GRAPH_RDLOCK
204 qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
205                       uint64_t end_offset, void **p_feature_table,
206                       int flags, bool *need_update_header, Error **errp)
207 {
208     BDRVQcow2State *s = bs->opaque;
209     QCowExtension ext;
210     uint64_t offset;
211     int ret;
212     Qcow2BitmapHeaderExt bitmaps_ext;
213 
214     if (need_update_header != NULL) {
215         *need_update_header = false;
216     }
217 
218 #ifdef DEBUG_EXT
219     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
220 #endif
221     offset = start_offset;
222     while (offset < end_offset) {
223 
224 #ifdef DEBUG_EXT
225         /* Sanity check */
226         if (offset > s->cluster_size)
227             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
228 
229         printf("attempting to read extended header in offset %lu\n", offset);
230 #endif
231 
232         ret = bdrv_co_pread(bs->file, offset, sizeof(ext), &ext, 0);
233         if (ret < 0) {
234             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
235                              "pread fail from offset %" PRIu64, offset);
236             return 1;
237         }
238         ext.magic = be32_to_cpu(ext.magic);
239         ext.len = be32_to_cpu(ext.len);
240         offset += sizeof(ext);
241 #ifdef DEBUG_EXT
242         printf("ext.magic = 0x%x\n", ext.magic);
243 #endif
244         if (offset > end_offset || ext.len > end_offset - offset) {
245             error_setg(errp, "Header extension too large");
246             return -EINVAL;
247         }
248 
249         switch (ext.magic) {
250         case QCOW2_EXT_MAGIC_END:
251             return 0;
252 
253         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
254             if (ext.len >= sizeof(bs->backing_format)) {
255                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
256                            " too large (>=%zu)", ext.len,
257                            sizeof(bs->backing_format));
258                 return 2;
259             }
260             ret = bdrv_co_pread(bs->file, offset, ext.len, bs->backing_format, 0);
261             if (ret < 0) {
262                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
263                                  "Could not read format name");
264                 return 3;
265             }
266             bs->backing_format[ext.len] = '\0';
267             s->image_backing_format = g_strdup(bs->backing_format);
268 #ifdef DEBUG_EXT
269             printf("Qcow2: Got format extension %s\n", bs->backing_format);
270 #endif
271             break;
272 
273         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
274             if (p_feature_table != NULL) {
275                 void *feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
276                 ret = bdrv_co_pread(bs->file, offset, ext.len, feature_table, 0);
277                 if (ret < 0) {
278                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
279                                      "Could not read table");
280                     g_free(feature_table);
281                     return ret;
282                 }
283 
284                 *p_feature_table = feature_table;
285             }
286             break;
287 
288         case QCOW2_EXT_MAGIC_CRYPTO_HEADER: {
289             unsigned int cflags = 0;
290             if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
291                 error_setg(errp, "CRYPTO header extension only "
292                            "expected with LUKS encryption method");
293                 return -EINVAL;
294             }
295             if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) {
296                 error_setg(errp, "CRYPTO header extension size %u, "
297                            "but expected size %zu", ext.len,
298                            sizeof(Qcow2CryptoHeaderExtension));
299                 return -EINVAL;
300             }
301 
302             ret = bdrv_co_pread(bs->file, offset, ext.len, &s->crypto_header, 0);
303             if (ret < 0) {
304                 error_setg_errno(errp, -ret,
305                                  "Unable to read CRYPTO header extension");
306                 return ret;
307             }
308             s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
309             s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
310 
311             if ((s->crypto_header.offset % s->cluster_size) != 0) {
312                 error_setg(errp, "Encryption header offset '%" PRIu64 "' is "
313                            "not a multiple of cluster size '%u'",
314                            s->crypto_header.offset, s->cluster_size);
315                 return -EINVAL;
316             }
317 
318             if (flags & BDRV_O_NO_IO) {
319                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
320             }
321             s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
322                                            qcow2_crypto_hdr_read_func,
323                                            bs, cflags, QCOW2_MAX_THREADS, errp);
324             if (!s->crypto) {
325                 return -EINVAL;
326             }
327         }   break;
328 
329         case QCOW2_EXT_MAGIC_BITMAPS:
330             if (ext.len != sizeof(bitmaps_ext)) {
331                 error_setg_errno(errp, -ret, "bitmaps_ext: "
332                                  "Invalid extension length");
333                 return -EINVAL;
334             }
335 
336             if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) {
337                 if (s->qcow_version < 3) {
338                     /* Let's be a bit more specific */
339                     warn_report("This qcow2 v2 image contains bitmaps, but "
340                                 "they may have been modified by a program "
341                                 "without persistent bitmap support; so now "
342                                 "they must all be considered inconsistent");
343                 } else {
344                     warn_report("a program lacking bitmap support "
345                                 "modified this file, so all bitmaps are now "
346                                 "considered inconsistent");
347                 }
348                 error_printf("Some clusters may be leaked, "
349                              "run 'qemu-img check -r' on the image "
350                              "file to fix.");
351                 if (need_update_header != NULL) {
352                     /* Updating is needed to drop invalid bitmap extension. */
353                     *need_update_header = true;
354                 }
355                 break;
356             }
357 
358             ret = bdrv_co_pread(bs->file, offset, ext.len, &bitmaps_ext, 0);
359             if (ret < 0) {
360                 error_setg_errno(errp, -ret, "bitmaps_ext: "
361                                  "Could not read ext header");
362                 return ret;
363             }
364 
365             if (bitmaps_ext.reserved32 != 0) {
366                 error_setg_errno(errp, -ret, "bitmaps_ext: "
367                                  "Reserved field is not zero");
368                 return -EINVAL;
369             }
370 
371             bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps);
372             bitmaps_ext.bitmap_directory_size =
373                 be64_to_cpu(bitmaps_ext.bitmap_directory_size);
374             bitmaps_ext.bitmap_directory_offset =
375                 be64_to_cpu(bitmaps_ext.bitmap_directory_offset);
376 
377             if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) {
378                 error_setg(errp,
379                            "bitmaps_ext: Image has %" PRIu32 " bitmaps, "
380                            "exceeding the QEMU supported maximum of %d",
381                            bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS);
382                 return -EINVAL;
383             }
384 
385             if (bitmaps_ext.nb_bitmaps == 0) {
386                 error_setg(errp, "found bitmaps extension with zero bitmaps");
387                 return -EINVAL;
388             }
389 
390             if (offset_into_cluster(s, bitmaps_ext.bitmap_directory_offset)) {
391                 error_setg(errp, "bitmaps_ext: "
392                                  "invalid bitmap directory offset");
393                 return -EINVAL;
394             }
395 
396             if (bitmaps_ext.bitmap_directory_size >
397                 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) {
398                 error_setg(errp, "bitmaps_ext: "
399                                  "bitmap directory size (%" PRIu64 ") exceeds "
400                                  "the maximum supported size (%d)",
401                                  bitmaps_ext.bitmap_directory_size,
402                                  QCOW2_MAX_BITMAP_DIRECTORY_SIZE);
403                 return -EINVAL;
404             }
405 
406             s->nb_bitmaps = bitmaps_ext.nb_bitmaps;
407             s->bitmap_directory_offset =
408                     bitmaps_ext.bitmap_directory_offset;
409             s->bitmap_directory_size =
410                     bitmaps_ext.bitmap_directory_size;
411 
412 #ifdef DEBUG_EXT
413             printf("Qcow2: Got bitmaps extension: "
414                    "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n",
415                    s->bitmap_directory_offset, s->nb_bitmaps);
416 #endif
417             break;
418 
419         case QCOW2_EXT_MAGIC_DATA_FILE:
420         {
421             s->image_data_file = g_malloc0(ext.len + 1);
422             ret = bdrv_co_pread(bs->file, offset, ext.len, s->image_data_file, 0);
423             if (ret < 0) {
424                 error_setg_errno(errp, -ret,
425                                  "ERROR: Could not read data file name");
426                 return ret;
427             }
428 #ifdef DEBUG_EXT
429             printf("Qcow2: Got external data file %s\n", s->image_data_file);
430 #endif
431             break;
432         }
433 
434         default:
435             /* unknown magic - save it in case we need to rewrite the header */
436             /* If you add a new feature, make sure to also update the fast
437              * path of qcow2_make_empty() to deal with it. */
438             {
439                 Qcow2UnknownHeaderExtension *uext;
440 
441                 uext = g_malloc0(sizeof(*uext)  + ext.len);
442                 uext->magic = ext.magic;
443                 uext->len = ext.len;
444                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
445 
446                 ret = bdrv_co_pread(bs->file, offset, uext->len, uext->data, 0);
447                 if (ret < 0) {
448                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
449                                      "Could not read data");
450                     return ret;
451                 }
452             }
453             break;
454         }
455 
456         offset += ((ext.len + 7) & ~7);
457     }
458 
459     return 0;
460 }
461 
462 static void cleanup_unknown_header_ext(BlockDriverState *bs)
463 {
464     BDRVQcow2State *s = bs->opaque;
465     Qcow2UnknownHeaderExtension *uext, *next;
466 
467     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
468         QLIST_REMOVE(uext, next);
469         g_free(uext);
470     }
471 }
472 
473 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
474                                        uint64_t mask)
475 {
476     g_autoptr(GString) features = g_string_sized_new(60);
477 
478     while (table && table->name[0] != '\0') {
479         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
480             if (mask & (1ULL << table->bit)) {
481                 if (features->len > 0) {
482                     g_string_append(features, ", ");
483                 }
484                 g_string_append_printf(features, "%.46s", table->name);
485                 mask &= ~(1ULL << table->bit);
486             }
487         }
488         table++;
489     }
490 
491     if (mask) {
492         if (features->len > 0) {
493             g_string_append(features, ", ");
494         }
495         g_string_append_printf(features,
496                                "Unknown incompatible feature: %" PRIx64, mask);
497     }
498 
499     error_setg(errp, "Unsupported qcow2 feature(s): %s", features->str);
500 }
501 
502 /*
503  * Sets the dirty bit and flushes afterwards if necessary.
504  *
505  * The incompatible_features bit is only set if the image file header was
506  * updated successfully.  Therefore it is not required to check the return
507  * value of this function.
508  */
509 int qcow2_mark_dirty(BlockDriverState *bs)
510 {
511     BDRVQcow2State *s = bs->opaque;
512     uint64_t val;
513     int ret;
514 
515     assert(s->qcow_version >= 3);
516 
517     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
518         return 0; /* already dirty */
519     }
520 
521     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
522     ret = bdrv_pwrite_sync(bs->file,
523                            offsetof(QCowHeader, incompatible_features),
524                            sizeof(val), &val, 0);
525     if (ret < 0) {
526         return ret;
527     }
528 
529     /* Only treat image as dirty if the header was updated successfully */
530     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
531     return 0;
532 }
533 
534 /*
535  * Clears the dirty bit and flushes before if necessary.  Only call this
536  * function when there are no pending requests, it does not guard against
537  * concurrent requests dirtying the image.
538  */
539 static int GRAPH_RDLOCK qcow2_mark_clean(BlockDriverState *bs)
540 {
541     BDRVQcow2State *s = bs->opaque;
542 
543     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
544         int ret;
545 
546         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
547 
548         ret = qcow2_flush_caches(bs);
549         if (ret < 0) {
550             return ret;
551         }
552 
553         return qcow2_update_header(bs);
554     }
555     return 0;
556 }
557 
558 /*
559  * Marks the image as corrupt.
560  */
561 int qcow2_mark_corrupt(BlockDriverState *bs)
562 {
563     BDRVQcow2State *s = bs->opaque;
564 
565     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
566     return qcow2_update_header(bs);
567 }
568 
569 /*
570  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
571  * before if necessary.
572  */
573 static int coroutine_fn GRAPH_RDLOCK
574 qcow2_mark_consistent(BlockDriverState *bs)
575 {
576     BDRVQcow2State *s = bs->opaque;
577 
578     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
579         int ret = qcow2_flush_caches(bs);
580         if (ret < 0) {
581             return ret;
582         }
583 
584         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
585         return qcow2_update_header(bs);
586     }
587     return 0;
588 }
589 
590 static void qcow2_add_check_result(BdrvCheckResult *out,
591                                    const BdrvCheckResult *src,
592                                    bool set_allocation_info)
593 {
594     out->corruptions += src->corruptions;
595     out->leaks += src->leaks;
596     out->check_errors += src->check_errors;
597     out->corruptions_fixed += src->corruptions_fixed;
598     out->leaks_fixed += src->leaks_fixed;
599 
600     if (set_allocation_info) {
601         out->image_end_offset = src->image_end_offset;
602         out->bfi = src->bfi;
603     }
604 }
605 
606 static int coroutine_fn GRAPH_RDLOCK
607 qcow2_co_check_locked(BlockDriverState *bs, BdrvCheckResult *result,
608                       BdrvCheckMode fix)
609 {
610     BdrvCheckResult snapshot_res = {};
611     BdrvCheckResult refcount_res = {};
612     int ret;
613 
614     memset(result, 0, sizeof(*result));
615 
616     ret = qcow2_check_read_snapshot_table(bs, &snapshot_res, fix);
617     if (ret < 0) {
618         qcow2_add_check_result(result, &snapshot_res, false);
619         return ret;
620     }
621 
622     ret = qcow2_check_refcounts(bs, &refcount_res, fix);
623     qcow2_add_check_result(result, &refcount_res, true);
624     if (ret < 0) {
625         qcow2_add_check_result(result, &snapshot_res, false);
626         return ret;
627     }
628 
629     ret = qcow2_check_fix_snapshot_table(bs, &snapshot_res, fix);
630     qcow2_add_check_result(result, &snapshot_res, false);
631     if (ret < 0) {
632         return ret;
633     }
634 
635     if (fix && result->check_errors == 0 && result->corruptions == 0) {
636         ret = qcow2_mark_clean(bs);
637         if (ret < 0) {
638             return ret;
639         }
640         return qcow2_mark_consistent(bs);
641     }
642     return ret;
643 }
644 
645 static int coroutine_fn GRAPH_RDLOCK
646 qcow2_co_check(BlockDriverState *bs, BdrvCheckResult *result,
647                BdrvCheckMode fix)
648 {
649     BDRVQcow2State *s = bs->opaque;
650     int ret;
651 
652     qemu_co_mutex_lock(&s->lock);
653     ret = qcow2_co_check_locked(bs, result, fix);
654     qemu_co_mutex_unlock(&s->lock);
655     return ret;
656 }
657 
658 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset,
659                          uint64_t entries, size_t entry_len,
660                          int64_t max_size_bytes, const char *table_name,
661                          Error **errp)
662 {
663     BDRVQcow2State *s = bs->opaque;
664 
665     if (entries > max_size_bytes / entry_len) {
666         error_setg(errp, "%s too large", table_name);
667         return -EFBIG;
668     }
669 
670     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
671      * because values will be passed to qemu functions taking int64_t. */
672     if ((INT64_MAX - entries * entry_len < offset) ||
673         (offset_into_cluster(s, offset) != 0)) {
674         error_setg(errp, "%s offset invalid", table_name);
675         return -EINVAL;
676     }
677 
678     return 0;
679 }
680 
681 static const char *const mutable_opts[] = {
682     QCOW2_OPT_LAZY_REFCOUNTS,
683     QCOW2_OPT_DISCARD_REQUEST,
684     QCOW2_OPT_DISCARD_SNAPSHOT,
685     QCOW2_OPT_DISCARD_OTHER,
686     QCOW2_OPT_DISCARD_NO_UNREF,
687     QCOW2_OPT_OVERLAP,
688     QCOW2_OPT_OVERLAP_TEMPLATE,
689     QCOW2_OPT_OVERLAP_MAIN_HEADER,
690     QCOW2_OPT_OVERLAP_ACTIVE_L1,
691     QCOW2_OPT_OVERLAP_ACTIVE_L2,
692     QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
693     QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
694     QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
695     QCOW2_OPT_OVERLAP_INACTIVE_L1,
696     QCOW2_OPT_OVERLAP_INACTIVE_L2,
697     QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
698     QCOW2_OPT_CACHE_SIZE,
699     QCOW2_OPT_L2_CACHE_SIZE,
700     QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
701     QCOW2_OPT_REFCOUNT_CACHE_SIZE,
702     QCOW2_OPT_CACHE_CLEAN_INTERVAL,
703     NULL
704 };
705 
706 static QemuOptsList qcow2_runtime_opts = {
707     .name = "qcow2",
708     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
709     .desc = {
710         {
711             .name = QCOW2_OPT_LAZY_REFCOUNTS,
712             .type = QEMU_OPT_BOOL,
713             .help = "Postpone refcount updates",
714         },
715         {
716             .name = QCOW2_OPT_DISCARD_REQUEST,
717             .type = QEMU_OPT_BOOL,
718             .help = "Pass guest discard requests to the layer below",
719         },
720         {
721             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
722             .type = QEMU_OPT_BOOL,
723             .help = "Generate discard requests when snapshot related space "
724                     "is freed",
725         },
726         {
727             .name = QCOW2_OPT_DISCARD_OTHER,
728             .type = QEMU_OPT_BOOL,
729             .help = "Generate discard requests when other clusters are freed",
730         },
731         {
732             .name = QCOW2_OPT_DISCARD_NO_UNREF,
733             .type = QEMU_OPT_BOOL,
734             .help = "Do not unreference discarded clusters",
735         },
736         {
737             .name = QCOW2_OPT_OVERLAP,
738             .type = QEMU_OPT_STRING,
739             .help = "Selects which overlap checks to perform from a range of "
740                     "templates (none, constant, cached, all)",
741         },
742         {
743             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
744             .type = QEMU_OPT_STRING,
745             .help = "Selects which overlap checks to perform from a range of "
746                     "templates (none, constant, cached, all)",
747         },
748         {
749             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
750             .type = QEMU_OPT_BOOL,
751             .help = "Check for unintended writes into the main qcow2 header",
752         },
753         {
754             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
755             .type = QEMU_OPT_BOOL,
756             .help = "Check for unintended writes into the active L1 table",
757         },
758         {
759             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
760             .type = QEMU_OPT_BOOL,
761             .help = "Check for unintended writes into an active L2 table",
762         },
763         {
764             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
765             .type = QEMU_OPT_BOOL,
766             .help = "Check for unintended writes into the refcount table",
767         },
768         {
769             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
770             .type = QEMU_OPT_BOOL,
771             .help = "Check for unintended writes into a refcount block",
772         },
773         {
774             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
775             .type = QEMU_OPT_BOOL,
776             .help = "Check for unintended writes into the snapshot table",
777         },
778         {
779             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
780             .type = QEMU_OPT_BOOL,
781             .help = "Check for unintended writes into an inactive L1 table",
782         },
783         {
784             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
785             .type = QEMU_OPT_BOOL,
786             .help = "Check for unintended writes into an inactive L2 table",
787         },
788         {
789             .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
790             .type = QEMU_OPT_BOOL,
791             .help = "Check for unintended writes into the bitmap directory",
792         },
793         {
794             .name = QCOW2_OPT_CACHE_SIZE,
795             .type = QEMU_OPT_SIZE,
796             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
797                     "cache size",
798         },
799         {
800             .name = QCOW2_OPT_L2_CACHE_SIZE,
801             .type = QEMU_OPT_SIZE,
802             .help = "Maximum L2 table cache size",
803         },
804         {
805             .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
806             .type = QEMU_OPT_SIZE,
807             .help = "Size of each entry in the L2 cache",
808         },
809         {
810             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
811             .type = QEMU_OPT_SIZE,
812             .help = "Maximum refcount block cache size",
813         },
814         {
815             .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
816             .type = QEMU_OPT_NUMBER,
817             .help = "Clean unused cache entries after this time (in seconds)",
818         },
819         BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
820             "ID of secret providing qcow2 AES key or LUKS passphrase"),
821         { /* end of list */ }
822     },
823 };
824 
825 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
826     [QCOW2_OL_MAIN_HEADER_BITNR]      = QCOW2_OPT_OVERLAP_MAIN_HEADER,
827     [QCOW2_OL_ACTIVE_L1_BITNR]        = QCOW2_OPT_OVERLAP_ACTIVE_L1,
828     [QCOW2_OL_ACTIVE_L2_BITNR]        = QCOW2_OPT_OVERLAP_ACTIVE_L2,
829     [QCOW2_OL_REFCOUNT_TABLE_BITNR]   = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
830     [QCOW2_OL_REFCOUNT_BLOCK_BITNR]   = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
831     [QCOW2_OL_SNAPSHOT_TABLE_BITNR]   = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
832     [QCOW2_OL_INACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_INACTIVE_L1,
833     [QCOW2_OL_INACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_INACTIVE_L2,
834     [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
835 };
836 
837 static void cache_clean_timer_cb(void *opaque)
838 {
839     BlockDriverState *bs = opaque;
840     BDRVQcow2State *s = bs->opaque;
841     qcow2_cache_clean_unused(s->l2_table_cache);
842     qcow2_cache_clean_unused(s->refcount_block_cache);
843     timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
844               (int64_t) s->cache_clean_interval * 1000);
845 }
846 
847 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
848 {
849     BDRVQcow2State *s = bs->opaque;
850     if (s->cache_clean_interval > 0) {
851         s->cache_clean_timer =
852             aio_timer_new_with_attrs(context, QEMU_CLOCK_VIRTUAL,
853                                      SCALE_MS, QEMU_TIMER_ATTR_EXTERNAL,
854                                      cache_clean_timer_cb, bs);
855         timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
856                   (int64_t) s->cache_clean_interval * 1000);
857     }
858 }
859 
860 static void cache_clean_timer_del(BlockDriverState *bs)
861 {
862     BDRVQcow2State *s = bs->opaque;
863     if (s->cache_clean_timer) {
864         timer_free(s->cache_clean_timer);
865         s->cache_clean_timer = NULL;
866     }
867 }
868 
869 static void qcow2_detach_aio_context(BlockDriverState *bs)
870 {
871     cache_clean_timer_del(bs);
872 }
873 
874 static void qcow2_attach_aio_context(BlockDriverState *bs,
875                                      AioContext *new_context)
876 {
877     cache_clean_timer_init(bs, new_context);
878 }
879 
880 static bool read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
881                              uint64_t *l2_cache_size,
882                              uint64_t *l2_cache_entry_size,
883                              uint64_t *refcount_cache_size, Error **errp)
884 {
885     BDRVQcow2State *s = bs->opaque;
886     uint64_t combined_cache_size, l2_cache_max_setting;
887     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
888     bool l2_cache_entry_size_set;
889     int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size;
890     uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
891     uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size);
892     /* An L2 table is always one cluster in size so the max cache size
893      * should be a multiple of the cluster size. */
894     uint64_t max_l2_cache = ROUND_UP(max_l2_entries * l2_entry_size(s),
895                                      s->cluster_size);
896 
897     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
898     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
899     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
900     l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE);
901 
902     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
903     l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE,
904                                              DEFAULT_L2_CACHE_MAX_SIZE);
905     *refcount_cache_size = qemu_opt_get_size(opts,
906                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
907 
908     *l2_cache_entry_size = qemu_opt_get_size(
909         opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size);
910 
911     *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting);
912 
913     if (combined_cache_size_set) {
914         if (l2_cache_size_set && refcount_cache_size_set) {
915             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
916                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
917                        "at the same time");
918             return false;
919         } else if (l2_cache_size_set &&
920                    (l2_cache_max_setting > combined_cache_size)) {
921             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
922                        QCOW2_OPT_CACHE_SIZE);
923             return false;
924         } else if (*refcount_cache_size > combined_cache_size) {
925             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
926                        QCOW2_OPT_CACHE_SIZE);
927             return false;
928         }
929 
930         if (l2_cache_size_set) {
931             *refcount_cache_size = combined_cache_size - *l2_cache_size;
932         } else if (refcount_cache_size_set) {
933             *l2_cache_size = combined_cache_size - *refcount_cache_size;
934         } else {
935             /* Assign as much memory as possible to the L2 cache, and
936              * use the remainder for the refcount cache */
937             if (combined_cache_size >= max_l2_cache + min_refcount_cache) {
938                 *l2_cache_size = max_l2_cache;
939                 *refcount_cache_size = combined_cache_size - *l2_cache_size;
940             } else {
941                 *refcount_cache_size =
942                     MIN(combined_cache_size, min_refcount_cache);
943                 *l2_cache_size = combined_cache_size - *refcount_cache_size;
944             }
945         }
946     }
947 
948     /*
949      * If the L2 cache is not enough to cover the whole disk then
950      * default to 4KB entries. Smaller entries reduce the cost of
951      * loads and evictions and increase I/O performance.
