xref: /openbmc/qemu/block/qcow2.c (revision 06329cce)
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 #include "block/block_int.h"
27 #include "sysemu/block-backend.h"
28 #include "qemu/module.h"
29 #include <zlib.h>
30 #include "block/qcow2.h"
31 #include "qemu/error-report.h"
32 #include "qapi/error.h"
33 #include "qapi/qmp/qerror.h"
34 #include "qapi/qmp/qdict.h"
35 #include "qapi/qmp/qstring.h"
36 #include "qapi-event.h"
37 #include "trace.h"
38 #include "qemu/option_int.h"
39 #include "qemu/cutils.h"
40 #include "qemu/bswap.h"
41 #include "qapi/opts-visitor.h"
42 #include "qapi-visit.h"
43 #include "block/crypto.h"
44 
45 /*
46   Differences with QCOW:
47 
48   - Support for multiple incremental snapshots.
49   - Memory management by reference counts.
50   - Clusters which have a reference count of one have the bit
51     QCOW_OFLAG_COPIED to optimize write performance.
52   - Size of compressed clusters is stored in sectors to reduce bit usage
53     in the cluster offsets.
54   - Support for storing additional data (such as the VM state) in the
55     snapshots.
56   - If a backing store is used, the cluster size is not constrained
57     (could be backported to QCOW).
58   - L2 tables have always a size of one cluster.
59 */
60 
61 
62 typedef struct {
63     uint32_t magic;
64     uint32_t len;
65 } QEMU_PACKED QCowExtension;
66 
67 #define  QCOW2_EXT_MAGIC_END 0
68 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
69 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
70 #define  QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
71 #define  QCOW2_EXT_MAGIC_BITMAPS 0x23852875
72 
73 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
74 {
75     const QCowHeader *cow_header = (const void *)buf;
76 
77     if (buf_size >= sizeof(QCowHeader) &&
78         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
79         be32_to_cpu(cow_header->version) >= 2)
80         return 100;
81     else
82         return 0;
83 }
84 
85 
86 static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset,
87                                           uint8_t *buf, size_t buflen,
88                                           void *opaque, Error **errp)
89 {
90     BlockDriverState *bs = opaque;
91     BDRVQcow2State *s = bs->opaque;
92     ssize_t ret;
93 
94     if ((offset + buflen) > s->crypto_header.length) {
95         error_setg(errp, "Request for data outside of extension header");
96         return -1;
97     }
98 
99     ret = bdrv_pread(bs->file,
100                      s->crypto_header.offset + offset, buf, buflen);
101     if (ret < 0) {
102         error_setg_errno(errp, -ret, "Could not read encryption header");
103         return -1;
104     }
105     return ret;
106 }
107 
108 
109 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen,
110                                           void *opaque, Error **errp)
111 {
112     BlockDriverState *bs = opaque;
113     BDRVQcow2State *s = bs->opaque;
114     int64_t ret;
115     int64_t clusterlen;
116 
117     ret = qcow2_alloc_clusters(bs, headerlen);
118     if (ret < 0) {
119         error_setg_errno(errp, -ret,
120                          "Cannot allocate cluster for LUKS header size %zu",
121                          headerlen);
122         return -1;
123     }
124 
125     s->crypto_header.length = headerlen;
126     s->crypto_header.offset = ret;
127 
128     /* Zero fill remaining space in cluster so it has predictable
129      * content in case of future spec changes */
130     clusterlen = size_to_clusters(s, headerlen) * s->cluster_size;
131     assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen) == 0);
132     ret = bdrv_pwrite_zeroes(bs->file,
133                              ret + headerlen,
134                              clusterlen - headerlen, 0);
135     if (ret < 0) {
136         error_setg_errno(errp, -ret, "Could not zero fill encryption header");
137         return -1;
138     }
139 
140     return ret;
141 }
142 
143 
144 static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset,
145                                            const uint8_t *buf, size_t buflen,
146                                            void *opaque, Error **errp)
147 {
148     BlockDriverState *bs = opaque;
149     BDRVQcow2State *s = bs->opaque;
150     ssize_t ret;
151 
152     if ((offset + buflen) > s->crypto_header.length) {
153         error_setg(errp, "Request for data outside of extension header");
154         return -1;
155     }
156 
157     ret = bdrv_pwrite(bs->file,
158                       s->crypto_header.offset + offset, buf, buflen);
159     if (ret < 0) {
160         error_setg_errno(errp, -ret, "Could not read encryption header");
161         return -1;
162     }
163     return ret;
164 }
165 
166 
167 /*
168  * read qcow2 extension and fill bs
169  * start reading from start_offset
170  * finish reading upon magic of value 0 or when end_offset reached
171  * unknown magic is skipped (future extension this version knows nothing about)
172  * return 0 upon success, non-0 otherwise
173  */
174 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
175                                  uint64_t end_offset, void **p_feature_table,
176                                  int flags, bool *need_update_header,
177                                  Error **errp)
178 {
179     BDRVQcow2State *s = bs->opaque;
180     QCowExtension ext;
181     uint64_t offset;
182     int ret;
183     Qcow2BitmapHeaderExt bitmaps_ext;
184 
185     if (need_update_header != NULL) {
186         *need_update_header = false;
187     }
188 
189 #ifdef DEBUG_EXT
190     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
191 #endif
192     offset = start_offset;
193     while (offset < end_offset) {
194 
195 #ifdef DEBUG_EXT
196         /* Sanity check */
197         if (offset > s->cluster_size)
198             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
199 
200         printf("attempting to read extended header in offset %lu\n", offset);
201 #endif
202 
203         ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
204         if (ret < 0) {
205             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
206                              "pread fail from offset %" PRIu64, offset);
207             return 1;
208         }
209         be32_to_cpus(&ext.magic);
210         be32_to_cpus(&ext.len);
211         offset += sizeof(ext);
212 #ifdef DEBUG_EXT
213         printf("ext.magic = 0x%x\n", ext.magic);
214 #endif
215         if (offset > end_offset || ext.len > end_offset - offset) {
216             error_setg(errp, "Header extension too large");
217             return -EINVAL;
218         }
219 
220         switch (ext.magic) {
221         case QCOW2_EXT_MAGIC_END:
222             return 0;
223 
224         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
225             if (ext.len >= sizeof(bs->backing_format)) {
226                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
227                            " too large (>=%zu)", ext.len,
228                            sizeof(bs->backing_format));
229                 return 2;
230             }
231             ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len);
232             if (ret < 0) {
233                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
234                                  "Could not read format name");
235                 return 3;
236             }
237             bs->backing_format[ext.len] = '\0';
238             s->image_backing_format = g_strdup(bs->backing_format);
239 #ifdef DEBUG_EXT
240             printf("Qcow2: Got format extension %s\n", bs->backing_format);
241 #endif
242             break;
243 
244         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
245             if (p_feature_table != NULL) {
246                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
247                 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
248                 if (ret < 0) {
249                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
250                                      "Could not read table");
251                     return ret;
252                 }
253 
254                 *p_feature_table = feature_table;
255             }
256             break;
257 
258         case QCOW2_EXT_MAGIC_CRYPTO_HEADER: {
259             unsigned int cflags = 0;
260             if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
261                 error_setg(errp, "CRYPTO header extension only "
262                            "expected with LUKS encryption method");
263                 return -EINVAL;
264             }
265             if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) {
266                 error_setg(errp, "CRYPTO header extension size %u, "
267                            "but expected size %zu", ext.len,
268                            sizeof(Qcow2CryptoHeaderExtension));
269                 return -EINVAL;
270             }
271 
272             ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len);
273             if (ret < 0) {
274                 error_setg_errno(errp, -ret,
275                                  "Unable to read CRYPTO header extension");
276                 return ret;
277             }
278             be64_to_cpus(&s->crypto_header.offset);
279             be64_to_cpus(&s->crypto_header.length);
280 
281             if ((s->crypto_header.offset % s->cluster_size) != 0) {
282                 error_setg(errp, "Encryption header offset '%" PRIu64 "' is "
283                            "not a multiple of cluster size '%u'",
284                            s->crypto_header.offset, s->cluster_size);
285                 return -EINVAL;
286             }
287 
288             if (flags & BDRV_O_NO_IO) {
289                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
290             }
291             s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
292                                            qcow2_crypto_hdr_read_func,
293                                            bs, cflags, errp);
294             if (!s->crypto) {
295                 return -EINVAL;
296             }
297         }   break;
298 
299         case QCOW2_EXT_MAGIC_BITMAPS:
300             if (ext.len != sizeof(bitmaps_ext)) {
301                 error_setg_errno(errp, -ret, "bitmaps_ext: "
302                                  "Invalid extension length");
303                 return -EINVAL;
304             }
305 
306             if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) {
307                 if (s->qcow_version < 3) {
308                     /* Let's be a bit more specific */
309                     warn_report("This qcow2 v2 image contains bitmaps, but "
310                                 "they may have been modified by a program "
311                                 "without persistent bitmap support; so now "
312                                 "they must all be considered inconsistent");
313                 } else {
314                     warn_report("a program lacking bitmap support "
315                                 "modified this file, so all bitmaps are now "
316                                 "considered inconsistent");
317                 }
318                 error_printf("Some clusters may be leaked, "
319                              "run 'qemu-img check -r' on the image "
320                              "file to fix.");
321                 if (need_update_header != NULL) {
322                     /* Updating is needed to drop invalid bitmap extension. */
323                     *need_update_header = true;
324                 }
325                 break;
326             }
327 
328             ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len);
329             if (ret < 0) {
330                 error_setg_errno(errp, -ret, "bitmaps_ext: "
331                                  "Could not read ext header");
332                 return ret;
333             }
334 
335             if (bitmaps_ext.reserved32 != 0) {
336                 error_setg_errno(errp, -ret, "bitmaps_ext: "
337                                  "Reserved field is not zero");
338                 return -EINVAL;
339             }
340 
341             be32_to_cpus(&bitmaps_ext.nb_bitmaps);
342             be64_to_cpus(&bitmaps_ext.bitmap_directory_size);
343             be64_to_cpus(&bitmaps_ext.bitmap_directory_offset);
344 
345             if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) {
346                 error_setg(errp,
347                            "bitmaps_ext: Image has %" PRIu32 " bitmaps, "
348                            "exceeding the QEMU supported maximum of %d",
349                            bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS);
350                 return -EINVAL;
351             }
352 
353             if (bitmaps_ext.nb_bitmaps == 0) {
354                 error_setg(errp, "found bitmaps extension with zero bitmaps");
355                 return -EINVAL;
356             }
357 
358             if (bitmaps_ext.bitmap_directory_offset & (s->cluster_size - 1)) {
359                 error_setg(errp, "bitmaps_ext: "
360                                  "invalid bitmap directory offset");
361                 return -EINVAL;
362             }
363 
364             if (bitmaps_ext.bitmap_directory_size >
365                 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) {
366                 error_setg(errp, "bitmaps_ext: "
367                                  "bitmap directory size (%" PRIu64 ") exceeds "
368                                  "the maximum supported size (%d)",
369                                  bitmaps_ext.bitmap_directory_size,
370                                  QCOW2_MAX_BITMAP_DIRECTORY_SIZE);
371                 return -EINVAL;
372             }
373 
374             s->nb_bitmaps = bitmaps_ext.nb_bitmaps;
375             s->bitmap_directory_offset =
376                     bitmaps_ext.bitmap_directory_offset;
377             s->bitmap_directory_size =
378                     bitmaps_ext.bitmap_directory_size;
379 
380 #ifdef DEBUG_EXT
381             printf("Qcow2: Got bitmaps extension: "
382                    "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n",
383                    s->bitmap_directory_offset, s->nb_bitmaps);
384 #endif
385             break;
386 
387         default:
388             /* unknown magic - save it in case we need to rewrite the header */
389             /* If you add a new feature, make sure to also update the fast
390              * path of qcow2_make_empty() to deal with it. */
391             {
392                 Qcow2UnknownHeaderExtension *uext;
393 
394                 uext = g_malloc0(sizeof(*uext)  + ext.len);
395                 uext->magic = ext.magic;
396                 uext->len = ext.len;
397                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
398 
399                 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
400                 if (ret < 0) {
401                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
402                                      "Could not read data");
403                     return ret;
404                 }
405             }
406             break;
407         }
408 
409         offset += ((ext.len + 7) & ~7);
410     }
411 
412     return 0;
413 }
414 
415 static void cleanup_unknown_header_ext(BlockDriverState *bs)
416 {
417     BDRVQcow2State *s = bs->opaque;
418     Qcow2UnknownHeaderExtension *uext, *next;
419 
420     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
421         QLIST_REMOVE(uext, next);
422         g_free(uext);
423     }
424 }
425 
426 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
427                                        uint64_t mask)
428 {
429     char *features = g_strdup("");
430     char *old;
431 
432     while (table && table->name[0] != '\0') {
433         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
434             if (mask & (1ULL << table->bit)) {
435                 old = features;
436                 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
437                                            table->name);
438                 g_free(old);
439                 mask &= ~(1ULL << table->bit);
440             }
441         }
442         table++;
443     }
444 
445     if (mask) {
446         old = features;
447         features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
448                                    old, *old ? ", " : "", mask);
449         g_free(old);
450     }
451 
452     error_setg(errp, "Unsupported qcow2 feature(s): %s", features);
453     g_free(features);
454 }
455 
456 /*
457  * Sets the dirty bit and flushes afterwards if necessary.
458  *
459  * The incompatible_features bit is only set if the image file header was
460  * updated successfully.  Therefore it is not required to check the return
461  * value of this function.
462  */
463 int qcow2_mark_dirty(BlockDriverState *bs)
464 {
465     BDRVQcow2State *s = bs->opaque;
466     uint64_t val;
467     int ret;
468 
469     assert(s->qcow_version >= 3);
470 
471     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
472         return 0; /* already dirty */
473     }
474 
475     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
476     ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
477                       &val, sizeof(val));
478     if (ret < 0) {
479         return ret;
480     }
481     ret = bdrv_flush(bs->file->bs);
482     if (ret < 0) {
483         return ret;
484     }
485 
486     /* Only treat image as dirty if the header was updated successfully */
487     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
488     return 0;
489 }
490 
491 /*
492  * Clears the dirty bit and flushes before if necessary.  Only call this
493  * function when there are no pending requests, it does not guard against
494  * concurrent requests dirtying the image.
495  */
496 static int qcow2_mark_clean(BlockDriverState *bs)
497 {
498     BDRVQcow2State *s = bs->opaque;
499 
500     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
501         int ret;
502 
503         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
504 
505         ret = bdrv_flush(bs);
506         if (ret < 0) {
507             return ret;
508         }
509 
510         return qcow2_update_header(bs);
511     }
512     return 0;
513 }
514 
515 /*
516  * Marks the image as corrupt.
517  */
518 int qcow2_mark_corrupt(BlockDriverState *bs)
519 {
520     BDRVQcow2State *s = bs->opaque;
521 
522     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
523     return qcow2_update_header(bs);
524 }
525 
526 /*
527  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
528  * before if necessary.
