xref: /openbmc/qemu/block/qcow2.c (revision e1ecf8c8)
1 /*
2  * Block driver for the QCOW version 2 format
3  *
4  * Copyright (c) 2004-2006 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 
25 #include "qemu/osdep.h"
26 
27 #include "block/qdict.h"
28 #include "sysemu/block-backend.h"
29 #include "qemu/main-loop.h"
30 #include "qemu/module.h"
31 #include "qcow2.h"
32 #include "qemu/error-report.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-events-block-core.h"
35 #include "qapi/qmp/qdict.h"
36 #include "qapi/qmp/qstring.h"
37 #include "trace.h"
38 #include "qemu/option_int.h"
39 #include "qemu/cutils.h"
40 #include "qemu/bswap.h"
41 #include "qapi/qobject-input-visitor.h"
42 #include "qapi/qapi-visit-block-core.h"
43 #include "crypto.h"
44 #include "block/aio_task.h"
45 
46 /*
47   Differences with QCOW:
48 
49   - Support for multiple incremental snapshots.
50   - Memory management by reference counts.
51   - Clusters which have a reference count of one have the bit
52     QCOW_OFLAG_COPIED to optimize write performance.
53   - Size of compressed clusters is stored in sectors to reduce bit usage
54     in the cluster offsets.
55   - Support for storing additional data (such as the VM state) in the
56     snapshots.
57   - If a backing store is used, the cluster size is not constrained
58     (could be backported to QCOW).
59   - L2 tables have always a size of one cluster.
60 */
61 
62 
63 typedef struct {
64     uint32_t magic;
65     uint32_t len;
66 } QEMU_PACKED QCowExtension;
67 
68 #define  QCOW2_EXT_MAGIC_END 0
69 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
70 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
71 #define  QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
72 #define  QCOW2_EXT_MAGIC_BITMAPS 0x23852875
73 #define  QCOW2_EXT_MAGIC_DATA_FILE 0x44415441
74 
75 static int coroutine_fn
76 qcow2_co_preadv_compressed(BlockDriverState *bs,
77                            uint64_t file_cluster_offset,
78                            uint64_t offset,
79                            uint64_t bytes,
80                            QEMUIOVector *qiov,
81                            size_t qiov_offset);
82 
83 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
84 {
85     const QCowHeader *cow_header = (const void *)buf;
86 
87     if (buf_size >= sizeof(QCowHeader) &&
88         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
89         be32_to_cpu(cow_header->version) >= 2)
90         return 100;
91     else
92         return 0;
93 }
94 
95 
96 static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset,
97                                           uint8_t *buf, size_t buflen,
98                                           void *opaque, Error **errp)
99 {
100     BlockDriverState *bs = opaque;
101     BDRVQcow2State *s = bs->opaque;
102     ssize_t ret;
103 
104     if ((offset + buflen) > s->crypto_header.length) {
105         error_setg(errp, "Request for data outside of extension header");
106         return -1;
107     }
108 
109     ret = bdrv_pread(bs->file,
110                      s->crypto_header.offset + offset, buf, buflen);
111     if (ret < 0) {
112         error_setg_errno(errp, -ret, "Could not read encryption header");
113         return -1;
114     }
115     return ret;
116 }
117 
118 
119 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen,
120                                           void *opaque, Error **errp)
121 {
122     BlockDriverState *bs = opaque;
123     BDRVQcow2State *s = bs->opaque;
124     int64_t ret;
125     int64_t clusterlen;
126 
127     ret = qcow2_alloc_clusters(bs, headerlen);
128     if (ret < 0) {
129         error_setg_errno(errp, -ret,
130                          "Cannot allocate cluster for LUKS header size %zu",
131                          headerlen);
132         return -1;
133     }
134 
135     s->crypto_header.length = headerlen;
136     s->crypto_header.offset = ret;
137 
138     /* Zero fill remaining space in cluster so it has predictable
139      * content in case of future spec changes */
140     clusterlen = size_to_clusters(s, headerlen) * s->cluster_size;
141     assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen, false) == 0);
142     ret = bdrv_pwrite_zeroes(bs->file,
143                              ret + headerlen,
144                              clusterlen - headerlen, 0);
145     if (ret < 0) {
146         error_setg_errno(errp, -ret, "Could not zero fill encryption header");
147         return -1;
148     }
149 
150     return ret;
151 }
152 
153 
154 static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset,
155                                            const uint8_t *buf, size_t buflen,
156                                            void *opaque, Error **errp)
157 {
158     BlockDriverState *bs = opaque;
159     BDRVQcow2State *s = bs->opaque;
160     ssize_t ret;
161 
162     if ((offset + buflen) > s->crypto_header.length) {
163         error_setg(errp, "Request for data outside of extension header");
164         return -1;
165     }
166 
167     ret = bdrv_pwrite(bs->file,
168                       s->crypto_header.offset + offset, buf, buflen);
169     if (ret < 0) {
170         error_setg_errno(errp, -ret, "Could not read encryption header");
171         return -1;
172     }
173     return ret;
174 }
175 
176 
177 /*
178  * read qcow2 extension and fill bs
179  * start reading from start_offset
180  * finish reading upon magic of value 0 or when end_offset reached
181  * unknown magic is skipped (future extension this version knows nothing about)
182  * return 0 upon success, non-0 otherwise
183  */
184 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
185                                  uint64_t end_offset, void **p_feature_table,
186                                  int flags, bool *need_update_header,
187                                  Error **errp)
188 {
189     BDRVQcow2State *s = bs->opaque;
190     QCowExtension ext;
191     uint64_t offset;
192     int ret;
193     Qcow2BitmapHeaderExt bitmaps_ext;
194 
195     if (need_update_header != NULL) {
196         *need_update_header = false;
197     }
198 
199 #ifdef DEBUG_EXT
200     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
201 #endif
202     offset = start_offset;
203     while (offset < end_offset) {
204 
205 #ifdef DEBUG_EXT
206         /* Sanity check */
207         if (offset > s->cluster_size)
208             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
209 
210         printf("attempting to read extended header in offset %lu\n", offset);
211 #endif
212 
213         ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
214         if (ret < 0) {
215             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
216                              "pread fail from offset %" PRIu64, offset);
217             return 1;
218         }
219         ext.magic = be32_to_cpu(ext.magic);
220         ext.len = be32_to_cpu(ext.len);
221         offset += sizeof(ext);
222 #ifdef DEBUG_EXT
223         printf("ext.magic = 0x%x\n", ext.magic);
224 #endif
225         if (offset > end_offset || ext.len > end_offset - offset) {
226             error_setg(errp, "Header extension too large");
227             return -EINVAL;
228         }
229 
230         switch (ext.magic) {
231         case QCOW2_EXT_MAGIC_END:
232             return 0;
233 
234         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
235             if (ext.len >= sizeof(bs->backing_format)) {
236                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
237                            " too large (>=%zu)", ext.len,
238                            sizeof(bs->backing_format));
239                 return 2;
240             }
241             ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len);
242             if (ret < 0) {
243                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
244                                  "Could not read format name");
245                 return 3;
246             }
247             bs->backing_format[ext.len] = '\0';
248             s->image_backing_format = g_strdup(bs->backing_format);
249 #ifdef DEBUG_EXT
250             printf("Qcow2: Got format extension %s\n", bs->backing_format);
251 #endif
252             break;
253 
254         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
255             if (p_feature_table != NULL) {
256                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
257                 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
258                 if (ret < 0) {
259                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
260                                      "Could not read table");
261                     return ret;
262                 }
263 
264                 *p_feature_table = feature_table;
265             }
266             break;
267 
268         case QCOW2_EXT_MAGIC_CRYPTO_HEADER: {
269             unsigned int cflags = 0;
270             if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
271                 error_setg(errp, "CRYPTO header extension only "
272                            "expected with LUKS encryption method");
273                 return -EINVAL;
274             }
275             if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) {
276                 error_setg(errp, "CRYPTO header extension size %u, "
277                            "but expected size %zu", ext.len,
278                            sizeof(Qcow2CryptoHeaderExtension));
279                 return -EINVAL;
280             }
281 
282             ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len);
283             if (ret < 0) {
284                 error_setg_errno(errp, -ret,
285                                  "Unable to read CRYPTO header extension");
286                 return ret;
287             }
288             s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
289             s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
290 
291             if ((s->crypto_header.offset % s->cluster_size) != 0) {
292                 error_setg(errp, "Encryption header offset '%" PRIu64 "' is "
293                            "not a multiple of cluster size '%u'",
294                            s->crypto_header.offset, s->cluster_size);
295                 return -EINVAL;
296             }
297 
298             if (flags & BDRV_O_NO_IO) {
299                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
300             }
301             s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
302                                            qcow2_crypto_hdr_read_func,
303                                            bs, cflags, QCOW2_MAX_THREADS, errp);
304             if (!s->crypto) {
305                 return -EINVAL;
306             }
307         }   break;
308 
309         case QCOW2_EXT_MAGIC_BITMAPS:
310             if (ext.len != sizeof(bitmaps_ext)) {
311                 error_setg_errno(errp, -ret, "bitmaps_ext: "
312                                  "Invalid extension length");
313                 return -EINVAL;
314             }
315 
316             if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) {
317                 if (s->qcow_version < 3) {
318                     /* Let's be a bit more specific */
319                     warn_report("This qcow2 v2 image contains bitmaps, but "
320                                 "they may have been modified by a program "
321                                 "without persistent bitmap support; so now "
322                                 "they must all be considered inconsistent");
323                 } else {
324                     warn_report("a program lacking bitmap support "
325                                 "modified this file, so all bitmaps are now "
326                                 "considered inconsistent");
327                 }
328                 error_printf("Some clusters may be leaked, "
329                              "run 'qemu-img check -r' on the image "
330                              "file to fix.");
331                 if (need_update_header != NULL) {
332                     /* Updating is needed to drop invalid bitmap extension. */
333                     *need_update_header = true;
334                 }
335                 break;
336             }
337 
338             ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len);
339             if (ret < 0) {
340                 error_setg_errno(errp, -ret, "bitmaps_ext: "
341                                  "Could not read ext header");
342                 return ret;
343             }
344 
345             if (bitmaps_ext.reserved32 != 0) {
346                 error_setg_errno(errp, -ret, "bitmaps_ext: "
347                                  "Reserved field is not zero");
348                 return -EINVAL;
349             }
350 
351             bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps);
352             bitmaps_ext.bitmap_directory_size =
353                 be64_to_cpu(bitmaps_ext.bitmap_directory_size);
354             bitmaps_ext.bitmap_directory_offset =
355                 be64_to_cpu(bitmaps_ext.bitmap_directory_offset);
356 
357             if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) {
358                 error_setg(errp,
359                            "bitmaps_ext: Image has %" PRIu32 " bitmaps, "
360                            "exceeding the QEMU supported maximum of %d",
361                            bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS);
362                 return -EINVAL;
363             }
364 
365             if (bitmaps_ext.nb_bitmaps == 0) {
366                 error_setg(errp, "found bitmaps extension with zero bitmaps");
367                 return -EINVAL;
368             }
369 
370             if (bitmaps_ext.bitmap_directory_offset & (s->cluster_size - 1)) {
371                 error_setg(errp, "bitmaps_ext: "
372                                  "invalid bitmap directory offset");
373                 return -EINVAL;
374             }
375 
376             if (bitmaps_ext.bitmap_directory_size >
377                 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) {
378                 error_setg(errp, "bitmaps_ext: "
379                                  "bitmap directory size (%" PRIu64 ") exceeds "
380                                  "the maximum supported size (%d)",
381                                  bitmaps_ext.bitmap_directory_size,
382                                  QCOW2_MAX_BITMAP_DIRECTORY_SIZE);
383                 return -EINVAL;
384             }
385 
386             s->nb_bitmaps = bitmaps_ext.nb_bitmaps;
387             s->bitmap_directory_offset =
388                     bitmaps_ext.bitmap_directory_offset;
389             s->bitmap_directory_size =
390                     bitmaps_ext.bitmap_directory_size;
391 
392 #ifdef DEBUG_EXT
393             printf("Qcow2: Got bitmaps extension: "
394                    "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n",
395                    s->bitmap_directory_offset, s->nb_bitmaps);
396 #endif
397             break;
398 
399         case QCOW2_EXT_MAGIC_DATA_FILE:
400         {
401             s->image_data_file = g_malloc0(ext.len + 1);
402             ret = bdrv_pread(bs->file, offset, s->image_data_file, ext.len);
403             if (ret < 0) {
404                 error_setg_errno(errp, -ret,
405                                  "ERROR: Could not read data file name");
406                 return ret;
407             }
408 #ifdef DEBUG_EXT
409             printf("Qcow2: Got external data file %s\n", s->image_data_file);
410 #endif
411             break;
412         }
413 
414         default:
415             /* unknown magic - save it in case we need to rewrite the header */
416             /* If you add a new feature, make sure to also update the fast
417              * path of qcow2_make_empty() to deal with it. */
418             {
419                 Qcow2UnknownHeaderExtension *uext;
420 
421                 uext = g_malloc0(sizeof(*uext)  + ext.len);
422                 uext->magic = ext.magic;
423                 uext->len = ext.len;
424                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
425 
426                 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
427                 if (ret < 0) {
428                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
429                                      "Could not read data");
430                     return ret;
431                 }
432             }
433             break;
434         }
435 
436         offset += ((ext.len + 7) & ~7);
437     }
438 
439     return 0;
440 }
441 
442 static void cleanup_unknown_header_ext(BlockDriverState *bs)
443 {
444     BDRVQcow2State *s = bs->opaque;
445     Qcow2UnknownHeaderExtension *uext, *next;
446 
447     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
448         QLIST_REMOVE(uext, next);
449         g_free(uext);
450     }
451 }
452 
453 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
454                                        uint64_t mask)
455 {
456     char *features = g_strdup("");
457     char *old;
458 
459     while (table && table->name[0] != '\0') {
460         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
461             if (mask & (1ULL << table->bit)) {
462                 old = features;
463                 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
464                                            table->name);
465                 g_free(old);
466                 mask &= ~(1ULL << table->bit);
467             }
468         }
469         table++;
470     }
471 
472     if (mask) {
473         old = features;
474         features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
475                                    old, *old ? ", " : "", mask);
476         g_free(old);
477     }
478 
479     error_setg(errp, "Unsupported qcow2 feature(s): %s", features);
480     g_free(features);
481 }
482 
483 /*
484  * Sets the dirty bit and flushes afterwards if necessary.
485  *
486  * The incompatible_features bit is only set if the image file header was
487  * updated successfully.  Therefore it is not required to check the return
488  * value of this function.
489  */
490 int qcow2_mark_dirty(BlockDriverState *bs)
491 {
492     BDRVQcow2State *s = bs->opaque;
493     uint64_t val;
494     int ret;
495 
496     assert(s->qcow_version >= 3);
497 
498     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
499         return 0; /* already dirty */
500     }
501 
502     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
503     ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
504                       &val, sizeof(val));
505     if (ret < 0) {
506         return ret;
507     }
508     ret = bdrv_flush(bs->file->bs);
509     if (ret < 0) {
510         return ret;
511     }
512 
513     /* Only treat image as dirty if the header was updated successfully */
514     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
515     return 0;
516 }
517 
518 /*
519  * Clears the dirty bit and flushes before if necessary.  Only call this
520  * function when there are no pending requests, it does not guard against
521  * concurrent requests dirtying the image.
522  */
523 static int qcow2_mark_clean(BlockDriverState *bs)
524 {
525     BDRVQcow2State *s = bs->opaque;
526 
527     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
528         int ret;
529 
530         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
531 
532         ret = qcow2_flush_caches(bs);
533         if (ret < 0) {
534             return ret;
535         }
536 
537         return qcow2_update_header(bs);
538     }
539     return 0;
540 }
541 
542 /*
543  * Marks the image as corrupt.
544  */
545 int qcow2_mark_corrupt(BlockDriverState *bs)
546 {
547     BDRVQcow2State *s = bs->opaque;
548 
549     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
550     return qcow2_update_header(bs);
551 }
552 
553 /*
554  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
555  * before if necessary.
556  */
557 int qcow2_mark_consistent(BlockDriverState *bs)
558 {
559     BDRVQcow2State *s = bs->opaque;
560 
561     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
562         int ret = qcow2_flush_caches(bs);
563         if (ret < 0) {
564             return ret;
565         }
566 
567         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
568         return qcow2_update_header(bs);
569     }
570     return 0;
571 }
572 
573 static int coroutine_fn qcow2_co_check_locked(BlockDriverState *bs,
574                                               BdrvCheckResult *result,
575                                               BdrvCheckMode fix)
576 {
577     int ret = qcow2_check_refcounts(bs, result, fix);
578     if (ret < 0) {
579         return ret;
580     }
581 
582     if (fix && result->check_errors == 0 && result->corruptions == 0) {
583         ret = qcow2_mark_clean(bs);
584         if (ret < 0) {
585             return ret;
586         }
587         return qcow2_mark_consistent(bs);
588     }
589     return ret;
590 }
591 
592 static int coroutine_fn qcow2_co_check(BlockDriverState *bs,
593                                        BdrvCheckResult *result,
594                                        BdrvCheckMode fix)
595 {
596     BDRVQcow2State *s = bs->opaque;
597     int ret;
598 
599     qemu_co_mutex_lock(&s->lock);
600     ret = qcow2_co_check_locked(bs, result, fix);
601     qemu_co_mutex_unlock(&s->lock);
602     return ret;
603 }
604 
605 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset,
606                          uint64_t entries, size_t entry_len,
607                          int64_t max_size_bytes, const char *table_name,
608                          Error **errp)
609 {
610     BDRVQcow2State *s = bs->opaque;
611 
612     if (entries > max_size_bytes / entry_len) {
613         error_setg(errp, "%s too large", table_name);
614         return -EFBIG;
615     }
616 
617     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
618      * because values will be passed to qemu functions taking int64_t. */
619     if ((INT64_MAX - entries * entry_len < offset) ||
620         (offset_into_cluster(s, offset) != 0)) {
621         error_setg(errp, "%s offset invalid", table_name);
622         return -EINVAL;
623     }
624 
625     return 0;
626 }
627 
628 static const char *const mutable_opts[] = {
629     QCOW2_OPT_LAZY_REFCOUNTS,
630     QCOW2_OPT_DISCARD_REQUEST,
631     QCOW2_OPT_DISCARD_SNAPSHOT,
632     QCOW2_OPT_DISCARD_OTHER,
633     QCOW2_OPT_OVERLAP,
634     QCOW2_OPT_OVERLAP_TEMPLATE,
635     QCOW2_OPT_OVERLAP_MAIN_HEADER,
636     QCOW2_OPT_OVERLAP_ACTIVE_L1,
637     QCOW2_OPT_OVERLAP_ACTIVE_L2,
638     QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
639     QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
640     QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
641     QCOW2_OPT_OVERLAP_INACTIVE_L1,
642     QCOW2_OPT_OVERLAP_INACTIVE_L2,
643     QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
644     QCOW2_OPT_CACHE_SIZE,
645     QCOW2_OPT_L2_CACHE_SIZE,
646     QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
647     QCOW2_OPT_REFCOUNT_CACHE_SIZE,
648     QCOW2_OPT_CACHE_CLEAN_INTERVAL,
649     NULL
650 };
651 
652 static QemuOptsList qcow2_runtime_opts = {
653     .name = "qcow2",
654     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
655     .desc = {
656         {
657             .name = QCOW2_OPT_LAZY_REFCOUNTS,
658             .type = QEMU_OPT_BOOL,
659             .help = "Postpone refcount updates",
660         },
661         {
662             .name = QCOW2_OPT_DISCARD_REQUEST,
663             .type = QEMU_OPT_BOOL,
664             .help = "Pass guest discard requests to the layer below",
665         },
666         {
667             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
668             .type = QEMU_OPT_BOOL,
669             .help = "Generate discard requests when snapshot related space "
670                     "is freed",
671         },
672         {
673             .name = QCOW2_OPT_DISCARD_OTHER,
674             .type = QEMU_OPT_BOOL,
675             .help = "Generate discard requests when other clusters are freed",
676         },
677         {
678             .name = QCOW2_OPT_OVERLAP,
679             .type = QEMU_OPT_STRING,
680             .help = "Selects which overlap checks to perform from a range of "
681                     "templates (none, constant, cached, all)",
682         },
683         {
684             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
685             .type = QEMU_OPT_STRING,
686             .help = "Selects which overlap checks to perform from a range of "
687                     "templates (none, constant, cached, all)",
688         },
689         {
690             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
691             .type = QEMU_OPT_BOOL,
692             .help = "Check for unintended writes into the main qcow2 header",
693         },
694         {
695             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
696             .type = QEMU_OPT_BOOL,
697             .help = "Check for unintended writes into the active L1 table",
698         },
699         {
700             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
701             .type = QEMU_OPT_BOOL,
702             .help = "Check for unintended writes into an active L2 table",
703         },
704         {
705             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
706             .type = QEMU_OPT_BOOL,
707             .help = "Check for unintended writes into the refcount table",
708         },
709         {
710             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
711             .type = QEMU_OPT_BOOL,
712             .help = "Check for unintended writes into a refcount block",
713         },
714         {
715             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
716             .type = QEMU_OPT_BOOL,
717             .help = "Check for unintended writes into the snapshot table",
718         },
719         {
720             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
721             .type = QEMU_OPT_BOOL,
722             .help = "Check for unintended writes into an inactive L1 table",
723         },
724         {
725             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
726             .type = QEMU_OPT_BOOL,
727             .help = "Check for unintended writes into an inactive L2 table",
728         },
729         {
730             .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
731             .type = QEMU_OPT_BOOL,
732             .help = "Check for unintended writes into the bitmap directory",
733         },
734         {
735             .name = QCOW2_OPT_CACHE_SIZE,
736             .type = QEMU_OPT_SIZE,
737             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
738                     "cache size",
739         },
740         {
741             .name = QCOW2_OPT_L2_CACHE_SIZE,
742             .type = QEMU_OPT_SIZE,
743             .help = "Maximum L2 table cache size",
744         },
745         {
746             .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
747             .type = QEMU_OPT_SIZE,
748             .help = "Size of each entry in the L2 cache",
749         },
750         {
751             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
752             .type = QEMU_OPT_SIZE,
753             .help = "Maximum refcount block cache size",
754         },
755         {
756             .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
757             .type = QEMU_OPT_NUMBER,
758             .help = "Clean unused cache entries after this time (in seconds)",
759         },
760         BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
761             "ID of secret providing qcow2 AES key or LUKS passphrase"),
762         { /* end of list */ }
763     },
764 };
765 
766 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
767     [QCOW2_OL_MAIN_HEADER_BITNR]      = QCOW2_OPT_OVERLAP_MAIN_HEADER,
768     [QCOW2_OL_ACTIVE_L1_BITNR]        = QCOW2_OPT_OVERLAP_ACTIVE_L1,
769     [QCOW2_OL_ACTIVE_L2_BITNR]        = QCOW2_OPT_OVERLAP_ACTIVE_L2,
770     [QCOW2_OL_REFCOUNT_TABLE_BITNR]   = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
771     [QCOW2_OL_REFCOUNT_BLOCK_BITNR]   = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
772     [QCOW2_OL_SNAPSHOT_TABLE_BITNR]   = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
773     [QCOW2_OL_INACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_INACTIVE_L1,
774     [QCOW2_OL_INACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_INACTIVE_L2,
775     [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
776 };
777 
778 static void cache_clean_timer_cb(void *opaque)
779 {
780     BlockDriverState *bs = opaque;
781     BDRVQcow2State *s = bs->opaque;
782     qcow2_cache_clean_unused(s->l2_table_cache);
783     qcow2_cache_clean_unused(s->refcount_block_cache);
784     timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
785               (int64_t) s->cache_clean_interval * 1000);
786 }
787 
788 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
789 {
790     BDRVQcow2State *s = bs->opaque;
791     if (s->cache_clean_interval > 0) {
792         s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
793                                              SCALE_MS, cache_clean_timer_cb,
794                                              bs);
795         timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
796                   (int64_t) s->cache_clean_interval * 1000);
797     }
798 }
799 
800 static void cache_clean_timer_del(BlockDriverState *bs)
801 {
802     BDRVQcow2State *s = bs->opaque;
803     if (s->cache_clean_timer) {
804         timer_del(s->cache_clean_timer);
805         timer_free(s->cache_clean_timer);
806         s->cache_clean_timer = NULL;
807     }
808 }
809 
810 static void qcow2_detach_aio_context(BlockDriverState *bs)
811 {
812     cache_clean_timer_del(bs);
813 }
814 
815 static void qcow2_attach_aio_context(BlockDriverState *bs,
816                                      AioContext *new_context)
817 {
818     cache_clean_timer_init(bs, new_context);
819 }
820 
821 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
822                              uint64_t *l2_cache_size,
823                              uint64_t *l2_cache_entry_size,
824                              uint64_t *refcount_cache_size, Error **errp)
825 {
826     BDRVQcow2State *s = bs->opaque;
827     uint64_t combined_cache_size, l2_cache_max_setting;
828     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
829     bool l2_cache_entry_size_set;
830     int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size;
831     uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
832     uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size);
833     /* An L2 table is always one cluster in size so the max cache size
834      * should be a multiple of the cluster size. */
835     uint64_t max_l2_cache = ROUND_UP(max_l2_entries * sizeof(uint64_t),
836                                      s->cluster_size);
837 
838     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
839     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
840     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
841     l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE);
842 
843     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
844     l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE,
845                                              DEFAULT_L2_CACHE_MAX_SIZE);
846     *refcount_cache_size = qemu_opt_get_size(opts,
847                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
848 
849     *l2_cache_entry_size = qemu_opt_get_size(
850         opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size);
851 
852     *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting);
853 
854     if (combined_cache_size_set) {
855         if (l2_cache_size_set && refcount_cache_size_set) {
856             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
857                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
858                        "at the same time");
859             return;
860         } else if (l2_cache_size_set &&
861                    (l2_cache_max_setting > combined_cache_size)) {
862             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
863                        QCOW2_OPT_CACHE_SIZE);
864             return;
865         } else if (*refcount_cache_size > combined_cache_size) {
866             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
867                        QCOW2_OPT_CACHE_SIZE);
868             return;
869         }
870 
871         if (l2_cache_size_set) {
872             *refcount_cache_size = combined_cache_size - *l2_cache_size;
873         } else if (refcount_cache_size_set) {
874             *l2_cache_size = combined_cache_size - *refcount_cache_size;
875         } else {
876             /* Assign as much memory as possible to the L2 cache, and
877              * use the remainder for the refcount cache */
878             if (combined_cache_size >= max_l2_cache + min_refcount_cache) {
879                 *l2_cache_size = max_l2_cache;
880                 *refcount_cache_size = combined_cache_size - *l2_cache_size;
881             } else {
882                 *refcount_cache_size =
883                     MIN(combined_cache_size, min_refcount_cache);
884                 *l2_cache_size = combined_cache_size - *refcount_cache_size;
885             }
886         }
887     }
888 
889     /*
890      * If the L2 cache is not enough to cover the whole disk then
891      * default to 4KB entries. Smaller entries reduce the cost of
892      * loads and evictions and increase I/O performance.
