xref: /openbmc/qemu/block/qcow2.c (revision 53d28455)
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 #define ZLIB_CONST
28 #include <zlib.h>
29 
30 #include "block/block_int.h"
31 #include "block/qdict.h"
32 #include "sysemu/block-backend.h"
33 #include "qemu/module.h"
34 #include "qcow2.h"
35 #include "qemu/error-report.h"
36 #include "qapi/error.h"
37 #include "qapi/qapi-events-block-core.h"
38 #include "qapi/qmp/qdict.h"
39 #include "qapi/qmp/qstring.h"
40 #include "trace.h"
41 #include "qemu/option_int.h"
42 #include "qemu/cutils.h"
43 #include "qemu/bswap.h"
44 #include "qapi/qobject-input-visitor.h"
45 #include "qapi/qapi-visit-block-core.h"
46 #include "crypto.h"
47 #include "block/thread-pool.h"
48 
49 /*
50   Differences with QCOW:
51 
52   - Support for multiple incremental snapshots.
53   - Memory management by reference counts.
54   - Clusters which have a reference count of one have the bit
55     QCOW_OFLAG_COPIED to optimize write performance.
56   - Size of compressed clusters is stored in sectors to reduce bit usage
57     in the cluster offsets.
58   - Support for storing additional data (such as the VM state) in the
59     snapshots.
60   - If a backing store is used, the cluster size is not constrained
61     (could be backported to QCOW).
62   - L2 tables have always a size of one cluster.
63 */
64 
65 
66 typedef struct {
67     uint32_t magic;
68     uint32_t len;
69 } QEMU_PACKED QCowExtension;
70 
71 #define  QCOW2_EXT_MAGIC_END 0
72 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
73 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
74 #define  QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
75 #define  QCOW2_EXT_MAGIC_BITMAPS 0x23852875
76 
77 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
78 {
79     const QCowHeader *cow_header = (const void *)buf;
80 
81     if (buf_size >= sizeof(QCowHeader) &&
82         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
83         be32_to_cpu(cow_header->version) >= 2)
84         return 100;
85     else
86         return 0;
87 }
88 
89 
90 static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset,
91                                           uint8_t *buf, size_t buflen,
92                                           void *opaque, Error **errp)
93 {
94     BlockDriverState *bs = opaque;
95     BDRVQcow2State *s = bs->opaque;
96     ssize_t ret;
97 
98     if ((offset + buflen) > s->crypto_header.length) {
99         error_setg(errp, "Request for data outside of extension header");
100         return -1;
101     }
102 
103     ret = bdrv_pread(bs->file,
104                      s->crypto_header.offset + offset, buf, buflen);
105     if (ret < 0) {
106         error_setg_errno(errp, -ret, "Could not read encryption header");
107         return -1;
108     }
109     return ret;
110 }
111 
112 
113 static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen,
114                                           void *opaque, Error **errp)
115 {
116     BlockDriverState *bs = opaque;
117     BDRVQcow2State *s = bs->opaque;
118     int64_t ret;
119     int64_t clusterlen;
120 
121     ret = qcow2_alloc_clusters(bs, headerlen);
122     if (ret < 0) {
123         error_setg_errno(errp, -ret,
124                          "Cannot allocate cluster for LUKS header size %zu",
125                          headerlen);
126         return -1;
127     }
128 
129     s->crypto_header.length = headerlen;
130     s->crypto_header.offset = ret;
131 
132     /* Zero fill remaining space in cluster so it has predictable
133      * content in case of future spec changes */
134     clusterlen = size_to_clusters(s, headerlen) * s->cluster_size;
135     assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen) == 0);
136     ret = bdrv_pwrite_zeroes(bs->file,
137                              ret + headerlen,
138                              clusterlen - headerlen, 0);
139     if (ret < 0) {
140         error_setg_errno(errp, -ret, "Could not zero fill encryption header");
141         return -1;
142     }
143 
144     return ret;
145 }
146 
147 
148 static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset,
149                                            const uint8_t *buf, size_t buflen,
150                                            void *opaque, Error **errp)
151 {
152     BlockDriverState *bs = opaque;
153     BDRVQcow2State *s = bs->opaque;
154     ssize_t ret;
155 
156     if ((offset + buflen) > s->crypto_header.length) {
157         error_setg(errp, "Request for data outside of extension header");
158         return -1;
159     }
160 
161     ret = bdrv_pwrite(bs->file,
162                       s->crypto_header.offset + offset, buf, buflen);
163     if (ret < 0) {
164         error_setg_errno(errp, -ret, "Could not read encryption header");
165         return -1;
166     }
167     return ret;
168 }
169 
170 
171 /*
172  * read qcow2 extension and fill bs
173  * start reading from start_offset
174  * finish reading upon magic of value 0 or when end_offset reached
175  * unknown magic is skipped (future extension this version knows nothing about)
176  * return 0 upon success, non-0 otherwise
177  */
178 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
179                                  uint64_t end_offset, void **p_feature_table,
180                                  int flags, bool *need_update_header,
181                                  Error **errp)
182 {
183     BDRVQcow2State *s = bs->opaque;
184     QCowExtension ext;
185     uint64_t offset;
186     int ret;
187     Qcow2BitmapHeaderExt bitmaps_ext;
188 
189     if (need_update_header != NULL) {
190         *need_update_header = false;
191     }
192 
193 #ifdef DEBUG_EXT
194     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
195 #endif
196     offset = start_offset;
197     while (offset < end_offset) {
198 
199 #ifdef DEBUG_EXT
200         /* Sanity check */
201         if (offset > s->cluster_size)
202             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
203 
204         printf("attempting to read extended header in offset %lu\n", offset);
205 #endif
206 
207         ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
208         if (ret < 0) {
209             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
210                              "pread fail from offset %" PRIu64, offset);
211             return 1;
212         }
213         be32_to_cpus(&ext.magic);
214         be32_to_cpus(&ext.len);
215         offset += sizeof(ext);
216 #ifdef DEBUG_EXT
217         printf("ext.magic = 0x%x\n", ext.magic);
218 #endif
219         if (offset > end_offset || ext.len > end_offset - offset) {
220             error_setg(errp, "Header extension too large");
221             return -EINVAL;
222         }
223 
224         switch (ext.magic) {
225         case QCOW2_EXT_MAGIC_END:
226             return 0;
227 
228         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
229             if (ext.len >= sizeof(bs->backing_format)) {
230                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
231                            " too large (>=%zu)", ext.len,
232                            sizeof(bs->backing_format));
233                 return 2;
234             }
235             ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len);
236             if (ret < 0) {
237                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
238                                  "Could not read format name");
239                 return 3;
240             }
241             bs->backing_format[ext.len] = '\0';
242             s->image_backing_format = g_strdup(bs->backing_format);
243 #ifdef DEBUG_EXT
244             printf("Qcow2: Got format extension %s\n", bs->backing_format);
245 #endif
246             break;
247 
248         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
249             if (p_feature_table != NULL) {
250                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
251                 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
252                 if (ret < 0) {
253                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
254                                      "Could not read table");
255                     return ret;
256                 }
257 
258                 *p_feature_table = feature_table;
259             }
260             break;
261 
262         case QCOW2_EXT_MAGIC_CRYPTO_HEADER: {
263             unsigned int cflags = 0;
264             if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
265                 error_setg(errp, "CRYPTO header extension only "
266                            "expected with LUKS encryption method");
267                 return -EINVAL;
268             }
269             if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) {
270                 error_setg(errp, "CRYPTO header extension size %u, "
271                            "but expected size %zu", ext.len,
272                            sizeof(Qcow2CryptoHeaderExtension));
273                 return -EINVAL;
274             }
275 
276             ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len);
277             if (ret < 0) {
278                 error_setg_errno(errp, -ret,
279                                  "Unable to read CRYPTO header extension");
280                 return ret;
281             }
282             be64_to_cpus(&s->crypto_header.offset);
283             be64_to_cpus(&s->crypto_header.length);
284 
285             if ((s->crypto_header.offset % s->cluster_size) != 0) {
286                 error_setg(errp, "Encryption header offset '%" PRIu64 "' is "
287                            "not a multiple of cluster size '%u'",
288                            s->crypto_header.offset, s->cluster_size);
289                 return -EINVAL;
290             }
291 
292             if (flags & BDRV_O_NO_IO) {
293                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
294             }
295             s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
296                                            qcow2_crypto_hdr_read_func,
297                                            bs, cflags, errp);
298             if (!s->crypto) {
299                 return -EINVAL;
300             }
301         }   break;
302 
303         case QCOW2_EXT_MAGIC_BITMAPS:
304             if (ext.len != sizeof(bitmaps_ext)) {
305                 error_setg_errno(errp, -ret, "bitmaps_ext: "
306                                  "Invalid extension length");
307                 return -EINVAL;
308             }
309 
310             if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) {
311                 if (s->qcow_version < 3) {
312                     /* Let's be a bit more specific */
313                     warn_report("This qcow2 v2 image contains bitmaps, but "
314                                 "they may have been modified by a program "
315                                 "without persistent bitmap support; so now "
316                                 "they must all be considered inconsistent");
317                 } else {
318                     warn_report("a program lacking bitmap support "
319                                 "modified this file, so all bitmaps are now "
320                                 "considered inconsistent");
321                 }
322                 error_printf("Some clusters may be leaked, "
323                              "run 'qemu-img check -r' on the image "
324                              "file to fix.");
325                 if (need_update_header != NULL) {
326                     /* Updating is needed to drop invalid bitmap extension. */
327                     *need_update_header = true;
328                 }
329                 break;
330             }
331 
332             ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len);
333             if (ret < 0) {
334                 error_setg_errno(errp, -ret, "bitmaps_ext: "
335                                  "Could not read ext header");
336                 return ret;
337             }
338 
339             if (bitmaps_ext.reserved32 != 0) {
340                 error_setg_errno(errp, -ret, "bitmaps_ext: "
341                                  "Reserved field is not zero");
342                 return -EINVAL;
343             }
344 
345             be32_to_cpus(&bitmaps_ext.nb_bitmaps);
346             be64_to_cpus(&bitmaps_ext.bitmap_directory_size);
347             be64_to_cpus(&bitmaps_ext.bitmap_directory_offset);
348 
349             if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) {
350                 error_setg(errp,
351                            "bitmaps_ext: Image has %" PRIu32 " bitmaps, "
352                            "exceeding the QEMU supported maximum of %d",
353                            bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS);
354                 return -EINVAL;
355             }
356 
357             if (bitmaps_ext.nb_bitmaps == 0) {
358                 error_setg(errp, "found bitmaps extension with zero bitmaps");
359                 return -EINVAL;
360             }
361 
362             if (bitmaps_ext.bitmap_directory_offset & (s->cluster_size - 1)) {
363                 error_setg(errp, "bitmaps_ext: "
364                                  "invalid bitmap directory offset");
365                 return -EINVAL;
366             }
367 
368             if (bitmaps_ext.bitmap_directory_size >
369                 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) {
370                 error_setg(errp, "bitmaps_ext: "
371                                  "bitmap directory size (%" PRIu64 ") exceeds "
372                                  "the maximum supported size (%d)",
373                                  bitmaps_ext.bitmap_directory_size,
374                                  QCOW2_MAX_BITMAP_DIRECTORY_SIZE);
375                 return -EINVAL;
376             }
377 
378             s->nb_bitmaps = bitmaps_ext.nb_bitmaps;
379             s->bitmap_directory_offset =
380                     bitmaps_ext.bitmap_directory_offset;
381             s->bitmap_directory_size =
382                     bitmaps_ext.bitmap_directory_size;
383 
384 #ifdef DEBUG_EXT
385             printf("Qcow2: Got bitmaps extension: "
386                    "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n",
387                    s->bitmap_directory_offset, s->nb_bitmaps);
388 #endif
389             break;
390 
391         default:
392             /* unknown magic - save it in case we need to rewrite the header */
393             /* If you add a new feature, make sure to also update the fast
394              * path of qcow2_make_empty() to deal with it. */
395             {
396                 Qcow2UnknownHeaderExtension *uext;
397 
398                 uext = g_malloc0(sizeof(*uext)  + ext.len);
399                 uext->magic = ext.magic;
400                 uext->len = ext.len;
401                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
402 
403                 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
404                 if (ret < 0) {
405                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
406                                      "Could not read data");
407                     return ret;
408                 }
409             }
410             break;
411         }
412 
413         offset += ((ext.len + 7) & ~7);
414     }
415 
416     return 0;
417 }
418 
419 static void cleanup_unknown_header_ext(BlockDriverState *bs)
420 {
421     BDRVQcow2State *s = bs->opaque;
422     Qcow2UnknownHeaderExtension *uext, *next;
423 
424     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
425         QLIST_REMOVE(uext, next);
426         g_free(uext);
427     }
428 }
429 
430 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
431                                        uint64_t mask)
432 {
433     char *features = g_strdup("");
434     char *old;
435 
436     while (table && table->name[0] != '\0') {
437         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
438             if (mask & (1ULL << table->bit)) {
439                 old = features;
440                 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
441                                            table->name);
442                 g_free(old);
443                 mask &= ~(1ULL << table->bit);
444             }
445         }
446         table++;
447     }
448 
449     if (mask) {
450         old = features;
451         features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
452                                    old, *old ? ", " : "", mask);
453         g_free(old);
454     }
455 
456     error_setg(errp, "Unsupported qcow2 feature(s): %s", features);
457     g_free(features);
458 }
459 
460 /*
461  * Sets the dirty bit and flushes afterwards if necessary.
462  *
463  * The incompatible_features bit is only set if the image file header was
464  * updated successfully.  Therefore it is not required to check the return
465  * value of this function.
466  */
467 int qcow2_mark_dirty(BlockDriverState *bs)
468 {
469     BDRVQcow2State *s = bs->opaque;
470     uint64_t val;
471     int ret;
472 
473     assert(s->qcow_version >= 3);
474 
475     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
476         return 0; /* already dirty */
477     }
478 
479     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
480     ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
481                       &val, sizeof(val));
482     if (ret < 0) {
483         return ret;
484     }
485     ret = bdrv_flush(bs->file->bs);
486     if (ret < 0) {
487         return ret;
488     }
489 
490     /* Only treat image as dirty if the header was updated successfully */
491     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
492     return 0;
493 }
494 
495 /*
496  * Clears the dirty bit and flushes before if necessary.  Only call this
497  * function when there are no pending requests, it does not guard against
498  * concurrent requests dirtying the image.
499  */
500 static int qcow2_mark_clean(BlockDriverState *bs)
501 {
502     BDRVQcow2State *s = bs->opaque;
503 
504     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
505         int ret;
506 
507         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
508 
509         ret = qcow2_flush_caches(bs);
510         if (ret < 0) {
511             return ret;
512         }
513 
514         return qcow2_update_header(bs);
515     }
516     return 0;
517 }
518 
519 /*
520  * Marks the image as corrupt.
521  */
522 int qcow2_mark_corrupt(BlockDriverState *bs)
523 {
524     BDRVQcow2State *s = bs->opaque;
525 
526     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
527     return qcow2_update_header(bs);
528 }
529 
530 /*
531  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
532  * before if necessary.