952      */
953     if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) {
954         *l2_cache_entry_size = MIN(s->cluster_size, 4096);
955     }
956 
957     /* l2_cache_size and refcount_cache_size are ensured to have at least
958      * their minimum values in qcow2_update_options_prepare() */
959 
960     if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) ||
961         *l2_cache_entry_size > s->cluster_size ||
962         !is_power_of_2(*l2_cache_entry_size)) {
963         error_setg(errp, "L2 cache entry size must be a power of two "
964                    "between %d and the cluster size (%d)",
965                    1 << MIN_CLUSTER_BITS, s->cluster_size);
966         return false;
967     }
968 
969     return true;
970 }
971 
972 typedef struct Qcow2ReopenState {
973     Qcow2Cache *l2_table_cache;
974     Qcow2Cache *refcount_block_cache;
975     int l2_slice_size; /* Number of entries in a slice of the L2 table */
976     bool use_lazy_refcounts;
977     int overlap_check;
978     bool discard_passthrough[QCOW2_DISCARD_MAX];
979     bool discard_no_unref;
980     uint64_t cache_clean_interval;
981     QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
982 } Qcow2ReopenState;
983 
984 static int GRAPH_RDLOCK
985 qcow2_update_options_prepare(BlockDriverState *bs, Qcow2ReopenState *r,
986                              QDict *options, int flags, Error **errp)
987 {
988     BDRVQcow2State *s = bs->opaque;
989     QemuOpts *opts = NULL;
990     const char *opt_overlap_check, *opt_overlap_check_template;
991     int overlap_check_template = 0;
992     uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size;
993     int i;
994     const char *encryptfmt;
995     QDict *encryptopts = NULL;
996     int ret;
997 
998     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
999     encryptfmt = qdict_get_try_str(encryptopts, "format");
1000 
1001     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
1002     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1003         ret = -EINVAL;
1004         goto fail;
1005     }
1006 
1007     /* get L2 table/refcount block cache size from command line options */
1008     if (!read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size,
1009                           &refcount_cache_size, errp)) {
1010         ret = -EINVAL;
1011         goto fail;
1012     }
1013 
1014     l2_cache_size /= l2_cache_entry_size;
1015     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
1016         l2_cache_size = MIN_L2_CACHE_SIZE;
1017     }
1018     if (l2_cache_size > INT_MAX) {
1019         error_setg(errp, "L2 cache size too big");
1020         ret = -EINVAL;
1021         goto fail;
1022     }
1023 
1024     refcount_cache_size /= s->cluster_size;
1025     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
1026         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
1027     }
1028     if (refcount_cache_size > INT_MAX) {
1029         error_setg(errp, "Refcount cache size too big");
1030         ret = -EINVAL;
1031         goto fail;
1032     }
1033 
1034     /* alloc new L2 table/refcount block cache, flush old one */
1035     if (s->l2_table_cache) {
1036         ret = qcow2_cache_flush(bs, s->l2_table_cache);
1037         if (ret) {
1038             error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
1039             goto fail;
1040         }
1041     }
1042 
1043     if (s->refcount_block_cache) {
1044         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1045         if (ret) {
1046             error_setg_errno(errp, -ret,
1047                              "Failed to flush the refcount block cache");
1048             goto fail;
1049         }
1050     }
1051 
1052     r->l2_slice_size = l2_cache_entry_size / l2_entry_size(s);
1053     r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size,
1054                                            l2_cache_entry_size);
1055     r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size,
1056                                                  s->cluster_size);
1057     if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
1058         error_setg(errp, "Could not allocate metadata caches");
1059         ret = -ENOMEM;
1060         goto fail;
1061     }
1062 
1063     /* New interval for cache cleanup timer */
1064     r->cache_clean_interval =
1065         qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
1066                             DEFAULT_CACHE_CLEAN_INTERVAL);
1067 #ifndef CONFIG_LINUX
1068     if (r->cache_clean_interval != 0) {
1069         error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL
1070                    " not supported on this host");
1071         ret = -EINVAL;
1072         goto fail;
1073     }
1074 #endif
1075     if (r->cache_clean_interval > UINT_MAX) {
1076         error_setg(errp, "Cache clean interval too big");
1077         ret = -EINVAL;
1078         goto fail;
1079     }
1080 
1081     /* lazy-refcounts; flush if going from enabled to disabled */
1082     r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
1083         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
1084     if (r->use_lazy_refcounts && s->qcow_version < 3) {
1085         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
1086                    "qemu 1.1 compatibility level");
1087         ret = -EINVAL;
1088         goto fail;
1089     }
1090 
1091     if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
1092         ret = qcow2_mark_clean(bs);
1093         if (ret < 0) {
1094             error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
1095             goto fail;
1096         }
1097     }
1098 
1099     /* Overlap check options */
1100     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
1101     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
1102     if (opt_overlap_check_template && opt_overlap_check &&
1103         strcmp(opt_overlap_check_template, opt_overlap_check))
1104     {
1105         error_setg(errp, "Conflicting values for qcow2 options '"
1106                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1107                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
1108         ret = -EINVAL;
1109         goto fail;
1110     }
1111     if (!opt_overlap_check) {
1112         opt_overlap_check = opt_overlap_check_template ?: "cached";
1113     }
1114 
1115     if (!strcmp(opt_overlap_check, "none")) {
1116         overlap_check_template = 0;
1117     } else if (!strcmp(opt_overlap_check, "constant")) {
1118         overlap_check_template = QCOW2_OL_CONSTANT;
1119     } else if (!strcmp(opt_overlap_check, "cached")) {
1120         overlap_check_template = QCOW2_OL_CACHED;
1121     } else if (!strcmp(opt_overlap_check, "all")) {
1122         overlap_check_template = QCOW2_OL_ALL;
1123     } else {
1124         error_setg(errp, "Unsupported value '%s' for qcow2 option "
1125                    "'overlap-check'. Allowed are any of the following: "
1126                    "none, constant, cached, all", opt_overlap_check);
1127         ret = -EINVAL;
1128         goto fail;
1129     }
1130 
1131     r->overlap_check = 0;
1132     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
1133         /* overlap-check defines a template bitmask, but every flag may be
1134          * overwritten through the associated boolean option */
1135         r->overlap_check |=
1136             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
1137                               overlap_check_template & (1 << i)) << i;
1138     }
1139 
1140     r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
1141     r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
1142     r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
1143         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
1144                           flags & BDRV_O_UNMAP);
1145     r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
1146         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
1147     r->discard_passthrough[QCOW2_DISCARD_OTHER] =
1148         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
1149 
1150     r->discard_no_unref = qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_NO_UNREF,
1151                                             false);
1152     if (r->discard_no_unref && s->qcow_version < 3) {
1153         error_setg(errp,
1154                    "discard-no-unref is only supported since qcow2 version 3");
1155         ret = -EINVAL;
1156         goto fail;
1157     }
1158 
1159     switch (s->crypt_method_header) {
1160     case QCOW_CRYPT_NONE:
1161         if (encryptfmt) {
1162             error_setg(errp, "No encryption in image header, but options "
1163                        "specified format '%s'", encryptfmt);
1164             ret = -EINVAL;
1165             goto fail;
1166         }
1167         break;
1168 
1169     case QCOW_CRYPT_AES:
1170         if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
1171             error_setg(errp,
1172                        "Header reported 'aes' encryption format but "
1173                        "options specify '%s'", encryptfmt);
1174             ret = -EINVAL;
1175             goto fail;
1176         }
1177         qdict_put_str(encryptopts, "format", "qcow");
1178         r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1179         if (!r->crypto_opts) {
1180             ret = -EINVAL;
1181             goto fail;
1182         }
1183         break;
1184 
1185     case QCOW_CRYPT_LUKS:
1186         if (encryptfmt && !g_str_equal(encryptfmt, "luks")) {
1187             error_setg(errp,
1188                        "Header reported 'luks' encryption format but "
1189                        "options specify '%s'", encryptfmt);
1190             ret = -EINVAL;
1191             goto fail;
1192         }
1193         qdict_put_str(encryptopts, "format", "luks");
1194         r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1195         if (!r->crypto_opts) {
1196             ret = -EINVAL;
1197             goto fail;
1198         }
1199         break;
1200 
1201     default:
1202         error_setg(errp, "Unsupported encryption method %d",
1203                    s->crypt_method_header);
1204         ret = -EINVAL;
1205         goto fail;
1206     }
1207 
1208     ret = 0;
1209 fail:
1210     qobject_unref(encryptopts);
1211     qemu_opts_del(opts);
1212     opts = NULL;
1213     return ret;
1214 }
1215 
1216 static void qcow2_update_options_commit(BlockDriverState *bs,
1217                                         Qcow2ReopenState *r)
1218 {
1219     BDRVQcow2State *s = bs->opaque;
1220     int i;
1221 
1222     if (s->l2_table_cache) {
1223         qcow2_cache_destroy(s->l2_table_cache);
1224     }
1225     if (s->refcount_block_cache) {
1226         qcow2_cache_destroy(s->refcount_block_cache);
1227     }
1228     s->l2_table_cache = r->l2_table_cache;
1229     s->refcount_block_cache = r->refcount_block_cache;
1230     s->l2_slice_size = r->l2_slice_size;
1231 
1232     s->overlap_check = r->overlap_check;
1233     s->use_lazy_refcounts = r->use_lazy_refcounts;
1234 
1235     for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
1236         s->discard_passthrough[i] = r->discard_passthrough[i];
1237     }
1238 
1239     s->discard_no_unref = r->discard_no_unref;
1240 
1241     if (s->cache_clean_interval != r->cache_clean_interval) {
1242         cache_clean_timer_del(bs);
1243         s->cache_clean_interval = r->cache_clean_interval;
1244         cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
1245     }
1246 
1247     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1248     s->crypto_opts = r->crypto_opts;
1249 }
1250 
1251 static void qcow2_update_options_abort(BlockDriverState *bs,
1252                                        Qcow2ReopenState *r)
1253 {
1254     if (r->l2_table_cache) {
1255         qcow2_cache_destroy(r->l2_table_cache);
1256     }
1257     if (r->refcount_block_cache) {
1258         qcow2_cache_destroy(r->refcount_block_cache);
1259     }
1260     qapi_free_QCryptoBlockOpenOptions(r->crypto_opts);
1261 }
1262 
1263 static int coroutine_fn GRAPH_RDLOCK
1264 qcow2_update_options(BlockDriverState *bs, QDict *options, int flags,
1265                      Error **errp)
1266 {
1267     Qcow2ReopenState r = {};
1268     int ret;
1269 
1270     ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
1271     if (ret >= 0) {
1272         qcow2_update_options_commit(bs, &r);
1273     } else {
1274         qcow2_update_options_abort(bs, &r);
1275     }
1276 
1277     return ret;
1278 }
1279 
1280 static int validate_compression_type(BDRVQcow2State *s, Error **errp)
1281 {
1282     switch (s->compression_type) {
1283     case QCOW2_COMPRESSION_TYPE_ZLIB:
1284 #ifdef CONFIG_ZSTD
1285     case QCOW2_COMPRESSION_TYPE_ZSTD:
1286 #endif
1287         break;
1288 
1289     default:
1290         error_setg(errp, "qcow2: unknown compression type: %u",
1291                    s->compression_type);
1292         return -ENOTSUP;
1293     }
1294 
1295     /*
1296      * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB
1297      * the incompatible feature flag must be set
1298      */
1299     if (s->compression_type == QCOW2_COMPRESSION_TYPE_ZLIB) {
1300         if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) {
1301             error_setg(errp, "qcow2: Compression type incompatible feature "
1302                              "bit must not be set");
1303             return -EINVAL;
1304         }
1305     } else {
1306         if (!(s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION)) {
1307             error_setg(errp, "qcow2: Compression type incompatible feature "
1308                              "bit must be set");
1309             return -EINVAL;
1310         }
1311     }
1312 
1313     return 0;
1314 }
1315 
1316 /* Called with s->lock held.  */
1317 static int coroutine_fn GRAPH_RDLOCK
1318 qcow2_do_open(BlockDriverState *bs, QDict *options, int flags,
1319               bool open_data_file, Error **errp)
1320 {
1321     ERRP_GUARD();
1322     BDRVQcow2State *s = bs->opaque;
1323     unsigned int len, i;
1324     int ret = 0;
1325     QCowHeader header;
1326     uint64_t ext_end;
1327     uint64_t l1_vm_state_index;
1328     bool update_header = false;
1329 
1330     ret = bdrv_co_pread(bs->file, 0, sizeof(header), &header, 0);
1331     if (ret < 0) {
1332         error_setg_errno(errp, -ret, "Could not read qcow2 header");
1333         goto fail;
1334     }
1335     header.magic = be32_to_cpu(header.magic);
1336     header.version = be32_to_cpu(header.version);
1337     header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
1338     header.backing_file_size = be32_to_cpu(header.backing_file_size);
1339     header.size = be64_to_cpu(header.size);
1340     header.cluster_bits = be32_to_cpu(header.cluster_bits);
1341     header.crypt_method = be32_to_cpu(header.crypt_method);
1342     header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
1343     header.l1_size = be32_to_cpu(header.l1_size);
1344     header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset);
1345     header.refcount_table_clusters =
1346         be32_to_cpu(header.refcount_table_clusters);
1347     header.snapshots_offset = be64_to_cpu(header.snapshots_offset);
1348     header.nb_snapshots = be32_to_cpu(header.nb_snapshots);
1349 
1350     if (header.magic != QCOW_MAGIC) {
1351         error_setg(errp, "Image is not in qcow2 format");
1352         ret = -EINVAL;
1353         goto fail;
1354     }
1355     if (header.version < 2 || header.version > 3) {
1356         error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
1357         ret = -ENOTSUP;
1358         goto fail;
1359     }
1360 
1361     s->qcow_version = header.version;
1362 
1363     /* Initialise cluster size */
1364     if (header.cluster_bits < MIN_CLUSTER_BITS ||
1365         header.cluster_bits > MAX_CLUSTER_BITS) {
1366         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
1367                    header.cluster_bits);
1368         ret = -EINVAL;
1369         goto fail;
1370     }
1371 
1372     s->cluster_bits = header.cluster_bits;
1373     s->cluster_size = 1 << s->cluster_bits;
1374 
1375     /* Initialise version 3 header fields */
1376     if (header.version == 2) {
1377         header.incompatible_features    = 0;
1378         header.compatible_features      = 0;
1379         header.autoclear_features       = 0;
1380         header.refcount_order           = 4;
1381         header.header_length            = 72;
1382     } else {
1383         header.incompatible_features =
1384             be64_to_cpu(header.incompatible_features);
1385         header.compatible_features = be64_to_cpu(header.compatible_features);
1386         header.autoclear_features = be64_to_cpu(header.autoclear_features);
1387         header.refcount_order = be32_to_cpu(header.refcount_order);
1388         header.header_length = be32_to_cpu(header.header_length);
1389 
1390         if (header.header_length < 104) {
1391             error_setg(errp, "qcow2 header too short");
1392             ret = -EINVAL;
1393             goto fail;
1394         }
1395     }
1396 
1397     if (header.header_length > s->cluster_size) {
1398         error_setg(errp, "qcow2 header exceeds cluster size");
1399         ret = -EINVAL;
1400         goto fail;
1401     }
1402 
1403     if (header.header_length > sizeof(header)) {
1404         s->unknown_header_fields_size = header.header_length - sizeof(header);
1405         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
1406         ret = bdrv_co_pread(bs->file, sizeof(header),
1407                             s->unknown_header_fields_size,
1408                             s->unknown_header_fields, 0);
1409         if (ret < 0) {
1410             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
1411                              "fields");
1412             goto fail;
1413         }
1414     }
1415 
1416     if (header.backing_file_offset > s->cluster_size) {
1417         error_setg(errp, "Invalid backing file offset");
1418         ret = -EINVAL;
1419         goto fail;
1420     }
1421 
1422     if (header.backing_file_offset) {
1423         ext_end = header.backing_file_offset;
1424     } else {
1425         ext_end = 1 << header.cluster_bits;
1426     }
1427 
1428     /* Handle feature bits */
1429     s->incompatible_features    = header.incompatible_features;
1430     s->compatible_features      = header.compatible_features;
1431     s->autoclear_features       = header.autoclear_features;
1432 
1433     /*
1434      * Handle compression type
1435      * Older qcow2 images don't contain the compression type header.
1436      * Distinguish them by the header length and use
1437      * the only valid (default) compression type in that case
1438      */
1439     if (header.header_length > offsetof(QCowHeader, compression_type)) {
1440         s->compression_type = header.compression_type;
1441     } else {
1442         s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
1443     }
1444 
1445     ret = validate_compression_type(s, errp);
1446     if (ret) {
1447         goto fail;
1448     }
1449 
1450     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
1451         void *feature_table = NULL;
1452         qcow2_read_extensions(bs, header.header_length, ext_end,
1453                               &feature_table, flags, NULL, NULL);
1454         report_unsupported_feature(errp, feature_table,
1455                                    s->incompatible_features &
1456                                    ~QCOW2_INCOMPAT_MASK);
1457         ret = -ENOTSUP;
1458         g_free(feature_table);
1459         goto fail;
1460     }
1461 
1462     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
1463         /* Corrupt images may not be written to unless they are being repaired
1464          */
1465         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
1466             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
1467                        "read/write");
1468             ret = -EACCES;
1469             goto fail;
1470         }
1471     }
1472 
1473     s->subclusters_per_cluster =
1474         has_subclusters(s) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER : 1;
1475     s->subcluster_size = s->cluster_size / s->subclusters_per_cluster;
1476     s->subcluster_bits = ctz32(s->subcluster_size);
1477 
1478     if (s->subcluster_size < (1 << MIN_CLUSTER_BITS)) {
1479         error_setg(errp, "Unsupported subcluster size: %d", s->subcluster_size);
1480         ret = -EINVAL;
1481         goto fail;
1482     }
1483 
1484     /* Check support for various header values */
1485     if (header.refcount_order > 6) {
1486         error_setg(errp, "Reference count entry width too large; may not "
1487                    "exceed 64 bits");
1488         ret = -EINVAL;
1489         goto fail;
1490     }
1491     s->refcount_order = header.refcount_order;
1492     s->refcount_bits = 1 << s->refcount_order;
1493     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1494     s->refcount_max += s->refcount_max - 1;
1495 
1496     s->crypt_method_header = header.crypt_method;
1497     if (s->crypt_method_header) {
1498         if (bdrv_uses_whitelist() &&
1499             s->crypt_method_header == QCOW_CRYPT_AES) {
1500             error_setg(errp,
1501                        "Use of AES-CBC encrypted qcow2 images is no longer "
1502                        "supported in system emulators");
1503             error_append_hint(errp,
1504                               "You can use 'qemu-img convert' to convert your "
1505                               "image to an alternative supported format, such "
1506                               "as unencrypted qcow2, or raw with the LUKS "
1507                               "format instead.\n");
1508             ret = -ENOSYS;
1509             goto fail;
1510         }
1511 
1512         if (s->crypt_method_header == QCOW_CRYPT_AES) {
1513             s->crypt_physical_offset = false;
1514         } else {
1515             /* Assuming LUKS and any future crypt methods we
1516              * add will all use physical offsets, due to the
1517              * fact that the alternative is insecure...  */
1518             s->crypt_physical_offset = true;
1519         }
1520 
1521         bs->encrypted = true;
1522     }
1523 
1524     s->l2_bits = s->cluster_bits - ctz32(l2_entry_size(s));
1525     s->l2_size = 1 << s->l2_bits;
1526     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1527     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1528     s->refcount_block_size = 1 << s->refcount_block_bits;
1529     bs->total_sectors = header.size / BDRV_SECTOR_SIZE;
1530     s->csize_shift = (62 - (s->cluster_bits - 8));
1531     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1532     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1533 
1534     s->refcount_table_offset = header.refcount_table_offset;
1535     s->refcount_table_size =
1536         header.refcount_table_clusters << (s->cluster_bits - 3);
1537 
1538     if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) {
1539         error_setg(errp, "Image does not contain a reference count table");
1540         ret = -EINVAL;
1541         goto fail;
1542     }
1543 
1544     ret = qcow2_validate_table(bs, s->refcount_table_offset,
1545                                header.refcount_table_clusters,
1546                                s->cluster_size, QCOW_MAX_REFTABLE_SIZE,
1547                                "Reference count table", errp);
1548     if (ret < 0) {
1549         goto fail;
1550     }
1551 
1552     if (!(flags & BDRV_O_CHECK)) {
1553         /*
1554          * The total size in bytes of the snapshot table is checked in
1555          * qcow2_read_snapshots() because the size of each snapshot is
1556          * variable and we don't know it yet.
1557          * Here we only check the offset and number of snapshots.
1558          */
1559         ret = qcow2_validate_table(bs, header.snapshots_offset,
1560                                    header.nb_snapshots,
1561                                    sizeof(QCowSnapshotHeader),
1562                                    sizeof(QCowSnapshotHeader) *
1563                                        QCOW_MAX_SNAPSHOTS,
1564                                    "Snapshot table", errp);
1565         if (ret < 0) {
1566             goto fail;
1567         }
1568     }
1569 
1570     /* read the level 1 table */
1571     ret = qcow2_validate_table(bs, header.l1_table_offset,
1572                                header.l1_size, L1E_SIZE,
1573                                QCOW_MAX_L1_SIZE, "Active L1 table", errp);
1574     if (ret < 0) {
1575         goto fail;
1576     }
1577     s->l1_size = header.l1_size;
1578     s->l1_table_offset = header.l1_table_offset;
1579 
1580     l1_vm_state_index = size_to_l1(s, header.size);
1581     if (l1_vm_state_index > INT_MAX) {
1582         error_setg(errp, "Image is too big");
1583         ret = -EFBIG;
1584         goto fail;
1585     }
1586     s->l1_vm_state_index = l1_vm_state_index;
1587 
1588     /* the L1 table must contain at least enough entries to put
1589        header.size bytes */
1590     if (s->l1_size < s->l1_vm_state_index) {
1591         error_setg(errp, "L1 table is too small");
1592         ret = -EINVAL;
1593         goto fail;
1594     }
1595 
1596     if (s->l1_size > 0) {
1597         s->l1_table = qemu_try_blockalign(bs->file->bs, s->l1_size * L1E_SIZE);
1598         if (s->l1_table == NULL) {
1599             error_setg(errp, "Could not allocate L1 table");
1600             ret = -ENOMEM;
1601             goto fail;
1602         }
1603         ret = bdrv_co_pread(bs->file, s->l1_table_offset, s->l1_size * L1E_SIZE,
1604                             s->l1_table, 0);
1605         if (ret < 0) {
1606             error_setg_errno(errp, -ret, "Could not read L1 table");
1607             goto fail;
1608         }
1609         for(i = 0;i < s->l1_size; i++) {
1610             s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
1611         }
1612     }
1613 
1614     /* Parse driver-specific options */
1615     ret = qcow2_update_options(bs, options, flags, errp);
1616     if (ret < 0) {
1617         goto fail;
1618     }
1619 
1620     s->flags = flags;
1621 
1622     ret = qcow2_refcount_init(bs);
1623     if (ret != 0) {
1624         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1625         goto fail;
1626     }
1627 
1628     QLIST_INIT(&s->cluster_allocs);
1629     QTAILQ_INIT(&s->discards);
1630 
1631     /* read qcow2 extensions */
1632     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1633                               flags, &update_header, errp)) {
1634         ret = -EINVAL;
1635         goto fail;
1636     }
1637 
1638     if (open_data_file) {
1639         /* Open external data file */
1640         bdrv_graph_co_rdunlock();
1641         s->data_file = bdrv_co_open_child(NULL, options, "data-file", bs,
1642                                           &child_of_bds, BDRV_CHILD_DATA,
1643                                           true, errp);
1644         bdrv_graph_co_rdlock();
1645         if (*errp) {
1646             ret = -EINVAL;
1647             goto fail;
1648         }
1649 
1650         if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) {
1651             if (!s->data_file && s->image_data_file) {
1652                 bdrv_graph_co_rdunlock();
1653                 s->data_file = bdrv_co_open_child(s->image_data_file, options,
1654                                                   "data-file", bs,
1655                                                   &child_of_bds,
1656                                                   BDRV_CHILD_DATA, false, errp);
1657                 bdrv_graph_co_rdlock();
1658                 if (!s->data_file) {
1659                     ret = -EINVAL;
1660                     goto fail;
1661                 }
1662             }
1663             if (!s->data_file) {
1664                 error_setg(errp, "'data-file' is required for this image");
1665                 ret = -EINVAL;
1666                 goto fail;
1667             }
1668 
1669             /* No data here */
1670             bs->file->role &= ~BDRV_CHILD_DATA;
1671 
1672             /* Must succeed because we have given up permissions if anything */
1673             bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1674         } else {
1675             if (s->data_file) {
1676                 error_setg(errp, "'data-file' can only be set for images with "
1677                                  "an external data file");
1678                 ret = -EINVAL;
1679                 goto fail;
1680             }
1681 
1682             s->data_file = bs->file;
1683 
1684             if (data_file_is_raw(bs)) {
1685                 error_setg(errp, "data-file-raw requires a data file");
1686                 ret = -EINVAL;
1687                 goto fail;
1688             }
1689         }
1690     }
1691 
1692     /* qcow2_read_extension may have set up the crypto context
1693      * if the crypt method needs a header region, some methods
1694      * don't need header extensions, so must check here
1695      */
1696     if (s->crypt_method_header && !s->crypto) {
1697         if (s->crypt_method_header == QCOW_CRYPT_AES) {
1698             unsigned int cflags = 0;
1699             if (flags & BDRV_O_NO_IO) {
1700                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1701             }
1702             s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
1703                                            NULL, NULL, cflags,
1704                                            QCOW2_MAX_THREADS, errp);
1705             if (!s->crypto) {
1706                 ret = -EINVAL;
1707                 goto fail;
1708             }
1709         } else if (!(flags & BDRV_O_NO_IO)) {
1710             error_setg(errp, "Missing CRYPTO header for crypt method %d",
1711                        s->crypt_method_header);
1712             ret = -EINVAL;
1713             goto fail;
1714         }
1715     }
1716 
1717     /* read the backing file name */
1718     if (header.backing_file_offset != 0) {
1719         len = header.backing_file_size;
1720         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1721             len >= sizeof(bs->backing_file)) {
1722             error_setg(errp, "Backing file name too long");
1723             ret = -EINVAL;
1724             goto fail;
1725         }
1726 
1727         s->image_backing_file = g_malloc(len + 1);
1728         ret = bdrv_co_pread(bs->file, header.backing_file_offset, len,
1729                             s->image_backing_file, 0);
1730         if (ret < 0) {
1731             error_setg_errno(errp, -ret, "Could not read backing file name");
1732             goto fail;
1733         }
1734         s->image_backing_file[len] = '\0';
1735 
1736         /*
1737          * Update only when something has changed.  This function is called by
1738          * qcow2_co_invalidate_cache(), and we do not want to reset
1739          * auto_backing_file unless necessary.