893      */
894     if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) {
895         *l2_cache_entry_size = MIN(s->cluster_size, 4096);
896     }
897 
898     /* l2_cache_size and refcount_cache_size are ensured to have at least
899      * their minimum values in qcow2_update_options_prepare() */
900 
901     if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) ||
902         *l2_cache_entry_size > s->cluster_size ||
903         !is_power_of_2(*l2_cache_entry_size)) {
904         error_setg(errp, "L2 cache entry size must be a power of two "
905                    "between %d and the cluster size (%d)",
906                    1 << MIN_CLUSTER_BITS, s->cluster_size);
907         return;
908     }
909 }
910 
911 typedef struct Qcow2ReopenState {
912     Qcow2Cache *l2_table_cache;
913     Qcow2Cache *refcount_block_cache;
914     int l2_slice_size; /* Number of entries in a slice of the L2 table */
915     bool use_lazy_refcounts;
916     int overlap_check;
917     bool discard_passthrough[QCOW2_DISCARD_MAX];
918     uint64_t cache_clean_interval;
919     QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
920 } Qcow2ReopenState;
921 
922 static int qcow2_update_options_prepare(BlockDriverState *bs,
923                                         Qcow2ReopenState *r,
924                                         QDict *options, int flags,
925                                         Error **errp)
926 {
927     BDRVQcow2State *s = bs->opaque;
928     QemuOpts *opts = NULL;
929     const char *opt_overlap_check, *opt_overlap_check_template;
930     int overlap_check_template = 0;
931     uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size;
932     int i;
933     const char *encryptfmt;
934     QDict *encryptopts = NULL;
935     Error *local_err = NULL;
936     int ret;
937 
938     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
939     encryptfmt = qdict_get_try_str(encryptopts, "format");
940 
941     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
942     qemu_opts_absorb_qdict(opts, options, &local_err);
943     if (local_err) {
944         error_propagate(errp, local_err);
945         ret = -EINVAL;
946         goto fail;
947     }
948 
949     /* get L2 table/refcount block cache size from command line options */
950     read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size,
951                      &refcount_cache_size, &local_err);
952     if (local_err) {
953         error_propagate(errp, local_err);
954         ret = -EINVAL;
955         goto fail;
956     }
957 
958     l2_cache_size /= l2_cache_entry_size;
959     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
960         l2_cache_size = MIN_L2_CACHE_SIZE;
961     }
962     if (l2_cache_size > INT_MAX) {
963         error_setg(errp, "L2 cache size too big");
964         ret = -EINVAL;
965         goto fail;
966     }
967 
968     refcount_cache_size /= s->cluster_size;
969     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
970         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
971     }
972     if (refcount_cache_size > INT_MAX) {
973         error_setg(errp, "Refcount cache size too big");
974         ret = -EINVAL;
975         goto fail;
976     }
977 
978     /* alloc new L2 table/refcount block cache, flush old one */
979     if (s->l2_table_cache) {
980         ret = qcow2_cache_flush(bs, s->l2_table_cache);
981         if (ret) {
982             error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
983             goto fail;
984         }
985     }
986 
987     if (s->refcount_block_cache) {
988         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
989         if (ret) {
990             error_setg_errno(errp, -ret,
991                              "Failed to flush the refcount block cache");
992             goto fail;
993         }
994     }
995 
996     r->l2_slice_size = l2_cache_entry_size / sizeof(uint64_t);
997     r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size,
998                                            l2_cache_entry_size);
999     r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size,
1000                                                  s->cluster_size);
1001     if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
1002         error_setg(errp, "Could not allocate metadata caches");
1003         ret = -ENOMEM;
1004         goto fail;
1005     }
1006 
1007     /* New interval for cache cleanup timer */
1008     r->cache_clean_interval =
1009         qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
1010                             DEFAULT_CACHE_CLEAN_INTERVAL);
1011 #ifndef CONFIG_LINUX
1012     if (r->cache_clean_interval != 0) {
1013         error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL
1014                    " not supported on this host");
1015         ret = -EINVAL;
1016         goto fail;
1017     }
1018 #endif
1019     if (r->cache_clean_interval > UINT_MAX) {
1020         error_setg(errp, "Cache clean interval too big");
1021         ret = -EINVAL;
1022         goto fail;
1023     }
1024 
1025     /* lazy-refcounts; flush if going from enabled to disabled */
1026     r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
1027         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
1028     if (r->use_lazy_refcounts && s->qcow_version < 3) {
1029         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
1030                    "qemu 1.1 compatibility level");
1031         ret = -EINVAL;
1032         goto fail;
1033     }
1034 
1035     if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
1036         ret = qcow2_mark_clean(bs);
1037         if (ret < 0) {
1038             error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
1039             goto fail;
1040         }
1041     }
1042 
1043     /* Overlap check options */
1044     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
1045     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
1046     if (opt_overlap_check_template && opt_overlap_check &&
1047         strcmp(opt_overlap_check_template, opt_overlap_check))
1048     {
1049         error_setg(errp, "Conflicting values for qcow2 options '"
1050                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1051                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
1052         ret = -EINVAL;
1053         goto fail;
1054     }
1055     if (!opt_overlap_check) {
1056         opt_overlap_check = opt_overlap_check_template ?: "cached";
1057     }
1058 
1059     if (!strcmp(opt_overlap_check, "none")) {
1060         overlap_check_template = 0;
1061     } else if (!strcmp(opt_overlap_check, "constant")) {
1062         overlap_check_template = QCOW2_OL_CONSTANT;
1063     } else if (!strcmp(opt_overlap_check, "cached")) {
1064         overlap_check_template = QCOW2_OL_CACHED;
1065     } else if (!strcmp(opt_overlap_check, "all")) {
1066         overlap_check_template = QCOW2_OL_ALL;
1067     } else {
1068         error_setg(errp, "Unsupported value '%s' for qcow2 option "
1069                    "'overlap-check'. Allowed are any of the following: "
1070                    "none, constant, cached, all", opt_overlap_check);
1071         ret = -EINVAL;
1072         goto fail;
1073     }
1074 
1075     r->overlap_check = 0;
1076     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
1077         /* overlap-check defines a template bitmask, but every flag may be
1078          * overwritten through the associated boolean option */
1079         r->overlap_check |=
1080             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
1081                               overlap_check_template & (1 << i)) << i;
1082     }
1083 
1084     r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
1085     r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
1086     r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
1087         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
1088                           flags & BDRV_O_UNMAP);
1089     r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
1090         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
1091     r->discard_passthrough[QCOW2_DISCARD_OTHER] =
1092         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
1093 
1094     switch (s->crypt_method_header) {
1095     case QCOW_CRYPT_NONE:
1096         if (encryptfmt) {
1097             error_setg(errp, "No encryption in image header, but options "
1098                        "specified format '%s'", encryptfmt);
1099             ret = -EINVAL;
1100             goto fail;
1101         }
1102         break;
1103 
1104     case QCOW_CRYPT_AES:
1105         if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
1106             error_setg(errp,
1107                        "Header reported 'aes' encryption format but "
1108                        "options specify '%s'", encryptfmt);
1109             ret = -EINVAL;
1110             goto fail;
1111         }
1112         qdict_put_str(encryptopts, "format", "qcow");
1113         r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1114         break;
1115 
1116     case QCOW_CRYPT_LUKS:
1117         if (encryptfmt && !g_str_equal(encryptfmt, "luks")) {
1118             error_setg(errp,
1119                        "Header reported 'luks' encryption format but "
1120                        "options specify '%s'", encryptfmt);
1121             ret = -EINVAL;
1122             goto fail;
1123         }
1124         qdict_put_str(encryptopts, "format", "luks");
1125         r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1126         break;
1127 
1128     default:
1129         error_setg(errp, "Unsupported encryption method %d",
1130                    s->crypt_method_header);
1131         break;
1132     }
1133     if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) {
1134         ret = -EINVAL;
1135         goto fail;
1136     }
1137 
1138     ret = 0;
1139 fail:
1140     qobject_unref(encryptopts);
1141     qemu_opts_del(opts);
1142     opts = NULL;
1143     return ret;
1144 }
1145 
1146 static void qcow2_update_options_commit(BlockDriverState *bs,
1147                                         Qcow2ReopenState *r)
1148 {
1149     BDRVQcow2State *s = bs->opaque;
1150     int i;
1151 
1152     if (s->l2_table_cache) {
1153         qcow2_cache_destroy(s->l2_table_cache);
1154     }
1155     if (s->refcount_block_cache) {
1156         qcow2_cache_destroy(s->refcount_block_cache);
1157     }
1158     s->l2_table_cache = r->l2_table_cache;
1159     s->refcount_block_cache = r->refcount_block_cache;
1160     s->l2_slice_size = r->l2_slice_size;
1161 
1162     s->overlap_check = r->overlap_check;
1163     s->use_lazy_refcounts = r->use_lazy_refcounts;
1164 
1165     for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
1166         s->discard_passthrough[i] = r->discard_passthrough[i];
1167     }
1168 
1169     if (s->cache_clean_interval != r->cache_clean_interval) {
1170         cache_clean_timer_del(bs);
1171         s->cache_clean_interval = r->cache_clean_interval;
1172         cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
1173     }
1174 
1175     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1176     s->crypto_opts = r->crypto_opts;
1177 }
1178 
1179 static void qcow2_update_options_abort(BlockDriverState *bs,
1180                                        Qcow2ReopenState *r)
1181 {
1182     if (r->l2_table_cache) {
1183         qcow2_cache_destroy(r->l2_table_cache);
1184     }
1185     if (r->refcount_block_cache) {
1186         qcow2_cache_destroy(r->refcount_block_cache);
1187     }
1188     qapi_free_QCryptoBlockOpenOptions(r->crypto_opts);
1189 }
1190 
1191 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
1192                                 int flags, Error **errp)
1193 {
1194     Qcow2ReopenState r = {};
1195     int ret;
1196 
1197     ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
1198     if (ret >= 0) {
1199         qcow2_update_options_commit(bs, &r);
1200     } else {
1201         qcow2_update_options_abort(bs, &r);
1202     }
1203 
1204     return ret;
1205 }
1206 
1207 /* Called with s->lock held.  */
1208 static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
1209                                       int flags, Error **errp)
1210 {
1211     BDRVQcow2State *s = bs->opaque;
1212     unsigned int len, i;
1213     int ret = 0;
1214     QCowHeader header;
1215     Error *local_err = NULL;
1216     uint64_t ext_end;
1217     uint64_t l1_vm_state_index;
1218     bool update_header = false;
1219 
1220     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
1221     if (ret < 0) {
1222         error_setg_errno(errp, -ret, "Could not read qcow2 header");
1223         goto fail;
1224     }
1225     header.magic = be32_to_cpu(header.magic);
1226     header.version = be32_to_cpu(header.version);
1227     header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
1228     header.backing_file_size = be32_to_cpu(header.backing_file_size);
1229     header.size = be64_to_cpu(header.size);
1230     header.cluster_bits = be32_to_cpu(header.cluster_bits);
1231     header.crypt_method = be32_to_cpu(header.crypt_method);
1232     header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
1233     header.l1_size = be32_to_cpu(header.l1_size);
1234     header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset);
1235     header.refcount_table_clusters =
1236         be32_to_cpu(header.refcount_table_clusters);
1237     header.snapshots_offset = be64_to_cpu(header.snapshots_offset);
1238     header.nb_snapshots = be32_to_cpu(header.nb_snapshots);
1239 
1240     if (header.magic != QCOW_MAGIC) {
1241         error_setg(errp, "Image is not in qcow2 format");
1242         ret = -EINVAL;
1243         goto fail;
1244     }
1245     if (header.version < 2 || header.version > 3) {
1246         error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
1247         ret = -ENOTSUP;
1248         goto fail;
1249     }
1250 
1251     s->qcow_version = header.version;
1252 
1253     /* Initialise cluster size */
1254     if (header.cluster_bits < MIN_CLUSTER_BITS ||
1255         header.cluster_bits > MAX_CLUSTER_BITS) {
1256         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
1257                    header.cluster_bits);
1258         ret = -EINVAL;
1259         goto fail;
1260     }
1261 
1262     s->cluster_bits = header.cluster_bits;
1263     s->cluster_size = 1 << s->cluster_bits;
1264 
1265     /* Initialise version 3 header fields */
1266     if (header.version == 2) {
1267         header.incompatible_features    = 0;
1268         header.compatible_features      = 0;
1269         header.autoclear_features       = 0;
1270         header.refcount_order           = 4;
1271         header.header_length            = 72;
1272     } else {
1273         header.incompatible_features =
1274             be64_to_cpu(header.incompatible_features);
1275         header.compatible_features = be64_to_cpu(header.compatible_features);
1276         header.autoclear_features = be64_to_cpu(header.autoclear_features);
1277         header.refcount_order = be32_to_cpu(header.refcount_order);
1278         header.header_length = be32_to_cpu(header.header_length);
1279 
1280         if (header.header_length < 104) {
1281             error_setg(errp, "qcow2 header too short");
1282             ret = -EINVAL;
1283             goto fail;
1284         }
1285     }
1286 
1287     if (header.header_length > s->cluster_size) {
1288         error_setg(errp, "qcow2 header exceeds cluster size");
1289         ret = -EINVAL;
1290         goto fail;
1291     }
1292 
1293     if (header.header_length > sizeof(header)) {
1294         s->unknown_header_fields_size = header.header_length - sizeof(header);
1295         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
1296         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
1297                          s->unknown_header_fields_size);
1298         if (ret < 0) {
1299             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
1300                              "fields");
1301             goto fail;
1302         }
1303     }
1304 
1305     if (header.backing_file_offset > s->cluster_size) {
1306         error_setg(errp, "Invalid backing file offset");
1307         ret = -EINVAL;
1308         goto fail;
1309     }
1310 
1311     if (header.backing_file_offset) {
1312         ext_end = header.backing_file_offset;
1313     } else {
1314         ext_end = 1 << header.cluster_bits;
1315     }
1316 
1317     /* Handle feature bits */
1318     s->incompatible_features    = header.incompatible_features;
1319     s->compatible_features      = header.compatible_features;
1320     s->autoclear_features       = header.autoclear_features;
1321 
1322     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
1323         void *feature_table = NULL;
1324         qcow2_read_extensions(bs, header.header_length, ext_end,
1325                               &feature_table, flags, NULL, NULL);
1326         report_unsupported_feature(errp, feature_table,
1327                                    s->incompatible_features &
1328                                    ~QCOW2_INCOMPAT_MASK);
1329         ret = -ENOTSUP;
1330         g_free(feature_table);
1331         goto fail;
1332     }
1333 
1334     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
1335         /* Corrupt images may not be written to unless they are being repaired
1336          */
1337         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
1338             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
1339                        "read/write");
1340             ret = -EACCES;
1341             goto fail;
1342         }
1343     }
1344 
1345     /* Check support for various header values */
1346     if (header.refcount_order > 6) {
1347         error_setg(errp, "Reference count entry width too large; may not "
1348                    "exceed 64 bits");
1349         ret = -EINVAL;
1350         goto fail;
1351     }
1352     s->refcount_order = header.refcount_order;
1353     s->refcount_bits = 1 << s->refcount_order;
1354     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1355     s->refcount_max += s->refcount_max - 1;
1356 
1357     s->crypt_method_header = header.crypt_method;
1358     if (s->crypt_method_header) {
1359         if (bdrv_uses_whitelist() &&
1360             s->crypt_method_header == QCOW_CRYPT_AES) {
1361             error_setg(errp,
1362                        "Use of AES-CBC encrypted qcow2 images is no longer "
1363                        "supported in system emulators");
1364             error_append_hint(errp,
1365                               "You can use 'qemu-img convert' to convert your "
1366                               "image to an alternative supported format, such "
1367                               "as unencrypted qcow2, or raw with the LUKS "
1368                               "format instead.\n");
1369             ret = -ENOSYS;
1370             goto fail;
1371         }
1372 
1373         if (s->crypt_method_header == QCOW_CRYPT_AES) {
1374             s->crypt_physical_offset = false;
1375         } else {
1376             /* Assuming LUKS and any future crypt methods we
1377              * add will all use physical offsets, due to the
1378              * fact that the alternative is insecure...  */
1379             s->crypt_physical_offset = true;
1380         }
1381 
1382         bs->encrypted = true;
1383     }
1384 
1385     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
1386     s->l2_size = 1 << s->l2_bits;
1387     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1388     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1389     s->refcount_block_size = 1 << s->refcount_block_bits;
1390     bs->total_sectors = header.size / BDRV_SECTOR_SIZE;
1391     s->csize_shift = (62 - (s->cluster_bits - 8));
1392     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1393     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1394 
1395     s->refcount_table_offset = header.refcount_table_offset;
1396     s->refcount_table_size =
1397         header.refcount_table_clusters << (s->cluster_bits - 3);
1398 
1399     if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) {
1400         error_setg(errp, "Image does not contain a reference count table");
1401         ret = -EINVAL;
1402         goto fail;
1403     }
1404 
1405     ret = qcow2_validate_table(bs, s->refcount_table_offset,
1406                                header.refcount_table_clusters,
1407                                s->cluster_size, QCOW_MAX_REFTABLE_SIZE,
1408                                "Reference count table", errp);
1409     if (ret < 0) {
1410         goto fail;
1411     }
1412 
1413     /* The total size in bytes of the snapshot table is checked in
1414      * qcow2_read_snapshots() because the size of each snapshot is
1415      * variable and we don't know it yet.