529  */
530 int qcow2_mark_consistent(BlockDriverState *bs)
531 {
532     BDRVQcow2State *s = bs->opaque;
533 
534     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
535         int ret = bdrv_flush(bs);
536         if (ret < 0) {
537             return ret;
538         }
539 
540         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
541         return qcow2_update_header(bs);
542     }
543     return 0;
544 }
545 
546 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
547                        BdrvCheckMode fix)
548 {
549     int ret = qcow2_check_refcounts(bs, result, fix);
550     if (ret < 0) {
551         return ret;
552     }
553 
554     if (fix && result->check_errors == 0 && result->corruptions == 0) {
555         ret = qcow2_mark_clean(bs);
556         if (ret < 0) {
557             return ret;
558         }
559         return qcow2_mark_consistent(bs);
560     }
561     return ret;
562 }
563 
564 static int validate_table_offset(BlockDriverState *bs, uint64_t offset,
565                                  uint64_t entries, size_t entry_len)
566 {
567     BDRVQcow2State *s = bs->opaque;
568     uint64_t size;
569 
570     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
571      * because values will be passed to qemu functions taking int64_t. */
572     if (entries > INT64_MAX / entry_len) {
573         return -EINVAL;
574     }
575 
576     size = entries * entry_len;
577 
578     if (INT64_MAX - size < offset) {
579         return -EINVAL;
580     }
581 
582     /* Tables must be cluster aligned */
583     if (offset_into_cluster(s, offset) != 0) {
584         return -EINVAL;
585     }
586 
587     return 0;
588 }
589 
590 static QemuOptsList qcow2_runtime_opts = {
591     .name = "qcow2",
592     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
593     .desc = {
594         {
595             .name = QCOW2_OPT_LAZY_REFCOUNTS,
596             .type = QEMU_OPT_BOOL,
597             .help = "Postpone refcount updates",
598         },
599         {
600             .name = QCOW2_OPT_DISCARD_REQUEST,
601             .type = QEMU_OPT_BOOL,
602             .help = "Pass guest discard requests to the layer below",
603         },
604         {
605             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
606             .type = QEMU_OPT_BOOL,
607             .help = "Generate discard requests when snapshot related space "
608                     "is freed",
609         },
610         {
611             .name = QCOW2_OPT_DISCARD_OTHER,
612             .type = QEMU_OPT_BOOL,
613             .help = "Generate discard requests when other clusters are freed",
614         },
615         {
616             .name = QCOW2_OPT_OVERLAP,
617             .type = QEMU_OPT_STRING,
618             .help = "Selects which overlap checks to perform from a range of "
619                     "templates (none, constant, cached, all)",
620         },
621         {
622             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
623             .type = QEMU_OPT_STRING,
624             .help = "Selects which overlap checks to perform from a range of "
625                     "templates (none, constant, cached, all)",
626         },
627         {
628             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
629             .type = QEMU_OPT_BOOL,
630             .help = "Check for unintended writes into the main qcow2 header",
631         },
632         {
633             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
634             .type = QEMU_OPT_BOOL,
635             .help = "Check for unintended writes into the active L1 table",
636         },
637         {
638             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
639             .type = QEMU_OPT_BOOL,
640             .help = "Check for unintended writes into an active L2 table",
641         },
642         {
643             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
644             .type = QEMU_OPT_BOOL,
645             .help = "Check for unintended writes into the refcount table",
646         },
647         {
648             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
649             .type = QEMU_OPT_BOOL,
650             .help = "Check for unintended writes into a refcount block",
651         },
652         {
653             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
654             .type = QEMU_OPT_BOOL,
655             .help = "Check for unintended writes into the snapshot table",
656         },
657         {
658             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
659             .type = QEMU_OPT_BOOL,
660             .help = "Check for unintended writes into an inactive L1 table",
661         },
662         {
663             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
664             .type = QEMU_OPT_BOOL,
665             .help = "Check for unintended writes into an inactive L2 table",
666         },
667         {
668             .name = QCOW2_OPT_CACHE_SIZE,
669             .type = QEMU_OPT_SIZE,
670             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
671                     "cache size",
672         },
673         {
674             .name = QCOW2_OPT_L2_CACHE_SIZE,
675             .type = QEMU_OPT_SIZE,
676             .help = "Maximum L2 table cache size",
677         },
678         {
679             .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
680             .type = QEMU_OPT_SIZE,
681             .help = "Size of each entry in the L2 cache",
682         },
683         {
684             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
685             .type = QEMU_OPT_SIZE,
686             .help = "Maximum refcount block cache size",
687         },
688         {
689             .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
690             .type = QEMU_OPT_NUMBER,
691             .help = "Clean unused cache entries after this time (in seconds)",
692         },
693         BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
694             "ID of secret providing qcow2 AES key or LUKS passphrase"),
695         { /* end of list */ }
696     },
697 };
698 
699 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
700     [QCOW2_OL_MAIN_HEADER_BITNR]    = QCOW2_OPT_OVERLAP_MAIN_HEADER,
701     [QCOW2_OL_ACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L1,
702     [QCOW2_OL_ACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L2,
703     [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
704     [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
705     [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
706     [QCOW2_OL_INACTIVE_L1_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L1,
707     [QCOW2_OL_INACTIVE_L2_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L2,
708 };
709 
710 static void cache_clean_timer_cb(void *opaque)
711 {
712     BlockDriverState *bs = opaque;
713     BDRVQcow2State *s = bs->opaque;
714     qcow2_cache_clean_unused(s->l2_table_cache);
715     qcow2_cache_clean_unused(s->refcount_block_cache);
716     timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
717               (int64_t) s->cache_clean_interval * 1000);
718 }
719 
720 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
721 {
722     BDRVQcow2State *s = bs->opaque;
723     if (s->cache_clean_interval > 0) {
724         s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
725                                              SCALE_MS, cache_clean_timer_cb,
726                                              bs);
727         timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
728                   (int64_t) s->cache_clean_interval * 1000);
729     }
730 }
731 
732 static void cache_clean_timer_del(BlockDriverState *bs)
733 {
734     BDRVQcow2State *s = bs->opaque;
735     if (s->cache_clean_timer) {
736         timer_del(s->cache_clean_timer);
737         timer_free(s->cache_clean_timer);
738         s->cache_clean_timer = NULL;
739     }
740 }
741 
742 static void qcow2_detach_aio_context(BlockDriverState *bs)
743 {
744     cache_clean_timer_del(bs);
745 }
746 
747 static void qcow2_attach_aio_context(BlockDriverState *bs,
748                                      AioContext *new_context)
749 {
750     cache_clean_timer_init(bs, new_context);
751 }
752 
753 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
754                              uint64_t *l2_cache_size,
755                              uint64_t *l2_cache_entry_size,
756                              uint64_t *refcount_cache_size, Error **errp)
757 {
758     BDRVQcow2State *s = bs->opaque;
759     uint64_t combined_cache_size;
760     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
761 
762     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
763     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
764     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
765 
766     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
767     *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0);
768     *refcount_cache_size = qemu_opt_get_size(opts,
769                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
770 
771     *l2_cache_entry_size = qemu_opt_get_size(
772         opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size);
773 
774     if (combined_cache_size_set) {
775         if (l2_cache_size_set && refcount_cache_size_set) {
776             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
777                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
778                        "the same time");
779             return;
780         } else if (*l2_cache_size > combined_cache_size) {
781             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
782                        QCOW2_OPT_CACHE_SIZE);
783             return;
784         } else if (*refcount_cache_size > combined_cache_size) {
785             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
786                        QCOW2_OPT_CACHE_SIZE);
787             return;
788         }
789 
790         if (l2_cache_size_set) {
791             *refcount_cache_size = combined_cache_size - *l2_cache_size;
792         } else if (refcount_cache_size_set) {
793             *l2_cache_size = combined_cache_size - *refcount_cache_size;
794         } else {
795             *refcount_cache_size = combined_cache_size
796                                  / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1);
797             *l2_cache_size = combined_cache_size - *refcount_cache_size;
798         }
799     } else {
800         if (!l2_cache_size_set && !refcount_cache_size_set) {
801             *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE,
802                                  (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
803                                  * s->cluster_size);
804             *refcount_cache_size = *l2_cache_size
805                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
806         } else if (!l2_cache_size_set) {
807             *l2_cache_size = *refcount_cache_size
808                            * DEFAULT_L2_REFCOUNT_SIZE_RATIO;
809         } else if (!refcount_cache_size_set) {
810             *refcount_cache_size = *l2_cache_size
811                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
812         }
813     }
814 
815     if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) ||
816         *l2_cache_entry_size > s->cluster_size ||
817         !is_power_of_2(*l2_cache_entry_size)) {
818         error_setg(errp, "L2 cache entry size must be a power of two "
819                    "between %d and the cluster size (%d)",
820                    1 << MIN_CLUSTER_BITS, s->cluster_size);
821         return;
822     }
823 }
824 
825 typedef struct Qcow2ReopenState {
826     Qcow2Cache *l2_table_cache;
827     Qcow2Cache *refcount_block_cache;
828     int l2_slice_size; /* Number of entries in a slice of the L2 table */
829     bool use_lazy_refcounts;
830     int overlap_check;
831     bool discard_passthrough[QCOW2_DISCARD_MAX];
832     uint64_t cache_clean_interval;
833     QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
834 } Qcow2ReopenState;
835 
836 static int qcow2_update_options_prepare(BlockDriverState *bs,
837                                         Qcow2ReopenState *r,
838                                         QDict *options, int flags,
839                                         Error **errp)
840 {
841     BDRVQcow2State *s = bs->opaque;
842     QemuOpts *opts = NULL;
843     const char *opt_overlap_check, *opt_overlap_check_template;
844     int overlap_check_template = 0;
845     uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size;
846     int i;
847     const char *encryptfmt;
848     QDict *encryptopts = NULL;
849     Error *local_err = NULL;
850     int ret;
851 
852     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
853     encryptfmt = qdict_get_try_str(encryptopts, "format");
854 
855     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
856     qemu_opts_absorb_qdict(opts, options, &local_err);
857     if (local_err) {
858         error_propagate(errp, local_err);
859         ret = -EINVAL;
860         goto fail;
861     }
862 
863     /* get L2 table/refcount block cache size from command line options */
864     read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size,
865                      &refcount_cache_size, &local_err);
866     if (local_err) {
867         error_propagate(errp, local_err);
868         ret = -EINVAL;
869         goto fail;
870     }
871 
872     l2_cache_size /= l2_cache_entry_size;
873     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
874         l2_cache_size = MIN_L2_CACHE_SIZE;
875     }
876     if (l2_cache_size > INT_MAX) {
877         error_setg(errp, "L2 cache size too big");
878         ret = -EINVAL;
879         goto fail;
880     }
881 
882     refcount_cache_size /= s->cluster_size;
883     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
884         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
885     }
886     if (refcount_cache_size > INT_MAX) {
887         error_setg(errp, "Refcount cache size too big");
888         ret = -EINVAL;
889         goto fail;
890     }
891 
892     /* alloc new L2 table/refcount block cache, flush old one */
893     if (s->l2_table_cache) {
894         ret = qcow2_cache_flush(bs, s->l2_table_cache);
895         if (ret) {
896             error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
897             goto fail;
898         }
899     }
900 
901     if (s->refcount_block_cache) {
902         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
903         if (ret) {
904             error_setg_errno(errp, -ret,
905                              "Failed to flush the refcount block cache");
906             goto fail;
907         }
908     }
909 
910     r->l2_slice_size = l2_cache_entry_size / sizeof(uint64_t);
911     r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size,
912                                            l2_cache_entry_size);
913     r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size,
914                                                  s->cluster_size);
915     if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
916         error_setg(errp, "Could not allocate metadata caches");
917         ret = -ENOMEM;
918         goto fail;
919     }
920 
921     /* New interval for cache cleanup timer */
922     r->cache_clean_interval =
923         qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
924                             s->cache_clean_interval);
925 #ifndef CONFIG_LINUX
926     if (r->cache_clean_interval != 0) {
927         error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL
928                    " not supported on this host");
929         ret = -EINVAL;
930         goto fail;
931     }
932 #endif
933     if (r->cache_clean_interval > UINT_MAX) {
934         error_setg(errp, "Cache clean interval too big");
935         ret = -EINVAL;
936         goto fail;
937     }
938 
939     /* lazy-refcounts; flush if going from enabled to disabled */
940     r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
941         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
942     if (r->use_lazy_refcounts && s->qcow_version < 3) {
943         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
944                    "qemu 1.1 compatibility level");
945         ret = -EINVAL;
946         goto fail;
947     }
948 
949     if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
950         ret = qcow2_mark_clean(bs);
951         if (ret < 0) {
952             error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
953             goto fail;
954         }
955     }
956 
957     /* Overlap check options */
958     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
959     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
960     if (opt_overlap_check_template && opt_overlap_check &&
961         strcmp(opt_overlap_check_template, opt_overlap_check))
962     {
963         error_setg(errp, "Conflicting values for qcow2 options '"
964                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
965                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
966         ret = -EINVAL;
967         goto fail;
968     }
969     if (!opt_overlap_check) {
970         opt_overlap_check = opt_overlap_check_template ?: "cached";
971     }
972 
973     if (!strcmp(opt_overlap_check, "none")) {
974         overlap_check_template = 0;
975     } else if (!strcmp(opt_overlap_check, "constant")) {
976         overlap_check_template = QCOW2_OL_CONSTANT;
977     } else if (!strcmp(opt_overlap_check, "cached")) {
978         overlap_check_template = QCOW2_OL_CACHED;
979     } else if (!strcmp(opt_overlap_check, "all")) {
980         overlap_check_template = QCOW2_OL_ALL;
981     } else {
982         error_setg(errp, "Unsupported value '%s' for qcow2 option "
983                    "'overlap-check'. Allowed are any of the following: "
984                    "none, constant, cached, all", opt_overlap_check);
985         ret = -EINVAL;
986         goto fail;
987     }
988 
989     r->overlap_check = 0;
990     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
991         /* overlap-check defines a template bitmask, but every flag may be
992          * overwritten through the associated boolean option */
993         r->overlap_check |=
994             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
995                               overlap_check_template & (1 << i)) << i;
996     }
997 
998     r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
999     r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
1000     r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
1001         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
1002                           flags & BDRV_O_UNMAP);
1003     r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
1004         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
1005     r->discard_passthrough[QCOW2_DISCARD_OTHER] =
1006         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
1007 
1008     switch (s->crypt_method_header) {
1009     case QCOW_CRYPT_NONE:
1010         if (encryptfmt) {
1011             error_setg(errp, "No encryption in image header, but options "
1012                        "specified format '%s'", encryptfmt);
1013             ret = -EINVAL;
1014             goto fail;
1015         }
1016         break;
1017 
1018     case QCOW_CRYPT_AES:
1019         if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
1020             error_setg(errp,
1021                        "Header reported 'aes' encryption format but "
1022                        "options specify '%s'", encryptfmt);
1023             ret = -EINVAL;
1024             goto fail;
1025         }
1026         qdict_del(encryptopts, "format");
1027         r->crypto_opts = block_crypto_open_opts_init(
1028             Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp);
1029         break;
1030 
1031     case QCOW_CRYPT_LUKS:
1032         if (encryptfmt && !g_str_equal(encryptfmt, "luks")) {
1033             error_setg(errp,
1034                        "Header reported 'luks' encryption format but "
1035                        "options specify '%s'", encryptfmt);
1036             ret = -EINVAL;
1037             goto fail;
1038         }
1039         qdict_del(encryptopts, "format");
1040         r->crypto_opts = block_crypto_open_opts_init(
1041             Q_CRYPTO_BLOCK_FORMAT_LUKS, encryptopts, errp);
1042         break;
1043 
1044     default:
1045         error_setg(errp, "Unsupported encryption method %d",
1046                    s->crypt_method_header);
1047         break;
1048     }
1049     if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) {
1050         ret = -EINVAL;
1051         goto fail;
1052     }
1053 
1054     ret = 0;
1055 fail:
1056     QDECREF(encryptopts);
1057     qemu_opts_del(opts);
1058     opts = NULL;
1059     return ret;
1060 }
1061 
1062 static void qcow2_update_options_commit(BlockDriverState *bs,
1063                                         Qcow2ReopenState *r)
1064 {
1065     BDRVQcow2State *s = bs->opaque;
1066     int i;
1067 
1068     if (s->l2_table_cache) {
1069         qcow2_cache_destroy(s->l2_table_cache);
1070     }
1071     if (s->refcount_block_cache) {
1072         qcow2_cache_destroy(s->refcount_block_cache);
1073     }
1074     s->l2_table_cache = r->l2_table_cache;
1075     s->refcount_block_cache = r->refcount_block_cache;
1076     s->l2_slice_size = r->l2_slice_size;
