xref: /openbmc/qemu/block/qcow2.c (revision ab938ae4)
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 #include "qemu/osdep.h"
25 #include "block/block_int.h"
26 #include "sysemu/block-backend.h"
27 #include "qemu/module.h"
28 #include <zlib.h>
29 #include "block/qcow2.h"
30 #include "qemu/error-report.h"
31 #include "qapi/qmp/qerror.h"
32 #include "qapi/qmp/qbool.h"
33 #include "qapi/qmp/types.h"
34 #include "qapi-event.h"
35 #include "trace.h"
36 #include "qemu/option_int.h"
37 #include "qemu/cutils.h"
38 #include "qemu/bswap.h"
39 #include "qapi/opts-visitor.h"
40 #include "qapi-visit.h"
41 #include "block/crypto.h"
42 
43 /*
44   Differences with QCOW:
45 
46   - Support for multiple incremental snapshots.
47   - Memory management by reference counts.
48   - Clusters which have a reference count of one have the bit
49     QCOW_OFLAG_COPIED to optimize write performance.
50   - Size of compressed clusters is stored in sectors to reduce bit usage
51     in the cluster offsets.
52   - Support for storing additional data (such as the VM state) in the
53     snapshots.
54   - If a backing store is used, the cluster size is not constrained
55     (could be backported to QCOW).
56   - L2 tables have always a size of one cluster.
57 */
58 
59 
60 typedef struct {
61     uint32_t magic;
62     uint32_t len;
63 } QEMU_PACKED QCowExtension;
64 
65 #define  QCOW2_EXT_MAGIC_END 0
66 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
67 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
68 #define  QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
69 #define  QCOW2_EXT_MAGIC_BITMAPS 0x23852875
70 
71 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
72 {
73     const QCowHeader *cow_header = (const void *)buf;
74 
75     if (buf_size >= sizeof(QCowHeader) &&
76         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
77         be32_to_cpu(cow_header->version) >= 2)
78         return 100;
79     else
80         return 0;
81 }
82 
83 
84 static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset,
85                                           uint8_t *buf, size_t buflen,
86                                           void *opaque, Error **errp)
87 {
88     BlockDriverState *bs = opaque;
89     BDRVQcow2State *s = bs->opaque;
90     ssize_t ret;
91 
92     if ((offset + buflen) > s->crypto_header.length) {
93         error_setg(errp, "Request for data outside of extension header");
94         return -1;
95     }
96 
97     ret = bdrv_pread(bs->file,
98                      s->crypto_header.offset + offset, buf, buflen);
99     if (ret < 0) {
100         error_setg_errno(errp, -ret, "Could not read encryption header");
101         return -1;
102     }
103     return ret;
104 }
105 
106 
107 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen,
108                                           void *opaque, Error **errp)
109 {
110     BlockDriverState *bs = opaque;
111     BDRVQcow2State *s = bs->opaque;
112     int64_t ret;
113     int64_t clusterlen;
114 
115     ret = qcow2_alloc_clusters(bs, headerlen);
116     if (ret < 0) {
117         error_setg_errno(errp, -ret,
118                          "Cannot allocate cluster for LUKS header size %zu",
119                          headerlen);
120         return -1;
121     }
122 
123     s->crypto_header.length = headerlen;
124     s->crypto_header.offset = ret;
125 
126     /* Zero fill remaining space in cluster so it has predictable
127      * content in case of future spec changes */
128     clusterlen = size_to_clusters(s, headerlen) * s->cluster_size;
129     ret = bdrv_pwrite_zeroes(bs->file,
130                              ret + headerlen,
131                              clusterlen - headerlen, 0);
132     if (ret < 0) {
133         error_setg_errno(errp, -ret, "Could not zero fill encryption header");
134         return -1;
135     }
136 
137     return ret;
138 }
139 
140 
141 static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset,
142                                            const uint8_t *buf, size_t buflen,
143                                            void *opaque, Error **errp)
144 {
145     BlockDriverState *bs = opaque;
146     BDRVQcow2State *s = bs->opaque;
147     ssize_t ret;
148 
149     if ((offset + buflen) > s->crypto_header.length) {
150         error_setg(errp, "Request for data outside of extension header");
151         return -1;
152     }
153 
154     ret = bdrv_pwrite(bs->file,
155                       s->crypto_header.offset + offset, buf, buflen);
156     if (ret < 0) {
157         error_setg_errno(errp, -ret, "Could not read encryption header");
158         return -1;
159     }
160     return ret;
161 }
162 
163 
164 /*
165  * read qcow2 extension and fill bs
166  * start reading from start_offset
167  * finish reading upon magic of value 0 or when end_offset reached
168  * unknown magic is skipped (future extension this version knows nothing about)
169  * return 0 upon success, non-0 otherwise
170  */
171 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
172                                  uint64_t end_offset, void **p_feature_table,
173                                  int flags, bool *need_update_header,
174                                  Error **errp)
175 {
176     BDRVQcow2State *s = bs->opaque;
177     QCowExtension ext;
178     uint64_t offset;
179     int ret;
180     Qcow2BitmapHeaderExt bitmaps_ext;
181 
182     if (need_update_header != NULL) {
183         *need_update_header = false;
184     }
185 
186 #ifdef DEBUG_EXT
187     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
188 #endif
189     offset = start_offset;
190     while (offset < end_offset) {
191 
192 #ifdef DEBUG_EXT
193         /* Sanity check */
194         if (offset > s->cluster_size)
195             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
196 
197         printf("attempting to read extended header in offset %lu\n", offset);
198 #endif
199 
200         ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
201         if (ret < 0) {
202             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
203                              "pread fail from offset %" PRIu64, offset);
204             return 1;
205         }
206         be32_to_cpus(&ext.magic);
207         be32_to_cpus(&ext.len);
208         offset += sizeof(ext);
209 #ifdef DEBUG_EXT
210         printf("ext.magic = 0x%x\n", ext.magic);
211 #endif
212         if (offset > end_offset || ext.len > end_offset - offset) {
213             error_setg(errp, "Header extension too large");
214             return -EINVAL;
215         }
216 
217         switch (ext.magic) {
218         case QCOW2_EXT_MAGIC_END:
219             return 0;
220 
221         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
222             if (ext.len >= sizeof(bs->backing_format)) {
223                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
224                            " too large (>=%zu)", ext.len,
225                            sizeof(bs->backing_format));
226                 return 2;
227             }
228             ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len);
229             if (ret < 0) {
230                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
231                                  "Could not read format name");
232                 return 3;
233             }
234             bs->backing_format[ext.len] = '\0';
235             s->image_backing_format = g_strdup(bs->backing_format);
236 #ifdef DEBUG_EXT
237             printf("Qcow2: Got format extension %s\n", bs->backing_format);
238 #endif
239             break;
240 
241         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
242             if (p_feature_table != NULL) {
243                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
244                 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
245                 if (ret < 0) {
246                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
247                                      "Could not read table");
248                     return ret;
249                 }
250 
251                 *p_feature_table = feature_table;
252             }
253             break;
254 
255         case QCOW2_EXT_MAGIC_CRYPTO_HEADER: {
256             unsigned int cflags = 0;
257             if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
258                 error_setg(errp, "CRYPTO header extension only "
259                            "expected with LUKS encryption method");
260                 return -EINVAL;
261             }
262             if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) {
263                 error_setg(errp, "CRYPTO header extension size %u, "
264                            "but expected size %zu", ext.len,
265                            sizeof(Qcow2CryptoHeaderExtension));
266                 return -EINVAL;
267             }
268 
269             ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len);
270             if (ret < 0) {
271                 error_setg_errno(errp, -ret,
272                                  "Unable to read CRYPTO header extension");
273                 return ret;
274             }
275             be64_to_cpus(&s->crypto_header.offset);
276             be64_to_cpus(&s->crypto_header.length);
277 
278             if ((s->crypto_header.offset % s->cluster_size) != 0) {
279                 error_setg(errp, "Encryption header offset '%" PRIu64 "' is "
280                            "not a multiple of cluster size '%u'",
281                            s->crypto_header.offset, s->cluster_size);
282                 return -EINVAL;
283             }
284 
285             if (flags & BDRV_O_NO_IO) {
286                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
287             }
288             s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
289                                            qcow2_crypto_hdr_read_func,
290                                            bs, cflags, errp);
291             if (!s->crypto) {
292                 return -EINVAL;
293             }
294         }   break;
295 
296         case QCOW2_EXT_MAGIC_BITMAPS:
297             if (ext.len != sizeof(bitmaps_ext)) {
298                 error_setg_errno(errp, -ret, "bitmaps_ext: "
299                                  "Invalid extension length");
300                 return -EINVAL;
301             }
302 
303             if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) {
304                 error_report("WARNING: a program lacking bitmap support "
305                              "modified this file, so all bitmaps are now "
306                              "considered inconsistent. Some clusters may be "
307                              "leaked, run 'qemu-img check -r' on the image "
308                              "file to fix.");
309                 if (need_update_header != NULL) {
310                     /* Updating is needed to drop invalid bitmap extension. */
311                     *need_update_header = true;
312                 }
313                 break;
314             }
315 
316             ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len);
317             if (ret < 0) {
318                 error_setg_errno(errp, -ret, "bitmaps_ext: "
319                                  "Could not read ext header");
320                 return ret;
321             }
322 
323             if (bitmaps_ext.reserved32 != 0) {
324                 error_setg_errno(errp, -ret, "bitmaps_ext: "
325                                  "Reserved field is not zero");
326                 return -EINVAL;
327             }
328 
329             be32_to_cpus(&bitmaps_ext.nb_bitmaps);
330             be64_to_cpus(&bitmaps_ext.bitmap_directory_size);
331             be64_to_cpus(&bitmaps_ext.bitmap_directory_offset);
332 
333             if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) {
334                 error_setg(errp,
335                            "bitmaps_ext: Image has %" PRIu32 " bitmaps, "
336                            "exceeding the QEMU supported maximum of %d",
337                            bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS);
338                 return -EINVAL;
339             }
340 
341             if (bitmaps_ext.nb_bitmaps == 0) {
342                 error_setg(errp, "found bitmaps extension with zero bitmaps");
343                 return -EINVAL;
344             }
345 
346             if (bitmaps_ext.bitmap_directory_offset & (s->cluster_size - 1)) {
347                 error_setg(errp, "bitmaps_ext: "
348                                  "invalid bitmap directory offset");
349                 return -EINVAL;
350             }
351 
352             if (bitmaps_ext.bitmap_directory_size >
353                 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) {
354                 error_setg(errp, "bitmaps_ext: "
355                                  "bitmap directory size (%" PRIu64 ") exceeds "
356                                  "the maximum supported size (%d)",
357                                  bitmaps_ext.bitmap_directory_size,
358                                  QCOW2_MAX_BITMAP_DIRECTORY_SIZE);
359                 return -EINVAL;
360             }
361 
362             s->nb_bitmaps = bitmaps_ext.nb_bitmaps;
363             s->bitmap_directory_offset =
364                     bitmaps_ext.bitmap_directory_offset;
365             s->bitmap_directory_size =
366                     bitmaps_ext.bitmap_directory_size;
367 
368 #ifdef DEBUG_EXT
369             printf("Qcow2: Got bitmaps extension: "
370                    "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n",
371                    s->bitmap_directory_offset, s->nb_bitmaps);
372 #endif
373             break;
374 
375         default:
376             /* unknown magic - save it in case we need to rewrite the header */
377             {
378                 Qcow2UnknownHeaderExtension *uext;
379 
380                 uext = g_malloc0(sizeof(*uext)  + ext.len);
381                 uext->magic = ext.magic;
382                 uext->len = ext.len;
383                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
384 
385                 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
386                 if (ret < 0) {
387                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
388                                      "Could not read data");
389                     return ret;
390                 }
391             }
392             break;
393         }
394 
395         offset += ((ext.len + 7) & ~7);
396     }
397 
398     return 0;
399 }
400 
401 static void cleanup_unknown_header_ext(BlockDriverState *bs)
402 {
403     BDRVQcow2State *s = bs->opaque;
404     Qcow2UnknownHeaderExtension *uext, *next;
405 
406     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
407         QLIST_REMOVE(uext, next);
408         g_free(uext);
409     }
410 }
411 
412 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
413                                        uint64_t mask)
414 {
415     char *features = g_strdup("");
416     char *old;
417 
418     while (table && table->name[0] != '\0') {
419         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
420             if (mask & (1ULL << table->bit)) {
421                 old = features;
422                 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
423                                            table->name);
424                 g_free(old);
425                 mask &= ~(1ULL << table->bit);
426             }
427         }
428         table++;
429     }
430 
431     if (mask) {
432         old = features;
433         features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
434                                    old, *old ? ", " : "", mask);
435         g_free(old);
436     }
437 
438     error_setg(errp, "Unsupported qcow2 feature(s): %s", features);
439     g_free(features);
440 }
441 
442 /*
443  * Sets the dirty bit and flushes afterwards if necessary.
444  *
445  * The incompatible_features bit is only set if the image file header was
446  * updated successfully.  Therefore it is not required to check the return
447  * value of this function.
448  */
449 int qcow2_mark_dirty(BlockDriverState *bs)
450 {
451     BDRVQcow2State *s = bs->opaque;
452     uint64_t val;
453     int ret;
454 
455     assert(s->qcow_version >= 3);
456 
457     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
458         return 0; /* already dirty */
459     }
460 
461     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
462     ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
463                       &val, sizeof(val));
464     if (ret < 0) {
465         return ret;
466     }
467     ret = bdrv_flush(bs->file->bs);
468     if (ret < 0) {
469         return ret;
470     }
471 
472     /* Only treat image as dirty if the header was updated successfully */
473     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
474     return 0;
475 }
476 
477 /*
478  * Clears the dirty bit and flushes before if necessary.  Only call this
479  * function when there are no pending requests, it does not guard against
480  * concurrent requests dirtying the image.
481  */
482 static int qcow2_mark_clean(BlockDriverState *bs)
483 {
484     BDRVQcow2State *s = bs->opaque;
485 
486     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
487         int ret;
488 
489         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
490 
491         ret = bdrv_flush(bs);
492         if (ret < 0) {
493             return ret;
494         }
495 
496         return qcow2_update_header(bs);
497     }
498     return 0;
499 }
500 
501 /*
502  * Marks the image as corrupt.
503  */
504 int qcow2_mark_corrupt(BlockDriverState *bs)
505 {
506     BDRVQcow2State *s = bs->opaque;
507 
508     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
509     return qcow2_update_header(bs);
510 }
511 
512 /*
513  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
514  * before if necessary.