1416      * Here we only check the offset and number of snapshots. */
1417     ret = qcow2_validate_table(bs, header.snapshots_offset,
1418                                header.nb_snapshots,
1419                                sizeof(QCowSnapshotHeader),
1420                                sizeof(QCowSnapshotHeader) * QCOW_MAX_SNAPSHOTS,
1421                                "Snapshot table", errp);
1422     if (ret < 0) {
1423         goto fail;
1424     }
1425 
1426     /* read the level 1 table */
1427     ret = qcow2_validate_table(bs, header.l1_table_offset,
1428                                header.l1_size, sizeof(uint64_t),
1429                                QCOW_MAX_L1_SIZE, "Active L1 table", errp);
1430     if (ret < 0) {
1431         goto fail;
1432     }
1433     s->l1_size = header.l1_size;
1434     s->l1_table_offset = header.l1_table_offset;
1435 
1436     l1_vm_state_index = size_to_l1(s, header.size);
1437     if (l1_vm_state_index > INT_MAX) {
1438         error_setg(errp, "Image is too big");
1439         ret = -EFBIG;
1440         goto fail;
1441     }
1442     s->l1_vm_state_index = l1_vm_state_index;
1443 
1444     /* the L1 table must contain at least enough entries to put
1445        header.size bytes */
1446     if (s->l1_size < s->l1_vm_state_index) {
1447         error_setg(errp, "L1 table is too small");
1448         ret = -EINVAL;
1449         goto fail;
1450     }
1451 
1452     if (s->l1_size > 0) {
1453         s->l1_table = qemu_try_blockalign(bs->file->bs,
1454             ROUND_UP(s->l1_size * sizeof(uint64_t), 512));
1455         if (s->l1_table == NULL) {
1456             error_setg(errp, "Could not allocate L1 table");
1457             ret = -ENOMEM;
1458             goto fail;
1459         }
1460         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
1461                          s->l1_size * sizeof(uint64_t));
1462         if (ret < 0) {
1463             error_setg_errno(errp, -ret, "Could not read L1 table");
1464             goto fail;
1465         }
1466         for(i = 0;i < s->l1_size; i++) {
1467             s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
1468         }
1469     }
1470 
1471     /* Parse driver-specific options */
1472     ret = qcow2_update_options(bs, options, flags, errp);
1473     if (ret < 0) {
1474         goto fail;
1475     }
1476 
1477     s->flags = flags;
1478 
1479     ret = qcow2_refcount_init(bs);
1480     if (ret != 0) {
1481         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1482         goto fail;
1483     }
1484 
1485     QLIST_INIT(&s->cluster_allocs);
1486     QTAILQ_INIT(&s->discards);
1487 
1488     /* read qcow2 extensions */
1489     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1490                               flags, &update_header, &local_err)) {
1491         error_propagate(errp, local_err);
1492         ret = -EINVAL;
1493         goto fail;
1494     }
1495 
1496     /* Open external data file */
1497     s->data_file = bdrv_open_child(NULL, options, "data-file", bs, &child_file,
1498                                    true, &local_err);
1499     if (local_err) {
1500         error_propagate(errp, local_err);
1501         ret = -EINVAL;
1502         goto fail;
1503     }
1504 
1505     if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) {
1506         if (!s->data_file && s->image_data_file) {
1507             s->data_file = bdrv_open_child(s->image_data_file, options,
1508                                            "data-file", bs, &child_file,
1509                                            false, errp);
1510             if (!s->data_file) {
1511                 ret = -EINVAL;
1512                 goto fail;
1513             }
1514         }
1515         if (!s->data_file) {
1516             error_setg(errp, "'data-file' is required for this image");
1517             ret = -EINVAL;
1518             goto fail;
1519         }
1520     } else {
1521         if (s->data_file) {
1522             error_setg(errp, "'data-file' can only be set for images with an "
1523                              "external data file");
1524             ret = -EINVAL;
1525             goto fail;
1526         }
1527 
1528         s->data_file = bs->file;
1529 
1530         if (data_file_is_raw(bs)) {
1531             error_setg(errp, "data-file-raw requires a data file");
1532             ret = -EINVAL;
1533             goto fail;
1534         }
1535     }
1536 
1537     /* qcow2_read_extension may have set up the crypto context
1538      * if the crypt method needs a header region, some methods
1539      * don't need header extensions, so must check here
1540      */
1541     if (s->crypt_method_header && !s->crypto) {
1542         if (s->crypt_method_header == QCOW_CRYPT_AES) {
1543             unsigned int cflags = 0;
1544             if (flags & BDRV_O_NO_IO) {
1545                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1546             }
1547             s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
1548                                            NULL, NULL, cflags,
1549                                            QCOW2_MAX_THREADS, errp);
1550             if (!s->crypto) {
1551                 ret = -EINVAL;
1552                 goto fail;
1553             }
1554         } else if (!(flags & BDRV_O_NO_IO)) {
1555             error_setg(errp, "Missing CRYPTO header for crypt method %d",
1556                        s->crypt_method_header);
1557             ret = -EINVAL;
1558             goto fail;
1559         }
1560     }
1561 
1562     /* read the backing file name */
1563     if (header.backing_file_offset != 0) {
1564         len = header.backing_file_size;
1565         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1566             len >= sizeof(bs->backing_file)) {
1567             error_setg(errp, "Backing file name too long");
1568             ret = -EINVAL;
1569             goto fail;
1570         }
1571         ret = bdrv_pread(bs->file, header.backing_file_offset,
1572                          bs->auto_backing_file, len);
1573         if (ret < 0) {
1574             error_setg_errno(errp, -ret, "Could not read backing file name");
1575             goto fail;
1576         }
1577         bs->auto_backing_file[len] = '\0';
1578         pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1579                 bs->auto_backing_file);
1580         s->image_backing_file = g_strdup(bs->auto_backing_file);
1581     }
1582 
1583     /* Internal snapshots */
1584     s->snapshots_offset = header.snapshots_offset;
1585     s->nb_snapshots = header.nb_snapshots;
1586 
1587     ret = qcow2_read_snapshots(bs);
1588     if (ret < 0) {
1589         error_setg_errno(errp, -ret, "Could not read snapshots");
1590         goto fail;
1591     }
1592 
1593     /* Clear unknown autoclear feature bits */
1594     update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
1595     update_header =
1596         update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE);
1597     if (update_header) {
1598         s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
1599     }
1600 
1601     /* == Handle persistent dirty bitmaps ==
1602      *
1603      * We want load dirty bitmaps in three cases:
1604      *
1605      * 1. Normal open of the disk in active mode, not related to invalidation
1606      *    after migration.
1607      *
1608      * 2. Invalidation of the target vm after pre-copy phase of migration, if
1609      *    bitmaps are _not_ migrating through migration channel, i.e.
1610      *    'dirty-bitmaps' capability is disabled.
1611      *
1612      * 3. Invalidation of source vm after failed or canceled migration.
1613      *    This is a very interesting case. There are two possible types of
1614      *    bitmaps:
1615      *
1616      *    A. Stored on inactivation and removed. They should be loaded from the
1617      *       image.
1618      *
1619      *    B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1620      *       the migration channel (with dirty-bitmaps capability).
1621      *
1622      *    On the other hand, there are two possible sub-cases:
1623      *
1624      *    3.1 disk was changed by somebody else while were inactive. In this
1625      *        case all in-RAM dirty bitmaps (both persistent and not) are
1626      *        definitely invalid. And we don't have any method to determine
1627      *        this.
1628      *
1629      *        Simple and safe thing is to just drop all the bitmaps of type B on
1630      *        inactivation. But in this case we lose bitmaps in valid 4.2 case.
1631      *
1632      *        On the other hand, resuming source vm, if disk was already changed
1633      *        is a bad thing anyway: not only bitmaps, the whole vm state is
1634      *        out of sync with disk.
1635      *
1636      *        This means, that user or management tool, who for some reason
1637      *        decided to resume source vm, after disk was already changed by
1638      *        target vm, should at least drop all dirty bitmaps by hand.
1639      *
1640      *        So, we can ignore this case for now, but TODO: "generation"
1641      *        extension for qcow2, to determine, that image was changed after
1642      *        last inactivation. And if it is changed, we will drop (or at least
1643      *        mark as 'invalid' all the bitmaps of type B, both persistent
1644      *        and not).
1645      *
1646      *    3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1647      *        to disk ('dirty-bitmaps' capability disabled), or not saved
1648      *        ('dirty-bitmaps' capability enabled), but we don't need to care
1649      *        of: let's load bitmaps as always: stored bitmaps will be loaded,
1650      *        and not stored has flag IN_USE=1 in the image and will be skipped
1651      *        on loading.
1652      *
1653      * One remaining possible case when we don't want load bitmaps:
1654      *
1655      * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1656      *    will be loaded on invalidation, no needs try loading them before)
1657      */
1658 
1659     if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) {
1660         /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1661         bool header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
1662 
1663         update_header = update_header && !header_updated;
1664     }
1665     if (local_err != NULL) {
1666         error_propagate(errp, local_err);
1667         ret = -EINVAL;
1668         goto fail;
1669     }
1670 
1671     if (update_header) {
1672         ret = qcow2_update_header(bs);
1673         if (ret < 0) {
1674             error_setg_errno(errp, -ret, "Could not update qcow2 header");
1675             goto fail;
1676         }
1677     }
1678 
1679     bs->supported_zero_flags = header.version >= 3 ? BDRV_REQ_MAY_UNMAP : 0;
1680 
1681     /* Repair image if dirty */
1682     if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1683         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1684         BdrvCheckResult result = {0};
1685 
1686         ret = qcow2_co_check_locked(bs, &result,
1687                                     BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1688         if (ret < 0 || result.check_errors) {
1689             if (ret >= 0) {
1690                 ret = -EIO;
1691             }
1692             error_setg_errno(errp, -ret, "Could not repair dirty image");
1693             goto fail;
1694         }
1695     }
1696 
1697 #ifdef DEBUG_ALLOC
1698     {
1699         BdrvCheckResult result = {0};
1700         qcow2_check_refcounts(bs, &result, 0);
1701     }
1702 #endif
1703 
1704     qemu_co_queue_init(&s->thread_task_queue);
1705 
1706     return ret;
1707 
1708  fail:
1709     g_free(s->image_data_file);
1710     if (has_data_file(bs)) {
1711         bdrv_unref_child(bs, s->data_file);
1712     }
1713     g_free(s->unknown_header_fields);
1714     cleanup_unknown_header_ext(bs);
1715     qcow2_free_snapshots(bs);
1716     qcow2_refcount_close(bs);
1717     qemu_vfree(s->l1_table);
1718     /* else pre-write overlap checks in cache_destroy may crash */
1719     s->l1_table = NULL;
1720     cache_clean_timer_del(bs);
1721     if (s->l2_table_cache) {
1722         qcow2_cache_destroy(s->l2_table_cache);
1723     }
1724     if (s->refcount_block_cache) {
1725         qcow2_cache_destroy(s->refcount_block_cache);
1726     }
1727     qcrypto_block_free(s->crypto);
1728     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1729     return ret;
1730 }
1731 
1732 typedef struct QCow2OpenCo {
1733     BlockDriverState *bs;
1734     QDict *options;
1735     int flags;
1736     Error **errp;
1737     int ret;
1738 } QCow2OpenCo;
1739 
1740 static void coroutine_fn qcow2_open_entry(void *opaque)
1741 {
1742     QCow2OpenCo *qoc = opaque;
1743     BDRVQcow2State *s = qoc->bs->opaque;
1744 
1745     qemu_co_mutex_lock(&s->lock);
1746     qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp);
1747     qemu_co_mutex_unlock(&s->lock);
1748 }
1749 
1750 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1751                       Error **errp)
1752 {
1753     BDRVQcow2State *s = bs->opaque;
1754     QCow2OpenCo qoc = {
1755         .bs = bs,
1756         .options = options,
1757         .flags = flags,
1758         .errp = errp,
1759         .ret = -EINPROGRESS
1760     };
1761 
1762     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
1763                                false, errp);
1764     if (!bs->file) {
1765         return -EINVAL;
1766     }
1767 
1768     /* Initialise locks */
1769     qemu_co_mutex_init(&s->lock);
1770 
1771     if (qemu_in_coroutine()) {
1772         /* From bdrv_co_create.  */
1773         qcow2_open_entry(&qoc);
1774     } else {
1775         assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1776         qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc));
1777         BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
1778     }
1779     return qoc.ret;
1780 }
1781 
1782 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1783 {
1784     BDRVQcow2State *s = bs->opaque;
1785 
1786     if (bs->encrypted) {
1787         /* Encryption works on a sector granularity */
1788         bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto);
1789     }
1790     bs->bl.pwrite_zeroes_alignment = s->cluster_size;
1791     bs->bl.pdiscard_alignment = s->cluster_size;
1792 }
1793 
1794 static int qcow2_reopen_prepare(BDRVReopenState *state,
1795                                 BlockReopenQueue *queue, Error **errp)
1796 {
1797     Qcow2ReopenState *r;
1798     int ret;
1799 
1800     r = g_new0(Qcow2ReopenState, 1);
1801     state->opaque = r;
1802 
1803     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1804                                        state->flags, errp);
1805     if (ret < 0) {
1806         goto fail;
1807     }
1808 
1809     /* We need to write out any unwritten data if we reopen read-only. */
1810     if ((state->flags & BDRV_O_RDWR) == 0) {
1811         ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1812         if (ret < 0) {
1813             goto fail;
1814         }
1815 
1816         ret = bdrv_flush(state->bs);
1817         if (ret < 0) {
1818             goto fail;
1819         }
1820 
1821         ret = qcow2_mark_clean(state->bs);
1822         if (ret < 0) {
1823             goto fail;
1824         }
1825     }
1826 
1827     return 0;
1828 
1829 fail:
1830     qcow2_update_options_abort(state->bs, r);
1831     g_free(r);
1832     return ret;
1833 }
1834 
1835 static void qcow2_reopen_commit(BDRVReopenState *state)
1836 {
1837     qcow2_update_options_commit(state->bs, state->opaque);
1838     g_free(state->opaque);
1839 }
1840 
1841 static void qcow2_reopen_abort(BDRVReopenState *state)
1842 {
1843     qcow2_update_options_abort(state->bs, state->opaque);
1844     g_free(state->opaque);
1845 }
1846 
1847 static void qcow2_join_options(QDict *options, QDict *old_options)
1848 {
1849     bool has_new_overlap_template =
1850         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1851         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1852     bool has_new_total_cache_size =
1853         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1854     bool has_all_cache_options;
1855 
1856     /* New overlap template overrides all old overlap options */
1857     if (has_new_overlap_template) {
1858         qdict_del(old_options, QCOW2_OPT_OVERLAP);
1859         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1860         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1861         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1862         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1863         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1864         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1865         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1866         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1867         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1868     }
1869 
1870     /* New total cache size overrides all old options */
1871     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1872         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1873         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1874     }
1875 
1876     qdict_join(options, old_options, false);
1877 
1878     /*
1879      * If after merging all cache size options are set, an old total size is
1880      * overwritten. Do keep all options, however, if all three are new. The
1881      * resulting error message is what we want to happen.
1882      */
1883     has_all_cache_options =
1884         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1885         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1886         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1887 
1888     if (has_all_cache_options && !has_new_total_cache_size) {
1889         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1890     }
1891 }
1892 
1893 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
1894                                               bool want_zero,
1895                                               int64_t offset, int64_t count,
1896                                               int64_t *pnum, int64_t *map,
1897                                               BlockDriverState **file)
1898 {
1899     BDRVQcow2State *s = bs->opaque;
1900     uint64_t cluster_offset;
1901     int index_in_cluster, ret;
1902     unsigned int bytes;
1903     int status = 0;
1904 
1905     if (!s->metadata_preallocation_checked) {
1906         ret = qcow2_detect_metadata_preallocation(bs);
1907         s->metadata_preallocation = (ret == 1);
1908         s->metadata_preallocation_checked = true;
1909     }
1910 
1911     bytes = MIN(INT_MAX, count);
1912     qemu_co_mutex_lock(&s->lock);
1913     ret = qcow2_get_cluster_offset(bs, offset, &bytes, &cluster_offset);
1914     qemu_co_mutex_unlock(&s->lock);
1915     if (ret < 0) {
1916         return ret;
1917     }
1918 
1919     *pnum = bytes;
1920 
1921     if ((ret == QCOW2_CLUSTER_NORMAL || ret == QCOW2_CLUSTER_ZERO_ALLOC) &&
1922         !s->crypto) {
1923         index_in_cluster = offset & (s->cluster_size - 1);
1924         *map = cluster_offset | index_in_cluster;
1925         *file = s->data_file->bs;
1926         status |= BDRV_BLOCK_OFFSET_VALID;
1927     }
1928     if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) {
1929         status |= BDRV_BLOCK_ZERO;
1930     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1931         status |= BDRV_BLOCK_DATA;
1932     }
1933     if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) &&
1934         (status & BDRV_BLOCK_OFFSET_VALID))
1935     {
1936         status |= BDRV_BLOCK_RECURSE;
1937     }
1938     return status;
1939 }
1940 
1941 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs,
1942                                             QCowL2Meta **pl2meta,
1943                                             bool link_l2)
1944 {
1945     int ret = 0;
1946     QCowL2Meta *l2meta = *pl2meta;
1947 
1948     while (l2meta != NULL) {
1949         QCowL2Meta *next;
1950 
1951         if (link_l2) {
1952             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1953             if (ret) {
1954                 goto out;
1955             }
1956         } else {
1957             qcow2_alloc_cluster_abort(bs, l2meta);
1958         }
1959 
1960         /* Take the request off the list of running requests */
1961         if (l2meta->nb_clusters != 0) {
1962             QLIST_REMOVE(l2meta, next_in_flight);
1963         }
1964 
1965         qemu_co_queue_restart_all(&l2meta->dependent_requests);
1966 
1967         next = l2meta->next;
1968         g_free(l2meta);
1969         l2meta = next;
1970     }
1971 out:
1972     *pl2meta = l2meta;
1973     return ret;
1974 }
1975 
1976 static coroutine_fn int
1977 qcow2_co_preadv_encrypted(BlockDriverState *bs,
1978                            uint64_t file_cluster_offset,
1979                            uint64_t offset,
1980                            uint64_t bytes,
1981                            QEMUIOVector *qiov,
1982                            uint64_t qiov_offset)
1983 {
1984     int ret;
1985     BDRVQcow2State *s = bs->opaque;
1986     uint8_t *buf;
1987 
1988     assert(bs->encrypted && s->crypto);
1989     assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1990 
1991     /*
1992      * For encrypted images, read everything into a temporary
1993      * contiguous buffer on which the AES functions can work.
1994      * Also, decryption in a separate buffer is better as it
1995      * prevents the guest from learning information about the
1996      * encrypted nature of the virtual disk.