1077 
1078     s->overlap_check = r->overlap_check;
1079     s->use_lazy_refcounts = r->use_lazy_refcounts;
1080 
1081     for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
1082         s->discard_passthrough[i] = r->discard_passthrough[i];
1083     }
1084 
1085     if (s->cache_clean_interval != r->cache_clean_interval) {
1086         cache_clean_timer_del(bs);
1087         s->cache_clean_interval = r->cache_clean_interval;
1088         cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
1089     }
1090 
1091     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1092     s->crypto_opts = r->crypto_opts;
1093 }
1094 
1095 static void qcow2_update_options_abort(BlockDriverState *bs,
1096                                        Qcow2ReopenState *r)
1097 {
1098     if (r->l2_table_cache) {
1099         qcow2_cache_destroy(r->l2_table_cache);
1100     }
1101     if (r->refcount_block_cache) {
1102         qcow2_cache_destroy(r->refcount_block_cache);
1103     }
1104     qapi_free_QCryptoBlockOpenOptions(r->crypto_opts);
1105 }
1106 
1107 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
1108                                 int flags, Error **errp)
1109 {
1110     Qcow2ReopenState r = {};
1111     int ret;
1112 
1113     ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
1114     if (ret >= 0) {
1115         qcow2_update_options_commit(bs, &r);
1116     } else {
1117         qcow2_update_options_abort(bs, &r);
1118     }
1119 
1120     return ret;
1121 }
1122 
1123 static int qcow2_do_open(BlockDriverState *bs, QDict *options, int flags,
1124                          Error **errp)
1125 {
1126     BDRVQcow2State *s = bs->opaque;
1127     unsigned int len, i;
1128     int ret = 0;
1129     QCowHeader header;
1130     Error *local_err = NULL;
1131     uint64_t ext_end;
1132     uint64_t l1_vm_state_index;
1133     bool update_header = false;
1134 
1135     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
1136     if (ret < 0) {
1137         error_setg_errno(errp, -ret, "Could not read qcow2 header");
1138         goto fail;
1139     }
1140     be32_to_cpus(&header.magic);
1141     be32_to_cpus(&header.version);
1142     be64_to_cpus(&header.backing_file_offset);
1143     be32_to_cpus(&header.backing_file_size);
1144     be64_to_cpus(&header.size);
1145     be32_to_cpus(&header.cluster_bits);
1146     be32_to_cpus(&header.crypt_method);
1147     be64_to_cpus(&header.l1_table_offset);
1148     be32_to_cpus(&header.l1_size);
1149     be64_to_cpus(&header.refcount_table_offset);
1150     be32_to_cpus(&header.refcount_table_clusters);
1151     be64_to_cpus(&header.snapshots_offset);
1152     be32_to_cpus(&header.nb_snapshots);
1153 
1154     if (header.magic != QCOW_MAGIC) {
1155         error_setg(errp, "Image is not in qcow2 format");
1156         ret = -EINVAL;
1157         goto fail;
1158     }
1159     if (header.version < 2 || header.version > 3) {
1160         error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
1161         ret = -ENOTSUP;
1162         goto fail;
1163     }
1164 
1165     s->qcow_version = header.version;
1166 
1167     /* Initialise cluster size */
1168     if (header.cluster_bits < MIN_CLUSTER_BITS ||
1169         header.cluster_bits > MAX_CLUSTER_BITS) {
1170         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
1171                    header.cluster_bits);
1172         ret = -EINVAL;
1173         goto fail;
1174     }
1175 
1176     s->cluster_bits = header.cluster_bits;
1177     s->cluster_size = 1 << s->cluster_bits;
1178     s->cluster_sectors = 1 << (s->cluster_bits - BDRV_SECTOR_BITS);
1179 
1180     /* Initialise version 3 header fields */
1181     if (header.version == 2) {
1182         header.incompatible_features    = 0;
1183         header.compatible_features      = 0;
1184         header.autoclear_features       = 0;
1185         header.refcount_order           = 4;
1186         header.header_length            = 72;
1187     } else {
1188         be64_to_cpus(&header.incompatible_features);
1189         be64_to_cpus(&header.compatible_features);
1190         be64_to_cpus(&header.autoclear_features);
1191         be32_to_cpus(&header.refcount_order);
1192         be32_to_cpus(&header.header_length);
1193 
1194         if (header.header_length < 104) {
1195             error_setg(errp, "qcow2 header too short");
1196             ret = -EINVAL;
1197             goto fail;
1198         }
1199     }
1200 
1201     if (header.header_length > s->cluster_size) {
1202         error_setg(errp, "qcow2 header exceeds cluster size");
1203         ret = -EINVAL;
1204         goto fail;
1205     }
1206 
1207     if (header.header_length > sizeof(header)) {
1208         s->unknown_header_fields_size = header.header_length - sizeof(header);
1209         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
1210         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
1211                          s->unknown_header_fields_size);
1212         if (ret < 0) {
1213             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
1214                              "fields");
1215             goto fail;
1216         }
1217     }
1218 
1219     if (header.backing_file_offset > s->cluster_size) {
1220         error_setg(errp, "Invalid backing file offset");
1221         ret = -EINVAL;
1222         goto fail;
1223     }
1224 
1225     if (header.backing_file_offset) {
1226         ext_end = header.backing_file_offset;
1227     } else {
1228         ext_end = 1 << header.cluster_bits;
1229     }
1230 
1231     /* Handle feature bits */
1232     s->incompatible_features    = header.incompatible_features;
1233     s->compatible_features      = header.compatible_features;
1234     s->autoclear_features       = header.autoclear_features;
1235 
1236     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
1237         void *feature_table = NULL;
1238         qcow2_read_extensions(bs, header.header_length, ext_end,
1239                               &feature_table, flags, NULL, NULL);
1240         report_unsupported_feature(errp, feature_table,
1241                                    s->incompatible_features &
1242                                    ~QCOW2_INCOMPAT_MASK);
1243         ret = -ENOTSUP;
1244         g_free(feature_table);
1245         goto fail;
1246     }
1247 
1248     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
1249         /* Corrupt images may not be written to unless they are being repaired
1250          */
1251         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
1252             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
1253                        "read/write");
1254             ret = -EACCES;
1255             goto fail;
1256         }
1257     }
1258 
1259     /* Check support for various header values */
1260     if (header.refcount_order > 6) {
1261         error_setg(errp, "Reference count entry width too large; may not "
1262                    "exceed 64 bits");
1263         ret = -EINVAL;
1264         goto fail;
1265     }
1266     s->refcount_order = header.refcount_order;
1267     s->refcount_bits = 1 << s->refcount_order;
1268     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1269     s->refcount_max += s->refcount_max - 1;
1270 
1271     s->crypt_method_header = header.crypt_method;
1272     if (s->crypt_method_header) {
1273         if (bdrv_uses_whitelist() &&
1274             s->crypt_method_header == QCOW_CRYPT_AES) {
1275             error_setg(errp,
1276                        "Use of AES-CBC encrypted qcow2 images is no longer "
1277                        "supported in system emulators");
1278             error_append_hint(errp,
1279                               "You can use 'qemu-img convert' to convert your "
1280                               "image to an alternative supported format, such "
1281                               "as unencrypted qcow2, or raw with the LUKS "
1282                               "format instead.\n");
1283             ret = -ENOSYS;
1284             goto fail;
1285         }
1286 
1287         if (s->crypt_method_header == QCOW_CRYPT_AES) {
1288             s->crypt_physical_offset = false;
1289         } else {
1290             /* Assuming LUKS and any future crypt methods we
1291              * add will all use physical offsets, due to the
1292              * fact that the alternative is insecure...  */
1293             s->crypt_physical_offset = true;
1294         }
1295 
1296         bs->encrypted = true;
1297     }
1298 
1299     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
1300     s->l2_size = 1 << s->l2_bits;
1301     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1302     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1303     s->refcount_block_size = 1 << s->refcount_block_bits;
1304     bs->total_sectors = header.size / 512;
1305     s->csize_shift = (62 - (s->cluster_bits - 8));
1306     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1307     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1308 
1309     s->refcount_table_offset = header.refcount_table_offset;
1310     s->refcount_table_size =
1311         header.refcount_table_clusters << (s->cluster_bits - 3);
1312 
1313     if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
1314         error_setg(errp, "Reference count table too large");
1315         ret = -EINVAL;
1316         goto fail;
1317     }
1318 
1319     if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) {
1320         error_setg(errp, "Image does not contain a reference count table");
1321         ret = -EINVAL;
1322         goto fail;
1323     }
1324 
1325     ret = validate_table_offset(bs, s->refcount_table_offset,
1326                                 s->refcount_table_size, sizeof(uint64_t));
1327     if (ret < 0) {
1328         error_setg(errp, "Invalid reference count table offset");
1329         goto fail;
1330     }
1331 
1332     /* Snapshot table offset/length */
1333     if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
1334         error_setg(errp, "Too many snapshots");
1335         ret = -EINVAL;
1336         goto fail;
1337     }
1338 
1339     ret = validate_table_offset(bs, header.snapshots_offset,
1340                                 header.nb_snapshots,
1341                                 sizeof(QCowSnapshotHeader));
1342     if (ret < 0) {
1343         error_setg(errp, "Invalid snapshot table offset");
1344         goto fail;
1345     }
1346 
1347     /* read the level 1 table */
1348     if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
1349         error_setg(errp, "Active L1 table too large");
1350         ret = -EFBIG;
1351         goto fail;
1352     }
1353     s->l1_size = header.l1_size;
1354 
1355     l1_vm_state_index = size_to_l1(s, header.size);
1356     if (l1_vm_state_index > INT_MAX) {
1357         error_setg(errp, "Image is too big");
1358         ret = -EFBIG;
1359         goto fail;
1360     }
1361     s->l1_vm_state_index = l1_vm_state_index;
1362 
1363     /* the L1 table must contain at least enough entries to put
1364        header.size bytes */
1365     if (s->l1_size < s->l1_vm_state_index) {
1366         error_setg(errp, "L1 table is too small");
1367         ret = -EINVAL;
1368         goto fail;
1369     }
1370 
1371     ret = validate_table_offset(bs, header.l1_table_offset,
1372                                 header.l1_size, sizeof(uint64_t));
1373     if (ret < 0) {
1374         error_setg(errp, "Invalid L1 table offset");
1375         goto fail;
1376     }
1377     s->l1_table_offset = header.l1_table_offset;
1378 
1379 
1380     if (s->l1_size > 0) {
1381         s->l1_table = qemu_try_blockalign(bs->file->bs,
1382             align_offset(s->l1_size * sizeof(uint64_t), 512));
1383         if (s->l1_table == NULL) {
1384             error_setg(errp, "Could not allocate L1 table");
1385             ret = -ENOMEM;
1386             goto fail;
1387         }
1388         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
1389                          s->l1_size * sizeof(uint64_t));
1390         if (ret < 0) {
1391             error_setg_errno(errp, -ret, "Could not read L1 table");
1392             goto fail;
1393         }
1394         for(i = 0;i < s->l1_size; i++) {
1395             be64_to_cpus(&s->l1_table[i]);
1396         }
1397     }
1398 
1399     /* Parse driver-specific options */
1400     ret = qcow2_update_options(bs, options, flags, errp);
1401     if (ret < 0) {
1402         goto fail;
1403     }
1404 
1405     s->cluster_cache_offset = -1;
1406     s->flags = flags;
1407 
1408     ret = qcow2_refcount_init(bs);
1409     if (ret != 0) {
1410         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1411         goto fail;
1412     }
1413 
1414     QLIST_INIT(&s->cluster_allocs);
1415     QTAILQ_INIT(&s->discards);
1416 
1417     /* read qcow2 extensions */
1418     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1419                               flags, &update_header, &local_err)) {
1420         error_propagate(errp, local_err);
1421         ret = -EINVAL;
1422         goto fail;
1423     }
1424 
1425     /* qcow2_read_extension may have set up the crypto context
1426      * if the crypt method needs a header region, some methods
1427      * don't need header extensions, so must check here
1428      */
1429     if (s->crypt_method_header && !s->crypto) {
1430         if (s->crypt_method_header == QCOW_CRYPT_AES) {
1431             unsigned int cflags = 0;
1432             if (flags & BDRV_O_NO_IO) {
1433                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1434             }
1435             s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
1436                                            NULL, NULL, cflags, errp);
1437             if (!s->crypto) {
1438                 ret = -EINVAL;
1439                 goto fail;
1440             }
1441         } else if (!(flags & BDRV_O_NO_IO)) {
1442             error_setg(errp, "Missing CRYPTO header for crypt method %d",
1443                        s->crypt_method_header);
1444             ret = -EINVAL;
1445             goto fail;
1446         }
1447     }
1448 
1449     /* read the backing file name */
1450     if (header.backing_file_offset != 0) {
1451         len = header.backing_file_size;
1452         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1453             len >= sizeof(bs->backing_file)) {
1454             error_setg(errp, "Backing file name too long");
1455             ret = -EINVAL;
1456             goto fail;
1457         }
1458         ret = bdrv_pread(bs->file, header.backing_file_offset,
1459                          bs->backing_file, len);
1460         if (ret < 0) {
1461             error_setg_errno(errp, -ret, "Could not read backing file name");
1462             goto fail;
1463         }
1464         bs->backing_file[len] = '\0';
1465         s->image_backing_file = g_strdup(bs->backing_file);
1466     }
1467 
1468     /* Internal snapshots */
1469     s->snapshots_offset = header.snapshots_offset;
1470     s->nb_snapshots = header.nb_snapshots;
1471 
1472     ret = qcow2_read_snapshots(bs);
1473     if (ret < 0) {
1474         error_setg_errno(errp, -ret, "Could not read snapshots");
1475         goto fail;
1476     }
1477 
1478     /* Clear unknown autoclear feature bits */
1479     update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
1480     update_header =
1481         update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE);
1482     if (update_header) {
1483         s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
1484     }
1485 
1486     if (qcow2_load_dirty_bitmaps(bs, &local_err)) {
1487         update_header = false;
1488     }
1489     if (local_err != NULL) {
1490         error_propagate(errp, local_err);
1491         ret = -EINVAL;
1492         goto fail;
1493     }
1494 
1495     if (update_header) {
1496         ret = qcow2_update_header(bs);
1497         if (ret < 0) {
1498             error_setg_errno(errp, -ret, "Could not update qcow2 header");
1499             goto fail;
1500         }
1501     }
1502 
1503     /* Initialise locks */
1504     qemu_co_mutex_init(&s->lock);
1505     bs->supported_zero_flags = header.version >= 3 ? BDRV_REQ_MAY_UNMAP : 0;
1506 
1507     /* Repair image if dirty */
1508     if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1509         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1510         BdrvCheckResult result = {0};
1511 
1512         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1513         if (ret < 0 || result.check_errors) {
1514             if (ret >= 0) {
1515                 ret = -EIO;
1516             }
1517             error_setg_errno(errp, -ret, "Could not repair dirty image");
1518             goto fail;
1519         }
1520     }
1521 
1522 #ifdef DEBUG_ALLOC
1523     {
1524         BdrvCheckResult result = {0};
1525         qcow2_check_refcounts(bs, &result, 0);
1526     }
1527 #endif
1528     return ret;
1529 
1530  fail:
1531     g_free(s->unknown_header_fields);
1532     cleanup_unknown_header_ext(bs);
1533     qcow2_free_snapshots(bs);
1534     qcow2_refcount_close(bs);
1535     qemu_vfree(s->l1_table);
1536     /* else pre-write overlap checks in cache_destroy may crash */
1537     s->l1_table = NULL;
1538     cache_clean_timer_del(bs);
1539     if (s->l2_table_cache) {
1540         qcow2_cache_destroy(s->l2_table_cache);
1541     }
1542     if (s->refcount_block_cache) {
1543         qcow2_cache_destroy(s->refcount_block_cache);
1544     }
1545     qcrypto_block_free(s->crypto);
1546     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1547     return ret;
1548 }
1549 
1550 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1551                       Error **errp)
1552 {
1553     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
1554                                false, errp);
1555     if (!bs->file) {
1556         return -EINVAL;
1557     }
1558 
1559     return qcow2_do_open(bs, options, flags, errp);
1560 }
1561 
1562 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1563 {
1564     BDRVQcow2State *s = bs->opaque;
1565 
1566     if (bs->encrypted) {
1567         /* Encryption works on a sector granularity */
1568         bs->bl.request_alignment = BDRV_SECTOR_SIZE;
1569     }
1570     bs->bl.pwrite_zeroes_alignment = s->cluster_size;
1571     bs->bl.pdiscard_alignment = s->cluster_size;
1572 }
1573 
1574 static int qcow2_reopen_prepare(BDRVReopenState *state,
1575                                 BlockReopenQueue *queue, Error **errp)
1576 {
1577     Qcow2ReopenState *r;
1578     int ret;
1579 
1580     r = g_new0(Qcow2ReopenState, 1);
1581     state->opaque = r;
1582 
1583     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1584                                        state->flags, errp);
1585     if (ret < 0) {
1586         goto fail;
1587     }
1588 
1589     /* We need to write out any unwritten data if we reopen read-only. */
1590     if ((state->flags & BDRV_O_RDWR) == 0) {
1591         ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1592         if (ret < 0) {
1593             goto fail;
1594         }
1595 
1596         ret = bdrv_flush(state->bs);
1597         if (ret < 0) {
1598             goto fail;
1599         }
1600 
1601         ret = qcow2_mark_clean(state->bs);
1602         if (ret < 0) {
1603             goto fail;
1604         }
1605     }
1606 
1607     return 0;
1608 
1609 fail:
1610     qcow2_update_options_abort(state->bs, r);
1611     g_free(r);
1612     return ret;
1613 }
1614 
1615 static void qcow2_reopen_commit(BDRVReopenState *state)
1616 {
1617     qcow2_update_options_commit(state->bs, state->opaque);
1618     g_free(state->opaque);
1619 }
1620 
1621 static void qcow2_reopen_abort(BDRVReopenState *state)
1622 {
1623     qcow2_update_options_abort(state->bs, state->opaque);
1624     g_free(state->opaque);
1625 }
1626 
1627 static void qcow2_join_options(QDict *options, QDict *old_options)
1628 {
1629     bool has_new_overlap_template =
1630         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1631         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1632     bool has_new_total_cache_size =
1633         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1634     bool has_all_cache_options;
1635 
1636     /* New overlap template overrides all old overlap options */
1637     if (has_new_overlap_template) {
1638         qdict_del(old_options, QCOW2_OPT_OVERLAP);
1639         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1640         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1641         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1642         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1643         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1644         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1645         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1646         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1647         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1648     }
1649 
1650     /* New total cache size overrides all old options */
1651     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1652         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1653         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1654     }
1655 
1656     qdict_join(options, old_options, false);
1657 
1658     /*
1659      * If after merging all cache size options are set, an old total size is
1660      * overwritten. Do keep all options, however, if all three are new. The
1661      * resulting error message is what we want to happen.