533  */
534 int qcow2_mark_consistent(BlockDriverState *bs)
535 {
536     BDRVQcow2State *s = bs->opaque;
537 
538     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
539         int ret = qcow2_flush_caches(bs);
540         if (ret < 0) {
541             return ret;
542         }
543 
544         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
545         return qcow2_update_header(bs);
546     }
547     return 0;
548 }
549 
550 static int coroutine_fn qcow2_co_check_locked(BlockDriverState *bs,
551                                               BdrvCheckResult *result,
552                                               BdrvCheckMode fix)
553 {
554     int ret = qcow2_check_refcounts(bs, result, fix);
555     if (ret < 0) {
556         return ret;
557     }
558 
559     if (fix && result->check_errors == 0 && result->corruptions == 0) {
560         ret = qcow2_mark_clean(bs);
561         if (ret < 0) {
562             return ret;
563         }
564         return qcow2_mark_consistent(bs);
565     }
566     return ret;
567 }
568 
569 static int coroutine_fn qcow2_co_check(BlockDriverState *bs,
570                                        BdrvCheckResult *result,
571                                        BdrvCheckMode fix)
572 {
573     BDRVQcow2State *s = bs->opaque;
574     int ret;
575 
576     qemu_co_mutex_lock(&s->lock);
577     ret = qcow2_co_check_locked(bs, result, fix);
578     qemu_co_mutex_unlock(&s->lock);
579     return ret;
580 }
581 
582 int qcow2_validate_table(BlockDriverState *bs, uint64_t offset,
583                          uint64_t entries, size_t entry_len,
584                          int64_t max_size_bytes, const char *table_name,
585                          Error **errp)
586 {
587     BDRVQcow2State *s = bs->opaque;
588 
589     if (entries > max_size_bytes / entry_len) {
590         error_setg(errp, "%s too large", table_name);
591         return -EFBIG;
592     }
593 
594     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
595      * because values will be passed to qemu functions taking int64_t. */
596     if ((INT64_MAX - entries * entry_len < offset) ||
597         (offset_into_cluster(s, offset) != 0)) {
598         error_setg(errp, "%s offset invalid", table_name);
599         return -EINVAL;
600     }
601 
602     return 0;
603 }
604 
605 static QemuOptsList qcow2_runtime_opts = {
606     .name = "qcow2",
607     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
608     .desc = {
609         {
610             .name = QCOW2_OPT_LAZY_REFCOUNTS,
611             .type = QEMU_OPT_BOOL,
612             .help = "Postpone refcount updates",
613         },
614         {
615             .name = QCOW2_OPT_DISCARD_REQUEST,
616             .type = QEMU_OPT_BOOL,
617             .help = "Pass guest discard requests to the layer below",
618         },
619         {
620             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
621             .type = QEMU_OPT_BOOL,
622             .help = "Generate discard requests when snapshot related space "
623                     "is freed",
624         },
625         {
626             .name = QCOW2_OPT_DISCARD_OTHER,
627             .type = QEMU_OPT_BOOL,
628             .help = "Generate discard requests when other clusters are freed",
629         },
630         {
631             .name = QCOW2_OPT_OVERLAP,
632             .type = QEMU_OPT_STRING,
633             .help = "Selects which overlap checks to perform from a range of "
634                     "templates (none, constant, cached, all)",
635         },
636         {
637             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
638             .type = QEMU_OPT_STRING,
639             .help = "Selects which overlap checks to perform from a range of "
640                     "templates (none, constant, cached, all)",
641         },
642         {
643             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
644             .type = QEMU_OPT_BOOL,
645             .help = "Check for unintended writes into the main qcow2 header",
646         },
647         {
648             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
649             .type = QEMU_OPT_BOOL,
650             .help = "Check for unintended writes into the active L1 table",
651         },
652         {
653             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
654             .type = QEMU_OPT_BOOL,
655             .help = "Check for unintended writes into an active L2 table",
656         },
657         {
658             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
659             .type = QEMU_OPT_BOOL,
660             .help = "Check for unintended writes into the refcount table",
661         },
662         {
663             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
664             .type = QEMU_OPT_BOOL,
665             .help = "Check for unintended writes into a refcount block",
666         },
667         {
668             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
669             .type = QEMU_OPT_BOOL,
670             .help = "Check for unintended writes into the snapshot table",
671         },
672         {
673             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
674             .type = QEMU_OPT_BOOL,
675             .help = "Check for unintended writes into an inactive L1 table",
676         },
677         {
678             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
679             .type = QEMU_OPT_BOOL,
680             .help = "Check for unintended writes into an inactive L2 table",
681         },
682         {
683             .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
684             .type = QEMU_OPT_BOOL,
685             .help = "Check for unintended writes into the bitmap directory",
686         },
687         {
688             .name = QCOW2_OPT_CACHE_SIZE,
689             .type = QEMU_OPT_SIZE,
690             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
691                     "cache size",
692         },
693         {
694             .name = QCOW2_OPT_L2_CACHE_SIZE,
695             .type = QEMU_OPT_SIZE,
696             .help = "Maximum L2 table cache size",
697         },
698         {
699             .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
700             .type = QEMU_OPT_SIZE,
701             .help = "Size of each entry in the L2 cache",
702         },
703         {
704             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
705             .type = QEMU_OPT_SIZE,
706             .help = "Maximum refcount block cache size",
707         },
708         {
709             .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
710             .type = QEMU_OPT_NUMBER,
711             .help = "Clean unused cache entries after this time (in seconds)",
712         },
713         BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
714             "ID of secret providing qcow2 AES key or LUKS passphrase"),
715         { /* end of list */ }
716     },
717 };
718 
719 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
720     [QCOW2_OL_MAIN_HEADER_BITNR]      = QCOW2_OPT_OVERLAP_MAIN_HEADER,
721     [QCOW2_OL_ACTIVE_L1_BITNR]        = QCOW2_OPT_OVERLAP_ACTIVE_L1,
722     [QCOW2_OL_ACTIVE_L2_BITNR]        = QCOW2_OPT_OVERLAP_ACTIVE_L2,
723     [QCOW2_OL_REFCOUNT_TABLE_BITNR]   = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
724     [QCOW2_OL_REFCOUNT_BLOCK_BITNR]   = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
725     [QCOW2_OL_SNAPSHOT_TABLE_BITNR]   = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
726     [QCOW2_OL_INACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_INACTIVE_L1,
727     [QCOW2_OL_INACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_INACTIVE_L2,
728     [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
729 };
730 
731 static void cache_clean_timer_cb(void *opaque)
732 {
733     BlockDriverState *bs = opaque;
734     BDRVQcow2State *s = bs->opaque;
735     qcow2_cache_clean_unused(s->l2_table_cache);
736     qcow2_cache_clean_unused(s->refcount_block_cache);
737     timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
738               (int64_t) s->cache_clean_interval * 1000);
739 }
740 
741 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
742 {
743     BDRVQcow2State *s = bs->opaque;
744     if (s->cache_clean_interval > 0) {
745         s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
746                                              SCALE_MS, cache_clean_timer_cb,
747                                              bs);
748         timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
749                   (int64_t) s->cache_clean_interval * 1000);
750     }
751 }
752 
753 static void cache_clean_timer_del(BlockDriverState *bs)
754 {
755     BDRVQcow2State *s = bs->opaque;
756     if (s->cache_clean_timer) {
757         timer_del(s->cache_clean_timer);
758         timer_free(s->cache_clean_timer);
759         s->cache_clean_timer = NULL;
760     }
761 }
762 
763 static void qcow2_detach_aio_context(BlockDriverState *bs)
764 {
765     cache_clean_timer_del(bs);
766 }
767 
768 static void qcow2_attach_aio_context(BlockDriverState *bs,
769                                      AioContext *new_context)
770 {
771     cache_clean_timer_init(bs, new_context);
772 }
773 
774 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
775                              uint64_t *l2_cache_size,
776                              uint64_t *l2_cache_entry_size,
777                              uint64_t *refcount_cache_size, Error **errp)
778 {
779     BDRVQcow2State *s = bs->opaque;
780     uint64_t combined_cache_size, l2_cache_max_setting;
781     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
782     int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size;
783     uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
784     uint64_t max_l2_cache = virtual_disk_size / (s->cluster_size / 8);
785 
786     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
787     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
788     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
789 
790     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
791     l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE,
792                                              DEFAULT_L2_CACHE_MAX_SIZE);
793     *refcount_cache_size = qemu_opt_get_size(opts,
794                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
795 
796     *l2_cache_entry_size = qemu_opt_get_size(
797         opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size);
798 
799     *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting);
800 
801     if (combined_cache_size_set) {
802         if (l2_cache_size_set && refcount_cache_size_set) {
803             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
804                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
805                        "at the same time");
806             return;
807         } else if (l2_cache_size_set &&
808                    (l2_cache_max_setting > combined_cache_size)) {
809             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
810                        QCOW2_OPT_CACHE_SIZE);
811             return;
812         } else if (*refcount_cache_size > combined_cache_size) {
813             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
814                        QCOW2_OPT_CACHE_SIZE);
815             return;
816         }
817 
818         if (l2_cache_size_set) {
819             *refcount_cache_size = combined_cache_size - *l2_cache_size;
820         } else if (refcount_cache_size_set) {
821             *l2_cache_size = combined_cache_size - *refcount_cache_size;
822         } else {
823             /* Assign as much memory as possible to the L2 cache, and
824              * use the remainder for the refcount cache */
825             if (combined_cache_size >= max_l2_cache + min_refcount_cache) {
826                 *l2_cache_size = max_l2_cache;
827                 *refcount_cache_size = combined_cache_size - *l2_cache_size;
828             } else {
829                 *refcount_cache_size =
830                     MIN(combined_cache_size, min_refcount_cache);
831                 *l2_cache_size = combined_cache_size - *refcount_cache_size;
832             }
833         }
834     }
835     /* l2_cache_size and refcount_cache_size are ensured to have at least
836      * their minimum values in qcow2_update_options_prepare() */
837 
838     if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) ||
839         *l2_cache_entry_size > s->cluster_size ||
840         !is_power_of_2(*l2_cache_entry_size)) {
841         error_setg(errp, "L2 cache entry size must be a power of two "
842                    "between %d and the cluster size (%d)",
843                    1 << MIN_CLUSTER_BITS, s->cluster_size);
844         return;
845     }
846 }
847 
848 typedef struct Qcow2ReopenState {
849     Qcow2Cache *l2_table_cache;
850     Qcow2Cache *refcount_block_cache;
851     int l2_slice_size; /* Number of entries in a slice of the L2 table */
852     bool use_lazy_refcounts;
853     int overlap_check;
854     bool discard_passthrough[QCOW2_DISCARD_MAX];
855     uint64_t cache_clean_interval;
856     QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
857 } Qcow2ReopenState;
858 
859 static int qcow2_update_options_prepare(BlockDriverState *bs,
860                                         Qcow2ReopenState *r,
861                                         QDict *options, int flags,
862                                         Error **errp)
863 {
864     BDRVQcow2State *s = bs->opaque;
865     QemuOpts *opts = NULL;
866     const char *opt_overlap_check, *opt_overlap_check_template;
867     int overlap_check_template = 0;
868     uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size;
869     int i;
870     const char *encryptfmt;
871     QDict *encryptopts = NULL;
872     Error *local_err = NULL;
873     int ret;
874 
875     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
876     encryptfmt = qdict_get_try_str(encryptopts, "format");
877 
878     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
879     qemu_opts_absorb_qdict(opts, options, &local_err);
880     if (local_err) {
881         error_propagate(errp, local_err);
882         ret = -EINVAL;
883         goto fail;
884     }
885 
886     /* get L2 table/refcount block cache size from command line options */
887     read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size,
888                      &refcount_cache_size, &local_err);
889     if (local_err) {
890         error_propagate(errp, local_err);
891         ret = -EINVAL;
892         goto fail;
893     }
894 
895     l2_cache_size /= l2_cache_entry_size;
896     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
897         l2_cache_size = MIN_L2_CACHE_SIZE;
898     }
899     if (l2_cache_size > INT_MAX) {
900         error_setg(errp, "L2 cache size too big");
901         ret = -EINVAL;
902         goto fail;
903     }
904 
905     refcount_cache_size /= s->cluster_size;
906     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
907         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
908     }
909     if (refcount_cache_size > INT_MAX) {
910         error_setg(errp, "Refcount cache size too big");
911         ret = -EINVAL;
912         goto fail;
913     }
914 
915     /* alloc new L2 table/refcount block cache, flush old one */
916     if (s->l2_table_cache) {
917         ret = qcow2_cache_flush(bs, s->l2_table_cache);
918         if (ret) {
919             error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
920             goto fail;
921         }
922     }
923 
924     if (s->refcount_block_cache) {
925         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
926         if (ret) {
927             error_setg_errno(errp, -ret,
928                              "Failed to flush the refcount block cache");
929             goto fail;
930         }
931     }
932 
933     r->l2_slice_size = l2_cache_entry_size / sizeof(uint64_t);
934     r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size,
935                                            l2_cache_entry_size);
936     r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size,
937                                                  s->cluster_size);
938     if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
939         error_setg(errp, "Could not allocate metadata caches");
940         ret = -ENOMEM;
941         goto fail;
942     }
943 
944     /* New interval for cache cleanup timer */
945     r->cache_clean_interval =
946         qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
947                             DEFAULT_CACHE_CLEAN_INTERVAL);
948 #ifndef CONFIG_LINUX
949     if (r->cache_clean_interval != 0) {
950         error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL
951                    " not supported on this host");
952         ret = -EINVAL;
953         goto fail;
954     }
955 #endif
956     if (r->cache_clean_interval > UINT_MAX) {
957         error_setg(errp, "Cache clean interval too big");
958         ret = -EINVAL;
959         goto fail;
960     }
961 
962     /* lazy-refcounts; flush if going from enabled to disabled */
963     r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
964         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
965     if (r->use_lazy_refcounts && s->qcow_version < 3) {
966         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
967                    "qemu 1.1 compatibility level");
968         ret = -EINVAL;
969         goto fail;
970     }
971 
972     if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
973         ret = qcow2_mark_clean(bs);
974         if (ret < 0) {
975             error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
976             goto fail;
977         }
978     }
979 
980     /* Overlap check options */
981     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
982     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
983     if (opt_overlap_check_template && opt_overlap_check &&
984         strcmp(opt_overlap_check_template, opt_overlap_check))
985     {
986         error_setg(errp, "Conflicting values for qcow2 options '"
987                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
988                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
989         ret = -EINVAL;
990         goto fail;
991     }
992     if (!opt_overlap_check) {
993         opt_overlap_check = opt_overlap_check_template ?: "cached";
994     }
995 
996     if (!strcmp(opt_overlap_check, "none")) {
997         overlap_check_template = 0;
998     } else if (!strcmp(opt_overlap_check, "constant")) {
999         overlap_check_template = QCOW2_OL_CONSTANT;
1000     } else if (!strcmp(opt_overlap_check, "cached")) {
1001         overlap_check_template = QCOW2_OL_CACHED;
1002     } else if (!strcmp(opt_overlap_check, "all")) {
1003         overlap_check_template = QCOW2_OL_ALL;
1004     } else {
1005         error_setg(errp, "Unsupported value '%s' for qcow2 option "
1006                    "'overlap-check'. Allowed are any of the following: "
1007                    "none, constant, cached, all", opt_overlap_check);
1008         ret = -EINVAL;
1009         goto fail;
1010     }
1011 
1012     r->overlap_check = 0;
1013     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
1014         /* overlap-check defines a template bitmask, but every flag may be
1015          * overwritten through the associated boolean option */
1016         r->overlap_check |=
1017             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
1018                               overlap_check_template & (1 << i)) << i;
1019     }
1020 
1021     r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
1022     r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
1023     r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
1024         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
1025                           flags & BDRV_O_UNMAP);
1026     r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
1027         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
1028     r->discard_passthrough[QCOW2_DISCARD_OTHER] =
1029         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
1030 
1031     switch (s->crypt_method_header) {
1032     case QCOW_CRYPT_NONE:
1033         if (encryptfmt) {
1034             error_setg(errp, "No encryption in image header, but options "
1035                        "specified format '%s'", encryptfmt);
1036             ret = -EINVAL;
1037             goto fail;
1038         }
1039         break;
1040 
1041     case QCOW_CRYPT_AES:
1042         if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
1043             error_setg(errp,
1044                        "Header reported 'aes' encryption format but "
1045                        "options specify '%s'", encryptfmt);
1046             ret = -EINVAL;
1047             goto fail;
1048         }
1049         qdict_put_str(encryptopts, "format", "qcow");
1050         r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1051         break;
1052 
1053     case QCOW_CRYPT_LUKS:
1054         if (encryptfmt && !g_str_equal(encryptfmt, "luks")) {
1055             error_setg(errp,
1056                        "Header reported 'luks' encryption format but "
1057                        "options specify '%s'", encryptfmt);
1058             ret = -EINVAL;
1059             goto fail;
1060         }
1061         qdict_put_str(encryptopts, "format", "luks");
1062         r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
1063         break;
1064 
1065     default:
1066         error_setg(errp, "Unsupported encryption method %d",
1067                    s->crypt_method_header);
1068         break;
1069     }
1070     if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) {
1071         ret = -EINVAL;
1072         goto fail;
1073     }
1074 
1075     ret = 0;
1076 fail:
1077     qobject_unref(encryptopts);
1078     qemu_opts_del(opts);
1079     opts = NULL;
1080     return ret;
1081 }
1082 
1083 static void qcow2_update_options_commit(BlockDriverState *bs,
1084                                         Qcow2ReopenState *r)
1085 {
1086     BDRVQcow2State *s = bs->opaque;
1087     int i;
1088 
1089     if (s->l2_table_cache) {
1090         qcow2_cache_destroy(s->l2_table_cache);
1091     }
1092     if (s->refcount_block_cache) {
1093         qcow2_cache_destroy(s->refcount_block_cache);
1094     }
1095     s->l2_table_cache = r->l2_table_cache;
1096     s->refcount_block_cache = r->refcount_block_cache;
1097     s->l2_slice_size = r->l2_slice_size;
1098 
1099     s->overlap_check = r->overlap_check;
1100     s->use_lazy_refcounts = r->use_lazy_refcounts;
1101 
1102     for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
1103         s->discard_passthrough[i] = r->discard_passthrough[i];
1104     }
1105 
1106     if (s->cache_clean_interval != r->cache_clean_interval) {
1107         cache_clean_timer_del(bs);
1108         s->cache_clean_interval = r->cache_clean_interval;
1109         cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
1110     }
1111 
1112     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1113     s->crypto_opts = r->crypto_opts;
1114 }
1115 
1116 static void qcow2_update_options_abort(BlockDriverState *bs,
1117                                        Qcow2ReopenState *r)
1118 {
1119     if (r->l2_table_cache) {
1120         qcow2_cache_destroy(r->l2_table_cache);
1121     }
1122     if (r->refcount_block_cache) {
1123         qcow2_cache_destroy(r->refcount_block_cache);
1124     }
1125     qapi_free_QCryptoBlockOpenOptions(r->crypto_opts);
1126 }
1127 
1128 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
1129                                 int flags, Error **errp)
1130 {
1131     Qcow2ReopenState r = {};
1132     int ret;
1133 
1134     ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
1135     if (ret >= 0) {
1136         qcow2_update_options_commit(bs, &r);
1137     } else {
1138         qcow2_update_options_abort(bs, &r);
1139     }
1140 
1141     return ret;
1142 }
1143 
1144 /* Called with s->lock held.  */