1997      */
1998 
1999     buf = qemu_try_blockalign(s->data_file->bs, bytes);
2000     if (buf == NULL) {
2001         return -ENOMEM;
2002     }
2003 
2004     BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2005     ret = bdrv_co_pread(s->data_file,
2006                         file_cluster_offset + offset_into_cluster(s, offset),
2007                         bytes, buf, 0);
2008     if (ret < 0) {
2009         goto fail;
2010     }
2011 
2012     assert(QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE));
2013     assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE));
2014     if (qcow2_co_decrypt(bs,
2015                          file_cluster_offset + offset_into_cluster(s, offset),
2016                          offset, buf, bytes) < 0)
2017     {
2018         ret = -EIO;
2019         goto fail;
2020     }
2021     qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes);
2022 
2023 fail:
2024     qemu_vfree(buf);
2025 
2026     return ret;
2027 }
2028 
2029 typedef struct Qcow2AioTask {
2030     AioTask task;
2031 
2032     BlockDriverState *bs;
2033     QCow2ClusterType cluster_type; /* only for read */
2034     uint64_t file_cluster_offset;
2035     uint64_t offset;
2036     uint64_t bytes;
2037     QEMUIOVector *qiov;
2038     uint64_t qiov_offset;
2039     QCowL2Meta *l2meta; /* only for write */
2040 } Qcow2AioTask;
2041 
2042 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task);
2043 static coroutine_fn int qcow2_add_task(BlockDriverState *bs,
2044                                        AioTaskPool *pool,
2045                                        AioTaskFunc func,
2046                                        QCow2ClusterType cluster_type,
2047                                        uint64_t file_cluster_offset,
2048                                        uint64_t offset,
2049                                        uint64_t bytes,
2050                                        QEMUIOVector *qiov,
2051                                        size_t qiov_offset,
2052                                        QCowL2Meta *l2meta)
2053 {
2054     Qcow2AioTask local_task;
2055     Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task;
2056 
2057     *task = (Qcow2AioTask) {
2058         .task.func = func,
2059         .bs = bs,
2060         .cluster_type = cluster_type,
2061         .qiov = qiov,
2062         .file_cluster_offset = file_cluster_offset,
2063         .offset = offset,
2064         .bytes = bytes,
2065         .qiov_offset = qiov_offset,
2066         .l2meta = l2meta,
2067     };
2068 
2069     trace_qcow2_add_task(qemu_coroutine_self(), bs, pool,
2070                          func == qcow2_co_preadv_task_entry ? "read" : "write",
2071                          cluster_type, file_cluster_offset, offset, bytes,
2072                          qiov, qiov_offset);
2073 
2074     if (!pool) {
2075         return func(&task->task);
2076     }
2077 
2078     aio_task_pool_start_task(pool, &task->task);
2079 
2080     return 0;
2081 }
2082 
2083 static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs,
2084                                              QCow2ClusterType cluster_type,
2085                                              uint64_t file_cluster_offset,
2086                                              uint64_t offset, uint64_t bytes,
2087                                              QEMUIOVector *qiov,
2088                                              size_t qiov_offset)
2089 {
2090     BDRVQcow2State *s = bs->opaque;
2091     int offset_in_cluster = offset_into_cluster(s, offset);
2092 
2093     switch (cluster_type) {
2094     case QCOW2_CLUSTER_ZERO_PLAIN:
2095     case QCOW2_CLUSTER_ZERO_ALLOC:
2096         /* Both zero types are handled in qcow2_co_preadv_part */
2097         g_assert_not_reached();
2098 
2099     case QCOW2_CLUSTER_UNALLOCATED:
2100         assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */
2101 
2102         BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
2103         return bdrv_co_preadv_part(bs->backing, offset, bytes,
2104                                    qiov, qiov_offset, 0);
2105 
2106     case QCOW2_CLUSTER_COMPRESSED:
2107         return qcow2_co_preadv_compressed(bs, file_cluster_offset,
2108                                           offset, bytes, qiov, qiov_offset);
2109 
2110     case QCOW2_CLUSTER_NORMAL:
2111         if ((file_cluster_offset & 511) != 0) {
2112             return -EIO;
2113         }
2114 
2115         if (bs->encrypted) {
2116             return qcow2_co_preadv_encrypted(bs, file_cluster_offset,
2117                                              offset, bytes, qiov, qiov_offset);
2118         }
2119 
2120         BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
2121         return bdrv_co_preadv_part(s->data_file,
2122                                    file_cluster_offset + offset_in_cluster,
2123                                    bytes, qiov, qiov_offset, 0);
2124 
2125     default:
2126         g_assert_not_reached();
2127     }
2128 
2129     g_assert_not_reached();
2130 }
2131 
2132 static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task)
2133 {
2134     Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2135 
2136     assert(!t->l2meta);
2137 
2138     return qcow2_co_preadv_task(t->bs, t->cluster_type, t->file_cluster_offset,
2139                                 t->offset, t->bytes, t->qiov, t->qiov_offset);
2140 }
2141 
2142 static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs,
2143                                              uint64_t offset, uint64_t bytes,
2144                                              QEMUIOVector *qiov,
2145                                              size_t qiov_offset, int flags)
2146 {
2147     BDRVQcow2State *s = bs->opaque;
2148     int ret = 0;
2149     unsigned int cur_bytes; /* number of bytes in current iteration */
2150     uint64_t cluster_offset = 0;
2151     AioTaskPool *aio = NULL;
2152 
2153     while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2154         /* prepare next request */
2155         cur_bytes = MIN(bytes, INT_MAX);
2156         if (s->crypto) {
2157             cur_bytes = MIN(cur_bytes,
2158                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2159         }
2160 
2161         qemu_co_mutex_lock(&s->lock);
2162         ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset);
2163         qemu_co_mutex_unlock(&s->lock);
2164         if (ret < 0) {
2165             goto out;
2166         }
2167 
2168         if (ret == QCOW2_CLUSTER_ZERO_PLAIN ||
2169             ret == QCOW2_CLUSTER_ZERO_ALLOC ||
2170             (ret == QCOW2_CLUSTER_UNALLOCATED && !bs->backing))
2171         {
2172             qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes);
2173         } else {
2174             if (!aio && cur_bytes != bytes) {
2175                 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2176             }
2177             ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, ret,
2178                                  cluster_offset, offset, cur_bytes,
2179                                  qiov, qiov_offset, NULL);
2180             if (ret < 0) {
2181                 goto out;
2182             }
2183         }
2184 
2185         bytes -= cur_bytes;
2186         offset += cur_bytes;
2187         qiov_offset += cur_bytes;
2188     }
2189 
2190 out:
2191     if (aio) {
2192         aio_task_pool_wait_all(aio);
2193         if (ret == 0) {
2194             ret = aio_task_pool_status(aio);
2195         }
2196         g_free(aio);
2197     }
2198 
2199     return ret;
2200 }
2201 
2202 /* Check if it's possible to merge a write request with the writing of
2203  * the data from the COW regions */
2204 static bool merge_cow(uint64_t offset, unsigned bytes,
2205                       QEMUIOVector *qiov, size_t qiov_offset,
2206                       QCowL2Meta *l2meta)
2207 {
2208     QCowL2Meta *m;
2209 
2210     for (m = l2meta; m != NULL; m = m->next) {
2211         /* If both COW regions are empty then there's nothing to merge */
2212         if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
2213             continue;
2214         }
2215 
2216         /* If COW regions are handled already, skip this too */
2217         if (m->skip_cow) {
2218             continue;
2219         }
2220 
2221         /* The data (middle) region must be immediately after the
2222          * start region */
2223         if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
2224             continue;
2225         }
2226 
2227         /* The end region must be immediately after the data (middle)
2228          * region */
2229         if (m->offset + m->cow_end.offset != offset + bytes) {
2230             continue;
2231         }
2232 
2233         /* Make sure that adding both COW regions to the QEMUIOVector
2234          * does not exceed IOV_MAX */
2235         if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) {
2236             continue;
2237         }
2238 
2239         m->data_qiov = qiov;
2240         m->data_qiov_offset = qiov_offset;
2241         return true;
2242     }
2243 
2244     return false;
2245 }
2246 
2247 static bool is_unallocated(BlockDriverState *bs, int64_t offset, int64_t bytes)
2248 {
2249     int64_t nr;
2250     return !bytes ||
2251         (!bdrv_is_allocated_above(bs, NULL, false, offset, bytes, &nr) &&
2252          nr == bytes);
2253 }
2254 
2255 static bool is_zero_cow(BlockDriverState *bs, QCowL2Meta *m)
2256 {
2257     /*
2258      * This check is designed for optimization shortcut so it must be
2259      * efficient.
2260      * Instead of is_zero(), use is_unallocated() as it is faster (but not
2261      * as accurate and can result in false negatives).
2262      */
2263     return is_unallocated(bs, m->offset + m->cow_start.offset,
2264                           m->cow_start.nb_bytes) &&
2265            is_unallocated(bs, m->offset + m->cow_end.offset,
2266                           m->cow_end.nb_bytes);
2267 }
2268 
2269 static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
2270 {
2271     BDRVQcow2State *s = bs->opaque;
2272     QCowL2Meta *m;
2273 
2274     if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) {
2275         return 0;
2276     }
2277 
2278     if (bs->encrypted) {
2279         return 0;
2280     }
2281 
2282     for (m = l2meta; m != NULL; m = m->next) {
2283         int ret;
2284 
2285         if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
2286             continue;
2287         }
2288 
2289         if (!is_zero_cow(bs, m)) {
2290             continue;
2291         }
2292 
2293         /*
2294          * instead of writing zero COW buffers,
2295          * efficiently zero out the whole clusters
2296          */
2297 
2298         ret = qcow2_pre_write_overlap_check(bs, 0, m->alloc_offset,
2299                                             m->nb_clusters * s->cluster_size,
2300                                             true);
2301         if (ret < 0) {
2302             return ret;
2303         }
2304 
2305         BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE);
2306         ret = bdrv_co_pwrite_zeroes(s->data_file, m->alloc_offset,
2307                                     m->nb_clusters * s->cluster_size,
2308                                     BDRV_REQ_NO_FALLBACK);
2309         if (ret < 0) {
2310             if (ret != -ENOTSUP && ret != -EAGAIN) {
2311                 return ret;
2312             }
2313             continue;
2314         }
2315 
2316         trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters);
2317         m->skip_cow = true;
2318     }
2319     return 0;
2320 }
2321 
2322 /*
2323  * qcow2_co_pwritev_task
2324  * Called with s->lock unlocked
2325  * l2meta  - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2326  *           not use it somehow after qcow2_co_pwritev_task() call
2327  */
2328 static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs,
2329                                               uint64_t file_cluster_offset,
2330                                               uint64_t offset, uint64_t bytes,
2331                                               QEMUIOVector *qiov,
2332                                               uint64_t qiov_offset,
2333                                               QCowL2Meta *l2meta)
2334 {
2335     int ret;
2336     BDRVQcow2State *s = bs->opaque;
2337     void *crypt_buf = NULL;
2338     int offset_in_cluster = offset_into_cluster(s, offset);
2339     QEMUIOVector encrypted_qiov;
2340 
2341     if (bs->encrypted) {
2342         assert(s->crypto);
2343         assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2344         crypt_buf = qemu_try_blockalign(bs->file->bs, bytes);
2345         if (crypt_buf == NULL) {
2346             ret = -ENOMEM;
2347             goto out_unlocked;
2348         }
2349         qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes);
2350 
2351         if (qcow2_co_encrypt(bs, file_cluster_offset + offset_in_cluster,
2352                              offset, crypt_buf, bytes) < 0)
2353         {
2354             ret = -EIO;
2355             goto out_unlocked;
2356         }
2357 
2358         qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes);
2359         qiov = &encrypted_qiov;
2360         qiov_offset = 0;
2361     }
2362 
2363     /* Try to efficiently initialize the physical space with zeroes */
2364     ret = handle_alloc_space(bs, l2meta);
2365     if (ret < 0) {
2366         goto out_unlocked;
2367     }
2368 
2369     /*
2370      * If we need to do COW, check if it's possible to merge the
2371      * writing of the guest data together with that of the COW regions.
2372      * If it's not possible (or not necessary) then write the
2373      * guest data now.
2374      */
2375     if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) {
2376         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
2377         trace_qcow2_writev_data(qemu_coroutine_self(),
2378                                 file_cluster_offset + offset_in_cluster);
2379         ret = bdrv_co_pwritev_part(s->data_file,
2380                                    file_cluster_offset + offset_in_cluster,
2381                                    bytes, qiov, qiov_offset, 0);
2382         if (ret < 0) {
2383             goto out_unlocked;
2384         }
2385     }
2386 
2387     qemu_co_mutex_lock(&s->lock);
2388 
2389     ret = qcow2_handle_l2meta(bs, &l2meta, true);
2390     goto out_locked;
2391 
2392 out_unlocked:
2393     qemu_co_mutex_lock(&s->lock);
2394 
2395 out_locked:
2396     qcow2_handle_l2meta(bs, &l2meta, false);
2397     qemu_co_mutex_unlock(&s->lock);
2398 
2399     qemu_vfree(crypt_buf);
2400 
2401     return ret;
2402 }
2403 
2404 static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task)
2405 {
2406     Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2407 
2408     assert(!t->cluster_type);
2409 
2410     return qcow2_co_pwritev_task(t->bs, t->file_cluster_offset,
2411                                  t->offset, t->bytes, t->qiov, t->qiov_offset,
2412                                  t->l2meta);
2413 }
2414 
2415 static coroutine_fn int qcow2_co_pwritev_part(
2416         BlockDriverState *bs, uint64_t offset, uint64_t bytes,
2417         QEMUIOVector *qiov, size_t qiov_offset, int flags)
2418 {
2419     BDRVQcow2State *s = bs->opaque;
2420     int offset_in_cluster;
2421     int ret;
2422     unsigned int cur_bytes; /* number of sectors in current iteration */
2423     uint64_t cluster_offset;
2424     QCowL2Meta *l2meta = NULL;
2425     AioTaskPool *aio = NULL;
2426 
2427     trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
2428 
2429     while (bytes != 0 && aio_task_pool_status(aio) == 0) {
2430 
2431         l2meta = NULL;
2432 
2433         trace_qcow2_writev_start_part(qemu_coroutine_self());
2434         offset_in_cluster = offset_into_cluster(s, offset);
2435         cur_bytes = MIN(bytes, INT_MAX);
2436         if (bs->encrypted) {
2437             cur_bytes = MIN(cur_bytes,
2438                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
2439                             - offset_in_cluster);
2440         }
2441 
2442         qemu_co_mutex_lock(&s->lock);
2443 
2444         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2445                                          &cluster_offset, &l2meta);
2446         if (ret < 0) {
2447             goto out_locked;
2448         }
2449 
2450         assert((cluster_offset & 511) == 0);
2451 
2452         ret = qcow2_pre_write_overlap_check(bs, 0,
2453                                             cluster_offset + offset_in_cluster,
2454                                             cur_bytes, true);
2455         if (ret < 0) {
2456             goto out_locked;
2457         }
2458 
2459         qemu_co_mutex_unlock(&s->lock);
2460 
2461         if (!aio && cur_bytes != bytes) {
2462             aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2463         }
2464         ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0,
2465                              cluster_offset, offset, cur_bytes,
2466                              qiov, qiov_offset, l2meta);
2467         l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */
2468         if (ret < 0) {
2469             goto fail_nometa;
2470         }
2471 
2472         bytes -= cur_bytes;
2473         offset += cur_bytes;
2474         qiov_offset += cur_bytes;
2475         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2476     }
2477     ret = 0;
2478 
2479     qemu_co_mutex_lock(&s->lock);
2480 
2481 out_locked:
2482     qcow2_handle_l2meta(bs, &l2meta, false);
2483 
2484     qemu_co_mutex_unlock(&s->lock);
2485 
2486 fail_nometa:
2487     if (aio) {
2488         aio_task_pool_wait_all(aio);
2489         if (ret == 0) {
2490             ret = aio_task_pool_status(aio);
2491         }
2492         g_free(aio);
2493     }
2494 
2495     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2496 
2497     return ret;
2498 }
2499 
2500 static int qcow2_inactivate(BlockDriverState *bs)
2501 {
2502     BDRVQcow2State *s = bs->opaque;
2503     int ret, result = 0;
2504     Error *local_err = NULL;
2505 
2506     qcow2_store_persistent_dirty_bitmaps(bs, &local_err);
2507     if (local_err != NULL) {
2508         result = -EINVAL;
2509         error_reportf_err(local_err, "Lost persistent bitmaps during "
2510                           "inactivation of node '%s': ",
2511                           bdrv_get_device_or_node_name(bs));
2512     }
2513 
2514     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2515     if (ret) {
2516         result = ret;
2517         error_report("Failed to flush the L2 table cache: %s",
2518                      strerror(-ret));
2519     }
2520 
2521     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2522     if (ret) {
2523         result = ret;
2524         error_report("Failed to flush the refcount block cache: %s",
2525                      strerror(-ret));
2526     }
2527 
2528     if (result == 0) {
2529         qcow2_mark_clean(bs);
2530     }
2531 
2532     return result;
2533 }
2534 
2535 static void qcow2_close(BlockDriverState *bs)
2536 {
2537     BDRVQcow2State *s = bs->opaque;
2538     qemu_vfree(s->l1_table);
2539     /* else pre-write overlap checks in cache_destroy may crash */
2540     s->l1_table = NULL;
2541 
2542     if (!(s->flags & BDRV_O_INACTIVE)) {
2543         qcow2_inactivate(bs);
2544     }
2545 
2546     cache_clean_timer_del(bs);
2547     qcow2_cache_destroy(s->l2_table_cache);
2548     qcow2_cache_destroy(s->refcount_block_cache);
2549 
2550     qcrypto_block_free(s->crypto);
2551     s->crypto = NULL;
2552 
2553     g_free(s->unknown_header_fields);
2554     cleanup_unknown_header_ext(bs);
2555 
2556     g_free(s->image_data_file);
2557     g_free(s->image_backing_file);
2558     g_free(s->image_backing_format);
2559 
2560     if (has_data_file(bs)) {
2561         bdrv_unref_child(bs, s->data_file);
2562     }
2563 
2564     qcow2_refcount_close(bs);
2565     qcow2_free_snapshots(bs);
2566 }
2567 
2568 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
2569                                                    Error **errp)
2570 {
2571     BDRVQcow2State *s = bs->opaque;
2572     int flags = s->flags;
2573     QCryptoBlock *crypto = NULL;
2574     QDict *options;
2575     Error *local_err = NULL;
2576     int ret;
2577 
2578     /*
2579      * Backing files are read-only which makes all of their metadata immutable,
2580      * that means we don't have to worry about reopening them here.
2581      */
2582 
2583     crypto = s->crypto;
2584     s->crypto = NULL;
2585 
2586     qcow2_close(bs);
2587 
2588     memset(s, 0, sizeof(BDRVQcow2State));
2589     options = qdict_clone_shallow(bs->options);
2590 
2591     flags &= ~BDRV_O_INACTIVE;
2592     qemu_co_mutex_lock(&s->lock);
2593     ret = qcow2_do_open(bs, options, flags, &local_err);
2594     qemu_co_mutex_unlock(&s->lock);
2595     qobject_unref(options);
2596     if (local_err) {
2597         error_propagate_prepend(errp, local_err,
2598                                 "Could not reopen qcow2 layer: ");
2599         bs->drv = NULL;
2600         return;
2601     } else if (ret < 0) {
2602         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
2603         bs->drv = NULL;
2604         return;
2605     }
2606 
2607     s->crypto = crypto;
2608 }
2609 
2610 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2611     size_t len, size_t buflen)
2612 {
2613     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2614     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2615 
2616     if (buflen < ext_len) {
2617         return -ENOSPC;
2618     }
2619 
2620     *ext_backing_fmt = (QCowExtension) {
2621         .magic  = cpu_to_be32(magic),
2622         .len    = cpu_to_be32(len),
2623     };
2624 
2625     if (len) {
2626         memcpy(buf + sizeof(QCowExtension), s, len);
2627     }
2628 
2629     return ext_len;
2630 }
2631 
2632 /*
2633  * Updates the qcow2 header, including the variable length parts of it, i.e.
2634  * the backing file name and all extensions. qcow2 was not designed to allow
2635  * such changes, so if we run out of space (we can only use the first cluster)
2636  * this function may fail.
2637  *
2638  * Returns 0 on success, -errno in error cases.