1662      */
1663     has_all_cache_options =
1664         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1665         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1666         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1667 
1668     if (has_all_cache_options && !has_new_total_cache_size) {
1669         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1670     }
1671 }
1672 
1673 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1674         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1675 {
1676     BDRVQcow2State *s = bs->opaque;
1677     uint64_t cluster_offset;
1678     int index_in_cluster, ret;
1679     unsigned int bytes;
1680     int64_t status = 0;
1681 
1682     bytes = MIN(INT_MAX, nb_sectors * BDRV_SECTOR_SIZE);
1683     qemu_co_mutex_lock(&s->lock);
1684     ret = qcow2_get_cluster_offset(bs, sector_num << BDRV_SECTOR_BITS, &bytes,
1685                                    &cluster_offset);
1686     qemu_co_mutex_unlock(&s->lock);
1687     if (ret < 0) {
1688         return ret;
1689     }
1690 
1691     *pnum = bytes >> BDRV_SECTOR_BITS;
1692 
1693     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1694         !s->crypto) {
1695         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1696         cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1697         *file = bs->file->bs;
1698         status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1699     }
1700     if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) {
1701         status |= BDRV_BLOCK_ZERO;
1702     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1703         status |= BDRV_BLOCK_DATA;
1704     }
1705     return status;
1706 }
1707 
1708 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset,
1709                                         uint64_t bytes, QEMUIOVector *qiov,
1710                                         int flags)
1711 {
1712     BDRVQcow2State *s = bs->opaque;
1713     int offset_in_cluster;
1714     int ret;
1715     unsigned int cur_bytes; /* number of bytes in current iteration */
1716     uint64_t cluster_offset = 0;
1717     uint64_t bytes_done = 0;
1718     QEMUIOVector hd_qiov;
1719     uint8_t *cluster_data = NULL;
1720 
1721     qemu_iovec_init(&hd_qiov, qiov->niov);
1722 
1723     qemu_co_mutex_lock(&s->lock);
1724 
1725     while (bytes != 0) {
1726 
1727         /* prepare next request */
1728         cur_bytes = MIN(bytes, INT_MAX);
1729         if (s->crypto) {
1730             cur_bytes = MIN(cur_bytes,
1731                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1732         }
1733 
1734         ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset);
1735         if (ret < 0) {
1736             goto fail;
1737         }
1738 
1739         offset_in_cluster = offset_into_cluster(s, offset);
1740 
1741         qemu_iovec_reset(&hd_qiov);
1742         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
1743 
1744         switch (ret) {
1745         case QCOW2_CLUSTER_UNALLOCATED:
1746 
1747             if (bs->backing) {
1748                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1749                 qemu_co_mutex_unlock(&s->lock);
1750                 ret = bdrv_co_preadv(bs->backing, offset, cur_bytes,
1751                                      &hd_qiov, 0);
1752                 qemu_co_mutex_lock(&s->lock);
1753                 if (ret < 0) {
1754                     goto fail;
1755                 }
1756             } else {
1757                 /* Note: in this case, no need to wait */
1758                 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1759             }
1760             break;
1761 
1762         case QCOW2_CLUSTER_ZERO_PLAIN:
1763         case QCOW2_CLUSTER_ZERO_ALLOC:
1764             qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1765             break;
1766 
1767         case QCOW2_CLUSTER_COMPRESSED:
1768             /* add AIO support for compressed blocks ? */
1769             ret = qcow2_decompress_cluster(bs, cluster_offset);
1770             if (ret < 0) {
1771                 goto fail;
1772             }
1773 
1774             qemu_iovec_from_buf(&hd_qiov, 0,
1775                                 s->cluster_cache + offset_in_cluster,
1776                                 cur_bytes);
1777             break;
1778 
1779         case QCOW2_CLUSTER_NORMAL:
1780             if ((cluster_offset & 511) != 0) {
1781                 ret = -EIO;
1782                 goto fail;
1783             }
1784 
1785             if (bs->encrypted) {
1786                 assert(s->crypto);
1787 
1788                 /*
1789                  * For encrypted images, read everything into a temporary
1790                  * contiguous buffer on which the AES functions can work.
1791                  */
1792                 if (!cluster_data) {
1793                     cluster_data =
1794                         qemu_try_blockalign(bs->file->bs,
1795                                             QCOW_MAX_CRYPT_CLUSTERS
1796                                             * s->cluster_size);
1797                     if (cluster_data == NULL) {
1798                         ret = -ENOMEM;
1799                         goto fail;
1800                     }
1801                 }
1802 
1803                 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1804                 qemu_iovec_reset(&hd_qiov);
1805                 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
1806             }
1807 
1808             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1809             qemu_co_mutex_unlock(&s->lock);
1810             ret = bdrv_co_preadv(bs->file,
1811                                  cluster_offset + offset_in_cluster,
1812                                  cur_bytes, &hd_qiov, 0);
1813             qemu_co_mutex_lock(&s->lock);
1814             if (ret < 0) {
1815                 goto fail;
1816             }
1817             if (bs->encrypted) {
1818                 assert(s->crypto);
1819                 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
1820                 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
1821                 if (qcrypto_block_decrypt(s->crypto,
1822                                           (s->crypt_physical_offset ?
1823                                            cluster_offset + offset_in_cluster :
1824                                            offset),
1825                                           cluster_data,
1826                                           cur_bytes,
1827                                           NULL) < 0) {
1828                     ret = -EIO;
1829                     goto fail;
1830                 }
1831                 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes);
1832             }
1833             break;
1834 
1835         default:
1836             g_assert_not_reached();
1837             ret = -EIO;
1838             goto fail;
1839         }
1840 
1841         bytes -= cur_bytes;
1842         offset += cur_bytes;
1843         bytes_done += cur_bytes;
1844     }
1845     ret = 0;
1846 
1847 fail:
1848     qemu_co_mutex_unlock(&s->lock);
1849 
1850     qemu_iovec_destroy(&hd_qiov);
1851     qemu_vfree(cluster_data);
1852 
1853     return ret;
1854 }
1855 
1856 /* Check if it's possible to merge a write request with the writing of
1857  * the data from the COW regions */
1858 static bool merge_cow(uint64_t offset, unsigned bytes,
1859                       QEMUIOVector *hd_qiov, QCowL2Meta *l2meta)
1860 {
1861     QCowL2Meta *m;
1862 
1863     for (m = l2meta; m != NULL; m = m->next) {
1864         /* If both COW regions are empty then there's nothing to merge */
1865         if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
1866             continue;
1867         }
1868 
1869         /* The data (middle) region must be immediately after the
1870          * start region */
1871         if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
1872             continue;
1873         }
1874 
1875         /* The end region must be immediately after the data (middle)
1876          * region */
1877         if (m->offset + m->cow_end.offset != offset + bytes) {
1878             continue;
1879         }
1880 
1881         /* Make sure that adding both COW regions to the QEMUIOVector
1882          * does not exceed IOV_MAX */
1883         if (hd_qiov->niov > IOV_MAX - 2) {
1884             continue;
1885         }
1886 
1887         m->data_qiov = hd_qiov;
1888         return true;
1889     }
1890 
1891     return false;
1892 }
1893 
1894 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset,
1895                                          uint64_t bytes, QEMUIOVector *qiov,
1896                                          int flags)
1897 {
1898     BDRVQcow2State *s = bs->opaque;
1899     int offset_in_cluster;
1900     int ret;
1901     unsigned int cur_bytes; /* number of sectors in current iteration */
1902     uint64_t cluster_offset;
1903     QEMUIOVector hd_qiov;
1904     uint64_t bytes_done = 0;
1905     uint8_t *cluster_data = NULL;
1906     QCowL2Meta *l2meta = NULL;
1907 
1908     trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
1909 
1910     qemu_iovec_init(&hd_qiov, qiov->niov);
1911 
1912     s->cluster_cache_offset = -1; /* disable compressed cache */
1913 
1914     qemu_co_mutex_lock(&s->lock);
1915 
1916     while (bytes != 0) {
1917 
1918         l2meta = NULL;
1919 
1920         trace_qcow2_writev_start_part(qemu_coroutine_self());
1921         offset_in_cluster = offset_into_cluster(s, offset);
1922         cur_bytes = MIN(bytes, INT_MAX);
1923         if (bs->encrypted) {
1924             cur_bytes = MIN(cur_bytes,
1925                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
1926                             - offset_in_cluster);
1927         }
1928 
1929         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
1930                                          &cluster_offset, &l2meta);
1931         if (ret < 0) {
1932             goto fail;
1933         }
1934 
1935         assert((cluster_offset & 511) == 0);
1936 
1937         qemu_iovec_reset(&hd_qiov);
1938         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
1939 
1940         if (bs->encrypted) {
1941             assert(s->crypto);
1942             if (!cluster_data) {
1943                 cluster_data = qemu_try_blockalign(bs->file->bs,
1944                                                    QCOW_MAX_CRYPT_CLUSTERS
1945                                                    * s->cluster_size);
1946                 if (cluster_data == NULL) {
1947                     ret = -ENOMEM;
1948                     goto fail;
1949                 }
1950             }
1951 
1952             assert(hd_qiov.size <=
1953                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1954             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1955 
1956             if (qcrypto_block_encrypt(s->crypto,
1957                                       (s->crypt_physical_offset ?
1958                                        cluster_offset + offset_in_cluster :
1959                                        offset),
1960                                       cluster_data,
1961                                       cur_bytes, NULL) < 0) {
1962                 ret = -EIO;
1963                 goto fail;
1964             }
1965 
1966             qemu_iovec_reset(&hd_qiov);
1967             qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
1968         }
1969 
1970         ret = qcow2_pre_write_overlap_check(bs, 0,
1971                 cluster_offset + offset_in_cluster, cur_bytes);
1972         if (ret < 0) {
1973             goto fail;
1974         }
1975 
1976         /* If we need to do COW, check if it's possible to merge the
1977          * writing of the guest data together with that of the COW regions.
1978          * If it's not possible (or not necessary) then write the
1979          * guest data now. */
1980         if (!merge_cow(offset, cur_bytes, &hd_qiov, l2meta)) {
1981             qemu_co_mutex_unlock(&s->lock);
1982             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1983             trace_qcow2_writev_data(qemu_coroutine_self(),
1984                                     cluster_offset + offset_in_cluster);
1985             ret = bdrv_co_pwritev(bs->file,
1986                                   cluster_offset + offset_in_cluster,
1987                                   cur_bytes, &hd_qiov, 0);
1988             qemu_co_mutex_lock(&s->lock);
1989             if (ret < 0) {
1990                 goto fail;
1991             }
1992         }
1993 
1994         while (l2meta != NULL) {
1995             QCowL2Meta *next;
1996 
1997             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1998             if (ret < 0) {
1999                 goto fail;
2000             }
2001 
2002             /* Take the request off the list of running requests */
2003             if (l2meta->nb_clusters != 0) {
2004                 QLIST_REMOVE(l2meta, next_in_flight);
2005             }
2006 
2007             qemu_co_queue_restart_all(&l2meta->dependent_requests);
2008 
2009             next = l2meta->next;
2010             g_free(l2meta);
2011             l2meta = next;
2012         }
2013 
2014         bytes -= cur_bytes;
2015         offset += cur_bytes;
2016         bytes_done += cur_bytes;
2017         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2018     }
2019     ret = 0;
2020 
2021 fail:
2022     while (l2meta != NULL) {
2023         QCowL2Meta *next;
2024 
2025         if (l2meta->nb_clusters != 0) {
2026             QLIST_REMOVE(l2meta, next_in_flight);
2027         }
2028         qemu_co_queue_restart_all(&l2meta->dependent_requests);
2029 
2030         next = l2meta->next;
2031         g_free(l2meta);
2032         l2meta = next;
2033     }
2034 
2035     qemu_co_mutex_unlock(&s->lock);
2036 
2037     qemu_iovec_destroy(&hd_qiov);
2038     qemu_vfree(cluster_data);
2039     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2040 
2041     return ret;
2042 }
2043 
2044 static int qcow2_inactivate(BlockDriverState *bs)
2045 {
2046     BDRVQcow2State *s = bs->opaque;
2047     int ret, result = 0;
2048     Error *local_err = NULL;
2049 
2050     qcow2_store_persistent_dirty_bitmaps(bs, &local_err);
2051     if (local_err != NULL) {
2052         result = -EINVAL;
2053         error_report_err(local_err);
2054         error_report("Persistent bitmaps are lost for node '%s'",
2055                      bdrv_get_device_or_node_name(bs));
2056     }
2057 
2058     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2059     if (ret) {
2060         result = ret;
2061         error_report("Failed to flush the L2 table cache: %s",
2062                      strerror(-ret));
2063     }
2064 
2065     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2066     if (ret) {
2067         result = ret;
2068         error_report("Failed to flush the refcount block cache: %s",
2069                      strerror(-ret));
2070     }
2071 
2072     if (result == 0) {
2073         qcow2_mark_clean(bs);
2074     }
2075 
2076     return result;
2077 }
2078 
2079 static void qcow2_close(BlockDriverState *bs)
2080 {
2081     BDRVQcow2State *s = bs->opaque;
2082     qemu_vfree(s->l1_table);
2083     /* else pre-write overlap checks in cache_destroy may crash */
2084     s->l1_table = NULL;
2085 
2086     if (!(s->flags & BDRV_O_INACTIVE)) {
2087         qcow2_inactivate(bs);
2088     }
2089 
2090     cache_clean_timer_del(bs);
2091     qcow2_cache_destroy(s->l2_table_cache);
2092     qcow2_cache_destroy(s->refcount_block_cache);
2093 
2094     qcrypto_block_free(s->crypto);
2095     s->crypto = NULL;
2096 
2097     g_free(s->unknown_header_fields);
2098     cleanup_unknown_header_ext(bs);
2099 
2100     g_free(s->image_backing_file);
2101     g_free(s->image_backing_format);
2102 
2103     g_free(s->cluster_cache);
2104     qemu_vfree(s->cluster_data);
2105     qcow2_refcount_close(bs);
2106     qcow2_free_snapshots(bs);
2107 }
2108 
2109 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
2110 {
2111     BDRVQcow2State *s = bs->opaque;
2112     int flags = s->flags;
2113     QCryptoBlock *crypto = NULL;
2114     QDict *options;
2115     Error *local_err = NULL;
2116     int ret;
2117 
2118     /*
2119      * Backing files are read-only which makes all of their metadata immutable,
2120      * that means we don't have to worry about reopening them here.
2121      */
2122 
2123     crypto = s->crypto;
2124     s->crypto = NULL;
2125 
2126     qcow2_close(bs);
2127 
2128     memset(s, 0, sizeof(BDRVQcow2State));
2129     options = qdict_clone_shallow(bs->options);
2130 
2131     flags &= ~BDRV_O_INACTIVE;
2132     ret = qcow2_do_open(bs, options, flags, &local_err);
2133     QDECREF(options);
2134     if (local_err) {
2135         error_propagate(errp, local_err);
2136         error_prepend(errp, "Could not reopen qcow2 layer: ");
2137         bs->drv = NULL;
2138         return;
2139     } else if (ret < 0) {
2140         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
2141         bs->drv = NULL;
2142         return;
2143     }
2144 
2145     s->crypto = crypto;
2146 }
2147 
2148 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2149     size_t len, size_t buflen)
2150 {
2151     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2152     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2153 
2154     if (buflen < ext_len) {
2155         return -ENOSPC;
2156     }
2157 
2158     *ext_backing_fmt = (QCowExtension) {
2159         .magic  = cpu_to_be32(magic),
2160         .len    = cpu_to_be32(len),
2161     };
2162 
2163     if (len) {
2164         memcpy(buf + sizeof(QCowExtension), s, len);
2165     }
2166 
2167     return ext_len;
2168 }
2169 
2170 /*
2171  * Updates the qcow2 header, including the variable length parts of it, i.e.
2172  * the backing file name and all extensions. qcow2 was not designed to allow
2173  * such changes, so if we run out of space (we can only use the first cluster)
2174  * this function may fail.
2175  *
2176  * Returns 0 on success, -errno in error cases.