515  */
516 int qcow2_mark_consistent(BlockDriverState *bs)
517 {
518     BDRVQcow2State *s = bs->opaque;
519 
520     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
521         int ret = bdrv_flush(bs);
522         if (ret < 0) {
523             return ret;
524         }
525 
526         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
527         return qcow2_update_header(bs);
528     }
529     return 0;
530 }
531 
532 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
533                        BdrvCheckMode fix)
534 {
535     int ret = qcow2_check_refcounts(bs, result, fix);
536     if (ret < 0) {
537         return ret;
538     }
539 
540     if (fix && result->check_errors == 0 && result->corruptions == 0) {
541         ret = qcow2_mark_clean(bs);
542         if (ret < 0) {
543             return ret;
544         }
545         return qcow2_mark_consistent(bs);
546     }
547     return ret;
548 }
549 
550 static int validate_table_offset(BlockDriverState *bs, uint64_t offset,
551                                  uint64_t entries, size_t entry_len)
552 {
553     BDRVQcow2State *s = bs->opaque;
554     uint64_t size;
555 
556     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
557      * because values will be passed to qemu functions taking int64_t. */
558     if (entries > INT64_MAX / entry_len) {
559         return -EINVAL;
560     }
561 
562     size = entries * entry_len;
563 
564     if (INT64_MAX - size < offset) {
565         return -EINVAL;
566     }
567 
568     /* Tables must be cluster aligned */
569     if (offset_into_cluster(s, offset) != 0) {
570         return -EINVAL;
571     }
572 
573     return 0;
574 }
575 
576 static QemuOptsList qcow2_runtime_opts = {
577     .name = "qcow2",
578     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
579     .desc = {
580         {
581             .name = QCOW2_OPT_LAZY_REFCOUNTS,
582             .type = QEMU_OPT_BOOL,
583             .help = "Postpone refcount updates",
584         },
585         {
586             .name = QCOW2_OPT_DISCARD_REQUEST,
587             .type = QEMU_OPT_BOOL,
588             .help = "Pass guest discard requests to the layer below",
589         },
590         {
591             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
592             .type = QEMU_OPT_BOOL,
593             .help = "Generate discard requests when snapshot related space "
594                     "is freed",
595         },
596         {
597             .name = QCOW2_OPT_DISCARD_OTHER,
598             .type = QEMU_OPT_BOOL,
599             .help = "Generate discard requests when other clusters are freed",
600         },
601         {
602             .name = QCOW2_OPT_OVERLAP,
603             .type = QEMU_OPT_STRING,
604             .help = "Selects which overlap checks to perform from a range of "
605                     "templates (none, constant, cached, all)",
606         },
607         {
608             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
609             .type = QEMU_OPT_STRING,
610             .help = "Selects which overlap checks to perform from a range of "
611                     "templates (none, constant, cached, all)",
612         },
613         {
614             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
615             .type = QEMU_OPT_BOOL,
616             .help = "Check for unintended writes into the main qcow2 header",
617         },
618         {
619             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
620             .type = QEMU_OPT_BOOL,
621             .help = "Check for unintended writes into the active L1 table",
622         },
623         {
624             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
625             .type = QEMU_OPT_BOOL,
626             .help = "Check for unintended writes into an active L2 table",
627         },
628         {
629             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
630             .type = QEMU_OPT_BOOL,
631             .help = "Check for unintended writes into the refcount table",
632         },
633         {
634             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
635             .type = QEMU_OPT_BOOL,
636             .help = "Check for unintended writes into a refcount block",
637         },
638         {
639             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
640             .type = QEMU_OPT_BOOL,
641             .help = "Check for unintended writes into the snapshot table",
642         },
643         {
644             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
645             .type = QEMU_OPT_BOOL,
646             .help = "Check for unintended writes into an inactive L1 table",
647         },
648         {
649             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
650             .type = QEMU_OPT_BOOL,
651             .help = "Check for unintended writes into an inactive L2 table",
652         },
653         {
654             .name = QCOW2_OPT_CACHE_SIZE,
655             .type = QEMU_OPT_SIZE,
656             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
657                     "cache size",
658         },
659         {
660             .name = QCOW2_OPT_L2_CACHE_SIZE,
661             .type = QEMU_OPT_SIZE,
662             .help = "Maximum L2 table cache size",
663         },
664         {
665             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
666             .type = QEMU_OPT_SIZE,
667             .help = "Maximum refcount block cache size",
668         },
669         {
670             .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
671             .type = QEMU_OPT_NUMBER,
672             .help = "Clean unused cache entries after this time (in seconds)",
673         },
674         BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
675             "ID of secret providing qcow2 AES key or LUKS passphrase"),
676         { /* end of list */ }
677     },
678 };
679 
680 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
681     [QCOW2_OL_MAIN_HEADER_BITNR]    = QCOW2_OPT_OVERLAP_MAIN_HEADER,
682     [QCOW2_OL_ACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L1,
683     [QCOW2_OL_ACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L2,
684     [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
685     [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
686     [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
687     [QCOW2_OL_INACTIVE_L1_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L1,
688     [QCOW2_OL_INACTIVE_L2_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L2,
689 };
690 
691 static void cache_clean_timer_cb(void *opaque)
692 {
693     BlockDriverState *bs = opaque;
694     BDRVQcow2State *s = bs->opaque;
695     qcow2_cache_clean_unused(bs, s->l2_table_cache);
696     qcow2_cache_clean_unused(bs, s->refcount_block_cache);
697     timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
698               (int64_t) s->cache_clean_interval * 1000);
699 }
700 
701 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
702 {
703     BDRVQcow2State *s = bs->opaque;
704     if (s->cache_clean_interval > 0) {
705         s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
706                                              SCALE_MS, cache_clean_timer_cb,
707                                              bs);
708         timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
709                   (int64_t) s->cache_clean_interval * 1000);
710     }
711 }
712 
713 static void cache_clean_timer_del(BlockDriverState *bs)
714 {
715     BDRVQcow2State *s = bs->opaque;
716     if (s->cache_clean_timer) {
717         timer_del(s->cache_clean_timer);
718         timer_free(s->cache_clean_timer);
719         s->cache_clean_timer = NULL;
720     }
721 }
722 
723 static void qcow2_detach_aio_context(BlockDriverState *bs)
724 {
725     cache_clean_timer_del(bs);
726 }
727 
728 static void qcow2_attach_aio_context(BlockDriverState *bs,
729                                      AioContext *new_context)
730 {
731     cache_clean_timer_init(bs, new_context);
732 }
733 
734 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
735                              uint64_t *l2_cache_size,
736                              uint64_t *refcount_cache_size, Error **errp)
737 {
738     BDRVQcow2State *s = bs->opaque;
739     uint64_t combined_cache_size;
740     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
741 
742     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
743     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
744     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
745 
746     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
747     *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0);
748     *refcount_cache_size = qemu_opt_get_size(opts,
749                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
750 
751     if (combined_cache_size_set) {
752         if (l2_cache_size_set && refcount_cache_size_set) {
753             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
754                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
755                        "the same time");
756             return;
757         } else if (*l2_cache_size > combined_cache_size) {
758             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
759                        QCOW2_OPT_CACHE_SIZE);
760             return;
761         } else if (*refcount_cache_size > combined_cache_size) {
762             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
763                        QCOW2_OPT_CACHE_SIZE);
764             return;
765         }
766 
767         if (l2_cache_size_set) {
768             *refcount_cache_size = combined_cache_size - *l2_cache_size;
769         } else if (refcount_cache_size_set) {
770             *l2_cache_size = combined_cache_size - *refcount_cache_size;
771         } else {
772             *refcount_cache_size = combined_cache_size
773                                  / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1);
774             *l2_cache_size = combined_cache_size - *refcount_cache_size;
775         }
776     } else {
777         if (!l2_cache_size_set && !refcount_cache_size_set) {
778             *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE,
779                                  (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
780                                  * s->cluster_size);
781             *refcount_cache_size = *l2_cache_size
782                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
783         } else if (!l2_cache_size_set) {
784             *l2_cache_size = *refcount_cache_size
785                            * DEFAULT_L2_REFCOUNT_SIZE_RATIO;
786         } else if (!refcount_cache_size_set) {
787             *refcount_cache_size = *l2_cache_size
788                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
789         }
790     }
791 }
792 
793 typedef struct Qcow2ReopenState {
794     Qcow2Cache *l2_table_cache;
795     Qcow2Cache *refcount_block_cache;
796     bool use_lazy_refcounts;
797     int overlap_check;
798     bool discard_passthrough[QCOW2_DISCARD_MAX];
799     uint64_t cache_clean_interval;
800     QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
801 } Qcow2ReopenState;
802 
803 static int qcow2_update_options_prepare(BlockDriverState *bs,
804                                         Qcow2ReopenState *r,
805                                         QDict *options, int flags,
806                                         Error **errp)
807 {
808     BDRVQcow2State *s = bs->opaque;
809     QemuOpts *opts = NULL;
810     const char *opt_overlap_check, *opt_overlap_check_template;
811     int overlap_check_template = 0;
812     uint64_t l2_cache_size, refcount_cache_size;
813     int i;
814     const char *encryptfmt;
815     QDict *encryptopts = NULL;
816     Error *local_err = NULL;
817     int ret;
818 
819     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
820     encryptfmt = qdict_get_try_str(encryptopts, "format");
821 
822     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
823     qemu_opts_absorb_qdict(opts, options, &local_err);
824     if (local_err) {
825         error_propagate(errp, local_err);
826         ret = -EINVAL;
827         goto fail;
828     }
829 
830     /* get L2 table/refcount block cache size from command line options */
831     read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size,
832                      &local_err);
833     if (local_err) {
834         error_propagate(errp, local_err);
835         ret = -EINVAL;
836         goto fail;
837     }
838 
839     l2_cache_size /= s->cluster_size;
840     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
841         l2_cache_size = MIN_L2_CACHE_SIZE;
842     }
843     if (l2_cache_size > INT_MAX) {
844         error_setg(errp, "L2 cache size too big");
845         ret = -EINVAL;
846         goto fail;
847     }
848 
849     refcount_cache_size /= s->cluster_size;
850     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
851         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
852     }
853     if (refcount_cache_size > INT_MAX) {
854         error_setg(errp, "Refcount cache size too big");
855         ret = -EINVAL;
856         goto fail;
857     }
858 
859     /* alloc new L2 table/refcount block cache, flush old one */
860     if (s->l2_table_cache) {
861         ret = qcow2_cache_flush(bs, s->l2_table_cache);
862         if (ret) {
863             error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
864             goto fail;
865         }
866     }
867 
868     if (s->refcount_block_cache) {
869         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
870         if (ret) {
871             error_setg_errno(errp, -ret,
872                              "Failed to flush the refcount block cache");
873             goto fail;
874         }
875     }
876 
877     r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size);
878     r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size);
879     if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
880         error_setg(errp, "Could not allocate metadata caches");
881         ret = -ENOMEM;
882         goto fail;
883     }
884 
885     /* New interval for cache cleanup timer */
886     r->cache_clean_interval =
887         qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
888                             s->cache_clean_interval);
889 #ifndef CONFIG_LINUX
890     if (r->cache_clean_interval != 0) {
891         error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL
892                    " not supported on this host");
893         ret = -EINVAL;
894         goto fail;
895     }
896 #endif
897     if (r->cache_clean_interval > UINT_MAX) {
898         error_setg(errp, "Cache clean interval too big");
899         ret = -EINVAL;
900         goto fail;
901     }
902 
903     /* lazy-refcounts; flush if going from enabled to disabled */
904     r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
905         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
906     if (r->use_lazy_refcounts && s->qcow_version < 3) {
907         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
908                    "qemu 1.1 compatibility level");
909         ret = -EINVAL;
910         goto fail;
911     }
912 
913     if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
914         ret = qcow2_mark_clean(bs);
915         if (ret < 0) {
916             error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
917             goto fail;
918         }
919     }
920 
921     /* Overlap check options */
922     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
923     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
924     if (opt_overlap_check_template && opt_overlap_check &&
925         strcmp(opt_overlap_check_template, opt_overlap_check))
926     {
927         error_setg(errp, "Conflicting values for qcow2 options '"
928                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
929                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
930         ret = -EINVAL;
931         goto fail;
932     }
933     if (!opt_overlap_check) {
934         opt_overlap_check = opt_overlap_check_template ?: "cached";
935     }
936 
937     if (!strcmp(opt_overlap_check, "none")) {
938         overlap_check_template = 0;
939     } else if (!strcmp(opt_overlap_check, "constant")) {
940         overlap_check_template = QCOW2_OL_CONSTANT;
941     } else if (!strcmp(opt_overlap_check, "cached")) {
942         overlap_check_template = QCOW2_OL_CACHED;
943     } else if (!strcmp(opt_overlap_check, "all")) {
944         overlap_check_template = QCOW2_OL_ALL;
945     } else {
946         error_setg(errp, "Unsupported value '%s' for qcow2 option "
947                    "'overlap-check'. Allowed are any of the following: "
948                    "none, constant, cached, all", opt_overlap_check);
949         ret = -EINVAL;
950         goto fail;
951     }
952 
953     r->overlap_check = 0;
954     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
955         /* overlap-check defines a template bitmask, but every flag may be
956          * overwritten through the associated boolean option */
957         r->overlap_check |=
958             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
959                               overlap_check_template & (1 << i)) << i;
960     }
961 
962     r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
963     r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
964     r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
965         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
966                           flags & BDRV_O_UNMAP);
967     r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
968         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
969     r->discard_passthrough[QCOW2_DISCARD_OTHER] =
970         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
971 
972     switch (s->crypt_method_header) {
973     case QCOW_CRYPT_NONE:
974         if (encryptfmt) {
975             error_setg(errp, "No encryption in image header, but options "
976                        "specified format '%s'", encryptfmt);
977             ret = -EINVAL;
978             goto fail;
979         }
980         break;
981 
982     case QCOW_CRYPT_AES:
983         if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
984             error_setg(errp,
985                        "Header reported 'aes' encryption format but "
986                        "options specify '%s'", encryptfmt);
987             ret = -EINVAL;
988             goto fail;
989         }
990         qdict_del(encryptopts, "format");
991         r->crypto_opts = block_crypto_open_opts_init(
992             Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp);
993         break;
994 
995     case QCOW_CRYPT_LUKS:
996         if (encryptfmt && !g_str_equal(encryptfmt, "luks")) {
997             error_setg(errp,
998                        "Header reported 'luks' encryption format but "
999                        "options specify '%s'", encryptfmt);
1000             ret = -EINVAL;
1001             goto fail;
1002         }
1003         qdict_del(encryptopts, "format");
1004         r->crypto_opts = block_crypto_open_opts_init(
1005             Q_CRYPTO_BLOCK_FORMAT_LUKS, encryptopts, errp);
1006         break;
1007 
1008     default:
1009         error_setg(errp, "Unsupported encryption method %d",
1010                    s->crypt_method_header);
1011         break;
1012     }
1013     if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) {
1014         ret = -EINVAL;
1015         goto fail;
1016     }
1017 
1018     ret = 0;
1019 fail:
1020     QDECREF(encryptopts);
1021     qemu_opts_del(opts);
1022     opts = NULL;
1023     return ret;
1024 }
1025 
1026 static void qcow2_update_options_commit(BlockDriverState *bs,
1027                                         Qcow2ReopenState *r)
1028 {
1029     BDRVQcow2State *s = bs->opaque;
1030     int i;
1031 
1032     if (s->l2_table_cache) {
1033         qcow2_cache_destroy(bs, s->l2_table_cache);
1034     }
1035     if (s->refcount_block_cache) {
1036         qcow2_cache_destroy(bs, s->refcount_block_cache);
1037     }
1038     s->l2_table_cache = r->l2_table_cache;
1039     s->refcount_block_cache = r->refcount_block_cache;
1040 
1041     s->overlap_check = r->overlap_check;
1042     s->use_lazy_refcounts = r->use_lazy_refcounts;
1043 
1044     for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
1045         s->discard_passthrough[i] = r->discard_passthrough[i];
1046     }
1047 
1048     if (s->cache_clean_interval != r->cache_clean_interval) {
1049         cache_clean_timer_del(bs);
1050         s->cache_clean_interval = r->cache_clean_interval;