1740          */
1741         if (!g_str_equal(s->image_backing_file, bs->backing_file)) {
1742             pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1743                     s->image_backing_file);
1744             pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
1745                     s->image_backing_file);
1746         }
1747     }
1748 
1749     /*
1750      * Internal snapshots; skip reading them in check mode, because
1751      * we do not need them then, and we do not want to abort because
1752      * of a broken table.
1753      */
1754     if (!(flags & BDRV_O_CHECK)) {
1755         s->snapshots_offset = header.snapshots_offset;
1756         s->nb_snapshots = header.nb_snapshots;
1757 
1758         ret = qcow2_read_snapshots(bs, errp);
1759         if (ret < 0) {
1760             goto fail;
1761         }
1762     }
1763 
1764     /* Clear unknown autoclear feature bits */
1765     update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
1766     update_header = update_header && bdrv_is_writable(bs);
1767     if (update_header) {
1768         s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
1769     }
1770 
1771     /* == Handle persistent dirty bitmaps ==
1772      *
1773      * We want load dirty bitmaps in three cases:
1774      *
1775      * 1. Normal open of the disk in active mode, not related to invalidation
1776      *    after migration.
1777      *
1778      * 2. Invalidation of the target vm after pre-copy phase of migration, if
1779      *    bitmaps are _not_ migrating through migration channel, i.e.
1780      *    'dirty-bitmaps' capability is disabled.
1781      *
1782      * 3. Invalidation of source vm after failed or canceled migration.
1783      *    This is a very interesting case. There are two possible types of
1784      *    bitmaps:
1785      *
1786      *    A. Stored on inactivation and removed. They should be loaded from the
1787      *       image.
1788      *
1789      *    B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1790      *       the migration channel (with dirty-bitmaps capability).
1791      *
1792      *    On the other hand, there are two possible sub-cases:
1793      *
1794      *    3.1 disk was changed by somebody else while were inactive. In this
1795      *        case all in-RAM dirty bitmaps (both persistent and not) are
1796      *        definitely invalid. And we don't have any method to determine
1797      *        this.
1798      *
1799      *        Simple and safe thing is to just drop all the bitmaps of type B on
1800      *        inactivation. But in this case we lose bitmaps in valid 4.2 case.
1801      *
1802      *        On the other hand, resuming source vm, if disk was already changed
1803      *        is a bad thing anyway: not only bitmaps, the whole vm state is
1804      *        out of sync with disk.
1805      *
1806      *        This means, that user or management tool, who for some reason
1807      *        decided to resume source vm, after disk was already changed by
1808      *        target vm, should at least drop all dirty bitmaps by hand.
1809      *
1810      *        So, we can ignore this case for now, but TODO: "generation"
1811      *        extension for qcow2, to determine, that image was changed after
1812      *        last inactivation. And if it is changed, we will drop (or at least
1813      *        mark as 'invalid' all the bitmaps of type B, both persistent
1814      *        and not).
1815      *
1816      *    3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1817      *        to disk ('dirty-bitmaps' capability disabled), or not saved
1818      *        ('dirty-bitmaps' capability enabled), but we don't need to care
1819      *        of: let's load bitmaps as always: stored bitmaps will be loaded,
1820      *        and not stored has flag IN_USE=1 in the image and will be skipped
1821      *        on loading.
1822      *
1823      * One remaining possible case when we don't want load bitmaps:
1824      *
1825      * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1826      *    will be loaded on invalidation, no needs try loading them before)
1827      */
1828 
1829     if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) {
1830         /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1831         bool header_updated;
1832         if (!qcow2_load_dirty_bitmaps(bs, &header_updated, errp)) {
1833             ret = -EINVAL;
1834             goto fail;
1835         }
1836 
1837         update_header = update_header && !header_updated;
1838     }
1839 
1840     if (update_header) {
1841         ret = qcow2_update_header(bs);
1842         if (ret < 0) {
1843             error_setg_errno(errp, -ret, "Could not update qcow2 header");
1844             goto fail;
1845         }
1846     }
1847 
1848     bs->supported_zero_flags = header.version >= 3 ?
1849                                BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0;
1850     bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
1851 
1852     /* Repair image if dirty */
1853     if (!(flags & BDRV_O_CHECK) && bdrv_is_writable(bs) &&
1854         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1855         BdrvCheckResult result = {0};
1856 
1857         ret = qcow2_co_check_locked(bs, &result,
1858                                     BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1859         if (ret < 0 || result.check_errors) {
1860             if (ret >= 0) {
1861                 ret = -EIO;
1862             }
1863             error_setg_errno(errp, -ret, "Could not repair dirty image");
1864             goto fail;
1865         }
1866     }
1867 
1868 #ifdef DEBUG_ALLOC
1869     {
1870         BdrvCheckResult result = {0};
1871         qcow2_check_refcounts(bs, &result, 0);
1872     }
1873 #endif
1874 
1875     qemu_co_queue_init(&s->thread_task_queue);
1876 
1877     return ret;
1878 
1879  fail:
1880     g_free(s->image_data_file);
1881     if (open_data_file && has_data_file(bs)) {
1882         bdrv_graph_co_rdunlock();
1883         bdrv_co_unref_child(bs, s->data_file);
1884         bdrv_graph_co_rdlock();
1885         s->data_file = NULL;
1886     }
1887     g_free(s->unknown_header_fields);
1888     cleanup_unknown_header_ext(bs);
1889     qcow2_free_snapshots(bs);
1890     qcow2_refcount_close(bs);
1891     qemu_vfree(s->l1_table);
1892     /* else pre-write overlap checks in cache_destroy may crash */
1893     s->l1_table = NULL;
1894     cache_clean_timer_del(bs);
1895     if (s->l2_table_cache) {
1896         qcow2_cache_destroy(s->l2_table_cache);
1897     }
1898     if (s->refcount_block_cache) {
1899         qcow2_cache_destroy(s->refcount_block_cache);
1900     }
1901     qcrypto_block_free(s->crypto);
1902     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1903     return ret;
1904 }
1905 
1906 typedef struct QCow2OpenCo {
1907     BlockDriverState *bs;
1908     QDict *options;
1909     int flags;
1910     Error **errp;
1911     int ret;
1912 } QCow2OpenCo;
1913 
1914 static void coroutine_fn qcow2_open_entry(void *opaque)
1915 {
1916     QCow2OpenCo *qoc = opaque;
1917     BDRVQcow2State *s = qoc->bs->opaque;
1918 
1919     GRAPH_RDLOCK_GUARD();
1920 
1921     qemu_co_mutex_lock(&s->lock);
1922     qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, true,
1923                              qoc->errp);
1924     qemu_co_mutex_unlock(&s->lock);
1925 
1926     aio_wait_kick();
1927 }
1928 
1929 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1930                       Error **errp)
1931 {
1932     BDRVQcow2State *s = bs->opaque;
1933     QCow2OpenCo qoc = {
1934         .bs = bs,
1935         .options = options,
1936         .flags = flags,
1937         .errp = errp,
1938         .ret = -EINPROGRESS
1939     };
1940     int ret;
1941 
1942     ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
1943     if (ret < 0) {
1944         return ret;
1945     }
1946 
1947     /* Initialise locks */
1948     qemu_co_mutex_init(&s->lock);
1949 
1950     assert(!qemu_in_coroutine());
1951     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1952 
1953     aio_co_enter(bdrv_get_aio_context(bs),
1954                  qemu_coroutine_create(qcow2_open_entry, &qoc));
1955     AIO_WAIT_WHILE_UNLOCKED(NULL, qoc.ret == -EINPROGRESS);
1956 
1957     return qoc.ret;
1958 }
1959 
1960 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1961 {
1962     BDRVQcow2State *s = bs->opaque;
1963 
1964     if (bs->encrypted) {
1965         /* Encryption works on a sector granularity */
1966         bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto);
1967     }
1968     bs->bl.pwrite_zeroes_alignment = s->subcluster_size;
1969     bs->bl.pdiscard_alignment = s->cluster_size;
1970 }
1971 
1972 static int GRAPH_UNLOCKED
1973 qcow2_reopen_prepare(BDRVReopenState *state,BlockReopenQueue *queue,
1974                      Error **errp)
1975 {
1976     BDRVQcow2State *s = state->bs->opaque;
1977     Qcow2ReopenState *r;
1978     int ret;
1979 
1980     GLOBAL_STATE_CODE();
1981     GRAPH_RDLOCK_GUARD_MAINLOOP();
1982 
1983     r = g_new0(Qcow2ReopenState, 1);
1984     state->opaque = r;
1985 
1986     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1987                                        state->flags, errp);
1988     if (ret < 0) {
1989         goto fail;
1990     }
1991 
1992     /* We need to write out any unwritten data if we reopen read-only. */
1993     if ((state->flags & BDRV_O_RDWR) == 0) {
1994         ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1995         if (ret < 0) {
1996             goto fail;
1997         }
1998 
1999         ret = bdrv_flush(state->bs);
2000         if (ret < 0) {
2001             goto fail;
2002         }
2003 
2004         ret = qcow2_mark_clean(state->bs);
2005         if (ret < 0) {
2006             goto fail;
2007         }
2008     }
2009 
2010     /*
2011      * Without an external data file, s->data_file points to the same BdrvChild
2012      * as bs->file. It needs to be resynced after reopen because bs->file may
2013      * be changed. We can't use it in the meantime.
2014      */
2015     if (!has_data_file(state->bs)) {
2016         assert(s->data_file == state->bs->file);
2017         s->data_file = NULL;
2018     }
2019 
2020     return 0;
2021 
2022 fail:
2023     qcow2_update_options_abort(state->bs, r);
2024     g_free(r);
2025     return ret;
2026 }
2027 
2028 static void qcow2_reopen_commit(BDRVReopenState *state)
2029 {
2030     BDRVQcow2State *s = state->bs->opaque;
2031 
2032     qcow2_update_options_commit(state->bs, state->opaque);
2033     if (!s->data_file) {
2034         /*
2035          * If we don't have an external data file, s->data_file was cleared by
2036          * qcow2_reopen_prepare() and needs to be updated.
2037          */
2038         s->data_file = state->bs->file;
2039     }
2040     g_free(state->opaque);
2041 }
2042 
2043 static void qcow2_reopen_commit_post(BDRVReopenState *state)
2044 {
2045     GRAPH_RDLOCK_GUARD_MAINLOOP();
2046 
2047     if (state->flags & BDRV_O_RDWR) {
2048         Error *local_err = NULL;
2049 
2050         if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) {
2051             /*
2052              * This is not fatal, bitmaps just left read-only, so all following
2053              * writes will fail. User can remove read-only bitmaps to unblock
2054              * writes or retry reopen.
2055              */
2056             error_reportf_err(local_err,
2057                               "%s: Failed to make dirty bitmaps writable: ",
2058                               bdrv_get_node_name(state->bs));
2059         }
2060     }
2061 }
2062 
2063 static void qcow2_reopen_abort(BDRVReopenState *state)
2064 {
2065     BDRVQcow2State *s = state->bs->opaque;
2066 
2067     if (!s->data_file) {
2068         /*
2069          * If we don't have an external data file, s->data_file was cleared by
2070          * qcow2_reopen_prepare() and needs to be restored.
2071          */
2072         s->data_file = state->bs->file;
2073     }
2074     qcow2_update_options_abort(state->bs, state->opaque);
2075     g_free(state->opaque);
2076 }
2077 
2078 static void qcow2_join_options(QDict *options, QDict *old_options)
2079 {
2080     bool has_new_overlap_template =
2081         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
2082         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
2083     bool has_new_total_cache_size =
2084         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
2085     bool has_all_cache_options;
2086 
2087     /* New overlap template overrides all old overlap options */
2088     if (has_new_overlap_template) {
2089         qdict_del(old_options, QCOW2_OPT_OVERLAP);
2090         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
2091         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
2092         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
2093         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
2094         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
2095         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
2096         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
2097         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
2098         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
2099     }
2100 
2101     /* New total cache size overrides all old options */
2102     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
2103         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
2104         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2105     }
2106 
2107     qdict_join(options, old_options, false);
2108 
2109     /*
2110      * If after merging all cache size options are set, an old total size is
2111      * overwritten. Do keep all options, however, if all three are new. The
2112      * resulting error message is what we want to happen.
2113      */
2114     has_all_cache_options =
2115         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
2116         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
2117         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2118 
2119     if (has_all_cache_options && !has_new_total_cache_size) {
2120         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
2121     }
2122 }
2123 
2124 static int coroutine_fn GRAPH_RDLOCK
2125 qcow2_co_block_status(BlockDriverState *bs, bool want_zero, int64_t offset,
2126                       int64_t count, int64_t *pnum, int64_t *map,
2127                       BlockDriverState **file)
2128 {
2129     BDRVQcow2State *s = bs->opaque;
2130     uint64_t host_offset;
2131     unsigned int bytes;
2132     QCow2SubclusterType type;
2133     int ret, status = 0;
2134 
2135     qemu_co_mutex_lock(&s->lock);
2136 
2137     if (!s->metadata_preallocation_checked) {
2138         ret = qcow2_detect_metadata_preallocation(bs);
2139         s->metadata_preallocation = (ret == 1);
2140         s->metadata_preallocation_checked = true;
2141     }
2142 
2143     bytes = MIN(INT_MAX, count);
2144     ret = qcow2_get_host_offset(bs, offset, &bytes, &host_offset, &type);
2145     qemu_co_mutex_unlock(&s->lock);
2146     if (ret < 0) {
2147         return ret;
2148     }
2149 
2150     *pnum = bytes;
2151 
2152     if ((type == QCOW2_SUBCLUSTER_NORMAL ||
2153          type == QCOW2_SUBCLUSTER_ZERO_ALLOC ||
2154          type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) && !s->crypto) {
2155         *map = host_offset;
2156         *file = s->data_file->bs;
2157         status |= BDRV_BLOCK_OFFSET_VALID;
2158     }
2159     if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN ||
2160         type == QCOW2_SUBCLUSTER_ZERO_ALLOC) {
2161         status |= BDRV_BLOCK_ZERO;
2162     } else if (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN &&
2163                type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) {
2164         status |= BDRV_BLOCK_DATA;
2165     }
2166     if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) &&
2167         (status & BDRV_BLOCK_OFFSET_VALID))
2168     {
2169         status |= BDRV_BLOCK_RECURSE;
2170     }
2171     if (type == QCOW2_SUBCLUSTER_COMPRESSED) {
2172         status |= BDRV_BLOCK_COMPRESSED;
2173     }
2174     return status;
2175 }
2176 
2177 static int coroutine_fn GRAPH_RDLOCK
2178 qcow2_handle_l2meta(BlockDriverState *bs, QCowL2Meta **pl2meta, bool link_l2)
2179 {
2180     int ret = 0;
2181     QCowL2Meta *l2meta = *pl2meta;
2182 
2183     while (l2meta != NULL) {
2184         QCowL2Meta *next;
2185 
2186         if (link_l2) {
2187             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
2188             if (ret) {
2189                 goto out;
2190             }
2191         } else {
2192             qcow2_alloc_cluster_abort(bs, l2meta);
2193         }
2194 
2195         /* Take the request off the list of running requests */
2196         QLIST_REMOVE(l2meta, next_in_flight);
2197 
2198         qemu_co_queue_restart_all(&l2meta->dependent_requests);
2199 
2200         next = l2meta->next;
2201         g_free(l2meta);
2202         l2meta = next;
2203     }
2204 out:
2205     *pl2meta = l2meta;
2206     return ret;
2207 }
2208 
2209 static int coroutine_fn GRAPH_RDLOCK
2210 qcow2_co_preadv_encrypted(BlockDriverState *bs,
2211                            uint64_t host_offset,
2212                            uint64_t offset,
2213                            uint64_t bytes,
2214                            QEMUIOVector *qiov,
2215                            uint64_t qiov_offset)
2216 {
2217     int ret;
2218     BDRVQcow2State *s = bs->opaque;
2219     uint8_t *buf;
2220 
2221     assert(bs->encrypted && s->crypto);
2222     assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2223 
2224     /*
2225      * For encrypted images, read everything into a temporary
2226      * contiguous buffer on which the AES functions can work.
2227      * Also, decryption in a separate buffer is better as it
2228      * prevents the guest from learning information about the
2229      * encrypted nature of the virtual disk.
2230      */
2231 
2232     buf = qemu_try_blockalign(s->data_file->bs, bytes);
2233     if (buf == NULL) {
2234         return -ENOMEM;
2235     }
2236 
2237     BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_AIO);
2238     ret = bdrv_co_pread(s->data_file, host_offset, bytes, buf, 0);
2239     if (ret < 0) {
2240         goto fail;
2241     }
2242 
2243     if (qcow2_co_decrypt(bs, host_offset, offset, buf, bytes) < 0)
2244     {
2245         ret = -EIO;
2246         goto fail;
2247     }
2248     qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes);
2249 
2250 fail:
2251     qemu_vfree(buf);
2252 
2253     return ret;
2254 }
2255 
2256 typedef struct Qcow2AioTask {
2257     AioTask task;
2258 
2259     BlockDriverState *bs;
2260     QCow2SubclusterType subcluster_type; /* only for read */
2261     uint64_t host_offset; /* or l2_entry for compressed read */
2262     uint64_t offset;
2263     uint64_t bytes;
2264     QEMUIOVector *qiov;
2265     uint64_t qiov_offset;
2266     QCowL2Meta *l2meta; /* only for write */
2267 } Qcow2AioTask;
2268 
2269 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task);
2270 static coroutine_fn int qcow2_add_task(BlockDriverState *bs,
2271                                        AioTaskPool *pool,
2272                                        AioTaskFunc func,
2273                                        QCow2SubclusterType subcluster_type,
2274                                        uint64_t host_offset,
2275                                        uint64_t offset,
2276                                        uint64_t bytes,
2277                                        QEMUIOVector *qiov,
2278                                        size_t qiov_offset,
2279                                        QCowL2Meta *l2meta)
2280 {
2281     Qcow2AioTask local_task;
2282     Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task;
2283 
2284     *task = (Qcow2AioTask) {
2285         .task.func = func,
2286         .bs = bs,
2287         .subcluster_type = subcluster_type,
2288         .qiov = qiov,
2289         .host_offset = host_offset,
2290         .offset = offset,
2291         .bytes = bytes,
2292         .qiov_offset = qiov_offset,
2293         .l2meta = l2meta,
2294     };
2295 
2296     trace_qcow2_add_task(qemu_coroutine_self(), bs, pool,
2297                          func == qcow2_co_preadv_task_entry ? "read" : "write",
2298                          subcluster_type, host_offset, offset, bytes,
2299                          qiov, qiov_offset);
2300 
2301     if (!pool) {
2302         return func(&task->task);
2303     }
2304 
2305     aio_task_pool_start_task(pool, &task->task);
2306 
2307     return 0;
2308 }
2309 
2310 static int coroutine_fn GRAPH_RDLOCK
2311 qcow2_co_preadv_task(BlockDriverState *bs, QCow2SubclusterType subc_type,
2312                      uint64_t host_offset, uint64_t offset, uint64_t bytes,
2313                      QEMUIOVector *qiov, size_t qiov_offset)
2314 {
2315     BDRVQcow2State *s = bs->opaque;
2316 
2317     switch (subc_type) {
2318     case QCOW2_SUBCLUSTER_ZERO_PLAIN:
2319     case QCOW2_SUBCLUSTER_ZERO_ALLOC:
2320         /* Both zero types are handled in qcow2_co_preadv_part */
2321         g_assert_not_reached();
2322 
2323     case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
2324     case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
2325         assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */
2326 
2327         BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
2328         return bdrv_co_preadv_part(bs->backing, offset, bytes,
2329                                    qiov, qiov_offset, 0);
2330 
2331     case QCOW2_SUBCLUSTER_COMPRESSED:
2332         return qcow2_co_preadv_compressed(bs, host_offset,
2333                                           offset, bytes, qiov, qiov_offset);
2334 
2335     case QCOW2_SUBCLUSTER_NORMAL:
2336         if (bs->encrypted) {
2337             return qcow2_co_preadv_encrypted(bs, host_offset,
2338                                              offset, bytes, qiov, qiov_offset);
2339         }
2340 
2341         BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_AIO);
2342         return bdrv_co_preadv_part(s->data_file, host_offset,
2343                                    bytes, qiov, qiov_offset, 0);
2344 
2345     default:
2346         g_assert_not_reached();
2347     }
2348 
2349     g_assert_not_reached();
2350 }
2351 
2352 /*
2353  * This function can count as GRAPH_RDLOCK because qcow2_co_preadv_part() holds
2354  * the graph lock and keeps it until this coroutine has terminated.
2355  */
2356 static int coroutine_fn GRAPH_RDLOCK qcow2_co_preadv_task_entry(AioTask *task)
2357 {
2358     Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2359 
2360     assert(!t->l2meta);
2361 
2362     return qcow2_co_preadv_task(t->bs, t->subcluster_type,
2363                                 t->host_offset, t->offset, t->bytes,
2364                                 t->qiov, t->qiov_offset);
2365 }
2366 
2367 static int coroutine_fn GRAPH_RDLOCK
2368 qcow2_co_preadv_part(BlockDriverState *bs, int64_t offset, int64_t bytes,
2369                      QEMUIOVector *qiov, size_t qiov_offset,
2370                      BdrvRequestFlags flags)
2371 {
2372     BDRVQcow2State *s = bs->opaque;
2373     int ret = 0;
2374     unsigned int cur_bytes; /* number of bytes in current iteration */
2375     uint64_t host_offset = 0;
2376     QCow2SubclusterType type;
2377     AioTaskPool *aio = NULL;
2378 
2379     while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2380         /* prepare next request */
2381         cur_bytes = MIN(bytes, INT_MAX);
2382         if (s->crypto) {
2383             cur_bytes = MIN(cur_bytes,
2384                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2385         }
2386 
2387         qemu_co_mutex_lock(&s->lock);
2388         ret = qcow2_get_host_offset(bs, offset, &cur_bytes,
2389                                     &host_offset, &type);
2390         qemu_co_mutex_unlock(&s->lock);
2391         if (ret < 0) {
2392             goto out;
2393         }
2394 
2395         if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN ||
2396             type == QCOW2_SUBCLUSTER_ZERO_ALLOC ||
2397             (type == QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && !bs->backing) ||
2398             (type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && !bs->backing))
2399         {
2400             qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes);
2401         } else {
2402             if (!aio && cur_bytes != bytes) {
2403                 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2404             }
2405             ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, type,
2406                                  host_offset, offset, cur_bytes,
2407                                  qiov, qiov_offset, NULL);
2408             if (ret < 0) {
2409                 goto out;
2410             }
2411         }
2412 
2413         bytes -= cur_bytes;
2414         offset += cur_bytes;
2415         qiov_offset += cur_bytes;
2416     }
2417 
2418 out:
2419     if (aio) {
2420         aio_task_pool_wait_all(aio);
2421         if (ret == 0) {
2422             ret = aio_task_pool_status(aio);
2423         }
2424         g_free(aio);
2425     }
2426 
2427     return ret;
2428 }
2429 
2430 /* Check if it's possible to merge a write request with the writing of
2431  * the data from the COW regions */
2432 static bool merge_cow(uint64_t offset, unsigned bytes,
2433                       QEMUIOVector *qiov, size_t qiov_offset,
2434                       QCowL2Meta *l2meta)
2435 {
2436     QCowL2Meta *m;
2437 
2438     for (m = l2meta; m != NULL; m = m->next) {
2439         /* If both COW regions are empty then there's nothing to merge */
2440         if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
2441             continue;
2442         }
2443 
2444         /* If COW regions are handled already, skip this too */
2445         if (m->skip_cow) {
2446             continue;
2447         }
2448 
2449         /*
2450          * The write request should start immediately after the first
2451          * COW region. This does not always happen because the area
2452          * touched by the request can be larger than the one defined
2453          * by @m (a single request can span an area consisting of a
2454          * mix of previously unallocated and allocated clusters, that
2455          * is why @l2meta is a list).
2456          */
2457         if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
2458             /* In this case the request starts before this region */
2459             assert(offset < l2meta_cow_start(m));
2460             assert(m->cow_start.nb_bytes == 0);
2461             continue;
2462         }
2463 
2464         /* The write request should end immediately before the second
2465          * COW region (see above for why it does not always happen) */
2466         if (m->offset + m->cow_end.offset != offset + bytes) {
2467             assert(offset + bytes > m->offset + m->cow_end.offset);
2468             assert(m->cow_end.nb_bytes == 0);
2469             continue;
2470         }
2471 
2472         /* Make sure that adding both COW regions to the QEMUIOVector
2473          * does not exceed IOV_MAX */
2474         if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) {
2475             continue;
2476         }
2477 
2478         m->data_qiov = qiov;
2479         m->data_qiov_offset = qiov_offset;
2480         return true;
2481     }
2482 
2483     return false;
2484 }
2485 
2486 /*
2487  * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error.