2177  */
2178 int qcow2_update_header(BlockDriverState *bs)
2179 {
2180     BDRVQcow2State *s = bs->opaque;
2181     QCowHeader *header;
2182     char *buf;
2183     size_t buflen = s->cluster_size;
2184     int ret;
2185     uint64_t total_size;
2186     uint32_t refcount_table_clusters;
2187     size_t header_length;
2188     Qcow2UnknownHeaderExtension *uext;
2189 
2190     buf = qemu_blockalign(bs, buflen);
2191 
2192     /* Header structure */
2193     header = (QCowHeader*) buf;
2194 
2195     if (buflen < sizeof(*header)) {
2196         ret = -ENOSPC;
2197         goto fail;
2198     }
2199 
2200     header_length = sizeof(*header) + s->unknown_header_fields_size;
2201     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2202     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2203 
2204     *header = (QCowHeader) {
2205         /* Version 2 fields */
2206         .magic                  = cpu_to_be32(QCOW_MAGIC),
2207         .version                = cpu_to_be32(s->qcow_version),
2208         .backing_file_offset    = 0,
2209         .backing_file_size      = 0,
2210         .cluster_bits           = cpu_to_be32(s->cluster_bits),
2211         .size                   = cpu_to_be64(total_size),
2212         .crypt_method           = cpu_to_be32(s->crypt_method_header),
2213         .l1_size                = cpu_to_be32(s->l1_size),
2214         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
2215         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
2216         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2217         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
2218         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
2219 
2220         /* Version 3 fields */
2221         .incompatible_features  = cpu_to_be64(s->incompatible_features),
2222         .compatible_features    = cpu_to_be64(s->compatible_features),
2223         .autoclear_features     = cpu_to_be64(s->autoclear_features),
2224         .refcount_order         = cpu_to_be32(s->refcount_order),
2225         .header_length          = cpu_to_be32(header_length),
2226     };
2227 
2228     /* For older versions, write a shorter header */
2229     switch (s->qcow_version) {
2230     case 2:
2231         ret = offsetof(QCowHeader, incompatible_features);
2232         break;
2233     case 3:
2234         ret = sizeof(*header);
2235         break;
2236     default:
2237         ret = -EINVAL;
2238         goto fail;
2239     }
2240 
2241     buf += ret;
2242     buflen -= ret;
2243     memset(buf, 0, buflen);
2244 
2245     /* Preserve any unknown field in the header */
2246     if (s->unknown_header_fields_size) {
2247         if (buflen < s->unknown_header_fields_size) {
2248             ret = -ENOSPC;
2249             goto fail;
2250         }
2251 
2252         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2253         buf += s->unknown_header_fields_size;
2254         buflen -= s->unknown_header_fields_size;
2255     }
2256 
2257     /* Backing file format header extension */
2258     if (s->image_backing_format) {
2259         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
2260                              s->image_backing_format,
2261                              strlen(s->image_backing_format),
2262                              buflen);
2263         if (ret < 0) {
2264             goto fail;
2265         }
2266 
2267         buf += ret;
2268         buflen -= ret;
2269     }
2270 
2271     /* Full disk encryption header pointer extension */
2272     if (s->crypto_header.offset != 0) {
2273         cpu_to_be64s(&s->crypto_header.offset);
2274         cpu_to_be64s(&s->crypto_header.length);
2275         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
2276                              &s->crypto_header, sizeof(s->crypto_header),
2277                              buflen);
2278         be64_to_cpus(&s->crypto_header.offset);
2279         be64_to_cpus(&s->crypto_header.length);
2280         if (ret < 0) {
2281             goto fail;
2282         }
2283         buf += ret;
2284         buflen -= ret;
2285     }
2286 
2287     /* Feature table */
2288     if (s->qcow_version >= 3) {
2289         Qcow2Feature features[] = {
2290             {
2291                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2292                 .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
2293                 .name = "dirty bit",
2294             },
2295             {
2296                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2297                 .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
2298                 .name = "corrupt bit",
2299             },
2300             {
2301                 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
2302                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
2303                 .name = "lazy refcounts",
2304             },
2305         };
2306 
2307         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
2308                              features, sizeof(features), buflen);
2309         if (ret < 0) {
2310             goto fail;
2311         }
2312         buf += ret;
2313         buflen -= ret;
2314     }
2315 
2316     /* Bitmap extension */
2317     if (s->nb_bitmaps > 0) {
2318         Qcow2BitmapHeaderExt bitmaps_header = {
2319             .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
2320             .bitmap_directory_size =
2321                     cpu_to_be64(s->bitmap_directory_size),
2322             .bitmap_directory_offset =
2323                     cpu_to_be64(s->bitmap_directory_offset)
2324         };
2325         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
2326                              &bitmaps_header, sizeof(bitmaps_header),
2327                              buflen);
2328         if (ret < 0) {
2329             goto fail;
2330         }
2331         buf += ret;
2332         buflen -= ret;
2333     }
2334 
2335     /* Keep unknown header extensions */
2336     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
2337         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
2338         if (ret < 0) {
2339             goto fail;
2340         }
2341 
2342         buf += ret;
2343         buflen -= ret;
2344     }
2345 
2346     /* End of header extensions */
2347     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
2348     if (ret < 0) {
2349         goto fail;
2350     }
2351 
2352     buf += ret;
2353     buflen -= ret;
2354 
2355     /* Backing file name */
2356     if (s->image_backing_file) {
2357         size_t backing_file_len = strlen(s->image_backing_file);
2358 
2359         if (buflen < backing_file_len) {
2360             ret = -ENOSPC;
2361             goto fail;
2362         }
2363 
2364         /* Using strncpy is ok here, since buf is not NUL-terminated. */
2365         strncpy(buf, s->image_backing_file, buflen);
2366 
2367         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
2368         header->backing_file_size   = cpu_to_be32(backing_file_len);
2369     }
2370 
2371     /* Write the new header */
2372     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
2373     if (ret < 0) {
2374         goto fail;
2375     }
2376 
2377     ret = 0;
2378 fail:
2379     qemu_vfree(header);
2380     return ret;
2381 }
2382 
2383 static int qcow2_change_backing_file(BlockDriverState *bs,
2384     const char *backing_file, const char *backing_fmt)
2385 {
2386     BDRVQcow2State *s = bs->opaque;
2387 
2388     if (backing_file && strlen(backing_file) > 1023) {
2389         return -EINVAL;
2390     }
2391 
2392     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2393     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2394 
2395     g_free(s->image_backing_file);
2396     g_free(s->image_backing_format);
2397 
2398     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
2399     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
2400 
2401     return qcow2_update_header(bs);
2402 }
2403 
2404 static int qcow2_crypt_method_from_format(const char *encryptfmt)
2405 {
2406     if (g_str_equal(encryptfmt, "luks")) {
2407         return QCOW_CRYPT_LUKS;
2408     } else if (g_str_equal(encryptfmt, "aes")) {
2409         return QCOW_CRYPT_AES;
2410     } else {
2411         return -EINVAL;
2412     }
2413 }
2414 
2415 static int qcow2_set_up_encryption(BlockDriverState *bs, const char *encryptfmt,
2416                                    QemuOpts *opts, Error **errp)
2417 {
2418     BDRVQcow2State *s = bs->opaque;
2419     QCryptoBlockCreateOptions *cryptoopts = NULL;
2420     QCryptoBlock *crypto = NULL;
2421     int ret = -EINVAL;
2422     QDict *options, *encryptopts;
2423     int fmt;
2424 
2425     options = qemu_opts_to_qdict(opts, NULL);
2426     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
2427     QDECREF(options);
2428 
2429     fmt = qcow2_crypt_method_from_format(encryptfmt);
2430 
2431     switch (fmt) {
2432     case QCOW_CRYPT_LUKS:
2433         cryptoopts = block_crypto_create_opts_init(
2434             Q_CRYPTO_BLOCK_FORMAT_LUKS, encryptopts, errp);
2435         break;
2436     case QCOW_CRYPT_AES:
2437         cryptoopts = block_crypto_create_opts_init(
2438             Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp);
2439         break;
2440     default:
2441         error_setg(errp, "Unknown encryption format '%s'", encryptfmt);
2442         break;
2443     }
2444     if (!cryptoopts) {
2445         ret = -EINVAL;
2446         goto out;
2447     }
2448     s->crypt_method_header = fmt;
2449 
2450     crypto = qcrypto_block_create(cryptoopts, "encrypt.",
2451                                   qcow2_crypto_hdr_init_func,
2452                                   qcow2_crypto_hdr_write_func,
2453                                   bs, errp);
2454     if (!crypto) {
2455         ret = -EINVAL;
2456         goto out;
2457     }
2458 
2459     ret = qcow2_update_header(bs);
2460     if (ret < 0) {
2461         error_setg_errno(errp, -ret, "Could not write encryption header");
2462         goto out;
2463     }
2464 
2465  out:
2466     QDECREF(encryptopts);
2467     qcrypto_block_free(crypto);
2468     qapi_free_QCryptoBlockCreateOptions(cryptoopts);
2469     return ret;
2470 }
2471 
2472 
2473 typedef struct PreallocCo {
2474     BlockDriverState *bs;
2475     uint64_t offset;
2476     uint64_t new_length;
2477 
2478     int ret;
2479 } PreallocCo;
2480 
2481 /**
2482  * Preallocates metadata structures for data clusters between @offset (in the
2483  * guest disk) and @new_length (which is thus generally the new guest disk
2484  * size).
2485  *
2486  * Returns: 0 on success, -errno on failure.
2487  */
2488 static void coroutine_fn preallocate_co(void *opaque)
2489 {
2490     PreallocCo *params = opaque;
2491     BlockDriverState *bs = params->bs;
2492     uint64_t offset = params->offset;
2493     uint64_t new_length = params->new_length;
2494     BDRVQcow2State *s = bs->opaque;
2495     uint64_t bytes;
2496     uint64_t host_offset = 0;
2497     unsigned int cur_bytes;
2498     int ret;
2499     QCowL2Meta *meta;
2500 
2501     qemu_co_mutex_lock(&s->lock);
2502 
2503     assert(offset <= new_length);
2504     bytes = new_length - offset;
2505 
2506     while (bytes) {
2507         cur_bytes = MIN(bytes, INT_MAX);
2508         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2509                                          &host_offset, &meta);
2510         if (ret < 0) {
2511             goto done;
2512         }
2513 
2514         while (meta) {
2515             QCowL2Meta *next = meta->next;
2516 
2517             ret = qcow2_alloc_cluster_link_l2(bs, meta);
2518             if (ret < 0) {
2519                 qcow2_free_any_clusters(bs, meta->alloc_offset,
2520                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
2521                 goto done;
2522             }
2523 
2524             /* There are no dependent requests, but we need to remove our
2525              * request from the list of in-flight requests */
2526             QLIST_REMOVE(meta, next_in_flight);
2527 
2528             g_free(meta);
2529             meta = next;
2530         }
2531 
2532         /* TODO Preallocate data if requested */
2533 
2534         bytes -= cur_bytes;
2535         offset += cur_bytes;
2536     }
2537 
2538     /*
2539      * It is expected that the image file is large enough to actually contain
2540      * all of the allocated clusters (otherwise we get failing reads after
2541      * EOF). Extend the image to the last allocated sector.
2542      */
2543     if (host_offset != 0) {
2544         uint8_t data = 0;
2545         ret = bdrv_pwrite(bs->file, (host_offset + cur_bytes) - 1,
2546                           &data, 1);
2547         if (ret < 0) {
2548             goto done;
2549         }
2550     }
2551 
2552     ret = 0;
2553 
2554 done:
2555     qemu_co_mutex_unlock(&s->lock);
2556     params->ret = ret;
2557 }
2558 
2559 static int preallocate(BlockDriverState *bs,
2560                        uint64_t offset, uint64_t new_length)
2561 {
2562     PreallocCo params = {
2563         .bs         = bs,
2564         .offset     = offset,
2565         .new_length = new_length,
2566         .ret        = -EINPROGRESS,
2567     };
2568 
2569     if (qemu_in_coroutine()) {
2570         preallocate_co(&params);
2571     } else {
2572         Coroutine *co = qemu_coroutine_create(preallocate_co, &params);
2573         bdrv_coroutine_enter(bs, co);
2574         BDRV_POLL_WHILE(bs, params.ret == -EINPROGRESS);
2575     }
2576     return params.ret;
2577 }
2578 
2579 /* qcow2_refcount_metadata_size:
2580  * @clusters: number of clusters to refcount (including data and L1/L2 tables)
2581  * @cluster_size: size of a cluster, in bytes
2582  * @refcount_order: refcount bits power-of-2 exponent
2583  * @generous_increase: allow for the refcount table to be 1.5x as large as it
2584  *                     needs to be
2585  *
2586  * Returns: Number of bytes required for refcount blocks and table metadata.
2587  */
2588 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
2589                                      int refcount_order, bool generous_increase,
2590                                      uint64_t *refblock_count)
2591 {
2592     /*
2593      * Every host cluster is reference-counted, including metadata (even
2594      * refcount metadata is recursively included).
2595      *
2596      * An accurate formula for the size of refcount metadata size is difficult
2597      * to derive.  An easier method of calculation is finding the fixed point
2598      * where no further refcount blocks or table clusters are required to
2599      * reference count every cluster.
2600      */
2601     int64_t blocks_per_table_cluster = cluster_size / sizeof(uint64_t);
2602     int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
2603     int64_t table = 0;  /* number of refcount table clusters */
2604     int64_t blocks = 0; /* number of refcount block clusters */
2605     int64_t last;
2606     int64_t n = 0;
2607 
2608     do {
2609         last = n;
2610         blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
2611         table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
2612         n = clusters + blocks + table;
2613 
2614         if (n == last && generous_increase) {
2615             clusters += DIV_ROUND_UP(table, 2);
2616             n = 0; /* force another loop */
2617             generous_increase = false;
2618         }
2619     } while (n != last);
2620 
2621     if (refblock_count) {
2622         *refblock_count = blocks;
2623     }
2624 
2625     return (blocks + table) * cluster_size;
2626 }
2627 
2628 /**
2629  * qcow2_calc_prealloc_size:
2630  * @total_size: virtual disk size in bytes
2631  * @cluster_size: cluster size in bytes
2632  * @refcount_order: refcount bits power-of-2 exponent
2633  *
2634  * Returns: Total number of bytes required for the fully allocated image
2635  * (including metadata).
2636  */
2637 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
2638                                         size_t cluster_size,
2639                                         int refcount_order)
2640 {
2641     int64_t meta_size = 0;
2642     uint64_t nl1e, nl2e;
2643     int64_t aligned_total_size = align_offset(total_size, cluster_size);
2644 
2645     /* header: 1 cluster */
2646     meta_size += cluster_size;
2647 
2648     /* total size of L2 tables */
2649     nl2e = aligned_total_size / cluster_size;
2650     nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
2651     meta_size += nl2e * sizeof(uint64_t);
2652 
2653     /* total size of L1 tables */
2654     nl1e = nl2e * sizeof(uint64_t) / cluster_size;
2655     nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
2656     meta_size += nl1e * sizeof(uint64_t);
2657 
2658     /* total size of refcount table and blocks */
2659     meta_size += qcow2_refcount_metadata_size(
2660             (meta_size + aligned_total_size) / cluster_size,
2661             cluster_size, refcount_order, false, NULL);
2662 
2663     return meta_size + aligned_total_size;
2664 }
2665 
2666 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, Error **errp)
2667 {
2668     size_t cluster_size;
2669     int cluster_bits;
2670 
2671     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2672                                          DEFAULT_CLUSTER_SIZE);
2673     cluster_bits = ctz32(cluster_size);
2674     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
2675         (1 << cluster_bits) != cluster_size)
2676     {
2677         error_setg(errp, "Cluster size must be a power of two between %d and "
2678                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
2679         return 0;
2680     }
2681     return cluster_size;
2682 }
2683 
2684 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
2685 {
2686     char *buf;
2687     int ret;
2688 
2689     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2690     if (!buf) {
2691         ret = 3; /* default */
2692     } else if (!strcmp(buf, "0.10")) {
2693         ret = 2;
2694     } else if (!strcmp(buf, "1.1")) {
2695         ret = 3;
2696     } else {
2697         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2698         ret = -EINVAL;
2699     }
2700     g_free(buf);
2701     return ret;
2702 }
2703 
2704 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
2705                                                 Error **errp)
2706 {
2707     uint64_t refcount_bits;
2708 
2709     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
2710     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2711         error_setg(errp, "Refcount width must be a power of two and may not "
2712                    "exceed 64 bits");
2713         return 0;
2714     }
2715 
2716     if (version < 3 && refcount_bits != 16) {
2717         error_setg(errp, "Different refcount widths than 16 bits require "
2718                    "compatibility level 1.1 or above (use compat=1.1 or "
2719                    "greater)");
2720         return 0;
2721     }
2722 
2723     return refcount_bits;
2724 }
2725 
2726 static int qcow2_create2(const char *filename, int64_t total_size,
2727                          const char *backing_file, const char *backing_format,
2728                          int flags, size_t cluster_size, PreallocMode prealloc,
2729                          QemuOpts *opts, int version, int refcount_order,
2730                          const char *encryptfmt, Error **errp)
2731 {
2732     QDict *options;
2733 
2734     /*
2735      * Open the image file and write a minimal qcow2 header.
2736      *
2737      * We keep things simple and start with a zero-sized image. We also
2738      * do without refcount blocks or a L1 table for now. We'll fix the
2739      * inconsistency later.
2740      *
2741      * We do need a refcount table because growing the refcount table means
2742      * allocating two new refcount blocks - the seconds of which would be at
2743      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2744      * size for any qcow2 image.
2745      */
2746     BlockBackend *blk;
2747     QCowHeader *header;
2748     uint64_t* refcount_table;
2749     Error *local_err = NULL;
2750     int ret;
2751 
2752     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
2753         int64_t prealloc_size =
2754             qcow2_calc_prealloc_size(total_size, cluster_size, refcount_order);
2755         qemu_opt_set_number(opts, BLOCK_OPT_SIZE, prealloc_size, &error_abort);
2756         qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_str(prealloc),
2757                      &error_abort);
2758     }
2759 
2760     ret = bdrv_create_file(filename, opts, &local_err);
2761     if (ret < 0) {
2762         error_propagate(errp, local_err);
2763         return ret;
2764     }
2765 
2766     blk = blk_new_open(filename, NULL, NULL,
2767                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2768                        &local_err);
2769     if (blk == NULL) {
2770         error_propagate(errp, local_err);
2771         return -EIO;
2772     }
2773 
2774     blk_set_allow_write_beyond_eof(blk, true);
2775 
2776     /* Write the header */
2777     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
2778     header = g_malloc0(cluster_size);
2779     *header = (QCowHeader) {
2780         .magic                      = cpu_to_be32(QCOW_MAGIC),
2781         .version                    = cpu_to_be32(version),
2782         .cluster_bits               = cpu_to_be32(ctz32(cluster_size)),
2783         .size                       = cpu_to_be64(0),
2784         .l1_table_offset            = cpu_to_be64(0),
2785         .l1_size                    = cpu_to_be32(0),
2786         .refcount_table_offset      = cpu_to_be64(cluster_size),
2787         .refcount_table_clusters    = cpu_to_be32(1),
2788         .refcount_order             = cpu_to_be32(refcount_order),
2789         .header_length              = cpu_to_be32(sizeof(*header)),
2790     };
2791 
2792     /* We'll update this to correct value later */
2793     header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
2794 
2795     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
2796         header->compatible_features |=
2797             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
2798     }
2799 
2800     ret = blk_pwrite(blk, 0, header, cluster_size, 0);
2801     g_free(header);
2802     if (ret < 0) {
2803         error_setg_errno(errp, -ret, "Could not write qcow2 header");
2804         goto out;
2805     }
2806 
2807     /* Write a refcount table with one refcount block */
2808     refcount_table = g_malloc0(2 * cluster_size);
2809     refcount_table[0] = cpu_to_be64(2 * cluster_size);
2810     ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
2811     g_free(refcount_table);
2812 
2813     if (ret < 0) {
2814         error_setg_errno(errp, -ret, "Could not write refcount table");
2815         goto out;
2816     }
2817 
2818     blk_unref(blk);
2819     blk = NULL;
2820 
2821     /*
2822      * And now open the image and make it consistent first (i.e. increase the
2823      * refcount of the cluster that is occupied by the header and the refcount
2824      * table)
2825      */
2826     options = qdict_new();
2827     qdict_put_str(options, "driver", "qcow2");
2828     blk = blk_new_open(filename, NULL, options,
2829                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
2830                        &local_err);
2831     if (blk == NULL) {
2832         error_propagate(errp, local_err);
2833         ret = -EIO;
2834         goto out;
2835     }
2836 
2837     ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
2838     if (ret < 0) {
2839         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
2840                          "header and refcount table");
2841         goto out;
2842 
2843     } else if (ret != 0) {
2844         error_report("Huh, first cluster in empty image is already in use?");
2845         abort();
2846     }
2847 
2848     /* Create a full header (including things like feature table) */
2849     ret = qcow2_update_header(blk_bs(blk));
2850     if (ret < 0) {
2851         error_setg_errno(errp, -ret, "Could not update qcow2 header");
2852         goto out;
2853     }
2854 
2855     /* Okay, now that we have a valid image, let's give it the right size */
2856     ret = blk_truncate(blk, total_size, PREALLOC_MODE_OFF, errp);
2857     if (ret < 0) {
2858         error_prepend(errp, "Could not resize image: ");
2859         goto out;
2860     }
2861 
2862     /* Want a backing file? There you go.*/
2863     if (backing_file) {
2864         ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format);
2865         if (ret < 0) {
2866             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
2867                              "with format '%s'", backing_file, backing_format);
2868             goto out;
2869         }
2870     }
2871 
2872     /* Want encryption? There you go. */
2873     if (encryptfmt) {
2874         ret = qcow2_set_up_encryption(blk_bs(blk), encryptfmt, opts, errp);
2875         if (ret < 0) {
2876             goto out;
2877         }
2878     }
2879 
2880     /* And if we're supposed to preallocate metadata, do that now */
2881     if (prealloc != PREALLOC_MODE_OFF) {
2882         ret = preallocate(blk_bs(blk), 0, total_size);
2883         if (ret < 0) {
2884             error_setg_errno(errp, -ret, "Could not preallocate metadata");
2885             goto out;
2886         }
2887     }
2888 
2889     blk_unref(blk);
2890     blk = NULL;
2891 
2892     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
2893      * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
2894      * have to setup decryption context. We're not doing any I/O on the top
2895      * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
2896      * not have effect.