1051         cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
1052     }
1053 
1054     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1055     s->crypto_opts = r->crypto_opts;
1056 }
1057 
1058 static void qcow2_update_options_abort(BlockDriverState *bs,
1059                                        Qcow2ReopenState *r)
1060 {
1061     if (r->l2_table_cache) {
1062         qcow2_cache_destroy(bs, r->l2_table_cache);
1063     }
1064     if (r->refcount_block_cache) {
1065         qcow2_cache_destroy(bs, r->refcount_block_cache);
1066     }
1067     qapi_free_QCryptoBlockOpenOptions(r->crypto_opts);
1068 }
1069 
1070 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
1071                                 int flags, Error **errp)
1072 {
1073     Qcow2ReopenState r = {};
1074     int ret;
1075 
1076     ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
1077     if (ret >= 0) {
1078         qcow2_update_options_commit(bs, &r);
1079     } else {
1080         qcow2_update_options_abort(bs, &r);
1081     }
1082 
1083     return ret;
1084 }
1085 
1086 static int qcow2_do_open(BlockDriverState *bs, QDict *options, int flags,
1087                          Error **errp)
1088 {
1089     BDRVQcow2State *s = bs->opaque;
1090     unsigned int len, i;
1091     int ret = 0;
1092     QCowHeader header;
1093     Error *local_err = NULL;
1094     uint64_t ext_end;
1095     uint64_t l1_vm_state_index;
1096     bool update_header = false;
1097 
1098     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
1099     if (ret < 0) {
1100         error_setg_errno(errp, -ret, "Could not read qcow2 header");
1101         goto fail;
1102     }
1103     be32_to_cpus(&header.magic);
1104     be32_to_cpus(&header.version);
1105     be64_to_cpus(&header.backing_file_offset);
1106     be32_to_cpus(&header.backing_file_size);
1107     be64_to_cpus(&header.size);
1108     be32_to_cpus(&header.cluster_bits);
1109     be32_to_cpus(&header.crypt_method);
1110     be64_to_cpus(&header.l1_table_offset);
1111     be32_to_cpus(&header.l1_size);
1112     be64_to_cpus(&header.refcount_table_offset);
1113     be32_to_cpus(&header.refcount_table_clusters);
1114     be64_to_cpus(&header.snapshots_offset);
1115     be32_to_cpus(&header.nb_snapshots);
1116 
1117     if (header.magic != QCOW_MAGIC) {
1118         error_setg(errp, "Image is not in qcow2 format");
1119         ret = -EINVAL;
1120         goto fail;
1121     }
1122     if (header.version < 2 || header.version > 3) {
1123         error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
1124         ret = -ENOTSUP;
1125         goto fail;
1126     }
1127 
1128     s->qcow_version = header.version;
1129 
1130     /* Initialise cluster size */
1131     if (header.cluster_bits < MIN_CLUSTER_BITS ||
1132         header.cluster_bits > MAX_CLUSTER_BITS) {
1133         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
1134                    header.cluster_bits);
1135         ret = -EINVAL;
1136         goto fail;
1137     }
1138 
1139     s->cluster_bits = header.cluster_bits;
1140     s->cluster_size = 1 << s->cluster_bits;
1141     s->cluster_sectors = 1 << (s->cluster_bits - 9);
1142 
1143     /* Initialise version 3 header fields */
1144     if (header.version == 2) {
1145         header.incompatible_features    = 0;
1146         header.compatible_features      = 0;
1147         header.autoclear_features       = 0;
1148         header.refcount_order           = 4;
1149         header.header_length            = 72;
1150     } else {
1151         be64_to_cpus(&header.incompatible_features);
1152         be64_to_cpus(&header.compatible_features);
1153         be64_to_cpus(&header.autoclear_features);
1154         be32_to_cpus(&header.refcount_order);
1155         be32_to_cpus(&header.header_length);
1156 
1157         if (header.header_length < 104) {
1158             error_setg(errp, "qcow2 header too short");
1159             ret = -EINVAL;
1160             goto fail;
1161         }
1162     }
1163 
1164     if (header.header_length > s->cluster_size) {
1165         error_setg(errp, "qcow2 header exceeds cluster size");
1166         ret = -EINVAL;
1167         goto fail;
1168     }
1169 
1170     if (header.header_length > sizeof(header)) {
1171         s->unknown_header_fields_size = header.header_length - sizeof(header);
1172         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
1173         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
1174                          s->unknown_header_fields_size);
1175         if (ret < 0) {
1176             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
1177                              "fields");
1178             goto fail;
1179         }
1180     }
1181 
1182     if (header.backing_file_offset > s->cluster_size) {
1183         error_setg(errp, "Invalid backing file offset");
1184         ret = -EINVAL;
1185         goto fail;
1186     }
1187 
1188     if (header.backing_file_offset) {
1189         ext_end = header.backing_file_offset;
1190     } else {
1191         ext_end = 1 << header.cluster_bits;
1192     }
1193 
1194     /* Handle feature bits */
1195     s->incompatible_features    = header.incompatible_features;
1196     s->compatible_features      = header.compatible_features;
1197     s->autoclear_features       = header.autoclear_features;
1198 
1199     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
1200         void *feature_table = NULL;
1201         qcow2_read_extensions(bs, header.header_length, ext_end,
1202                               &feature_table, flags, NULL, NULL);
1203         report_unsupported_feature(errp, feature_table,
1204                                    s->incompatible_features &
1205                                    ~QCOW2_INCOMPAT_MASK);
1206         ret = -ENOTSUP;
1207         g_free(feature_table);
1208         goto fail;
1209     }
1210 
1211     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
1212         /* Corrupt images may not be written to unless they are being repaired
1213          */
1214         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
1215             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
1216                        "read/write");
1217             ret = -EACCES;
1218             goto fail;
1219         }
1220     }
1221 
1222     /* Check support for various header values */
1223     if (header.refcount_order > 6) {
1224         error_setg(errp, "Reference count entry width too large; may not "
1225                    "exceed 64 bits");
1226         ret = -EINVAL;
1227         goto fail;
1228     }
1229     s->refcount_order = header.refcount_order;
1230     s->refcount_bits = 1 << s->refcount_order;
1231     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1232     s->refcount_max += s->refcount_max - 1;
1233 
1234     s->crypt_method_header = header.crypt_method;
1235     if (s->crypt_method_header) {
1236         if (bdrv_uses_whitelist() &&
1237             s->crypt_method_header == QCOW_CRYPT_AES) {
1238             error_setg(errp,
1239                        "Use of AES-CBC encrypted qcow2 images is no longer "
1240                        "supported in system emulators");
1241             error_append_hint(errp,
1242                               "You can use 'qemu-img convert' to convert your "
1243                               "image to an alternative supported format, such "
1244                               "as unencrypted qcow2, or raw with the LUKS "
1245                               "format instead.\n");
1246             ret = -ENOSYS;
1247             goto fail;
1248         }
1249 
1250         if (s->crypt_method_header == QCOW_CRYPT_AES) {
1251             s->crypt_physical_offset = false;
1252         } else {
1253             /* Assuming LUKS and any future crypt methods we
1254              * add will all use physical offsets, due to the
1255              * fact that the alternative is insecure...  */
1256             s->crypt_physical_offset = true;
1257         }
1258 
1259         bs->encrypted = true;
1260     }
1261 
1262     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
1263     s->l2_size = 1 << s->l2_bits;
1264     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1265     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1266     s->refcount_block_size = 1 << s->refcount_block_bits;
1267     bs->total_sectors = header.size / 512;
1268     s->csize_shift = (62 - (s->cluster_bits - 8));
1269     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1270     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1271 
1272     s->refcount_table_offset = header.refcount_table_offset;
1273     s->refcount_table_size =
1274         header.refcount_table_clusters << (s->cluster_bits - 3);
1275 
1276     if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
1277         error_setg(errp, "Reference count table too large");
1278         ret = -EINVAL;
1279         goto fail;
1280     }
1281 
1282     ret = validate_table_offset(bs, s->refcount_table_offset,
1283                                 s->refcount_table_size, sizeof(uint64_t));
1284     if (ret < 0) {
1285         error_setg(errp, "Invalid reference count table offset");
1286         goto fail;
1287     }
1288 
1289     /* Snapshot table offset/length */
1290     if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
1291         error_setg(errp, "Too many snapshots");
1292         ret = -EINVAL;
1293         goto fail;
1294     }
1295 
1296     ret = validate_table_offset(bs, header.snapshots_offset,
1297                                 header.nb_snapshots,
1298                                 sizeof(QCowSnapshotHeader));
1299     if (ret < 0) {
1300         error_setg(errp, "Invalid snapshot table offset");
1301         goto fail;
1302     }
1303 
1304     /* read the level 1 table */
1305     if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
1306         error_setg(errp, "Active L1 table too large");
1307         ret = -EFBIG;
1308         goto fail;
1309     }
1310     s->l1_size = header.l1_size;
1311 
1312     l1_vm_state_index = size_to_l1(s, header.size);
1313     if (l1_vm_state_index > INT_MAX) {
1314         error_setg(errp, "Image is too big");
1315         ret = -EFBIG;
1316         goto fail;
1317     }
1318     s->l1_vm_state_index = l1_vm_state_index;
1319 
1320     /* the L1 table must contain at least enough entries to put
1321        header.size bytes */
1322     if (s->l1_size < s->l1_vm_state_index) {
1323         error_setg(errp, "L1 table is too small");
1324         ret = -EINVAL;
1325         goto fail;
1326     }
1327 
1328     ret = validate_table_offset(bs, header.l1_table_offset,
1329                                 header.l1_size, sizeof(uint64_t));
1330     if (ret < 0) {
1331         error_setg(errp, "Invalid L1 table offset");
1332         goto fail;
1333     }
1334     s->l1_table_offset = header.l1_table_offset;
1335 
1336 
1337     if (s->l1_size > 0) {
1338         s->l1_table = qemu_try_blockalign(bs->file->bs,
1339             align_offset(s->l1_size * sizeof(uint64_t), 512));
1340         if (s->l1_table == NULL) {
1341             error_setg(errp, "Could not allocate L1 table");
1342             ret = -ENOMEM;
1343             goto fail;
1344         }
1345         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
1346                          s->l1_size * sizeof(uint64_t));
1347         if (ret < 0) {
1348             error_setg_errno(errp, -ret, "Could not read L1 table");
1349             goto fail;
1350         }
1351         for(i = 0;i < s->l1_size; i++) {
1352             be64_to_cpus(&s->l1_table[i]);
1353         }
1354     }
1355 
1356     /* Parse driver-specific options */
1357     ret = qcow2_update_options(bs, options, flags, errp);
1358     if (ret < 0) {
1359         goto fail;
1360     }
1361 
1362     s->cluster_cache_offset = -1;
1363     s->flags = flags;
1364 
1365     ret = qcow2_refcount_init(bs);
1366     if (ret != 0) {
1367         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1368         goto fail;
1369     }
1370 
1371     QLIST_INIT(&s->cluster_allocs);
1372     QTAILQ_INIT(&s->discards);
1373 
1374     /* read qcow2 extensions */
1375     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1376                               flags, &update_header, &local_err)) {
1377         error_propagate(errp, local_err);
1378         ret = -EINVAL;
1379         goto fail;
1380     }
1381 
1382     /* qcow2_read_extension may have set up the crypto context
1383      * if the crypt method needs a header region, some methods
1384      * don't need header extensions, so must check here
1385      */
1386     if (s->crypt_method_header && !s->crypto) {
1387         if (s->crypt_method_header == QCOW_CRYPT_AES) {
1388             unsigned int cflags = 0;
1389             if (flags & BDRV_O_NO_IO) {
1390                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1391             }
1392             s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
1393                                            NULL, NULL, cflags, errp);
1394             if (!s->crypto) {
1395                 ret = -EINVAL;
1396                 goto fail;
1397             }
1398         } else if (!(flags & BDRV_O_NO_IO)) {
1399             error_setg(errp, "Missing CRYPTO header for crypt method %d",
1400                        s->crypt_method_header);
1401             ret = -EINVAL;
1402             goto fail;
1403         }
1404     }
1405 
1406     /* read the backing file name */
1407     if (header.backing_file_offset != 0) {
1408         len = header.backing_file_size;
1409         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1410             len >= sizeof(bs->backing_file)) {
1411             error_setg(errp, "Backing file name too long");
1412             ret = -EINVAL;
1413             goto fail;
1414         }
1415         ret = bdrv_pread(bs->file, header.backing_file_offset,
1416                          bs->backing_file, len);
1417         if (ret < 0) {
1418             error_setg_errno(errp, -ret, "Could not read backing file name");
1419             goto fail;
1420         }
1421         bs->backing_file[len] = '\0';
1422         s->image_backing_file = g_strdup(bs->backing_file);
1423     }
1424 
1425     /* Internal snapshots */
1426     s->snapshots_offset = header.snapshots_offset;
1427     s->nb_snapshots = header.nb_snapshots;
1428 
1429     ret = qcow2_read_snapshots(bs);
1430     if (ret < 0) {
1431         error_setg_errno(errp, -ret, "Could not read snapshots");
1432         goto fail;
1433     }
1434 
1435     /* Clear unknown autoclear feature bits */
1436     update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
1437     update_header =
1438         update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE);
1439     if (update_header) {
1440         s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
1441     }
1442 
1443     if (qcow2_load_autoloading_dirty_bitmaps(bs, &local_err)) {
1444         update_header = false;
1445     }
1446     if (local_err != NULL) {
1447         error_propagate(errp, local_err);
1448         ret = -EINVAL;
1449         goto fail;
1450     }
1451 
1452     if (update_header) {
1453         ret = qcow2_update_header(bs);
1454         if (ret < 0) {
1455             error_setg_errno(errp, -ret, "Could not update qcow2 header");
1456             goto fail;
1457         }
1458     }
1459 
1460     /* Initialise locks */
1461     qemu_co_mutex_init(&s->lock);
1462     bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
1463 
1464     /* Repair image if dirty */
1465     if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1466         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1467         BdrvCheckResult result = {0};
1468 
1469         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1470         if (ret < 0) {
1471             error_setg_errno(errp, -ret, "Could not repair dirty image");
1472             goto fail;
1473         }
1474     }
1475 
1476 #ifdef DEBUG_ALLOC
1477     {
1478         BdrvCheckResult result = {0};
1479         qcow2_check_refcounts(bs, &result, 0);
1480     }
1481 #endif
1482     return ret;
1483 
1484  fail:
1485     g_free(s->unknown_header_fields);
1486     cleanup_unknown_header_ext(bs);
1487     qcow2_free_snapshots(bs);
1488     qcow2_refcount_close(bs);
1489     qemu_vfree(s->l1_table);
1490     /* else pre-write overlap checks in cache_destroy may crash */
1491     s->l1_table = NULL;
1492     cache_clean_timer_del(bs);
1493     if (s->l2_table_cache) {
1494         qcow2_cache_destroy(bs, s->l2_table_cache);
1495     }
1496     if (s->refcount_block_cache) {
1497         qcow2_cache_destroy(bs, s->refcount_block_cache);
1498     }
1499     qcrypto_block_free(s->crypto);
1500     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1501     return ret;
1502 }
1503 
1504 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1505                       Error **errp)
1506 {
1507     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
1508                                false, errp);
1509     if (!bs->file) {
1510         return -EINVAL;
1511     }
1512 
1513     return qcow2_do_open(bs, options, flags, errp);
1514 }
1515 
1516 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1517 {
1518     BDRVQcow2State *s = bs->opaque;
1519 
1520     if (bs->encrypted) {
1521         /* Encryption works on a sector granularity */
1522         bs->bl.request_alignment = BDRV_SECTOR_SIZE;
1523     }
1524     bs->bl.pwrite_zeroes_alignment = s->cluster_size;
1525     bs->bl.pdiscard_alignment = s->cluster_size;
1526 }
1527 
1528 static int qcow2_reopen_prepare(BDRVReopenState *state,
1529                                 BlockReopenQueue *queue, Error **errp)
1530 {
1531     Qcow2ReopenState *r;
1532     int ret;
1533 
1534     r = g_new0(Qcow2ReopenState, 1);
1535     state->opaque = r;
1536 
1537     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1538                                        state->flags, errp);
1539     if (ret < 0) {
1540         goto fail;
1541     }
1542 
1543     /* We need to write out any unwritten data if we reopen read-only. */
1544     if ((state->flags & BDRV_O_RDWR) == 0) {
1545         ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1546         if (ret < 0) {
1547             goto fail;
1548         }
1549 
1550         ret = bdrv_flush(state->bs);
1551         if (ret < 0) {
1552             goto fail;
1553         }
1554 
1555         ret = qcow2_mark_clean(state->bs);
1556         if (ret < 0) {
1557             goto fail;
1558         }
1559     }
1560 
1561     return 0;
1562 
1563 fail:
1564     qcow2_update_options_abort(state->bs, r);
1565     g_free(r);
1566     return ret;
1567 }
1568 
1569 static void qcow2_reopen_commit(BDRVReopenState *state)
1570 {
1571     qcow2_update_options_commit(state->bs, state->opaque);
1572     g_free(state->opaque);
1573 }
1574 
1575 static void qcow2_reopen_abort(BDRVReopenState *state)
1576 {
1577     qcow2_update_options_abort(state->bs, state->opaque);
1578     g_free(state->opaque);
1579 }
1580 
1581 static void qcow2_join_options(QDict *options, QDict *old_options)
1582 {
1583     bool has_new_overlap_template =
1584         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1585         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1586     bool has_new_total_cache_size =
1587         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1588     bool has_all_cache_options;
1589 
1590     /* New overlap template overrides all old overlap options */
1591     if (has_new_overlap_template) {
1592         qdict_del(old_options, QCOW2_OPT_OVERLAP);
1593         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1594         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1595         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1596         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1597         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1598         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1599         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1600         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1601         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1602     }
1603 
1604     /* New total cache size overrides all old options */
1605     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1606         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1607         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1608     }
1609 
1610     qdict_join(options, old_options, false);
1611 
1612     /*
1613      * If after merging all cache size options are set, an old total size is
1614      * overwritten. Do keep all options, however, if all three are new. The
1615      * resulting error message is what we want to happen.