1145 static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
1146                                       int flags, Error **errp)
1147 {
1148     BDRVQcow2State *s = bs->opaque;
1149     unsigned int len, i;
1150     int ret = 0;
1151     QCowHeader header;
1152     Error *local_err = NULL;
1153     uint64_t ext_end;
1154     uint64_t l1_vm_state_index;
1155     bool update_header = false;
1156 
1157     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
1158     if (ret < 0) {
1159         error_setg_errno(errp, -ret, "Could not read qcow2 header");
1160         goto fail;
1161     }
1162     be32_to_cpus(&header.magic);
1163     be32_to_cpus(&header.version);
1164     be64_to_cpus(&header.backing_file_offset);
1165     be32_to_cpus(&header.backing_file_size);
1166     be64_to_cpus(&header.size);
1167     be32_to_cpus(&header.cluster_bits);
1168     be32_to_cpus(&header.crypt_method);
1169     be64_to_cpus(&header.l1_table_offset);
1170     be32_to_cpus(&header.l1_size);
1171     be64_to_cpus(&header.refcount_table_offset);
1172     be32_to_cpus(&header.refcount_table_clusters);
1173     be64_to_cpus(&header.snapshots_offset);
1174     be32_to_cpus(&header.nb_snapshots);
1175 
1176     if (header.magic != QCOW_MAGIC) {
1177         error_setg(errp, "Image is not in qcow2 format");
1178         ret = -EINVAL;
1179         goto fail;
1180     }
1181     if (header.version < 2 || header.version > 3) {
1182         error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
1183         ret = -ENOTSUP;
1184         goto fail;
1185     }
1186 
1187     s->qcow_version = header.version;
1188 
1189     /* Initialise cluster size */
1190     if (header.cluster_bits < MIN_CLUSTER_BITS ||
1191         header.cluster_bits > MAX_CLUSTER_BITS) {
1192         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
1193                    header.cluster_bits);
1194         ret = -EINVAL;
1195         goto fail;
1196     }
1197 
1198     s->cluster_bits = header.cluster_bits;
1199     s->cluster_size = 1 << s->cluster_bits;
1200     s->cluster_sectors = 1 << (s->cluster_bits - BDRV_SECTOR_BITS);
1201 
1202     /* Initialise version 3 header fields */
1203     if (header.version == 2) {
1204         header.incompatible_features    = 0;
1205         header.compatible_features      = 0;
1206         header.autoclear_features       = 0;
1207         header.refcount_order           = 4;
1208         header.header_length            = 72;
1209     } else {
1210         be64_to_cpus(&header.incompatible_features);
1211         be64_to_cpus(&header.compatible_features);
1212         be64_to_cpus(&header.autoclear_features);
1213         be32_to_cpus(&header.refcount_order);
1214         be32_to_cpus(&header.header_length);
1215 
1216         if (header.header_length < 104) {
1217             error_setg(errp, "qcow2 header too short");
1218             ret = -EINVAL;
1219             goto fail;
1220         }
1221     }
1222 
1223     if (header.header_length > s->cluster_size) {
1224         error_setg(errp, "qcow2 header exceeds cluster size");
1225         ret = -EINVAL;
1226         goto fail;
1227     }
1228 
1229     if (header.header_length > sizeof(header)) {
1230         s->unknown_header_fields_size = header.header_length - sizeof(header);
1231         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
1232         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
1233                          s->unknown_header_fields_size);
1234         if (ret < 0) {
1235             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
1236                              "fields");
1237             goto fail;
1238         }
1239     }
1240 
1241     if (header.backing_file_offset > s->cluster_size) {
1242         error_setg(errp, "Invalid backing file offset");
1243         ret = -EINVAL;
1244         goto fail;
1245     }
1246 
1247     if (header.backing_file_offset) {
1248         ext_end = header.backing_file_offset;
1249     } else {
1250         ext_end = 1 << header.cluster_bits;
1251     }
1252 
1253     /* Handle feature bits */
1254     s->incompatible_features    = header.incompatible_features;
1255     s->compatible_features      = header.compatible_features;
1256     s->autoclear_features       = header.autoclear_features;
1257 
1258     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
1259         void *feature_table = NULL;
1260         qcow2_read_extensions(bs, header.header_length, ext_end,
1261                               &feature_table, flags, NULL, NULL);
1262         report_unsupported_feature(errp, feature_table,
1263                                    s->incompatible_features &
1264                                    ~QCOW2_INCOMPAT_MASK);
1265         ret = -ENOTSUP;
1266         g_free(feature_table);
1267         goto fail;
1268     }
1269 
1270     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
1271         /* Corrupt images may not be written to unless they are being repaired
1272          */
1273         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
1274             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
1275                        "read/write");
1276             ret = -EACCES;
1277             goto fail;
1278         }
1279     }
1280 
1281     /* Check support for various header values */
1282     if (header.refcount_order > 6) {
1283         error_setg(errp, "Reference count entry width too large; may not "
1284                    "exceed 64 bits");
1285         ret = -EINVAL;
1286         goto fail;
1287     }
1288     s->refcount_order = header.refcount_order;
1289     s->refcount_bits = 1 << s->refcount_order;
1290     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1291     s->refcount_max += s->refcount_max - 1;
1292 
1293     s->crypt_method_header = header.crypt_method;
1294     if (s->crypt_method_header) {
1295         if (bdrv_uses_whitelist() &&
1296             s->crypt_method_header == QCOW_CRYPT_AES) {
1297             error_setg(errp,
1298                        "Use of AES-CBC encrypted qcow2 images is no longer "
1299                        "supported in system emulators");
1300             error_append_hint(errp,
1301                               "You can use 'qemu-img convert' to convert your "
1302                               "image to an alternative supported format, such "
1303                               "as unencrypted qcow2, or raw with the LUKS "
1304                               "format instead.\n");
1305             ret = -ENOSYS;
1306             goto fail;
1307         }
1308 
1309         if (s->crypt_method_header == QCOW_CRYPT_AES) {
1310             s->crypt_physical_offset = false;
1311         } else {
1312             /* Assuming LUKS and any future crypt methods we
1313              * add will all use physical offsets, due to the
1314              * fact that the alternative is insecure...  */
1315             s->crypt_physical_offset = true;
1316         }
1317 
1318         bs->encrypted = true;
1319     }
1320 
1321     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
1322     s->l2_size = 1 << s->l2_bits;
1323     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1324     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1325     s->refcount_block_size = 1 << s->refcount_block_bits;
1326     bs->total_sectors = header.size / BDRV_SECTOR_SIZE;
1327     s->csize_shift = (62 - (s->cluster_bits - 8));
1328     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1329     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1330 
1331     s->refcount_table_offset = header.refcount_table_offset;
1332     s->refcount_table_size =
1333         header.refcount_table_clusters << (s->cluster_bits - 3);
1334 
1335     if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) {
1336         error_setg(errp, "Image does not contain a reference count table");
1337         ret = -EINVAL;
1338         goto fail;
1339     }
1340 
1341     ret = qcow2_validate_table(bs, s->refcount_table_offset,
1342                                header.refcount_table_clusters,
1343                                s->cluster_size, QCOW_MAX_REFTABLE_SIZE,
1344                                "Reference count table", errp);
1345     if (ret < 0) {
1346         goto fail;
1347     }
1348 
1349     /* The total size in bytes of the snapshot table is checked in
1350      * qcow2_read_snapshots() because the size of each snapshot is
1351      * variable and we don't know it yet.
1352      * Here we only check the offset and number of snapshots. */
1353     ret = qcow2_validate_table(bs, header.snapshots_offset,
1354                                header.nb_snapshots,
1355                                sizeof(QCowSnapshotHeader),
1356                                sizeof(QCowSnapshotHeader) * QCOW_MAX_SNAPSHOTS,
1357                                "Snapshot table", errp);
1358     if (ret < 0) {
1359         goto fail;
1360     }
1361 
1362     /* read the level 1 table */
1363     ret = qcow2_validate_table(bs, header.l1_table_offset,
1364                                header.l1_size, sizeof(uint64_t),
1365                                QCOW_MAX_L1_SIZE, "Active L1 table", errp);
1366     if (ret < 0) {
1367         goto fail;
1368     }
1369     s->l1_size = header.l1_size;
1370     s->l1_table_offset = header.l1_table_offset;
1371 
1372     l1_vm_state_index = size_to_l1(s, header.size);
1373     if (l1_vm_state_index > INT_MAX) {
1374         error_setg(errp, "Image is too big");
1375         ret = -EFBIG;
1376         goto fail;
1377     }
1378     s->l1_vm_state_index = l1_vm_state_index;
1379 
1380     /* the L1 table must contain at least enough entries to put
1381        header.size bytes */
1382     if (s->l1_size < s->l1_vm_state_index) {
1383         error_setg(errp, "L1 table is too small");
1384         ret = -EINVAL;
1385         goto fail;
1386     }
1387 
1388     if (s->l1_size > 0) {
1389         s->l1_table = qemu_try_blockalign(bs->file->bs,
1390             ROUND_UP(s->l1_size * sizeof(uint64_t), 512));
1391         if (s->l1_table == NULL) {
1392             error_setg(errp, "Could not allocate L1 table");
1393             ret = -ENOMEM;
1394             goto fail;
1395         }
1396         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
1397                          s->l1_size * sizeof(uint64_t));
1398         if (ret < 0) {
1399             error_setg_errno(errp, -ret, "Could not read L1 table");
1400             goto fail;
1401         }
1402         for(i = 0;i < s->l1_size; i++) {
1403             be64_to_cpus(&s->l1_table[i]);
1404         }
1405     }
1406 
1407     /* Parse driver-specific options */
1408     ret = qcow2_update_options(bs, options, flags, errp);
1409     if (ret < 0) {
1410         goto fail;
1411     }
1412 
1413     s->cluster_cache_offset = -1;
1414     s->flags = flags;
1415 
1416     ret = qcow2_refcount_init(bs);
1417     if (ret != 0) {
1418         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1419         goto fail;
1420     }
1421 
1422     QLIST_INIT(&s->cluster_allocs);
1423     QTAILQ_INIT(&s->discards);
1424 
1425     /* read qcow2 extensions */
1426     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1427                               flags, &update_header, &local_err)) {
1428         error_propagate(errp, local_err);
1429         ret = -EINVAL;
1430         goto fail;
1431     }
1432 
1433     /* qcow2_read_extension may have set up the crypto context
1434      * if the crypt method needs a header region, some methods
1435      * don't need header extensions, so must check here
1436      */
1437     if (s->crypt_method_header && !s->crypto) {
1438         if (s->crypt_method_header == QCOW_CRYPT_AES) {
1439             unsigned int cflags = 0;
1440             if (flags & BDRV_O_NO_IO) {
1441                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1442             }
1443             s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
1444                                            NULL, NULL, cflags, errp);
1445             if (!s->crypto) {
1446                 ret = -EINVAL;
1447                 goto fail;
1448             }
1449         } else if (!(flags & BDRV_O_NO_IO)) {
1450             error_setg(errp, "Missing CRYPTO header for crypt method %d",
1451                        s->crypt_method_header);
1452             ret = -EINVAL;
1453             goto fail;
1454         }
1455     }
1456 
1457     /* read the backing file name */
1458     if (header.backing_file_offset != 0) {
1459         len = header.backing_file_size;
1460         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1461             len >= sizeof(bs->backing_file)) {
1462             error_setg(errp, "Backing file name too long");
1463             ret = -EINVAL;
1464             goto fail;
1465         }
1466         ret = bdrv_pread(bs->file, header.backing_file_offset,
1467                          bs->backing_file, len);
1468         if (ret < 0) {
1469             error_setg_errno(errp, -ret, "Could not read backing file name");
1470             goto fail;
1471         }
1472         bs->backing_file[len] = '\0';
1473         s->image_backing_file = g_strdup(bs->backing_file);
1474     }
1475 
1476     /* Internal snapshots */
1477     s->snapshots_offset = header.snapshots_offset;
1478     s->nb_snapshots = header.nb_snapshots;
1479 
1480     ret = qcow2_read_snapshots(bs);
1481     if (ret < 0) {
1482         error_setg_errno(errp, -ret, "Could not read snapshots");
1483         goto fail;
1484     }
1485 
1486     /* Clear unknown autoclear feature bits */
1487     update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
1488     update_header =
1489         update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE);
1490     if (update_header) {
1491         s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
1492     }
1493 
1494     /* == Handle persistent dirty bitmaps ==
1495      *
1496      * We want load dirty bitmaps in three cases:
1497      *
1498      * 1. Normal open of the disk in active mode, not related to invalidation
1499      *    after migration.
1500      *
1501      * 2. Invalidation of the target vm after pre-copy phase of migration, if
1502      *    bitmaps are _not_ migrating through migration channel, i.e.
1503      *    'dirty-bitmaps' capability is disabled.
1504      *
1505      * 3. Invalidation of source vm after failed or canceled migration.
1506      *    This is a very interesting case. There are two possible types of
1507      *    bitmaps:
1508      *
1509      *    A. Stored on inactivation and removed. They should be loaded from the
1510      *       image.
1511      *
1512      *    B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1513      *       the migration channel (with dirty-bitmaps capability).
1514      *
1515      *    On the other hand, there are two possible sub-cases:
1516      *
1517      *    3.1 disk was changed by somebody else while were inactive. In this
1518      *        case all in-RAM dirty bitmaps (both persistent and not) are
1519      *        definitely invalid. And we don't have any method to determine
1520      *        this.
1521      *
1522      *        Simple and safe thing is to just drop all the bitmaps of type B on
1523      *        inactivation. But in this case we lose bitmaps in valid 4.2 case.
1524      *
1525      *        On the other hand, resuming source vm, if disk was already changed
1526      *        is a bad thing anyway: not only bitmaps, the whole vm state is
1527      *        out of sync with disk.
1528      *
1529      *        This means, that user or management tool, who for some reason
1530      *        decided to resume source vm, after disk was already changed by
1531      *        target vm, should at least drop all dirty bitmaps by hand.
1532      *
1533      *        So, we can ignore this case for now, but TODO: "generation"
1534      *        extension for qcow2, to determine, that image was changed after
1535      *        last inactivation. And if it is changed, we will drop (or at least
1536      *        mark as 'invalid' all the bitmaps of type B, both persistent
1537      *        and not).
1538      *
1539      *    3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1540      *        to disk ('dirty-bitmaps' capability disabled), or not saved
1541      *        ('dirty-bitmaps' capability enabled), but we don't need to care
1542      *        of: let's load bitmaps as always: stored bitmaps will be loaded,
1543      *        and not stored has flag IN_USE=1 in the image and will be skipped
1544      *        on loading.
1545      *
1546      * One remaining possible case when we don't want load bitmaps:
1547      *
1548      * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1549      *    will be loaded on invalidation, no needs try loading them before)
1550      */
1551 
1552     if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) {
1553         /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1554         bool header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
1555 
1556         update_header = update_header && !header_updated;
1557     }
1558     if (local_err != NULL) {
1559         error_propagate(errp, local_err);
1560         ret = -EINVAL;
1561         goto fail;
1562     }
1563 
1564     if (update_header) {
1565         ret = qcow2_update_header(bs);
1566         if (ret < 0) {
1567             error_setg_errno(errp, -ret, "Could not update qcow2 header");
1568             goto fail;
1569         }
1570     }
1571 
1572     bs->supported_zero_flags = header.version >= 3 ? BDRV_REQ_MAY_UNMAP : 0;
1573 
1574     /* Repair image if dirty */
1575     if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1576         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1577         BdrvCheckResult result = {0};
1578 
1579         ret = qcow2_co_check_locked(bs, &result,
1580                                     BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1581         if (ret < 0 || result.check_errors) {
1582             if (ret >= 0) {
1583                 ret = -EIO;
1584             }
1585             error_setg_errno(errp, -ret, "Could not repair dirty image");
1586             goto fail;
1587         }
1588     }
1589 
1590 #ifdef DEBUG_ALLOC
1591     {
1592         BdrvCheckResult result = {0};
1593         qcow2_check_refcounts(bs, &result, 0);
1594     }
1595 #endif
1596 
1597     qemu_co_queue_init(&s->compress_wait_queue);
1598 
1599     return ret;
1600 
1601  fail:
1602     g_free(s->unknown_header_fields);
1603     cleanup_unknown_header_ext(bs);
1604     qcow2_free_snapshots(bs);
1605     qcow2_refcount_close(bs);
1606     qemu_vfree(s->l1_table);
1607     /* else pre-write overlap checks in cache_destroy may crash */
1608     s->l1_table = NULL;
1609     cache_clean_timer_del(bs);
1610     if (s->l2_table_cache) {
1611         qcow2_cache_destroy(s->l2_table_cache);
1612     }
1613     if (s->refcount_block_cache) {
1614         qcow2_cache_destroy(s->refcount_block_cache);
1615     }
1616     qcrypto_block_free(s->crypto);
1617     qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1618     return ret;
1619 }
1620 
1621 typedef struct QCow2OpenCo {
1622     BlockDriverState *bs;
1623     QDict *options;
1624     int flags;
1625     Error **errp;
1626     int ret;
1627 } QCow2OpenCo;
1628 
1629 static void coroutine_fn qcow2_open_entry(void *opaque)
1630 {
1631     QCow2OpenCo *qoc = opaque;
1632     BDRVQcow2State *s = qoc->bs->opaque;
1633 
1634     qemu_co_mutex_lock(&s->lock);
1635     qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp);
1636     qemu_co_mutex_unlock(&s->lock);
1637 }
1638 
1639 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1640                       Error **errp)
1641 {
1642     BDRVQcow2State *s = bs->opaque;
1643     QCow2OpenCo qoc = {
1644         .bs = bs,
1645         .options = options,
1646         .flags = flags,
1647         .errp = errp,
1648         .ret = -EINPROGRESS
1649     };
1650 
1651     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
1652                                false, errp);
1653     if (!bs->file) {
1654         return -EINVAL;
1655     }
1656 
1657     /* Initialise locks */
1658     qemu_co_mutex_init(&s->lock);
1659 
1660     if (qemu_in_coroutine()) {
1661         /* From bdrv_co_create.  */
1662         qcow2_open_entry(&qoc);
1663     } else {
1664         qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc));
1665         BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
1666     }
1667     return qoc.ret;
1668 }
1669 
1670 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1671 {
1672     BDRVQcow2State *s = bs->opaque;
1673 
1674     if (bs->encrypted) {
1675         /* Encryption works on a sector granularity */
1676         bs->bl.request_alignment = BDRV_SECTOR_SIZE;
1677     }
1678     bs->bl.pwrite_zeroes_alignment = s->cluster_size;
1679     bs->bl.pdiscard_alignment = s->cluster_size;
1680 }
1681 
1682 static int qcow2_reopen_prepare(BDRVReopenState *state,
1683                                 BlockReopenQueue *queue, Error **errp)
1684 {
1685     Qcow2ReopenState *r;
1686     int ret;
1687 
1688     r = g_new0(Qcow2ReopenState, 1);
1689     state->opaque = r;
1690 
1691     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1692                                        state->flags, errp);
1693     if (ret < 0) {
1694         goto fail;
1695     }
1696 
1697     /* We need to write out any unwritten data if we reopen read-only. */
1698     if ((state->flags & BDRV_O_RDWR) == 0) {
1699         ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1700         if (ret < 0) {
1701             goto fail;
1702         }
1703 
1704         ret = bdrv_flush(state->bs);
1705         if (ret < 0) {
1706             goto fail;
1707         }
1708 
1709         ret = qcow2_mark_clean(state->bs);
1710         if (ret < 0) {
1711             goto fail;
1712         }
1713     }
1714 
1715     return 0;
1716 
1717 fail:
1718     qcow2_update_options_abort(state->bs, r);
1719     g_free(r);
1720     return ret;
1721 }
1722 
1723 static void qcow2_reopen_commit(BDRVReopenState *state)
1724 {
1725     qcow2_update_options_commit(state->bs, state->opaque);
1726     g_free(state->opaque);
1727 }
1728 
1729 static void qcow2_reopen_abort(BDRVReopenState *state)
1730 {
1731     qcow2_update_options_abort(state->bs, state->opaque);
1732     g_free(state->opaque);
1733 }
1734 
1735 static void qcow2_join_options(QDict *options, QDict *old_options)
1736 {
1737     bool has_new_overlap_template =
1738         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1739         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1740     bool has_new_total_cache_size =
1741         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1742     bool has_all_cache_options;
1743 
1744     /* New overlap template overrides all old overlap options */
1745     if (has_new_overlap_template) {
1746         qdict_del(old_options, QCOW2_OPT_OVERLAP);
1747         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1748         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1749         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1750         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1751         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1752         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1753         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1754         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1755         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1756     }
1757 
1758     /* New total cache size overrides all old options */
1759     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1760         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1761         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1762     }
1763 
1764     qdict_join(options, old_options, false);
1765 
1766     /*
1767      * If after merging all cache size options are set, an old total size is
1768      * overwritten. Do keep all options, however, if all three are new. The
1769      * resulting error message is what we want to happen.