2897      */
2898     options = qdict_new();
2899     qdict_put_str(options, "driver", "qcow2");
2900     blk = blk_new_open(filename, NULL, options,
2901                        BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO,
2902                        &local_err);
2903     if (blk == NULL) {
2904         error_propagate(errp, local_err);
2905         ret = -EIO;
2906         goto out;
2907     }
2908 
2909     ret = 0;
2910 out:
2911     if (blk) {
2912         blk_unref(blk);
2913     }
2914     return ret;
2915 }
2916 
2917 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
2918 {
2919     char *backing_file = NULL;
2920     char *backing_fmt = NULL;
2921     char *buf = NULL;
2922     uint64_t size = 0;
2923     int flags = 0;
2924     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
2925     PreallocMode prealloc;
2926     int version;
2927     uint64_t refcount_bits;
2928     int refcount_order;
2929     char *encryptfmt = NULL;
2930     Error *local_err = NULL;
2931     int ret;
2932 
2933     /* Read out options */
2934     size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2935                     BDRV_SECTOR_SIZE);
2936     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2937     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2938     encryptfmt = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
2939     if (encryptfmt) {
2940         if (qemu_opt_get(opts, BLOCK_OPT_ENCRYPT)) {
2941             error_setg(errp, "Options " BLOCK_OPT_ENCRYPT " and "
2942                        BLOCK_OPT_ENCRYPT_FORMAT " are mutually exclusive");
2943             ret = -EINVAL;
2944             goto finish;
2945         }
2946     } else if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2947         encryptfmt = g_strdup("aes");
2948     }
2949     cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err);
2950     if (local_err) {
2951         error_propagate(errp, local_err);
2952         ret = -EINVAL;
2953         goto finish;
2954     }
2955     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2956     prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
2957                                PREALLOC_MODE_OFF, &local_err);
2958     if (local_err) {
2959         error_propagate(errp, local_err);
2960         ret = -EINVAL;
2961         goto finish;
2962     }
2963 
2964     version = qcow2_opt_get_version_del(opts, &local_err);
2965     if (local_err) {
2966         error_propagate(errp, local_err);
2967         ret = -EINVAL;
2968         goto finish;
2969     }
2970 
2971     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2972         flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2973     }
2974 
2975     if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2976         error_setg(errp, "Backing file and preallocation cannot be used at "
2977                    "the same time");
2978         ret = -EINVAL;
2979         goto finish;
2980     }
2981 
2982     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2983         error_setg(errp, "Lazy refcounts only supported with compatibility "
2984                    "level 1.1 and above (use compat=1.1 or greater)");
2985         ret = -EINVAL;
2986         goto finish;
2987     }
2988 
2989     refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
2990     if (local_err) {
2991         error_propagate(errp, local_err);
2992         ret = -EINVAL;
2993         goto finish;
2994     }
2995 
2996     refcount_order = ctz32(refcount_bits);
2997 
2998     ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2999                         cluster_size, prealloc, opts, version, refcount_order,
3000                         encryptfmt, &local_err);
3001     error_propagate(errp, local_err);
3002 
3003 finish:
3004     g_free(backing_file);
3005     g_free(backing_fmt);
3006     g_free(encryptfmt);
3007     g_free(buf);
3008     return ret;
3009 }
3010 
3011 
3012 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes)
3013 {
3014     int64_t nr;
3015     int res;
3016 
3017     /* Clamp to image length, before checking status of underlying sectors */
3018     if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
3019         bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset;
3020     }
3021 
3022     if (!bytes) {
3023         return true;
3024     }
3025     res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL);
3026     return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == bytes;
3027 }
3028 
3029 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
3030     int64_t offset, int bytes, BdrvRequestFlags flags)
3031 {
3032     int ret;
3033     BDRVQcow2State *s = bs->opaque;
3034 
3035     uint32_t head = offset % s->cluster_size;
3036     uint32_t tail = (offset + bytes) % s->cluster_size;
3037 
3038     trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes);
3039     if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) {
3040         tail = 0;
3041     }
3042 
3043     if (head || tail) {
3044         uint64_t off;
3045         unsigned int nr;
3046 
3047         assert(head + bytes <= s->cluster_size);
3048 
3049         /* check whether remainder of cluster already reads as zero */
3050         if (!(is_zero(bs, offset - head, head) &&
3051               is_zero(bs, offset + bytes,
3052                       tail ? s->cluster_size - tail : 0))) {
3053             return -ENOTSUP;
3054         }
3055 
3056         qemu_co_mutex_lock(&s->lock);
3057         /* We can have new write after previous check */
3058         offset = QEMU_ALIGN_DOWN(offset, s->cluster_size);
3059         bytes = s->cluster_size;
3060         nr = s->cluster_size;
3061         ret = qcow2_get_cluster_offset(bs, offset, &nr, &off);
3062         if (ret != QCOW2_CLUSTER_UNALLOCATED &&
3063             ret != QCOW2_CLUSTER_ZERO_PLAIN &&
3064             ret != QCOW2_CLUSTER_ZERO_ALLOC) {
3065             qemu_co_mutex_unlock(&s->lock);
3066             return -ENOTSUP;
3067         }
3068     } else {
3069         qemu_co_mutex_lock(&s->lock);
3070     }
3071 
3072     trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes);
3073 
3074     /* Whatever is left can use real zero clusters */
3075     ret = qcow2_cluster_zeroize(bs, offset, bytes, flags);
3076     qemu_co_mutex_unlock(&s->lock);
3077 
3078     return ret;
3079 }
3080 
3081 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
3082                                           int64_t offset, int bytes)
3083 {
3084     int ret;
3085     BDRVQcow2State *s = bs->opaque;
3086 
3087     if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
3088         assert(bytes < s->cluster_size);
3089         /* Ignore partial clusters, except for the special case of the
3090          * complete partial cluster at the end of an unaligned file */
3091         if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
3092             offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) {
3093             return -ENOTSUP;
3094         }
3095     }
3096 
3097     qemu_co_mutex_lock(&s->lock);
3098     ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST,
3099                                 false);
3100     qemu_co_mutex_unlock(&s->lock);
3101     return ret;
3102 }
3103 
3104 static int qcow2_truncate(BlockDriverState *bs, int64_t offset,
3105                           PreallocMode prealloc, Error **errp)
3106 {
3107     BDRVQcow2State *s = bs->opaque;
3108     uint64_t old_length;
3109     int64_t new_l1_size;
3110     int ret;
3111 
3112     if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA &&
3113         prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL)
3114     {
3115         error_setg(errp, "Unsupported preallocation mode '%s'",
3116                    PreallocMode_str(prealloc));
3117         return -ENOTSUP;
3118     }
3119 
3120     if (offset & 511) {
3121         error_setg(errp, "The new size must be a multiple of 512");
3122         return -EINVAL;
3123     }
3124 
3125     /* cannot proceed if image has snapshots */
3126     if (s->nb_snapshots) {
3127         error_setg(errp, "Can't resize an image which has snapshots");
3128         return -ENOTSUP;
3129     }
3130 
3131     /* cannot proceed if image has bitmaps */
3132     if (s->nb_bitmaps) {
3133         /* TODO: resize bitmaps in the image */
3134         error_setg(errp, "Can't resize an image which has bitmaps");
3135         return -ENOTSUP;
3136     }
3137 
3138     old_length = bs->total_sectors * 512;
3139     new_l1_size = size_to_l1(s, offset);
3140 
3141     if (offset < old_length) {
3142         int64_t last_cluster, old_file_size;
3143         if (prealloc != PREALLOC_MODE_OFF) {
3144             error_setg(errp,
3145                        "Preallocation can't be used for shrinking an image");
3146             return -EINVAL;
3147         }
3148 
3149         ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size),
3150                                     old_length - ROUND_UP(offset,
3151                                                           s->cluster_size),
3152                                     QCOW2_DISCARD_ALWAYS, true);
3153         if (ret < 0) {
3154             error_setg_errno(errp, -ret, "Failed to discard cropped clusters");
3155             return ret;
3156         }
3157 
3158         ret = qcow2_shrink_l1_table(bs, new_l1_size);
3159         if (ret < 0) {
3160             error_setg_errno(errp, -ret,
3161                              "Failed to reduce the number of L2 tables");
3162             return ret;
3163         }
3164 
3165         ret = qcow2_shrink_reftable(bs);
3166         if (ret < 0) {
3167             error_setg_errno(errp, -ret,
3168                              "Failed to discard unused refblocks");
3169             return ret;
3170         }
3171 
3172         old_file_size = bdrv_getlength(bs->file->bs);
3173         if (old_file_size < 0) {
3174             error_setg_errno(errp, -old_file_size,
3175                              "Failed to inquire current file length");
3176             return old_file_size;
3177         }
3178         last_cluster = qcow2_get_last_cluster(bs, old_file_size);
3179         if (last_cluster < 0) {
3180             error_setg_errno(errp, -last_cluster,
3181                              "Failed to find the last cluster");
3182             return last_cluster;
3183         }
3184         if ((last_cluster + 1) * s->cluster_size < old_file_size) {
3185             Error *local_err = NULL;
3186 
3187             bdrv_truncate(bs->file, (last_cluster + 1) * s->cluster_size,
3188                           PREALLOC_MODE_OFF, &local_err);
3189             if (local_err) {
3190                 warn_reportf_err(local_err,
3191                                  "Failed to truncate the tail of the image: ");
3192             }
3193         }
3194     } else {
3195         ret = qcow2_grow_l1_table(bs, new_l1_size, true);
3196         if (ret < 0) {
3197             error_setg_errno(errp, -ret, "Failed to grow the L1 table");
3198             return ret;
3199         }
3200     }
3201 
3202     switch (prealloc) {
3203     case PREALLOC_MODE_OFF:
3204         break;
3205 
3206     case PREALLOC_MODE_METADATA:
3207         ret = preallocate(bs, old_length, offset);
3208         if (ret < 0) {
3209             error_setg_errno(errp, -ret, "Preallocation failed");
3210             return ret;
3211         }
3212         break;
3213 
3214     case PREALLOC_MODE_FALLOC:
3215     case PREALLOC_MODE_FULL:
3216     {
3217         int64_t allocation_start, host_offset, guest_offset;
3218         int64_t clusters_allocated;
3219         int64_t old_file_size, new_file_size;
3220         uint64_t nb_new_data_clusters, nb_new_l2_tables;
3221 
3222         old_file_size = bdrv_getlength(bs->file->bs);
3223         if (old_file_size < 0) {
3224             error_setg_errno(errp, -old_file_size,
3225                              "Failed to inquire current file length");
3226             return old_file_size;
3227         }
3228         old_file_size = ROUND_UP(old_file_size, s->cluster_size);
3229 
3230         nb_new_data_clusters = DIV_ROUND_UP(offset - old_length,
3231                                             s->cluster_size);
3232 
3233         /* This is an overestimation; we will not actually allocate space for
3234          * these in the file but just make sure the new refcount structures are
3235          * able to cover them so we will not have to allocate new refblocks
3236          * while entering the data blocks in the potentially new L2 tables.