1616      */
1617     has_all_cache_options =
1618         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1619         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1620         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1621 
1622     if (has_all_cache_options && !has_new_total_cache_size) {
1623         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1624     }
1625 }
1626 
1627 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1628         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1629 {
1630     BDRVQcow2State *s = bs->opaque;
1631     uint64_t cluster_offset;
1632     int index_in_cluster, ret;
1633     unsigned int bytes;
1634     int64_t status = 0;
1635 
1636     bytes = MIN(INT_MAX, nb_sectors * BDRV_SECTOR_SIZE);
1637     qemu_co_mutex_lock(&s->lock);
1638     ret = qcow2_get_cluster_offset(bs, sector_num << 9, &bytes,
1639                                    &cluster_offset);
1640     qemu_co_mutex_unlock(&s->lock);
1641     if (ret < 0) {
1642         return ret;
1643     }
1644 
1645     *pnum = bytes >> BDRV_SECTOR_BITS;
1646 
1647     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1648         !s->crypto) {
1649         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1650         cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1651         *file = bs->file->bs;
1652         status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1653     }
1654     if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) {
1655         status |= BDRV_BLOCK_ZERO;
1656     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1657         status |= BDRV_BLOCK_DATA;
1658     }
1659     return status;
1660 }
1661 
1662 /* handle reading after the end of the backing file */
1663 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
1664                         int64_t offset, int bytes)
1665 {
1666     uint64_t bs_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1667     int n1;
1668 
1669     if ((offset + bytes) <= bs_size) {
1670         return bytes;
1671     }
1672 
1673     if (offset >= bs_size) {
1674         n1 = 0;
1675     } else {
1676         n1 = bs_size - offset;
1677     }
1678 
1679     qemu_iovec_memset(qiov, n1, 0, bytes - n1);
1680 
1681     return n1;
1682 }
1683 
1684 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset,
1685                                         uint64_t bytes, QEMUIOVector *qiov,
1686                                         int flags)
1687 {
1688     BDRVQcow2State *s = bs->opaque;
1689     int offset_in_cluster, n1;
1690     int ret;
1691     unsigned int cur_bytes; /* number of bytes in current iteration */
1692     uint64_t cluster_offset = 0;
1693     uint64_t bytes_done = 0;
1694     QEMUIOVector hd_qiov;
1695     uint8_t *cluster_data = NULL;
1696 
1697     qemu_iovec_init(&hd_qiov, qiov->niov);
1698 
1699     qemu_co_mutex_lock(&s->lock);
1700 
1701     while (bytes != 0) {
1702 
1703         /* prepare next request */
1704         cur_bytes = MIN(bytes, INT_MAX);
1705         if (s->crypto) {
1706             cur_bytes = MIN(cur_bytes,
1707                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1708         }
1709 
1710         ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset);
1711         if (ret < 0) {
1712             goto fail;
1713         }
1714 
1715         offset_in_cluster = offset_into_cluster(s, offset);
1716 
1717         qemu_iovec_reset(&hd_qiov);
1718         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
1719 
1720         switch (ret) {
1721         case QCOW2_CLUSTER_UNALLOCATED:
1722 
1723             if (bs->backing) {
1724                 /* read from the base image */
1725                 n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov,
1726                                          offset, cur_bytes);
1727                 if (n1 > 0) {
1728                     QEMUIOVector local_qiov;
1729 
1730                     qemu_iovec_init(&local_qiov, hd_qiov.niov);
1731                     qemu_iovec_concat(&local_qiov, &hd_qiov, 0, n1);
1732 
1733                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1734                     qemu_co_mutex_unlock(&s->lock);
1735                     ret = bdrv_co_preadv(bs->backing, offset, n1,
1736                                          &local_qiov, 0);
1737                     qemu_co_mutex_lock(&s->lock);
1738 
1739                     qemu_iovec_destroy(&local_qiov);
1740 
1741                     if (ret < 0) {
1742                         goto fail;
1743                     }
1744                 }
1745             } else {
1746                 /* Note: in this case, no need to wait */
1747                 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1748             }
1749             break;
1750 
1751         case QCOW2_CLUSTER_ZERO_PLAIN:
1752         case QCOW2_CLUSTER_ZERO_ALLOC:
1753             qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1754             break;
1755 
1756         case QCOW2_CLUSTER_COMPRESSED:
1757             /* add AIO support for compressed blocks ? */
1758             ret = qcow2_decompress_cluster(bs, cluster_offset);
1759             if (ret < 0) {
1760                 goto fail;
1761             }
1762 
1763             qemu_iovec_from_buf(&hd_qiov, 0,
1764                                 s->cluster_cache + offset_in_cluster,
1765                                 cur_bytes);
1766             break;
1767 
1768         case QCOW2_CLUSTER_NORMAL:
1769             if ((cluster_offset & 511) != 0) {
1770                 ret = -EIO;
1771                 goto fail;
1772             }
1773 
1774             if (bs->encrypted) {
1775                 assert(s->crypto);
1776 
1777                 /*
1778                  * For encrypted images, read everything into a temporary
1779                  * contiguous buffer on which the AES functions can work.
1780                  */
1781                 if (!cluster_data) {
1782                     cluster_data =
1783                         qemu_try_blockalign(bs->file->bs,
1784                                             QCOW_MAX_CRYPT_CLUSTERS
1785                                             * s->cluster_size);
1786                     if (cluster_data == NULL) {
1787                         ret = -ENOMEM;
1788                         goto fail;
1789                     }
1790                 }
1791 
1792                 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1793                 qemu_iovec_reset(&hd_qiov);
1794                 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
1795             }
1796 
1797             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1798             qemu_co_mutex_unlock(&s->lock);
1799             ret = bdrv_co_preadv(bs->file,
1800                                  cluster_offset + offset_in_cluster,
1801                                  cur_bytes, &hd_qiov, 0);
1802             qemu_co_mutex_lock(&s->lock);
1803             if (ret < 0) {
1804                 goto fail;
1805             }
1806             if (bs->encrypted) {
1807                 assert(s->crypto);
1808                 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
1809                 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
1810                 if (qcrypto_block_decrypt(s->crypto,
1811                                           (s->crypt_physical_offset ?
1812                                            cluster_offset + offset_in_cluster :
1813                                            offset) >> BDRV_SECTOR_BITS,
1814                                           cluster_data,
1815                                           cur_bytes,
1816                                           NULL) < 0) {
1817                     ret = -EIO;
1818                     goto fail;
1819                 }
1820                 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes);
1821             }
1822             break;
1823 
1824         default:
1825             g_assert_not_reached();
1826             ret = -EIO;
1827             goto fail;
1828         }
1829 
1830         bytes -= cur_bytes;
1831         offset += cur_bytes;
1832         bytes_done += cur_bytes;
1833     }
1834     ret = 0;
1835 
1836 fail:
1837     qemu_co_mutex_unlock(&s->lock);
1838 
1839     qemu_iovec_destroy(&hd_qiov);
1840     qemu_vfree(cluster_data);
1841 
1842     return ret;
1843 }
1844 
1845 /* Check if it's possible to merge a write request with the writing of
1846  * the data from the COW regions */
1847 static bool merge_cow(uint64_t offset, unsigned bytes,
1848                       QEMUIOVector *hd_qiov, QCowL2Meta *l2meta)
1849 {
1850     QCowL2Meta *m;
1851 
1852     for (m = l2meta; m != NULL; m = m->next) {
1853         /* If both COW regions are empty then there's nothing to merge */
1854         if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
1855             continue;
1856         }
1857 
1858         /* The data (middle) region must be immediately after the
1859          * start region */
1860         if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
1861             continue;
1862         }
1863 
1864         /* The end region must be immediately after the data (middle)
1865          * region */
1866         if (m->offset + m->cow_end.offset != offset + bytes) {
1867             continue;
1868         }
1869 
1870         /* Make sure that adding both COW regions to the QEMUIOVector
1871          * does not exceed IOV_MAX */
1872         if (hd_qiov->niov > IOV_MAX - 2) {
1873             continue;
1874         }
1875 
1876         m->data_qiov = hd_qiov;
1877         return true;
1878     }
1879 
1880     return false;
1881 }
1882 
1883 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset,
1884                                          uint64_t bytes, QEMUIOVector *qiov,
1885                                          int flags)
1886 {
1887     BDRVQcow2State *s = bs->opaque;
1888     int offset_in_cluster;
1889     int ret;
1890     unsigned int cur_bytes; /* number of sectors in current iteration */
1891     uint64_t cluster_offset;
1892     QEMUIOVector hd_qiov;
1893     uint64_t bytes_done = 0;
1894     uint8_t *cluster_data = NULL;
1895     QCowL2Meta *l2meta = NULL;
1896 
1897     trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
1898 
1899     qemu_iovec_init(&hd_qiov, qiov->niov);
1900 
1901     s->cluster_cache_offset = -1; /* disable compressed cache */
1902 
1903     qemu_co_mutex_lock(&s->lock);
1904 
1905     while (bytes != 0) {
1906 
1907         l2meta = NULL;
1908 
1909         trace_qcow2_writev_start_part(qemu_coroutine_self());
1910         offset_in_cluster = offset_into_cluster(s, offset);
1911         cur_bytes = MIN(bytes, INT_MAX);
1912         if (bs->encrypted) {
1913             cur_bytes = MIN(cur_bytes,
1914                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
1915                             - offset_in_cluster);
1916         }
1917 
1918         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
1919                                          &cluster_offset, &l2meta);
1920         if (ret < 0) {
1921             goto fail;
1922         }
1923 
1924         assert((cluster_offset & 511) == 0);
1925 
1926         qemu_iovec_reset(&hd_qiov);
1927         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
1928 
1929         if (bs->encrypted) {
1930             assert(s->crypto);
1931             if (!cluster_data) {
1932                 cluster_data = qemu_try_blockalign(bs->file->bs,
1933                                                    QCOW_MAX_CRYPT_CLUSTERS
1934                                                    * s->cluster_size);
1935                 if (cluster_data == NULL) {
1936                     ret = -ENOMEM;
1937                     goto fail;
1938                 }
1939             }
1940 
1941             assert(hd_qiov.size <=
1942                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1943             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1944 
1945             if (qcrypto_block_encrypt(s->crypto,
1946                                       (s->crypt_physical_offset ?
1947                                        cluster_offset + offset_in_cluster :
1948                                        offset) >> BDRV_SECTOR_BITS,
1949                                       cluster_data,
1950                                       cur_bytes, NULL) < 0) {
1951                 ret = -EIO;
1952                 goto fail;
1953             }
1954 
1955             qemu_iovec_reset(&hd_qiov);
1956             qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
1957         }
1958 
1959         ret = qcow2_pre_write_overlap_check(bs, 0,
1960                 cluster_offset + offset_in_cluster, cur_bytes);
1961         if (ret < 0) {
1962             goto fail;
1963         }
1964 
1965         /* If we need to do COW, check if it's possible to merge the
1966          * writing of the guest data together with that of the COW regions.
1967          * If it's not possible (or not necessary) then write the
1968          * guest data now. */
1969         if (!merge_cow(offset, cur_bytes, &hd_qiov, l2meta)) {
1970             qemu_co_mutex_unlock(&s->lock);
1971             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1972             trace_qcow2_writev_data(qemu_coroutine_self(),
1973                                     cluster_offset + offset_in_cluster);
1974             ret = bdrv_co_pwritev(bs->file,
1975                                   cluster_offset + offset_in_cluster,
1976                                   cur_bytes, &hd_qiov, 0);
1977             qemu_co_mutex_lock(&s->lock);
1978             if (ret < 0) {
1979                 goto fail;
1980             }
1981         }
1982 
1983         while (l2meta != NULL) {
1984             QCowL2Meta *next;
1985 
1986             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1987             if (ret < 0) {
1988                 goto fail;
1989             }
1990 
1991             /* Take the request off the list of running requests */
1992             if (l2meta->nb_clusters != 0) {
1993                 QLIST_REMOVE(l2meta, next_in_flight);
1994             }
1995 
1996             qemu_co_queue_restart_all(&l2meta->dependent_requests);
1997 
1998             next = l2meta->next;
1999             g_free(l2meta);
2000             l2meta = next;
2001         }
2002 
2003         bytes -= cur_bytes;
2004         offset += cur_bytes;
2005         bytes_done += cur_bytes;
2006         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2007     }
2008     ret = 0;
2009 
2010 fail:
2011     while (l2meta != NULL) {
2012         QCowL2Meta *next;
2013 
2014         if (l2meta->nb_clusters != 0) {
2015             QLIST_REMOVE(l2meta, next_in_flight);
2016         }
2017         qemu_co_queue_restart_all(&l2meta->dependent_requests);
2018 
2019         next = l2meta->next;
2020         g_free(l2meta);
2021         l2meta = next;
2022     }
2023 
2024     qemu_co_mutex_unlock(&s->lock);
2025 
2026     qemu_iovec_destroy(&hd_qiov);
2027     qemu_vfree(cluster_data);
2028     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2029 
2030     return ret;
2031 }
2032 
2033 static int qcow2_inactivate(BlockDriverState *bs)
2034 {
2035     BDRVQcow2State *s = bs->opaque;
2036     int ret, result = 0;
2037     Error *local_err = NULL;
2038 
2039     qcow2_store_persistent_dirty_bitmaps(bs, &local_err);
2040     if (local_err != NULL) {
2041         result = -EINVAL;
2042         error_report_err(local_err);
2043         error_report("Persistent bitmaps are lost for node '%s'",
2044                      bdrv_get_device_or_node_name(bs));
2045     }
2046 
2047     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2048     if (ret) {
2049         result = ret;
2050         error_report("Failed to flush the L2 table cache: %s",
2051                      strerror(-ret));
2052     }
2053 
2054     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2055     if (ret) {
2056         result = ret;
2057         error_report("Failed to flush the refcount block cache: %s",
2058                      strerror(-ret));
2059     }
2060 
2061     if (result == 0) {
2062         qcow2_mark_clean(bs);
2063     }
2064 
2065     return result;
2066 }
2067 
2068 static void qcow2_close(BlockDriverState *bs)
2069 {
2070     BDRVQcow2State *s = bs->opaque;
2071     qemu_vfree(s->l1_table);
2072     /* else pre-write overlap checks in cache_destroy may crash */
2073     s->l1_table = NULL;
2074 
2075     if (!(s->flags & BDRV_O_INACTIVE)) {
2076         qcow2_inactivate(bs);
2077     }
2078 
2079     cache_clean_timer_del(bs);
2080     qcow2_cache_destroy(bs, s->l2_table_cache);
2081     qcow2_cache_destroy(bs, s->refcount_block_cache);
2082 
2083     qcrypto_block_free(s->crypto);
2084     s->crypto = NULL;
2085 
2086     g_free(s->unknown_header_fields);
2087     cleanup_unknown_header_ext(bs);
2088 
2089     g_free(s->image_backing_file);
2090     g_free(s->image_backing_format);
2091 
2092     g_free(s->cluster_cache);
2093     qemu_vfree(s->cluster_data);
2094     qcow2_refcount_close(bs);
2095     qcow2_free_snapshots(bs);
2096 }
2097 
2098 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
2099 {
2100     BDRVQcow2State *s = bs->opaque;
2101     int flags = s->flags;
2102     QCryptoBlock *crypto = NULL;
2103     QDict *options;
2104     Error *local_err = NULL;
2105     int ret;
2106 
2107     /*
2108      * Backing files are read-only which makes all of their metadata immutable,
2109      * that means we don't have to worry about reopening them here.
2110      */
2111 
2112     crypto = s->crypto;
2113     s->crypto = NULL;
2114 
2115     qcow2_close(bs);
2116 
2117     memset(s, 0, sizeof(BDRVQcow2State));
2118     options = qdict_clone_shallow(bs->options);
2119 
2120     flags &= ~BDRV_O_INACTIVE;
2121     ret = qcow2_do_open(bs, options, flags, &local_err);
2122     QDECREF(options);
2123     if (local_err) {
2124         error_propagate(errp, local_err);
2125         error_prepend(errp, "Could not reopen qcow2 layer: ");
2126         bs->drv = NULL;
2127         return;
2128     } else if (ret < 0) {
2129         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
2130         bs->drv = NULL;
2131         return;
2132     }
2133 
2134     s->crypto = crypto;
2135 }
2136 
2137 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2138     size_t len, size_t buflen)
2139 {
2140     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2141     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2142 
2143     if (buflen < ext_len) {
2144         return -ENOSPC;
2145     }
2146 
2147     *ext_backing_fmt = (QCowExtension) {
2148         .magic  = cpu_to_be32(magic),
2149         .len    = cpu_to_be32(len),
2150     };
2151 
2152     if (len) {
2153         memcpy(buf + sizeof(QCowExtension), s, len);
2154     }
2155 
2156     return ext_len;
2157 }
2158 
2159 /*
2160  * Updates the qcow2 header, including the variable length parts of it, i.e.
2161  * the backing file name and all extensions. qcow2 was not designed to allow
2162  * such changes, so if we run out of space (we can only use the first cluster)
2163  * this function may fail.
2164  *
2165  * Returns 0 on success, -errno in error cases.