1770      */
1771     has_all_cache_options =
1772         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1773         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1774         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1775 
1776     if (has_all_cache_options && !has_new_total_cache_size) {
1777         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1778     }
1779 }
1780 
1781 static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
1782                                               bool want_zero,
1783                                               int64_t offset, int64_t count,
1784                                               int64_t *pnum, int64_t *map,
1785                                               BlockDriverState **file)
1786 {
1787     BDRVQcow2State *s = bs->opaque;
1788     uint64_t cluster_offset;
1789     int index_in_cluster, ret;
1790     unsigned int bytes;
1791     int status = 0;
1792 
1793     bytes = MIN(INT_MAX, count);
1794     qemu_co_mutex_lock(&s->lock);
1795     ret = qcow2_get_cluster_offset(bs, offset, &bytes, &cluster_offset);
1796     qemu_co_mutex_unlock(&s->lock);
1797     if (ret < 0) {
1798         return ret;
1799     }
1800 
1801     *pnum = bytes;
1802 
1803     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1804         !s->crypto) {
1805         index_in_cluster = offset & (s->cluster_size - 1);
1806         *map = cluster_offset | index_in_cluster;
1807         *file = bs->file->bs;
1808         status |= BDRV_BLOCK_OFFSET_VALID;
1809     }
1810     if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) {
1811         status |= BDRV_BLOCK_ZERO;
1812     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1813         status |= BDRV_BLOCK_DATA;
1814     }
1815     return status;
1816 }
1817 
1818 static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs,
1819                                             QCowL2Meta **pl2meta,
1820                                             bool link_l2)
1821 {
1822     int ret = 0;
1823     QCowL2Meta *l2meta = *pl2meta;
1824 
1825     while (l2meta != NULL) {
1826         QCowL2Meta *next;
1827 
1828         if (link_l2) {
1829             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1830             if (ret) {
1831                 goto out;
1832             }
1833         } else {
1834             qcow2_alloc_cluster_abort(bs, l2meta);
1835         }
1836 
1837         /* Take the request off the list of running requests */
1838         if (l2meta->nb_clusters != 0) {
1839             QLIST_REMOVE(l2meta, next_in_flight);
1840         }
1841 
1842         qemu_co_queue_restart_all(&l2meta->dependent_requests);
1843 
1844         next = l2meta->next;
1845         g_free(l2meta);
1846         l2meta = next;
1847     }
1848 out:
1849     *pl2meta = l2meta;
1850     return ret;
1851 }
1852 
1853 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset,
1854                                         uint64_t bytes, QEMUIOVector *qiov,
1855                                         int flags)
1856 {
1857     BDRVQcow2State *s = bs->opaque;
1858     int offset_in_cluster;
1859     int ret;
1860     unsigned int cur_bytes; /* number of bytes in current iteration */
1861     uint64_t cluster_offset = 0;
1862     uint64_t bytes_done = 0;
1863     QEMUIOVector hd_qiov;
1864     uint8_t *cluster_data = NULL;
1865 
1866     qemu_iovec_init(&hd_qiov, qiov->niov);
1867 
1868     qemu_co_mutex_lock(&s->lock);
1869 
1870     while (bytes != 0) {
1871 
1872         /* prepare next request */
1873         cur_bytes = MIN(bytes, INT_MAX);
1874         if (s->crypto) {
1875             cur_bytes = MIN(cur_bytes,
1876                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1877         }
1878 
1879         ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset);
1880         if (ret < 0) {
1881             goto fail;
1882         }
1883 
1884         offset_in_cluster = offset_into_cluster(s, offset);
1885 
1886         qemu_iovec_reset(&hd_qiov);
1887         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
1888 
1889         switch (ret) {
1890         case QCOW2_CLUSTER_UNALLOCATED:
1891 
1892             if (bs->backing) {
1893                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1894                 qemu_co_mutex_unlock(&s->lock);
1895                 ret = bdrv_co_preadv(bs->backing, offset, cur_bytes,
1896                                      &hd_qiov, 0);
1897                 qemu_co_mutex_lock(&s->lock);
1898                 if (ret < 0) {
1899                     goto fail;
1900                 }
1901             } else {
1902                 /* Note: in this case, no need to wait */
1903                 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1904             }
1905             break;
1906 
1907         case QCOW2_CLUSTER_ZERO_PLAIN:
1908         case QCOW2_CLUSTER_ZERO_ALLOC:
1909             qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1910             break;
1911 
1912         case QCOW2_CLUSTER_COMPRESSED:
1913             /* add AIO support for compressed blocks ? */
1914             ret = qcow2_decompress_cluster(bs, cluster_offset);
1915             if (ret < 0) {
1916                 goto fail;
1917             }
1918 
1919             qemu_iovec_from_buf(&hd_qiov, 0,
1920                                 s->cluster_cache + offset_in_cluster,
1921                                 cur_bytes);
1922             break;
1923 
1924         case QCOW2_CLUSTER_NORMAL:
1925             if ((cluster_offset & 511) != 0) {
1926                 ret = -EIO;
1927                 goto fail;
1928             }
1929 
1930             if (bs->encrypted) {
1931                 assert(s->crypto);
1932 
1933                 /*
1934                  * For encrypted images, read everything into a temporary
1935                  * contiguous buffer on which the AES functions can work.
1936                  */
1937                 if (!cluster_data) {
1938                     cluster_data =
1939                         qemu_try_blockalign(bs->file->bs,
1940                                             QCOW_MAX_CRYPT_CLUSTERS
1941                                             * s->cluster_size);
1942                     if (cluster_data == NULL) {
1943                         ret = -ENOMEM;
1944                         goto fail;
1945                     }
1946                 }
1947 
1948                 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1949                 qemu_iovec_reset(&hd_qiov);
1950                 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
1951             }
1952 
1953             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1954             qemu_co_mutex_unlock(&s->lock);
1955             ret = bdrv_co_preadv(bs->file,
1956                                  cluster_offset + offset_in_cluster,
1957                                  cur_bytes, &hd_qiov, 0);
1958             qemu_co_mutex_lock(&s->lock);
1959             if (ret < 0) {
1960                 goto fail;
1961             }
1962             if (bs->encrypted) {
1963                 assert(s->crypto);
1964                 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
1965                 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
1966                 if (qcrypto_block_decrypt(s->crypto,
1967                                           (s->crypt_physical_offset ?
1968                                            cluster_offset + offset_in_cluster :
1969                                            offset),
1970                                           cluster_data,
1971                                           cur_bytes,
1972                                           NULL) < 0) {
1973                     ret = -EIO;
1974                     goto fail;
1975                 }
1976                 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes);
1977             }
1978             break;
1979 
1980         default:
1981             g_assert_not_reached();
1982             ret = -EIO;
1983             goto fail;
1984         }
1985 
1986         bytes -= cur_bytes;
1987         offset += cur_bytes;
1988         bytes_done += cur_bytes;
1989     }
1990     ret = 0;
1991 
1992 fail:
1993     qemu_co_mutex_unlock(&s->lock);
1994 
1995     qemu_iovec_destroy(&hd_qiov);
1996     qemu_vfree(cluster_data);
1997 
1998     return ret;
1999 }
2000 
2001 /* Check if it's possible to merge a write request with the writing of
2002  * the data from the COW regions */
2003 static bool merge_cow(uint64_t offset, unsigned bytes,
2004                       QEMUIOVector *hd_qiov, QCowL2Meta *l2meta)
2005 {
2006     QCowL2Meta *m;
2007 
2008     for (m = l2meta; m != NULL; m = m->next) {
2009         /* If both COW regions are empty then there's nothing to merge */
2010         if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
2011             continue;
2012         }
2013 
2014         /* The data (middle) region must be immediately after the
2015          * start region */
2016         if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
2017             continue;
2018         }
2019 
2020         /* The end region must be immediately after the data (middle)
2021          * region */
2022         if (m->offset + m->cow_end.offset != offset + bytes) {
2023             continue;
2024         }
2025 
2026         /* Make sure that adding both COW regions to the QEMUIOVector
2027          * does not exceed IOV_MAX */
2028         if (hd_qiov->niov > IOV_MAX - 2) {
2029             continue;
2030         }
2031 
2032         m->data_qiov = hd_qiov;
2033         return true;
2034     }
2035 
2036     return false;
2037 }
2038 
2039 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset,
2040                                          uint64_t bytes, QEMUIOVector *qiov,
2041                                          int flags)
2042 {
2043     BDRVQcow2State *s = bs->opaque;
2044     int offset_in_cluster;
2045     int ret;
2046     unsigned int cur_bytes; /* number of sectors in current iteration */
2047     uint64_t cluster_offset;
2048     QEMUIOVector hd_qiov;
2049     uint64_t bytes_done = 0;
2050     uint8_t *cluster_data = NULL;
2051     QCowL2Meta *l2meta = NULL;
2052 
2053     trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
2054 
2055     qemu_iovec_init(&hd_qiov, qiov->niov);
2056 
2057     s->cluster_cache_offset = -1; /* disable compressed cache */
2058 
2059     qemu_co_mutex_lock(&s->lock);
2060 
2061     while (bytes != 0) {
2062 
2063         l2meta = NULL;
2064 
2065         trace_qcow2_writev_start_part(qemu_coroutine_self());
2066         offset_in_cluster = offset_into_cluster(s, offset);
2067         cur_bytes = MIN(bytes, INT_MAX);
2068         if (bs->encrypted) {
2069             cur_bytes = MIN(cur_bytes,
2070                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
2071                             - offset_in_cluster);
2072         }
2073 
2074         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2075                                          &cluster_offset, &l2meta);
2076         if (ret < 0) {
2077             goto fail;
2078         }
2079 
2080         assert((cluster_offset & 511) == 0);
2081 
2082         qemu_iovec_reset(&hd_qiov);
2083         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
2084 
2085         if (bs->encrypted) {
2086             assert(s->crypto);
2087             if (!cluster_data) {
2088                 cluster_data = qemu_try_blockalign(bs->file->bs,
2089                                                    QCOW_MAX_CRYPT_CLUSTERS
2090                                                    * s->cluster_size);
2091                 if (cluster_data == NULL) {
2092                     ret = -ENOMEM;
2093                     goto fail;
2094                 }
2095             }
2096 
2097             assert(hd_qiov.size <=
2098                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2099             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
2100 
2101             if (qcrypto_block_encrypt(s->crypto,
2102                                       (s->crypt_physical_offset ?
2103                                        cluster_offset + offset_in_cluster :
2104                                        offset),
2105                                       cluster_data,
2106                                       cur_bytes, NULL) < 0) {
2107                 ret = -EIO;
2108                 goto fail;
2109             }
2110 
2111             qemu_iovec_reset(&hd_qiov);
2112             qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
2113         }
2114 
2115         ret = qcow2_pre_write_overlap_check(bs, 0,
2116                 cluster_offset + offset_in_cluster, cur_bytes);
2117         if (ret < 0) {
2118             goto fail;
2119         }
2120 
2121         /* If we need to do COW, check if it's possible to merge the
2122          * writing of the guest data together with that of the COW regions.
2123          * If it's not possible (or not necessary) then write the
2124          * guest data now. */
2125         if (!merge_cow(offset, cur_bytes, &hd_qiov, l2meta)) {
2126             qemu_co_mutex_unlock(&s->lock);
2127             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
2128             trace_qcow2_writev_data(qemu_coroutine_self(),
2129                                     cluster_offset + offset_in_cluster);
2130             ret = bdrv_co_pwritev(bs->file,
2131                                   cluster_offset + offset_in_cluster,
2132                                   cur_bytes, &hd_qiov, 0);
2133             qemu_co_mutex_lock(&s->lock);
2134             if (ret < 0) {
2135                 goto fail;
2136             }
2137         }
2138 
2139         ret = qcow2_handle_l2meta(bs, &l2meta, true);
2140         if (ret) {
2141             goto fail;
2142         }
2143 
2144         bytes -= cur_bytes;
2145         offset += cur_bytes;
2146         bytes_done += cur_bytes;
2147         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
2148     }
2149     ret = 0;
2150 
2151 fail:
2152     qcow2_handle_l2meta(bs, &l2meta, false);
2153 
2154     qemu_co_mutex_unlock(&s->lock);
2155 
2156     qemu_iovec_destroy(&hd_qiov);
2157     qemu_vfree(cluster_data);
2158     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
2159 
2160     return ret;
2161 }
2162 
2163 static int qcow2_inactivate(BlockDriverState *bs)
2164 {
2165     BDRVQcow2State *s = bs->opaque;
2166     int ret, result = 0;
2167     Error *local_err = NULL;
2168 
2169     qcow2_store_persistent_dirty_bitmaps(bs, &local_err);
2170     if (local_err != NULL) {
2171         result = -EINVAL;
2172         error_reportf_err(local_err, "Lost persistent bitmaps during "
2173                           "inactivation of node '%s': ",
2174                           bdrv_get_device_or_node_name(bs));
2175     }
2176 
2177     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2178     if (ret) {
2179         result = ret;
2180         error_report("Failed to flush the L2 table cache: %s",
2181                      strerror(-ret));
2182     }
2183 
2184     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2185     if (ret) {
2186         result = ret;
2187         error_report("Failed to flush the refcount block cache: %s",
2188                      strerror(-ret));
2189     }
2190 
2191     if (result == 0) {
2192         qcow2_mark_clean(bs);
2193     }
2194 
2195     return result;
2196 }
2197 
2198 static void qcow2_close(BlockDriverState *bs)
2199 {
2200     BDRVQcow2State *s = bs->opaque;
2201     qemu_vfree(s->l1_table);
2202     /* else pre-write overlap checks in cache_destroy may crash */
2203     s->l1_table = NULL;
2204 
2205     if (!(s->flags & BDRV_O_INACTIVE)) {
2206         qcow2_inactivate(bs);
2207     }
2208 
2209     cache_clean_timer_del(bs);
2210     qcow2_cache_destroy(s->l2_table_cache);
2211     qcow2_cache_destroy(s->refcount_block_cache);
2212 
2213     qcrypto_block_free(s->crypto);
2214     s->crypto = NULL;
2215 
2216     g_free(s->unknown_header_fields);
2217     cleanup_unknown_header_ext(bs);
2218 
2219     g_free(s->image_backing_file);
2220     g_free(s->image_backing_format);
2221 
2222     g_free(s->cluster_cache);
2223     qemu_vfree(s->cluster_data);
2224     qcow2_refcount_close(bs);
2225     qcow2_free_snapshots(bs);
2226 }
2227 
2228 static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
2229                                                    Error **errp)
2230 {
2231     BDRVQcow2State *s = bs->opaque;
2232     int flags = s->flags;
2233     QCryptoBlock *crypto = NULL;
2234     QDict *options;
2235     Error *local_err = NULL;
2236     int ret;
2237 
2238     /*
2239      * Backing files are read-only which makes all of their metadata immutable,
2240      * that means we don't have to worry about reopening them here.
2241      */
2242 
2243     crypto = s->crypto;
2244     s->crypto = NULL;
2245 
2246     qcow2_close(bs);
2247 
2248     memset(s, 0, sizeof(BDRVQcow2State));
2249     options = qdict_clone_shallow(bs->options);
2250 
2251     flags &= ~BDRV_O_INACTIVE;
2252     qemu_co_mutex_lock(&s->lock);
2253     ret = qcow2_do_open(bs, options, flags, &local_err);
2254     qemu_co_mutex_unlock(&s->lock);
2255     qobject_unref(options);
2256     if (local_err) {
2257         error_propagate_prepend(errp, local_err,
2258                                 "Could not reopen qcow2 layer: ");
2259         bs->drv = NULL;
2260         return;
2261     } else if (ret < 0) {
2262         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
2263         bs->drv = NULL;
2264         return;
2265     }
2266 
2267     s->crypto = crypto;
2268 }
2269 
2270 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2271     size_t len, size_t buflen)
2272 {
2273     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2274     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
2275 
2276     if (buflen < ext_len) {
2277         return -ENOSPC;
2278     }
2279 
2280     *ext_backing_fmt = (QCowExtension) {
2281         .magic  = cpu_to_be32(magic),
2282         .len    = cpu_to_be32(len),
2283     };
2284 
2285     if (len) {
2286         memcpy(buf + sizeof(QCowExtension), s, len);
2287     }
2288 
2289     return ext_len;
2290 }
2291 
2292 /*
2293  * Updates the qcow2 header, including the variable length parts of it, i.e.