2488  * Note that returning 0 does not guarantee non-zero data.
2489  */
2490 static int coroutine_fn GRAPH_RDLOCK
2491 is_zero_cow(BlockDriverState *bs, QCowL2Meta *m)
2492 {
2493     /*
2494      * This check is designed for optimization shortcut so it must be
2495      * efficient.
2496      * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is
2497      * faster (but not as accurate and can result in false negatives).
2498      */
2499     int ret = bdrv_co_is_zero_fast(bs, m->offset + m->cow_start.offset,
2500                                    m->cow_start.nb_bytes);
2501     if (ret <= 0) {
2502         return ret;
2503     }
2504 
2505     return bdrv_co_is_zero_fast(bs, m->offset + m->cow_end.offset,
2506                                 m->cow_end.nb_bytes);
2507 }
2508 
2509 static int coroutine_fn GRAPH_RDLOCK
2510 handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
2511 {
2512     BDRVQcow2State *s = bs->opaque;
2513     QCowL2Meta *m;
2514 
2515     if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) {
2516         return 0;
2517     }
2518 
2519     if (bs->encrypted) {
2520         return 0;
2521     }
2522 
2523     for (m = l2meta; m != NULL; m = m->next) {
2524         int ret;
2525         uint64_t start_offset = m->alloc_offset + m->cow_start.offset;
2526         unsigned nb_bytes = m->cow_end.offset + m->cow_end.nb_bytes -
2527             m->cow_start.offset;
2528 
2529         if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
2530             continue;
2531         }
2532 
2533         ret = is_zero_cow(bs, m);
2534         if (ret < 0) {
2535             return ret;
2536         } else if (ret == 0) {
2537             continue;
2538         }
2539 
2540         /*
2541          * instead of writing zero COW buffers,
2542          * efficiently zero out the whole clusters
2543          */
2544 
2545         ret = qcow2_pre_write_overlap_check(bs, 0, start_offset, nb_bytes,
2546                                             true);
2547         if (ret < 0) {
2548             return ret;
2549         }
2550 
2551         BLKDBG_CO_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE);
2552         ret = bdrv_co_pwrite_zeroes(s->data_file, start_offset, nb_bytes,
2553                                     BDRV_REQ_NO_FALLBACK);
2554         if (ret < 0) {
2555             if (ret != -ENOTSUP && ret != -EAGAIN) {
2556                 return ret;
2557             }
2558             continue;
2559         }
2560 
2561         trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters);
2562         m->skip_cow = true;
2563     }
2564     return 0;
2565 }
2566 
2567 /*
2568  * qcow2_co_pwritev_task
2569  * Called with s->lock unlocked
2570  * l2meta  - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2571  *           not use it somehow after qcow2_co_pwritev_task() call
2572  */
2573 static coroutine_fn GRAPH_RDLOCK
2574 int qcow2_co_pwritev_task(BlockDriverState *bs, uint64_t host_offset,
2575                           uint64_t offset, uint64_t bytes, QEMUIOVector *qiov,
2576                           uint64_t qiov_offset, QCowL2Meta *l2meta)
2577 {
2578     int ret;
2579     BDRVQcow2State *s = bs->opaque;
2580     void *crypt_buf = NULL;
2581     QEMUIOVector encrypted_qiov;
2582 
2583     if (bs->encrypted) {
2584         assert(s->crypto);
2585         assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2586         crypt_buf = qemu_try_blockalign(bs->file->bs, bytes);
2587         if (crypt_buf == NULL) {
2588             ret = -ENOMEM;
2589             goto out_unlocked;
2590         }
2591         qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes);
2592 
2593         if (qcow2_co_encrypt(bs, host_offset, offset, crypt_buf, bytes) < 0) {
2594             ret = -EIO;
2595             goto out_unlocked;
2596         }
2597 
2598         qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes);
2599         qiov = &encrypted_qiov;
2600         qiov_offset = 0;
2601     }
2602 
2603     /* Try to efficiently initialize the physical space with zeroes */
2604     ret = handle_alloc_space(bs, l2meta);
2605     if (ret < 0) {
2606         goto out_unlocked;
2607     }
2608 
2609     /*
2610      * If we need to do COW, check if it's possible to merge the
2611      * writing of the guest data together with that of the COW regions.
2612      * If it's not possible (or not necessary) then write the
2613      * guest data now.
2614      */
2615     if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) {
2616         BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_AIO);
2617         trace_qcow2_writev_data(qemu_coroutine_self(), host_offset);
2618         ret = bdrv_co_pwritev_part(s->data_file, host_offset,
2619                                    bytes, qiov, qiov_offset, 0);
2620         if (ret < 0) {
2621             goto out_unlocked;
2622         }
2623     }
2624 
2625     qemu_co_mutex_lock(&s->lock);
2626 
2627     ret = qcow2_handle_l2meta(bs, &l2meta, true);
2628     goto out_locked;
2629 
2630 out_unlocked:
2631     qemu_co_mutex_lock(&s->lock);
2632 
2633 out_locked:
2634     qcow2_handle_l2meta(bs, &l2meta, false);
2635     qemu_co_mutex_unlock(&s->lock);
2636 
2637     qemu_vfree(crypt_buf);
2638 
2639     return ret;
2640 }
2641 
2642 /*
2643  * This function can count as GRAPH_RDLOCK because qcow2_co_pwritev_part() holds
2644  * the graph lock and keeps it until this coroutine has terminated.
2645  */
2646 static coroutine_fn GRAPH_RDLOCK int qcow2_co_pwritev_task_entry(AioTask *task)
2647 {
2648     Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2649 
2650     assert(!t->subcluster_type);
2651 
2652     return qcow2_co_pwritev_task(t->bs, t->host_offset,
2653                                  t->offset, t->bytes, t->qiov, t->qiov_offset,
2654                                  t->l2meta);
2655 }
2656 
2657 static int coroutine_fn GRAPH_RDLOCK
2658 qcow2_co_pwritev_part(BlockDriverState *bs, int64_t offset, int64_t bytes,
2659                       QEMUIOVector *qiov, size_t qiov_offset,
2660                       BdrvRequestFlags flags)
2661 {
2662     BDRVQcow2State *s = bs->opaque;
2663     int offset_in_cluster;
2664     int ret;
2665     unsigned int cur_bytes; /* number of sectors in current iteration */
2666     uint64_t host_offset;
2667     QCowL2Meta *l2meta = NULL;
2668     AioTaskPool *aio = NULL;
2669 
2670     trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
2671 
2672     while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2673 
2674         l2meta = NULL;
2675 
2676         trace_qcow2_writev_start_part(qemu_coroutine_self());
2677         offset_in_cluster = offset_into_cluster(s, offset);
2678         cur_bytes = MIN(bytes, INT_MAX);
2679         if (bs->encrypted) {
2680             cur_bytes = MIN(cur_bytes,
2681                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
2682                             - offset_in_cluster);
2683         }
2684 
2685         qemu_co_mutex_lock(&s->lock);
2686 
2687         ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes,
2688                                       &host_offset, &l2meta);
2689         if (ret < 0) {
2690             goto out_locked;
2691         }
2692 
2693         ret = qcow2_pre_write_overlap_check(bs, 0, host_offset,
2694                                             cur_bytes, true);
2695         if (ret < 0) {
2696             goto out_locked;
2697         }
2698 
2699         qemu_co_mutex_unlock(&s->lock);
2700 
2701         if (!aio && cur_bytes != bytes) {
2702             aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2703         }
2704         ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0,
2705                              host_offset, offset,
2706                              cur_bytes, qiov, qiov_offset, l2meta);
2707         l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */
2708         if (ret < 0) {
2709             goto fail_nometa;
2710         }
2711 
2712         bytes -= cur_bytes;
2713         offset += cur_bytes;
2714         qiov_offset += cur_bytes;
2715         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2716     }
2717     ret = 0;
2718 
2719     qemu_co_mutex_lock(&s->lock);
2720 
2721 out_locked:
2722     qcow2_handle_l2meta(bs, &l2meta, false);
2723 
2724     qemu_co_mutex_unlock(&s->lock);
2725 
2726 fail_nometa:
2727     if (aio) {
2728         aio_task_pool_wait_all(aio);
2729         if (ret == 0) {
2730             ret = aio_task_pool_status(aio);
2731         }
2732         g_free(aio);
2733     }
2734 
2735     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2736 
2737     return ret;
2738 }
2739 
2740 static int GRAPH_RDLOCK qcow2_inactivate(BlockDriverState *bs)
2741 {
2742     BDRVQcow2State *s = bs->opaque;
2743     int ret, result = 0;
2744     Error *local_err = NULL;
2745 
2746     qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err);
2747     if (local_err != NULL) {
2748         result = -EINVAL;
2749         error_reportf_err(local_err, "Lost persistent bitmaps during "
2750                           "inactivation of node '%s': ",
2751                           bdrv_get_device_or_node_name(bs));
2752     }
2753 
2754     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2755     if (ret) {
2756         result = ret;
2757         error_report("Failed to flush the L2 table cache: %s",
2758                      strerror(-ret));
2759     }
2760 
2761     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2762     if (ret) {
2763         result = ret;
2764         error_report("Failed to flush the refcount block cache: %s",
2765                      strerror(-ret));
2766     }
2767 
2768     if (result == 0) {
2769         qcow2_mark_clean(bs);
2770     }
2771 
2772     return result;
2773 }
2774 
2775 static void coroutine_mixed_fn GRAPH_RDLOCK
2776 qcow2_do_close(BlockDriverState *bs, bool close_data_file)
2777 {
2778     BDRVQcow2State *s = bs->opaque;
2779     qemu_vfree(s->l1_table);
2780     /* else pre-write overlap checks in cache_destroy may crash */
2781     s->l1_table = NULL;
2782 
2783     if (!(s->flags & BDRV_O_INACTIVE)) {
2784         qcow2_inactivate(bs);
2785     }
2786 
2787     cache_clean_timer_del(bs);
2788     qcow2_cache_destroy(s->l2_table_cache);
2789     qcow2_cache_destroy(s->refcount_block_cache);
2790 
2791     qcrypto_block_free(s->crypto);
2792     s->crypto = NULL;
2793     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
2794 
2795     g_free(s->unknown_header_fields);
2796     cleanup_unknown_header_ext(bs);
2797 
2798     g_free(s->image_data_file);
2799     g_free(s->image_backing_file);
2800     g_free(s->image_backing_format);
2801 
2802     if (close_data_file && has_data_file(bs)) {
2803         GLOBAL_STATE_CODE();
2804         bdrv_graph_rdunlock_main_loop();
2805         bdrv_graph_wrlock(NULL);
2806         bdrv_unref_child(bs, s->data_file);
2807         bdrv_graph_wrunlock();
2808         s->data_file = NULL;
2809         bdrv_graph_rdlock_main_loop();
2810     }
2811 
2812     qcow2_refcount_close(bs);
2813     qcow2_free_snapshots(bs);
2814 }
2815 
2816 static void GRAPH_UNLOCKED qcow2_close(BlockDriverState *bs)
2817 {
2818     GLOBAL_STATE_CODE();
2819     GRAPH_RDLOCK_GUARD_MAINLOOP();
2820 
2821     qcow2_do_close(bs, true);
2822 }
2823 
2824 static void coroutine_fn GRAPH_RDLOCK
2825 qcow2_co_invalidate_cache(BlockDriverState *bs, Error **errp)
2826 {
2827     ERRP_GUARD();
2828     BDRVQcow2State *s = bs->opaque;
2829     BdrvChild *data_file;
2830     int flags = s->flags;
2831     QCryptoBlock *crypto = NULL;
2832     QDict *options;
2833     int ret;
2834 
2835     /*
2836      * Backing files are read-only which makes all of their metadata immutable,
2837      * that means we don't have to worry about reopening them here.
2838      */
2839 
2840     crypto = s->crypto;
2841     s->crypto = NULL;
2842 
2843     /*
2844      * Do not reopen s->data_file (i.e., have qcow2_do_close() not close it,
2845      * and then prevent qcow2_do_open() from opening it), because this function
2846      * runs in the I/O path and as such we must not invoke global-state
2847      * functions like bdrv_unref_child() and bdrv_open_child().
2848      */
2849 
2850     qcow2_do_close(bs, false);
2851 
2852     data_file = s->data_file;
2853     memset(s, 0, sizeof(BDRVQcow2State));
2854     s->data_file = data_file;
2855 
2856     options = qdict_clone_shallow(bs->options);
2857 
2858     flags &= ~BDRV_O_INACTIVE;
2859     qemu_co_mutex_lock(&s->lock);
2860     ret = qcow2_do_open(bs, options, flags, false, errp);
2861     qemu_co_mutex_unlock(&s->lock);
2862     qobject_unref(options);
2863     if (ret < 0) {
2864         error_prepend(errp, "Could not reopen qcow2 layer: ");
2865         bs->drv = NULL;
2866         return;
2867     }
2868 
2869     s->crypto = crypto;
2870 }
2871 
2872 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2873     size_t len, size_t buflen)
2874 {
2875     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2876     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2877 
2878     if (buflen < ext_len) {
2879         return -ENOSPC;
2880     }
2881 
2882     *ext_backing_fmt = (QCowExtension) {
2883         .magic  = cpu_to_be32(magic),
2884         .len    = cpu_to_be32(len),
2885     };
2886 
2887     if (len) {
2888         memcpy(buf + sizeof(QCowExtension), s, len);
2889     }
2890 
2891     return ext_len;
2892 }
2893 
2894 /*
2895  * Updates the qcow2 header, including the variable length parts of it, i.e.
2896  * the backing file name and all extensions. qcow2 was not designed to allow
2897  * such changes, so if we run out of space (we can only use the first cluster)
2898  * this function may fail.
2899  *
2900  * Returns 0 on success, -errno in error cases.
2901  */
2902 int qcow2_update_header(BlockDriverState *bs)
2903 {
2904     BDRVQcow2State *s = bs->opaque;
2905     QCowHeader *header;
2906     char *buf;
2907     size_t buflen = s->cluster_size;
2908     int ret;
2909     uint64_t total_size;
2910     uint32_t refcount_table_clusters;
2911     size_t header_length;
2912     Qcow2UnknownHeaderExtension *uext;
2913 
2914     buf = qemu_blockalign(bs, buflen);
2915 
2916     /* Header structure */
2917     header = (QCowHeader*) buf;
2918 
2919     if (buflen < sizeof(*header)) {
2920         ret = -ENOSPC;
2921         goto fail;
2922     }
2923 
2924     header_length = sizeof(*header) + s->unknown_header_fields_size;
2925     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2926     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2927 
2928     ret = validate_compression_type(s, NULL);
2929     if (ret) {
2930         goto fail;
2931     }
2932 
2933     *header = (QCowHeader) {
2934         /* Version 2 fields */
2935         .magic                  = cpu_to_be32(QCOW_MAGIC),
2936         .version                = cpu_to_be32(s->qcow_version),
2937         .backing_file_offset    = 0,
2938         .backing_file_size      = 0,
2939         .cluster_bits           = cpu_to_be32(s->cluster_bits),
2940         .size                   = cpu_to_be64(total_size),
2941         .crypt_method           = cpu_to_be32(s->crypt_method_header),
2942         .l1_size                = cpu_to_be32(s->l1_size),
2943         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
2944         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
2945         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2946         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
2947         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
2948 
2949         /* Version 3 fields */
2950         .incompatible_features  = cpu_to_be64(s->incompatible_features),
2951         .compatible_features    = cpu_to_be64(s->compatible_features),
2952         .autoclear_features     = cpu_to_be64(s->autoclear_features),
2953         .refcount_order         = cpu_to_be32(s->refcount_order),
2954         .header_length          = cpu_to_be32(header_length),
2955         .compression_type       = s->compression_type,
2956     };
2957 
2958     /* For older versions, write a shorter header */
2959     switch (s->qcow_version) {
2960     case 2:
2961         ret = offsetof(QCowHeader, incompatible_features);
2962         break;
2963     case 3:
2964         ret = sizeof(*header);
2965         break;
2966     default:
2967         ret = -EINVAL;
2968         goto fail;
2969     }
2970 
2971     buf += ret;
2972     buflen -= ret;
2973     memset(buf, 0, buflen);
2974 
2975     /* Preserve any unknown field in the header */
2976     if (s->unknown_header_fields_size) {
2977         if (buflen < s->unknown_header_fields_size) {
2978             ret = -ENOSPC;
2979             goto fail;
2980         }
2981 
2982         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2983         buf += s->unknown_header_fields_size;
2984         buflen -= s->unknown_header_fields_size;
2985     }
2986 
2987     /* Backing file format header extension */
2988     if (s->image_backing_format) {
2989         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
2990                              s->image_backing_format,
2991                              strlen(s->image_backing_format),
2992                              buflen);
2993         if (ret < 0) {
2994             goto fail;
2995         }
2996 
2997         buf += ret;
2998         buflen -= ret;
2999     }
3000 
3001     /* External data file header extension */
3002     if (has_data_file(bs) && s->image_data_file) {
3003         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE,
3004                              s->image_data_file, strlen(s->image_data_file),
3005                              buflen);
3006         if (ret < 0) {
3007             goto fail;
3008         }
3009 
3010         buf += ret;
3011         buflen -= ret;
3012     }
3013 
3014     /* Full disk encryption header pointer extension */
3015     if (s->crypto_header.offset != 0) {
3016         s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset);
3017         s->crypto_header.length = cpu_to_be64(s->crypto_header.length);
3018         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
3019                              &s->crypto_header, sizeof(s->crypto_header),
3020                              buflen);
3021         s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
3022         s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
3023         if (ret < 0) {
3024             goto fail;
3025         }
3026         buf += ret;
3027         buflen -= ret;
3028     }
3029 
3030     /*
3031      * Feature table.  A mere 8 feature names occupies 392 bytes, and
3032      * when coupled with the v3 minimum header of 104 bytes plus the
3033      * 8-byte end-of-extension marker, that would leave only 8 bytes
3034      * for a backing file name in an image with 512-byte clusters.
3035      * Thus, we choose to omit this header for cluster sizes 4k and
3036      * smaller.
3037      */
3038     if (s->qcow_version >= 3 && s->cluster_size > 4096) {
3039         static const Qcow2Feature features[] = {
3040             {
3041                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3042                 .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
3043                 .name = "dirty bit",
3044             },
3045             {
3046                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3047                 .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
3048                 .name = "corrupt bit",
3049             },
3050             {
3051                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3052                 .bit  = QCOW2_INCOMPAT_DATA_FILE_BITNR,
3053                 .name = "external data file",
3054             },
3055             {
3056                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3057                 .bit  = QCOW2_INCOMPAT_COMPRESSION_BITNR,
3058                 .name = "compression type",
3059             },
3060             {
3061                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3062                 .bit  = QCOW2_INCOMPAT_EXTL2_BITNR,
3063                 .name = "extended L2 entries",
3064             },
3065             {
3066                 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
3067                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
3068                 .name = "lazy refcounts",
3069             },
3070             {
3071                 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
3072                 .bit  = QCOW2_AUTOCLEAR_BITMAPS_BITNR,
3073                 .name = "bitmaps",
3074             },
3075             {
3076                 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
3077                 .bit  = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR,
3078                 .name = "raw external data",
3079             },
3080         };
3081 
3082         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
3083                              features, sizeof(features), buflen);
3084         if (ret < 0) {
3085             goto fail;
3086         }
3087         buf += ret;
3088         buflen -= ret;
3089     }
3090 
3091     /* Bitmap extension */
3092     if (s->nb_bitmaps > 0) {
3093         Qcow2BitmapHeaderExt bitmaps_header = {
3094             .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
3095             .bitmap_directory_size =
3096                     cpu_to_be64(s->bitmap_directory_size),
3097             .bitmap_directory_offset =
3098                     cpu_to_be64(s->bitmap_directory_offset)
3099         };
3100         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
3101                              &bitmaps_header, sizeof(bitmaps_header),
3102                              buflen);
3103         if (ret < 0) {
3104             goto fail;
3105         }
3106         buf += ret;
3107         buflen -= ret;
3108     }
3109 
3110     /* Keep unknown header extensions */
3111     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
3112         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
3113         if (ret < 0) {
3114             goto fail;
3115         }
3116 
3117         buf += ret;
3118         buflen -= ret;
3119     }
3120 
3121     /* End of header extensions */
3122     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
3123     if (ret < 0) {
3124         goto fail;
3125     }
3126 
3127     buf += ret;
3128     buflen -= ret;
3129 
3130     /* Backing file name */
3131     if (s->image_backing_file) {
3132         size_t backing_file_len = strlen(s->image_backing_file);
3133 
3134         if (buflen < backing_file_len) {
3135             ret = -ENOSPC;
3136             goto fail;
3137         }
3138 
3139         /* Using strncpy is ok here, since buf is not NUL-terminated. */
3140         strncpy(buf, s->image_backing_file, buflen);
3141 
3142         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
3143         header->backing_file_size   = cpu_to_be32(backing_file_len);
3144     }
3145 
3146     /* Write the new header */
3147     ret = bdrv_pwrite(bs->file, 0, s->cluster_size, header, 0);
3148     if (ret < 0) {
3149         goto fail;
3150     }
3151 
3152     ret = 0;
3153 fail:
3154     qemu_vfree(header);
3155     return ret;
3156 }
3157 
3158 static int qcow2_change_backing_file(BlockDriverState *bs,
3159     const char *backing_file, const char *backing_fmt)
3160 {
3161     BDRVQcow2State *s = bs->opaque;
3162 
3163     /* Adding a backing file means that the external data file alone won't be
3164      * enough to make sense of the content */
3165     if (backing_file && data_file_is_raw(bs)) {
3166         return -EINVAL;
3167     }
3168 
3169     if (backing_file && strlen(backing_file) > 1023) {
3170         return -EINVAL;
3171     }
3172 
3173     pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3174             backing_file ?: "");
3175     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3176     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3177 
3178     g_free(s->image_backing_file);
3179     g_free(s->image_backing_format);
3180 
3181     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
3182     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
3183 
3184     return qcow2_update_header(bs);
3185 }
3186 
3187 static int coroutine_fn GRAPH_RDLOCK
3188 qcow2_set_up_encryption(BlockDriverState *bs,
3189                         QCryptoBlockCreateOptions *cryptoopts,
3190                         Error **errp)
3191 {
3192     BDRVQcow2State *s = bs->opaque;
3193     QCryptoBlock *crypto = NULL;
3194     int fmt, ret;
3195 
3196     switch (cryptoopts->format) {
3197     case Q_CRYPTO_BLOCK_FORMAT_LUKS:
3198         fmt = QCOW_CRYPT_LUKS;
3199         break;
3200     case Q_CRYPTO_BLOCK_FORMAT_QCOW:
3201         fmt = QCOW_CRYPT_AES;
3202         break;
3203     default:
3204         error_setg(errp, "Crypto format not supported in qcow2");
3205         return -EINVAL;
3206     }
3207 
3208     s->crypt_method_header = fmt;
3209 
3210     crypto = qcrypto_block_create(cryptoopts, "encrypt.",
3211                                   qcow2_crypto_hdr_init_func,
3212                                   qcow2_crypto_hdr_write_func,
3213                                   bs, errp);
3214     if (!crypto) {
3215         return -EINVAL;
3216     }
3217 
3218     ret = qcow2_update_header(bs);
3219     if (ret < 0) {
3220         error_setg_errno(errp, -ret, "Could not write encryption header");
3221         goto out;
3222     }
3223 
3224     ret = 0;
3225  out:
3226     qcrypto_block_free(crypto);
3227     return ret;
3228 }
3229 
3230 /**
3231  * Preallocates metadata structures for data clusters between @offset (in the
3232  * guest disk) and @new_length (which is thus generally the new guest disk
3233  * size).
3234  *
3235  * Returns: 0 on success, -errno on failure.
3236  */
3237 static int coroutine_fn GRAPH_RDLOCK
3238 preallocate_co(BlockDriverState *bs, uint64_t offset, uint64_t new_length,
3239                PreallocMode mode, Error **errp)
3240 {
3241     BDRVQcow2State *s = bs->opaque;
3242     uint64_t bytes;
3243     uint64_t host_offset = 0;
3244     int64_t file_length;
3245     unsigned int cur_bytes;
3246     int ret;
3247     QCowL2Meta *meta = NULL, *m;
3248 
3249     assert(offset <= new_length);
3250     bytes = new_length - offset;
3251 
3252     while (bytes) {
3253         cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size));
3254         ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes,
3255                                       &host_offset, &meta);
3256         if (ret < 0) {
3257             error_setg_errno(errp, -ret, "Allocating clusters failed");
3258             goto out;
3259         }
3260 
3261         for (m = meta; m != NULL; m = m->next) {
3262             m->prealloc = true;
3263         }
3264 
3265         ret = qcow2_handle_l2meta(bs, &meta, true);
3266         if (ret < 0) {
3267             error_setg_errno(errp, -ret, "Mapping clusters failed");
3268             goto out;
3269         }
3270 
3271         /* TODO Preallocate data if requested */
3272 
3273         bytes -= cur_bytes;
3274         offset += cur_bytes;
3275     }
3276 
3277     /*
3278      * It is expected that the image file is large enough to actually contain
3279      * all of the allocated clusters (otherwise we get failing reads after
3280      * EOF). Extend the image to the last allocated sector.