2166  */
2167 int qcow2_update_header(BlockDriverState *bs)
2168 {
2169     BDRVQcow2State *s = bs->opaque;
2170     QCowHeader *header;
2171     char *buf;
2172     size_t buflen = s->cluster_size;
2173     int ret;
2174     uint64_t total_size;
2175     uint32_t refcount_table_clusters;
2176     size_t header_length;
2177     Qcow2UnknownHeaderExtension *uext;
2178 
2179     buf = qemu_blockalign(bs, buflen);
2180 
2181     /* Header structure */
2182     header = (QCowHeader*) buf;
2183 
2184     if (buflen < sizeof(*header)) {
2185         ret = -ENOSPC;
2186         goto fail;
2187     }
2188 
2189     header_length = sizeof(*header) + s->unknown_header_fields_size;
2190     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2191     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2192 
2193     *header = (QCowHeader) {
2194         /* Version 2 fields */
2195         .magic                  = cpu_to_be32(QCOW_MAGIC),
2196         .version                = cpu_to_be32(s->qcow_version),
2197         .backing_file_offset    = 0,
2198         .backing_file_size      = 0,
2199         .cluster_bits           = cpu_to_be32(s->cluster_bits),
2200         .size                   = cpu_to_be64(total_size),
2201         .crypt_method           = cpu_to_be32(s->crypt_method_header),
2202         .l1_size                = cpu_to_be32(s->l1_size),
2203         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
2204         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
2205         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2206         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
2207         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
2208 
2209         /* Version 3 fields */
2210         .incompatible_features  = cpu_to_be64(s->incompatible_features),
2211         .compatible_features    = cpu_to_be64(s->compatible_features),
2212         .autoclear_features     = cpu_to_be64(s->autoclear_features),
2213         .refcount_order         = cpu_to_be32(s->refcount_order),
2214         .header_length          = cpu_to_be32(header_length),
2215     };
2216 
2217     /* For older versions, write a shorter header */
2218     switch (s->qcow_version) {
2219     case 2:
2220         ret = offsetof(QCowHeader, incompatible_features);
2221         break;
2222     case 3:
2223         ret = sizeof(*header);
2224         break;
2225     default:
2226         ret = -EINVAL;
2227         goto fail;
2228     }
2229 
2230     buf += ret;
2231     buflen -= ret;
2232     memset(buf, 0, buflen);
2233 
2234     /* Preserve any unknown field in the header */
2235     if (s->unknown_header_fields_size) {
2236         if (buflen < s->unknown_header_fields_size) {
2237             ret = -ENOSPC;
2238             goto fail;
2239         }
2240 
2241         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2242         buf += s->unknown_header_fields_size;
2243         buflen -= s->unknown_header_fields_size;
2244     }
2245 
2246     /* Backing file format header extension */
2247     if (s->image_backing_format) {
2248         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
2249                              s->image_backing_format,
2250                              strlen(s->image_backing_format),
2251                              buflen);
2252         if (ret < 0) {
2253             goto fail;
2254         }
2255 
2256         buf += ret;
2257         buflen -= ret;
2258     }
2259 
2260     /* Full disk encryption header pointer extension */
2261     if (s->crypto_header.offset != 0) {
2262         cpu_to_be64s(&s->crypto_header.offset);
2263         cpu_to_be64s(&s->crypto_header.length);
2264         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
2265                              &s->crypto_header, sizeof(s->crypto_header),
2266                              buflen);
2267         be64_to_cpus(&s->crypto_header.offset);
2268         be64_to_cpus(&s->crypto_header.length);
2269         if (ret < 0) {
2270             goto fail;
2271         }
2272         buf += ret;
2273         buflen -= ret;
2274     }
2275 
2276     /* Feature table */
2277     if (s->qcow_version >= 3) {
2278         Qcow2Feature features[] = {
2279             {
2280                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2281                 .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
2282                 .name = "dirty bit",
2283             },
2284             {
2285                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2286                 .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
2287                 .name = "corrupt bit",
2288             },
2289             {
2290                 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
2291                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
2292                 .name = "lazy refcounts",
2293             },
2294         };
2295 
2296         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
2297                              features, sizeof(features), buflen);
2298         if (ret < 0) {
2299             goto fail;
2300         }
2301         buf += ret;
2302         buflen -= ret;
2303     }
2304 
2305     /* Bitmap extension */
2306     if (s->nb_bitmaps > 0) {
2307         Qcow2BitmapHeaderExt bitmaps_header = {
2308             .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
2309             .bitmap_directory_size =
2310                     cpu_to_be64(s->bitmap_directory_size),
2311             .bitmap_directory_offset =
2312                     cpu_to_be64(s->bitmap_directory_offset)
2313         };
2314         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
2315                              &bitmaps_header, sizeof(bitmaps_header),
2316                              buflen);
2317         if (ret < 0) {
2318             goto fail;
2319         }
2320         buf += ret;
2321         buflen -= ret;
2322     }
2323 
2324     /* Keep unknown header extensions */
2325     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
2326         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
2327         if (ret < 0) {
2328             goto fail;
2329         }
2330 
2331         buf += ret;
2332         buflen -= ret;
2333     }
2334 
2335     /* End of header extensions */
2336     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
2337     if (ret < 0) {
2338         goto fail;
2339     }
2340 
2341     buf += ret;
2342     buflen -= ret;
2343 
2344     /* Backing file name */
2345     if (s->image_backing_file) {
2346         size_t backing_file_len = strlen(s->image_backing_file);
2347 
2348         if (buflen < backing_file_len) {
2349             ret = -ENOSPC;
2350             goto fail;
2351         }
2352 
2353         /* Using strncpy is ok here, since buf is not NUL-terminated. */
2354         strncpy(buf, s->image_backing_file, buflen);
2355 
2356         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
2357         header->backing_file_size   = cpu_to_be32(backing_file_len);
2358     }
2359 
2360     /* Write the new header */
2361     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
2362     if (ret < 0) {
2363         goto fail;
2364     }
2365 
2366     ret = 0;
2367 fail:
2368     qemu_vfree(header);
2369     return ret;
2370 }
2371 
2372 static int qcow2_change_backing_file(BlockDriverState *bs,
2373     const char *backing_file, const char *backing_fmt)
2374 {
2375     BDRVQcow2State *s = bs->opaque;
2376 
2377     if (backing_file && strlen(backing_file) > 1023) {
2378         return -EINVAL;
2379     }
2380 
2381     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2382     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2383 
2384     g_free(s->image_backing_file);
2385     g_free(s->image_backing_format);
2386 
2387     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
2388     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
2389 
2390     return qcow2_update_header(bs);
2391 }
2392 
2393 static int qcow2_crypt_method_from_format(const char *encryptfmt)
2394 {
2395     if (g_str_equal(encryptfmt, "luks")) {
2396         return QCOW_CRYPT_LUKS;
2397     } else if (g_str_equal(encryptfmt, "aes")) {
2398         return QCOW_CRYPT_AES;
2399     } else {
2400         return -EINVAL;
2401     }
2402 }
2403 
2404 static int qcow2_set_up_encryption(BlockDriverState *bs, const char *encryptfmt,
2405                                    QemuOpts *opts, Error **errp)
2406 {
2407     BDRVQcow2State *s = bs->opaque;
2408     QCryptoBlockCreateOptions *cryptoopts = NULL;
2409     QCryptoBlock *crypto = NULL;
2410     int ret = -EINVAL;
2411     QDict *options, *encryptopts;
2412     int fmt;
2413 
2414     options = qemu_opts_to_qdict(opts, NULL);
2415     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
2416     QDECREF(options);
2417 
2418     fmt = qcow2_crypt_method_from_format(encryptfmt);
2419 
2420     switch (fmt) {
2421     case QCOW_CRYPT_LUKS:
2422         cryptoopts = block_crypto_create_opts_init(
2423             Q_CRYPTO_BLOCK_FORMAT_LUKS, encryptopts, errp);
2424         break;
2425     case QCOW_CRYPT_AES:
2426         cryptoopts = block_crypto_create_opts_init(
2427             Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp);
2428         break;
2429     default:
2430         error_setg(errp, "Unknown encryption format '%s'", encryptfmt);
2431         break;
2432     }
2433     if (!cryptoopts) {
2434         ret = -EINVAL;
2435         goto out;
2436     }
2437     s->crypt_method_header = fmt;
2438 
2439     crypto = qcrypto_block_create(cryptoopts, "encrypt.",
2440                                   qcow2_crypto_hdr_init_func,
2441                                   qcow2_crypto_hdr_write_func,
2442                                   bs, errp);
2443     if (!crypto) {
2444         ret = -EINVAL;
2445         goto out;
2446     }
2447 
2448     ret = qcow2_update_header(bs);
2449     if (ret < 0) {
2450         error_setg_errno(errp, -ret, "Could not write encryption header");
2451         goto out;
2452     }
2453 
2454  out:
2455     QDECREF(encryptopts);
2456     qcrypto_block_free(crypto);
2457     qapi_free_QCryptoBlockCreateOptions(cryptoopts);
2458     return ret;
2459 }
2460 
2461 
2462 /**
2463  * Preallocates metadata structures for data clusters between @offset (in the
2464  * guest disk) and @new_length (which is thus generally the new guest disk
2465  * size).
2466  *
2467  * Returns: 0 on success, -errno on failure.
2468  */
2469 static int preallocate(BlockDriverState *bs,
2470                        uint64_t offset, uint64_t new_length)
2471 {
2472     BDRVQcow2State *s = bs->opaque;
2473     uint64_t bytes;
2474     uint64_t host_offset = 0;
2475     unsigned int cur_bytes;
2476     int ret;
2477     QCowL2Meta *meta;
2478 
2479     if (qemu_in_coroutine()) {
2480         qemu_co_mutex_lock(&s->lock);
2481     }
2482 
2483     assert(offset <= new_length);
2484     bytes = new_length - offset;
2485 
2486     while (bytes) {
2487         cur_bytes = MIN(bytes, INT_MAX);
2488         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2489                                          &host_offset, &meta);
2490         if (ret < 0) {
2491             goto done;
2492         }
2493 
2494         while (meta) {
2495             QCowL2Meta *next = meta->next;
2496 
2497             ret = qcow2_alloc_cluster_link_l2(bs, meta);
2498             if (ret < 0) {
2499                 qcow2_free_any_clusters(bs, meta->alloc_offset,
2500                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
2501                 goto done;
2502             }
2503 
2504             /* There are no dependent requests, but we need to remove our
2505              * request from the list of in-flight requests */
2506             QLIST_REMOVE(meta, next_in_flight);
2507 
2508             g_free(meta);
2509             meta = next;
2510         }
2511 
2512         /* TODO Preallocate data if requested */
2513 
2514         bytes -= cur_bytes;
2515         offset += cur_bytes;
2516     }
2517 
2518     /*
2519      * It is expected that the image file is large enough to actually contain
2520      * all of the allocated clusters (otherwise we get failing reads after
2521      * EOF). Extend the image to the last allocated sector.
2522      */
2523     if (host_offset != 0) {
2524         uint8_t data = 0;
2525         ret = bdrv_pwrite(bs->file, (host_offset + cur_bytes) - 1,
2526                           &data, 1);
2527         if (ret < 0) {
2528             goto done;
2529         }
2530     }
2531 
2532     ret = 0;
2533 
2534 done:
2535     if (qemu_in_coroutine()) {
2536         qemu_co_mutex_unlock(&s->lock);
2537     }
2538     return ret;
2539 }
2540 
2541 /* qcow2_refcount_metadata_size:
2542  * @clusters: number of clusters to refcount (including data and L1/L2 tables)
2543  * @cluster_size: size of a cluster, in bytes
2544  * @refcount_order: refcount bits power-of-2 exponent
2545  * @generous_increase: allow for the refcount table to be 1.5x as large as it
2546  *                     needs to be
2547  *
2548  * Returns: Number of bytes required for refcount blocks and table metadata.
2549  */
2550 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
2551                                      int refcount_order, bool generous_increase,
2552                                      uint64_t *refblock_count)
2553 {
2554     /*
2555      * Every host cluster is reference-counted, including metadata (even
2556      * refcount metadata is recursively included).
2557      *
2558      * An accurate formula for the size of refcount metadata size is difficult
2559      * to derive.  An easier method of calculation is finding the fixed point
2560      * where no further refcount blocks or table clusters are required to
2561      * reference count every cluster.
2562      */
2563     int64_t blocks_per_table_cluster = cluster_size / sizeof(uint64_t);
2564     int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
2565     int64_t table = 0;  /* number of refcount table clusters */
2566     int64_t blocks = 0; /* number of refcount block clusters */
2567     int64_t last;
2568     int64_t n = 0;
2569 
2570     do {
2571         last = n;
2572         blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
2573         table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
2574         n = clusters + blocks + table;
2575 
2576         if (n == last && generous_increase) {
2577             clusters += DIV_ROUND_UP(table, 2);
2578             n = 0; /* force another loop */
2579             generous_increase = false;
2580         }
2581     } while (n != last);
2582 
2583     if (refblock_count) {
2584         *refblock_count = blocks;
2585     }
2586 
2587     return (blocks + table) * cluster_size;
2588 }
2589 
2590 /**
2591  * qcow2_calc_prealloc_size:
2592  * @total_size: virtual disk size in bytes
2593  * @cluster_size: cluster size in bytes
2594  * @refcount_order: refcount bits power-of-2 exponent
2595  *
2596  * Returns: Total number of bytes required for the fully allocated image
2597  * (including metadata).
2598  */
2599 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
2600                                         size_t cluster_size,
2601                                         int refcount_order)
2602 {
2603     int64_t meta_size = 0;
2604     uint64_t nl1e, nl2e;
2605     int64_t aligned_total_size = align_offset(total_size, cluster_size);
2606 
2607     /* header: 1 cluster */
2608     meta_size += cluster_size;
2609 
2610     /* total size of L2 tables */
2611     nl2e = aligned_total_size / cluster_size;
2612     nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
2613     meta_size += nl2e * sizeof(uint64_t);
2614 
2615     /* total size of L1 tables */
2616     nl1e = nl2e * sizeof(uint64_t) / cluster_size;
2617     nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
2618     meta_size += nl1e * sizeof(uint64_t);
2619 
2620     /* total size of refcount table and blocks */
2621     meta_size += qcow2_refcount_metadata_size(
2622             (meta_size + aligned_total_size) / cluster_size,
2623             cluster_size, refcount_order, false, NULL);
2624 
2625     return meta_size + aligned_total_size;
2626 }
2627 
2628 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, Error **errp)
2629 {
2630     size_t cluster_size;
2631     int cluster_bits;
2632 
2633     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2634                                          DEFAULT_CLUSTER_SIZE);
2635     cluster_bits = ctz32(cluster_size);
2636     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
2637         (1 << cluster_bits) != cluster_size)
2638     {
2639         error_setg(errp, "Cluster size must be a power of two between %d and "
2640                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
2641         return 0;
2642     }
2643     return cluster_size;
2644 }
2645 
2646 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
2647 {
2648     char *buf;
2649     int ret;
2650 
2651     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2652     if (!buf) {
2653         ret = 3; /* default */
2654     } else if (!strcmp(buf, "0.10")) {
2655         ret = 2;
2656     } else if (!strcmp(buf, "1.1")) {
2657         ret = 3;
2658     } else {
2659         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2660         ret = -EINVAL;
2661     }
2662     g_free(buf);
2663     return ret;
2664 }
2665 
2666 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
2667                                                 Error **errp)
2668 {
2669     uint64_t refcount_bits;
2670 
2671     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
2672     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2673         error_setg(errp, "Refcount width must be a power of two and may not "
2674                    "exceed 64 bits");
2675         return 0;
2676     }
2677 
2678     if (version < 3 && refcount_bits != 16) {
2679         error_setg(errp, "Different refcount widths than 16 bits require "
2680                    "compatibility level 1.1 or above (use compat=1.1 or "
2681                    "greater)");
2682         return 0;
2683     }
2684 
2685     return refcount_bits;
2686 }
2687 
2688 static int qcow2_create2(const char *filename, int64_t total_size,
2689                          const char *backing_file, const char *backing_format,
2690                          int flags, size_t cluster_size, PreallocMode prealloc,
2691                          QemuOpts *opts, int version, int refcount_order,
2692                          const char *encryptfmt, Error **errp)
2693 {
2694     QDict *options;
2695 
2696     /*
2697      * Open the image file and write a minimal qcow2 header.
2698      *
2699      * We keep things simple and start with a zero-sized image. We also
2700      * do without refcount blocks or a L1 table for now. We'll fix the
2701      * inconsistency later.
2702      *
2703      * We do need a refcount table because growing the refcount table means
2704      * allocating two new refcount blocks - the seconds of which would be at
2705      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2706      * size for any qcow2 image.
2707      */
2708     BlockBackend *blk;
2709     QCowHeader *header;
2710     uint64_t* refcount_table;
2711     Error *local_err = NULL;
2712     int ret;
2713 
2714     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
2715         int64_t prealloc_size =
2716             qcow2_calc_prealloc_size(total_size, cluster_size, refcount_order);
2717         qemu_opt_set_number(opts, BLOCK_OPT_SIZE, prealloc_size, &error_abort);
2718         qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_str(prealloc),
2719                      &error_abort);
2720     }
2721 
2722     ret = bdrv_create_file(filename, opts, &local_err);
2723     if (ret < 0) {
2724         error_propagate(errp, local_err);
2725         return ret;
2726     }
2727 
2728     blk = blk_new_open(filename, NULL, NULL,
2729                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2730                        &local_err);
2731     if (blk == NULL) {
2732         error_propagate(errp, local_err);
2733         return -EIO;
2734     }
2735 
2736     blk_set_allow_write_beyond_eof(blk, true);
2737 
2738     /* Write the header */
2739     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
2740     header = g_malloc0(cluster_size);
2741     *header = (QCowHeader) {
2742         .magic                      = cpu_to_be32(QCOW_MAGIC),
2743         .version                    = cpu_to_be32(version),
2744         .cluster_bits               = cpu_to_be32(ctz32(cluster_size)),
2745         .size                       = cpu_to_be64(0),
2746         .l1_table_offset            = cpu_to_be64(0),
2747         .l1_size                    = cpu_to_be32(0),
2748         .refcount_table_offset      = cpu_to_be64(cluster_size),
2749         .refcount_table_clusters    = cpu_to_be32(1),
2750         .refcount_order             = cpu_to_be32(refcount_order),
2751         .header_length              = cpu_to_be32(sizeof(*header)),
2752     };
2753 
2754     /* We'll update this to correct value later */
2755     header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
2756 
2757     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
2758         header->compatible_features |=
2759             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
2760     }
2761 
2762     ret = blk_pwrite(blk, 0, header, cluster_size, 0);
2763     g_free(header);
2764     if (ret < 0) {
2765         error_setg_errno(errp, -ret, "Could not write qcow2 header");
2766         goto out;
2767     }
2768 
2769     /* Write a refcount table with one refcount block */
2770     refcount_table = g_malloc0(2 * cluster_size);
2771     refcount_table[0] = cpu_to_be64(2 * cluster_size);
2772     ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
2773     g_free(refcount_table);
2774 
2775     if (ret < 0) {
2776         error_setg_errno(errp, -ret, "Could not write refcount table");
2777         goto out;
2778     }
2779 
2780     blk_unref(blk);
2781     blk = NULL;
2782 
2783     /*
2784      * And now open the image and make it consistent first (i.e. increase the
2785      * refcount of the cluster that is occupied by the header and the refcount
2786      * table)
2787      */
2788     options = qdict_new();
2789     qdict_put_str(options, "driver", "qcow2");
2790     blk = blk_new_open(filename, NULL, options,
2791                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
2792                        &local_err);
2793     if (blk == NULL) {
2794         error_propagate(errp, local_err);
2795         ret = -EIO;
2796         goto out;
2797     }
2798 
2799     ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
2800     if (ret < 0) {
2801         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
2802                          "header and refcount table");
2803         goto out;
2804 
2805     } else if (ret != 0) {
2806         error_report("Huh, first cluster in empty image is already in use?");
2807         abort();
2808     }
2809 
2810     /* Create a full header (including things like feature table) */
2811     ret = qcow2_update_header(blk_bs(blk));
2812     if (ret < 0) {
2813         error_setg_errno(errp, -ret, "Could not update qcow2 header");
2814         goto out;
2815     }
2816 
2817     /* Okay, now that we have a valid image, let's give it the right size */
2818     ret = blk_truncate(blk, total_size, PREALLOC_MODE_OFF, errp);
2819     if (ret < 0) {
2820         error_prepend(errp, "Could not resize image: ");
2821         goto out;
2822     }
2823 
2824     /* Want a backing file? There you go.*/
2825     if (backing_file) {
2826         ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format);
2827         if (ret < 0) {
2828             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
2829                              "with format '%s'", backing_file, backing_format);
2830             goto out;
2831         }
2832     }
2833 
2834     /* Want encryption? There you go. */
2835     if (encryptfmt) {
2836         ret = qcow2_set_up_encryption(blk_bs(blk), encryptfmt, opts, errp);
2837         if (ret < 0) {
2838             goto out;
2839         }
2840     }
2841 
2842     /* And if we're supposed to preallocate metadata, do that now */
2843     if (prealloc != PREALLOC_MODE_OFF) {
2844         ret = preallocate(blk_bs(blk), 0, total_size);
2845         if (ret < 0) {
2846             error_setg_errno(errp, -ret, "Could not preallocate metadata");
2847             goto out;
2848         }
2849     }
2850 
2851     blk_unref(blk);
2852     blk = NULL;
2853 
2854     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
2855      * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
2856      * have to setup decryption context. We're not doing any I/O on the top
2857      * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
2858      * not have effect.