2294  * the backing file name and all extensions. qcow2 was not designed to allow
2295  * such changes, so if we run out of space (we can only use the first cluster)
2296  * this function may fail.
2297  *
2298  * Returns 0 on success, -errno in error cases.
2299  */
2300 int qcow2_update_header(BlockDriverState *bs)
2301 {
2302     BDRVQcow2State *s = bs->opaque;
2303     QCowHeader *header;
2304     char *buf;
2305     size_t buflen = s->cluster_size;
2306     int ret;
2307     uint64_t total_size;
2308     uint32_t refcount_table_clusters;
2309     size_t header_length;
2310     Qcow2UnknownHeaderExtension *uext;
2311 
2312     buf = qemu_blockalign(bs, buflen);
2313 
2314     /* Header structure */
2315     header = (QCowHeader*) buf;
2316 
2317     if (buflen < sizeof(*header)) {
2318         ret = -ENOSPC;
2319         goto fail;
2320     }
2321 
2322     header_length = sizeof(*header) + s->unknown_header_fields_size;
2323     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2324     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2325 
2326     *header = (QCowHeader) {
2327         /* Version 2 fields */
2328         .magic                  = cpu_to_be32(QCOW_MAGIC),
2329         .version                = cpu_to_be32(s->qcow_version),
2330         .backing_file_offset    = 0,
2331         .backing_file_size      = 0,
2332         .cluster_bits           = cpu_to_be32(s->cluster_bits),
2333         .size                   = cpu_to_be64(total_size),
2334         .crypt_method           = cpu_to_be32(s->crypt_method_header),
2335         .l1_size                = cpu_to_be32(s->l1_size),
2336         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
2337         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
2338         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2339         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
2340         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
2341 
2342         /* Version 3 fields */
2343         .incompatible_features  = cpu_to_be64(s->incompatible_features),
2344         .compatible_features    = cpu_to_be64(s->compatible_features),
2345         .autoclear_features     = cpu_to_be64(s->autoclear_features),
2346         .refcount_order         = cpu_to_be32(s->refcount_order),
2347         .header_length          = cpu_to_be32(header_length),
2348     };
2349 
2350     /* For older versions, write a shorter header */
2351     switch (s->qcow_version) {
2352     case 2:
2353         ret = offsetof(QCowHeader, incompatible_features);
2354         break;
2355     case 3:
2356         ret = sizeof(*header);
2357         break;
2358     default:
2359         ret = -EINVAL;
2360         goto fail;
2361     }
2362 
2363     buf += ret;
2364     buflen -= ret;
2365     memset(buf, 0, buflen);
2366 
2367     /* Preserve any unknown field in the header */
2368     if (s->unknown_header_fields_size) {
2369         if (buflen < s->unknown_header_fields_size) {
2370             ret = -ENOSPC;
2371             goto fail;
2372         }
2373 
2374         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2375         buf += s->unknown_header_fields_size;
2376         buflen -= s->unknown_header_fields_size;
2377     }
2378 
2379     /* Backing file format header extension */
2380     if (s->image_backing_format) {
2381         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
2382                              s->image_backing_format,
2383                              strlen(s->image_backing_format),
2384                              buflen);
2385         if (ret < 0) {
2386             goto fail;
2387         }
2388 
2389         buf += ret;
2390         buflen -= ret;
2391     }
2392 
2393     /* Full disk encryption header pointer extension */
2394     if (s->crypto_header.offset != 0) {
2395         cpu_to_be64s(&s->crypto_header.offset);
2396         cpu_to_be64s(&s->crypto_header.length);
2397         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
2398                              &s->crypto_header, sizeof(s->crypto_header),
2399                              buflen);
2400         be64_to_cpus(&s->crypto_header.offset);
2401         be64_to_cpus(&s->crypto_header.length);
2402         if (ret < 0) {
2403             goto fail;
2404         }
2405         buf += ret;
2406         buflen -= ret;
2407     }
2408 
2409     /* Feature table */
2410     if (s->qcow_version >= 3) {
2411         Qcow2Feature features[] = {
2412             {
2413                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2414                 .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
2415                 .name = "dirty bit",
2416             },
2417             {
2418                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2419                 .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
2420                 .name = "corrupt bit",
2421             },
2422             {
2423                 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
2424                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
2425                 .name = "lazy refcounts",
2426             },
2427         };
2428 
2429         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
2430                              features, sizeof(features), buflen);
2431         if (ret < 0) {
2432             goto fail;
2433         }
2434         buf += ret;
2435         buflen -= ret;
2436     }
2437 
2438     /* Bitmap extension */
2439     if (s->nb_bitmaps > 0) {
2440         Qcow2BitmapHeaderExt bitmaps_header = {
2441             .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
2442             .bitmap_directory_size =
2443                     cpu_to_be64(s->bitmap_directory_size),
2444             .bitmap_directory_offset =
2445                     cpu_to_be64(s->bitmap_directory_offset)
2446         };
2447         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
2448                              &bitmaps_header, sizeof(bitmaps_header),
2449                              buflen);
2450         if (ret < 0) {
2451             goto fail;
2452         }
2453         buf += ret;
2454         buflen -= ret;
2455     }
2456 
2457     /* Keep unknown header extensions */
2458     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
2459         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
2460         if (ret < 0) {
2461             goto fail;
2462         }
2463 
2464         buf += ret;
2465         buflen -= ret;
2466     }
2467 
2468     /* End of header extensions */
2469     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
2470     if (ret < 0) {
2471         goto fail;
2472     }
2473 
2474     buf += ret;
2475     buflen -= ret;
2476 
2477     /* Backing file name */
2478     if (s->image_backing_file) {
2479         size_t backing_file_len = strlen(s->image_backing_file);
2480 
2481         if (buflen < backing_file_len) {
2482             ret = -ENOSPC;
2483             goto fail;
2484         }
2485 
2486         /* Using strncpy is ok here, since buf is not NUL-terminated. */
2487         strncpy(buf, s->image_backing_file, buflen);
2488 
2489         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
2490         header->backing_file_size   = cpu_to_be32(backing_file_len);
2491     }
2492 
2493     /* Write the new header */
2494     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
2495     if (ret < 0) {
2496         goto fail;
2497     }
2498 
2499     ret = 0;
2500 fail:
2501     qemu_vfree(header);
2502     return ret;
2503 }
2504 
2505 static int qcow2_change_backing_file(BlockDriverState *bs,
2506     const char *backing_file, const char *backing_fmt)
2507 {
2508     BDRVQcow2State *s = bs->opaque;
2509 
2510     if (backing_file && strlen(backing_file) > 1023) {
2511         return -EINVAL;
2512     }
2513 
2514     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2515     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2516 
2517     g_free(s->image_backing_file);
2518     g_free(s->image_backing_format);
2519 
2520     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
2521     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
2522 
2523     return qcow2_update_header(bs);
2524 }
2525 
2526 static int qcow2_crypt_method_from_format(const char *encryptfmt)
2527 {
2528     if (g_str_equal(encryptfmt, "luks")) {
2529         return QCOW_CRYPT_LUKS;
2530     } else if (g_str_equal(encryptfmt, "aes")) {
2531         return QCOW_CRYPT_AES;
2532     } else {
2533         return -EINVAL;
2534     }
2535 }
2536 
2537 static int qcow2_set_up_encryption(BlockDriverState *bs,
2538                                    QCryptoBlockCreateOptions *cryptoopts,
2539                                    Error **errp)
2540 {
2541     BDRVQcow2State *s = bs->opaque;
2542     QCryptoBlock *crypto = NULL;
2543     int fmt, ret;
2544 
2545     switch (cryptoopts->format) {
2546     case Q_CRYPTO_BLOCK_FORMAT_LUKS:
2547         fmt = QCOW_CRYPT_LUKS;
2548         break;
2549     case Q_CRYPTO_BLOCK_FORMAT_QCOW:
2550         fmt = QCOW_CRYPT_AES;
2551         break;
2552     default:
2553         error_setg(errp, "Crypto format not supported in qcow2");
2554         return -EINVAL;
2555     }
2556 
2557     s->crypt_method_header = fmt;
2558 
2559     crypto = qcrypto_block_create(cryptoopts, "encrypt.",
2560                                   qcow2_crypto_hdr_init_func,
2561                                   qcow2_crypto_hdr_write_func,
2562                                   bs, errp);
2563     if (!crypto) {
2564         return -EINVAL;
2565     }
2566 
2567     ret = qcow2_update_header(bs);
2568     if (ret < 0) {
2569         error_setg_errno(errp, -ret, "Could not write encryption header");
2570         goto out;
2571     }
2572 
2573     ret = 0;
2574  out:
2575     qcrypto_block_free(crypto);
2576     return ret;
2577 }
2578 
2579 /**
2580  * Preallocates metadata structures for data clusters between @offset (in the
2581  * guest disk) and @new_length (which is thus generally the new guest disk
2582  * size).
2583  *
2584  * Returns: 0 on success, -errno on failure.
2585  */
2586 static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset,
2587                                        uint64_t new_length)
2588 {
2589     uint64_t bytes;
2590     uint64_t host_offset = 0;
2591     unsigned int cur_bytes;
2592     int ret;
2593     QCowL2Meta *meta;
2594 
2595     assert(offset <= new_length);
2596     bytes = new_length - offset;
2597 
2598     while (bytes) {
2599         cur_bytes = MIN(bytes, INT_MAX);
2600         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2601                                          &host_offset, &meta);
2602         if (ret < 0) {
2603             return ret;
2604         }
2605 
2606         while (meta) {
2607             QCowL2Meta *next = meta->next;
2608 
2609             ret = qcow2_alloc_cluster_link_l2(bs, meta);
2610             if (ret < 0) {
2611                 qcow2_free_any_clusters(bs, meta->alloc_offset,
2612                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
2613                 return ret;
2614             }
2615 
2616             /* There are no dependent requests, but we need to remove our
2617              * request from the list of in-flight requests */
2618             QLIST_REMOVE(meta, next_in_flight);
2619 
2620             g_free(meta);
2621             meta = next;
2622         }
2623 
2624         /* TODO Preallocate data if requested */
2625 
2626         bytes -= cur_bytes;
2627         offset += cur_bytes;
2628     }
2629 
2630     /*
2631      * It is expected that the image file is large enough to actually contain
2632      * all of the allocated clusters (otherwise we get failing reads after
2633      * EOF). Extend the image to the last allocated sector.
2634      */
2635     if (host_offset != 0) {
2636         uint8_t data = 0;
2637         ret = bdrv_pwrite(bs->file, (host_offset + cur_bytes) - 1,
2638                           &data, 1);
2639         if (ret < 0) {
2640             return ret;
2641         }
2642     }
2643 
2644     return 0;
2645 }
2646 
2647 /* qcow2_refcount_metadata_size:
2648  * @clusters: number of clusters to refcount (including data and L1/L2 tables)
2649  * @cluster_size: size of a cluster, in bytes
2650  * @refcount_order: refcount bits power-of-2 exponent
2651  * @generous_increase: allow for the refcount table to be 1.5x as large as it
2652  *                     needs to be
2653  *
2654  * Returns: Number of bytes required for refcount blocks and table metadata.
2655  */
2656 int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
2657                                      int refcount_order, bool generous_increase,
2658                                      uint64_t *refblock_count)
2659 {
2660     /*
2661      * Every host cluster is reference-counted, including metadata (even
2662      * refcount metadata is recursively included).
2663      *
2664      * An accurate formula for the size of refcount metadata size is difficult
2665      * to derive.  An easier method of calculation is finding the fixed point
2666      * where no further refcount blocks or table clusters are required to
2667      * reference count every cluster.
2668      */
2669     int64_t blocks_per_table_cluster = cluster_size / sizeof(uint64_t);
2670     int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
2671     int64_t table = 0;  /* number of refcount table clusters */
2672     int64_t blocks = 0; /* number of refcount block clusters */
2673     int64_t last;
2674     int64_t n = 0;
2675 
2676     do {
2677         last = n;
2678         blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
2679         table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
2680         n = clusters + blocks + table;
2681 
2682         if (n == last && generous_increase) {
2683             clusters += DIV_ROUND_UP(table, 2);
2684             n = 0; /* force another loop */
2685             generous_increase = false;
2686         }
2687     } while (n != last);
2688 
2689     if (refblock_count) {
2690         *refblock_count = blocks;
2691     }
2692 
2693     return (blocks + table) * cluster_size;
2694 }
2695 
2696 /**
2697  * qcow2_calc_prealloc_size:
2698  * @total_size: virtual disk size in bytes
2699  * @cluster_size: cluster size in bytes
2700  * @refcount_order: refcount bits power-of-2 exponent
2701  *
2702  * Returns: Total number of bytes required for the fully allocated image
2703  * (including metadata).
2704  */
2705 static int64_t qcow2_calc_prealloc_size(int64_t total_size,
2706                                         size_t cluster_size,
2707                                         int refcount_order)
2708 {
2709     int64_t meta_size = 0;
2710     uint64_t nl1e, nl2e;
2711     int64_t aligned_total_size = ROUND_UP(total_size, cluster_size);
2712 
2713     /* header: 1 cluster */
2714     meta_size += cluster_size;
2715 
2716     /* total size of L2 tables */
2717     nl2e = aligned_total_size / cluster_size;
2718     nl2e = ROUND_UP(nl2e, cluster_size / sizeof(uint64_t));
2719     meta_size += nl2e * sizeof(uint64_t);
2720 
2721     /* total size of L1 tables */
2722     nl1e = nl2e * sizeof(uint64_t) / cluster_size;
2723     nl1e = ROUND_UP(nl1e, cluster_size / sizeof(uint64_t));
2724     meta_size += nl1e * sizeof(uint64_t);
2725 
2726     /* total size of refcount table and blocks */
2727     meta_size += qcow2_refcount_metadata_size(
2728             (meta_size + aligned_total_size) / cluster_size,
2729             cluster_size, refcount_order, false, NULL);
2730 
2731     return meta_size + aligned_total_size;
2732 }
2733 
2734 static bool validate_cluster_size(size_t cluster_size, Error **errp)
2735 {
2736     int cluster_bits = ctz32(cluster_size);
2737     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
2738         (1 << cluster_bits) != cluster_size)
2739     {
2740         error_setg(errp, "Cluster size must be a power of two between %d and "
2741                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
2742         return false;
2743     }
2744     return true;
2745 }
2746 
2747 static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, Error **errp)
2748 {
2749     size_t cluster_size;
2750 
2751     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2752                                          DEFAULT_CLUSTER_SIZE);
2753     if (!validate_cluster_size(cluster_size, errp)) {
2754         return 0;
2755     }
2756     return cluster_size;
2757 }
2758 
2759 static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
2760 {
2761     char *buf;
2762     int ret;
2763 
2764     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2765     if (!buf) {
2766         ret = 3; /* default */
2767     } else if (!strcmp(buf, "0.10")) {
2768         ret = 2;
2769     } else if (!strcmp(buf, "1.1")) {
2770         ret = 3;
2771     } else {
2772         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2773         ret = -EINVAL;
2774     }
2775     g_free(buf);
2776     return ret;
2777 }
2778 
2779 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
2780                                                 Error **errp)
2781 {
2782     uint64_t refcount_bits;
2783 
2784     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
2785     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2786         error_setg(errp, "Refcount width must be a power of two and may not "
2787                    "exceed 64 bits");
2788         return 0;
2789     }
2790 
2791     if (version < 3 && refcount_bits != 16) {
2792         error_setg(errp, "Different refcount widths than 16 bits require "
2793                    "compatibility level 1.1 or above (use compat=1.1 or "
2794                    "greater)");
2795         return 0;
2796     }
2797 
2798     return refcount_bits;
2799 }
2800 
2801 static int coroutine_fn
2802 qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
2803 {
2804     BlockdevCreateOptionsQcow2 *qcow2_opts;
2805     QDict *options;
2806 
2807     /*
2808      * Open the image file and write a minimal qcow2 header.
2809      *
2810      * We keep things simple and start with a zero-sized image. We also
2811      * do without refcount blocks or a L1 table for now. We'll fix the
2812      * inconsistency later.
2813      *
2814      * We do need a refcount table because growing the refcount table means
2815      * allocating two new refcount blocks - the seconds of which would be at
2816      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2817      * size for any qcow2 image.