3237          * (We do not actually care where the L2 tables are placed. Maybe they
3238          *  are already allocated or they can be placed somewhere before
3239          *  @old_file_size. It does not matter because they will be fully
3240          *  allocated automatically, so they do not need to be covered by the
3241          *  preallocation. All that matters is that we will not have to allocate
3242          *  new refcount structures for them.) */
3243         nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters,
3244                                         s->cluster_size / sizeof(uint64_t));
3245         /* The cluster range may not be aligned to L2 boundaries, so add one L2
3246          * table for a potential head/tail */
3247         nb_new_l2_tables++;
3248 
3249         allocation_start = qcow2_refcount_area(bs, old_file_size,
3250                                                nb_new_data_clusters +
3251                                                nb_new_l2_tables,
3252                                                true, 0, 0);
3253         if (allocation_start < 0) {
3254             error_setg_errno(errp, -allocation_start,
3255                              "Failed to resize refcount structures");
3256             return allocation_start;
3257         }
3258 
3259         clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start,
3260                                                      nb_new_data_clusters);
3261         if (clusters_allocated < 0) {
3262             error_setg_errno(errp, -clusters_allocated,
3263                              "Failed to allocate data clusters");
3264             return -clusters_allocated;
3265         }
3266 
3267         assert(clusters_allocated == nb_new_data_clusters);
3268 
3269         /* Allocate the data area */
3270         new_file_size = allocation_start +
3271                         nb_new_data_clusters * s->cluster_size;
3272         ret = bdrv_truncate(bs->file, new_file_size, prealloc, errp);
3273         if (ret < 0) {
3274             error_prepend(errp, "Failed to resize underlying file: ");
3275             qcow2_free_clusters(bs, allocation_start,
3276                                 nb_new_data_clusters * s->cluster_size,
3277                                 QCOW2_DISCARD_OTHER);
3278             return ret;
3279         }
3280 
3281         /* Create the necessary L2 entries */
3282         host_offset = allocation_start;
3283         guest_offset = old_length;
3284         while (nb_new_data_clusters) {
3285             int64_t nb_clusters = MIN(
3286                 nb_new_data_clusters,
3287                 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset));
3288             QCowL2Meta allocation = {
3289                 .offset       = guest_offset,
3290                 .alloc_offset = host_offset,
3291                 .nb_clusters  = nb_clusters,
3292             };
3293             qemu_co_queue_init(&allocation.dependent_requests);
3294 
3295             ret = qcow2_alloc_cluster_link_l2(bs, &allocation);
3296             if (ret < 0) {
3297                 error_setg_errno(errp, -ret, "Failed to update L2 tables");
3298                 qcow2_free_clusters(bs, host_offset,
3299                                     nb_new_data_clusters * s->cluster_size,
3300                                     QCOW2_DISCARD_OTHER);
3301                 return ret;
3302             }
3303 
3304             guest_offset += nb_clusters * s->cluster_size;
3305             host_offset += nb_clusters * s->cluster_size;
3306             nb_new_data_clusters -= nb_clusters;
3307         }
3308         break;
3309     }
3310 
3311     default:
3312         g_assert_not_reached();
3313     }
3314 
3315     if (prealloc != PREALLOC_MODE_OFF) {
3316         /* Flush metadata before actually changing the image size */
3317         ret = bdrv_flush(bs);
3318         if (ret < 0) {
3319             error_setg_errno(errp, -ret,
3320                              "Failed to flush the preallocated area to disk");
3321             return ret;
3322         }
3323     }
3324 
3325     /* write updated header.size */
3326     offset = cpu_to_be64(offset);
3327     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
3328                            &offset, sizeof(uint64_t));
3329     if (ret < 0) {
3330         error_setg_errno(errp, -ret, "Failed to update the image size");
3331         return ret;
3332     }
3333 
3334     s->l1_vm_state_index = new_l1_size;
3335     return 0;
3336 }
3337 
3338 /* XXX: put compressed sectors first, then all the cluster aligned
3339    tables to avoid losing bytes in alignment */
3340 static coroutine_fn int
3341 qcow2_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
3342                             uint64_t bytes, QEMUIOVector *qiov)
3343 {
3344     BDRVQcow2State *s = bs->opaque;
3345     QEMUIOVector hd_qiov;
3346     struct iovec iov;
3347     z_stream strm;
3348     int ret, out_len;
3349     uint8_t *buf, *out_buf;
3350     int64_t cluster_offset;
3351 
3352     if (bytes == 0) {
3353         /* align end of file to a sector boundary to ease reading with
3354            sector based I/Os */
3355         cluster_offset = bdrv_getlength(bs->file->bs);
3356         if (cluster_offset < 0) {
3357             return cluster_offset;
3358         }
3359         return bdrv_truncate(bs->file, cluster_offset, PREALLOC_MODE_OFF, NULL);
3360     }
3361 
3362     if (offset_into_cluster(s, offset)) {
3363         return -EINVAL;
3364     }
3365 
3366     buf = qemu_blockalign(bs, s->cluster_size);
3367     if (bytes != s->cluster_size) {
3368         if (bytes > s->cluster_size ||
3369             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
3370         {
3371             qemu_vfree(buf);
3372             return -EINVAL;
3373         }
3374         /* Zero-pad last write if image size is not cluster aligned */
3375         memset(buf + bytes, 0, s->cluster_size - bytes);
3376     }
3377     qemu_iovec_to_buf(qiov, 0, buf, bytes);
3378 
3379     out_buf = g_malloc(s->cluster_size);
3380 
3381     /* best compression, small window, no zlib header */
3382     memset(&strm, 0, sizeof(strm));
3383     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
3384                        Z_DEFLATED, -12,
3385                        9, Z_DEFAULT_STRATEGY);
3386     if (ret != 0) {
3387         ret = -EINVAL;
3388         goto fail;
3389     }
3390 
3391     strm.avail_in = s->cluster_size;
3392     strm.next_in = (uint8_t *)buf;
3393     strm.avail_out = s->cluster_size;
3394     strm.next_out = out_buf;
3395 
3396     ret = deflate(&strm, Z_FINISH);
3397     if (ret != Z_STREAM_END && ret != Z_OK) {
3398         deflateEnd(&strm);
3399         ret = -EINVAL;
3400         goto fail;
3401     }
3402     out_len = strm.next_out - out_buf;
3403 
3404     deflateEnd(&strm);
3405 
3406     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
3407         /* could not compress: write normal cluster */
3408         ret = qcow2_co_pwritev(bs, offset, bytes, qiov, 0);
3409         if (ret < 0) {
3410             goto fail;
3411         }
3412         goto success;
3413     }
3414 
3415     qemu_co_mutex_lock(&s->lock);
3416     cluster_offset =
3417         qcow2_alloc_compressed_cluster_offset(bs, offset, out_len);
3418     if (!cluster_offset) {
3419         qemu_co_mutex_unlock(&s->lock);
3420         ret = -EIO;
3421         goto fail;
3422     }
3423     cluster_offset &= s->cluster_offset_mask;
3424 
3425     ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
3426     qemu_co_mutex_unlock(&s->lock);
3427     if (ret < 0) {
3428         goto fail;
3429     }
3430 
3431     iov = (struct iovec) {
3432         .iov_base   = out_buf,
3433         .iov_len    = out_len,
3434     };
3435     qemu_iovec_init_external(&hd_qiov, &iov, 1);
3436 
3437     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
3438     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
3439     if (ret < 0) {
3440         goto fail;
3441     }
3442 success:
3443     ret = 0;
3444 fail:
3445     qemu_vfree(buf);
3446     g_free(out_buf);
3447     return ret;
3448 }
3449 
3450 static int make_completely_empty(BlockDriverState *bs)
3451 {
3452     BDRVQcow2State *s = bs->opaque;
3453     Error *local_err = NULL;
3454     int ret, l1_clusters;
3455     int64_t offset;
3456     uint64_t *new_reftable = NULL;
3457     uint64_t rt_entry, l1_size2;
3458     struct {
3459         uint64_t l1_offset;
3460         uint64_t reftable_offset;
3461         uint32_t reftable_clusters;
3462     } QEMU_PACKED l1_ofs_rt_ofs_cls;
3463 
3464     ret = qcow2_cache_empty(bs, s->l2_table_cache);
3465     if (ret < 0) {
3466         goto fail;
3467     }
3468 
3469     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
3470     if (ret < 0) {
3471         goto fail;
3472     }
3473 
3474     /* Refcounts will be broken utterly */
3475     ret = qcow2_mark_dirty(bs);
3476     if (ret < 0) {
3477         goto fail;
3478     }
3479 
3480     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
3481 
3482     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
3483     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
3484 
3485     /* After this call, neither the in-memory nor the on-disk refcount
3486      * information accurately describe the actual references */
3487 
3488     ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
3489                              l1_clusters * s->cluster_size, 0);
3490     if (ret < 0) {
3491         goto fail_broken_refcounts;
3492     }
3493     memset(s->l1_table, 0, l1_size2);
3494 
3495     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
3496 
3497     /* Overwrite enough clusters at the beginning of the sectors to place
3498      * the refcount table, a refcount block and the L1 table in; this may
3499      * overwrite parts of the existing refcount and L1 table, which is not
3500      * an issue because the dirty flag is set, complete data loss is in fact
3501      * desired and partial data loss is consequently fine as well */
3502     ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
3503                              (2 + l1_clusters) * s->cluster_size, 0);
3504     /* This call (even if it failed overall) may have overwritten on-disk
3505      * refcount structures; in that case, the in-memory refcount information
3506      * will probably differ from the on-disk information which makes the BDS
3507      * unusable */
3508     if (ret < 0) {
3509         goto fail_broken_refcounts;
3510     }
3511 
3512     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
3513     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
3514 
3515     /* "Create" an empty reftable (one cluster) directly after the image
3516      * header and an empty L1 table three clusters after the image header;
3517      * the cluster between those two will be used as the first refblock */
3518     l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
3519     l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
3520     l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
3521     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
3522                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
3523     if (ret < 0) {
3524         goto fail_broken_refcounts;
3525     }
3526 
3527     s->l1_table_offset = 3 * s->cluster_size;
3528 
3529     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
3530     if (!new_reftable) {
3531         ret = -ENOMEM;
3532         goto fail_broken_refcounts;
3533     }
3534 
3535     s->refcount_table_offset = s->cluster_size;
3536     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
3537     s->max_refcount_table_index = 0;
3538 
3539     g_free(s->refcount_table);
3540     s->refcount_table = new_reftable;
3541     new_reftable = NULL;
3542 
3543     /* Now the in-memory refcount information again corresponds to the on-disk
3544      * information (reftable is empty and no refblocks (the refblock cache is
3545      * empty)); however, this means some clusters (e.g. the image header) are
3546      * referenced, but not refcounted, but the normal qcow2 code assumes that
3547      * the in-memory information is always correct */
3548 
3549     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
3550 
3551     /* Enter the first refblock into the reftable */
3552     rt_entry = cpu_to_be64(2 * s->cluster_size);
3553     ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
3554                            &rt_entry, sizeof(rt_entry));
3555     if (ret < 0) {
3556         goto fail_broken_refcounts;
3557     }
3558     s->refcount_table[0] = 2 * s->cluster_size;
3559 
3560     s->free_cluster_index = 0;
3561     assert(3 + l1_clusters <= s->refcount_block_size);
3562     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
3563     if (offset < 0) {
3564         ret = offset;
3565         goto fail_broken_refcounts;
3566     } else if (offset > 0) {
3567         error_report("First cluster in emptied image is in use");
3568         abort();
3569     }
3570 
3571     /* Now finally the in-memory information corresponds to the on-disk
3572      * structures and is correct */
3573     ret = qcow2_mark_clean(bs);
3574     if (ret < 0) {
3575         goto fail;
3576     }
3577 
3578     ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size,
3579                         PREALLOC_MODE_OFF, &local_err);
3580     if (ret < 0) {
3581         error_report_err(local_err);
3582         goto fail;
3583     }
3584 
3585     return 0;
3586 
3587 fail_broken_refcounts:
3588     /* The BDS is unusable at this point. If we wanted to make it usable, we
3589      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
3590      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
3591      * again. However, because the functions which could have caused this error
3592      * path to be taken are used by those functions as well, it's very likely
3593      * that that sequence will fail as well. Therefore, just eject the BDS. */
3594     bs->drv = NULL;
3595 
3596 fail:
3597     g_free(new_reftable);
3598     return ret;
3599 }
3600 
3601 static int qcow2_make_empty(BlockDriverState *bs)
3602 {
3603     BDRVQcow2State *s = bs->opaque;
3604     uint64_t offset, end_offset;
3605     int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
3606     int l1_clusters, ret = 0;
3607 
3608     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
3609 
3610     if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps &&
3611         3 + l1_clusters <= s->refcount_block_size &&
3612         s->crypt_method_header != QCOW_CRYPT_LUKS) {
3613         /* The following function only works for qcow2 v3 images (it
3614          * requires the dirty flag) and only as long as there are no
3615          * features that reserve extra clusters (such as snapshots,
3616          * LUKS header, or persistent bitmaps), because it completely
3617          * empties the image.  Furthermore, the L1 table and three
3618          * additional clusters (image header, refcount table, one
3619          * refcount block) have to fit inside one refcount block. */
3620         return make_completely_empty(bs);
3621     }
3622 
3623     /* This fallback code simply discards every active cluster; this is slow,
3624      * but works in all cases */
3625     end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
3626     for (offset = 0; offset < end_offset; offset += step) {
3627         /* As this function is generally used after committing an external
3628          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
3629          * default action for this kind of discard is to pass the discard,
3630          * which will ideally result in an actually smaller image file, as
3631          * is probably desired. */
3632         ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
3633                                     QCOW2_DISCARD_SNAPSHOT, true);
3634         if (ret < 0) {
3635             break;
3636         }
3637     }
3638 
3639     return ret;
3640 }
3641 
3642 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
3643 {
3644     BDRVQcow2State *s = bs->opaque;
3645     int ret;
3646 
3647     qemu_co_mutex_lock(&s->lock);
3648     ret = qcow2_cache_write(bs, s->l2_table_cache);
3649     if (ret < 0) {
3650         qemu_co_mutex_unlock(&s->lock);
3651         return ret;
3652     }
3653 
3654     if (qcow2_need_accurate_refcounts(s)) {
3655         ret = qcow2_cache_write(bs, s->refcount_block_cache);
3656         if (ret < 0) {
3657             qemu_co_mutex_unlock(&s->lock);
3658             return ret;
3659         }
3660     }
3661     qemu_co_mutex_unlock(&s->lock);
3662 
3663     return 0;
3664 }
3665 
3666 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs,
3667                                        Error **errp)
3668 {
3669     Error *local_err = NULL;
3670     BlockMeasureInfo *info;
3671     uint64_t required = 0; /* bytes that contribute to required size */
3672     uint64_t virtual_size; /* disk size as seen by guest */
3673     uint64_t refcount_bits;
3674     uint64_t l2_tables;
3675     size_t cluster_size;
3676     int version;
3677     char *optstr;
3678     PreallocMode prealloc;
3679     bool has_backing_file;
3680 
3681     /* Parse image creation options */
3682     cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err);
3683     if (local_err) {
3684         goto err;
3685     }
3686 
3687     version = qcow2_opt_get_version_del(opts, &local_err);
3688     if (local_err) {
3689         goto err;
3690     }
3691 
3692     refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
3693     if (local_err) {
3694         goto err;
3695     }
3696 
3697     optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
3698     prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr,
3699                                PREALLOC_MODE_OFF, &local_err);
3700     g_free(optstr);
3701     if (local_err) {
3702         goto err;
3703     }
3704 
3705     optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
3706     has_backing_file = !!optstr;
3707     g_free(optstr);
3708 
3709     virtual_size = align_offset(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
3710                                 cluster_size);
3711 
3712     /* Check that virtual disk size is valid */
3713     l2_tables = DIV_ROUND_UP(virtual_size / cluster_size,
3714                              cluster_size / sizeof(uint64_t));
3715     if (l2_tables * sizeof(uint64_t) > QCOW_MAX_L1_SIZE) {
3716         error_setg(&local_err, "The image size is too large "
3717                                "(try using a larger cluster size)");
3718         goto err;
3719     }
3720 
3721     /* Account for input image */
3722     if (in_bs) {
3723         int64_t ssize = bdrv_getlength(in_bs);
3724         if (ssize < 0) {
3725             error_setg_errno(&local_err, -ssize,
3726                              "Unable to get image virtual_size");
3727             goto err;
3728         }
3729 
3730         virtual_size = align_offset(ssize, cluster_size);
3731 
3732         if (has_backing_file) {
3733             /* We don't how much of the backing chain is shared by the input
3734              * image and the new image file.  In the worst case the new image's
3735              * backing file has nothing in common with the input image.  Be
3736              * conservative and assume all clusters need to be written.
3737              */
3738             required = virtual_size;
3739         } else {
3740             int64_t offset;
3741             int64_t pnum = 0;
3742 
3743             for (offset = 0; offset < ssize; offset += pnum) {
3744                 int ret;
3745 
3746                 ret = bdrv_block_status_above(in_bs, NULL, offset,
3747                                               ssize - offset, &pnum, NULL,
3748                                               NULL);
3749                 if (ret < 0) {
3750                     error_setg_errno(&local_err, -ret,
3751                                      "Unable to get block status");
3752                     goto err;
3753                 }
3754 
3755                 if (ret & BDRV_BLOCK_ZERO) {
3756                     /* Skip zero regions (safe with no backing file) */
3757                 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
3758                            (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
3759                     /* Extend pnum to end of cluster for next iteration */
3760                     pnum = ROUND_UP(offset + pnum, cluster_size) - offset;
3761 
3762                     /* Count clusters we've seen */
3763                     required += offset % cluster_size + pnum;
3764                 }
3765             }
3766         }
3767     }
3768 
3769     /* Take into account preallocation.  Nothing special is needed for
3770      * PREALLOC_MODE_METADATA since metadata is always counted.
3771      */
3772     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
3773         required = virtual_size;
3774     }
3775 
3776     info = g_new(BlockMeasureInfo, 1);
3777     info->fully_allocated =
3778         qcow2_calc_prealloc_size(virtual_size, cluster_size,
3779                                  ctz32(refcount_bits));
3780 
3781     /* Remove data clusters that are not required.  This overestimates the
3782      * required size because metadata needed for the fully allocated file is
3783      * still counted.
3784      */
3785     info->required = info->fully_allocated - virtual_size + required;
3786     return info;
3787 
3788 err:
3789     error_propagate(errp, local_err);
3790     return NULL;
3791 }
3792 
3793 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3794 {
3795     BDRVQcow2State *s = bs->opaque;
3796     bdi->unallocated_blocks_are_zero = true;
3797     bdi->cluster_size = s->cluster_size;
3798     bdi->vm_state_offset = qcow2_vm_state_offset(s);
3799     return 0;
3800 }
3801 
3802 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
3803 {
3804     BDRVQcow2State *s = bs->opaque;
3805     ImageInfoSpecific *spec_info;
3806     QCryptoBlockInfo *encrypt_info = NULL;
3807 
3808     if (s->crypto != NULL) {
3809         encrypt_info = qcrypto_block_get_info(s->crypto, &error_abort);
3810     }
3811 
3812     spec_info = g_new(ImageInfoSpecific, 1);
3813     *spec_info = (ImageInfoSpecific){
3814         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
3815         .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1),
3816     };
3817     if (s->qcow_version == 2) {
3818         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
3819             .compat             = g_strdup("0.10"),
3820             .refcount_bits      = s->refcount_bits,
3821         };
3822     } else if (s->qcow_version == 3) {
3823         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
3824             .compat             = g_strdup("1.1"),
3825             .lazy_refcounts     = s->compatible_features &
3826                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
3827             .has_lazy_refcounts = true,
3828             .corrupt            = s->incompatible_features &
3829                                   QCOW2_INCOMPAT_CORRUPT,
3830             .has_corrupt        = true,
3831             .refcount_bits      = s->refcount_bits,
3832         };
3833     } else {
3834         /* if this assertion fails, this probably means a new version was
3835          * added without having it covered here */
3836         assert(false);
3837     }
3838 
3839     if (encrypt_info) {
3840         ImageInfoSpecificQCow2Encryption *qencrypt =
3841             g_new(ImageInfoSpecificQCow2Encryption, 1);
3842         switch (encrypt_info->format) {
3843         case Q_CRYPTO_BLOCK_FORMAT_QCOW:
3844             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES;
3845             qencrypt->u.aes = encrypt_info->u.qcow;
3846             break;
3847         case Q_CRYPTO_BLOCK_FORMAT_LUKS:
3848             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS;
3849             qencrypt->u.luks = encrypt_info->u.luks;
3850             break;
3851         default:
3852             abort();
3853         }
3854         /* Since we did shallow copy above, erase any pointers
3855          * in the original info */
3856         memset(&encrypt_info->u, 0, sizeof(encrypt_info->u));
3857         qapi_free_QCryptoBlockInfo(encrypt_info);
3858 
3859         spec_info->u.qcow2.data->has_encrypt = true;
3860         spec_info->u.qcow2.data->encrypt = qencrypt;
3861     }
3862 
3863     return spec_info;
3864 }
3865 
3866 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
3867                               int64_t pos)
3868 {
3869     BDRVQcow2State *s = bs->opaque;
3870 
3871     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
3872     return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos,
3873                                     qiov->size, qiov, 0);
3874 }
3875 
3876 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
3877                               int64_t pos)
3878 {
3879     BDRVQcow2State *s = bs->opaque;
3880 
3881     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
3882     return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos,
3883                                    qiov->size, qiov, 0);
3884 }
3885 
3886 /*
3887  * Downgrades an image's version. To achieve this, any incompatible features
3888  * have to be removed.