2859      */
2860     options = qdict_new();
2861     qdict_put_str(options, "driver", "qcow2");
2862     blk = blk_new_open(filename, NULL, options,
2863                        BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO,
2864                        &local_err);
2865     if (blk == NULL) {
2866         error_propagate(errp, local_err);
2867         ret = -EIO;
2868         goto out;
2869     }
2870 
2871     ret = 0;
2872 out:
2873     if (blk) {
2874         blk_unref(blk);
2875     }
2876     return ret;
2877 }
2878 
2879 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
2880 {
2881     char *backing_file = NULL;
2882     char *backing_fmt = NULL;
2883     char *buf = NULL;
2884     uint64_t size = 0;
2885     int flags = 0;
2886     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
2887     PreallocMode prealloc;
2888     int version;
2889     uint64_t refcount_bits;
2890     int refcount_order;
2891     char *encryptfmt = NULL;
2892     Error *local_err = NULL;
2893     int ret;
2894 
2895     /* Read out options */
2896     size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2897                     BDRV_SECTOR_SIZE);
2898     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2899     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2900     encryptfmt = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
2901     if (encryptfmt) {
2902         if (qemu_opt_get(opts, BLOCK_OPT_ENCRYPT)) {
2903             error_setg(errp, "Options " BLOCK_OPT_ENCRYPT " and "
2904                        BLOCK_OPT_ENCRYPT_FORMAT " are mutually exclusive");
2905             ret = -EINVAL;
2906             goto finish;
2907         }
2908     } else if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2909         encryptfmt = g_strdup("aes");
2910     }
2911     cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err);
2912     if (local_err) {
2913         error_propagate(errp, local_err);
2914         ret = -EINVAL;
2915         goto finish;
2916     }
2917     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2918     prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
2919                                PREALLOC_MODE_OFF, &local_err);
2920     if (local_err) {
2921         error_propagate(errp, local_err);
2922         ret = -EINVAL;
2923         goto finish;
2924     }
2925 
2926     version = qcow2_opt_get_version_del(opts, &local_err);
2927     if (local_err) {
2928         error_propagate(errp, local_err);
2929         ret = -EINVAL;
2930         goto finish;
2931     }
2932 
2933     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2934         flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2935     }
2936 
2937     if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2938         error_setg(errp, "Backing file and preallocation cannot be used at "
2939                    "the same time");
2940         ret = -EINVAL;
2941         goto finish;
2942     }
2943 
2944     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2945         error_setg(errp, "Lazy refcounts only supported with compatibility "
2946                    "level 1.1 and above (use compat=1.1 or greater)");
2947         ret = -EINVAL;
2948         goto finish;
2949     }
2950 
2951     refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
2952     if (local_err) {
2953         error_propagate(errp, local_err);
2954         ret = -EINVAL;
2955         goto finish;
2956     }
2957 
2958     refcount_order = ctz32(refcount_bits);
2959 
2960     ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2961                         cluster_size, prealloc, opts, version, refcount_order,
2962                         encryptfmt, &local_err);
2963     error_propagate(errp, local_err);
2964 
2965 finish:
2966     g_free(backing_file);
2967     g_free(backing_fmt);
2968     g_free(encryptfmt);
2969     g_free(buf);
2970     return ret;
2971 }
2972 
2973 
2974 static bool is_zero_sectors(BlockDriverState *bs, int64_t start,
2975                             uint32_t count)
2976 {
2977     int nr;
2978     BlockDriverState *file;
2979     int64_t res;
2980 
2981     if (start + count > bs->total_sectors) {
2982         count = bs->total_sectors - start;
2983     }
2984 
2985     if (!count) {
2986         return true;
2987     }
2988     res = bdrv_get_block_status_above(bs, NULL, start, count,
2989                                       &nr, &file);
2990     return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == count;
2991 }
2992 
2993 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
2994     int64_t offset, int bytes, BdrvRequestFlags flags)
2995 {
2996     int ret;
2997     BDRVQcow2State *s = bs->opaque;
2998 
2999     uint32_t head = offset % s->cluster_size;
3000     uint32_t tail = (offset + bytes) % s->cluster_size;
3001 
3002     trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes);
3003     if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) {
3004         tail = 0;
3005     }
3006 
3007     if (head || tail) {
3008         int64_t cl_start = (offset - head) >> BDRV_SECTOR_BITS;
3009         uint64_t off;
3010         unsigned int nr;
3011 
3012         assert(head + bytes <= s->cluster_size);
3013 
3014         /* check whether remainder of cluster already reads as zero */
3015         if (!(is_zero_sectors(bs, cl_start,
3016                               DIV_ROUND_UP(head, BDRV_SECTOR_SIZE)) &&
3017               is_zero_sectors(bs, (offset + bytes) >> BDRV_SECTOR_BITS,
3018                               DIV_ROUND_UP(-tail & (s->cluster_size - 1),
3019                                            BDRV_SECTOR_SIZE)))) {
3020             return -ENOTSUP;
3021         }
3022 
3023         qemu_co_mutex_lock(&s->lock);
3024         /* We can have new write after previous check */
3025         offset = cl_start << BDRV_SECTOR_BITS;
3026         bytes = s->cluster_size;
3027         nr = s->cluster_size;
3028         ret = qcow2_get_cluster_offset(bs, offset, &nr, &off);
3029         if (ret != QCOW2_CLUSTER_UNALLOCATED &&
3030             ret != QCOW2_CLUSTER_ZERO_PLAIN &&
3031             ret != QCOW2_CLUSTER_ZERO_ALLOC) {
3032             qemu_co_mutex_unlock(&s->lock);
3033             return -ENOTSUP;
3034         }
3035     } else {
3036         qemu_co_mutex_lock(&s->lock);
3037     }
3038 
3039     trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes);
3040 
3041     /* Whatever is left can use real zero clusters */
3042     ret = qcow2_cluster_zeroize(bs, offset, bytes, flags);
3043     qemu_co_mutex_unlock(&s->lock);
3044 
3045     return ret;
3046 }
3047 
3048 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
3049                                           int64_t offset, int bytes)
3050 {
3051     int ret;
3052     BDRVQcow2State *s = bs->opaque;
3053 
3054     if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
3055         assert(bytes < s->cluster_size);
3056         /* Ignore partial clusters, except for the special case of the
3057          * complete partial cluster at the end of an unaligned file */
3058         if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
3059             offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) {
3060             return -ENOTSUP;
3061         }
3062     }
3063 
3064     qemu_co_mutex_lock(&s->lock);
3065     ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST,
3066                                 false);
3067     qemu_co_mutex_unlock(&s->lock);
3068     return ret;
3069 }
3070 
3071 static int qcow2_truncate(BlockDriverState *bs, int64_t offset,
3072                           PreallocMode prealloc, Error **errp)
3073 {
3074     BDRVQcow2State *s = bs->opaque;
3075     uint64_t old_length;
3076     int64_t new_l1_size;
3077     int ret;
3078 
3079     if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA &&
3080         prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL)
3081     {
3082         error_setg(errp, "Unsupported preallocation mode '%s'",
3083                    PreallocMode_str(prealloc));
3084         return -ENOTSUP;
3085     }
3086 
3087     if (offset & 511) {
3088         error_setg(errp, "The new size must be a multiple of 512");
3089         return -EINVAL;
3090     }
3091 
3092     /* cannot proceed if image has snapshots */
3093     if (s->nb_snapshots) {
3094         error_setg(errp, "Can't resize an image which has snapshots");
3095         return -ENOTSUP;
3096     }
3097 
3098     /* cannot proceed if image has bitmaps */
3099     if (s->nb_bitmaps) {
3100         /* TODO: resize bitmaps in the image */
3101         error_setg(errp, "Can't resize an image which has bitmaps");
3102         return -ENOTSUP;
3103     }
3104 
3105     old_length = bs->total_sectors * 512;
3106 
3107     /* shrinking is currently not supported */
3108     if (offset < old_length) {
3109         error_setg(errp, "qcow2 doesn't support shrinking images yet");
3110         return -ENOTSUP;
3111     }
3112 
3113     new_l1_size = size_to_l1(s, offset);
3114     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
3115     if (ret < 0) {
3116         error_setg_errno(errp, -ret, "Failed to grow the L1 table");
3117         return ret;
3118     }
3119 
3120     switch (prealloc) {
3121     case PREALLOC_MODE_OFF:
3122         break;
3123 
3124     case PREALLOC_MODE_METADATA:
3125         ret = preallocate(bs, old_length, offset);
3126         if (ret < 0) {
3127             error_setg_errno(errp, -ret, "Preallocation failed");
3128             return ret;
3129         }
3130         break;
3131 
3132     case PREALLOC_MODE_FALLOC:
3133     case PREALLOC_MODE_FULL:
3134     {
3135         int64_t allocation_start, host_offset, guest_offset;
3136         int64_t clusters_allocated;
3137         int64_t old_file_size, new_file_size;
3138         uint64_t nb_new_data_clusters, nb_new_l2_tables;
3139 
3140         old_file_size = bdrv_getlength(bs->file->bs);
3141         if (old_file_size < 0) {
3142             error_setg_errno(errp, -old_file_size,
3143                              "Failed to inquire current file length");
3144             return ret;
3145         }
3146 
3147         nb_new_data_clusters = DIV_ROUND_UP(offset - old_length,
3148                                             s->cluster_size);
3149 
3150         /* This is an overestimation; we will not actually allocate space for
3151          * these in the file but just make sure the new refcount structures are
3152          * able to cover them so we will not have to allocate new refblocks
3153          * while entering the data blocks in the potentially new L2 tables.
3154          * (We do not actually care where the L2 tables are placed. Maybe they
3155          *  are already allocated or they can be placed somewhere before
3156          *  @old_file_size. It does not matter because they will be fully
3157          *  allocated automatically, so they do not need to be covered by the
3158          *  preallocation. All that matters is that we will not have to allocate
3159          *  new refcount structures for them.) */
3160         nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters,
3161                                         s->cluster_size / sizeof(uint64_t));
3162         /* The cluster range may not be aligned to L2 boundaries, so add one L2
3163          * table for a potential head/tail */
3164         nb_new_l2_tables++;
3165 
3166         allocation_start = qcow2_refcount_area(bs, old_file_size,
3167                                                nb_new_data_clusters +
3168                                                nb_new_l2_tables,
3169                                                true, 0, 0);
3170         if (allocation_start < 0) {
3171             error_setg_errno(errp, -allocation_start,
3172                              "Failed to resize refcount structures");
3173             return -allocation_start;
3174         }
3175 
3176         clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start,
3177                                                      nb_new_data_clusters);
3178         if (clusters_allocated < 0) {
3179             error_setg_errno(errp, -clusters_allocated,
3180                              "Failed to allocate data clusters");
3181             return -clusters_allocated;
3182         }
3183 
3184         assert(clusters_allocated == nb_new_data_clusters);
3185 
3186         /* Allocate the data area */
3187         new_file_size = allocation_start +
3188                         nb_new_data_clusters * s->cluster_size;
3189         ret = bdrv_truncate(bs->file, new_file_size, prealloc, errp);
3190         if (ret < 0) {
3191             error_prepend(errp, "Failed to resize underlying file: ");
3192             qcow2_free_clusters(bs, allocation_start,
3193                                 nb_new_data_clusters * s->cluster_size,
3194                                 QCOW2_DISCARD_OTHER);
3195             return ret;
3196         }
3197 
3198         /* Create the necessary L2 entries */
3199         host_offset = allocation_start;
3200         guest_offset = old_length;
3201         while (nb_new_data_clusters) {
3202             int64_t guest_cluster = guest_offset >> s->cluster_bits;
3203             int64_t nb_clusters = MIN(nb_new_data_clusters,
3204                                       s->l2_size - guest_cluster % s->l2_size);
3205             QCowL2Meta allocation = {
3206                 .offset       = guest_offset,
3207                 .alloc_offset = host_offset,
3208                 .nb_clusters  = nb_clusters,
3209             };
3210             qemu_co_queue_init(&allocation.dependent_requests);
3211 
3212             ret = qcow2_alloc_cluster_link_l2(bs, &allocation);
3213             if (ret < 0) {
3214                 error_setg_errno(errp, -ret, "Failed to update L2 tables");
3215                 qcow2_free_clusters(bs, host_offset,
3216                                     nb_new_data_clusters * s->cluster_size,
3217                                     QCOW2_DISCARD_OTHER);
3218                 return ret;
3219             }
3220 
3221             guest_offset += nb_clusters * s->cluster_size;
3222             host_offset += nb_clusters * s->cluster_size;
3223             nb_new_data_clusters -= nb_clusters;
3224         }
3225         break;
3226     }
3227 
3228     default:
3229         g_assert_not_reached();
3230     }
3231 
3232     if (prealloc != PREALLOC_MODE_OFF) {
3233         /* Flush metadata before actually changing the image size */
3234         ret = bdrv_flush(bs);
3235         if (ret < 0) {
3236             error_setg_errno(errp, -ret,
3237                              "Failed to flush the preallocated area to disk");
3238             return ret;
3239         }
3240     }
3241 
3242     /* write updated header.size */
3243     offset = cpu_to_be64(offset);
3244     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
3245                            &offset, sizeof(uint64_t));
3246     if (ret < 0) {
3247         error_setg_errno(errp, -ret, "Failed to update the image size");
3248         return ret;
3249     }
3250 
3251     s->l1_vm_state_index = new_l1_size;
3252     return 0;
3253 }
3254 
3255 /* XXX: put compressed sectors first, then all the cluster aligned
3256    tables to avoid losing bytes in alignment */
3257 static coroutine_fn int
3258 qcow2_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
3259                             uint64_t bytes, QEMUIOVector *qiov)
3260 {
3261     BDRVQcow2State *s = bs->opaque;
3262     QEMUIOVector hd_qiov;
3263     struct iovec iov;
3264     z_stream strm;
3265     int ret, out_len;
3266     uint8_t *buf, *out_buf;
3267     int64_t cluster_offset;
3268 
3269     if (bytes == 0) {
3270         /* align end of file to a sector boundary to ease reading with
3271            sector based I/Os */
3272         cluster_offset = bdrv_getlength(bs->file->bs);
3273         if (cluster_offset < 0) {
3274             return cluster_offset;
3275         }
3276         return bdrv_truncate(bs->file, cluster_offset, PREALLOC_MODE_OFF, NULL);
3277     }
3278 
3279     buf = qemu_blockalign(bs, s->cluster_size);
3280     if (bytes != s->cluster_size) {
3281         if (bytes > s->cluster_size ||
3282             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
3283         {
3284             qemu_vfree(buf);
3285             return -EINVAL;
3286         }
3287         /* Zero-pad last write if image size is not cluster aligned */
3288         memset(buf + bytes, 0, s->cluster_size - bytes);
3289     }
3290     qemu_iovec_to_buf(qiov, 0, buf, bytes);
3291 
3292     out_buf = g_malloc(s->cluster_size);
3293 
3294     /* best compression, small window, no zlib header */
3295     memset(&strm, 0, sizeof(strm));
3296     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
3297                        Z_DEFLATED, -12,
3298                        9, Z_DEFAULT_STRATEGY);
3299     if (ret != 0) {
3300         ret = -EINVAL;
3301         goto fail;
3302     }
3303 
3304     strm.avail_in = s->cluster_size;
3305     strm.next_in = (uint8_t *)buf;
3306     strm.avail_out = s->cluster_size;
3307     strm.next_out = out_buf;
3308 
3309     ret = deflate(&strm, Z_FINISH);
3310     if (ret != Z_STREAM_END && ret != Z_OK) {
3311         deflateEnd(&strm);
3312         ret = -EINVAL;
3313         goto fail;
3314     }
3315     out_len = strm.next_out - out_buf;
3316 
3317     deflateEnd(&strm);
3318 
3319     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
3320         /* could not compress: write normal cluster */
3321         ret = qcow2_co_pwritev(bs, offset, bytes, qiov, 0);
3322         if (ret < 0) {
3323             goto fail;
3324         }
3325         goto success;
3326     }
3327 
3328     qemu_co_mutex_lock(&s->lock);
3329     cluster_offset =
3330         qcow2_alloc_compressed_cluster_offset(bs, offset, out_len);
3331     if (!cluster_offset) {
3332         qemu_co_mutex_unlock(&s->lock);
3333         ret = -EIO;
3334         goto fail;
3335     }
3336     cluster_offset &= s->cluster_offset_mask;
3337 
3338     ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
3339     qemu_co_mutex_unlock(&s->lock);
3340     if (ret < 0) {