3281      */
3282     file_length = bdrv_co_getlength(s->data_file->bs);
3283     if (file_length < 0) {
3284         error_setg_errno(errp, -file_length, "Could not get file size");
3285         ret = file_length;
3286         goto out;
3287     }
3288 
3289     if (host_offset + cur_bytes > file_length) {
3290         if (mode == PREALLOC_MODE_METADATA) {
3291             mode = PREALLOC_MODE_OFF;
3292         }
3293         ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false,
3294                                mode, 0, errp);
3295         if (ret < 0) {
3296             goto out;
3297         }
3298     }
3299 
3300     ret = 0;
3301 
3302 out:
3303     qcow2_handle_l2meta(bs, &meta, false);
3304     return ret;
3305 }
3306 
3307 /* qcow2_refcount_metadata_size:
3308  * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3309  * @cluster_size: size of a cluster, in bytes
3310  * @refcount_order: refcount bits power-of-2 exponent
3311  * @generous_increase: allow for the refcount table to be 1.5x as large as it
3312  *                     needs to be
3313  *
3314  * Returns: Number of bytes required for refcount blocks and table metadata.
3315  */
3316 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
3317                                      int refcount_order, bool generous_increase,
3318                                      uint64_t *refblock_count)
3319 {
3320     /*
3321      * Every host cluster is reference-counted, including metadata (even
3322      * refcount metadata is recursively included).
3323      *
3324      * An accurate formula for the size of refcount metadata size is difficult
3325      * to derive.  An easier method of calculation is finding the fixed point
3326      * where no further refcount blocks or table clusters are required to
3327      * reference count every cluster.
3328      */
3329     int64_t blocks_per_table_cluster = cluster_size / REFTABLE_ENTRY_SIZE;
3330     int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
3331     int64_t table = 0;  /* number of refcount table clusters */
3332     int64_t blocks = 0; /* number of refcount block clusters */
3333     int64_t last;
3334     int64_t n = 0;
3335 
3336     do {
3337         last = n;
3338         blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
3339         table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
3340         n = clusters + blocks + table;
3341 
3342         if (n == last && generous_increase) {
3343             clusters += DIV_ROUND_UP(table, 2);
3344             n = 0; /* force another loop */
3345             generous_increase = false;
3346         }
3347     } while (n != last);
3348 
3349     if (refblock_count) {
3350         *refblock_count = blocks;
3351     }
3352 
3353     return (blocks + table) * cluster_size;
3354 }
3355 
3356 /**
3357  * qcow2_calc_prealloc_size:
3358  * @total_size: virtual disk size in bytes
3359  * @cluster_size: cluster size in bytes
3360  * @refcount_order: refcount bits power-of-2 exponent
3361  * @extended_l2: true if the image has extended L2 entries
3362  *
3363  * Returns: Total number of bytes required for the fully allocated image
3364  * (including metadata).
3365  */
3366 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
3367                                         size_t cluster_size,
3368                                         int refcount_order,
3369                                         bool extended_l2)
3370 {
3371     int64_t meta_size = 0;
3372     uint64_t nl1e, nl2e;
3373     int64_t aligned_total_size = ROUND_UP(total_size, cluster_size);
3374     size_t l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL;
3375 
3376     /* header: 1 cluster */
3377     meta_size += cluster_size;
3378 
3379     /* total size of L2 tables */
3380     nl2e = aligned_total_size / cluster_size;
3381     nl2e = ROUND_UP(nl2e, cluster_size / l2e_size);
3382     meta_size += nl2e * l2e_size;
3383 
3384     /* total size of L1 tables */
3385     nl1e = nl2e * l2e_size / cluster_size;
3386     nl1e = ROUND_UP(nl1e, cluster_size / L1E_SIZE);
3387     meta_size += nl1e * L1E_SIZE;
3388 
3389     /* total size of refcount table and blocks */
3390     meta_size += qcow2_refcount_metadata_size(
3391             (meta_size + aligned_total_size) / cluster_size,
3392             cluster_size, refcount_order, false, NULL);
3393 
3394     return meta_size + aligned_total_size;
3395 }
3396 
3397 static bool validate_cluster_size(size_t cluster_size, bool extended_l2,
3398                                   Error **errp)
3399 {
3400     int cluster_bits = ctz32(cluster_size);
3401     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
3402         (1 << cluster_bits) != cluster_size)
3403     {
3404         error_setg(errp, "Cluster size must be a power of two between %d and "
3405                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
3406         return false;
3407     }
3408 
3409     if (extended_l2) {
3410         unsigned min_cluster_size =
3411             (1 << MIN_CLUSTER_BITS) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER;
3412         if (cluster_size < min_cluster_size) {
3413             error_setg(errp, "Extended L2 entries are only supported with "
3414                        "cluster sizes of at least %u bytes", min_cluster_size);
3415             return false;
3416         }
3417     }
3418 
3419     return true;
3420 }
3421 
3422 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, bool extended_l2,
3423                                              Error **errp)
3424 {
3425     size_t cluster_size;
3426 
3427     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
3428                                          DEFAULT_CLUSTER_SIZE);
3429     if (!validate_cluster_size(cluster_size, extended_l2, errp)) {
3430         return 0;
3431     }
3432     return cluster_size;
3433 }
3434 
3435 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
3436 {
3437     char *buf;
3438     int ret;
3439 
3440     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
3441     if (!buf) {
3442         ret = 3; /* default */
3443     } else if (!strcmp(buf, "0.10")) {
3444         ret = 2;
3445     } else if (!strcmp(buf, "1.1")) {
3446         ret = 3;
3447     } else {
3448         error_setg(errp, "Invalid compatibility level: '%s'", buf);
3449         ret = -EINVAL;
3450     }
3451     g_free(buf);
3452     return ret;
3453 }
3454 
3455 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
3456                                                 Error **errp)
3457 {
3458     uint64_t refcount_bits;
3459 
3460     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
3461     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
3462         error_setg(errp, "Refcount width must be a power of two and may not "
3463                    "exceed 64 bits");
3464         return 0;
3465     }
3466 
3467     if (version < 3 && refcount_bits != 16) {
3468         error_setg(errp, "Different refcount widths than 16 bits require "
3469                    "compatibility level 1.1 or above (use compat=1.1 or "
3470                    "greater)");
3471         return 0;
3472     }
3473 
3474     return refcount_bits;
3475 }
3476 
3477 static int coroutine_fn GRAPH_UNLOCKED
3478 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
3479 {
3480     BlockdevCreateOptionsQcow2 *qcow2_opts;
3481     QDict *options;
3482 
3483     /*
3484      * Open the image file and write a minimal qcow2 header.
3485      *
3486      * We keep things simple and start with a zero-sized image. We also
3487      * do without refcount blocks or a L1 table for now. We'll fix the
3488      * inconsistency later.
3489      *
3490      * We do need a refcount table because growing the refcount table means
3491      * allocating two new refcount blocks - the second of which would be at
3492      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3493      * size for any qcow2 image.
3494      */
3495     BlockBackend *blk = NULL;
3496     BlockDriverState *bs = NULL;
3497     BlockDriverState *data_bs = NULL;
3498     QCowHeader *header;
3499     size_t cluster_size;
3500     int version;
3501     int refcount_order;
3502     uint64_t *refcount_table;
3503     int ret;
3504     uint8_t compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
3505 
3506     assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2);
3507     qcow2_opts = &create_options->u.qcow2;
3508 
3509     bs = bdrv_co_open_blockdev_ref(qcow2_opts->file, errp);
3510     if (bs == NULL) {
3511         return -EIO;
3512     }
3513 
3514     /* Validate options and set default values */
3515     if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) {
3516         error_setg(errp, "Image size must be a multiple of %u bytes",
3517                    (unsigned) BDRV_SECTOR_SIZE);
3518         ret = -EINVAL;
3519         goto out;
3520     }
3521 
3522     if (qcow2_opts->has_version) {
3523         switch (qcow2_opts->version) {
3524         case BLOCKDEV_QCOW2_VERSION_V2:
3525             version = 2;
3526             break;
3527         case BLOCKDEV_QCOW2_VERSION_V3:
3528             version = 3;
3529             break;
3530         default:
3531             g_assert_not_reached();
3532         }
3533     } else {
3534         version = 3;
3535     }
3536 
3537     if (qcow2_opts->has_cluster_size) {
3538         cluster_size = qcow2_opts->cluster_size;
3539     } else {
3540         cluster_size = DEFAULT_CLUSTER_SIZE;
3541     }
3542 
3543     if (!qcow2_opts->has_extended_l2) {
3544         qcow2_opts->extended_l2 = false;
3545     }
3546     if (qcow2_opts->extended_l2) {
3547         if (version < 3) {
3548             error_setg(errp, "Extended L2 entries are only supported with "
3549                        "compatibility level 1.1 and above (use version=v3 or "
3550                        "greater)");
3551             ret = -EINVAL;
3552             goto out;
3553         }
3554     }
3555 
3556     if (!validate_cluster_size(cluster_size, qcow2_opts->extended_l2, errp)) {
3557         ret = -EINVAL;
3558         goto out;
3559     }
3560 
3561     if (!qcow2_opts->has_preallocation) {
3562         qcow2_opts->preallocation = PREALLOC_MODE_OFF;
3563     }
3564     if (qcow2_opts->backing_file &&
3565         qcow2_opts->preallocation != PREALLOC_MODE_OFF &&
3566         !qcow2_opts->extended_l2)
3567     {
3568         error_setg(errp, "Backing file and preallocation can only be used at "
3569                    "the same time if extended_l2 is on");
3570         ret = -EINVAL;
3571         goto out;
3572     }
3573     if (qcow2_opts->has_backing_fmt && !qcow2_opts->backing_file) {
3574         error_setg(errp, "Backing format cannot be used without backing file");
3575         ret = -EINVAL;
3576         goto out;
3577     }
3578 
3579     if (!qcow2_opts->has_lazy_refcounts) {
3580         qcow2_opts->lazy_refcounts = false;
3581     }
3582     if (version < 3 && qcow2_opts->lazy_refcounts) {
3583         error_setg(errp, "Lazy refcounts only supported with compatibility "
3584                    "level 1.1 and above (use version=v3 or greater)");
3585         ret = -EINVAL;
3586         goto out;
3587     }
3588 
3589     if (!qcow2_opts->has_refcount_bits) {
3590         qcow2_opts->refcount_bits = 16;
3591     }
3592     if (qcow2_opts->refcount_bits > 64 ||
3593         !is_power_of_2(qcow2_opts->refcount_bits))
3594     {
3595         error_setg(errp, "Refcount width must be a power of two and may not "
3596                    "exceed 64 bits");
3597         ret = -EINVAL;
3598         goto out;
3599     }
3600     if (version < 3 && qcow2_opts->refcount_bits != 16) {
3601         error_setg(errp, "Different refcount widths than 16 bits require "
3602                    "compatibility level 1.1 or above (use version=v3 or "
3603                    "greater)");
3604         ret = -EINVAL;
3605         goto out;
3606     }
3607     refcount_order = ctz32(qcow2_opts->refcount_bits);
3608 
3609     if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) {
3610         error_setg(errp, "data-file-raw requires data-file");
3611         ret = -EINVAL;
3612         goto out;
3613     }
3614     if (qcow2_opts->data_file_raw && qcow2_opts->backing_file) {
3615         error_setg(errp, "Backing file and data-file-raw cannot be used at "
3616                    "the same time");
3617         ret = -EINVAL;
3618         goto out;
3619     }
3620     if (qcow2_opts->data_file_raw &&
3621         qcow2_opts->preallocation == PREALLOC_MODE_OFF)
3622     {
3623         /*
3624          * data-file-raw means that "the external data file can be
3625          * read as a consistent standalone raw image without looking
3626          * at the qcow2 metadata."  It does not say that the metadata
3627          * must be ignored, though (and the qcow2 driver in fact does
3628          * not ignore it), so the L1/L2 tables must be present and
3629          * give a 1:1 mapping, so you get the same result regardless
3630          * of whether you look at the metadata or whether you ignore
3631          * it.
3632          */
3633         qcow2_opts->preallocation = PREALLOC_MODE_METADATA;
3634 
3635         /*
3636          * Cannot use preallocation with backing files, but giving a
3637          * backing file when specifying data_file_raw is an error
3638          * anyway.
3639          */
3640         assert(!qcow2_opts->backing_file);
3641     }
3642 
3643     if (qcow2_opts->data_file) {
3644         if (version < 3) {
3645             error_setg(errp, "External data files are only supported with "
3646                        "compatibility level 1.1 and above (use version=v3 or "
3647                        "greater)");
3648             ret = -EINVAL;
3649             goto out;
3650         }
3651         data_bs = bdrv_co_open_blockdev_ref(qcow2_opts->data_file, errp);
3652         if (data_bs == NULL) {
3653             ret = -EIO;
3654             goto out;
3655         }
3656     }
3657 
3658     if (qcow2_opts->has_compression_type &&
3659         qcow2_opts->compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) {
3660 
3661         ret = -EINVAL;
3662 
3663         if (version < 3) {
3664             error_setg(errp, "Non-zlib compression type is only supported with "
3665                        "compatibility level 1.1 and above (use version=v3 or "
3666                        "greater)");
3667             goto out;
3668         }
3669 
3670         switch (qcow2_opts->compression_type) {
3671 #ifdef CONFIG_ZSTD
3672         case QCOW2_COMPRESSION_TYPE_ZSTD:
3673             break;
3674 #endif
3675         default:
3676             error_setg(errp, "Unknown compression type");
3677             goto out;
3678         }
3679 
3680         compression_type = qcow2_opts->compression_type;
3681     }
3682 
3683     /* Create BlockBackend to write to the image */
3684     blk = blk_co_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
3685                              errp);
3686     if (!blk) {
3687         ret = -EPERM;
3688         goto out;
3689     }
3690     blk_set_allow_write_beyond_eof(blk, true);
3691 
3692     /* Write the header */
3693     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
3694     header = g_malloc0(cluster_size);
3695     *header = (QCowHeader) {
3696         .magic                      = cpu_to_be32(QCOW_MAGIC),
3697         .version                    = cpu_to_be32(version),
3698         .cluster_bits               = cpu_to_be32(ctz32(cluster_size)),
3699         .size                       = cpu_to_be64(0),
3700         .l1_table_offset            = cpu_to_be64(0),
3701         .l1_size                    = cpu_to_be32(0),
3702         .refcount_table_offset      = cpu_to_be64(cluster_size),
3703         .refcount_table_clusters    = cpu_to_be32(1),
3704         .refcount_order             = cpu_to_be32(refcount_order),
3705         /* don't deal with endianness since compression_type is 1 byte long */
3706         .compression_type           = compression_type,
3707         .header_length              = cpu_to_be32(sizeof(*header)),
3708     };
3709 
3710     /* We'll update this to correct value later */
3711     header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
3712 
3713     if (qcow2_opts->lazy_refcounts) {
3714         header->compatible_features |=
3715             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
3716     }
3717     if (data_bs) {
3718         header->incompatible_features |=
3719             cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE);
3720     }
3721     if (qcow2_opts->data_file_raw) {
3722         header->autoclear_features |=
3723             cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW);
3724     }
3725     if (compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) {
3726         header->incompatible_features |=
3727             cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION);
3728     }
3729 
3730     if (qcow2_opts->extended_l2) {
3731         header->incompatible_features |=
3732             cpu_to_be64(QCOW2_INCOMPAT_EXTL2);
3733     }
3734 
3735     ret = blk_co_pwrite(blk, 0, cluster_size, header, 0);
3736     g_free(header);
3737     if (ret < 0) {
3738         error_setg_errno(errp, -ret, "Could not write qcow2 header");
3739         goto out;
3740     }
3741 
3742     /* Write a refcount table with one refcount block */
3743     refcount_table = g_malloc0(2 * cluster_size);
3744     refcount_table[0] = cpu_to_be64(2 * cluster_size);
3745     ret = blk_co_pwrite(blk, cluster_size, 2 * cluster_size, refcount_table, 0);
3746     g_free(refcount_table);
3747 
3748     if (ret < 0) {
3749         error_setg_errno(errp, -ret, "Could not write refcount table");
3750         goto out;
3751     }
3752 
3753     blk_co_unref(blk);
3754     blk = NULL;
3755 
3756     /*
3757      * And now open the image and make it consistent first (i.e. increase the
3758      * refcount of the cluster that is occupied by the header and the refcount
3759      * table)
3760      */
3761     options = qdict_new();
3762     qdict_put_str(options, "driver", "qcow2");
3763     qdict_put_str(options, "file", bs->node_name);
3764     if (data_bs) {
3765         qdict_put_str(options, "data-file", data_bs->node_name);
3766     }
3767     blk = blk_co_new_open(NULL, NULL, options,
3768                           BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
3769                           errp);
3770     if (blk == NULL) {
3771         ret = -EIO;
3772         goto out;
3773     }
3774 
3775     bdrv_graph_co_rdlock();
3776     ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
3777     if (ret < 0) {
3778         bdrv_graph_co_rdunlock();
3779         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
3780                          "header and refcount table");
3781         goto out;
3782 
3783     } else if (ret != 0) {
3784         error_report("Huh, first cluster in empty image is already in use?");
3785         abort();
3786     }
3787 
3788     /* Set the external data file if necessary */
3789     if (data_bs) {
3790         BDRVQcow2State *s = blk_bs(blk)->opaque;
3791         s->image_data_file = g_strdup(data_bs->filename);
3792     }
3793 
3794     /* Create a full header (including things like feature table) */
3795     ret = qcow2_update_header(blk_bs(blk));
3796     bdrv_graph_co_rdunlock();
3797 
3798     if (ret < 0) {
3799         error_setg_errno(errp, -ret, "Could not update qcow2 header");
3800         goto out;
3801     }
3802 
3803     /* Okay, now that we have a valid image, let's give it the right size */
3804     ret = blk_co_truncate(blk, qcow2_opts->size, false,
3805                           qcow2_opts->preallocation, 0, errp);
3806     if (ret < 0) {
3807         error_prepend(errp, "Could not resize image: ");
3808         goto out;
3809     }
3810 
3811     /* Want a backing file? There you go. */
3812     if (qcow2_opts->backing_file) {
3813         const char *backing_format = NULL;
3814 
3815         if (qcow2_opts->has_backing_fmt) {
3816             backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt);
3817         }
3818 
3819         ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file,
3820                                        backing_format, false);
3821         if (ret < 0) {
3822             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
3823                              "with format '%s'", qcow2_opts->backing_file,
3824                              backing_format);
3825             goto out;
3826         }
3827     }
3828 
3829     /* Want encryption? There you go. */
3830     if (qcow2_opts->encrypt) {
3831         bdrv_graph_co_rdlock();
3832         ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp);
3833         bdrv_graph_co_rdunlock();
3834 
3835         if (ret < 0) {
3836             goto out;
3837         }
3838     }
3839 
3840     blk_co_unref(blk);
3841     blk = NULL;
3842 
3843     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3844      * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3845      * have to setup decryption context. We're not doing any I/O on the top
3846      * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3847      * not have effect.
3848      */
3849     options = qdict_new();
3850     qdict_put_str(options, "driver", "qcow2");
3851     qdict_put_str(options, "file", bs->node_name);
3852     if (data_bs) {
3853         qdict_put_str(options, "data-file", data_bs->node_name);
3854     }
3855     blk = blk_co_new_open(NULL, NULL, options,
3856                           BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO,
3857                           errp);
3858     if (blk == NULL) {
3859         ret = -EIO;
3860         goto out;
3861     }
3862 
3863     ret = 0;
3864 out:
3865     blk_co_unref(blk);
3866     bdrv_co_unref(bs);
3867     bdrv_co_unref(data_bs);
3868     return ret;
3869 }
3870 
3871 static int coroutine_fn GRAPH_UNLOCKED
3872 qcow2_co_create_opts(BlockDriver *drv, const char *filename, QemuOpts *opts,
3873                      Error **errp)
3874 {
3875     BlockdevCreateOptions *create_options = NULL;
3876     QDict *qdict;
3877     Visitor *v;
3878     BlockDriverState *bs = NULL;
3879     BlockDriverState *data_bs = NULL;
3880     const char *val;
3881     int ret;
3882 
3883     /* Only the keyval visitor supports the dotted syntax needed for
3884      * encryption, so go through a QDict before getting a QAPI type. Ignore
3885      * options meant for the protocol layer so that the visitor doesn't
3886      * complain. */
3887     qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts,
3888                                         true);
3889 
3890     /* Handle encryption options */
3891     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
3892     if (val && !strcmp(val, "on")) {
3893         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
3894     } else if (val && !strcmp(val, "off")) {
3895         qdict_del(qdict, BLOCK_OPT_ENCRYPT);
3896     }
3897 
3898     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
3899     if (val && !strcmp(val, "aes")) {
3900         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
3901     }
3902 
3903     /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3904      * version=v2/v3 below. */
3905     val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL);
3906     if (val && !strcmp(val, "0.10")) {
3907         qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2");
3908     } else if (val && !strcmp(val, "1.1")) {
3909         qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3");
3910     }
3911 
3912     /* Change legacy command line options into QMP ones */
3913     static const QDictRenames opt_renames[] = {
3914         { BLOCK_OPT_BACKING_FILE,       "backing-file" },
3915         { BLOCK_OPT_BACKING_FMT,        "backing-fmt" },
3916         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
3917         { BLOCK_OPT_LAZY_REFCOUNTS,     "lazy-refcounts" },
3918         { BLOCK_OPT_EXTL2,              "extended-l2" },
3919         { BLOCK_OPT_REFCOUNT_BITS,      "refcount-bits" },
3920         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
3921         { BLOCK_OPT_COMPAT_LEVEL,       "version" },
3922         { BLOCK_OPT_DATA_FILE_RAW,      "data-file-raw" },
3923         { BLOCK_OPT_COMPRESSION_TYPE,   "compression-type" },
3924         { NULL, NULL },
3925     };
3926 
3927     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
3928         ret = -EINVAL;
3929         goto finish;
3930     }
3931 
3932     /* Create and open the file (protocol layer) */
3933     ret = bdrv_co_create_file(filename, opts, errp);
3934     if (ret < 0) {
3935         goto finish;
3936     }
3937 
3938     bs = bdrv_co_open(filename, NULL, NULL,
3939                       BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
3940     if (bs == NULL) {
3941         ret = -EIO;
3942         goto finish;
3943     }
3944 
3945     /* Create and open an external data file (protocol layer) */
3946     val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE);
3947     if (val) {
3948         ret = bdrv_co_create_file(val, opts, errp);
3949         if (ret < 0) {
3950             goto finish;
3951         }
3952 
3953         data_bs = bdrv_co_open(val, NULL, NULL,
3954                                BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
3955                                errp);
3956         if (data_bs == NULL) {
3957             ret = -EIO;
3958             goto finish;
3959         }
3960 
3961         qdict_del(qdict, BLOCK_OPT_DATA_FILE);
3962         qdict_put_str(qdict, "data-file", data_bs->node_name);
3963     }
3964 
3965     /* Set 'driver' and 'node' options */
3966     qdict_put_str(qdict, "driver", "qcow2");
3967     qdict_put_str(qdict, "file", bs->node_name);
3968 
3969     /* Now get the QAPI type BlockdevCreateOptions */
3970     v = qobject_input_visitor_new_flat_confused(qdict, errp);
3971     if (!v) {
3972         ret = -EINVAL;
3973         goto finish;
3974     }
3975 
3976     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
3977     visit_free(v);
3978     if (!create_options) {
3979         ret = -EINVAL;
3980         goto finish;
3981     }
3982 
3983     /* Silently round up size */
3984     create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size,
3985                                             BDRV_SECTOR_SIZE);
3986 
3987     /* Create the qcow2 image (format layer) */
3988     ret = qcow2_co_create(create_options, errp);
3989 finish:
3990     if (ret < 0) {
3991         bdrv_graph_co_rdlock();
3992         bdrv_co_delete_file_noerr(bs);
3993         bdrv_co_delete_file_noerr(data_bs);
3994         bdrv_graph_co_rdunlock();
3995     } else {
3996         ret = 0;
3997     }
3998 
3999     qobject_unref(qdict);
4000     bdrv_co_unref(bs);
4001     bdrv_co_unref(data_bs);
4002     qapi_free_BlockdevCreateOptions(create_options);
4003     return ret;
4004 }
4005 
4006 
4007 static bool coroutine_fn GRAPH_RDLOCK
4008 is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes)
4009 {
4010     int64_t nr;
4011     int res;
4012 
4013     /* Clamp to image length, before checking status of underlying sectors */
4014     if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
4015         bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset;
4016     }
4017 
4018     if (!bytes) {
4019         return true;
4020     }
4021 
4022     /*
4023      * bdrv_block_status_above doesn't merge different types of zeros, for
4024      * example, zeros which come from the region which is unallocated in
4025      * the whole backing chain, and zeros which come because of a short
4026      * backing file. So, we need a loop.