3889  */
3890 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
3891                            BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
3892 {
3893     BDRVQcow2State *s = bs->opaque;
3894     int current_version = s->qcow_version;
3895     int ret;
3896 
3897     if (target_version == current_version) {
3898         return 0;
3899     } else if (target_version > current_version) {
3900         return -EINVAL;
3901     } else if (target_version != 2) {
3902         return -EINVAL;
3903     }
3904 
3905     if (s->refcount_order != 4) {
3906         error_report("compat=0.10 requires refcount_bits=16");
3907         return -ENOTSUP;
3908     }
3909 
3910     /* clear incompatible features */
3911     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
3912         ret = qcow2_mark_clean(bs);
3913         if (ret < 0) {
3914             return ret;
3915         }
3916     }
3917 
3918     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
3919      * the first place; if that happens nonetheless, returning -ENOTSUP is the
3920      * best thing to do anyway */
3921 
3922     if (s->incompatible_features) {
3923         return -ENOTSUP;
3924     }
3925 
3926     /* since we can ignore compatible features, we can set them to 0 as well */
3927     s->compatible_features = 0;
3928     /* if lazy refcounts have been used, they have already been fixed through
3929      * clearing the dirty flag */
3930 
3931     /* clearing autoclear features is trivial */
3932     s->autoclear_features = 0;
3933 
3934     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
3935     if (ret < 0) {
3936         return ret;
3937     }
3938 
3939     s->qcow_version = target_version;
3940     ret = qcow2_update_header(bs);
3941     if (ret < 0) {
3942         s->qcow_version = current_version;
3943         return ret;
3944     }
3945     return 0;
3946 }
3947 
3948 typedef enum Qcow2AmendOperation {
3949     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
3950      * statically initialized to so that the helper CB can discern the first
3951      * invocation from an operation change */
3952     QCOW2_NO_OPERATION = 0,
3953 
3954     QCOW2_CHANGING_REFCOUNT_ORDER,
3955     QCOW2_DOWNGRADING,
3956 } Qcow2AmendOperation;
3957 
3958 typedef struct Qcow2AmendHelperCBInfo {
3959     /* The code coordinating the amend operations should only modify
3960      * these four fields; the rest will be managed by the CB */
3961     BlockDriverAmendStatusCB *original_status_cb;
3962     void *original_cb_opaque;
3963 
3964     Qcow2AmendOperation current_operation;
3965 
3966     /* Total number of operations to perform (only set once) */
3967     int total_operations;
3968 
3969     /* The following fields are managed by the CB */
3970 
3971     /* Number of operations completed */
3972     int operations_completed;
3973 
3974     /* Cumulative offset of all completed operations */
3975     int64_t offset_completed;
3976 
3977     Qcow2AmendOperation last_operation;
3978     int64_t last_work_size;
3979 } Qcow2AmendHelperCBInfo;
3980 
3981 static void qcow2_amend_helper_cb(BlockDriverState *bs,
3982                                   int64_t operation_offset,
3983                                   int64_t operation_work_size, void *opaque)
3984 {
3985     Qcow2AmendHelperCBInfo *info = opaque;
3986     int64_t current_work_size;
3987     int64_t projected_work_size;
3988 
3989     if (info->current_operation != info->last_operation) {
3990         if (info->last_operation != QCOW2_NO_OPERATION) {
3991             info->offset_completed += info->last_work_size;
3992             info->operations_completed++;
3993         }
3994 
3995         info->last_operation = info->current_operation;
3996     }
3997 
3998     assert(info->total_operations > 0);
3999     assert(info->operations_completed < info->total_operations);
4000 
4001     info->last_work_size = operation_work_size;
4002 
4003     current_work_size = info->offset_completed + operation_work_size;
4004 
4005     /* current_work_size is the total work size for (operations_completed + 1)
4006      * operations (which includes this one), so multiply it by the number of
4007      * operations not covered and divide it by the number of operations
4008      * covered to get a projection for the operations not covered */
4009     projected_work_size = current_work_size * (info->total_operations -
4010                                                info->operations_completed - 1)
4011                                             / (info->operations_completed + 1);
4012 
4013     info->original_status_cb(bs, info->offset_completed + operation_offset,
4014                              current_work_size + projected_work_size,
4015                              info->original_cb_opaque);
4016 }
4017 
4018 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
4019                                BlockDriverAmendStatusCB *status_cb,
4020                                void *cb_opaque)
4021 {
4022     BDRVQcow2State *s = bs->opaque;
4023     int old_version = s->qcow_version, new_version = old_version;
4024     uint64_t new_size = 0;
4025     const char *backing_file = NULL, *backing_format = NULL;
4026     bool lazy_refcounts = s->use_lazy_refcounts;
4027     const char *compat = NULL;
4028     uint64_t cluster_size = s->cluster_size;
4029     bool encrypt;
4030     int encformat;
4031     int refcount_bits = s->refcount_bits;
4032     Error *local_err = NULL;
4033     int ret;
4034     QemuOptDesc *desc = opts->list->desc;
4035     Qcow2AmendHelperCBInfo helper_cb_info;
4036 
4037     while (desc && desc->name) {
4038         if (!qemu_opt_find(opts, desc->name)) {
4039             /* only change explicitly defined options */
4040             desc++;
4041             continue;
4042         }
4043 
4044         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
4045             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
4046             if (!compat) {
4047                 /* preserve default */
4048             } else if (!strcmp(compat, "0.10")) {
4049                 new_version = 2;
4050             } else if (!strcmp(compat, "1.1")) {
4051                 new_version = 3;
4052             } else {
4053                 error_report("Unknown compatibility level %s", compat);
4054                 return -EINVAL;
4055             }
4056         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
4057             error_report("Cannot change preallocation mode");
4058             return -ENOTSUP;
4059         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
4060             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
4061         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
4062             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4063         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
4064             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4065         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
4066             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
4067                                         !!s->crypto);
4068 
4069             if (encrypt != !!s->crypto) {
4070                 error_report("Changing the encryption flag is not supported");
4071                 return -ENOTSUP;
4072             }
4073         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT_FORMAT)) {
4074             encformat = qcow2_crypt_method_from_format(
4075                 qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT));
4076 
4077             if (encformat != s->crypt_method_header) {
4078                 error_report("Changing the encryption format is not supported");
4079                 return -ENOTSUP;
4080             }
4081         } else if (g_str_has_prefix(desc->name, "encrypt.")) {
4082             error_report("Changing the encryption parameters is not supported");
4083             return -ENOTSUP;
4084         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
4085             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
4086                                              cluster_size);
4087             if (cluster_size != s->cluster_size) {
4088                 error_report("Changing the cluster size is not supported");
4089                 return -ENOTSUP;
4090             }
4091         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
4092             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
4093                                                lazy_refcounts);
4094         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
4095             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
4096                                                 refcount_bits);
4097 
4098             if (refcount_bits <= 0 || refcount_bits > 64 ||
4099                 !is_power_of_2(refcount_bits))
4100             {
4101                 error_report("Refcount width must be a power of two and may "
4102                              "not exceed 64 bits");
4103                 return -EINVAL;
4104             }
4105         } else {
4106             /* if this point is reached, this probably means a new option was
4107              * added without having it covered here */
4108             abort();
4109         }
4110 
4111         desc++;
4112     }
4113 
4114     helper_cb_info = (Qcow2AmendHelperCBInfo){
4115         .original_status_cb = status_cb,
4116         .original_cb_opaque = cb_opaque,
4117         .total_operations = (new_version < old_version)
4118                           + (s->refcount_bits != refcount_bits)
4119     };
4120 
4121     /* Upgrade first (some features may require compat=1.1) */
4122     if (new_version > old_version) {
4123         s->qcow_version = new_version;
4124         ret = qcow2_update_header(bs);
4125         if (ret < 0) {
4126             s->qcow_version = old_version;
4127             return ret;
4128         }
4129     }
4130 
4131     if (s->refcount_bits != refcount_bits) {
4132         int refcount_order = ctz32(refcount_bits);
4133 
4134         if (new_version < 3 && refcount_bits != 16) {
4135             error_report("Different refcount widths than 16 bits require "
4136                          "compatibility level 1.1 or above (use compat=1.1 or "
4137                          "greater)");
4138             return -EINVAL;
4139         }
4140 
4141         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
4142         ret = qcow2_change_refcount_order(bs, refcount_order,
4143                                           &qcow2_amend_helper_cb,
4144                                           &helper_cb_info, &local_err);
4145         if (ret < 0) {
4146             error_report_err(local_err);
4147             return ret;
4148         }
4149     }
4150 
4151     if (backing_file || backing_format) {
4152         ret = qcow2_change_backing_file(bs,
4153                     backing_file ?: s->image_backing_file,
4154                     backing_format ?: s->image_backing_format);
4155         if (ret < 0) {
4156             return ret;
4157         }
4158     }
4159 
4160     if (s->use_lazy_refcounts != lazy_refcounts) {
4161         if (lazy_refcounts) {
4162             if (new_version < 3) {
4163                 error_report("Lazy refcounts only supported with compatibility "
4164                              "level 1.1 and above (use compat=1.1 or greater)");
4165                 return -EINVAL;
4166             }
4167             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
4168             ret = qcow2_update_header(bs);
4169             if (ret < 0) {
4170                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
4171                 return ret;
4172             }
4173             s->use_lazy_refcounts = true;
4174         } else {
4175             /* make image clean first */
4176             ret = qcow2_mark_clean(bs);
4177             if (ret < 0) {
4178                 return ret;
4179             }
4180             /* now disallow lazy refcounts */
4181             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
4182             ret = qcow2_update_header(bs);
4183             if (ret < 0) {
4184                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
4185                 return ret;
4186             }
4187             s->use_lazy_refcounts = false;
4188         }
4189     }
4190 
4191     if (new_size) {
4192         BlockBackend *blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
4193         ret = blk_insert_bs(blk, bs, &local_err);
4194         if (ret < 0) {
4195             error_report_err(local_err);
4196             blk_unref(blk);
4197             return ret;
4198         }
4199 
4200         ret = blk_truncate(blk, new_size, PREALLOC_MODE_OFF, &local_err);
4201         blk_unref(blk);
4202         if (ret < 0) {
4203             error_report_err(local_err);
4204             return ret;
4205         }
4206     }
4207 
4208     /* Downgrade last (so unsupported features can be removed before) */
4209     if (new_version < old_version) {
4210         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
4211         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
4212                               &helper_cb_info);
4213         if (ret < 0) {
4214             return ret;
4215         }
4216     }
4217 
4218     return 0;
4219 }
4220 
4221 /*
4222  * If offset or size are negative, respectively, they will not be included in
4223  * the BLOCK_IMAGE_CORRUPTED event emitted.
4224  * fatal will be ignored for read-only BDS; corruptions found there will always
4225  * be considered non-fatal.
4226  */
4227 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
4228                              int64_t size, const char *message_format, ...)
4229 {
4230     BDRVQcow2State *s = bs->opaque;
4231     const char *node_name;
4232     char *message;
4233     va_list ap;
4234 
4235     fatal = fatal && !bs->read_only;
4236 
4237     if (s->signaled_corruption &&
4238         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
4239     {
4240         return;
4241     }
4242 
4243     va_start(ap, message_format);
4244     message = g_strdup_vprintf(message_format, ap);
4245     va_end(ap);
4246 
4247     if (fatal) {
4248         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
4249                 "corruption events will be suppressed\n", message);
4250     } else {
4251         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
4252                 "corruption events will be suppressed\n", message);
4253     }
4254 
4255     node_name = bdrv_get_node_name(bs);
4256     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
4257                                           *node_name != '\0', node_name,
4258                                           message, offset >= 0, offset,
4259                                           size >= 0, size,
4260                                           fatal, &error_abort);
4261     g_free(message);
4262 
4263     if (fatal) {
4264         qcow2_mark_corrupt(bs);
4265         bs->drv = NULL; /* make BDS unusable */
4266     }
4267 
4268     s->signaled_corruption = true;
4269 }
4270 
4271 static QemuOptsList qcow2_create_opts = {
4272     .name = "qcow2-create-opts",
4273     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
4274     .desc = {
4275         {
4276             .name = BLOCK_OPT_SIZE,
4277             .type = QEMU_OPT_SIZE,
4278             .help = "Virtual disk size"
4279         },
4280         {
4281             .name = BLOCK_OPT_COMPAT_LEVEL,
4282             .type = QEMU_OPT_STRING,
4283             .help = "Compatibility level (0.10 or 1.1)"
4284         },
4285         {
4286             .name = BLOCK_OPT_BACKING_FILE,
4287             .type = QEMU_OPT_STRING,
4288             .help = "File name of a base image"
4289         },
4290         {
4291             .name = BLOCK_OPT_BACKING_FMT,
4292             .type = QEMU_OPT_STRING,
4293             .help = "Image format of the base image"
4294         },
4295         {
4296             .name = BLOCK_OPT_ENCRYPT,
4297             .type = QEMU_OPT_BOOL,
4298             .help = "Encrypt the image with format 'aes'. (Deprecated "
4299                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
4300         },
4301         {
4302             .name = BLOCK_OPT_ENCRYPT_FORMAT,
4303             .type = QEMU_OPT_STRING,
4304             .help = "Encrypt the image, format choices: 'aes', 'luks'",
4305         },
4306         BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
4307             "ID of secret providing qcow AES key or LUKS passphrase"),
4308         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
4309         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
4310         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
4311         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
4312         BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
4313         BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
4314         {
4315             .name = BLOCK_OPT_CLUSTER_SIZE,
4316             .type = QEMU_OPT_SIZE,
4317             .help = "qcow2 cluster size",
4318             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
4319         },
4320         {
4321             .name = BLOCK_OPT_PREALLOC,
4322             .type = QEMU_OPT_STRING,
4323             .help = "Preallocation mode (allowed values: off, metadata, "
4324                     "falloc, full)"
4325         },
4326         {
4327             .name = BLOCK_OPT_LAZY_REFCOUNTS,
4328             .type = QEMU_OPT_BOOL,
4329             .help = "Postpone refcount updates",
4330             .def_value_str = "off"
4331         },
4332         {
4333             .name = BLOCK_OPT_REFCOUNT_BITS,
4334             .type = QEMU_OPT_NUMBER,
4335             .help = "Width of a reference count entry in bits",
4336             .def_value_str = "16"
4337         },
4338         { /* end of list */ }
4339     }
4340 };
4341 
4342 BlockDriver bdrv_qcow2 = {
4343     .format_name        = "qcow2",
4344     .instance_size      = sizeof(BDRVQcow2State),
4345     .bdrv_probe         = qcow2_probe,
4346     .bdrv_open          = qcow2_open,
4347     .bdrv_close         = qcow2_close,
4348     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
4349     .bdrv_reopen_commit   = qcow2_reopen_commit,
4350     .bdrv_reopen_abort    = qcow2_reopen_abort,
4351     .bdrv_join_options    = qcow2_join_options,
4352     .bdrv_child_perm      = bdrv_format_default_perms,
4353     .bdrv_create        = qcow2_create,
4354     .bdrv_has_zero_init = bdrv_has_zero_init_1,
4355     .bdrv_co_get_block_status = qcow2_co_get_block_status,
4356 
4357     .bdrv_co_preadv         = qcow2_co_preadv,
4358     .bdrv_co_pwritev        = qcow2_co_pwritev,
4359     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
4360 
4361     .bdrv_co_pwrite_zeroes  = qcow2_co_pwrite_zeroes,
4362     .bdrv_co_pdiscard       = qcow2_co_pdiscard,
4363     .bdrv_truncate          = qcow2_truncate,
4364     .bdrv_co_pwritev_compressed = qcow2_co_pwritev_compressed,
4365     .bdrv_make_empty        = qcow2_make_empty,
4366 
4367     .bdrv_snapshot_create   = qcow2_snapshot_create,
4368     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
4369     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
4370     .bdrv_snapshot_list     = qcow2_snapshot_list,
4371     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
4372     .bdrv_measure           = qcow2_measure,
4373     .bdrv_get_info          = qcow2_get_info,
4374     .bdrv_get_specific_info = qcow2_get_specific_info,
4375 
4376     .bdrv_save_vmstate    = qcow2_save_vmstate,
4377     .bdrv_load_vmstate    = qcow2_load_vmstate,
4378 
4379     .supports_backing           = true,
4380     .bdrv_change_backing_file   = qcow2_change_backing_file,
4381 
4382     .bdrv_refresh_limits        = qcow2_refresh_limits,
4383     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
4384     .bdrv_inactivate            = qcow2_inactivate,
4385 
4386     .create_opts         = &qcow2_create_opts,
4387     .bdrv_check          = qcow2_check,
4388     .bdrv_amend_options  = qcow2_amend_options,
4389 
4390     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
4391     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
4392 
4393     .bdrv_reopen_bitmaps_rw = qcow2_reopen_bitmaps_rw,
4394     .bdrv_can_store_new_dirty_bitmap = qcow2_can_store_new_dirty_bitmap,
4395     .bdrv_remove_persistent_dirty_bitmap = qcow2_remove_persistent_dirty_bitmap,
4396 };
4397 
4398 static void bdrv_qcow2_init(void)
4399 {
4400     bdrv_register(&bdrv_qcow2);
4401 }
4402 
4403 block_init(bdrv_qcow2_init);
4404