2639  */
2640 int qcow2_update_header(BlockDriverState *bs)
2641 {
2642     BDRVQcow2State *s = bs->opaque;
2643     QCowHeader *header;
2644     char *buf;
2645     size_t buflen = s->cluster_size;
2646     int ret;
2647     uint64_t total_size;
2648     uint32_t refcount_table_clusters;
2649     size_t header_length;
2650     Qcow2UnknownHeaderExtension *uext;
2651 
2652     buf = qemu_blockalign(bs, buflen);
2653 
2654     /* Header structure */
2655     header = (QCowHeader*) buf;
2656 
2657     if (buflen < sizeof(*header)) {
2658         ret = -ENOSPC;
2659         goto fail;
2660     }
2661 
2662     header_length = sizeof(*header) + s->unknown_header_fields_size;
2663     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2664     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2665 
2666     *header = (QCowHeader) {
2667         /* Version 2 fields */
2668         .magic                  = cpu_to_be32(QCOW_MAGIC),
2669         .version                = cpu_to_be32(s->qcow_version),
2670         .backing_file_offset    = 0,
2671         .backing_file_size      = 0,
2672         .cluster_bits           = cpu_to_be32(s->cluster_bits),
2673         .size                   = cpu_to_be64(total_size),
2674         .crypt_method           = cpu_to_be32(s->crypt_method_header),
2675         .l1_size                = cpu_to_be32(s->l1_size),
2676         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
2677         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
2678         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2679         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
2680         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
2681 
2682         /* Version 3 fields */
2683         .incompatible_features  = cpu_to_be64(s->incompatible_features),
2684         .compatible_features    = cpu_to_be64(s->compatible_features),
2685         .autoclear_features     = cpu_to_be64(s->autoclear_features),
2686         .refcount_order         = cpu_to_be32(s->refcount_order),
2687         .header_length          = cpu_to_be32(header_length),
2688     };
2689 
2690     /* For older versions, write a shorter header */
2691     switch (s->qcow_version) {
2692     case 2:
2693         ret = offsetof(QCowHeader, incompatible_features);
2694         break;
2695     case 3:
2696         ret = sizeof(*header);
2697         break;
2698     default:
2699         ret = -EINVAL;
2700         goto fail;
2701     }
2702 
2703     buf += ret;
2704     buflen -= ret;
2705     memset(buf, 0, buflen);
2706 
2707     /* Preserve any unknown field in the header */
2708     if (s->unknown_header_fields_size) {
2709         if (buflen < s->unknown_header_fields_size) {
2710             ret = -ENOSPC;
2711             goto fail;
2712         }
2713 
2714         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2715         buf += s->unknown_header_fields_size;
2716         buflen -= s->unknown_header_fields_size;
2717     }
2718 
2719     /* Backing file format header extension */
2720     if (s->image_backing_format) {
2721         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
2722                              s->image_backing_format,
2723                              strlen(s->image_backing_format),
2724                              buflen);
2725         if (ret < 0) {
2726             goto fail;
2727         }
2728 
2729         buf += ret;
2730         buflen -= ret;
2731     }
2732 
2733     /* External data file header extension */
2734     if (has_data_file(bs) && s->image_data_file) {
2735         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE,
2736                              s->image_data_file, strlen(s->image_data_file),
2737                              buflen);
2738         if (ret < 0) {
2739             goto fail;
2740         }
2741 
2742         buf += ret;
2743         buflen -= ret;
2744     }
2745 
2746     /* Full disk encryption header pointer extension */
2747     if (s->crypto_header.offset != 0) {
2748         s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset);
2749         s->crypto_header.length = cpu_to_be64(s->crypto_header.length);
2750         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
2751                              &s->crypto_header, sizeof(s->crypto_header),
2752                              buflen);
2753         s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
2754         s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
2755         if (ret < 0) {
2756             goto fail;
2757         }
2758         buf += ret;
2759         buflen -= ret;
2760     }
2761 
2762     /* Feature table */
2763     if (s->qcow_version >= 3) {
2764         Qcow2Feature features[] = {
2765             {
2766                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2767                 .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
2768                 .name = "dirty bit",
2769             },
2770             {
2771                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2772                 .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
2773                 .name = "corrupt bit",
2774             },
2775             {
2776                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2777                 .bit  = QCOW2_INCOMPAT_DATA_FILE_BITNR,
2778                 .name = "external data file",
2779             },
2780             {
2781                 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
2782                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
2783                 .name = "lazy refcounts",
2784             },
2785         };
2786 
2787         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
2788                              features, sizeof(features), buflen);
2789         if (ret < 0) {
2790             goto fail;
2791         }
2792         buf += ret;
2793         buflen -= ret;
2794     }
2795 
2796     /* Bitmap extension */
2797     if (s->nb_bitmaps > 0) {
2798         Qcow2BitmapHeaderExt bitmaps_header = {
2799             .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
2800             .bitmap_directory_size =
2801                     cpu_to_be64(s->bitmap_directory_size),
2802             .bitmap_directory_offset =
2803                     cpu_to_be64(s->bitmap_directory_offset)
2804         };
2805         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
2806                              &bitmaps_header, sizeof(bitmaps_header),
2807                              buflen);
2808         if (ret < 0) {
2809             goto fail;
2810         }
2811         buf += ret;
2812         buflen -= ret;
2813     }
2814 
2815     /* Keep unknown header extensions */
2816     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
2817         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
2818         if (ret < 0) {
2819             goto fail;
2820         }
2821 
2822         buf += ret;
2823         buflen -= ret;
2824     }
2825 
2826     /* End of header extensions */
2827     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
2828     if (ret < 0) {
2829         goto fail;
2830     }
2831 
2832     buf += ret;
2833     buflen -= ret;
2834 
2835     /* Backing file name */
2836     if (s->image_backing_file) {
2837         size_t backing_file_len = strlen(s->image_backing_file);
2838 
2839         if (buflen < backing_file_len) {
2840             ret = -ENOSPC;
2841             goto fail;
2842         }
2843 
2844         /* Using strncpy is ok here, since buf is not NUL-terminated. */
2845         strncpy(buf, s->image_backing_file, buflen);
2846 
2847         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
2848         header->backing_file_size   = cpu_to_be32(backing_file_len);
2849     }
2850 
2851     /* Write the new header */
2852     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
2853     if (ret < 0) {
2854         goto fail;
2855     }
2856 
2857     ret = 0;
2858 fail:
2859     qemu_vfree(header);
2860     return ret;
2861 }
2862 
2863 static int qcow2_change_backing_file(BlockDriverState *bs,
2864     const char *backing_file, const char *backing_fmt)
2865 {
2866     BDRVQcow2State *s = bs->opaque;
2867 
2868     /* Adding a backing file means that the external data file alone won't be
2869      * enough to make sense of the content */
2870     if (backing_file && data_file_is_raw(bs)) {
2871         return -EINVAL;
2872     }
2873 
2874     if (backing_file && strlen(backing_file) > 1023) {
2875         return -EINVAL;
2876     }
2877 
2878     pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2879             backing_file ?: "");
2880     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2881     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2882 
2883     g_free(s->image_backing_file);
2884     g_free(s->image_backing_format);
2885 
2886     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
2887     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
2888 
2889     return qcow2_update_header(bs);
2890 }
2891 
2892 static int qcow2_crypt_method_from_format(const char *encryptfmt)
2893 {
2894     if (g_str_equal(encryptfmt, "luks")) {
2895         return QCOW_CRYPT_LUKS;
2896     } else if (g_str_equal(encryptfmt, "aes")) {
2897         return QCOW_CRYPT_AES;
2898     } else {
2899         return -EINVAL;
2900     }
2901 }
2902 
2903 static int qcow2_set_up_encryption(BlockDriverState *bs,
2904                                    QCryptoBlockCreateOptions *cryptoopts,
2905                                    Error **errp)
2906 {
2907     BDRVQcow2State *s = bs->opaque;
2908     QCryptoBlock *crypto = NULL;
2909     int fmt, ret;
2910 
2911     switch (cryptoopts->format) {
2912     case Q_CRYPTO_BLOCK_FORMAT_LUKS:
2913         fmt = QCOW_CRYPT_LUKS;
2914         break;
2915     case Q_CRYPTO_BLOCK_FORMAT_QCOW:
2916         fmt = QCOW_CRYPT_AES;
2917         break;
2918     default:
2919         error_setg(errp, "Crypto format not supported in qcow2");
2920         return -EINVAL;
2921     }
2922 
2923     s->crypt_method_header = fmt;
2924 
2925     crypto = qcrypto_block_create(cryptoopts, "encrypt.",
2926                                   qcow2_crypto_hdr_init_func,
2927                                   qcow2_crypto_hdr_write_func,
2928                                   bs, errp);
2929     if (!crypto) {
2930         return -EINVAL;
2931     }
2932 
2933     ret = qcow2_update_header(bs);
2934     if (ret < 0) {
2935         error_setg_errno(errp, -ret, "Could not write encryption header");
2936         goto out;
2937     }
2938 
2939     ret = 0;
2940  out:
2941     qcrypto_block_free(crypto);
2942     return ret;
2943 }
2944 
2945 /**
2946  * Preallocates metadata structures for data clusters between @offset (in the
2947  * guest disk) and @new_length (which is thus generally the new guest disk
2948  * size).
2949  *
2950  * Returns: 0 on success, -errno on failure.
2951  */
2952 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset,
2953                                        uint64_t new_length, PreallocMode mode,
2954                                        Error **errp)
2955 {
2956     BDRVQcow2State *s = bs->opaque;
2957     uint64_t bytes;
2958     uint64_t host_offset = 0;
2959     int64_t file_length;
2960     unsigned int cur_bytes;
2961     int ret;
2962     QCowL2Meta *meta;
2963 
2964     assert(offset <= new_length);
2965     bytes = new_length - offset;
2966 
2967     while (bytes) {
2968         cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size));
2969         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2970                                          &host_offset, &meta);
2971         if (ret < 0) {
2972             error_setg_errno(errp, -ret, "Allocating clusters failed");
2973             return ret;
2974         }
2975 
2976         while (meta) {
2977             QCowL2Meta *next = meta->next;
2978 
2979             ret = qcow2_alloc_cluster_link_l2(bs, meta);
2980             if (ret < 0) {
2981                 error_setg_errno(errp, -ret, "Mapping clusters failed");
2982                 qcow2_free_any_clusters(bs, meta->alloc_offset,
2983                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
2984                 return ret;
2985             }
2986 
2987             /* There are no dependent requests, but we need to remove our
2988              * request from the list of in-flight requests */
2989             QLIST_REMOVE(meta, next_in_flight);
2990 
2991             g_free(meta);
2992             meta = next;
2993         }
2994 
2995         /* TODO Preallocate data if requested */
2996 
2997         bytes -= cur_bytes;
2998         offset += cur_bytes;
2999     }
3000 
3001     /*
3002      * It is expected that the image file is large enough to actually contain
3003      * all of the allocated clusters (otherwise we get failing reads after
3004      * EOF). Extend the image to the last allocated sector.
3005      */
3006     file_length = bdrv_getlength(s->data_file->bs);
3007     if (file_length < 0) {
3008         error_setg_errno(errp, -file_length, "Could not get file size");
3009         return file_length;
3010     }
3011 
3012     if (host_offset + cur_bytes > file_length) {
3013         if (mode == PREALLOC_MODE_METADATA) {
3014             mode = PREALLOC_MODE_OFF;
3015         }
3016         ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, mode,
3017                                errp);
3018         if (ret < 0) {
3019             return ret;
3020         }
3021     }
3022 
3023     return 0;
3024 }
3025 
3026 /* qcow2_refcount_metadata_size:
3027  * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3028  * @cluster_size: size of a cluster, in bytes
3029  * @refcount_order: refcount bits power-of-2 exponent
3030  * @generous_increase: allow for the refcount table to be 1.5x as large as it
3031  *                     needs to be
3032  *
3033  * Returns: Number of bytes required for refcount blocks and table metadata.
3034  */
3035 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
3036                                      int refcount_order, bool generous_increase,
3037                                      uint64_t *refblock_count)
3038 {
3039     /*
3040      * Every host cluster is reference-counted, including metadata (even
3041      * refcount metadata is recursively included).
3042      *
3043      * An accurate formula for the size of refcount metadata size is difficult
3044      * to derive.  An easier method of calculation is finding the fixed point
3045      * where no further refcount blocks or table clusters are required to
3046      * reference count every cluster.
3047      */
3048     int64_t blocks_per_table_cluster = cluster_size / sizeof(uint64_t);
3049     int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
3050     int64_t table = 0;  /* number of refcount table clusters */
3051     int64_t blocks = 0; /* number of refcount block clusters */
3052     int64_t last;
3053     int64_t n = 0;
3054 
3055     do {
3056         last = n;
3057         blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
3058         table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
3059         n = clusters + blocks + table;
3060 
3061         if (n == last && generous_increase) {
3062             clusters += DIV_ROUND_UP(table, 2);
3063             n = 0; /* force another loop */
3064             generous_increase = false;
3065         }
3066     } while (n != last);
3067 
3068     if (refblock_count) {
3069         *refblock_count = blocks;
3070     }
3071 
3072     return (blocks + table) * cluster_size;
3073 }
3074 
3075 /**
3076  * qcow2_calc_prealloc_size:
3077  * @total_size: virtual disk size in bytes
3078  * @cluster_size: cluster size in bytes
3079  * @refcount_order: refcount bits power-of-2 exponent
3080  *
3081  * Returns: Total number of bytes required for the fully allocated image
3082  * (including metadata).
3083  */
3084 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
3085                                         size_t cluster_size,
3086                                         int refcount_order)
3087 {
3088     int64_t meta_size = 0;
3089     uint64_t nl1e, nl2e;
3090     int64_t aligned_total_size = ROUND_UP(total_size, cluster_size);
3091 
3092     /* header: 1 cluster */
3093     meta_size += cluster_size;
3094 
3095     /* total size of L2 tables */
3096     nl2e = aligned_total_size / cluster_size;
3097     nl2e = ROUND_UP(nl2e, cluster_size / sizeof(uint64_t));
3098     meta_size += nl2e * sizeof(uint64_t);
3099 
3100     /* total size of L1 tables */
3101     nl1e = nl2e * sizeof(uint64_t) / cluster_size;
3102     nl1e = ROUND_UP(nl1e, cluster_size / sizeof(uint64_t));
3103     meta_size += nl1e * sizeof(uint64_t);
3104 
3105     /* total size of refcount table and blocks */
3106     meta_size += qcow2_refcount_metadata_size(
3107             (meta_size + aligned_total_size) / cluster_size,
3108             cluster_size, refcount_order, false, NULL);
3109 
3110     return meta_size + aligned_total_size;
3111 }
3112 
3113 static bool validate_cluster_size(size_t cluster_size, Error **errp)
3114 {
3115     int cluster_bits = ctz32(cluster_size);
3116     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
3117         (1 << cluster_bits) != cluster_size)
3118     {
3119         error_setg(errp, "Cluster size must be a power of two between %d and "
3120                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
3121         return false;
3122     }
3123     return true;
3124 }
3125 
3126 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, Error **errp)
3127 {
3128     size_t cluster_size;
3129 
3130     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
3131                                          DEFAULT_CLUSTER_SIZE);
3132     if (!validate_cluster_size(cluster_size, errp)) {
3133         return 0;
3134     }
3135     return cluster_size;
3136 }
3137 
3138 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
3139 {
3140     char *buf;
3141     int ret;
3142 
3143     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
3144     if (!buf) {
3145         ret = 3; /* default */
3146     } else if (!strcmp(buf, "0.10")) {
3147         ret = 2;
3148     } else if (!strcmp(buf, "1.1")) {
3149         ret = 3;
3150     } else {
3151         error_setg(errp, "Invalid compatibility level: '%s'", buf);
3152         ret = -EINVAL;
3153     }
3154     g_free(buf);
3155     return ret;
3156 }
3157 
3158 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
3159                                                 Error **errp)
3160 {
3161     uint64_t refcount_bits;
3162 
3163     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
3164     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
3165         error_setg(errp, "Refcount width must be a power of two and may not "
3166                    "exceed 64 bits");
3167         return 0;
3168     }
3169 
3170     if (version < 3 && refcount_bits != 16) {
3171         error_setg(errp, "Different refcount widths than 16 bits require "
3172                    "compatibility level 1.1 or above (use compat=1.1 or "
3173                    "greater)");
3174         return 0;
3175     }
3176 
3177     return refcount_bits;
3178 }
3179 
3180 static int coroutine_fn
3181 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
3182 {
3183     BlockdevCreateOptionsQcow2 *qcow2_opts;
3184     QDict *options;
3185 
3186     /*
3187      * Open the image file and write a minimal qcow2 header.
3188      *
3189      * We keep things simple and start with a zero-sized image. We also
3190      * do without refcount blocks or a L1 table for now. We'll fix the
3191      * inconsistency later.
3192      *
3193      * We do need a refcount table because growing the refcount table means
3194      * allocating two new refcount blocks - the seconds of which would be at
3195      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3196      * size for any qcow2 image.
3197      */
3198     BlockBackend *blk = NULL;
3199     BlockDriverState *bs = NULL;
3200     BlockDriverState *data_bs = NULL;
3201     QCowHeader *header;
3202     size_t cluster_size;
3203     int version;
3204     int refcount_order;
3205     uint64_t* refcount_table;
3206     Error *local_err = NULL;
3207     int ret;
3208 
3209     assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2);
3210     qcow2_opts = &create_options->u.qcow2;
3211 
3212     bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp);
3213     if (bs == NULL) {
3214         return -EIO;
3215     }
3216 
3217     /* Validate options and set default values */
3218     if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) {
3219         error_setg(errp, "Image size must be a multiple of 512 bytes");
3220         ret = -EINVAL;
3221         goto out;
3222     }
3223 
3224     if (qcow2_opts->has_version) {
3225         switch (qcow2_opts->version) {
3226         case BLOCKDEV_QCOW2_VERSION_V2:
3227             version = 2;
3228             break;
3229         case BLOCKDEV_QCOW2_VERSION_V3:
3230             version = 3;
3231             break;
3232         default:
3233             g_assert_not_reached();
3234         }
3235     } else {
3236         version = 3;
3237     }
3238 
3239     if (qcow2_opts->has_cluster_size) {
3240         cluster_size = qcow2_opts->cluster_size;
3241     } else {
3242         cluster_size = DEFAULT_CLUSTER_SIZE;
3243     }
3244 
3245     if (!validate_cluster_size(cluster_size, errp)) {
3246         ret = -EINVAL;
3247         goto out;
3248     }
3249 
3250     if (!qcow2_opts->has_preallocation) {
3251         qcow2_opts->preallocation = PREALLOC_MODE_OFF;
3252     }
3253     if (qcow2_opts->has_backing_file &&
3254         qcow2_opts->preallocation != PREALLOC_MODE_OFF)
3255     {
3256         error_setg(errp, "Backing file and preallocation cannot be used at "
3257                    "the same time");
3258         ret = -EINVAL;
3259         goto out;
3260     }
3261     if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) {
3262         error_setg(errp, "Backing format cannot be used without backing file");
3263         ret = -EINVAL;
3264         goto out;
3265     }
3266 
3267     if (!qcow2_opts->has_lazy_refcounts) {
3268         qcow2_opts->lazy_refcounts = false;
3269     }
3270     if (version < 3 && qcow2_opts->lazy_refcounts) {
3271         error_setg(errp, "Lazy refcounts only supported with compatibility "
3272                    "level 1.1 and above (use version=v3 or greater)");
3273         ret = -EINVAL;
3274         goto out;
3275     }
3276 
3277     if (!qcow2_opts->has_refcount_bits) {
3278         qcow2_opts->refcount_bits = 16;
3279     }
3280     if (qcow2_opts->refcount_bits > 64 ||
3281         !is_power_of_2(qcow2_opts->refcount_bits))
3282     {
3283         error_setg(errp, "Refcount width must be a power of two and may not "
3284                    "exceed 64 bits");
3285         ret = -EINVAL;
3286         goto out;
3287     }
3288     if (version < 3 && qcow2_opts->refcount_bits != 16) {
3289         error_setg(errp, "Different refcount widths than 16 bits require "
3290                    "compatibility level 1.1 or above (use version=v3 or "
3291                    "greater)");
3292         ret = -EINVAL;
3293         goto out;
3294     }
3295     refcount_order = ctz32(qcow2_opts->refcount_bits);
3296 
3297     if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) {
3298         error_setg(errp, "data-file-raw requires data-file");
3299         ret = -EINVAL;
3300         goto out;
3301     }
3302     if (qcow2_opts->data_file_raw && qcow2_opts->has_backing_file) {
3303         error_setg(errp, "Backing file and data-file-raw cannot be used at "
3304                    "the same time");
3305         ret = -EINVAL;
3306         goto out;
3307     }
3308 
3309     if (qcow2_opts->data_file) {
3310         if (version < 3) {
3311             error_setg(errp, "External data files are only supported with "
3312                        "compatibility level 1.1 and above (use version=v3 or "
3313                        "greater)");
3314             ret = -EINVAL;
3315             goto out;
3316         }
3317         data_bs = bdrv_open_blockdev_ref(qcow2_opts->data_file, errp);
3318         if (data_bs == NULL) {
3319             ret = -EIO;
3320             goto out;
3321         }
3322     }
3323 
3324     /* Create BlockBackend to write to the image */
3325     blk = blk_new(bdrv_get_aio_context(bs),
3326                   BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
3327     ret = blk_insert_bs(blk, bs, errp);
3328     if (ret < 0) {
3329         goto out;
3330     }
3331     blk_set_allow_write_beyond_eof(blk, true);
3332 
3333     /* Clear the protocol layer and preallocate it if necessary */
3334     ret = blk_truncate(blk, 0, PREALLOC_MODE_OFF, errp);
3335     if (ret < 0) {
3336         goto out;
3337     }
3338 
3339     /* Write the header */
3340     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
3341     header = g_malloc0(cluster_size);
3342     *header = (QCowHeader) {
3343         .magic                      = cpu_to_be32(QCOW_MAGIC),
3344         .version                    = cpu_to_be32(version),
3345         .cluster_bits               = cpu_to_be32(ctz32(cluster_size)),
3346         .size                       = cpu_to_be64(0),
3347         .l1_table_offset            = cpu_to_be64(0),
3348         .l1_size                    = cpu_to_be32(0),
3349         .refcount_table_offset      = cpu_to_be64(cluster_size),
3350         .refcount_table_clusters    = cpu_to_be32(1),
3351         .refcount_order             = cpu_to_be32(refcount_order),
3352         .header_length              = cpu_to_be32(sizeof(*header)),
3353     };
3354 
3355     /* We'll update this to correct value later */
3356     header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
3357 
3358     if (qcow2_opts->lazy_refcounts) {
3359         header->compatible_features |=
3360             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
3361     }
3362     if (data_bs) {
3363         header->incompatible_features |=
3364             cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE);
3365     }
3366     if (qcow2_opts->data_file_raw) {
3367         header->autoclear_features |=
3368             cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW);
3369     }
3370 
3371     ret = blk_pwrite(blk, 0, header, cluster_size, 0);
3372     g_free(header);
3373     if (ret < 0) {
3374         error_setg_errno(errp, -ret, "Could not write qcow2 header");
3375         goto out;
3376     }
3377 
3378     /* Write a refcount table with one refcount block */
3379     refcount_table = g_malloc0(2 * cluster_size);
3380     refcount_table[0] = cpu_to_be64(2 * cluster_size);
3381     ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
3382     g_free(refcount_table);
3383 
3384     if (ret < 0) {
3385         error_setg_errno(errp, -ret, "Could not write refcount table");
3386         goto out;
3387     }
3388 
3389     blk_unref(blk);
3390     blk = NULL;
3391 
3392     /*
3393      * And now open the image and make it consistent first (i.e. increase the
3394      * refcount of the cluster that is occupied by the header and the refcount
3395      * table)
3396      */
3397     options = qdict_new();
3398     qdict_put_str(options, "driver", "qcow2");
3399     qdict_put_str(options, "file", bs->node_name);
3400     if (data_bs) {
3401         qdict_put_str(options, "data-file", data_bs->node_name);
3402     }
3403     blk = blk_new_open(NULL, NULL, options,
3404                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
3405                        &local_err);
3406     if (blk == NULL) {
3407         error_propagate(errp, local_err);
3408         ret = -EIO;
3409         goto out;
3410     }
3411 
3412     ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
3413     if (ret < 0) {
3414         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
3415                          "header and refcount table");
3416         goto out;
3417 
3418     } else if (ret != 0) {
3419         error_report("Huh, first cluster in empty image is already in use?");
3420         abort();
3421     }
3422 
3423     /* Set the external data file if necessary */
3424     if (data_bs) {
3425         BDRVQcow2State *s = blk_bs(blk)->opaque;
3426         s->image_data_file = g_strdup(data_bs->filename);
3427     }
3428 
3429     /* Create a full header (including things like feature table) */
3430     ret = qcow2_update_header(blk_bs(blk));
3431     if (ret < 0) {
3432         error_setg_errno(errp, -ret, "Could not update qcow2 header");
3433         goto out;
3434     }
3435 
3436     /* Okay, now that we have a valid image, let's give it the right size */
3437     ret = blk_truncate(blk, qcow2_opts->size, qcow2_opts->preallocation, errp);
3438     if (ret < 0) {
3439         error_prepend(errp, "Could not resize image: ");
3440         goto out;
3441     }
3442 
3443     /* Want a backing file? There you go.*/
3444     if (qcow2_opts->has_backing_file) {
3445         const char *backing_format = NULL;
3446 
3447         if (qcow2_opts->has_backing_fmt) {
3448             backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt);
3449         }
3450 
3451         ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file,
3452                                        backing_format);
3453         if (ret < 0) {
3454             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
3455                              "with format '%s'", qcow2_opts->backing_file,
3456                              backing_format);
3457             goto out;
3458         }
3459     }
3460 
3461     /* Want encryption? There you go. */
3462     if (qcow2_opts->has_encrypt) {
3463         ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp);
3464         if (ret < 0) {
3465             goto out;
3466         }
3467     }
3468 
3469     blk_unref(blk);
3470     blk = NULL;
3471 
3472     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3473      * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3474      * have to setup decryption context. We're not doing any I/O on the top
3475      * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3476      * not have effect.