2818      */
2819     BlockBackend *blk = NULL;
2820     BlockDriverState *bs = NULL;
2821     QCowHeader *header;
2822     size_t cluster_size;
2823     int version;
2824     int refcount_order;
2825     uint64_t* refcount_table;
2826     Error *local_err = NULL;
2827     int ret;
2828 
2829     assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2);
2830     qcow2_opts = &create_options->u.qcow2;
2831 
2832     bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp);
2833     if (bs == NULL) {
2834         return -EIO;
2835     }
2836 
2837     /* Validate options and set default values */
2838     if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) {
2839         error_setg(errp, "Image size must be a multiple of 512 bytes");
2840         ret = -EINVAL;
2841         goto out;
2842     }
2843 
2844     if (qcow2_opts->has_version) {
2845         switch (qcow2_opts->version) {
2846         case BLOCKDEV_QCOW2_VERSION_V2:
2847             version = 2;
2848             break;
2849         case BLOCKDEV_QCOW2_VERSION_V3:
2850             version = 3;
2851             break;
2852         default:
2853             g_assert_not_reached();
2854         }
2855     } else {
2856         version = 3;
2857     }
2858 
2859     if (qcow2_opts->has_cluster_size) {
2860         cluster_size = qcow2_opts->cluster_size;
2861     } else {
2862         cluster_size = DEFAULT_CLUSTER_SIZE;
2863     }
2864 
2865     if (!validate_cluster_size(cluster_size, errp)) {
2866         ret = -EINVAL;
2867         goto out;
2868     }
2869 
2870     if (!qcow2_opts->has_preallocation) {
2871         qcow2_opts->preallocation = PREALLOC_MODE_OFF;
2872     }
2873     if (qcow2_opts->has_backing_file &&
2874         qcow2_opts->preallocation != PREALLOC_MODE_OFF)
2875     {
2876         error_setg(errp, "Backing file and preallocation cannot be used at "
2877                    "the same time");
2878         ret = -EINVAL;
2879         goto out;
2880     }
2881     if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) {
2882         error_setg(errp, "Backing format cannot be used without backing file");
2883         ret = -EINVAL;
2884         goto out;
2885     }
2886 
2887     if (!qcow2_opts->has_lazy_refcounts) {
2888         qcow2_opts->lazy_refcounts = false;
2889     }
2890     if (version < 3 && qcow2_opts->lazy_refcounts) {
2891         error_setg(errp, "Lazy refcounts only supported with compatibility "
2892                    "level 1.1 and above (use version=v3 or greater)");
2893         ret = -EINVAL;
2894         goto out;
2895     }
2896 
2897     if (!qcow2_opts->has_refcount_bits) {
2898         qcow2_opts->refcount_bits = 16;
2899     }
2900     if (qcow2_opts->refcount_bits > 64 ||
2901         !is_power_of_2(qcow2_opts->refcount_bits))
2902     {
2903         error_setg(errp, "Refcount width must be a power of two and may not "
2904                    "exceed 64 bits");
2905         ret = -EINVAL;
2906         goto out;
2907     }
2908     if (version < 3 && qcow2_opts->refcount_bits != 16) {
2909         error_setg(errp, "Different refcount widths than 16 bits require "
2910                    "compatibility level 1.1 or above (use version=v3 or "
2911                    "greater)");
2912         ret = -EINVAL;
2913         goto out;
2914     }
2915     refcount_order = ctz32(qcow2_opts->refcount_bits);
2916 
2917 
2918     /* Create BlockBackend to write to the image */
2919     blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
2920     ret = blk_insert_bs(blk, bs, errp);
2921     if (ret < 0) {
2922         goto out;
2923     }
2924     blk_set_allow_write_beyond_eof(blk, true);
2925 
2926     /* Clear the protocol layer and preallocate it if necessary */
2927     ret = blk_truncate(blk, 0, PREALLOC_MODE_OFF, errp);
2928     if (ret < 0) {
2929         goto out;
2930     }
2931 
2932     if (qcow2_opts->preallocation == PREALLOC_MODE_FULL ||
2933         qcow2_opts->preallocation == PREALLOC_MODE_FALLOC)
2934     {
2935         int64_t prealloc_size =
2936             qcow2_calc_prealloc_size(qcow2_opts->size, cluster_size,
2937                                      refcount_order);
2938 
2939         ret = blk_truncate(blk, prealloc_size, qcow2_opts->preallocation, errp);
2940         if (ret < 0) {
2941             goto out;
2942         }
2943     }
2944 
2945     /* Write the header */
2946     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
2947     header = g_malloc0(cluster_size);
2948     *header = (QCowHeader) {
2949         .magic                      = cpu_to_be32(QCOW_MAGIC),
2950         .version                    = cpu_to_be32(version),
2951         .cluster_bits               = cpu_to_be32(ctz32(cluster_size)),
2952         .size                       = cpu_to_be64(0),
2953         .l1_table_offset            = cpu_to_be64(0),
2954         .l1_size                    = cpu_to_be32(0),
2955         .refcount_table_offset      = cpu_to_be64(cluster_size),
2956         .refcount_table_clusters    = cpu_to_be32(1),
2957         .refcount_order             = cpu_to_be32(refcount_order),
2958         .header_length              = cpu_to_be32(sizeof(*header)),
2959     };
2960 
2961     /* We'll update this to correct value later */
2962     header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
2963 
2964     if (qcow2_opts->lazy_refcounts) {
2965         header->compatible_features |=
2966             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
2967     }
2968 
2969     ret = blk_pwrite(blk, 0, header, cluster_size, 0);
2970     g_free(header);
2971     if (ret < 0) {
2972         error_setg_errno(errp, -ret, "Could not write qcow2 header");
2973         goto out;
2974     }
2975 
2976     /* Write a refcount table with one refcount block */
2977     refcount_table = g_malloc0(2 * cluster_size);
2978     refcount_table[0] = cpu_to_be64(2 * cluster_size);
2979     ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
2980     g_free(refcount_table);
2981 
2982     if (ret < 0) {
2983         error_setg_errno(errp, -ret, "Could not write refcount table");
2984         goto out;
2985     }
2986 
2987     blk_unref(blk);
2988     blk = NULL;
2989 
2990     /*
2991      * And now open the image and make it consistent first (i.e. increase the
2992      * refcount of the cluster that is occupied by the header and the refcount
2993      * table)
2994      */
2995     options = qdict_new();
2996     qdict_put_str(options, "driver", "qcow2");
2997     qdict_put_str(options, "file", bs->node_name);
2998     blk = blk_new_open(NULL, NULL, options,
2999                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
3000                        &local_err);
3001     if (blk == NULL) {
3002         error_propagate(errp, local_err);
3003         ret = -EIO;
3004         goto out;
3005     }
3006 
3007     ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
3008     if (ret < 0) {
3009         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
3010                          "header and refcount table");
3011         goto out;
3012 
3013     } else if (ret != 0) {
3014         error_report("Huh, first cluster in empty image is already in use?");
3015         abort();
3016     }
3017 
3018     /* Create a full header (including things like feature table) */
3019     ret = qcow2_update_header(blk_bs(blk));
3020     if (ret < 0) {
3021         error_setg_errno(errp, -ret, "Could not update qcow2 header");
3022         goto out;
3023     }
3024 
3025     /* Okay, now that we have a valid image, let's give it the right size */
3026     ret = blk_truncate(blk, qcow2_opts->size, PREALLOC_MODE_OFF, errp);
3027     if (ret < 0) {
3028         error_prepend(errp, "Could not resize image: ");
3029         goto out;
3030     }
3031 
3032     /* Want a backing file? There you go.*/
3033     if (qcow2_opts->has_backing_file) {
3034         const char *backing_format = NULL;
3035 
3036         if (qcow2_opts->has_backing_fmt) {
3037             backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt);
3038         }
3039 
3040         ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file,
3041                                        backing_format);
3042         if (ret < 0) {
3043             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
3044                              "with format '%s'", qcow2_opts->backing_file,
3045                              backing_format);
3046             goto out;
3047         }
3048     }
3049 
3050     /* Want encryption? There you go. */
3051     if (qcow2_opts->has_encrypt) {
3052         ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp);
3053         if (ret < 0) {
3054             goto out;
3055         }
3056     }
3057 
3058     /* And if we're supposed to preallocate metadata, do that now */
3059     if (qcow2_opts->preallocation != PREALLOC_MODE_OFF) {
3060         BDRVQcow2State *s = blk_bs(blk)->opaque;
3061         qemu_co_mutex_lock(&s->lock);
3062         ret = preallocate_co(blk_bs(blk), 0, qcow2_opts->size);
3063         qemu_co_mutex_unlock(&s->lock);
3064 
3065         if (ret < 0) {
3066             error_setg_errno(errp, -ret, "Could not preallocate metadata");
3067             goto out;
3068         }
3069     }
3070 
3071     blk_unref(blk);
3072     blk = NULL;
3073 
3074     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3075      * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3076      * have to setup decryption context. We're not doing any I/O on the top
3077      * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3078      * not have effect.
3079      */
3080     options = qdict_new();
3081     qdict_put_str(options, "driver", "qcow2");
3082     qdict_put_str(options, "file", bs->node_name);
3083     blk = blk_new_open(NULL, NULL, options,
3084                        BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO,
3085                        &local_err);
3086     if (blk == NULL) {
3087         error_propagate(errp, local_err);
3088         ret = -EIO;
3089         goto out;
3090     }
3091 
3092     ret = 0;
3093 out:
3094     blk_unref(blk);
3095     bdrv_unref(bs);
3096     return ret;
3097 }
3098 
3099 static int coroutine_fn qcow2_co_create_opts(const char *filename, QemuOpts *opts,
3100                                              Error **errp)
3101 {
3102     BlockdevCreateOptions *create_options = NULL;
3103     QDict *qdict;
3104     Visitor *v;
3105     BlockDriverState *bs = NULL;
3106     Error *local_err = NULL;
3107     const char *val;
3108     int ret;
3109 
3110     /* Only the keyval visitor supports the dotted syntax needed for
3111      * encryption, so go through a QDict before getting a QAPI type. Ignore
3112      * options meant for the protocol layer so that the visitor doesn't
3113      * complain. */
3114     qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts,
3115                                         true);
3116 
3117     /* Handle encryption options */
3118     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
3119     if (val && !strcmp(val, "on")) {
3120         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
3121     } else if (val && !strcmp(val, "off")) {
3122         qdict_del(qdict, BLOCK_OPT_ENCRYPT);
3123     }
3124 
3125     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
3126     if (val && !strcmp(val, "aes")) {
3127         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
3128     }
3129 
3130     /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3131      * version=v2/v3 below. */
3132     val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL);
3133     if (val && !strcmp(val, "0.10")) {
3134         qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2");
3135     } else if (val && !strcmp(val, "1.1")) {
3136         qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3");
3137     }
3138 
3139     /* Change legacy command line options into QMP ones */
3140     static const QDictRenames opt_renames[] = {
3141         { BLOCK_OPT_BACKING_FILE,       "backing-file" },
3142         { BLOCK_OPT_BACKING_FMT,        "backing-fmt" },
3143         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
3144         { BLOCK_OPT_LAZY_REFCOUNTS,     "lazy-refcounts" },
3145         { BLOCK_OPT_REFCOUNT_BITS,      "refcount-bits" },
3146         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
3147         { BLOCK_OPT_COMPAT_LEVEL,       "version" },
3148         { NULL, NULL },
3149     };
3150 
3151     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
3152         ret = -EINVAL;
3153         goto finish;
3154     }
3155 
3156     /* Create and open the file (protocol layer) */
3157     ret = bdrv_create_file(filename, opts, errp);
3158     if (ret < 0) {
3159         goto finish;
3160     }
3161 
3162     bs = bdrv_open(filename, NULL, NULL,
3163                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
3164     if (bs == NULL) {
3165         ret = -EIO;
3166         goto finish;
3167     }
3168 
3169     /* Set 'driver' and 'node' options */
3170     qdict_put_str(qdict, "driver", "qcow2");
3171     qdict_put_str(qdict, "file", bs->node_name);
3172 
3173     /* Now get the QAPI type BlockdevCreateOptions */
3174     v = qobject_input_visitor_new_flat_confused(qdict, errp);
3175     if (!v) {
3176         ret = -EINVAL;
3177         goto finish;
3178     }
3179 
3180     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
3181     visit_free(v);
3182 
3183     if (local_err) {
3184         error_propagate(errp, local_err);
3185         ret = -EINVAL;
3186         goto finish;
3187     }
3188 
3189     /* Silently round up size */
3190     create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size,
3191                                             BDRV_SECTOR_SIZE);
3192 
3193     /* Create the qcow2 image (format layer) */
3194     ret = qcow2_co_create(create_options, errp);
3195     if (ret < 0) {
3196         goto finish;
3197     }
3198 
3199     ret = 0;
3200 finish:
3201     qobject_unref(qdict);
3202     bdrv_unref(bs);
3203     qapi_free_BlockdevCreateOptions(create_options);
3204     return ret;
3205 }
3206 
3207 
3208 static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes)
3209 {
3210     int64_t nr;
3211     int res;
3212 
3213     /* Clamp to image length, before checking status of underlying sectors */
3214     if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
3215         bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset;
3216     }
3217 
3218     if (!bytes) {
3219         return true;
3220     }
3221     res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL);
3222     return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == bytes;
3223 }
3224 
3225 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
3226     int64_t offset, int bytes, BdrvRequestFlags flags)
3227 {
3228     int ret;
3229     BDRVQcow2State *s = bs->opaque;
3230 
3231     uint32_t head = offset % s->cluster_size;
3232     uint32_t tail = (offset + bytes) % s->cluster_size;
3233 
3234     trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes);
3235     if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) {
3236         tail = 0;
3237     }
3238 
3239     if (head || tail) {
3240         uint64_t off;
3241         unsigned int nr;
3242 
3243         assert(head + bytes <= s->cluster_size);
3244 
3245         /* check whether remainder of cluster already reads as zero */
3246         if (!(is_zero(bs, offset - head, head) &&
3247               is_zero(bs, offset + bytes,
3248                       tail ? s->cluster_size - tail : 0))) {
3249             return -ENOTSUP;
3250         }
3251 
3252         qemu_co_mutex_lock(&s->lock);
3253         /* We can have new write after previous check */
3254         offset = QEMU_ALIGN_DOWN(offset, s->cluster_size);
3255         bytes = s->cluster_size;
3256         nr = s->cluster_size;
3257         ret = qcow2_get_cluster_offset(bs, offset, &nr, &off);
3258         if (ret != QCOW2_CLUSTER_UNALLOCATED &&
3259             ret != QCOW2_CLUSTER_ZERO_PLAIN &&
3260             ret != QCOW2_CLUSTER_ZERO_ALLOC) {
3261             qemu_co_mutex_unlock(&s->lock);
3262             return -ENOTSUP;
3263         }
3264     } else {
3265         qemu_co_mutex_lock(&s->lock);
3266     }
3267 
3268     trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes);
3269 
3270     /* Whatever is left can use real zero clusters */
3271     ret = qcow2_cluster_zeroize(bs, offset, bytes, flags);
3272     qemu_co_mutex_unlock(&s->lock);
3273 
3274     return ret;
3275 }
3276 
3277 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
3278                                           int64_t offset, int bytes)
3279 {
3280     int ret;
3281     BDRVQcow2State *s = bs->opaque;
3282 
3283     if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
3284         assert(bytes < s->cluster_size);
3285         /* Ignore partial clusters, except for the special case of the
3286          * complete partial cluster at the end of an unaligned file */
3287         if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
3288             offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) {
3289             return -ENOTSUP;
3290         }
3291     }
3292 
3293     qemu_co_mutex_lock(&s->lock);
3294     ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST,
3295                                 false);
3296     qemu_co_mutex_unlock(&s->lock);
3297     return ret;
3298 }
3299 
3300 static int coroutine_fn
3301 qcow2_co_copy_range_from(BlockDriverState *bs,
3302                          BdrvChild *src, uint64_t src_offset,
3303                          BdrvChild *dst, uint64_t dst_offset,
3304                          uint64_t bytes, BdrvRequestFlags read_flags,
3305                          BdrvRequestFlags write_flags)
3306 {
3307     BDRVQcow2State *s = bs->opaque;
3308     int ret;
3309     unsigned int cur_bytes; /* number of bytes in current iteration */
3310     BdrvChild *child = NULL;
3311     BdrvRequestFlags cur_write_flags;
3312 
3313     assert(!bs->encrypted);
3314     qemu_co_mutex_lock(&s->lock);
3315 
3316     while (bytes != 0) {
3317         uint64_t copy_offset = 0;
3318         /* prepare next request */
3319         cur_bytes = MIN(bytes, INT_MAX);
3320         cur_write_flags = write_flags;
3321 
3322         ret = qcow2_get_cluster_offset(bs, src_offset, &cur_bytes, &copy_offset);
3323         if (ret < 0) {
3324             goto out;
3325         }
3326 
3327         switch (ret) {
3328         case QCOW2_CLUSTER_UNALLOCATED:
3329             if (bs->backing && bs->backing->bs) {
3330                 int64_t backing_length = bdrv_getlength(bs->backing->bs);
3331                 if (src_offset >= backing_length) {
3332                     cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3333                 } else {
3334                     child = bs->backing;
3335                     cur_bytes = MIN(cur_bytes, backing_length - src_offset);
3336                     copy_offset = src_offset;
3337                 }
3338             } else {
3339                 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3340             }
3341             break;
3342 
3343         case QCOW2_CLUSTER_ZERO_PLAIN:
3344         case QCOW2_CLUSTER_ZERO_ALLOC:
3345             cur_write_flags |= BDRV_REQ_ZERO_WRITE;
3346             break;
3347 
3348         case QCOW2_CLUSTER_COMPRESSED:
3349             ret = -ENOTSUP;
3350             goto out;
3351 
3352         case QCOW2_CLUSTER_NORMAL:
3353             child = bs->file;
3354             copy_offset += offset_into_cluster(s, src_offset);
3355             if ((copy_offset & 511) != 0) {
3356                 ret = -EIO;
3357                 goto out;
3358             }
3359             break;
3360 
3361         default:
3362             abort();
3363         }
3364         qemu_co_mutex_unlock(&s->lock);
3365         ret = bdrv_co_copy_range_from(child,
3366                                       copy_offset,
3367                                       dst, dst_offset,
3368                                       cur_bytes, read_flags, cur_write_flags);
3369         qemu_co_mutex_lock(&s->lock);
3370         if (ret < 0) {
3371             goto out;
3372         }
3373 
3374         bytes -= cur_bytes;
3375         src_offset += cur_bytes;
3376         dst_offset += cur_bytes;
3377     }
3378     ret = 0;
3379 
3380 out:
3381     qemu_co_mutex_unlock(&s->lock);
3382     return ret;
3383 }
3384 
3385 static int coroutine_fn
3386 qcow2_co_copy_range_to(BlockDriverState *bs,
3387                        BdrvChild *src, uint64_t src_offset,
3388                        BdrvChild *dst, uint64_t dst_offset,
3389                        uint64_t bytes, BdrvRequestFlags read_flags,
3390                        BdrvRequestFlags write_flags)
3391 {
3392     BDRVQcow2State *s = bs->opaque;
3393     int offset_in_cluster;
3394     int ret;
3395     unsigned int cur_bytes; /* number of sectors in current iteration */
3396     uint64_t cluster_offset;
3397     QCowL2Meta *l2meta = NULL;
3398 
3399     assert(!bs->encrypted);
3400     s->cluster_cache_offset = -1; /* disable compressed cache */
3401 
3402     qemu_co_mutex_lock(&s->lock);
3403 
3404     while (bytes != 0) {
3405 
3406         l2meta = NULL;
3407 
3408         offset_in_cluster = offset_into_cluster(s, dst_offset);
3409         cur_bytes = MIN(bytes, INT_MAX);
3410 
3411         /* TODO:
3412          * If src->bs == dst->bs, we could simply copy by incrementing
3413          * the refcnt, without copying user data.