3341         goto fail;
3342     }
3343 
3344     iov = (struct iovec) {
3345         .iov_base   = out_buf,
3346         .iov_len    = out_len,
3347     };
3348     qemu_iovec_init_external(&hd_qiov, &iov, 1);
3349 
3350     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
3351     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
3352     if (ret < 0) {
3353         goto fail;
3354     }
3355 success:
3356     ret = 0;
3357 fail:
3358     qemu_vfree(buf);
3359     g_free(out_buf);
3360     return ret;
3361 }
3362 
3363 static int make_completely_empty(BlockDriverState *bs)
3364 {
3365     BDRVQcow2State *s = bs->opaque;
3366     Error *local_err = NULL;
3367     int ret, l1_clusters;
3368     int64_t offset;
3369     uint64_t *new_reftable = NULL;
3370     uint64_t rt_entry, l1_size2;
3371     struct {
3372         uint64_t l1_offset;
3373         uint64_t reftable_offset;
3374         uint32_t reftable_clusters;
3375     } QEMU_PACKED l1_ofs_rt_ofs_cls;
3376 
3377     ret = qcow2_cache_empty(bs, s->l2_table_cache);
3378     if (ret < 0) {
3379         goto fail;
3380     }
3381 
3382     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
3383     if (ret < 0) {
3384         goto fail;
3385     }
3386 
3387     /* Refcounts will be broken utterly */
3388     ret = qcow2_mark_dirty(bs);
3389     if (ret < 0) {
3390         goto fail;
3391     }
3392 
3393     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
3394 
3395     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
3396     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
3397 
3398     /* After this call, neither the in-memory nor the on-disk refcount
3399      * information accurately describe the actual references */
3400 
3401     ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
3402                              l1_clusters * s->cluster_size, 0);
3403     if (ret < 0) {
3404         goto fail_broken_refcounts;
3405     }
3406     memset(s->l1_table, 0, l1_size2);
3407 
3408     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
3409 
3410     /* Overwrite enough clusters at the beginning of the sectors to place
3411      * the refcount table, a refcount block and the L1 table in; this may
3412      * overwrite parts of the existing refcount and L1 table, which is not
3413      * an issue because the dirty flag is set, complete data loss is in fact
3414      * desired and partial data loss is consequently fine as well */
3415     ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
3416                              (2 + l1_clusters) * s->cluster_size, 0);
3417     /* This call (even if it failed overall) may have overwritten on-disk
3418      * refcount structures; in that case, the in-memory refcount information
3419      * will probably differ from the on-disk information which makes the BDS
3420      * unusable */
3421     if (ret < 0) {
3422         goto fail_broken_refcounts;
3423     }
3424 
3425     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
3426     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
3427 
3428     /* "Create" an empty reftable (one cluster) directly after the image
3429      * header and an empty L1 table three clusters after the image header;
3430      * the cluster between those two will be used as the first refblock */
3431     l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
3432     l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
3433     l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
3434     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
3435                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
3436     if (ret < 0) {
3437         goto fail_broken_refcounts;
3438     }
3439 
3440     s->l1_table_offset = 3 * s->cluster_size;
3441 
3442     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
3443     if (!new_reftable) {
3444         ret = -ENOMEM;
3445         goto fail_broken_refcounts;
3446     }
3447 
3448     s->refcount_table_offset = s->cluster_size;
3449     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
3450     s->max_refcount_table_index = 0;
3451 
3452     g_free(s->refcount_table);
3453     s->refcount_table = new_reftable;
3454     new_reftable = NULL;
3455 
3456     /* Now the in-memory refcount information again corresponds to the on-disk
3457      * information (reftable is empty and no refblocks (the refblock cache is
3458      * empty)); however, this means some clusters (e.g. the image header) are
3459      * referenced, but not refcounted, but the normal qcow2 code assumes that
3460      * the in-memory information is always correct */
3461 
3462     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
3463 
3464     /* Enter the first refblock into the reftable */
3465     rt_entry = cpu_to_be64(2 * s->cluster_size);
3466     ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
3467                            &rt_entry, sizeof(rt_entry));
3468     if (ret < 0) {
3469         goto fail_broken_refcounts;
3470     }
3471     s->refcount_table[0] = 2 * s->cluster_size;
3472 
3473     s->free_cluster_index = 0;
3474     assert(3 + l1_clusters <= s->refcount_block_size);
3475     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
3476     if (offset < 0) {
3477         ret = offset;
3478         goto fail_broken_refcounts;
3479     } else if (offset > 0) {
3480         error_report("First cluster in emptied image is in use");
3481         abort();
3482     }
3483 
3484     /* Now finally the in-memory information corresponds to the on-disk
3485      * structures and is correct */
3486     ret = qcow2_mark_clean(bs);
3487     if (ret < 0) {
3488         goto fail;
3489     }
3490 
3491     ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size,
3492                         PREALLOC_MODE_OFF, &local_err);
3493     if (ret < 0) {
3494         error_report_err(local_err);
3495         goto fail;
3496     }
3497 
3498     return 0;
3499 
3500 fail_broken_refcounts:
3501     /* The BDS is unusable at this point. If we wanted to make it usable, we
3502      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
3503      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
3504      * again. However, because the functions which could have caused this error
3505      * path to be taken are used by those functions as well, it's very likely
3506      * that that sequence will fail as well. Therefore, just eject the BDS. */
3507     bs->drv = NULL;
3508 
3509 fail:
3510     g_free(new_reftable);
3511     return ret;
3512 }
3513 
3514 static int qcow2_make_empty(BlockDriverState *bs)
3515 {
3516     BDRVQcow2State *s = bs->opaque;
3517     uint64_t offset, end_offset;
3518     int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
3519     int l1_clusters, ret = 0;
3520 
3521     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
3522 
3523     if (s->qcow_version >= 3 && !s->snapshots &&
3524         3 + l1_clusters <= s->refcount_block_size) {
3525         /* The following function only works for qcow2 v3 images (it requires
3526          * the dirty flag) and only as long as there are no snapshots (because
3527          * it completely empties the image). Furthermore, the L1 table and three
3528          * additional clusters (image header, refcount table, one refcount
3529          * block) have to fit inside one refcount block. */
3530         return make_completely_empty(bs);
3531     }
3532 
3533     /* This fallback code simply discards every active cluster; this is slow,
3534      * but works in all cases */
3535     end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
3536     for (offset = 0; offset < end_offset; offset += step) {
3537         /* As this function is generally used after committing an external
3538          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
3539          * default action for this kind of discard is to pass the discard,
3540          * which will ideally result in an actually smaller image file, as
3541          * is probably desired. */
3542         ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
3543                                     QCOW2_DISCARD_SNAPSHOT, true);
3544         if (ret < 0) {
3545             break;
3546         }
3547     }
3548 
3549     return ret;
3550 }
3551 
3552 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
3553 {
3554     BDRVQcow2State *s = bs->opaque;
3555     int ret;
3556 
3557     qemu_co_mutex_lock(&s->lock);
3558     ret = qcow2_cache_write(bs, s->l2_table_cache);
3559     if (ret < 0) {
3560         qemu_co_mutex_unlock(&s->lock);
3561         return ret;
3562     }
3563 
3564     if (qcow2_need_accurate_refcounts(s)) {
3565         ret = qcow2_cache_write(bs, s->refcount_block_cache);
3566         if (ret < 0) {
3567             qemu_co_mutex_unlock(&s->lock);
3568             return ret;
3569         }
3570     }
3571     qemu_co_mutex_unlock(&s->lock);
3572 
3573     return 0;
3574 }
3575 
3576 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs,
3577                                        Error **errp)
3578 {
3579     Error *local_err = NULL;
3580     BlockMeasureInfo *info;
3581     uint64_t required = 0; /* bytes that contribute to required size */
3582     uint64_t virtual_size; /* disk size as seen by guest */
3583     uint64_t refcount_bits;
3584     uint64_t l2_tables;
3585     size_t cluster_size;
3586     int version;
3587     char *optstr;
3588     PreallocMode prealloc;
3589     bool has_backing_file;
3590 
3591     /* Parse image creation options */
3592     cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err);
3593     if (local_err) {
3594         goto err;
3595     }
3596 
3597     version = qcow2_opt_get_version_del(opts, &local_err);
3598     if (local_err) {
3599         goto err;
3600     }
3601 
3602     refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
3603     if (local_err) {
3604         goto err;
3605     }
3606 
3607     optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
3608     prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr,
3609                                PREALLOC_MODE_OFF, &local_err);
3610     g_free(optstr);
3611     if (local_err) {
3612         goto err;
3613     }
3614 
3615     optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
3616     has_backing_file = !!optstr;
3617     g_free(optstr);
3618 
3619     virtual_size = align_offset(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
3620                                 cluster_size);
3621 
3622     /* Check that virtual disk size is valid */
3623     l2_tables = DIV_ROUND_UP(virtual_size / cluster_size,
3624                              cluster_size / sizeof(uint64_t));
3625     if (l2_tables * sizeof(uint64_t) > QCOW_MAX_L1_SIZE) {
3626         error_setg(&local_err, "The image size is too large "
3627                                "(try using a larger cluster size)");
3628         goto err;
3629     }
3630 
3631     /* Account for input image */
3632     if (in_bs) {
3633         int64_t ssize = bdrv_getlength(in_bs);
3634         if (ssize < 0) {
3635             error_setg_errno(&local_err, -ssize,
3636                              "Unable to get image virtual_size");
3637             goto err;
3638         }
3639 
3640         virtual_size = align_offset(ssize, cluster_size);
3641 
3642         if (has_backing_file) {
3643             /* We don't how much of the backing chain is shared by the input
3644              * image and the new image file.  In the worst case the new image's
3645              * backing file has nothing in common with the input image.  Be
3646              * conservative and assume all clusters need to be written.
3647              */
3648             required = virtual_size;
3649         } else {
3650             int cluster_sectors = cluster_size / BDRV_SECTOR_SIZE;
3651             int64_t sector_num;
3652             int pnum = 0;
3653 
3654             for (sector_num = 0;
3655                  sector_num < ssize / BDRV_SECTOR_SIZE;
3656                  sector_num += pnum) {
3657                 int nb_sectors = MIN(ssize / BDRV_SECTOR_SIZE - sector_num,
3658                                      BDRV_REQUEST_MAX_SECTORS);
3659                 BlockDriverState *file;
3660                 int64_t ret;
3661 
3662                 ret = bdrv_get_block_status_above(in_bs, NULL,
3663                                                   sector_num, nb_sectors,
3664                                                   &pnum, &file);
3665                 if (ret < 0) {
3666                     error_setg_errno(&local_err, -ret,
3667                                      "Unable to get block status");
3668                     goto err;
3669                 }
3670 
3671                 if (ret & BDRV_BLOCK_ZERO) {
3672                     /* Skip zero regions (safe with no backing file) */
3673                 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
3674                            (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
3675                     /* Extend pnum to end of cluster for next iteration */
3676                     pnum = ROUND_UP(sector_num + pnum, cluster_sectors) -
3677                            sector_num;
3678 
3679                     /* Count clusters we've seen */
3680                     required += (sector_num % cluster_sectors + pnum) *
3681                                 BDRV_SECTOR_SIZE;
3682                 }
3683             }
3684         }
3685     }
3686 
3687     /* Take into account preallocation.  Nothing special is needed for
3688      * PREALLOC_MODE_METADATA since metadata is always counted.
3689      */
3690     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
3691         required = virtual_size;
3692     }
3693 
3694     info = g_new(BlockMeasureInfo, 1);
3695     info->fully_allocated =
3696         qcow2_calc_prealloc_size(virtual_size, cluster_size,
3697                                  ctz32(refcount_bits));
3698 
3699     /* Remove data clusters that are not required.  This overestimates the
3700      * required size because metadata needed for the fully allocated file is
3701      * still counted.
3702      */
3703     info->required = info->fully_allocated - virtual_size + required;
3704     return info;
3705 
3706 err:
3707     error_propagate(errp, local_err);
3708     return NULL;
3709 }
3710 
3711 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3712 {
3713     BDRVQcow2State *s = bs->opaque;
3714     bdi->unallocated_blocks_are_zero = true;
3715     bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
3716     bdi->cluster_size = s->cluster_size;
3717     bdi->vm_state_offset = qcow2_vm_state_offset(s);
3718     return 0;
3719 }
3720 
3721 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
3722 {
3723     BDRVQcow2State *s = bs->opaque;
3724     ImageInfoSpecific *spec_info;
3725     QCryptoBlockInfo *encrypt_info = NULL;
3726 
3727     if (s->crypto != NULL) {
3728         encrypt_info = qcrypto_block_get_info(s->crypto, &error_abort);
3729     }
3730 
3731     spec_info = g_new(ImageInfoSpecific, 1);
3732     *spec_info = (ImageInfoSpecific){
3733         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
3734         .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1),
3735     };
3736     if (s->qcow_version == 2) {
3737         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
3738             .compat             = g_strdup("0.10"),
3739             .refcount_bits      = s->refcount_bits,
3740         };
3741     } else if (s->qcow_version == 3) {
3742         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
3743             .compat             = g_strdup("1.1"),
3744             .lazy_refcounts     = s->compatible_features &
3745                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
3746             .has_lazy_refcounts = true,
3747             .corrupt            = s->incompatible_features &
3748                                   QCOW2_INCOMPAT_CORRUPT,
3749             .has_corrupt        = true,
3750             .refcount_bits      = s->refcount_bits,
3751         };
3752     } else {
3753         /* if this assertion fails, this probably means a new version was
3754          * added without having it covered here */
3755         assert(false);
3756     }
3757 
3758     if (encrypt_info) {
3759         ImageInfoSpecificQCow2Encryption *qencrypt =
3760             g_new(ImageInfoSpecificQCow2Encryption, 1);
3761         switch (encrypt_info->format) {
3762         case Q_CRYPTO_BLOCK_FORMAT_QCOW:
3763             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES;
3764             qencrypt->u.aes = encrypt_info->u.qcow;
3765             break;
3766         case Q_CRYPTO_BLOCK_FORMAT_LUKS:
3767             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS;
3768             qencrypt->u.luks = encrypt_info->u.luks;
3769             break;
3770         default:
3771             abort();
3772         }
3773         /* Since we did shallow copy above, erase any pointers
3774          * in the original info */
3775         memset(&encrypt_info->u, 0, sizeof(encrypt_info->u));
3776         qapi_free_QCryptoBlockInfo(encrypt_info);
3777 
3778         spec_info->u.qcow2.data->has_encrypt = true;
3779         spec_info->u.qcow2.data->encrypt = qencrypt;
3780     }
3781 
3782     return spec_info;
3783 }
3784 
3785 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
3786                               int64_t pos)
3787 {
3788     BDRVQcow2State *s = bs->opaque;
3789 
3790     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
3791     return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos,
3792                                     qiov->size, qiov, 0);
3793 }
3794 
3795 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
3796                               int64_t pos)
3797 {
3798     BDRVQcow2State *s = bs->opaque;
3799 
3800     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
3801     return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos,
3802                                    qiov->size, qiov, 0);
3803 }
3804 
3805 /*
3806  * Downgrades an image's version. To achieve this, any incompatible features
3807  * have to be removed.