4027      */
4028     do {
4029         res = bdrv_co_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL);
4030         offset += nr;
4031         bytes -= nr;
4032     } while (res >= 0 && (res & BDRV_BLOCK_ZERO) && nr && bytes);
4033 
4034     return res >= 0 && (res & BDRV_BLOCK_ZERO) && bytes == 0;
4035 }
4036 
4037 static int coroutine_fn GRAPH_RDLOCK
4038 qcow2_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
4039                        BdrvRequestFlags flags)
4040 {
4041     int ret;
4042     BDRVQcow2State *s = bs->opaque;
4043 
4044     uint32_t head = offset_into_subcluster(s, offset);
4045     uint32_t tail = ROUND_UP(offset + bytes, s->subcluster_size) -
4046         (offset + bytes);
4047 
4048     trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes);
4049     if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) {
4050         tail = 0;
4051     }
4052 
4053     if (head || tail) {
4054         uint64_t off;
4055         unsigned int nr;
4056         QCow2SubclusterType type;
4057 
4058         assert(head + bytes + tail <= s->subcluster_size);
4059 
4060         /* check whether remainder of cluster already reads as zero */
4061         if (!(is_zero(bs, offset - head, head) &&
4062               is_zero(bs, offset + bytes, tail))) {
4063             return -ENOTSUP;
4064         }
4065 
4066         qemu_co_mutex_lock(&s->lock);
4067         /* We can have new write after previous check */
4068         offset -= head;
4069         bytes = s->subcluster_size;
4070         nr = s->subcluster_size;
4071         ret = qcow2_get_host_offset(bs, offset, &nr, &off, &type);
4072         if (ret < 0 ||
4073             (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN &&
4074              type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC &&
4075              type != QCOW2_SUBCLUSTER_ZERO_PLAIN &&
4076              type != QCOW2_SUBCLUSTER_ZERO_ALLOC)) {
4077             qemu_co_mutex_unlock(&s->lock);
4078             return ret < 0 ? ret : -ENOTSUP;
4079         }
4080     } else {
4081         qemu_co_mutex_lock(&s->lock);
4082     }
4083 
4084     trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes);
4085 
4086     /* Whatever is left can use real zero subclusters */
4087     ret = qcow2_subcluster_zeroize(bs, offset, bytes, flags);
4088     qemu_co_mutex_unlock(&s->lock);
4089 
4090     return ret;
4091 }
4092 
4093 static int coroutine_fn GRAPH_RDLOCK
4094 qcow2_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
4095 {
4096     int ret;
4097     BDRVQcow2State *s = bs->opaque;
4098 
4099     /* If the image does not support QCOW_OFLAG_ZERO then discarding
4100      * clusters could expose stale data from the backing file. */
4101     if (s->qcow_version < 3 && bs->backing) {
4102         return -ENOTSUP;
4103     }
4104 
4105     if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
4106         assert(bytes < s->cluster_size);
4107         /* Ignore partial clusters, except for the special case of the
4108          * complete partial cluster at the end of an unaligned file */
4109         if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
4110             offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) {
4111             return -ENOTSUP;
4112         }
4113     }
4114 
4115     qemu_co_mutex_lock(&s->lock);
4116     ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST,
4117                                 false);
4118     qemu_co_mutex_unlock(&s->lock);
4119     return ret;
4120 }
4121 
4122 static int coroutine_fn GRAPH_RDLOCK
4123 qcow2_co_copy_range_from(BlockDriverState *bs,
4124                          BdrvChild *src, int64_t src_offset,
4125                          BdrvChild *dst, int64_t dst_offset,
4126                          int64_t bytes, BdrvRequestFlags read_flags,
4127                          BdrvRequestFlags write_flags)
4128 {
4129     BDRVQcow2State *s = bs->opaque;
4130     int ret;
4131     unsigned int cur_bytes; /* number of bytes in current iteration */
4132     BdrvChild *child = NULL;
4133     BdrvRequestFlags cur_write_flags;
4134 
4135     assert(!bs->encrypted);
4136     qemu_co_mutex_lock(&s->lock);
4137 
4138     while (bytes != 0) {
4139         uint64_t copy_offset = 0;
4140         QCow2SubclusterType type;
4141         /* prepare next request */
4142         cur_bytes = MIN(bytes, INT_MAX);
4143         cur_write_flags = write_flags;
4144 
4145         ret = qcow2_get_host_offset(bs, src_offset, &cur_bytes,
4146                                     &copy_offset, &type);
4147         if (ret < 0) {
4148             goto out;
4149         }
4150 
4151         switch (type) {
4152         case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
4153         case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
4154             if (bs->backing && bs->backing->bs) {
4155                 int64_t backing_length = bdrv_co_getlength(bs->backing->bs);
4156                 if (src_offset >= backing_length) {
4157                     cur_write_flags |= BDRV_REQ_ZERO_WRITE;
4158                 } else {
4159                     child = bs->backing;
4160                     cur_bytes = MIN(cur_bytes, backing_length - src_offset);
4161                     copy_offset = src_offset;
4162                 }
4163             } else {
4164                 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
4165             }
4166             break;
4167 
4168         case QCOW2_SUBCLUSTER_ZERO_PLAIN:
4169         case QCOW2_SUBCLUSTER_ZERO_ALLOC:
4170             cur_write_flags |= BDRV_REQ_ZERO_WRITE;
4171             break;
4172 
4173         case QCOW2_SUBCLUSTER_COMPRESSED:
4174             ret = -ENOTSUP;
4175             goto out;
4176 
4177         case QCOW2_SUBCLUSTER_NORMAL:
4178             child = s->data_file;
4179             break;
4180 
4181         default:
4182             abort();
4183         }
4184         qemu_co_mutex_unlock(&s->lock);
4185         ret = bdrv_co_copy_range_from(child,
4186                                       copy_offset,
4187                                       dst, dst_offset,
4188                                       cur_bytes, read_flags, cur_write_flags);
4189         qemu_co_mutex_lock(&s->lock);
4190         if (ret < 0) {
4191             goto out;
4192         }
4193 
4194         bytes -= cur_bytes;
4195         src_offset += cur_bytes;
4196         dst_offset += cur_bytes;
4197     }
4198     ret = 0;
4199 
4200 out:
4201     qemu_co_mutex_unlock(&s->lock);
4202     return ret;
4203 }
4204 
4205 static int coroutine_fn GRAPH_RDLOCK
4206 qcow2_co_copy_range_to(BlockDriverState *bs,
4207                        BdrvChild *src, int64_t src_offset,
4208                        BdrvChild *dst, int64_t dst_offset,
4209                        int64_t bytes, BdrvRequestFlags read_flags,
4210                        BdrvRequestFlags write_flags)
4211 {
4212     BDRVQcow2State *s = bs->opaque;
4213     int ret;
4214     unsigned int cur_bytes; /* number of sectors in current iteration */
4215     uint64_t host_offset;
4216     QCowL2Meta *l2meta = NULL;
4217 
4218     assert(!bs->encrypted);
4219 
4220     qemu_co_mutex_lock(&s->lock);
4221 
4222     while (bytes != 0) {
4223 
4224         l2meta = NULL;
4225 
4226         cur_bytes = MIN(bytes, INT_MAX);
4227 
4228         /* TODO:
4229          * If src->bs == dst->bs, we could simply copy by incrementing
4230          * the refcnt, without copying user data.
4231          * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
4232         ret = qcow2_alloc_host_offset(bs, dst_offset, &cur_bytes,
4233                                       &host_offset, &l2meta);
4234         if (ret < 0) {
4235             goto fail;
4236         }
4237 
4238         ret = qcow2_pre_write_overlap_check(bs, 0, host_offset, cur_bytes,
4239                                             true);
4240         if (ret < 0) {
4241             goto fail;
4242         }
4243 
4244         qemu_co_mutex_unlock(&s->lock);
4245         ret = bdrv_co_copy_range_to(src, src_offset, s->data_file, host_offset,
4246                                     cur_bytes, read_flags, write_flags);
4247         qemu_co_mutex_lock(&s->lock);
4248         if (ret < 0) {
4249             goto fail;
4250         }
4251 
4252         ret = qcow2_handle_l2meta(bs, &l2meta, true);
4253         if (ret) {
4254             goto fail;
4255         }
4256 
4257         bytes -= cur_bytes;
4258         src_offset += cur_bytes;
4259         dst_offset += cur_bytes;
4260     }
4261     ret = 0;
4262 
4263 fail:
4264     qcow2_handle_l2meta(bs, &l2meta, false);
4265 
4266     qemu_co_mutex_unlock(&s->lock);
4267 
4268     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
4269 
4270     return ret;
4271 }
4272 
4273 static int coroutine_fn GRAPH_RDLOCK
4274 qcow2_co_truncate(BlockDriverState *bs, int64_t offset, bool exact,
4275                   PreallocMode prealloc, BdrvRequestFlags flags, Error **errp)
4276 {
4277     BDRVQcow2State *s = bs->opaque;
4278     uint64_t old_length;
4279     int64_t new_l1_size;
4280     int ret;
4281     QDict *options;
4282 
4283     if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA &&
4284         prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL)
4285     {
4286         error_setg(errp, "Unsupported preallocation mode '%s'",
4287                    PreallocMode_str(prealloc));
4288         return -ENOTSUP;
4289     }
4290 
4291     if (!QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)) {
4292         error_setg(errp, "The new size must be a multiple of %u",
4293                    (unsigned) BDRV_SECTOR_SIZE);
4294         return -EINVAL;
4295     }
4296 
4297     qemu_co_mutex_lock(&s->lock);
4298 
4299     /*
4300      * Even though we store snapshot size for all images, it was not
4301      * required until v3, so it is not safe to proceed for v2.
4302      */
4303     if (s->nb_snapshots && s->qcow_version < 3) {
4304         error_setg(errp, "Can't resize a v2 image which has snapshots");
4305         ret = -ENOTSUP;
4306         goto fail;
4307     }
4308 
4309     /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */
4310     if (qcow2_truncate_bitmaps_check(bs, errp)) {
4311         ret = -ENOTSUP;
4312         goto fail;
4313     }
4314 
4315     old_length = bs->total_sectors * BDRV_SECTOR_SIZE;
4316     new_l1_size = size_to_l1(s, offset);
4317 
4318     if (offset < old_length) {
4319         int64_t last_cluster, old_file_size;
4320         if (prealloc != PREALLOC_MODE_OFF) {
4321             error_setg(errp,
4322                        "Preallocation can't be used for shrinking an image");
4323             ret = -EINVAL;
4324             goto fail;
4325         }
4326 
4327         ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size),
4328                                     old_length - ROUND_UP(offset,
4329                                                           s->cluster_size),
4330                                     QCOW2_DISCARD_ALWAYS, true);
4331         if (ret < 0) {
4332             error_setg_errno(errp, -ret, "Failed to discard cropped clusters");
4333             goto fail;
4334         }
4335 
4336         ret = qcow2_shrink_l1_table(bs, new_l1_size);
4337         if (ret < 0) {
4338             error_setg_errno(errp, -ret,
4339                              "Failed to reduce the number of L2 tables");
4340             goto fail;
4341         }
4342 
4343         ret = qcow2_shrink_reftable(bs);
4344         if (ret < 0) {
4345             error_setg_errno(errp, -ret,
4346                              "Failed to discard unused refblocks");
4347             goto fail;
4348         }
4349 
4350         old_file_size = bdrv_co_getlength(bs->file->bs);
4351         if (old_file_size < 0) {
4352             error_setg_errno(errp, -old_file_size,
4353                              "Failed to inquire current file length");
4354             ret = old_file_size;
4355             goto fail;
4356         }
4357         last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4358         if (last_cluster < 0) {
4359             error_setg_errno(errp, -last_cluster,
4360                              "Failed to find the last cluster");
4361             ret = last_cluster;
4362             goto fail;
4363         }
4364         if ((last_cluster + 1) * s->cluster_size < old_file_size) {
4365             Error *local_err = NULL;
4366 
4367             /*
4368              * Do not pass @exact here: It will not help the user if
4369              * we get an error here just because they wanted to shrink
4370              * their qcow2 image (on a block device) with qemu-img.
4371              * (And on the qcow2 layer, the @exact requirement is
4372              * always fulfilled, so there is no need to pass it on.)
4373              */
4374             bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size,
4375                              false, PREALLOC_MODE_OFF, 0, &local_err);
4376             if (local_err) {
4377                 warn_reportf_err(local_err,
4378                                  "Failed to truncate the tail of the image: ");
4379             }
4380         }
4381     } else {
4382         ret = qcow2_grow_l1_table(bs, new_l1_size, true);
4383         if (ret < 0) {
4384             error_setg_errno(errp, -ret, "Failed to grow the L1 table");
4385             goto fail;
4386         }
4387 
4388         if (data_file_is_raw(bs) && prealloc == PREALLOC_MODE_OFF) {
4389             /*
4390              * When creating a qcow2 image with data-file-raw, we enforce
4391              * at least prealloc=metadata, so that the L1/L2 tables are
4392              * fully allocated and reading from the data file will return
4393              * the same data as reading from the qcow2 image.  When the
4394              * image is grown, we must consequently preallocate the
4395              * metadata structures to cover the added area.
4396              */
4397             prealloc = PREALLOC_MODE_METADATA;
4398         }
4399     }
4400 
4401     switch (prealloc) {
4402     case PREALLOC_MODE_OFF:
4403         if (has_data_file(bs)) {
4404             /*
4405              * If the caller wants an exact resize, the external data
4406              * file should be resized to the exact target size, too,
4407              * so we pass @exact here.
4408              */
4409             ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, 0,
4410                                    errp);
4411             if (ret < 0) {
4412                 goto fail;
4413             }
4414         }
4415         break;
4416 
4417     case PREALLOC_MODE_METADATA:
4418         ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4419         if (ret < 0) {
4420             goto fail;
4421         }
4422         break;
4423 
4424     case PREALLOC_MODE_FALLOC:
4425     case PREALLOC_MODE_FULL:
4426     {
4427         int64_t allocation_start, host_offset, guest_offset;
4428         int64_t clusters_allocated;
4429         int64_t old_file_size, last_cluster, new_file_size;
4430         uint64_t nb_new_data_clusters, nb_new_l2_tables;
4431         bool subclusters_need_allocation = false;
4432 
4433         /* With a data file, preallocation means just allocating the metadata
4434          * and forwarding the truncate request to the data file */
4435         if (has_data_file(bs)) {
4436             ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4437             if (ret < 0) {
4438                 goto fail;
4439             }
4440             break;
4441         }
4442 
4443         old_file_size = bdrv_co_getlength(bs->file->bs);
4444         if (old_file_size < 0) {
4445             error_setg_errno(errp, -old_file_size,
4446                              "Failed to inquire current file length");
4447             ret = old_file_size;
4448             goto fail;
4449         }
4450 
4451         last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4452         if (last_cluster >= 0) {
4453             old_file_size = (last_cluster + 1) * s->cluster_size;
4454         } else {
4455             old_file_size = ROUND_UP(old_file_size, s->cluster_size);
4456         }
4457 
4458         nb_new_data_clusters = (ROUND_UP(offset, s->cluster_size) -
4459             start_of_cluster(s, old_length)) >> s->cluster_bits;
4460 
4461         /* This is an overestimation; we will not actually allocate space for
4462          * these in the file but just make sure the new refcount structures are
4463          * able to cover them so we will not have to allocate new refblocks
4464          * while entering the data blocks in the potentially new L2 tables.
4465          * (We do not actually care where the L2 tables are placed. Maybe they
4466          *  are already allocated or they can be placed somewhere before
4467          *  @old_file_size. It does not matter because they will be fully
4468          *  allocated automatically, so they do not need to be covered by the
4469          *  preallocation. All that matters is that we will not have to allocate
4470          *  new refcount structures for them.) */
4471         nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters,
4472                                         s->cluster_size / l2_entry_size(s));
4473         /* The cluster range may not be aligned to L2 boundaries, so add one L2
4474          * table for a potential head/tail */
4475         nb_new_l2_tables++;
4476 
4477         allocation_start = qcow2_refcount_area(bs, old_file_size,
4478                                                nb_new_data_clusters +
4479                                                nb_new_l2_tables,
4480                                                true, 0, 0);
4481         if (allocation_start < 0) {
4482             error_setg_errno(errp, -allocation_start,
4483                              "Failed to resize refcount structures");
4484             ret = allocation_start;
4485             goto fail;
4486         }
4487 
4488         clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start,
4489                                                      nb_new_data_clusters);
4490         if (clusters_allocated < 0) {
4491             error_setg_errno(errp, -clusters_allocated,
4492                              "Failed to allocate data clusters");
4493             ret = clusters_allocated;
4494             goto fail;
4495         }
4496 
4497         assert(clusters_allocated == nb_new_data_clusters);
4498 
4499         /* Allocate the data area */
4500         new_file_size = allocation_start +
4501                         nb_new_data_clusters * s->cluster_size;
4502         /*
4503          * Image file grows, so @exact does not matter.
4504          *
4505          * If we need to zero out the new area, try first whether the protocol
4506          * driver can already take care of this.
4507          */
4508         if (flags & BDRV_REQ_ZERO_WRITE) {
4509             ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc,
4510                                    BDRV_REQ_ZERO_WRITE, NULL);
4511             if (ret >= 0) {
4512                 flags &= ~BDRV_REQ_ZERO_WRITE;
4513                 /* Ensure that we read zeroes and not backing file data */
4514                 subclusters_need_allocation = true;
4515             }
4516         } else {
4517             ret = -1;
4518         }
4519         if (ret < 0) {
4520             ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0,
4521                                    errp);
4522         }
4523         if (ret < 0) {
4524             error_prepend(errp, "Failed to resize underlying file: ");
4525             qcow2_free_clusters(bs, allocation_start,
4526                                 nb_new_data_clusters * s->cluster_size,
4527                                 QCOW2_DISCARD_OTHER);
4528             goto fail;
4529         }
4530 
4531         /* Create the necessary L2 entries */
4532         host_offset = allocation_start;
4533         guest_offset = old_length;
4534         while (nb_new_data_clusters) {
4535             int64_t nb_clusters = MIN(
4536                 nb_new_data_clusters,
4537                 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset));
4538             unsigned cow_start_length = offset_into_cluster(s, guest_offset);
4539             QCowL2Meta allocation;
4540             guest_offset = start_of_cluster(s, guest_offset);
4541             allocation = (QCowL2Meta) {
4542                 .offset       = guest_offset,
4543                 .alloc_offset = host_offset,
4544                 .nb_clusters  = nb_clusters,
4545                 .cow_start    = {
4546                     .offset       = 0,
4547                     .nb_bytes     = cow_start_length,
4548                 },
4549                 .cow_end      = {
4550                     .offset       = nb_clusters << s->cluster_bits,
4551                     .nb_bytes     = 0,
4552                 },
4553                 .prealloc     = !subclusters_need_allocation,
4554             };
4555             qemu_co_queue_init(&allocation.dependent_requests);
4556 
4557             ret = qcow2_alloc_cluster_link_l2(bs, &allocation);
4558             if (ret < 0) {
4559                 error_setg_errno(errp, -ret, "Failed to update L2 tables");
4560                 qcow2_free_clusters(bs, host_offset,
4561                                     nb_new_data_clusters * s->cluster_size,
4562                                     QCOW2_DISCARD_OTHER);
4563                 goto fail;
4564             }
4565 
4566             guest_offset += nb_clusters * s->cluster_size;
4567             host_offset += nb_clusters * s->cluster_size;
4568             nb_new_data_clusters -= nb_clusters;
4569         }
4570         break;
4571     }
4572 
4573     default:
4574         g_assert_not_reached();
4575     }
4576 
4577     if ((flags & BDRV_REQ_ZERO_WRITE) && offset > old_length) {
4578         uint64_t zero_start = QEMU_ALIGN_UP(old_length, s->subcluster_size);
4579 
4580         /*
4581          * Use zero clusters as much as we can. qcow2_subcluster_zeroize()
4582          * requires a subcluster-aligned start. The end may be unaligned if
4583          * it is at the end of the image (which it is here).
4584          */
4585         if (offset > zero_start) {
4586             ret = qcow2_subcluster_zeroize(bs, zero_start, offset - zero_start,
4587                                            0);
4588             if (ret < 0) {
4589                 error_setg_errno(errp, -ret, "Failed to zero out new clusters");
4590                 goto fail;
4591             }
4592         }
4593 
4594         /* Write explicit zeros for the unaligned head */
4595         if (zero_start > old_length) {
4596             uint64_t len = MIN(zero_start, offset) - old_length;
4597             uint8_t *buf = qemu_blockalign0(bs, len);
4598             QEMUIOVector qiov;
4599             qemu_iovec_init_buf(&qiov, buf, len);
4600 
4601             qemu_co_mutex_unlock(&s->lock);
4602             ret = qcow2_co_pwritev_part(bs, old_length, len, &qiov, 0, 0);
4603             qemu_co_mutex_lock(&s->lock);
4604 
4605             qemu_vfree(buf);
4606             if (ret < 0) {
4607                 error_setg_errno(errp, -ret, "Failed to zero out the new area");
4608                 goto fail;
4609             }
4610         }
4611     }
4612 
4613     if (prealloc != PREALLOC_MODE_OFF) {
4614         /* Flush metadata before actually changing the image size */
4615         ret = qcow2_write_caches(bs);
4616         if (ret < 0) {
4617             error_setg_errno(errp, -ret,
4618                              "Failed to flush the preallocated area to disk");
4619             goto fail;
4620         }
4621     }
4622 
4623     bs->total_sectors = offset / BDRV_SECTOR_SIZE;
4624 
4625     /* write updated header.size */
4626     offset = cpu_to_be64(offset);
4627     ret = bdrv_co_pwrite_sync(bs->file, offsetof(QCowHeader, size),
4628                               sizeof(offset), &offset, 0);
4629     if (ret < 0) {
4630         error_setg_errno(errp, -ret, "Failed to update the image size");
4631         goto fail;
4632     }
4633 
4634     s->l1_vm_state_index = new_l1_size;
4635 
4636     /* Update cache sizes */
4637     options = qdict_clone_shallow(bs->options);
4638     ret = qcow2_update_options(bs, options, s->flags, errp);
4639     qobject_unref(options);
4640     if (ret < 0) {
4641         goto fail;
4642     }
4643     ret = 0;
4644 fail:
4645     qemu_co_mutex_unlock(&s->lock);
4646     return ret;
4647 }
4648 
4649 static int coroutine_fn GRAPH_RDLOCK
4650 qcow2_co_pwritev_compressed_task(BlockDriverState *bs,
4651                                  uint64_t offset, uint64_t bytes,
4652                                  QEMUIOVector *qiov, size_t qiov_offset)
4653 {
4654     BDRVQcow2State *s = bs->opaque;
4655     int ret;
4656     ssize_t out_len;
4657     uint8_t *buf, *out_buf;
4658     uint64_t cluster_offset;
4659 
4660     assert(bytes == s->cluster_size || (bytes < s->cluster_size &&
4661            (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS)));
4662 
4663     buf = qemu_blockalign(bs, s->cluster_size);
4664     if (bytes < s->cluster_size) {
4665         /* Zero-pad last write if image size is not cluster aligned */
4666         memset(buf + bytes, 0, s->cluster_size - bytes);
4667     }
4668     qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes);
4669 
4670     out_buf = g_malloc(s->cluster_size);
4671 
4672     out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1,
4673                                 buf, s->cluster_size);
4674     if (out_len == -ENOMEM) {
4675         /* could not compress: write normal cluster */
4676         ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0);
4677         if (ret < 0) {
4678             goto fail;
4679         }
4680         goto success;
4681     } else if (out_len < 0) {
4682         ret = -EINVAL;
4683         goto fail;
4684     }
4685 
4686     qemu_co_mutex_lock(&s->lock);
4687     ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len,
4688                                                 &cluster_offset);
4689     if (ret < 0) {
4690         qemu_co_mutex_unlock(&s->lock);
4691         goto fail;
4692     }
4693 
4694     ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true);
4695     qemu_co_mutex_unlock(&s->lock);
4696     if (ret < 0) {
4697         goto fail;
4698     }
4699 
4700     BLKDBG_CO_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED);
4701     ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0);
4702     if (ret < 0) {
4703         goto fail;
4704     }
4705 success:
4706     ret = 0;
4707 fail:
4708     qemu_vfree(buf);
4709     g_free(out_buf);
4710     return ret;
4711 }
4712 
4713 /*
4714  * This function can count as GRAPH_RDLOCK because
4715  * qcow2_co_pwritev_compressed_part() holds the graph lock and keeps it until
4716  * this coroutine has terminated.