3477      */
3478     options = qdict_new();
3479     qdict_put_str(options, "driver", "qcow2");
3480     qdict_put_str(options, "file", bs->node_name);
3481     if (data_bs) {
3482         qdict_put_str(options, "data-file", data_bs->node_name);
3483     }
3484     blk = blk_new_open(NULL, NULL, options,
3485                        BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO,
3486                        &local_err);
3487     if (blk == NULL) {
3488         error_propagate(errp, local_err);
3489         ret = -EIO;
3490         goto out;
3491     }
3492 
3493     ret = 0;
3494 out:
3495     blk_unref(blk);
3496     bdrv_unref(bs);
3497     bdrv_unref(data_bs);
3498     return ret;
3499 }
3500 
3501 static int coroutine_fn qcow2_co_create_opts(const char *filename, QemuOpts *opts,
3502                                              Error **errp)
3503 {
3504     BlockdevCreateOptions *create_options = NULL;
3505     QDict *qdict;
3506     Visitor *v;
3507     BlockDriverState *bs = NULL;
3508     BlockDriverState *data_bs = NULL;
3509     Error *local_err = NULL;
3510     const char *val;
3511     int ret;
3512 
3513     /* Only the keyval visitor supports the dotted syntax needed for
3514      * encryption, so go through a QDict before getting a QAPI type. Ignore
3515      * options meant for the protocol layer so that the visitor doesn't
3516      * complain. */
3517     qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts,
3518                                         true);
3519 
3520     /* Handle encryption options */
3521     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
3522     if (val && !strcmp(val, "on")) {
3523         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
3524     } else if (val && !strcmp(val, "off")) {
3525         qdict_del(qdict, BLOCK_OPT_ENCRYPT);
3526     }
3527 
3528     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
3529     if (val && !strcmp(val, "aes")) {
3530         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
3531     }
3532 
3533     /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3534      * version=v2/v3 below. */
3535     val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL);
3536     if (val && !strcmp(val, "0.10")) {
3537         qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2");
3538     } else if (val && !strcmp(val, "1.1")) {
3539         qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3");
3540     }
3541 
3542     /* Change legacy command line options into QMP ones */
3543     static const QDictRenames opt_renames[] = {
3544         { BLOCK_OPT_BACKING_FILE,       "backing-file" },
3545         { BLOCK_OPT_BACKING_FMT,        "backing-fmt" },
3546         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
3547         { BLOCK_OPT_LAZY_REFCOUNTS,     "lazy-refcounts" },
3548         { BLOCK_OPT_REFCOUNT_BITS,      "refcount-bits" },
3549         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
3550         { BLOCK_OPT_COMPAT_LEVEL,       "version" },
3551         { BLOCK_OPT_DATA_FILE_RAW,      "data-file-raw" },
3552         { NULL, NULL },
3553     };
3554 
3555     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
3556         ret = -EINVAL;
3557         goto finish;
3558     }
3559 
3560     /* Create and open the file (protocol layer) */
3561     ret = bdrv_create_file(filename, opts, errp);
3562     if (ret < 0) {
3563         goto finish;
3564     }
3565 
3566     bs = bdrv_open(filename, NULL, NULL,
3567                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
3568     if (bs == NULL) {
3569         ret = -EIO;
3570         goto finish;
3571     }
3572 
3573     /* Create and open an external data file (protocol layer) */
3574     val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE);
3575     if (val) {
3576         ret = bdrv_create_file(val, opts, errp);
3577         if (ret < 0) {
3578             goto finish;
3579         }
3580 
3581         data_bs = bdrv_open(val, NULL, NULL,
3582                             BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
3583                             errp);
3584         if (data_bs == NULL) {
3585             ret = -EIO;
3586             goto finish;
3587         }
3588 
3589         qdict_del(qdict, BLOCK_OPT_DATA_FILE);
3590         qdict_put_str(qdict, "data-file", data_bs->node_name);
3591     }
3592 
3593     /* Set 'driver' and 'node' options */
3594     qdict_put_str(qdict, "driver", "qcow2");
3595     qdict_put_str(qdict, "file", bs->node_name);
3596 
3597     /* Now get the QAPI type BlockdevCreateOptions */
3598     v = qobject_input_visitor_new_flat_confused(qdict, errp);
3599     if (!v) {
3600         ret = -EINVAL;
3601         goto finish;
3602     }
3603 
3604     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
3605     visit_free(v);
3606 
3607     if (local_err) {
3608         error_propagate(errp, local_err);
3609         ret = -EINVAL;
3610         goto finish;
3611     }
3612 
3613     /* Silently round up size */
3614     create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size,
3615                                             BDRV_SECTOR_SIZE);
3616 
3617     /* Create the qcow2 image (format layer) */
3618     ret = qcow2_co_create(create_options, errp);
3619     if (ret < 0) {
3620         goto finish;
3621     }
3622 
3623     ret = 0;
3624 finish:
3625     qobject_unref(qdict);
3626     bdrv_unref(bs);
3627     bdrv_unref(data_bs);
3628     qapi_free_BlockdevCreateOptions(create_options);
3629     return ret;
3630 }
3631 
3632 
3633 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes)
3634 {
3635     int64_t nr;
3636     int res;
3637 
3638     /* Clamp to image length, before checking status of underlying sectors */
3639     if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
3640         bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset;
3641     }
3642 
3643     if (!bytes) {
3644         return true;
3645     }
3646     res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL);
3647     return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == bytes;
3648 }
3649 
3650 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
3651     int64_t offset, int bytes, BdrvRequestFlags flags)
3652 {
3653     int ret;
3654     BDRVQcow2State *s = bs->opaque;
3655 
3656     uint32_t head = offset % s->cluster_size;
3657     uint32_t tail = (offset + bytes) % s->cluster_size;
3658 
3659     trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes);
3660     if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) {
3661         tail = 0;
3662     }
3663 
3664     if (head || tail) {
3665         uint64_t off;
3666         unsigned int nr;
3667 
3668         assert(head + bytes <= s->cluster_size);
3669 
3670         /* check whether remainder of cluster already reads as zero */
3671         if (!(is_zero(bs, offset - head, head) &&
3672               is_zero(bs, offset + bytes,
3673                       tail ? s->cluster_size - tail : 0))) {
3674             return -ENOTSUP;
3675         }
3676 
3677         qemu_co_mutex_lock(&s->lock);
3678         /* We can have new write after previous check */
3679         offset = QEMU_ALIGN_DOWN(offset, s->cluster_size);
3680         bytes = s->cluster_size;
3681         nr = s->cluster_size;
3682         ret = qcow2_get_cluster_offset(bs, offset, &nr, &off);
3683         if (ret != QCOW2_CLUSTER_UNALLOCATED &&
3684             ret != QCOW2_CLUSTER_ZERO_PLAIN &&
3685             ret != QCOW2_CLUSTER_ZERO_ALLOC) {
3686             qemu_co_mutex_unlock(&s->lock);
3687             return -ENOTSUP;
3688         }
3689     } else {
3690         qemu_co_mutex_lock(&s->lock);
3691     }
3692 
3693     trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes);
3694 
3695     /* Whatever is left can use real zero clusters */
3696     ret = qcow2_cluster_zeroize(bs, offset, bytes, flags);
3697     qemu_co_mutex_unlock(&s->lock);
3698 
3699     return ret;
3700 }
3701 
3702 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
3703                                           int64_t offset, int bytes)
3704 {
3705     int ret;
3706     BDRVQcow2State *s = bs->opaque;
3707 
3708     if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
3709         assert(bytes < s->cluster_size);
3710         /* Ignore partial clusters, except for the special case of the
3711          * complete partial cluster at the end of an unaligned file */
3712         if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
3713             offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) {
3714             return -ENOTSUP;
3715         }
3716     }
3717 
3718     qemu_co_mutex_lock(&s->lock);
3719     ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST,
3720                                 false);
3721     qemu_co_mutex_unlock(&s->lock);
3722     return ret;
3723 }
3724 
3725 static int coroutine_fn
3726 qcow2_co_copy_range_from(BlockDriverState *bs,
3727                          BdrvChild *src, uint64_t src_offset,
3728                          BdrvChild *dst, uint64_t dst_offset,
3729                          uint64_t bytes, BdrvRequestFlags read_flags,
3730                          BdrvRequestFlags write_flags)
3731 {
3732     BDRVQcow2State *s = bs->opaque;
3733     int ret;
3734     unsigned int cur_bytes; /* number of bytes in current iteration */
3735     BdrvChild *child = NULL;
3736     BdrvRequestFlags cur_write_flags;
3737 
3738     assert(!bs->encrypted);
3739     qemu_co_mutex_lock(&s->lock);
3740 
3741     while (bytes != 0) {
3742         uint64_t copy_offset = 0;
3743         /* prepare next request */
3744         cur_bytes = MIN(bytes, INT_MAX);
3745         cur_write_flags = write_flags;
3746 
3747         ret = qcow2_get_cluster_offset(bs, src_offset, &cur_bytes, &copy_offset);
3748         if (ret < 0) {
3749             goto out;
3750         }
3751 
3752         switch (ret) {
3753         case QCOW2_CLUSTER_UNALLOCATED:
3754             if (bs->backing && bs->backing->bs) {
3755                 int64_t backing_length = bdrv_getlength(bs->backing->bs);
3756                 if (src_offset >= backing_length) {
3757                     cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3758                 } else {
3759                     child = bs->backing;
3760                     cur_bytes = MIN(cur_bytes, backing_length - src_offset);
3761                     copy_offset = src_offset;
3762                 }
3763             } else {
3764                 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3765             }
3766             break;
3767 
3768         case QCOW2_CLUSTER_ZERO_PLAIN:
3769         case QCOW2_CLUSTER_ZERO_ALLOC:
3770             cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3771             break;
3772 
3773         case QCOW2_CLUSTER_COMPRESSED:
3774             ret = -ENOTSUP;
3775             goto out;
3776 
3777         case QCOW2_CLUSTER_NORMAL:
3778             child = s->data_file;
3779             copy_offset += offset_into_cluster(s, src_offset);
3780             if ((copy_offset & 511) != 0) {
3781                 ret = -EIO;
3782                 goto out;
3783             }
3784             break;
3785 
3786         default:
3787             abort();
3788         }
3789         qemu_co_mutex_unlock(&s->lock);
3790         ret = bdrv_co_copy_range_from(child,
3791                                       copy_offset,
3792                                       dst, dst_offset,
3793                                       cur_bytes, read_flags, cur_write_flags);
3794         qemu_co_mutex_lock(&s->lock);
3795         if (ret < 0) {
3796             goto out;
3797         }
3798 
3799         bytes -= cur_bytes;
3800         src_offset += cur_bytes;
3801         dst_offset += cur_bytes;
3802     }
3803     ret = 0;
3804 
3805 out:
3806     qemu_co_mutex_unlock(&s->lock);
3807     return ret;
3808 }
3809 
3810 static int coroutine_fn
3811 qcow2_co_copy_range_to(BlockDriverState *bs,
3812                        BdrvChild *src, uint64_t src_offset,
3813                        BdrvChild *dst, uint64_t dst_offset,
3814                        uint64_t bytes, BdrvRequestFlags read_flags,
3815                        BdrvRequestFlags write_flags)
3816 {
3817     BDRVQcow2State *s = bs->opaque;
3818     int offset_in_cluster;
3819     int ret;
3820     unsigned int cur_bytes; /* number of sectors in current iteration */
3821     uint64_t cluster_offset;
3822     QCowL2Meta *l2meta = NULL;
3823 
3824     assert(!bs->encrypted);
3825 
3826     qemu_co_mutex_lock(&s->lock);
3827 
3828     while (bytes != 0) {
3829 
3830         l2meta = NULL;
3831 
3832         offset_in_cluster = offset_into_cluster(s, dst_offset);
3833         cur_bytes = MIN(bytes, INT_MAX);
3834 
3835         /* TODO:
3836          * If src->bs == dst->bs, we could simply copy by incrementing
3837          * the refcnt, without copying user data.
3838          * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
3839         ret = qcow2_alloc_cluster_offset(bs, dst_offset, &cur_bytes,
3840                                          &cluster_offset, &l2meta);
3841         if (ret < 0) {
3842             goto fail;
3843         }
3844 
3845         assert((cluster_offset & 511) == 0);
3846 
3847         ret = qcow2_pre_write_overlap_check(bs, 0,
3848                 cluster_offset + offset_in_cluster, cur_bytes, true);
3849         if (ret < 0) {
3850             goto fail;
3851         }
3852 
3853         qemu_co_mutex_unlock(&s->lock);
3854         ret = bdrv_co_copy_range_to(src, src_offset,
3855                                     s->data_file,
3856                                     cluster_offset + offset_in_cluster,
3857                                     cur_bytes, read_flags, write_flags);
3858         qemu_co_mutex_lock(&s->lock);
3859         if (ret < 0) {
3860             goto fail;
3861         }
3862 
3863         ret = qcow2_handle_l2meta(bs, &l2meta, true);
3864         if (ret) {
3865             goto fail;
3866         }
3867 
3868         bytes -= cur_bytes;
3869         src_offset += cur_bytes;
3870         dst_offset += cur_bytes;
3871     }
3872     ret = 0;
3873 
3874 fail:
3875     qcow2_handle_l2meta(bs, &l2meta, false);
3876 
3877     qemu_co_mutex_unlock(&s->lock);
3878 
3879     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
3880 
3881     return ret;
3882 }
3883 
3884 static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,
3885                                           PreallocMode prealloc, Error **errp)
3886 {
3887     BDRVQcow2State *s = bs->opaque;
3888     uint64_t old_length;
3889     int64_t new_l1_size;
3890     int ret;
3891     QDict *options;
3892 
3893     if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA &&
3894         prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL)
3895     {
3896         error_setg(errp, "Unsupported preallocation mode '%s'",
3897                    PreallocMode_str(prealloc));
3898         return -ENOTSUP;
3899     }
3900 
3901     if (offset & 511) {
3902         error_setg(errp, "The new size must be a multiple of 512");
3903         return -EINVAL;
3904     }
3905 
3906     qemu_co_mutex_lock(&s->lock);
3907 
3908     /* cannot proceed if image has snapshots */
3909     if (s->nb_snapshots) {
3910         error_setg(errp, "Can't resize an image which has snapshots");
3911         ret = -ENOTSUP;
3912         goto fail;
3913     }
3914 
3915     /* cannot proceed if image has bitmaps */
3916     if (qcow2_truncate_bitmaps_check(bs, errp)) {
3917         ret = -ENOTSUP;
3918         goto fail;
3919     }
3920 
3921     old_length = bs->total_sectors * BDRV_SECTOR_SIZE;
3922     new_l1_size = size_to_l1(s, offset);
3923 
3924     if (offset < old_length) {
3925         int64_t last_cluster, old_file_size;
3926         if (prealloc != PREALLOC_MODE_OFF) {
3927             error_setg(errp,
3928                        "Preallocation can't be used for shrinking an image");
3929             ret = -EINVAL;
3930             goto fail;
3931         }
3932 
3933         ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size),
3934                                     old_length - ROUND_UP(offset,
3935                                                           s->cluster_size),
3936                                     QCOW2_DISCARD_ALWAYS, true);
3937         if (ret < 0) {
3938             error_setg_errno(errp, -ret, "Failed to discard cropped clusters");
3939             goto fail;
3940         }
3941 
3942         ret = qcow2_shrink_l1_table(bs, new_l1_size);
3943         if (ret < 0) {
3944             error_setg_errno(errp, -ret,
3945                              "Failed to reduce the number of L2 tables");
3946             goto fail;
3947         }
3948 
3949         ret = qcow2_shrink_reftable(bs);
3950         if (ret < 0) {
3951             error_setg_errno(errp, -ret,
3952                              "Failed to discard unused refblocks");
3953             goto fail;
3954         }
3955 
3956         old_file_size = bdrv_getlength(bs->file->bs);
3957         if (old_file_size < 0) {
3958             error_setg_errno(errp, -old_file_size,
3959                              "Failed to inquire current file length");
3960             ret = old_file_size;
3961             goto fail;
3962         }
3963         last_cluster = qcow2_get_last_cluster(bs, old_file_size);
3964         if (last_cluster < 0) {
3965             error_setg_errno(errp, -last_cluster,
3966                              "Failed to find the last cluster");
3967             ret = last_cluster;
3968             goto fail;
3969         }
3970         if ((last_cluster + 1) * s->cluster_size < old_file_size) {
3971             Error *local_err = NULL;
3972 
3973             bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size,
3974                              PREALLOC_MODE_OFF, &local_err);
3975             if (local_err) {
3976                 warn_reportf_err(local_err,
3977                                  "Failed to truncate the tail of the image: ");
3978             }
3979         }
3980     } else {
3981         ret = qcow2_grow_l1_table(bs, new_l1_size, true);
3982         if (ret < 0) {
3983             error_setg_errno(errp, -ret, "Failed to grow the L1 table");
3984             goto fail;
3985         }
3986     }
3987 
3988     switch (prealloc) {
3989     case PREALLOC_MODE_OFF:
3990         if (has_data_file(bs)) {
3991             ret = bdrv_co_truncate(s->data_file, offset, prealloc, errp);
3992             if (ret < 0) {
3993                 goto fail;
3994             }
3995         }
3996         break;
3997 
3998     case PREALLOC_MODE_METADATA:
3999         ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4000         if (ret < 0) {
4001             goto fail;
4002         }
4003         break;
4004 
4005     case PREALLOC_MODE_FALLOC:
4006     case PREALLOC_MODE_FULL:
4007     {
4008         int64_t allocation_start, host_offset, guest_offset;
4009         int64_t clusters_allocated;
4010         int64_t old_file_size, new_file_size;
4011         uint64_t nb_new_data_clusters, nb_new_l2_tables;
4012 
4013         /* With a data file, preallocation means just allocating the metadata
4014          * and forwarding the truncate request to the data file */
4015         if (has_data_file(bs)) {
4016             ret = preallocate_co(bs, old_length, offset, prealloc, errp);
4017             if (ret < 0) {
4018                 goto fail;
4019             }
4020             break;
4021         }
4022 
4023         old_file_size = bdrv_getlength(bs->file->bs);
4024         if (old_file_size < 0) {
4025             error_setg_errno(errp, -old_file_size,
4026                              "Failed to inquire current file length");
4027             ret = old_file_size;
4028             goto fail;
4029         }
4030         old_file_size = ROUND_UP(old_file_size, s->cluster_size);
4031 
4032         nb_new_data_clusters = DIV_ROUND_UP(offset - old_length,
4033                                             s->cluster_size);
4034 
4035         /* This is an overestimation; we will not actually allocate space for
4036          * these in the file but just make sure the new refcount structures are
4037          * able to cover them so we will not have to allocate new refblocks
4038          * while entering the data blocks in the potentially new L2 tables.