3808  */
3809 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
3810                            BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
3811 {
3812     BDRVQcow2State *s = bs->opaque;
3813     int current_version = s->qcow_version;
3814     int ret;
3815 
3816     if (target_version == current_version) {
3817         return 0;
3818     } else if (target_version > current_version) {
3819         return -EINVAL;
3820     } else if (target_version != 2) {
3821         return -EINVAL;
3822     }
3823 
3824     if (s->refcount_order != 4) {
3825         error_report("compat=0.10 requires refcount_bits=16");
3826         return -ENOTSUP;
3827     }
3828 
3829     /* clear incompatible features */
3830     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
3831         ret = qcow2_mark_clean(bs);
3832         if (ret < 0) {
3833             return ret;
3834         }
3835     }
3836 
3837     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
3838      * the first place; if that happens nonetheless, returning -ENOTSUP is the
3839      * best thing to do anyway */
3840 
3841     if (s->incompatible_features) {
3842         return -ENOTSUP;
3843     }
3844 
3845     /* since we can ignore compatible features, we can set them to 0 as well */
3846     s->compatible_features = 0;
3847     /* if lazy refcounts have been used, they have already been fixed through
3848      * clearing the dirty flag */
3849 
3850     /* clearing autoclear features is trivial */
3851     s->autoclear_features = 0;
3852 
3853     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
3854     if (ret < 0) {
3855         return ret;
3856     }
3857 
3858     s->qcow_version = target_version;
3859     ret = qcow2_update_header(bs);
3860     if (ret < 0) {
3861         s->qcow_version = current_version;
3862         return ret;
3863     }
3864     return 0;
3865 }
3866 
3867 typedef enum Qcow2AmendOperation {
3868     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
3869      * statically initialized to so that the helper CB can discern the first
3870      * invocation from an operation change */
3871     QCOW2_NO_OPERATION = 0,
3872 
3873     QCOW2_CHANGING_REFCOUNT_ORDER,
3874     QCOW2_DOWNGRADING,
3875 } Qcow2AmendOperation;
3876 
3877 typedef struct Qcow2AmendHelperCBInfo {
3878     /* The code coordinating the amend operations should only modify
3879      * these four fields; the rest will be managed by the CB */
3880     BlockDriverAmendStatusCB *original_status_cb;
3881     void *original_cb_opaque;
3882 
3883     Qcow2AmendOperation current_operation;
3884 
3885     /* Total number of operations to perform (only set once) */
3886     int total_operations;
3887 
3888     /* The following fields are managed by the CB */
3889 
3890     /* Number of operations completed */
3891     int operations_completed;
3892 
3893     /* Cumulative offset of all completed operations */
3894     int64_t offset_completed;
3895 
3896     Qcow2AmendOperation last_operation;
3897     int64_t last_work_size;
3898 } Qcow2AmendHelperCBInfo;
3899 
3900 static void qcow2_amend_helper_cb(BlockDriverState *bs,
3901                                   int64_t operation_offset,
3902                                   int64_t operation_work_size, void *opaque)
3903 {
3904     Qcow2AmendHelperCBInfo *info = opaque;
3905     int64_t current_work_size;
3906     int64_t projected_work_size;
3907 
3908     if (info->current_operation != info->last_operation) {
3909         if (info->last_operation != QCOW2_NO_OPERATION) {
3910             info->offset_completed += info->last_work_size;
3911             info->operations_completed++;
3912         }
3913 
3914         info->last_operation = info->current_operation;
3915     }
3916 
3917     assert(info->total_operations > 0);
3918     assert(info->operations_completed < info->total_operations);
3919 
3920     info->last_work_size = operation_work_size;
3921 
3922     current_work_size = info->offset_completed + operation_work_size;
3923 
3924     /* current_work_size is the total work size for (operations_completed + 1)
3925      * operations (which includes this one), so multiply it by the number of
3926      * operations not covered and divide it by the number of operations
3927      * covered to get a projection for the operations not covered */
3928     projected_work_size = current_work_size * (info->total_operations -
3929                                                info->operations_completed - 1)
3930                                             / (info->operations_completed + 1);
3931 
3932     info->original_status_cb(bs, info->offset_completed + operation_offset,
3933                              current_work_size + projected_work_size,
3934                              info->original_cb_opaque);
3935 }
3936 
3937 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
3938                                BlockDriverAmendStatusCB *status_cb,
3939                                void *cb_opaque)
3940 {
3941     BDRVQcow2State *s = bs->opaque;
3942     int old_version = s->qcow_version, new_version = old_version;
3943     uint64_t new_size = 0;
3944     const char *backing_file = NULL, *backing_format = NULL;
3945     bool lazy_refcounts = s->use_lazy_refcounts;
3946     const char *compat = NULL;
3947     uint64_t cluster_size = s->cluster_size;
3948     bool encrypt;
3949     int encformat;
3950     int refcount_bits = s->refcount_bits;
3951     Error *local_err = NULL;
3952     int ret;
3953     QemuOptDesc *desc = opts->list->desc;
3954     Qcow2AmendHelperCBInfo helper_cb_info;
3955 
3956     while (desc && desc->name) {
3957         if (!qemu_opt_find(opts, desc->name)) {
3958             /* only change explicitly defined options */
3959             desc++;
3960             continue;
3961         }
3962 
3963         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
3964             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
3965             if (!compat) {
3966                 /* preserve default */
3967             } else if (!strcmp(compat, "0.10")) {
3968                 new_version = 2;
3969             } else if (!strcmp(compat, "1.1")) {
3970                 new_version = 3;
3971             } else {
3972                 error_report("Unknown compatibility level %s", compat);
3973                 return -EINVAL;
3974             }
3975         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
3976             error_report("Cannot change preallocation mode");
3977             return -ENOTSUP;
3978         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
3979             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
3980         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
3981             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
3982         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
3983             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
3984         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
3985             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
3986                                         !!s->crypto);
3987 
3988             if (encrypt != !!s->crypto) {
3989                 error_report("Changing the encryption flag is not supported");
3990                 return -ENOTSUP;
3991             }
3992         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT_FORMAT)) {
3993             encformat = qcow2_crypt_method_from_format(
3994                 qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT));
3995 
3996             if (encformat != s->crypt_method_header) {
3997                 error_report("Changing the encryption format is not supported");
3998                 return -ENOTSUP;
3999             }
4000         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
4001             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
4002                                              cluster_size);
4003             if (cluster_size != s->cluster_size) {
4004                 error_report("Changing the cluster size is not supported");
4005                 return -ENOTSUP;
4006             }
4007         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
4008             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
4009                                                lazy_refcounts);
4010         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
4011             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
4012                                                 refcount_bits);
4013 
4014             if (refcount_bits <= 0 || refcount_bits > 64 ||
4015                 !is_power_of_2(refcount_bits))
4016             {
4017                 error_report("Refcount width must be a power of two and may "
4018                              "not exceed 64 bits");
4019                 return -EINVAL;
4020             }
4021         } else {
4022             /* if this point is reached, this probably means a new option was
4023              * added without having it covered here */
4024             abort();
4025         }
4026 
4027         desc++;
4028     }
4029 
4030     helper_cb_info = (Qcow2AmendHelperCBInfo){
4031         .original_status_cb = status_cb,
4032         .original_cb_opaque = cb_opaque,
4033         .total_operations = (new_version < old_version)
4034                           + (s->refcount_bits != refcount_bits)
4035     };
4036 
4037     /* Upgrade first (some features may require compat=1.1) */
4038     if (new_version > old_version) {
4039         s->qcow_version = new_version;
4040         ret = qcow2_update_header(bs);
4041         if (ret < 0) {
4042             s->qcow_version = old_version;
4043             return ret;
4044         }
4045     }
4046 
4047     if (s->refcount_bits != refcount_bits) {
4048         int refcount_order = ctz32(refcount_bits);
4049 
4050         if (new_version < 3 && refcount_bits != 16) {
4051             error_report("Different refcount widths than 16 bits require "
4052                          "compatibility level 1.1 or above (use compat=1.1 or "
4053                          "greater)");
4054             return -EINVAL;
4055         }
4056 
4057         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
4058         ret = qcow2_change_refcount_order(bs, refcount_order,
4059                                           &qcow2_amend_helper_cb,
4060                                           &helper_cb_info, &local_err);
4061         if (ret < 0) {
4062             error_report_err(local_err);
4063             return ret;
4064         }
4065     }
4066 
4067     if (backing_file || backing_format) {
4068         ret = qcow2_change_backing_file(bs,
4069                     backing_file ?: s->image_backing_file,
4070                     backing_format ?: s->image_backing_format);
4071         if (ret < 0) {
4072             return ret;
4073         }
4074     }
4075 
4076     if (s->use_lazy_refcounts != lazy_refcounts) {
4077         if (lazy_refcounts) {
4078             if (new_version < 3) {
4079                 error_report("Lazy refcounts only supported with compatibility "
4080                              "level 1.1 and above (use compat=1.1 or greater)");
4081                 return -EINVAL;
4082             }
4083             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
4084             ret = qcow2_update_header(bs);
4085             if (ret < 0) {
4086                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
4087                 return ret;
4088             }
4089             s->use_lazy_refcounts = true;
4090         } else {
4091             /* make image clean first */
4092             ret = qcow2_mark_clean(bs);
4093             if (ret < 0) {
4094                 return ret;
4095             }
4096             /* now disallow lazy refcounts */
4097             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
4098             ret = qcow2_update_header(bs);
4099             if (ret < 0) {
4100                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
4101                 return ret;
4102             }
4103             s->use_lazy_refcounts = false;
4104         }
4105     }
4106 
4107     if (new_size) {
4108         BlockBackend *blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
4109         ret = blk_insert_bs(blk, bs, &local_err);
4110         if (ret < 0) {
4111             error_report_err(local_err);
4112             blk_unref(blk);
4113             return ret;
4114         }
4115 
4116         ret = blk_truncate(blk, new_size, PREALLOC_MODE_OFF, &local_err);
4117         blk_unref(blk);
4118         if (ret < 0) {
4119             error_report_err(local_err);
4120             return ret;
4121         }
4122     }
4123 
4124     /* Downgrade last (so unsupported features can be removed before) */
4125     if (new_version < old_version) {
4126         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
4127         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
4128                               &helper_cb_info);
4129         if (ret < 0) {
4130             return ret;
4131         }
4132     }
4133 
4134     return 0;
4135 }
4136 
4137 /*
4138  * If offset or size are negative, respectively, they will not be included in
4139  * the BLOCK_IMAGE_CORRUPTED event emitted.
4140  * fatal will be ignored for read-only BDS; corruptions found there will always
4141  * be considered non-fatal.
4142  */
4143 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
4144                              int64_t size, const char *message_format, ...)
4145 {
4146     BDRVQcow2State *s = bs->opaque;
4147     const char *node_name;
4148     char *message;
4149     va_list ap;
4150 
4151     fatal = fatal && !bs->read_only;
4152 
4153     if (s->signaled_corruption &&
4154         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
4155     {
4156         return;
4157     }
4158 
4159     va_start(ap, message_format);
4160     message = g_strdup_vprintf(message_format, ap);
4161     va_end(ap);
4162 
4163     if (fatal) {
4164         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
4165                 "corruption events will be suppressed\n", message);
4166     } else {
4167         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
4168                 "corruption events will be suppressed\n", message);
4169     }
4170 
4171     node_name = bdrv_get_node_name(bs);
4172     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
4173                                           *node_name != '\0', node_name,
4174                                           message, offset >= 0, offset,
4175                                           size >= 0, size,
4176                                           fatal, &error_abort);
4177     g_free(message);
4178 
4179     if (fatal) {
4180         qcow2_mark_corrupt(bs);
4181         bs->drv = NULL; /* make BDS unusable */
4182     }
4183 
4184     s->signaled_corruption = true;
4185 }
4186 
4187 static QemuOptsList qcow2_create_opts = {
4188     .name = "qcow2-create-opts",
4189     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
4190     .desc = {
4191         {
4192             .name = BLOCK_OPT_SIZE,
4193             .type = QEMU_OPT_SIZE,
4194             .help = "Virtual disk size"
4195         },
4196         {
4197             .name = BLOCK_OPT_COMPAT_LEVEL,
4198             .type = QEMU_OPT_STRING,
4199             .help = "Compatibility level (0.10 or 1.1)"
4200         },
4201         {
4202             .name = BLOCK_OPT_BACKING_FILE,
4203             .type = QEMU_OPT_STRING,
4204             .help = "File name of a base image"
4205         },
4206         {
4207             .name = BLOCK_OPT_BACKING_FMT,
4208             .type = QEMU_OPT_STRING,
4209             .help = "Image format of the base image"
4210         },
4211         {
4212             .name = BLOCK_OPT_ENCRYPT,
4213             .type = QEMU_OPT_BOOL,
4214             .help = "Encrypt the image with format 'aes'. (Deprecated "
4215                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
4216         },
4217         {
4218             .name = BLOCK_OPT_ENCRYPT_FORMAT,
4219             .type = QEMU_OPT_STRING,
4220             .help = "Encrypt the image, format choices: 'aes', 'luks'",
4221         },
4222         BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
4223             "ID of secret providing qcow AES key or LUKS passphrase"),
4224         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
4225         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
4226         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
4227         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
4228         BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
4229         BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
4230         {
4231             .name = BLOCK_OPT_CLUSTER_SIZE,
4232             .type = QEMU_OPT_SIZE,
4233             .help = "qcow2 cluster size",
4234             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
4235         },
4236         {
4237             .name = BLOCK_OPT_PREALLOC,
4238             .type = QEMU_OPT_STRING,
4239             .help = "Preallocation mode (allowed values: off, metadata, "
4240                     "falloc, full)"
4241         },
4242         {
4243             .name = BLOCK_OPT_LAZY_REFCOUNTS,
4244             .type = QEMU_OPT_BOOL,
4245             .help = "Postpone refcount updates",
4246             .def_value_str = "off"
4247         },
4248         {
4249             .name = BLOCK_OPT_REFCOUNT_BITS,
4250             .type = QEMU_OPT_NUMBER,
4251             .help = "Width of a reference count entry in bits",
4252             .def_value_str = "16"
4253         },
4254         { /* end of list */ }
4255     }
4256 };
4257 
4258 BlockDriver bdrv_qcow2 = {
4259     .format_name        = "qcow2",
4260     .instance_size      = sizeof(BDRVQcow2State),
4261     .bdrv_probe         = qcow2_probe,
4262     .bdrv_open          = qcow2_open,
4263     .bdrv_close         = qcow2_close,
4264     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
4265     .bdrv_reopen_commit   = qcow2_reopen_commit,
4266     .bdrv_reopen_abort    = qcow2_reopen_abort,
4267     .bdrv_join_options    = qcow2_join_options,
4268     .bdrv_child_perm      = bdrv_format_default_perms,
4269     .bdrv_create        = qcow2_create,
4270     .bdrv_has_zero_init = bdrv_has_zero_init_1,
4271     .bdrv_co_get_block_status = qcow2_co_get_block_status,
4272 
4273     .bdrv_co_preadv         = qcow2_co_preadv,
4274     .bdrv_co_pwritev        = qcow2_co_pwritev,
4275     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
4276 
4277     .bdrv_co_pwrite_zeroes  = qcow2_co_pwrite_zeroes,
4278     .bdrv_co_pdiscard       = qcow2_co_pdiscard,
4279     .bdrv_truncate          = qcow2_truncate,
4280     .bdrv_co_pwritev_compressed = qcow2_co_pwritev_compressed,
4281     .bdrv_make_empty        = qcow2_make_empty,
4282 
4283     .bdrv_snapshot_create   = qcow2_snapshot_create,
4284     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
4285     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
4286     .bdrv_snapshot_list     = qcow2_snapshot_list,
4287     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
4288     .bdrv_measure           = qcow2_measure,
4289     .bdrv_get_info          = qcow2_get_info,
4290     .bdrv_get_specific_info = qcow2_get_specific_info,
4291 
4292     .bdrv_save_vmstate    = qcow2_save_vmstate,
4293     .bdrv_load_vmstate    = qcow2_load_vmstate,
4294 
4295     .supports_backing           = true,
4296     .bdrv_change_backing_file   = qcow2_change_backing_file,
4297 
4298     .bdrv_refresh_limits        = qcow2_refresh_limits,
4299     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
4300     .bdrv_inactivate            = qcow2_inactivate,
4301 
4302     .create_opts         = &qcow2_create_opts,
4303     .bdrv_check          = qcow2_check,
4304     .bdrv_amend_options  = qcow2_amend_options,
4305 
4306     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
4307     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
4308 
4309     .bdrv_reopen_bitmaps_rw = qcow2_reopen_bitmaps_rw,
4310     .bdrv_can_store_new_dirty_bitmap = qcow2_can_store_new_dirty_bitmap,
4311     .bdrv_remove_persistent_dirty_bitmap = qcow2_remove_persistent_dirty_bitmap,
4312 };
4313 
4314 static void bdrv_qcow2_init(void)
4315 {
4316     bdrv_register(&bdrv_qcow2);
4317 }
4318 
4319 block_init(bdrv_qcow2_init);
4320