4717  */
4718 static int coroutine_fn GRAPH_RDLOCK
4719 qcow2_co_pwritev_compressed_task_entry(AioTask *task)
4720 {
4721     Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
4722 
4723     assert(!t->subcluster_type && !t->l2meta);
4724 
4725     return qcow2_co_pwritev_compressed_task(t->bs, t->offset, t->bytes, t->qiov,
4726                                             t->qiov_offset);
4727 }
4728 
4729 /*
4730  * XXX: put compressed sectors first, then all the cluster aligned
4731  * tables to avoid losing bytes in alignment
4732  */
4733 static int coroutine_fn GRAPH_RDLOCK
4734 qcow2_co_pwritev_compressed_part(BlockDriverState *bs,
4735                                  int64_t offset, int64_t bytes,
4736                                  QEMUIOVector *qiov, size_t qiov_offset)
4737 {
4738     BDRVQcow2State *s = bs->opaque;
4739     AioTaskPool *aio = NULL;
4740     int ret = 0;
4741 
4742     if (has_data_file(bs)) {
4743         return -ENOTSUP;
4744     }
4745 
4746     if (bytes == 0) {
4747         /*
4748          * align end of file to a sector boundary to ease reading with
4749          * sector based I/Os
4750          */
4751         int64_t len = bdrv_co_getlength(bs->file->bs);
4752         if (len < 0) {
4753             return len;
4754         }
4755         return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, 0,
4756                                 NULL);
4757     }
4758 
4759     if (offset_into_cluster(s, offset)) {
4760         return -EINVAL;
4761     }
4762 
4763     if (offset_into_cluster(s, bytes) &&
4764         (offset + bytes) != (bs->total_sectors << BDRV_SECTOR_BITS)) {
4765         return -EINVAL;
4766     }
4767 
4768     while (bytes && aio_task_pool_status(aio) == 0) {
4769         uint64_t chunk_size = MIN(bytes, s->cluster_size);
4770 
4771         if (!aio && chunk_size != bytes) {
4772             aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
4773         }
4774 
4775         ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_compressed_task_entry,
4776                              0, 0, offset, chunk_size, qiov, qiov_offset, NULL);
4777         if (ret < 0) {
4778             break;
4779         }
4780         qiov_offset += chunk_size;
4781         offset += chunk_size;
4782         bytes -= chunk_size;
4783     }
4784 
4785     if (aio) {
4786         aio_task_pool_wait_all(aio);
4787         if (ret == 0) {
4788             ret = aio_task_pool_status(aio);
4789         }
4790         g_free(aio);
4791     }
4792 
4793     return ret;
4794 }
4795 
4796 static int coroutine_fn GRAPH_RDLOCK
4797 qcow2_co_preadv_compressed(BlockDriverState *bs,
4798                            uint64_t l2_entry,
4799                            uint64_t offset,
4800                            uint64_t bytes,
4801                            QEMUIOVector *qiov,
4802                            size_t qiov_offset)
4803 {
4804     BDRVQcow2State *s = bs->opaque;
4805     int ret = 0, csize;
4806     uint64_t coffset;
4807     uint8_t *buf, *out_buf;
4808     int offset_in_cluster = offset_into_cluster(s, offset);
4809 
4810     qcow2_parse_compressed_l2_entry(bs, l2_entry, &coffset, &csize);
4811 
4812     buf = g_try_malloc(csize);
4813     if (!buf) {
4814         return -ENOMEM;
4815     }
4816 
4817     out_buf = qemu_blockalign(bs, s->cluster_size);
4818 
4819     BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
4820     ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0);
4821     if (ret < 0) {
4822         goto fail;
4823     }
4824 
4825     if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) {
4826         ret = -EIO;
4827         goto fail;
4828     }
4829 
4830     qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes);
4831 
4832 fail:
4833     qemu_vfree(out_buf);
4834     g_free(buf);
4835 
4836     return ret;
4837 }
4838 
4839 static int GRAPH_RDLOCK make_completely_empty(BlockDriverState *bs)
4840 {
4841     BDRVQcow2State *s = bs->opaque;
4842     Error *local_err = NULL;
4843     int ret, l1_clusters;
4844     int64_t offset;
4845     uint64_t *new_reftable = NULL;
4846     uint64_t rt_entry, l1_size2;
4847     struct {
4848         uint64_t l1_offset;
4849         uint64_t reftable_offset;
4850         uint32_t reftable_clusters;
4851     } QEMU_PACKED l1_ofs_rt_ofs_cls;
4852 
4853     ret = qcow2_cache_empty(bs, s->l2_table_cache);
4854     if (ret < 0) {
4855         goto fail;
4856     }
4857 
4858     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
4859     if (ret < 0) {
4860         goto fail;
4861     }
4862 
4863     /* Refcounts will be broken utterly */
4864     ret = qcow2_mark_dirty(bs);
4865     if (ret < 0) {
4866         goto fail;
4867     }
4868 
4869     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4870 
4871     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE);
4872     l1_size2 = (uint64_t)s->l1_size * L1E_SIZE;
4873 
4874     /* After this call, neither the in-memory nor the on-disk refcount
4875      * information accurately describe the actual references */
4876 
4877     ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
4878                              l1_clusters * s->cluster_size, 0);
4879     if (ret < 0) {
4880         goto fail_broken_refcounts;
4881     }
4882     memset(s->l1_table, 0, l1_size2);
4883 
4884     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
4885 
4886     /* Overwrite enough clusters at the beginning of the sectors to place
4887      * the refcount table, a refcount block and the L1 table in; this may
4888      * overwrite parts of the existing refcount and L1 table, which is not
4889      * an issue because the dirty flag is set, complete data loss is in fact
4890      * desired and partial data loss is consequently fine as well */
4891     ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
4892                              (2 + l1_clusters) * s->cluster_size, 0);
4893     /* This call (even if it failed overall) may have overwritten on-disk
4894      * refcount structures; in that case, the in-memory refcount information
4895      * will probably differ from the on-disk information which makes the BDS
4896      * unusable */
4897     if (ret < 0) {
4898         goto fail_broken_refcounts;
4899     }
4900 
4901     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4902     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
4903 
4904     /* "Create" an empty reftable (one cluster) directly after the image
4905      * header and an empty L1 table three clusters after the image header;
4906      * the cluster between those two will be used as the first refblock */
4907     l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
4908     l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
4909     l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
4910     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
4911                            sizeof(l1_ofs_rt_ofs_cls), &l1_ofs_rt_ofs_cls, 0);
4912     if (ret < 0) {
4913         goto fail_broken_refcounts;
4914     }
4915 
4916     s->l1_table_offset = 3 * s->cluster_size;
4917 
4918     new_reftable = g_try_new0(uint64_t, s->cluster_size / REFTABLE_ENTRY_SIZE);
4919     if (!new_reftable) {
4920         ret = -ENOMEM;
4921         goto fail_broken_refcounts;
4922     }
4923 
4924     s->refcount_table_offset = s->cluster_size;
4925     s->refcount_table_size   = s->cluster_size / REFTABLE_ENTRY_SIZE;
4926     s->max_refcount_table_index = 0;
4927 
4928     g_free(s->refcount_table);
4929     s->refcount_table = new_reftable;
4930     new_reftable = NULL;
4931 
4932     /* Now the in-memory refcount information again corresponds to the on-disk
4933      * information (reftable is empty and no refblocks (the refblock cache is
4934      * empty)); however, this means some clusters (e.g. the image header) are
4935      * referenced, but not refcounted, but the normal qcow2 code assumes that
4936      * the in-memory information is always correct */
4937 
4938     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
4939 
4940     /* Enter the first refblock into the reftable */
4941     rt_entry = cpu_to_be64(2 * s->cluster_size);
4942     ret = bdrv_pwrite_sync(bs->file, s->cluster_size, sizeof(rt_entry),
4943                            &rt_entry, 0);
4944     if (ret < 0) {
4945         goto fail_broken_refcounts;
4946     }
4947     s->refcount_table[0] = 2 * s->cluster_size;
4948 
4949     s->free_cluster_index = 0;
4950     assert(3 + l1_clusters <= s->refcount_block_size);
4951     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
4952     if (offset < 0) {
4953         ret = offset;
4954         goto fail_broken_refcounts;
4955     } else if (offset > 0) {
4956         error_report("First cluster in emptied image is in use");
4957         abort();
4958     }
4959 
4960     /* Now finally the in-memory information corresponds to the on-disk
4961      * structures and is correct */
4962     ret = qcow2_mark_clean(bs);
4963     if (ret < 0) {
4964         goto fail;
4965     }
4966 
4967     ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false,
4968                         PREALLOC_MODE_OFF, 0, &local_err);
4969     if (ret < 0) {
4970         error_report_err(local_err);
4971         goto fail;
4972     }
4973 
4974     return 0;
4975 
4976 fail_broken_refcounts:
4977     /* The BDS is unusable at this point. If we wanted to make it usable, we
4978      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4979      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4980      * again. However, because the functions which could have caused this error
4981      * path to be taken are used by those functions as well, it's very likely
4982      * that that sequence will fail as well. Therefore, just eject the BDS. */
4983     bs->drv = NULL;
4984 
4985 fail:
4986     g_free(new_reftable);
4987     return ret;
4988 }
4989 
4990 static int GRAPH_RDLOCK qcow2_make_empty(BlockDriverState *bs)
4991 {
4992     BDRVQcow2State *s = bs->opaque;
4993     uint64_t offset, end_offset;
4994     int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
4995     int l1_clusters, ret = 0;
4996 
4997     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE);
4998 
4999     if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps &&
5000         3 + l1_clusters <= s->refcount_block_size &&
5001         s->crypt_method_header != QCOW_CRYPT_LUKS &&
5002         !has_data_file(bs)) {
5003         /* The following function only works for qcow2 v3 images (it
5004          * requires the dirty flag) and only as long as there are no
5005          * features that reserve extra clusters (such as snapshots,
5006          * LUKS header, or persistent bitmaps), because it completely
5007          * empties the image.  Furthermore, the L1 table and three
5008          * additional clusters (image header, refcount table, one
5009          * refcount block) have to fit inside one refcount block. It
5010          * only resets the image file, i.e. does not work with an
5011          * external data file. */
5012         return make_completely_empty(bs);
5013     }
5014 
5015     /* This fallback code simply discards every active cluster; this is slow,
5016      * but works in all cases */
5017     end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
5018     for (offset = 0; offset < end_offset; offset += step) {
5019         /* As this function is generally used after committing an external
5020          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
5021          * default action for this kind of discard is to pass the discard,
5022          * which will ideally result in an actually smaller image file, as
5023          * is probably desired. */
5024         ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
5025                                     QCOW2_DISCARD_SNAPSHOT, true);
5026         if (ret < 0) {
5027             break;
5028         }
5029     }
5030 
5031     return ret;
5032 }
5033 
5034 static coroutine_fn GRAPH_RDLOCK int qcow2_co_flush_to_os(BlockDriverState *bs)
5035 {
5036     BDRVQcow2State *s = bs->opaque;
5037     int ret;
5038 
5039     qemu_co_mutex_lock(&s->lock);
5040     ret = qcow2_write_caches(bs);
5041     qemu_co_mutex_unlock(&s->lock);
5042 
5043     return ret;
5044 }
5045 
5046 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs,
5047                                        Error **errp)
5048 {
5049     Error *local_err = NULL;
5050     BlockMeasureInfo *info;
5051     uint64_t required = 0; /* bytes that contribute to required size */
5052     uint64_t virtual_size; /* disk size as seen by guest */
5053     uint64_t refcount_bits;
5054     uint64_t l2_tables;
5055     uint64_t luks_payload_size = 0;
5056     size_t cluster_size;
5057     int version;
5058     char *optstr;
5059     PreallocMode prealloc;
5060     bool has_backing_file;
5061     bool has_luks;
5062     bool extended_l2;
5063     size_t l2e_size;
5064 
5065     /* Parse image creation options */
5066     extended_l2 = qemu_opt_get_bool_del(opts, BLOCK_OPT_EXTL2, false);
5067 
5068     cluster_size = qcow2_opt_get_cluster_size_del(opts, extended_l2,
5069                                                   &local_err);
5070     if (local_err) {
5071         goto err;
5072     }
5073 
5074     version = qcow2_opt_get_version_del(opts, &local_err);
5075     if (local_err) {
5076         goto err;
5077     }
5078 
5079     refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
5080     if (local_err) {
5081         goto err;
5082     }
5083 
5084     optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
5085     prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr,
5086                                PREALLOC_MODE_OFF, &local_err);
5087     g_free(optstr);
5088     if (local_err) {
5089         goto err;
5090     }
5091 
5092     optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
5093     has_backing_file = !!optstr;
5094     g_free(optstr);
5095 
5096     optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
5097     has_luks = optstr && strcmp(optstr, "luks") == 0;
5098     g_free(optstr);
5099 
5100     if (has_luks) {
5101         g_autoptr(QCryptoBlockCreateOptions) create_opts = NULL;
5102         QDict *cryptoopts = qcow2_extract_crypto_opts(opts, "luks", errp);
5103         size_t headerlen;
5104 
5105         create_opts = block_crypto_create_opts_init(cryptoopts, errp);
5106         qobject_unref(cryptoopts);
5107         if (!create_opts) {
5108             goto err;
5109         }
5110 
5111         if (!qcrypto_block_calculate_payload_offset(create_opts,
5112                                                     "encrypt.",
5113                                                     &headerlen,
5114                                                     &local_err)) {
5115             goto err;
5116         }
5117 
5118         luks_payload_size = ROUND_UP(headerlen, cluster_size);
5119     }
5120 
5121     virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
5122     virtual_size = ROUND_UP(virtual_size, cluster_size);
5123 
5124     /* Check that virtual disk size is valid */
5125     l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL;
5126     l2_tables = DIV_ROUND_UP(virtual_size / cluster_size,
5127                              cluster_size / l2e_size);
5128     if (l2_tables * L1E_SIZE > QCOW_MAX_L1_SIZE) {
5129         error_setg(&local_err, "The image size is too large "
5130                                "(try using a larger cluster size)");
5131         goto err;
5132     }
5133 
5134     /* Account for input image */
5135     if (in_bs) {
5136         int64_t ssize = bdrv_getlength(in_bs);
5137         if (ssize < 0) {
5138             error_setg_errno(&local_err, -ssize,
5139                              "Unable to get image virtual_size");
5140             goto err;
5141         }
5142 
5143         virtual_size = ROUND_UP(ssize, cluster_size);
5144 
5145         if (has_backing_file) {
5146             /* We don't how much of the backing chain is shared by the input
5147              * image and the new image file.  In the worst case the new image's
5148              * backing file has nothing in common with the input image.  Be
5149              * conservative and assume all clusters need to be written.
5150              */
5151             required = virtual_size;
5152         } else {
5153             int64_t offset;
5154             int64_t pnum = 0;
5155 
5156             for (offset = 0; offset < ssize; offset += pnum) {
5157                 int ret;
5158 
5159                 ret = bdrv_block_status_above(in_bs, NULL, offset,
5160                                               ssize - offset, &pnum, NULL,
5161                                               NULL);
5162                 if (ret < 0) {
5163                     error_setg_errno(&local_err, -ret,
5164                                      "Unable to get block status");
5165                     goto err;
5166                 }
5167 
5168                 if (ret & BDRV_BLOCK_ZERO) {
5169                     /* Skip zero regions (safe with no backing file) */
5170                 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
5171                            (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
5172                     /* Extend pnum to end of cluster for next iteration */
5173                     pnum = ROUND_UP(offset + pnum, cluster_size) - offset;
5174 
5175                     /* Count clusters we've seen */
5176                     required += offset % cluster_size + pnum;
5177                 }
5178             }
5179         }
5180     }
5181 
5182     /* Take into account preallocation.  Nothing special is needed for
5183      * PREALLOC_MODE_METADATA since metadata is always counted.
5184      */
5185     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
5186         required = virtual_size;
5187     }
5188 
5189     info = g_new0(BlockMeasureInfo, 1);
5190     info->fully_allocated = luks_payload_size +
5191         qcow2_calc_prealloc_size(virtual_size, cluster_size,
5192                                  ctz32(refcount_bits), extended_l2);
5193 
5194     /*
5195      * Remove data clusters that are not required.  This overestimates the
5196      * required size because metadata needed for the fully allocated file is
5197      * still counted.  Show bitmaps only if both source and destination
5198      * would support them.
5199      */
5200     info->required = info->fully_allocated - virtual_size + required;
5201     info->has_bitmaps = version >= 3 && in_bs &&
5202         bdrv_supports_persistent_dirty_bitmap(in_bs);
5203     if (info->has_bitmaps) {
5204         info->bitmaps = qcow2_get_persistent_dirty_bitmap_size(in_bs,
5205                                                                cluster_size);
5206     }
5207     return info;
5208 
5209 err:
5210     error_propagate(errp, local_err);
5211     return NULL;
5212 }
5213 
5214 static int coroutine_fn
5215 qcow2_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
5216 {
5217     BDRVQcow2State *s = bs->opaque;
5218     bdi->cluster_size = s->cluster_size;
5219     bdi->subcluster_size = s->subcluster_size;
5220     bdi->vm_state_offset = qcow2_vm_state_offset(s);
5221     bdi->is_dirty = s->incompatible_features & QCOW2_INCOMPAT_DIRTY;
5222     return 0;
5223 }
5224 
5225 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs,
5226                                                   Error **errp)
5227 {
5228     BDRVQcow2State *s = bs->opaque;
5229     ImageInfoSpecific *spec_info;
5230     QCryptoBlockInfo *encrypt_info = NULL;
5231 
5232     if (s->crypto != NULL) {
5233         encrypt_info = qcrypto_block_get_info(s->crypto, errp);
5234         if (!encrypt_info) {
5235             return NULL;
5236         }
5237     }
5238 
5239     spec_info = g_new(ImageInfoSpecific, 1);
5240     *spec_info = (ImageInfoSpecific){
5241         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
5242         .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1),
5243     };
5244     if (s->qcow_version == 2) {
5245         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
5246             .compat             = g_strdup("0.10"),
5247             .refcount_bits      = s->refcount_bits,
5248         };
5249     } else if (s->qcow_version == 3) {
5250         Qcow2BitmapInfoList *bitmaps;
5251         if (!qcow2_get_bitmap_info_list(bs, &bitmaps, errp)) {
5252             qapi_free_ImageInfoSpecific(spec_info);
5253             qapi_free_QCryptoBlockInfo(encrypt_info);
5254             return NULL;
5255         }
5256         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
5257             .compat             = g_strdup("1.1"),
5258             .lazy_refcounts     = s->compatible_features &
5259                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
5260             .has_lazy_refcounts = true,
5261             .corrupt            = s->incompatible_features &
5262                                   QCOW2_INCOMPAT_CORRUPT,
5263             .has_corrupt        = true,
5264             .has_extended_l2    = true,
5265             .extended_l2        = has_subclusters(s),
5266             .refcount_bits      = s->refcount_bits,
5267             .has_bitmaps        = !!bitmaps,
5268             .bitmaps            = bitmaps,
5269             .data_file          = g_strdup(s->image_data_file),
5270             .has_data_file_raw  = has_data_file(bs),
5271             .data_file_raw      = data_file_is_raw(bs),
5272             .compression_type   = s->compression_type,
5273         };
5274     } else {
5275         /* if this assertion fails, this probably means a new version was
5276          * added without having it covered here */
5277         assert(false);
5278     }
5279 
5280     if (encrypt_info) {
5281         ImageInfoSpecificQCow2Encryption *qencrypt =
5282             g_new(ImageInfoSpecificQCow2Encryption, 1);
5283         switch (encrypt_info->format) {
5284         case Q_CRYPTO_BLOCK_FORMAT_QCOW:
5285             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES;
5286             break;
5287         case Q_CRYPTO_BLOCK_FORMAT_LUKS:
5288             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS;
5289             qencrypt->u.luks = encrypt_info->u.luks;
5290             break;
5291         default:
5292             abort();
5293         }
5294         /* Since we did shallow copy above, erase any pointers
5295          * in the original info */
5296         memset(&encrypt_info->u, 0, sizeof(encrypt_info->u));
5297         qapi_free_QCryptoBlockInfo(encrypt_info);
5298 
5299         spec_info->u.qcow2.data->encrypt = qencrypt;
5300     }
5301 
5302     return spec_info;
5303 }
5304 
5305 static int coroutine_mixed_fn qcow2_has_zero_init(BlockDriverState *bs)
5306 {
5307     BDRVQcow2State *s = bs->opaque;
5308     bool preallocated;
5309 
5310     if (qemu_in_coroutine()) {
5311         qemu_co_mutex_lock(&s->lock);
5312     }
5313     /*
5314      * Check preallocation status: Preallocated images have all L2
5315      * tables allocated, nonpreallocated images have none.  It is
5316      * therefore enough to check the first one.
5317      */
5318     preallocated = s->l1_size > 0 && s->l1_table[0] != 0;
5319     if (qemu_in_coroutine()) {
5320         qemu_co_mutex_unlock(&s->lock);
5321     }
5322 
5323     if (!preallocated) {
5324         return 1;
5325     } else if (bs->encrypted) {
5326         return 0;
5327     } else {
5328         return bdrv_has_zero_init(s->data_file->bs);
5329     }
5330 }
5331 
5332 /*
5333  * Check the request to vmstate. On success return
5334  *      qcow2_vm_state_offset(bs) + @pos
5335  */
5336 static int64_t qcow2_check_vmstate_request(BlockDriverState *bs,
5337                                            QEMUIOVector *qiov, int64_t pos)
5338 {
5339     BDRVQcow2State *s = bs->opaque;
5340     int64_t vmstate_offset = qcow2_vm_state_offset(s);
5341     int ret;
5342 
5343     /* Incoming requests must be OK */
5344     bdrv_check_qiov_request(pos, qiov->size, qiov, 0, &error_abort);
5345 
5346     if (INT64_MAX - pos < vmstate_offset) {
5347         return -EIO;
5348     }
5349 
5350     pos += vmstate_offset;
5351     ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL);
5352     if (ret < 0) {
5353         return ret;
5354     }
5355 
5356     return pos;
5357 }
5358 
5359 static int coroutine_fn GRAPH_RDLOCK
5360 qcow2_co_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
5361 {
5362     int64_t offset = qcow2_check_vmstate_request(bs, qiov, pos);
5363     if (offset < 0) {
5364         return offset;
5365     }
5366 
5367     BLKDBG_CO_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
5368     return bs->drv->bdrv_co_pwritev_part(bs, offset, qiov->size, qiov, 0, 0);
5369 }
5370 
5371 static int coroutine_fn GRAPH_RDLOCK
5372 qcow2_co_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
5373 {
5374     int64_t offset = qcow2_check_vmstate_request(bs, qiov, pos);
5375     if (offset < 0) {
5376         return offset;
5377     }
5378 
5379     BLKDBG_CO_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
5380     return bs->drv->bdrv_co_preadv_part(bs, offset, qiov->size, qiov, 0, 0);
5381 }
5382 
5383 static int GRAPH_RDLOCK qcow2_has_compressed_clusters(BlockDriverState *bs)
5384 {
5385     int64_t offset = 0;
5386     int64_t bytes = bdrv_getlength(bs);
5387 
5388     if (bytes < 0) {
5389         return bytes;
5390     }
5391 
5392     while (bytes != 0) {
5393         int ret;
5394         QCow2SubclusterType type;
5395         unsigned int cur_bytes = MIN(INT_MAX, bytes);
5396         uint64_t host_offset;
5397 
5398         ret = qcow2_get_host_offset(bs, offset, &cur_bytes, &host_offset,
5399                                     &type);
5400         if (ret < 0) {
5401             return ret;
5402         }
5403 
5404         if (type == QCOW2_SUBCLUSTER_COMPRESSED) {
5405             return 1;
5406         }
5407 
5408         offset += cur_bytes;
5409         bytes -= cur_bytes;
5410     }
5411 
5412     return 0;
5413 }
5414 
5415 /*
5416  * Downgrades an image's version. To achieve this, any incompatible features
5417  * have to be removed.
5418  */
5419 static int GRAPH_RDLOCK
5420 qcow2_downgrade(BlockDriverState *bs, int target_version,
5421                 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5422                 Error **errp)
5423 {
5424     BDRVQcow2State *s = bs->opaque;
5425     int current_version = s->qcow_version;
5426     int ret;
5427     int i;
5428 
5429     /* This is qcow2_downgrade(), not qcow2_upgrade() */
5430     assert(target_version < current_version);
5431 
5432     /* There are no other versions (now) that you can downgrade to */
5433     assert(target_version == 2);
5434 
5435     if (s->refcount_order != 4) {
5436         error_setg(errp, "compat=0.10 requires refcount_bits=16");
5437         return -ENOTSUP;
5438     }
5439 
5440     if (has_data_file(bs)) {
5441         error_setg(errp, "Cannot downgrade an image with a data file");
5442         return -ENOTSUP;
5443     }
5444 
5445     /*
5446      * If any internal snapshot has a different size than the current
5447      * image size, or VM state size that exceeds 32 bits, downgrading
5448      * is unsafe.  Even though we would still use v3-compliant output
5449      * to preserve that data, other v2 programs might not realize
5450      * those optional fields are important.
5451      */
5452     for (i = 0; i < s->nb_snapshots; i++) {
5453         if (s->snapshots[i].vm_state_size > UINT32_MAX ||
5454             s->snapshots[i].disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) {
5455             error_setg(errp, "Internal snapshots prevent downgrade of image");
5456             return -ENOTSUP;
5457         }
5458     }
5459 
5460     /* clear incompatible features */
5461     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
5462         ret = qcow2_mark_clean(bs);
5463         if (ret < 0) {
5464             error_setg_errno(errp, -ret, "Failed to make the image clean");
5465             return ret;
5466         }
5467     }
5468 
5469     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
5470      * the first place; if that happens nonetheless, returning -ENOTSUP is the
5471      * best thing to do anyway */
5472 
5473     if (s->incompatible_features & ~QCOW2_INCOMPAT_COMPRESSION) {
5474         error_setg(errp, "Cannot downgrade an image with incompatible features "
5475                    "0x%" PRIx64 " set",
5476                    s->incompatible_features & ~QCOW2_INCOMPAT_COMPRESSION);
5477         return -ENOTSUP;
5478     }
5479 
5480     /* since we can ignore compatible features, we can set them to 0 as well */
5481     s->compatible_features = 0;
5482     /* if lazy refcounts have been used, they have already been fixed through
5483      * clearing the dirty flag */
5484 
5485     /* clearing autoclear features is trivial */
5486     s->autoclear_features = 0;
5487 
5488     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
5489     if (ret < 0) {
5490         error_setg_errno(errp, -ret, "Failed to turn zero into data clusters");
5491         return ret;
5492     }
5493 
5494     if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) {
5495         ret = qcow2_has_compressed_clusters(bs);
5496         if (ret < 0) {
5497             error_setg(errp, "Failed to check block status");
5498             return -EINVAL;
5499         }
5500         if (ret) {
5501             error_setg(errp, "Cannot downgrade an image with zstd compression "
5502                        "type and existing compressed clusters");
5503             return -ENOTSUP;
5504         }
5505         /*
5506          * No compressed clusters for now, so just chose default zlib
5507          * compression.