4039          * (We do not actually care where the L2 tables are placed. Maybe they
4040          *  are already allocated or they can be placed somewhere before
4041          *  @old_file_size. It does not matter because they will be fully
4042          *  allocated automatically, so they do not need to be covered by the
4043          *  preallocation. All that matters is that we will not have to allocate
4044          *  new refcount structures for them.) */
4045         nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters,
4046                                         s->cluster_size / sizeof(uint64_t));
4047         /* The cluster range may not be aligned to L2 boundaries, so add one L2
4048          * table for a potential head/tail */
4049         nb_new_l2_tables++;
4050 
4051         allocation_start = qcow2_refcount_area(bs, old_file_size,
4052                                                nb_new_data_clusters +
4053                                                nb_new_l2_tables,
4054                                                true, 0, 0);
4055         if (allocation_start < 0) {
4056             error_setg_errno(errp, -allocation_start,
4057                              "Failed to resize refcount structures");
4058             ret = allocation_start;
4059             goto fail;
4060         }
4061 
4062         clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start,
4063                                                      nb_new_data_clusters);
4064         if (clusters_allocated < 0) {
4065             error_setg_errno(errp, -clusters_allocated,
4066                              "Failed to allocate data clusters");
4067             ret = clusters_allocated;
4068             goto fail;
4069         }
4070 
4071         assert(clusters_allocated == nb_new_data_clusters);
4072 
4073         /* Allocate the data area */
4074         new_file_size = allocation_start +
4075                         nb_new_data_clusters * s->cluster_size;
4076         ret = bdrv_co_truncate(bs->file, new_file_size, prealloc, errp);
4077         if (ret < 0) {
4078             error_prepend(errp, "Failed to resize underlying file: ");
4079             qcow2_free_clusters(bs, allocation_start,
4080                                 nb_new_data_clusters * s->cluster_size,
4081                                 QCOW2_DISCARD_OTHER);
4082             goto fail;
4083         }
4084 
4085         /* Create the necessary L2 entries */
4086         host_offset = allocation_start;
4087         guest_offset = old_length;
4088         while (nb_new_data_clusters) {
4089             int64_t nb_clusters = MIN(
4090                 nb_new_data_clusters,
4091                 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset));
4092             QCowL2Meta allocation = {
4093                 .offset       = guest_offset,
4094                 .alloc_offset = host_offset,
4095                 .nb_clusters  = nb_clusters,
4096             };
4097             qemu_co_queue_init(&allocation.dependent_requests);
4098 
4099             ret = qcow2_alloc_cluster_link_l2(bs, &allocation);
4100             if (ret < 0) {
4101                 error_setg_errno(errp, -ret, "Failed to update L2 tables");
4102                 qcow2_free_clusters(bs, host_offset,
4103                                     nb_new_data_clusters * s->cluster_size,
4104                                     QCOW2_DISCARD_OTHER);
4105                 goto fail;
4106             }
4107 
4108             guest_offset += nb_clusters * s->cluster_size;
4109             host_offset += nb_clusters * s->cluster_size;
4110             nb_new_data_clusters -= nb_clusters;
4111         }
4112         break;
4113     }
4114 
4115     default:
4116         g_assert_not_reached();
4117     }
4118 
4119     if (prealloc != PREALLOC_MODE_OFF) {
4120         /* Flush metadata before actually changing the image size */
4121         ret = qcow2_write_caches(bs);
4122         if (ret < 0) {
4123             error_setg_errno(errp, -ret,
4124                              "Failed to flush the preallocated area to disk");
4125             goto fail;
4126         }
4127     }
4128 
4129     bs->total_sectors = offset / BDRV_SECTOR_SIZE;
4130 
4131     /* write updated header.size */
4132     offset = cpu_to_be64(offset);
4133     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
4134                            &offset, sizeof(uint64_t));
4135     if (ret < 0) {
4136         error_setg_errno(errp, -ret, "Failed to update the image size");
4137         goto fail;
4138     }
4139 
4140     s->l1_vm_state_index = new_l1_size;
4141 
4142     /* Update cache sizes */
4143     options = qdict_clone_shallow(bs->options);
4144     ret = qcow2_update_options(bs, options, s->flags, errp);
4145     qobject_unref(options);
4146     if (ret < 0) {
4147         goto fail;
4148     }
4149     ret = 0;
4150 fail:
4151     qemu_co_mutex_unlock(&s->lock);
4152     return ret;
4153 }
4154 
4155 /* XXX: put compressed sectors first, then all the cluster aligned
4156    tables to avoid losing bytes in alignment */
4157 static coroutine_fn int
4158 qcow2_co_pwritev_compressed_part(BlockDriverState *bs,
4159                                  uint64_t offset, uint64_t bytes,
4160                                  QEMUIOVector *qiov, size_t qiov_offset)
4161 {
4162     BDRVQcow2State *s = bs->opaque;
4163     int ret;
4164     ssize_t out_len;
4165     uint8_t *buf, *out_buf;
4166     uint64_t cluster_offset;
4167 
4168     if (has_data_file(bs)) {
4169         return -ENOTSUP;
4170     }
4171 
4172     if (bytes == 0) {
4173         /* align end of file to a sector boundary to ease reading with
4174            sector based I/Os */
4175         int64_t len = bdrv_getlength(bs->file->bs);
4176         if (len < 0) {
4177             return len;
4178         }
4179         return bdrv_co_truncate(bs->file, len, PREALLOC_MODE_OFF, NULL);
4180     }
4181 
4182     if (offset_into_cluster(s, offset)) {
4183         return -EINVAL;
4184     }
4185 
4186     buf = qemu_blockalign(bs, s->cluster_size);
4187     if (bytes != s->cluster_size) {
4188         if (bytes > s->cluster_size ||
4189             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
4190         {
4191             qemu_vfree(buf);
4192             return -EINVAL;
4193         }
4194         /* Zero-pad last write if image size is not cluster aligned */
4195         memset(buf + bytes, 0, s->cluster_size - bytes);
4196     }
4197     qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes);
4198 
4199     out_buf = g_malloc(s->cluster_size);
4200 
4201     out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1,
4202                                 buf, s->cluster_size);
4203     if (out_len == -ENOMEM) {
4204         /* could not compress: write normal cluster */
4205         ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0);
4206         if (ret < 0) {
4207             goto fail;
4208         }
4209         goto success;
4210     } else if (out_len < 0) {
4211         ret = -EINVAL;
4212         goto fail;
4213     }
4214 
4215     qemu_co_mutex_lock(&s->lock);
4216     ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len,
4217                                                 &cluster_offset);
4218     if (ret < 0) {
4219         qemu_co_mutex_unlock(&s->lock);
4220         goto fail;
4221     }
4222 
4223     ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true);
4224     qemu_co_mutex_unlock(&s->lock);
4225     if (ret < 0) {
4226         goto fail;
4227     }
4228 
4229     BLKDBG_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED);
4230     ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0);
4231     if (ret < 0) {
4232         goto fail;
4233     }
4234 success:
4235     ret = 0;
4236 fail:
4237     qemu_vfree(buf);
4238     g_free(out_buf);
4239     return ret;
4240 }
4241 
4242 static int coroutine_fn
4243 qcow2_co_preadv_compressed(BlockDriverState *bs,
4244                            uint64_t file_cluster_offset,
4245                            uint64_t offset,
4246                            uint64_t bytes,
4247                            QEMUIOVector *qiov,
4248                            size_t qiov_offset)
4249 {
4250     BDRVQcow2State *s = bs->opaque;
4251     int ret = 0, csize, nb_csectors;
4252     uint64_t coffset;
4253     uint8_t *buf, *out_buf;
4254     int offset_in_cluster = offset_into_cluster(s, offset);
4255 
4256     coffset = file_cluster_offset & s->cluster_offset_mask;
4257     nb_csectors = ((file_cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
4258     csize = nb_csectors * QCOW2_COMPRESSED_SECTOR_SIZE -
4259         (coffset & ~QCOW2_COMPRESSED_SECTOR_MASK);
4260 
4261     buf = g_try_malloc(csize);
4262     if (!buf) {
4263         return -ENOMEM;
4264     }
4265 
4266     out_buf = qemu_blockalign(bs, s->cluster_size);
4267 
4268     BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
4269     ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0);
4270     if (ret < 0) {
4271         goto fail;
4272     }
4273 
4274     if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) {
4275         ret = -EIO;
4276         goto fail;
4277     }
4278 
4279     qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes);
4280 
4281 fail:
4282     qemu_vfree(out_buf);
4283     g_free(buf);
4284 
4285     return ret;
4286 }
4287 
4288 static int make_completely_empty(BlockDriverState *bs)
4289 {
4290     BDRVQcow2State *s = bs->opaque;
4291     Error *local_err = NULL;
4292     int ret, l1_clusters;
4293     int64_t offset;
4294     uint64_t *new_reftable = NULL;
4295     uint64_t rt_entry, l1_size2;
4296     struct {
4297         uint64_t l1_offset;
4298         uint64_t reftable_offset;
4299         uint32_t reftable_clusters;
4300     } QEMU_PACKED l1_ofs_rt_ofs_cls;
4301 
4302     ret = qcow2_cache_empty(bs, s->l2_table_cache);
4303     if (ret < 0) {
4304         goto fail;
4305     }
4306 
4307     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
4308     if (ret < 0) {
4309         goto fail;
4310     }
4311 
4312     /* Refcounts will be broken utterly */
4313     ret = qcow2_mark_dirty(bs);
4314     if (ret < 0) {
4315         goto fail;
4316     }
4317 
4318     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4319 
4320     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
4321     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
4322 
4323     /* After this call, neither the in-memory nor the on-disk refcount
4324      * information accurately describe the actual references */
4325 
4326     ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
4327                              l1_clusters * s->cluster_size, 0);
4328     if (ret < 0) {
4329         goto fail_broken_refcounts;
4330     }
4331     memset(s->l1_table, 0, l1_size2);
4332 
4333     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
4334 
4335     /* Overwrite enough clusters at the beginning of the sectors to place
4336      * the refcount table, a refcount block and the L1 table in; this may
4337      * overwrite parts of the existing refcount and L1 table, which is not
4338      * an issue because the dirty flag is set, complete data loss is in fact
4339      * desired and partial data loss is consequently fine as well */
4340     ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
4341                              (2 + l1_clusters) * s->cluster_size, 0);
4342     /* This call (even if it failed overall) may have overwritten on-disk
4343      * refcount structures; in that case, the in-memory refcount information
4344      * will probably differ from the on-disk information which makes the BDS
4345      * unusable */
4346     if (ret < 0) {
4347         goto fail_broken_refcounts;
4348     }
4349 
4350     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4351     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
4352 
4353     /* "Create" an empty reftable (one cluster) directly after the image
4354      * header and an empty L1 table three clusters after the image header;
4355      * the cluster between those two will be used as the first refblock */
4356     l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
4357     l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
4358     l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
4359     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
4360                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
4361     if (ret < 0) {
4362         goto fail_broken_refcounts;
4363     }
4364 
4365     s->l1_table_offset = 3 * s->cluster_size;
4366 
4367     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
4368     if (!new_reftable) {
4369         ret = -ENOMEM;
4370         goto fail_broken_refcounts;
4371     }
4372 
4373     s->refcount_table_offset = s->cluster_size;
4374     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
4375     s->max_refcount_table_index = 0;
4376 
4377     g_free(s->refcount_table);
4378     s->refcount_table = new_reftable;
4379     new_reftable = NULL;
4380 
4381     /* Now the in-memory refcount information again corresponds to the on-disk
4382      * information (reftable is empty and no refblocks (the refblock cache is
4383      * empty)); however, this means some clusters (e.g. the image header) are
4384      * referenced, but not refcounted, but the normal qcow2 code assumes that
4385      * the in-memory information is always correct */
4386 
4387     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
4388 
4389     /* Enter the first refblock into the reftable */
4390     rt_entry = cpu_to_be64(2 * s->cluster_size);
4391     ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
4392                            &rt_entry, sizeof(rt_entry));
4393     if (ret < 0) {
4394         goto fail_broken_refcounts;
4395     }
4396     s->refcount_table[0] = 2 * s->cluster_size;
4397 
4398     s->free_cluster_index = 0;
4399     assert(3 + l1_clusters <= s->refcount_block_size);
4400     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
4401     if (offset < 0) {
4402         ret = offset;
4403         goto fail_broken_refcounts;
4404     } else if (offset > 0) {
4405         error_report("First cluster in emptied image is in use");
4406         abort();
4407     }
4408 
4409     /* Now finally the in-memory information corresponds to the on-disk
4410      * structures and is correct */
4411     ret = qcow2_mark_clean(bs);
4412     if (ret < 0) {
4413         goto fail;
4414     }
4415 
4416     ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size,
4417                         PREALLOC_MODE_OFF, &local_err);
4418     if (ret < 0) {
4419         error_report_err(local_err);
4420         goto fail;
4421     }
4422 
4423     return 0;
4424 
4425 fail_broken_refcounts:
4426     /* The BDS is unusable at this point. If we wanted to make it usable, we
4427      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4428      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4429      * again. However, because the functions which could have caused this error
4430      * path to be taken are used by those functions as well, it's very likely
4431      * that that sequence will fail as well. Therefore, just eject the BDS. */
4432     bs->drv = NULL;
4433 
4434 fail:
4435     g_free(new_reftable);
4436     return ret;
4437 }
4438 
4439 static int qcow2_make_empty(BlockDriverState *bs)
4440 {
4441     BDRVQcow2State *s = bs->opaque;
4442     uint64_t offset, end_offset;
4443     int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
4444     int l1_clusters, ret = 0;
4445 
4446     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
4447 
4448     if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps &&
4449         3 + l1_clusters <= s->refcount_block_size &&
4450         s->crypt_method_header != QCOW_CRYPT_LUKS &&
4451         !has_data_file(bs)) {
4452         /* The following function only works for qcow2 v3 images (it
4453          * requires the dirty flag) and only as long as there are no
4454          * features that reserve extra clusters (such as snapshots,
4455          * LUKS header, or persistent bitmaps), because it completely
4456          * empties the image.  Furthermore, the L1 table and three
4457          * additional clusters (image header, refcount table, one
4458          * refcount block) have to fit inside one refcount block. It
4459          * only resets the image file, i.e. does not work with an
4460          * external data file. */
4461         return make_completely_empty(bs);
4462     }
4463 
4464     /* This fallback code simply discards every active cluster; this is slow,
4465      * but works in all cases */
4466     end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
4467     for (offset = 0; offset < end_offset; offset += step) {
4468         /* As this function is generally used after committing an external
4469          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4470          * default action for this kind of discard is to pass the discard,
4471          * which will ideally result in an actually smaller image file, as
4472          * is probably desired. */
4473         ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
4474                                     QCOW2_DISCARD_SNAPSHOT, true);
4475         if (ret < 0) {
4476             break;
4477         }
4478     }
4479 
4480     return ret;
4481 }
4482 
4483 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
4484 {
4485     BDRVQcow2State *s = bs->opaque;
4486     int ret;
4487 
4488     qemu_co_mutex_lock(&s->lock);
4489     ret = qcow2_write_caches(bs);
4490     qemu_co_mutex_unlock(&s->lock);
4491 
4492     return ret;
4493 }
4494 
4495 static ssize_t qcow2_measure_crypto_hdr_init_func(QCryptoBlock *block,
4496         size_t headerlen, void *opaque, Error **errp)
4497 {
4498     size_t *headerlenp = opaque;
4499 
4500     /* Stash away the payload size */
4501     *headerlenp = headerlen;
4502     return 0;
4503 }
4504 
4505 static ssize_t qcow2_measure_crypto_hdr_write_func(QCryptoBlock *block,
4506         size_t offset, const uint8_t *buf, size_t buflen,
4507         void *opaque, Error **errp)
4508 {
4509     /* Discard the bytes, we're not actually writing to an image */
4510     return buflen;
4511 }
4512 
4513 /* Determine the number of bytes for the LUKS payload */
4514 static bool qcow2_measure_luks_headerlen(QemuOpts *opts, size_t *len,
4515                                          Error **errp)
4516 {
4517     QDict *opts_qdict;
4518     QDict *cryptoopts_qdict;
4519     QCryptoBlockCreateOptions *cryptoopts;
4520     QCryptoBlock *crypto;
4521 
4522     /* Extract "encrypt." options into a qdict */
4523     opts_qdict = qemu_opts_to_qdict(opts, NULL);
4524     qdict_extract_subqdict(opts_qdict, &cryptoopts_qdict, "encrypt.");
4525     qobject_unref(opts_qdict);
4526 
4527     /* Build QCryptoBlockCreateOptions object from qdict */
4528     qdict_put_str(cryptoopts_qdict, "format", "luks");
4529     cryptoopts = block_crypto_create_opts_init(cryptoopts_qdict, errp);
4530     qobject_unref(cryptoopts_qdict);
4531     if (!cryptoopts) {
4532         return false;
4533     }
4534 
4535     /* Fake LUKS creation in order to determine the payload size */
4536     crypto = qcrypto_block_create(cryptoopts, "encrypt.",
4537                                   qcow2_measure_crypto_hdr_init_func,
4538                                   qcow2_measure_crypto_hdr_write_func,
4539                                   len, errp);
4540     qapi_free_QCryptoBlockCreateOptions(cryptoopts);
4541     if (!crypto) {
4542         return false;
4543     }
4544 
4545     qcrypto_block_free(crypto);
4546     return true;
4547 }
4548 
4549 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs,
4550                                        Error **errp)
4551 {
4552     Error *local_err = NULL;
4553     BlockMeasureInfo *info;
4554     uint64_t required = 0; /* bytes that contribute to required size */
4555     uint64_t virtual_size; /* disk size as seen by guest */
4556     uint64_t refcount_bits;
4557     uint64_t l2_tables;
4558     uint64_t luks_payload_size = 0;
4559     size_t cluster_size;
4560     int version;
4561     char *optstr;
4562     PreallocMode prealloc;
4563     bool has_backing_file;
4564     bool has_luks;
4565 
4566     /* Parse image creation options */
4567     cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err);
4568     if (local_err) {
4569         goto err;
4570     }
4571 
4572     version = qcow2_opt_get_version_del(opts, &local_err);
4573     if (local_err) {
4574         goto err;
4575     }
4576 
4577     refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
4578     if (local_err) {
4579         goto err;
4580     }
4581 
4582     optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
4583     prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr,
4584                                PREALLOC_MODE_OFF, &local_err);
4585     g_free(optstr);
4586     if (local_err) {
4587         goto err;
4588     }
4589 
4590     optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
4591     has_backing_file = !!optstr;
4592     g_free(optstr);
4593 
4594     optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
4595     has_luks = optstr && strcmp(optstr, "luks") == 0;
4596     g_free(optstr);
4597 
4598     if (has_luks) {
4599         size_t headerlen;
4600 
4601         if (!qcow2_measure_luks_headerlen(opts, &headerlen, &local_err)) {
4602             goto err;
4603         }
4604 
4605         luks_payload_size = ROUND_UP(headerlen, cluster_size);
4606     }
4607 
4608     virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
4609     virtual_size = ROUND_UP(virtual_size, cluster_size);
4610 
4611     /* Check that virtual disk size is valid */
4612     l2_tables = DIV_ROUND_UP(virtual_size / cluster_size,
4613                              cluster_size / sizeof(uint64_t));
4614     if (l2_tables * sizeof(uint64_t) > QCOW_MAX_L1_SIZE) {
4615         error_setg(&local_err, "The image size is too large "
4616                                "(try using a larger cluster size)");
4617         goto err;
4618     }
4619 
4620     /* Account for input image */
4621     if (in_bs) {
4622         int64_t ssize = bdrv_getlength(in_bs);
4623         if (ssize < 0) {
4624             error_setg_errno(&local_err, -ssize,
4625                              "Unable to get image virtual_size");
4626             goto err;
4627         }
4628 
4629         virtual_size = ROUND_UP(ssize, cluster_size);
4630 
4631         if (has_backing_file) {
4632             /* We don't how much of the backing chain is shared by the input
4633              * image and the new image file.  In the worst case the new image's
4634              * backing file has nothing in common with the input image.  Be
4635              * conservative and assume all clusters need to be written.
4636              */
4637             required = virtual_size;
4638         } else {
4639             int64_t offset;
4640             int64_t pnum = 0;
4641 
4642             for (offset = 0; offset < ssize; offset += pnum) {
4643                 int ret;
4644 
4645                 ret = bdrv_block_status_above(in_bs, NULL, offset,
4646                                               ssize - offset, &pnum, NULL,
4647                                               NULL);
4648                 if (ret < 0) {
4649                     error_setg_errno(&local_err, -ret,
4650                                      "Unable to get block status");
4651                     goto err;
4652                 }
4653 
4654                 if (ret & BDRV_BLOCK_ZERO) {
4655                     /* Skip zero regions (safe with no backing file) */
4656                 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
4657                            (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
4658                     /* Extend pnum to end of cluster for next iteration */
4659                     pnum = ROUND_UP(offset + pnum, cluster_size) - offset;
4660 
4661                     /* Count clusters we've seen */
4662                     required += offset % cluster_size + pnum;
4663                 }
4664             }
4665         }
4666     }
4667 
4668     /* Take into account preallocation.  Nothing special is needed for
4669      * PREALLOC_MODE_METADATA since metadata is always counted.
4670      */
4671     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
4672         required = virtual_size;
4673     }
4674 
4675     info = g_new(BlockMeasureInfo, 1);
4676     info->fully_allocated =
4677         qcow2_calc_prealloc_size(virtual_size, cluster_size,
4678                                  ctz32(refcount_bits)) + luks_payload_size;
4679 
4680     /* Remove data clusters that are not required.  This overestimates the
4681      * required size because metadata needed for the fully allocated file is
4682      * still counted.
4683      */
4684     info->required = info->fully_allocated - virtual_size + required;
4685     return info;
4686 
4687 err:
4688     error_propagate(errp, local_err);
4689     return NULL;
4690 }
4691 
4692 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4693 {
4694     BDRVQcow2State *s = bs->opaque;
4695     bdi->unallocated_blocks_are_zero = true;
4696     bdi->cluster_size = s->cluster_size;
4697     bdi->vm_state_offset = qcow2_vm_state_offset(s);
4698     return 0;
4699 }
4700 
4701 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs,
4702                                                   Error **errp)
4703 {
4704     BDRVQcow2State *s = bs->opaque;
4705     ImageInfoSpecific *spec_info;
4706     QCryptoBlockInfo *encrypt_info = NULL;
4707     Error *local_err = NULL;
4708 
4709     if (s->crypto != NULL) {
4710         encrypt_info = qcrypto_block_get_info(s->crypto, &local_err);
4711         if (local_err) {
4712             error_propagate(errp, local_err);
4713             return NULL;
4714         }
4715     }
4716 
4717     spec_info = g_new(ImageInfoSpecific, 1);
4718     *spec_info = (ImageInfoSpecific){
4719         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
4720         .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1),
4721     };
4722     if (s->qcow_version == 2) {
4723         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
4724             .compat             = g_strdup("0.10"),
4725             .refcount_bits      = s->refcount_bits,
4726         };
4727     } else if (s->qcow_version == 3) {
4728         Qcow2BitmapInfoList *bitmaps;
4729         bitmaps = qcow2_get_bitmap_info_list(bs, &local_err);
4730         if (local_err) {
4731             error_propagate(errp, local_err);
4732             qapi_free_ImageInfoSpecific(spec_info);
4733             return NULL;
4734         }
4735         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
4736             .compat             = g_strdup("1.1"),
4737             .lazy_refcounts     = s->compatible_features &
4738                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
4739             .has_lazy_refcounts = true,
4740             .corrupt            = s->incompatible_features &
4741                                   QCOW2_INCOMPAT_CORRUPT,
4742             .has_corrupt        = true,
4743             .refcount_bits      = s->refcount_bits,
4744             .has_bitmaps        = !!bitmaps,
4745             .bitmaps            = bitmaps,
4746             .has_data_file      = !!s->image_data_file,
4747             .data_file          = g_strdup(s->image_data_file),
4748             .has_data_file_raw  = has_data_file(bs),
4749             .data_file_raw      = data_file_is_raw(bs),
4750         };
4751     } else {
4752         /* if this assertion fails, this probably means a new version was
4753          * added without having it covered here */
4754         assert(false);
4755     }
4756 
4757     if (encrypt_info) {
4758         ImageInfoSpecificQCow2Encryption *qencrypt =
4759             g_new(ImageInfoSpecificQCow2Encryption, 1);
4760         switch (encrypt_info->format) {
4761         case Q_CRYPTO_BLOCK_FORMAT_QCOW:
4762             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES;
4763             break;
4764         case Q_CRYPTO_BLOCK_FORMAT_LUKS:
4765             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS;
4766             qencrypt->u.luks = encrypt_info->u.luks;
4767             break;
4768         default:
4769             abort();
4770         }
4771         /* Since we did shallow copy above, erase any pointers
4772          * in the original info */
4773         memset(&encrypt_info->u, 0, sizeof(encrypt_info->u));
4774         qapi_free_QCryptoBlockInfo(encrypt_info);
4775 
4776         spec_info->u.qcow2.data->has_encrypt = true;
4777         spec_info->u.qcow2.data->encrypt = qencrypt;
4778     }
4779 
4780     return spec_info;
4781 }
4782 
4783 static int qcow2_has_zero_init(BlockDriverState *bs)
4784 {
4785     BDRVQcow2State *s = bs->opaque;
4786     bool preallocated;
4787 
4788     if (qemu_in_coroutine()) {
4789         qemu_co_mutex_lock(&s->lock);
4790     }
4791     /*
4792      * Check preallocation status: Preallocated images have all L2
4793      * tables allocated, nonpreallocated images have none.  It is
4794      * therefore enough to check the first one.