3414          * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
3415         ret = qcow2_alloc_cluster_offset(bs, dst_offset, &cur_bytes,
3416                                          &cluster_offset, &l2meta);
3417         if (ret < 0) {
3418             goto fail;
3419         }
3420 
3421         assert((cluster_offset & 511) == 0);
3422 
3423         ret = qcow2_pre_write_overlap_check(bs, 0,
3424                 cluster_offset + offset_in_cluster, cur_bytes);
3425         if (ret < 0) {
3426             goto fail;
3427         }
3428 
3429         qemu_co_mutex_unlock(&s->lock);
3430         ret = bdrv_co_copy_range_to(src, src_offset,
3431                                     bs->file,
3432                                     cluster_offset + offset_in_cluster,
3433                                     cur_bytes, read_flags, write_flags);
3434         qemu_co_mutex_lock(&s->lock);
3435         if (ret < 0) {
3436             goto fail;
3437         }
3438 
3439         ret = qcow2_handle_l2meta(bs, &l2meta, true);
3440         if (ret) {
3441             goto fail;
3442         }
3443 
3444         bytes -= cur_bytes;
3445         src_offset += cur_bytes;
3446         dst_offset += cur_bytes;
3447     }
3448     ret = 0;
3449 
3450 fail:
3451     qcow2_handle_l2meta(bs, &l2meta, false);
3452 
3453     qemu_co_mutex_unlock(&s->lock);
3454 
3455     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
3456 
3457     return ret;
3458 }
3459 
3460 static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,
3461                                           PreallocMode prealloc, Error **errp)
3462 {
3463     BDRVQcow2State *s = bs->opaque;
3464     uint64_t old_length;
3465     int64_t new_l1_size;
3466     int ret;
3467     QDict *options;
3468 
3469     if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA &&
3470         prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL)
3471     {
3472         error_setg(errp, "Unsupported preallocation mode '%s'",
3473                    PreallocMode_str(prealloc));
3474         return -ENOTSUP;
3475     }
3476 
3477     if (offset & 511) {
3478         error_setg(errp, "The new size must be a multiple of 512");
3479         return -EINVAL;
3480     }
3481 
3482     qemu_co_mutex_lock(&s->lock);
3483 
3484     /* cannot proceed if image has snapshots */
3485     if (s->nb_snapshots) {
3486         error_setg(errp, "Can't resize an image which has snapshots");
3487         ret = -ENOTSUP;
3488         goto fail;
3489     }
3490 
3491     /* cannot proceed if image has bitmaps */
3492     if (s->nb_bitmaps) {
3493         /* TODO: resize bitmaps in the image */
3494         error_setg(errp, "Can't resize an image which has bitmaps");
3495         ret = -ENOTSUP;
3496         goto fail;
3497     }
3498 
3499     old_length = bs->total_sectors * BDRV_SECTOR_SIZE;
3500     new_l1_size = size_to_l1(s, offset);
3501 
3502     if (offset < old_length) {
3503         int64_t last_cluster, old_file_size;
3504         if (prealloc != PREALLOC_MODE_OFF) {
3505             error_setg(errp,
3506                        "Preallocation can't be used for shrinking an image");
3507             ret = -EINVAL;
3508             goto fail;
3509         }
3510 
3511         ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size),
3512                                     old_length - ROUND_UP(offset,
3513                                                           s->cluster_size),
3514                                     QCOW2_DISCARD_ALWAYS, true);
3515         if (ret < 0) {
3516             error_setg_errno(errp, -ret, "Failed to discard cropped clusters");
3517             goto fail;
3518         }
3519 
3520         ret = qcow2_shrink_l1_table(bs, new_l1_size);
3521         if (ret < 0) {
3522             error_setg_errno(errp, -ret,
3523                              "Failed to reduce the number of L2 tables");
3524             goto fail;
3525         }
3526 
3527         ret = qcow2_shrink_reftable(bs);
3528         if (ret < 0) {
3529             error_setg_errno(errp, -ret,
3530                              "Failed to discard unused refblocks");
3531             goto fail;
3532         }
3533 
3534         old_file_size = bdrv_getlength(bs->file->bs);
3535         if (old_file_size < 0) {
3536             error_setg_errno(errp, -old_file_size,
3537                              "Failed to inquire current file length");
3538             ret = old_file_size;
3539             goto fail;
3540         }
3541         last_cluster = qcow2_get_last_cluster(bs, old_file_size);
3542         if (last_cluster < 0) {
3543             error_setg_errno(errp, -last_cluster,
3544                              "Failed to find the last cluster");
3545             ret = last_cluster;
3546             goto fail;
3547         }
3548         if ((last_cluster + 1) * s->cluster_size < old_file_size) {
3549             Error *local_err = NULL;
3550 
3551             bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size,
3552                              PREALLOC_MODE_OFF, &local_err);
3553             if (local_err) {
3554                 warn_reportf_err(local_err,
3555                                  "Failed to truncate the tail of the image: ");
3556             }
3557         }
3558     } else {
3559         ret = qcow2_grow_l1_table(bs, new_l1_size, true);
3560         if (ret < 0) {
3561             error_setg_errno(errp, -ret, "Failed to grow the L1 table");
3562             goto fail;
3563         }
3564     }
3565 
3566     switch (prealloc) {
3567     case PREALLOC_MODE_OFF:
3568         break;
3569 
3570     case PREALLOC_MODE_METADATA:
3571         ret = preallocate_co(bs, old_length, offset);
3572         if (ret < 0) {
3573             error_setg_errno(errp, -ret, "Preallocation failed");
3574             goto fail;
3575         }
3576         break;
3577 
3578     case PREALLOC_MODE_FALLOC:
3579     case PREALLOC_MODE_FULL:
3580     {
3581         int64_t allocation_start, host_offset, guest_offset;
3582         int64_t clusters_allocated;
3583         int64_t old_file_size, new_file_size;
3584         uint64_t nb_new_data_clusters, nb_new_l2_tables;
3585 
3586         old_file_size = bdrv_getlength(bs->file->bs);
3587         if (old_file_size < 0) {
3588             error_setg_errno(errp, -old_file_size,
3589                              "Failed to inquire current file length");
3590             ret = old_file_size;
3591             goto fail;
3592         }
3593         old_file_size = ROUND_UP(old_file_size, s->cluster_size);
3594 
3595         nb_new_data_clusters = DIV_ROUND_UP(offset - old_length,
3596                                             s->cluster_size);
3597 
3598         /* This is an overestimation; we will not actually allocate space for
3599          * these in the file but just make sure the new refcount structures are
3600          * able to cover them so we will not have to allocate new refblocks
3601          * while entering the data blocks in the potentially new L2 tables.
3602          * (We do not actually care where the L2 tables are placed. Maybe they
3603          *  are already allocated or they can be placed somewhere before
3604          *  @old_file_size. It does not matter because they will be fully
3605          *  allocated automatically, so they do not need to be covered by the
3606          *  preallocation. All that matters is that we will not have to allocate
3607          *  new refcount structures for them.) */
3608         nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters,
3609                                         s->cluster_size / sizeof(uint64_t));
3610         /* The cluster range may not be aligned to L2 boundaries, so add one L2
3611          * table for a potential head/tail */
3612         nb_new_l2_tables++;
3613 
3614         allocation_start = qcow2_refcount_area(bs, old_file_size,
3615                                                nb_new_data_clusters +
3616                                                nb_new_l2_tables,
3617                                                true, 0, 0);
3618         if (allocation_start < 0) {
3619             error_setg_errno(errp, -allocation_start,
3620                              "Failed to resize refcount structures");
3621             ret = allocation_start;
3622             goto fail;
3623         }
3624 
3625         clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start,
3626                                                      nb_new_data_clusters);
3627         if (clusters_allocated < 0) {
3628             error_setg_errno(errp, -clusters_allocated,
3629                              "Failed to allocate data clusters");
3630             ret = clusters_allocated;
3631             goto fail;
3632         }
3633 
3634         assert(clusters_allocated == nb_new_data_clusters);
3635 
3636         /* Allocate the data area */
3637         new_file_size = allocation_start +
3638                         nb_new_data_clusters * s->cluster_size;
3639         ret = bdrv_co_truncate(bs->file, new_file_size, prealloc, errp);
3640         if (ret < 0) {
3641             error_prepend(errp, "Failed to resize underlying file: ");
3642             qcow2_free_clusters(bs, allocation_start,
3643                                 nb_new_data_clusters * s->cluster_size,
3644                                 QCOW2_DISCARD_OTHER);
3645             goto fail;
3646         }
3647 
3648         /* Create the necessary L2 entries */
3649         host_offset = allocation_start;
3650         guest_offset = old_length;
3651         while (nb_new_data_clusters) {
3652             int64_t nb_clusters = MIN(
3653                 nb_new_data_clusters,
3654                 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset));
3655             QCowL2Meta allocation = {
3656                 .offset       = guest_offset,
3657                 .alloc_offset = host_offset,
3658                 .nb_clusters  = nb_clusters,
3659             };
3660             qemu_co_queue_init(&allocation.dependent_requests);
3661 
3662             ret = qcow2_alloc_cluster_link_l2(bs, &allocation);
3663             if (ret < 0) {
3664                 error_setg_errno(errp, -ret, "Failed to update L2 tables");
3665                 qcow2_free_clusters(bs, host_offset,
3666                                     nb_new_data_clusters * s->cluster_size,
3667                                     QCOW2_DISCARD_OTHER);
3668                 goto fail;
3669             }
3670 
3671             guest_offset += nb_clusters * s->cluster_size;
3672             host_offset += nb_clusters * s->cluster_size;
3673             nb_new_data_clusters -= nb_clusters;
3674         }
3675         break;
3676     }
3677 
3678     default:
3679         g_assert_not_reached();
3680     }
3681 
3682     if (prealloc != PREALLOC_MODE_OFF) {
3683         /* Flush metadata before actually changing the image size */
3684         ret = qcow2_write_caches(bs);
3685         if (ret < 0) {
3686             error_setg_errno(errp, -ret,
3687                              "Failed to flush the preallocated area to disk");
3688             goto fail;
3689         }
3690     }
3691 
3692     bs->total_sectors = offset / BDRV_SECTOR_SIZE;
3693 
3694     /* write updated header.size */
3695     offset = cpu_to_be64(offset);
3696     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
3697                            &offset, sizeof(uint64_t));
3698     if (ret < 0) {
3699         error_setg_errno(errp, -ret, "Failed to update the image size");
3700         goto fail;
3701     }
3702 
3703     s->l1_vm_state_index = new_l1_size;
3704 
3705     /* Update cache sizes */
3706     options = qdict_clone_shallow(bs->options);
3707     ret = qcow2_update_options(bs, options, s->flags, errp);
3708     qobject_unref(options);
3709     if (ret < 0) {
3710         goto fail;
3711     }
3712     ret = 0;
3713 fail:
3714     qemu_co_mutex_unlock(&s->lock);
3715     return ret;
3716 }
3717 
3718 /*
3719  * qcow2_compress()
3720  *
3721  * @dest - destination buffer, at least of @size-1 bytes
3722  * @src - source buffer, @size bytes
3723  *
3724  * Returns: compressed size on success
3725  *          -1 if compression is inefficient
3726  *          -2 on any other error
3727  */
3728 static ssize_t qcow2_compress(void *dest, const void *src, size_t size)
3729 {
3730     ssize_t ret;
3731     z_stream strm;
3732 
3733     /* best compression, small window, no zlib header */
3734     memset(&strm, 0, sizeof(strm));
3735     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, Z_DEFLATED,
3736                        -12, 9, Z_DEFAULT_STRATEGY);
3737     if (ret != 0) {
3738         return -2;
3739     }
3740 
3741     /* strm.next_in is not const in old zlib versions, such as those used on
3742      * OpenBSD/NetBSD, so cast the const away */
3743     strm.avail_in = size;
3744     strm.next_in = (void *) src;
3745     strm.avail_out = size - 1;
3746     strm.next_out = dest;
3747 
3748     ret = deflate(&strm, Z_FINISH);
3749     if (ret == Z_STREAM_END) {
3750         ret = size - 1 - strm.avail_out;
3751     } else {
3752         ret = (ret == Z_OK ? -1 : -2);
3753     }
3754 
3755     deflateEnd(&strm);
3756 
3757     return ret;
3758 }
3759 
3760 #define MAX_COMPRESS_THREADS 4
3761 
3762 typedef struct Qcow2CompressData {
3763     void *dest;
3764     const void *src;
3765     size_t size;
3766     ssize_t ret;
3767 } Qcow2CompressData;
3768 
3769 static int qcow2_compress_pool_func(void *opaque)
3770 {
3771     Qcow2CompressData *data = opaque;
3772 
3773     data->ret = qcow2_compress(data->dest, data->src, data->size);
3774 
3775     return 0;
3776 }
3777 
3778 static void qcow2_compress_complete(void *opaque, int ret)
3779 {
3780     qemu_coroutine_enter(opaque);
3781 }
3782 
3783 /* See qcow2_compress definition for parameters description */
3784 static ssize_t qcow2_co_compress(BlockDriverState *bs,
3785                                  void *dest, const void *src, size_t size)
3786 {
3787     BDRVQcow2State *s = bs->opaque;
3788     BlockAIOCB *acb;
3789     ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
3790     Qcow2CompressData arg = {
3791         .dest = dest,
3792         .src = src,
3793         .size = size,
3794     };
3795 
3796     while (s->nb_compress_threads >= MAX_COMPRESS_THREADS) {
3797         qemu_co_queue_wait(&s->compress_wait_queue, NULL);
3798     }
3799 
3800     s->nb_compress_threads++;
3801     acb = thread_pool_submit_aio(pool, qcow2_compress_pool_func, &arg,
3802                                  qcow2_compress_complete,
3803                                  qemu_coroutine_self());
3804 
3805     if (!acb) {
3806         s->nb_compress_threads--;
3807         return -EINVAL;
3808     }
3809     qemu_coroutine_yield();
3810     s->nb_compress_threads--;
3811     qemu_co_queue_next(&s->compress_wait_queue);
3812 
3813     return arg.ret;
3814 }
3815 
3816 /* XXX: put compressed sectors first, then all the cluster aligned
3817    tables to avoid losing bytes in alignment */
3818 static coroutine_fn int
3819 qcow2_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
3820                             uint64_t bytes, QEMUIOVector *qiov)
3821 {
3822     BDRVQcow2State *s = bs->opaque;
3823     QEMUIOVector hd_qiov;
3824     struct iovec iov;
3825     int ret;
3826     size_t out_len;
3827     uint8_t *buf, *out_buf;
3828     int64_t cluster_offset;
3829 
3830     if (bytes == 0) {
3831         /* align end of file to a sector boundary to ease reading with
3832            sector based I/Os */
3833         cluster_offset = bdrv_getlength(bs->file->bs);
3834         if (cluster_offset < 0) {
3835             return cluster_offset;
3836         }
3837         return bdrv_co_truncate(bs->file, cluster_offset, PREALLOC_MODE_OFF,
3838                                 NULL);
3839     }
3840 
3841     if (offset_into_cluster(s, offset)) {
3842         return -EINVAL;
3843     }
3844 
3845     buf = qemu_blockalign(bs, s->cluster_size);
3846     if (bytes != s->cluster_size) {
3847         if (bytes > s->cluster_size ||
3848             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
3849         {
3850             qemu_vfree(buf);
3851             return -EINVAL;
3852         }
3853         /* Zero-pad last write if image size is not cluster aligned */
3854         memset(buf + bytes, 0, s->cluster_size - bytes);
3855     }
3856     qemu_iovec_to_buf(qiov, 0, buf, bytes);
3857 
3858     out_buf = g_malloc(s->cluster_size);
3859 
3860     out_len = qcow2_co_compress(bs, out_buf, buf, s->cluster_size);
3861     if (out_len == -2) {
3862         ret = -EINVAL;
3863         goto fail;
3864     } else if (out_len == -1) {
3865         /* could not compress: write normal cluster */
3866         ret = qcow2_co_pwritev(bs, offset, bytes, qiov, 0);
3867         if (ret < 0) {
3868             goto fail;
3869         }
3870         goto success;
3871     }
3872 
3873     qemu_co_mutex_lock(&s->lock);
3874     cluster_offset =
3875         qcow2_alloc_compressed_cluster_offset(bs, offset, out_len);
3876     if (!cluster_offset) {
3877         qemu_co_mutex_unlock(&s->lock);
3878         ret = -EIO;
3879         goto fail;
3880     }
3881     cluster_offset &= s->cluster_offset_mask;
3882 
3883     ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
3884     qemu_co_mutex_unlock(&s->lock);
3885     if (ret < 0) {
3886         goto fail;
3887     }
3888 
3889     iov = (struct iovec) {
3890         .iov_base   = out_buf,
3891         .iov_len    = out_len,
3892     };
3893     qemu_iovec_init_external(&hd_qiov, &iov, 1);
3894 
3895     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
3896     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
3897     if (ret < 0) {
3898         goto fail;
3899     }
3900 success:
3901     ret = 0;
3902 fail:
3903     qemu_vfree(buf);
3904     g_free(out_buf);
3905     return ret;
3906 }
3907 
3908 static int make_completely_empty(BlockDriverState *bs)
3909 {
3910     BDRVQcow2State *s = bs->opaque;
3911     Error *local_err = NULL;
3912     int ret, l1_clusters;
3913     int64_t offset;
3914     uint64_t *new_reftable = NULL;
3915     uint64_t rt_entry, l1_size2;
3916     struct {
3917         uint64_t l1_offset;
3918         uint64_t reftable_offset;
3919         uint32_t reftable_clusters;
3920     } QEMU_PACKED l1_ofs_rt_ofs_cls;
3921 
3922     ret = qcow2_cache_empty(bs, s->l2_table_cache);
3923     if (ret < 0) {
3924         goto fail;
3925     }
3926 
3927     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
3928     if (ret < 0) {
3929         goto fail;
3930     }
3931 
3932     /* Refcounts will be broken utterly */
3933     ret = qcow2_mark_dirty(bs);
3934     if (ret < 0) {
3935         goto fail;
3936     }
3937 
3938     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
3939 
3940     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
3941     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
3942 
3943     /* After this call, neither the in-memory nor the on-disk refcount
3944      * information accurately describe the actual references */
3945 
3946     ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
3947                              l1_clusters * s->cluster_size, 0);
3948     if (ret < 0) {
3949         goto fail_broken_refcounts;
3950     }
3951     memset(s->l1_table, 0, l1_size2);
3952 
3953     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
3954 
3955     /* Overwrite enough clusters at the beginning of the sectors to place
3956      * the refcount table, a refcount block and the L1 table in; this may
3957      * overwrite parts of the existing refcount and L1 table, which is not
3958      * an issue because the dirty flag is set, complete data loss is in fact
3959      * desired and partial data loss is consequently fine as well */
3960     ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
3961                              (2 + l1_clusters) * s->cluster_size, 0);
3962     /* This call (even if it failed overall) may have overwritten on-disk
3963      * refcount structures; in that case, the in-memory refcount information