5508          */
5509         s->incompatible_features &= ~QCOW2_INCOMPAT_COMPRESSION;
5510         s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
5511     }
5512 
5513     assert(s->incompatible_features == 0);
5514 
5515     s->qcow_version = target_version;
5516     ret = qcow2_update_header(bs);
5517     if (ret < 0) {
5518         s->qcow_version = current_version;
5519         error_setg_errno(errp, -ret, "Failed to update the image header");
5520         return ret;
5521     }
5522     return 0;
5523 }
5524 
5525 /*
5526  * Upgrades an image's version.  While newer versions encompass all
5527  * features of older versions, some things may have to be presented
5528  * differently.
5529  */
5530 static int GRAPH_RDLOCK
5531 qcow2_upgrade(BlockDriverState *bs, int target_version,
5532               BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5533               Error **errp)
5534 {
5535     BDRVQcow2State *s = bs->opaque;
5536     bool need_snapshot_update;
5537     int current_version = s->qcow_version;
5538     int i;
5539     int ret;
5540 
5541     /* This is qcow2_upgrade(), not qcow2_downgrade() */
5542     assert(target_version > current_version);
5543 
5544     /* There are no other versions (yet) that you can upgrade to */
5545     assert(target_version == 3);
5546 
5547     status_cb(bs, 0, 2, cb_opaque);
5548 
5549     /*
5550      * In v2, snapshots do not need to have extra data.  v3 requires
5551      * the 64-bit VM state size and the virtual disk size to be
5552      * present.
5553      * qcow2_write_snapshots() will always write the list in the
5554      * v3-compliant format.
5555      */
5556     need_snapshot_update = false;
5557     for (i = 0; i < s->nb_snapshots; i++) {
5558         if (s->snapshots[i].extra_data_size <
5559             sizeof_field(QCowSnapshotExtraData, vm_state_size_large) +
5560             sizeof_field(QCowSnapshotExtraData, disk_size))
5561         {
5562             need_snapshot_update = true;
5563             break;
5564         }
5565     }
5566     if (need_snapshot_update) {
5567         ret = qcow2_write_snapshots(bs);
5568         if (ret < 0) {
5569             error_setg_errno(errp, -ret, "Failed to update the snapshot table");
5570             return ret;
5571         }
5572     }
5573     status_cb(bs, 1, 2, cb_opaque);
5574 
5575     s->qcow_version = target_version;
5576     ret = qcow2_update_header(bs);
5577     if (ret < 0) {
5578         s->qcow_version = current_version;
5579         error_setg_errno(errp, -ret, "Failed to update the image header");
5580         return ret;
5581     }
5582     status_cb(bs, 2, 2, cb_opaque);
5583 
5584     return 0;
5585 }
5586 
5587 typedef enum Qcow2AmendOperation {
5588     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5589      * statically initialized to so that the helper CB can discern the first
5590      * invocation from an operation change */
5591     QCOW2_NO_OPERATION = 0,
5592 
5593     QCOW2_UPGRADING,
5594     QCOW2_UPDATING_ENCRYPTION,
5595     QCOW2_CHANGING_REFCOUNT_ORDER,
5596     QCOW2_DOWNGRADING,
5597 } Qcow2AmendOperation;
5598 
5599 typedef struct Qcow2AmendHelperCBInfo {
5600     /* The code coordinating the amend operations should only modify
5601      * these four fields; the rest will be managed by the CB */
5602     BlockDriverAmendStatusCB *original_status_cb;
5603     void *original_cb_opaque;
5604 
5605     Qcow2AmendOperation current_operation;
5606 
5607     /* Total number of operations to perform (only set once) */
5608     int total_operations;
5609 
5610     /* The following fields are managed by the CB */
5611 
5612     /* Number of operations completed */
5613     int operations_completed;
5614 
5615     /* Cumulative offset of all completed operations */
5616     int64_t offset_completed;
5617 
5618     Qcow2AmendOperation last_operation;
5619     int64_t last_work_size;
5620 } Qcow2AmendHelperCBInfo;
5621 
5622 static void qcow2_amend_helper_cb(BlockDriverState *bs,
5623                                   int64_t operation_offset,
5624                                   int64_t operation_work_size, void *opaque)
5625 {
5626     Qcow2AmendHelperCBInfo *info = opaque;
5627     int64_t current_work_size;
5628     int64_t projected_work_size;
5629 
5630     if (info->current_operation != info->last_operation) {
5631         if (info->last_operation != QCOW2_NO_OPERATION) {
5632             info->offset_completed += info->last_work_size;
5633             info->operations_completed++;
5634         }
5635 
5636         info->last_operation = info->current_operation;
5637     }
5638 
5639     assert(info->total_operations > 0);
5640     assert(info->operations_completed < info->total_operations);
5641 
5642     info->last_work_size = operation_work_size;
5643 
5644     current_work_size = info->offset_completed + operation_work_size;
5645 
5646     /* current_work_size is the total work size for (operations_completed + 1)
5647      * operations (which includes this one), so multiply it by the number of
5648      * operations not covered and divide it by the number of operations
5649      * covered to get a projection for the operations not covered */
5650     projected_work_size = current_work_size * (info->total_operations -
5651                                                info->operations_completed - 1)
5652                                             / (info->operations_completed + 1);
5653 
5654     info->original_status_cb(bs, info->offset_completed + operation_offset,
5655                              current_work_size + projected_work_size,
5656                              info->original_cb_opaque);
5657 }
5658 
5659 static int GRAPH_RDLOCK
5660 qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
5661                     BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5662                     bool force, Error **errp)
5663 {
5664     BDRVQcow2State *s = bs->opaque;
5665     int old_version = s->qcow_version, new_version = old_version;
5666     uint64_t new_size = 0;
5667     const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL;
5668     bool lazy_refcounts = s->use_lazy_refcounts;
5669     bool data_file_raw = data_file_is_raw(bs);
5670     const char *compat = NULL;
5671     int refcount_bits = s->refcount_bits;
5672     int ret;
5673     QemuOptDesc *desc = opts->list->desc;
5674     Qcow2AmendHelperCBInfo helper_cb_info;
5675     bool encryption_update = false;
5676 
5677     while (desc && desc->name) {
5678         if (!qemu_opt_find(opts, desc->name)) {
5679             /* only change explicitly defined options */
5680             desc++;
5681             continue;
5682         }
5683 
5684         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
5685             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
5686             if (!compat) {
5687                 /* preserve default */
5688             } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) {
5689                 new_version = 2;
5690             } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) {
5691                 new_version = 3;
5692             } else {
5693                 error_setg(errp, "Unknown compatibility level %s", compat);
5694                 return -EINVAL;
5695             }
5696         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
5697             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
5698         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
5699             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5700         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
5701             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5702         } else if (g_str_has_prefix(desc->name, "encrypt.")) {
5703             if (!s->crypto) {
5704                 error_setg(errp,
5705                            "Can't amend encryption options - encryption not present");
5706                 return -EINVAL;
5707             }
5708             if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
5709                 error_setg(errp,
5710                            "Only LUKS encryption options can be amended");
5711                 return -ENOTSUP;
5712             }
5713             encryption_update = true;
5714         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
5715             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
5716                                                lazy_refcounts);
5717         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
5718             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
5719                                                 refcount_bits);
5720 
5721             if (refcount_bits <= 0 || refcount_bits > 64 ||
5722                 !is_power_of_2(refcount_bits))
5723             {
5724                 error_setg(errp, "Refcount width must be a power of two and "
5725                            "may not exceed 64 bits");
5726                 return -EINVAL;
5727             }
5728         } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) {
5729             data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE);
5730             if (data_file && !has_data_file(bs)) {
5731                 error_setg(errp, "data-file can only be set for images that "
5732                                  "use an external data file");
5733                 return -EINVAL;
5734             }
5735         } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) {
5736             data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW,
5737                                               data_file_raw);
5738             if (data_file_raw && !data_file_is_raw(bs)) {
5739                 error_setg(errp, "data-file-raw cannot be set on existing "
5740                                  "images");
5741                 return -EINVAL;
5742             }
5743         } else {
5744             /* if this point is reached, this probably means a new option was
5745              * added without having it covered here */
5746             abort();
5747         }
5748 
5749         desc++;
5750     }
5751 
5752     helper_cb_info = (Qcow2AmendHelperCBInfo){
5753         .original_status_cb = status_cb,
5754         .original_cb_opaque = cb_opaque,
5755         .total_operations = (new_version != old_version)
5756                           + (s->refcount_bits != refcount_bits) +
5757                             (encryption_update == true)
5758     };
5759 
5760     /* Upgrade first (some features may require compat=1.1) */
5761     if (new_version > old_version) {
5762         helper_cb_info.current_operation = QCOW2_UPGRADING;
5763         ret = qcow2_upgrade(bs, new_version, &qcow2_amend_helper_cb,
5764                             &helper_cb_info, errp);
5765         if (ret < 0) {
5766             return ret;
5767         }
5768     }
5769 
5770     if (encryption_update) {
5771         QDict *amend_opts_dict;
5772         QCryptoBlockAmendOptions *amend_opts;
5773 
5774         helper_cb_info.current_operation = QCOW2_UPDATING_ENCRYPTION;
5775         amend_opts_dict = qcow2_extract_crypto_opts(opts, "luks", errp);
5776         if (!amend_opts_dict) {
5777             return -EINVAL;
5778         }
5779         amend_opts = block_crypto_amend_opts_init(amend_opts_dict, errp);
5780         qobject_unref(amend_opts_dict);
5781         if (!amend_opts) {
5782             return -EINVAL;
5783         }
5784         ret = qcrypto_block_amend_options(s->crypto,
5785                                           qcow2_crypto_hdr_read_func,
5786                                           qcow2_crypto_hdr_write_func,
5787                                           bs,
5788                                           amend_opts,
5789                                           force,
5790                                           errp);
5791         qapi_free_QCryptoBlockAmendOptions(amend_opts);
5792         if (ret < 0) {
5793             return ret;
5794         }
5795     }
5796 
5797     if (s->refcount_bits != refcount_bits) {
5798         int refcount_order = ctz32(refcount_bits);
5799 
5800         if (new_version < 3 && refcount_bits != 16) {
5801             error_setg(errp, "Refcount widths other than 16 bits require "
5802                        "compatibility level 1.1 or above (use compat=1.1 or "
5803                        "greater)");
5804             return -EINVAL;
5805         }
5806 
5807         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
5808         ret = qcow2_change_refcount_order(bs, refcount_order,
5809                                           &qcow2_amend_helper_cb,
5810                                           &helper_cb_info, errp);
5811         if (ret < 0) {
5812             return ret;
5813         }
5814     }
5815 
5816     /* data-file-raw blocks backing files, so clear it first if requested */
5817     if (data_file_raw) {
5818         s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5819     } else {
5820         s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5821     }
5822 
5823     if (data_file) {
5824         g_free(s->image_data_file);
5825         s->image_data_file = *data_file ? g_strdup(data_file) : NULL;
5826     }
5827 
5828     ret = qcow2_update_header(bs);
5829     if (ret < 0) {
5830         error_setg_errno(errp, -ret, "Failed to update the image header");
5831         return ret;
5832     }
5833 
5834     if (backing_file || backing_format) {
5835         if (g_strcmp0(backing_file, s->image_backing_file) ||
5836             g_strcmp0(backing_format, s->image_backing_format)) {
5837             error_setg(errp, "Cannot amend the backing file");
5838             error_append_hint(errp,
5839                               "You can use 'qemu-img rebase' instead.\n");
5840             return -EINVAL;
5841         }
5842     }
5843 
5844     if (s->use_lazy_refcounts != lazy_refcounts) {
5845         if (lazy_refcounts) {
5846             if (new_version < 3) {
5847                 error_setg(errp, "Lazy refcounts only supported with "
5848                            "compatibility level 1.1 and above (use compat=1.1 "
5849                            "or greater)");
5850                 return -EINVAL;
5851             }
5852             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5853             ret = qcow2_update_header(bs);
5854             if (ret < 0) {
5855                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5856                 error_setg_errno(errp, -ret, "Failed to update the image header");
5857                 return ret;
5858             }
5859             s->use_lazy_refcounts = true;
5860         } else {
5861             /* make image clean first */
5862             ret = qcow2_mark_clean(bs);
5863             if (ret < 0) {
5864                 error_setg_errno(errp, -ret, "Failed to make the image clean");
5865                 return ret;
5866             }
5867             /* now disallow lazy refcounts */
5868             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5869             ret = qcow2_update_header(bs);
5870             if (ret < 0) {
5871                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5872                 error_setg_errno(errp, -ret, "Failed to update the image header");
5873                 return ret;
5874             }
5875             s->use_lazy_refcounts = false;
5876         }
5877     }
5878 
5879     if (new_size) {
5880         BlockBackend *blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL,
5881                                             errp);
5882         if (!blk) {
5883             return -EPERM;
5884         }
5885 
5886         /*
5887          * Amending image options should ensure that the image has
5888          * exactly the given new values, so pass exact=true here.
5889          */
5890         ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, 0, errp);
5891         blk_unref(blk);
5892         if (ret < 0) {
5893             return ret;
5894         }
5895     }
5896 
5897     /* Downgrade last (so unsupported features can be removed before) */
5898     if (new_version < old_version) {
5899         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
5900         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
5901                               &helper_cb_info, errp);
5902         if (ret < 0) {
5903             return ret;
5904         }
5905     }
5906 
5907     return 0;
5908 }
5909 
5910 static int coroutine_fn qcow2_co_amend(BlockDriverState *bs,
5911                                        BlockdevAmendOptions *opts,
5912                                        bool force,
5913                                        Error **errp)
5914 {
5915     BlockdevAmendOptionsQcow2 *qopts = &opts->u.qcow2;
5916     BDRVQcow2State *s = bs->opaque;
5917     int ret = 0;
5918 
5919     if (qopts->encrypt) {
5920         if (!s->crypto) {
5921             error_setg(errp, "image is not encrypted, can't amend");
5922             return -EOPNOTSUPP;
5923         }
5924 
5925         if (qopts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_LUKS) {
5926             error_setg(errp,
5927                        "Amend can't be used to change the qcow2 encryption format");
5928             return -EOPNOTSUPP;
5929         }
5930 
5931         if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
5932             error_setg(errp,
5933                        "Only LUKS encryption options can be amended for qcow2 with blockdev-amend");
5934             return -EOPNOTSUPP;
5935         }
5936 
5937         ret = qcrypto_block_amend_options(s->crypto,
5938                                           qcow2_crypto_hdr_read_func,
5939                                           qcow2_crypto_hdr_write_func,
5940                                           bs,
5941                                           qopts->encrypt,
5942                                           force,
5943                                           errp);
5944     }
5945     return ret;
5946 }
5947 
5948 /*
5949  * If offset or size are negative, respectively, they will not be included in
5950  * the BLOCK_IMAGE_CORRUPTED event emitted.
5951  * fatal will be ignored for read-only BDS; corruptions found there will always
5952  * be considered non-fatal.
5953  */
5954 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
5955                              int64_t size, const char *message_format, ...)
5956 {
5957     BDRVQcow2State *s = bs->opaque;
5958     const char *node_name;
5959     char *message;
5960     va_list ap;
5961 
5962     fatal = fatal && bdrv_is_writable(bs);
5963 
5964     if (s->signaled_corruption &&
5965         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
5966     {
5967         return;
5968     }
5969 
5970     va_start(ap, message_format);
5971     message = g_strdup_vprintf(message_format, ap);
5972     va_end(ap);
5973 
5974     if (fatal) {
5975         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
5976                 "corruption events will be suppressed\n", message);
5977     } else {
5978         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
5979                 "corruption events will be suppressed\n", message);
5980     }
5981 
5982     node_name = bdrv_get_node_name(bs);
5983     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
5984                                           *node_name ? node_name : NULL,
5985                                           message, offset >= 0, offset,
5986                                           size >= 0, size,
5987                                           fatal);
5988     g_free(message);
5989 
5990     if (fatal) {
5991         qcow2_mark_corrupt(bs);
5992         bs->drv = NULL; /* make BDS unusable */
5993     }
5994 
5995     s->signaled_corruption = true;
5996 }
5997 
5998 #define QCOW_COMMON_OPTIONS                                         \
5999     {                                                               \
6000         .name = BLOCK_OPT_SIZE,                                     \
6001         .type = QEMU_OPT_SIZE,                                      \
6002         .help = "Virtual disk size"                                 \
6003     },                                                              \
6004     {                                                               \
6005         .name = BLOCK_OPT_COMPAT_LEVEL,                             \
6006         .type = QEMU_OPT_STRING,                                    \
6007         .help = "Compatibility level (v2 [0.10] or v3 [1.1])"       \
6008     },                                                              \
6009     {                                                               \
6010         .name = BLOCK_OPT_BACKING_FILE,                             \
6011         .type = QEMU_OPT_STRING,                                    \
6012         .help = "File name of a base image"                         \
6013     },                                                              \
6014     {                                                               \
6015         .name = BLOCK_OPT_BACKING_FMT,                              \
6016         .type = QEMU_OPT_STRING,                                    \
6017         .help = "Image format of the base image"                    \
6018     },                                                              \
6019     {                                                               \
6020         .name = BLOCK_OPT_DATA_FILE,                                \
6021         .type = QEMU_OPT_STRING,                                    \
6022         .help = "File name of an external data file"                \
6023     },                                                              \
6024     {                                                               \
6025         .name = BLOCK_OPT_DATA_FILE_RAW,                            \
6026         .type = QEMU_OPT_BOOL,                                      \
6027         .help = "The external data file must stay valid "           \
6028                 "as a raw image"                                    \
6029     },                                                              \
6030     {                                                               \
6031         .name = BLOCK_OPT_LAZY_REFCOUNTS,                           \
6032         .type = QEMU_OPT_BOOL,                                      \
6033         .help = "Postpone refcount updates",                        \
6034         .def_value_str = "off"                                      \
6035     },                                                              \
6036     {                                                               \
6037         .name = BLOCK_OPT_REFCOUNT_BITS,                            \
6038         .type = QEMU_OPT_NUMBER,                                    \
6039         .help = "Width of a reference count entry in bits",         \
6040         .def_value_str = "16"                                       \
6041     }
6042 
6043 static QemuOptsList qcow2_create_opts = {
6044     .name = "qcow2-create-opts",
6045     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
6046     .desc = {
6047         {                                                               \
6048             .name = BLOCK_OPT_ENCRYPT,                                  \
6049             .type = QEMU_OPT_BOOL,                                      \
6050             .help = "Encrypt the image with format 'aes'. (Deprecated " \
6051                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",    \
6052         },                                                              \
6053         {                                                               \
6054             .name = BLOCK_OPT_ENCRYPT_FORMAT,                           \
6055             .type = QEMU_OPT_STRING,                                    \
6056             .help = "Encrypt the image, format choices: 'aes', 'luks'", \
6057         },                                                              \
6058         BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",                     \
6059             "ID of secret providing qcow AES key or LUKS passphrase"),  \
6060         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),               \
6061         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),              \
6062         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),                \
6063         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),           \
6064         BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),                 \
6065         BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),                \
6066         {                                                               \
6067             .name = BLOCK_OPT_CLUSTER_SIZE,                             \
6068             .type = QEMU_OPT_SIZE,                                      \
6069             .help = "qcow2 cluster size",                               \
6070             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)            \
6071         },                                                              \
6072         {                                                               \
6073             .name = BLOCK_OPT_EXTL2,                                    \
6074             .type = QEMU_OPT_BOOL,                                      \
6075             .help = "Extended L2 tables",                               \
6076             .def_value_str = "off"                                      \
6077         },                                                              \
6078         {                                                               \
6079             .name = BLOCK_OPT_PREALLOC,                                 \
6080             .type = QEMU_OPT_STRING,                                    \
6081             .help = "Preallocation mode (allowed values: off, "         \
6082                     "metadata, falloc, full)"                           \
6083         },                                                              \
6084         {                                                               \
6085             .name = BLOCK_OPT_COMPRESSION_TYPE,                         \
6086             .type = QEMU_OPT_STRING,                                    \
6087             .help = "Compression method used for image cluster "        \
6088                     "compression",                                      \
6089             .def_value_str = "zlib"                                     \
6090         },
6091         QCOW_COMMON_OPTIONS,
6092         { /* end of list */ }
6093     }
6094 };
6095 
6096 static QemuOptsList qcow2_amend_opts = {
6097     .name = "qcow2-amend-opts",
6098     .head = QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts.head),
6099     .desc = {
6100         BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."),
6101         BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."),
6102         BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."),
6103         BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."),
6104         BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
6105         QCOW_COMMON_OPTIONS,
6106         { /* end of list */ }
6107     }
6108 };
6109 
6110 static const char *const qcow2_strong_runtime_opts[] = {
6111     "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
6112 
6113     NULL
6114 };
6115 
6116 BlockDriver bdrv_qcow2 = {
6117     .format_name        = "qcow2",
6118     .instance_size      = sizeof(BDRVQcow2State),
6119     .bdrv_probe         = qcow2_probe,
6120     .bdrv_open          = qcow2_open,
6121     .bdrv_close         = qcow2_close,
6122     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
6123     .bdrv_reopen_commit   = qcow2_reopen_commit,
6124     .bdrv_reopen_commit_post = qcow2_reopen_commit_post,
6125     .bdrv_reopen_abort    = qcow2_reopen_abort,
6126     .bdrv_join_options    = qcow2_join_options,
6127     .bdrv_child_perm      = bdrv_default_perms,
6128     .bdrv_co_create_opts  = qcow2_co_create_opts,
6129     .bdrv_co_create       = qcow2_co_create,
6130     .bdrv_has_zero_init   = qcow2_has_zero_init,
6131     .bdrv_co_block_status = qcow2_co_block_status,
6132 
6133     .bdrv_co_preadv_part    = qcow2_co_preadv_part,
6134     .bdrv_co_pwritev_part   = qcow2_co_pwritev_part,
6135     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
6136 
6137     .bdrv_co_pwrite_zeroes  = qcow2_co_pwrite_zeroes,
6138     .bdrv_co_pdiscard       = qcow2_co_pdiscard,
6139     .bdrv_co_copy_range_from = qcow2_co_copy_range_from,
6140     .bdrv_co_copy_range_to  = qcow2_co_copy_range_to,
6141     .bdrv_co_truncate       = qcow2_co_truncate,
6142     .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part,
6143     .bdrv_make_empty        = qcow2_make_empty,
6144 
6145     .bdrv_snapshot_create   = qcow2_snapshot_create,
6146     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
6147     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
6148     .bdrv_snapshot_list     = qcow2_snapshot_list,
6149     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
6150     .bdrv_measure           = qcow2_measure,
6151     .bdrv_co_get_info       = qcow2_co_get_info,
6152     .bdrv_get_specific_info = qcow2_get_specific_info,
6153 
6154     .bdrv_co_save_vmstate   = qcow2_co_save_vmstate,
6155     .bdrv_co_load_vmstate   = qcow2_co_load_vmstate,
6156 
6157     .is_format                  = true,
6158     .supports_backing           = true,
6159     .bdrv_change_backing_file   = qcow2_change_backing_file,
6160 
6161     .bdrv_refresh_limits        = qcow2_refresh_limits,
6162     .bdrv_co_invalidate_cache   = qcow2_co_invalidate_cache,
6163     .bdrv_inactivate            = qcow2_inactivate,
6164 
6165     .create_opts         = &qcow2_create_opts,
6166     .amend_opts          = &qcow2_amend_opts,
6167     .strong_runtime_opts = qcow2_strong_runtime_opts,
6168     .mutable_opts        = mutable_opts,
6169     .bdrv_co_check       = qcow2_co_check,
6170     .bdrv_amend_options  = qcow2_amend_options,
6171     .bdrv_co_amend       = qcow2_co_amend,
6172 
6173     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
6174     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
6175 
6176     .bdrv_supports_persistent_dirty_bitmap =
6177             qcow2_supports_persistent_dirty_bitmap,
6178     .bdrv_co_can_store_new_dirty_bitmap = qcow2_co_can_store_new_dirty_bitmap,
6179     .bdrv_co_remove_persistent_dirty_bitmap =
6180             qcow2_co_remove_persistent_dirty_bitmap,
6181 };
6182 
6183 static void bdrv_qcow2_init(void)
6184 {
6185     bdrv_register(&bdrv_qcow2);
6186 }
6187 
6188 block_init(bdrv_qcow2_init);
6189