4795      */
4796     preallocated = s->l1_size > 0 && s->l1_table[0] != 0;
4797     if (qemu_in_coroutine()) {
4798         qemu_co_mutex_unlock(&s->lock);
4799     }
4800 
4801     if (!preallocated) {
4802         return 1;
4803     } else if (bs->encrypted) {
4804         return 0;
4805     } else {
4806         return bdrv_has_zero_init(s->data_file->bs);
4807     }
4808 }
4809 
4810 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
4811                               int64_t pos)
4812 {
4813     BDRVQcow2State *s = bs->opaque;
4814 
4815     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
4816     return bs->drv->bdrv_co_pwritev_part(bs, qcow2_vm_state_offset(s) + pos,
4817                                          qiov->size, qiov, 0, 0);
4818 }
4819 
4820 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
4821                               int64_t pos)
4822 {
4823     BDRVQcow2State *s = bs->opaque;
4824 
4825     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
4826     return bs->drv->bdrv_co_preadv_part(bs, qcow2_vm_state_offset(s) + pos,
4827                                         qiov->size, qiov, 0, 0);
4828 }
4829 
4830 /*
4831  * Downgrades an image's version. To achieve this, any incompatible features
4832  * have to be removed.
4833  */
4834 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
4835                            BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
4836                            Error **errp)
4837 {
4838     BDRVQcow2State *s = bs->opaque;
4839     int current_version = s->qcow_version;
4840     int ret;
4841 
4842     /* This is qcow2_downgrade(), not qcow2_upgrade() */
4843     assert(target_version < current_version);
4844 
4845     /* There are no other versions (now) that you can downgrade to */
4846     assert(target_version == 2);
4847 
4848     if (s->refcount_order != 4) {
4849         error_setg(errp, "compat=0.10 requires refcount_bits=16");
4850         return -ENOTSUP;
4851     }
4852 
4853     if (has_data_file(bs)) {
4854         error_setg(errp, "Cannot downgrade an image with a data file");
4855         return -ENOTSUP;
4856     }
4857 
4858     /* clear incompatible features */
4859     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
4860         ret = qcow2_mark_clean(bs);
4861         if (ret < 0) {
4862             error_setg_errno(errp, -ret, "Failed to make the image clean");
4863             return ret;
4864         }
4865     }
4866 
4867     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
4868      * the first place; if that happens nonetheless, returning -ENOTSUP is the
4869      * best thing to do anyway */
4870 
4871     if (s->incompatible_features) {
4872         error_setg(errp, "Cannot downgrade an image with incompatible features "
4873                    "%#" PRIx64 " set", s->incompatible_features);
4874         return -ENOTSUP;
4875     }
4876 
4877     /* since we can ignore compatible features, we can set them to 0 as well */
4878     s->compatible_features = 0;
4879     /* if lazy refcounts have been used, they have already been fixed through
4880      * clearing the dirty flag */
4881 
4882     /* clearing autoclear features is trivial */
4883     s->autoclear_features = 0;
4884 
4885     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
4886     if (ret < 0) {
4887         error_setg_errno(errp, -ret, "Failed to turn zero into data clusters");
4888         return ret;
4889     }
4890 
4891     s->qcow_version = target_version;
4892     ret = qcow2_update_header(bs);
4893     if (ret < 0) {
4894         s->qcow_version = current_version;
4895         error_setg_errno(errp, -ret, "Failed to update the image header");
4896         return ret;
4897     }
4898     return 0;
4899 }
4900 
4901 typedef enum Qcow2AmendOperation {
4902     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
4903      * statically initialized to so that the helper CB can discern the first
4904      * invocation from an operation change */
4905     QCOW2_NO_OPERATION = 0,
4906 
4907     QCOW2_CHANGING_REFCOUNT_ORDER,
4908     QCOW2_DOWNGRADING,
4909 } Qcow2AmendOperation;
4910 
4911 typedef struct Qcow2AmendHelperCBInfo {
4912     /* The code coordinating the amend operations should only modify
4913      * these four fields; the rest will be managed by the CB */
4914     BlockDriverAmendStatusCB *original_status_cb;
4915     void *original_cb_opaque;
4916 
4917     Qcow2AmendOperation current_operation;
4918 
4919     /* Total number of operations to perform (only set once) */
4920     int total_operations;
4921 
4922     /* The following fields are managed by the CB */
4923 
4924     /* Number of operations completed */
4925     int operations_completed;
4926 
4927     /* Cumulative offset of all completed operations */
4928     int64_t offset_completed;
4929 
4930     Qcow2AmendOperation last_operation;
4931     int64_t last_work_size;
4932 } Qcow2AmendHelperCBInfo;
4933 
4934 static void qcow2_amend_helper_cb(BlockDriverState *bs,
4935                                   int64_t operation_offset,
4936                                   int64_t operation_work_size, void *opaque)
4937 {
4938     Qcow2AmendHelperCBInfo *info = opaque;
4939     int64_t current_work_size;
4940     int64_t projected_work_size;
4941 
4942     if (info->current_operation != info->last_operation) {
4943         if (info->last_operation != QCOW2_NO_OPERATION) {
4944             info->offset_completed += info->last_work_size;
4945             info->operations_completed++;
4946         }
4947 
4948         info->last_operation = info->current_operation;
4949     }
4950 
4951     assert(info->total_operations > 0);
4952     assert(info->operations_completed < info->total_operations);
4953 
4954     info->last_work_size = operation_work_size;
4955 
4956     current_work_size = info->offset_completed + operation_work_size;
4957 
4958     /* current_work_size is the total work size for (operations_completed + 1)
4959      * operations (which includes this one), so multiply it by the number of
4960      * operations not covered and divide it by the number of operations
4961      * covered to get a projection for the operations not covered */
4962     projected_work_size = current_work_size * (info->total_operations -
4963                                                info->operations_completed - 1)
4964                                             / (info->operations_completed + 1);
4965 
4966     info->original_status_cb(bs, info->offset_completed + operation_offset,
4967                              current_work_size + projected_work_size,
4968                              info->original_cb_opaque);
4969 }
4970 
4971 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
4972                                BlockDriverAmendStatusCB *status_cb,
4973                                void *cb_opaque,
4974                                Error **errp)
4975 {
4976     BDRVQcow2State *s = bs->opaque;
4977     int old_version = s->qcow_version, new_version = old_version;
4978     uint64_t new_size = 0;
4979     const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL;
4980     bool lazy_refcounts = s->use_lazy_refcounts;
4981     bool data_file_raw = data_file_is_raw(bs);
4982     const char *compat = NULL;
4983     uint64_t cluster_size = s->cluster_size;
4984     bool encrypt;
4985     int encformat;
4986     int refcount_bits = s->refcount_bits;
4987     int ret;
4988     QemuOptDesc *desc = opts->list->desc;
4989     Qcow2AmendHelperCBInfo helper_cb_info;
4990 
4991     while (desc && desc->name) {
4992         if (!qemu_opt_find(opts, desc->name)) {
4993             /* only change explicitly defined options */
4994             desc++;
4995             continue;
4996         }
4997 
4998         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
4999             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
5000             if (!compat) {
5001                 /* preserve default */
5002             } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) {
5003                 new_version = 2;
5004             } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) {
5005                 new_version = 3;
5006             } else {
5007                 error_setg(errp, "Unknown compatibility level %s", compat);
5008                 return -EINVAL;
5009             }
5010         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
5011             error_setg(errp, "Cannot change preallocation mode");
5012             return -ENOTSUP;
5013         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
5014             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
5015         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
5016             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5017         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
5018             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5019         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
5020             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
5021                                         !!s->crypto);
5022 
5023             if (encrypt != !!s->crypto) {
5024                 error_setg(errp,
5025                            "Changing the encryption flag is not supported");
5026                 return -ENOTSUP;
5027             }
5028         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT_FORMAT)) {
5029             encformat = qcow2_crypt_method_from_format(
5030                 qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT));
5031 
5032             if (encformat != s->crypt_method_header) {
5033                 error_setg(errp,
5034                            "Changing the encryption format is not supported");
5035                 return -ENOTSUP;
5036             }
5037         } else if (g_str_has_prefix(desc->name, "encrypt.")) {
5038             error_setg(errp,
5039                        "Changing the encryption parameters is not supported");
5040             return -ENOTSUP;
5041         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
5042             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
5043                                              cluster_size);
5044             if (cluster_size != s->cluster_size) {
5045                 error_setg(errp, "Changing the cluster size is not supported");
5046                 return -ENOTSUP;
5047             }
5048         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
5049             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
5050                                                lazy_refcounts);
5051         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
5052             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
5053                                                 refcount_bits);
5054 
5055             if (refcount_bits <= 0 || refcount_bits > 64 ||
5056                 !is_power_of_2(refcount_bits))
5057             {
5058                 error_setg(errp, "Refcount width must be a power of two and "
5059                            "may not exceed 64 bits");
5060                 return -EINVAL;
5061             }
5062         } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) {
5063             data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE);
5064             if (data_file && !has_data_file(bs)) {
5065                 error_setg(errp, "data-file can only be set for images that "
5066                                  "use an external data file");
5067                 return -EINVAL;
5068             }
5069         } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) {
5070             data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW,
5071                                               data_file_raw);
5072             if (data_file_raw && !data_file_is_raw(bs)) {
5073                 error_setg(errp, "data-file-raw cannot be set on existing "
5074                                  "images");
5075                 return -EINVAL;
5076             }
5077         } else {
5078             /* if this point is reached, this probably means a new option was
5079              * added without having it covered here */
5080             abort();
5081         }
5082 
5083         desc++;
5084     }
5085 
5086     helper_cb_info = (Qcow2AmendHelperCBInfo){
5087         .original_status_cb = status_cb,
5088         .original_cb_opaque = cb_opaque,
5089         .total_operations = (new_version < old_version)
5090                           + (s->refcount_bits != refcount_bits)
5091     };
5092 
5093     /* Upgrade first (some features may require compat=1.1) */
5094     if (new_version > old_version) {
5095         s->qcow_version = new_version;
5096         ret = qcow2_update_header(bs);
5097         if (ret < 0) {
5098             s->qcow_version = old_version;
5099             error_setg_errno(errp, -ret, "Failed to update the image header");
5100             return ret;
5101         }
5102     }
5103 
5104     if (s->refcount_bits != refcount_bits) {
5105         int refcount_order = ctz32(refcount_bits);
5106 
5107         if (new_version < 3 && refcount_bits != 16) {
5108             error_setg(errp, "Refcount widths other than 16 bits require "
5109                        "compatibility level 1.1 or above (use compat=1.1 or "
5110                        "greater)");
5111             return -EINVAL;
5112         }
5113 
5114         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
5115         ret = qcow2_change_refcount_order(bs, refcount_order,
5116                                           &qcow2_amend_helper_cb,
5117                                           &helper_cb_info, errp);
5118         if (ret < 0) {
5119             return ret;
5120         }
5121     }
5122 
5123     /* data-file-raw blocks backing files, so clear it first if requested */
5124     if (data_file_raw) {
5125         s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5126     } else {
5127         s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5128     }
5129 
5130     if (data_file) {
5131         g_free(s->image_data_file);
5132         s->image_data_file = *data_file ? g_strdup(data_file) : NULL;
5133     }
5134 
5135     ret = qcow2_update_header(bs);
5136     if (ret < 0) {
5137         error_setg_errno(errp, -ret, "Failed to update the image header");
5138         return ret;
5139     }
5140 
5141     if (backing_file || backing_format) {
5142         ret = qcow2_change_backing_file(bs,
5143                     backing_file ?: s->image_backing_file,
5144                     backing_format ?: s->image_backing_format);
5145         if (ret < 0) {
5146             error_setg_errno(errp, -ret, "Failed to change the backing file");
5147             return ret;
5148         }
5149     }
5150 
5151     if (s->use_lazy_refcounts != lazy_refcounts) {
5152         if (lazy_refcounts) {
5153             if (new_version < 3) {
5154                 error_setg(errp, "Lazy refcounts only supported with "
5155                            "compatibility level 1.1 and above (use compat=1.1 "
5156                            "or greater)");
5157                 return -EINVAL;
5158             }
5159             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5160             ret = qcow2_update_header(bs);
5161             if (ret < 0) {
5162                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5163                 error_setg_errno(errp, -ret, "Failed to update the image header");
5164                 return ret;
5165             }
5166             s->use_lazy_refcounts = true;
5167         } else {
5168             /* make image clean first */
5169             ret = qcow2_mark_clean(bs);
5170             if (ret < 0) {
5171                 error_setg_errno(errp, -ret, "Failed to make the image clean");
5172                 return ret;
5173             }
5174             /* now disallow lazy refcounts */
5175             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5176             ret = qcow2_update_header(bs);
5177             if (ret < 0) {
5178                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5179                 error_setg_errno(errp, -ret, "Failed to update the image header");
5180                 return ret;
5181             }
5182             s->use_lazy_refcounts = false;
5183         }
5184     }
5185 
5186     if (new_size) {
5187         BlockBackend *blk = blk_new(bdrv_get_aio_context(bs),
5188                                     BLK_PERM_RESIZE, BLK_PERM_ALL);
5189         ret = blk_insert_bs(blk, bs, errp);
5190         if (ret < 0) {
5191             blk_unref(blk);
5192             return ret;
5193         }
5194 
5195         ret = blk_truncate(blk, new_size, PREALLOC_MODE_OFF, errp);
5196         blk_unref(blk);
5197         if (ret < 0) {
5198             return ret;
5199         }
5200     }
5201 
5202     /* Downgrade last (so unsupported features can be removed before) */
5203     if (new_version < old_version) {
5204         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
5205         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
5206                               &helper_cb_info, errp);
5207         if (ret < 0) {
5208             return ret;
5209         }
5210     }
5211 
5212     return 0;
5213 }
5214 
5215 /*
5216  * If offset or size are negative, respectively, they will not be included in
5217  * the BLOCK_IMAGE_CORRUPTED event emitted.
5218  * fatal will be ignored for read-only BDS; corruptions found there will always
5219  * be considered non-fatal.
5220  */
5221 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
5222                              int64_t size, const char *message_format, ...)
5223 {
5224     BDRVQcow2State *s = bs->opaque;
5225     const char *node_name;
5226     char *message;
5227     va_list ap;
5228 
5229     fatal = fatal && bdrv_is_writable(bs);
5230 
5231     if (s->signaled_corruption &&
5232         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
5233     {
5234         return;
5235     }
5236 
5237     va_start(ap, message_format);
5238     message = g_strdup_vprintf(message_format, ap);
5239     va_end(ap);
5240 
5241     if (fatal) {
5242         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
5243                 "corruption events will be suppressed\n", message);
5244     } else {
5245         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
5246                 "corruption events will be suppressed\n", message);
5247     }
5248 
5249     node_name = bdrv_get_node_name(bs);
5250     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
5251                                           *node_name != '\0', node_name,
5252                                           message, offset >= 0, offset,
5253                                           size >= 0, size,
5254                                           fatal);
5255     g_free(message);
5256 
5257     if (fatal) {
5258         qcow2_mark_corrupt(bs);
5259         bs->drv = NULL; /* make BDS unusable */
5260     }
5261 
5262     s->signaled_corruption = true;
5263 }
5264 
5265 static QemuOptsList qcow2_create_opts = {
5266     .name = "qcow2-create-opts",
5267     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
5268     .desc = {
5269         {
5270             .name = BLOCK_OPT_SIZE,
5271             .type = QEMU_OPT_SIZE,
5272             .help = "Virtual disk size"
5273         },
5274         {
5275             .name = BLOCK_OPT_COMPAT_LEVEL,
5276             .type = QEMU_OPT_STRING,
5277             .help = "Compatibility level (v2 [0.10] or v3 [1.1])"
5278         },
5279         {
5280             .name = BLOCK_OPT_BACKING_FILE,
5281             .type = QEMU_OPT_STRING,
5282             .help = "File name of a base image"
5283         },
5284         {
5285             .name = BLOCK_OPT_BACKING_FMT,
5286             .type = QEMU_OPT_STRING,
5287             .help = "Image format of the base image"
5288         },
5289         {
5290             .name = BLOCK_OPT_DATA_FILE,
5291             .type = QEMU_OPT_STRING,
5292             .help = "File name of an external data file"
5293         },
5294         {
5295             .name = BLOCK_OPT_DATA_FILE_RAW,
5296             .type = QEMU_OPT_BOOL,
5297             .help = "The external data file must stay valid as a raw image"
5298         },
5299         {
5300             .name = BLOCK_OPT_ENCRYPT,
5301             .type = QEMU_OPT_BOOL,
5302             .help = "Encrypt the image with format 'aes'. (Deprecated "
5303                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
5304         },
5305         {
5306             .name = BLOCK_OPT_ENCRYPT_FORMAT,
5307             .type = QEMU_OPT_STRING,
5308             .help = "Encrypt the image, format choices: 'aes', 'luks'",
5309         },
5310         BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
5311             "ID of secret providing qcow AES key or LUKS passphrase"),
5312         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
5313         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
5314         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
5315         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
5316         BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
5317         BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
5318         {
5319             .name = BLOCK_OPT_CLUSTER_SIZE,
5320             .type = QEMU_OPT_SIZE,
5321             .help = "qcow2 cluster size",
5322             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
5323         },
5324         {
5325             .name = BLOCK_OPT_PREALLOC,
5326             .type = QEMU_OPT_STRING,
5327             .help = "Preallocation mode (allowed values: off, metadata, "
5328                     "falloc, full)"
5329         },
5330         {
5331             .name = BLOCK_OPT_LAZY_REFCOUNTS,
5332             .type = QEMU_OPT_BOOL,
5333             .help = "Postpone refcount updates",
5334             .def_value_str = "off"
5335         },
5336         {
5337             .name = BLOCK_OPT_REFCOUNT_BITS,
5338             .type = QEMU_OPT_NUMBER,
5339             .help = "Width of a reference count entry in bits",
5340             .def_value_str = "16"
5341         },
5342         { /* end of list */ }
5343     }
5344 };
5345 
5346 static const char *const qcow2_strong_runtime_opts[] = {
5347     "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
5348 
5349     NULL
5350 };
5351 
5352 BlockDriver bdrv_qcow2 = {
5353     .format_name        = "qcow2",
5354     .instance_size      = sizeof(BDRVQcow2State),
5355     .bdrv_probe         = qcow2_probe,
5356     .bdrv_open          = qcow2_open,
5357     .bdrv_close         = qcow2_close,
5358     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
5359     .bdrv_reopen_commit   = qcow2_reopen_commit,
5360     .bdrv_reopen_abort    = qcow2_reopen_abort,
5361     .bdrv_join_options    = qcow2_join_options,
5362     .bdrv_child_perm      = bdrv_format_default_perms,
5363     .bdrv_co_create_opts  = qcow2_co_create_opts,
5364     .bdrv_co_create       = qcow2_co_create,
5365     .bdrv_has_zero_init   = qcow2_has_zero_init,
5366     .bdrv_has_zero_init_truncate = bdrv_has_zero_init_1,
5367     .bdrv_co_block_status = qcow2_co_block_status,
5368 
5369     .bdrv_co_preadv_part    = qcow2_co_preadv_part,
5370     .bdrv_co_pwritev_part   = qcow2_co_pwritev_part,
5371     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
5372 
5373     .bdrv_co_pwrite_zeroes  = qcow2_co_pwrite_zeroes,
5374     .bdrv_co_pdiscard       = qcow2_co_pdiscard,
5375     .bdrv_co_copy_range_from = qcow2_co_copy_range_from,
5376     .bdrv_co_copy_range_to  = qcow2_co_copy_range_to,
5377     .bdrv_co_truncate       = qcow2_co_truncate,
5378     .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part,
5379     .bdrv_make_empty        = qcow2_make_empty,
5380 
5381     .bdrv_snapshot_create   = qcow2_snapshot_create,
5382     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
5383     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
5384     .bdrv_snapshot_list     = qcow2_snapshot_list,
5385     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
5386     .bdrv_measure           = qcow2_measure,
5387     .bdrv_get_info          = qcow2_get_info,
5388     .bdrv_get_specific_info = qcow2_get_specific_info,
5389 
5390     .bdrv_save_vmstate    = qcow2_save_vmstate,
5391     .bdrv_load_vmstate    = qcow2_load_vmstate,
5392 
5393     .supports_backing           = true,
5394     .bdrv_change_backing_file   = qcow2_change_backing_file,
5395 
5396     .bdrv_refresh_limits        = qcow2_refresh_limits,
5397     .bdrv_co_invalidate_cache   = qcow2_co_invalidate_cache,
5398     .bdrv_inactivate            = qcow2_inactivate,
5399 
5400     .create_opts         = &qcow2_create_opts,
5401     .strong_runtime_opts = qcow2_strong_runtime_opts,
5402     .mutable_opts        = mutable_opts,
5403     .bdrv_co_check       = qcow2_co_check,
5404     .bdrv_amend_options  = qcow2_amend_options,
5405 
5406     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
5407     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
5408 
5409     .bdrv_reopen_bitmaps_rw = qcow2_reopen_bitmaps_rw,
5410     .bdrv_can_store_new_dirty_bitmap = qcow2_can_store_new_dirty_bitmap,
5411     .bdrv_remove_persistent_dirty_bitmap = qcow2_remove_persistent_dirty_bitmap,
5412 };
5413 
5414 static void bdrv_qcow2_init(void)
5415 {
5416     bdrv_register(&bdrv_qcow2);
5417 }
5418 
5419 block_init(bdrv_qcow2_init);
5420