3964      * will probably differ from the on-disk information which makes the BDS
3965      * unusable */
3966     if (ret < 0) {
3967         goto fail_broken_refcounts;
3968     }
3969 
3970     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
3971     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
3972 
3973     /* "Create" an empty reftable (one cluster) directly after the image
3974      * header and an empty L1 table three clusters after the image header;
3975      * the cluster between those two will be used as the first refblock */
3976     l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
3977     l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
3978     l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
3979     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
3980                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
3981     if (ret < 0) {
3982         goto fail_broken_refcounts;
3983     }
3984 
3985     s->l1_table_offset = 3 * s->cluster_size;
3986 
3987     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
3988     if (!new_reftable) {
3989         ret = -ENOMEM;
3990         goto fail_broken_refcounts;
3991     }
3992 
3993     s->refcount_table_offset = s->cluster_size;
3994     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
3995     s->max_refcount_table_index = 0;
3996 
3997     g_free(s->refcount_table);
3998     s->refcount_table = new_reftable;
3999     new_reftable = NULL;
4000 
4001     /* Now the in-memory refcount information again corresponds to the on-disk
4002      * information (reftable is empty and no refblocks (the refblock cache is
4003      * empty)); however, this means some clusters (e.g. the image header) are
4004      * referenced, but not refcounted, but the normal qcow2 code assumes that
4005      * the in-memory information is always correct */
4006 
4007     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
4008 
4009     /* Enter the first refblock into the reftable */
4010     rt_entry = cpu_to_be64(2 * s->cluster_size);
4011     ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
4012                            &rt_entry, sizeof(rt_entry));
4013     if (ret < 0) {
4014         goto fail_broken_refcounts;
4015     }
4016     s->refcount_table[0] = 2 * s->cluster_size;
4017 
4018     s->free_cluster_index = 0;
4019     assert(3 + l1_clusters <= s->refcount_block_size);
4020     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
4021     if (offset < 0) {
4022         ret = offset;
4023         goto fail_broken_refcounts;
4024     } else if (offset > 0) {
4025         error_report("First cluster in emptied image is in use");
4026         abort();
4027     }
4028 
4029     /* Now finally the in-memory information corresponds to the on-disk
4030      * structures and is correct */
4031     ret = qcow2_mark_clean(bs);
4032     if (ret < 0) {
4033         goto fail;
4034     }
4035 
4036     ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size,
4037                         PREALLOC_MODE_OFF, &local_err);
4038     if (ret < 0) {
4039         error_report_err(local_err);
4040         goto fail;
4041     }
4042 
4043     return 0;
4044 
4045 fail_broken_refcounts:
4046     /* The BDS is unusable at this point. If we wanted to make it usable, we
4047      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4048      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4049      * again. However, because the functions which could have caused this error
4050      * path to be taken are used by those functions as well, it's very likely
4051      * that that sequence will fail as well. Therefore, just eject the BDS. */
4052     bs->drv = NULL;
4053 
4054 fail:
4055     g_free(new_reftable);
4056     return ret;
4057 }
4058 
4059 static int qcow2_make_empty(BlockDriverState *bs)
4060 {
4061     BDRVQcow2State *s = bs->opaque;
4062     uint64_t offset, end_offset;
4063     int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
4064     int l1_clusters, ret = 0;
4065 
4066     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
4067 
4068     if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps &&
4069         3 + l1_clusters <= s->refcount_block_size &&
4070         s->crypt_method_header != QCOW_CRYPT_LUKS) {
4071         /* The following function only works for qcow2 v3 images (it
4072          * requires the dirty flag) and only as long as there are no
4073          * features that reserve extra clusters (such as snapshots,
4074          * LUKS header, or persistent bitmaps), because it completely
4075          * empties the image.  Furthermore, the L1 table and three
4076          * additional clusters (image header, refcount table, one
4077          * refcount block) have to fit inside one refcount block. */
4078         return make_completely_empty(bs);
4079     }
4080 
4081     /* This fallback code simply discards every active cluster; this is slow,
4082      * but works in all cases */
4083     end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
4084     for (offset = 0; offset < end_offset; offset += step) {
4085         /* As this function is generally used after committing an external
4086          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4087          * default action for this kind of discard is to pass the discard,
4088          * which will ideally result in an actually smaller image file, as
4089          * is probably desired. */
4090         ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
4091                                     QCOW2_DISCARD_SNAPSHOT, true);
4092         if (ret < 0) {
4093             break;
4094         }
4095     }
4096 
4097     return ret;
4098 }
4099 
4100 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
4101 {
4102     BDRVQcow2State *s = bs->opaque;
4103     int ret;
4104 
4105     qemu_co_mutex_lock(&s->lock);
4106     ret = qcow2_write_caches(bs);
4107     qemu_co_mutex_unlock(&s->lock);
4108 
4109     return ret;
4110 }
4111 
4112 static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs,
4113                                        Error **errp)
4114 {
4115     Error *local_err = NULL;
4116     BlockMeasureInfo *info;
4117     uint64_t required = 0; /* bytes that contribute to required size */
4118     uint64_t virtual_size; /* disk size as seen by guest */
4119     uint64_t refcount_bits;
4120     uint64_t l2_tables;
4121     size_t cluster_size;
4122     int version;
4123     char *optstr;
4124     PreallocMode prealloc;
4125     bool has_backing_file;
4126 
4127     /* Parse image creation options */
4128     cluster_size = qcow2_opt_get_cluster_size_del(opts, &local_err);
4129     if (local_err) {
4130         goto err;
4131     }
4132 
4133     version = qcow2_opt_get_version_del(opts, &local_err);
4134     if (local_err) {
4135         goto err;
4136     }
4137 
4138     refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
4139     if (local_err) {
4140         goto err;
4141     }
4142 
4143     optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
4144     prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr,
4145                                PREALLOC_MODE_OFF, &local_err);
4146     g_free(optstr);
4147     if (local_err) {
4148         goto err;
4149     }
4150 
4151     optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
4152     has_backing_file = !!optstr;
4153     g_free(optstr);
4154 
4155     virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
4156     virtual_size = ROUND_UP(virtual_size, cluster_size);
4157 
4158     /* Check that virtual disk size is valid */
4159     l2_tables = DIV_ROUND_UP(virtual_size / cluster_size,
4160                              cluster_size / sizeof(uint64_t));
4161     if (l2_tables * sizeof(uint64_t) > QCOW_MAX_L1_SIZE) {
4162         error_setg(&local_err, "The image size is too large "
4163                                "(try using a larger cluster size)");
4164         goto err;
4165     }
4166 
4167     /* Account for input image */
4168     if (in_bs) {
4169         int64_t ssize = bdrv_getlength(in_bs);
4170         if (ssize < 0) {
4171             error_setg_errno(&local_err, -ssize,
4172                              "Unable to get image virtual_size");
4173             goto err;
4174         }
4175 
4176         virtual_size = ROUND_UP(ssize, cluster_size);
4177 
4178         if (has_backing_file) {
4179             /* We don't how much of the backing chain is shared by the input
4180              * image and the new image file.  In the worst case the new image's
4181              * backing file has nothing in common with the input image.  Be
4182              * conservative and assume all clusters need to be written.
4183              */
4184             required = virtual_size;
4185         } else {
4186             int64_t offset;
4187             int64_t pnum = 0;
4188 
4189             for (offset = 0; offset < ssize; offset += pnum) {
4190                 int ret;
4191 
4192                 ret = bdrv_block_status_above(in_bs, NULL, offset,
4193                                               ssize - offset, &pnum, NULL,
4194                                               NULL);
4195                 if (ret < 0) {
4196                     error_setg_errno(&local_err, -ret,
4197                                      "Unable to get block status");
4198                     goto err;
4199                 }
4200 
4201                 if (ret & BDRV_BLOCK_ZERO) {
4202                     /* Skip zero regions (safe with no backing file) */
4203                 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
4204                            (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
4205                     /* Extend pnum to end of cluster for next iteration */
4206                     pnum = ROUND_UP(offset + pnum, cluster_size) - offset;
4207 
4208                     /* Count clusters we've seen */
4209                     required += offset % cluster_size + pnum;
4210                 }
4211             }
4212         }
4213     }
4214 
4215     /* Take into account preallocation.  Nothing special is needed for
4216      * PREALLOC_MODE_METADATA since metadata is always counted.
4217      */
4218     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
4219         required = virtual_size;
4220     }
4221 
4222     info = g_new(BlockMeasureInfo, 1);
4223     info->fully_allocated =
4224         qcow2_calc_prealloc_size(virtual_size, cluster_size,
4225                                  ctz32(refcount_bits));
4226 
4227     /* Remove data clusters that are not required.  This overestimates the
4228      * required size because metadata needed for the fully allocated file is
4229      * still counted.
4230      */
4231     info->required = info->fully_allocated - virtual_size + required;
4232     return info;
4233 
4234 err:
4235     error_propagate(errp, local_err);
4236     return NULL;
4237 }
4238 
4239 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4240 {
4241     BDRVQcow2State *s = bs->opaque;
4242     bdi->unallocated_blocks_are_zero = true;
4243     bdi->cluster_size = s->cluster_size;
4244     bdi->vm_state_offset = qcow2_vm_state_offset(s);
4245     return 0;
4246 }
4247 
4248 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
4249 {
4250     BDRVQcow2State *s = bs->opaque;
4251     ImageInfoSpecific *spec_info;
4252     QCryptoBlockInfo *encrypt_info = NULL;
4253 
4254     if (s->crypto != NULL) {
4255         encrypt_info = qcrypto_block_get_info(s->crypto, &error_abort);
4256     }
4257 
4258     spec_info = g_new(ImageInfoSpecific, 1);
4259     *spec_info = (ImageInfoSpecific){
4260         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
4261         .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1),
4262     };
4263     if (s->qcow_version == 2) {
4264         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
4265             .compat             = g_strdup("0.10"),
4266             .refcount_bits      = s->refcount_bits,
4267         };
4268     } else if (s->qcow_version == 3) {
4269         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
4270             .compat             = g_strdup("1.1"),
4271             .lazy_refcounts     = s->compatible_features &
4272                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
4273             .has_lazy_refcounts = true,
4274             .corrupt            = s->incompatible_features &
4275                                   QCOW2_INCOMPAT_CORRUPT,
4276             .has_corrupt        = true,
4277             .refcount_bits      = s->refcount_bits,
4278         };
4279     } else {
4280         /* if this assertion fails, this probably means a new version was
4281          * added without having it covered here */
4282         assert(false);
4283     }
4284 
4285     if (encrypt_info) {
4286         ImageInfoSpecificQCow2Encryption *qencrypt =
4287             g_new(ImageInfoSpecificQCow2Encryption, 1);
4288         switch (encrypt_info->format) {
4289         case Q_CRYPTO_BLOCK_FORMAT_QCOW:
4290             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES;
4291             break;
4292         case Q_CRYPTO_BLOCK_FORMAT_LUKS:
4293             qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS;
4294             qencrypt->u.luks = encrypt_info->u.luks;
4295             break;
4296         default:
4297             abort();
4298         }
4299         /* Since we did shallow copy above, erase any pointers
4300          * in the original info */
4301         memset(&encrypt_info->u, 0, sizeof(encrypt_info->u));
4302         qapi_free_QCryptoBlockInfo(encrypt_info);
4303 
4304         spec_info->u.qcow2.data->has_encrypt = true;
4305         spec_info->u.qcow2.data->encrypt = qencrypt;
4306     }
4307 
4308     return spec_info;
4309 }
4310 
4311 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
4312                               int64_t pos)
4313 {
4314     BDRVQcow2State *s = bs->opaque;
4315 
4316     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
4317     return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos,
4318                                     qiov->size, qiov, 0);
4319 }
4320 
4321 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
4322                               int64_t pos)
4323 {
4324     BDRVQcow2State *s = bs->opaque;
4325 
4326     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
4327     return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos,
4328                                    qiov->size, qiov, 0);
4329 }
4330 
4331 /*
4332  * Downgrades an image's version. To achieve this, any incompatible features
4333  * have to be removed.
4334  */
4335 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
4336                            BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
4337                            Error **errp)
4338 {
4339     BDRVQcow2State *s = bs->opaque;
4340     int current_version = s->qcow_version;
4341     int ret;
4342 
4343     /* This is qcow2_downgrade(), not qcow2_upgrade() */
4344     assert(target_version < current_version);
4345 
4346     /* There are no other versions (now) that you can downgrade to */
4347     assert(target_version == 2);
4348 
4349     if (s->refcount_order != 4) {
4350         error_setg(errp, "compat=0.10 requires refcount_bits=16");
4351         return -ENOTSUP;
4352     }
4353 
4354     /* clear incompatible features */
4355     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
4356         ret = qcow2_mark_clean(bs);
4357         if (ret < 0) {
4358             error_setg_errno(errp, -ret, "Failed to make the image clean");
4359             return ret;
4360         }
4361     }
4362 
4363     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
4364      * the first place; if that happens nonetheless, returning -ENOTSUP is the
4365      * best thing to do anyway */
4366 
4367     if (s->incompatible_features) {
4368         error_setg(errp, "Cannot downgrade an image with incompatible features "
4369                    "%#" PRIx64 " set", s->incompatible_features);
4370         return -ENOTSUP;
4371     }
4372 
4373     /* since we can ignore compatible features, we can set them to 0 as well */
4374     s->compatible_features = 0;
4375     /* if lazy refcounts have been used, they have already been fixed through
4376      * clearing the dirty flag */
4377 
4378     /* clearing autoclear features is trivial */
4379     s->autoclear_features = 0;
4380 
4381     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
4382     if (ret < 0) {
4383         error_setg_errno(errp, -ret, "Failed to turn zero into data clusters");
4384         return ret;
4385     }
4386 
4387     s->qcow_version = target_version;
4388     ret = qcow2_update_header(bs);
4389     if (ret < 0) {
4390         s->qcow_version = current_version;
4391         error_setg_errno(errp, -ret, "Failed to update the image header");
4392         return ret;
4393     }
4394     return 0;
4395 }
4396 
4397 typedef enum Qcow2AmendOperation {
4398     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
4399      * statically initialized to so that the helper CB can discern the first
4400      * invocation from an operation change */
4401     QCOW2_NO_OPERATION = 0,
4402 
4403     QCOW2_CHANGING_REFCOUNT_ORDER,
4404     QCOW2_DOWNGRADING,
4405 } Qcow2AmendOperation;
4406 
4407 typedef struct Qcow2AmendHelperCBInfo {
4408     /* The code coordinating the amend operations should only modify
4409      * these four fields; the rest will be managed by the CB */
4410     BlockDriverAmendStatusCB *original_status_cb;
4411     void *original_cb_opaque;
4412 
4413     Qcow2AmendOperation current_operation;
4414 
4415     /* Total number of operations to perform (only set once) */
4416     int total_operations;
4417 
4418     /* The following fields are managed by the CB */
4419 
4420     /* Number of operations completed */
4421     int operations_completed;
4422 
4423     /* Cumulative offset of all completed operations */
4424     int64_t offset_completed;
4425 
4426     Qcow2AmendOperation last_operation;
4427     int64_t last_work_size;
4428 } Qcow2AmendHelperCBInfo;
4429 
4430 static void qcow2_amend_helper_cb(BlockDriverState *bs,
4431                                   int64_t operation_offset,
4432                                   int64_t operation_work_size, void *opaque)
4433 {
4434     Qcow2AmendHelperCBInfo *info = opaque;
4435     int64_t current_work_size;
4436     int64_t projected_work_size;
4437 
4438     if (info->current_operation != info->last_operation) {
4439         if (info->last_operation != QCOW2_NO_OPERATION) {
4440             info->offset_completed += info->last_work_size;
4441             info->operations_completed++;
4442         }
4443 
4444         info->last_operation = info->current_operation;
4445     }
4446 
4447     assert(info->total_operations > 0);
4448     assert(info->operations_completed < info->total_operations);
4449 
4450     info->last_work_size = operation_work_size;
4451 
4452     current_work_size = info->offset_completed + operation_work_size;
4453 
4454     /* current_work_size is the total work size for (operations_completed + 1)
4455      * operations (which includes this one), so multiply it by the number of
4456      * operations not covered and divide it by the number of operations
4457      * covered to get a projection for the operations not covered */
4458     projected_work_size = current_work_size * (info->total_operations -
4459                                                info->operations_completed - 1)
4460                                             / (info->operations_completed + 1);
4461 
4462     info->original_status_cb(bs, info->offset_completed + operation_offset,
4463                              current_work_size + projected_work_size,
4464                              info->original_cb_opaque);
4465 }
4466 
4467 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
4468                                BlockDriverAmendStatusCB *status_cb,
4469                                void *cb_opaque,
4470                                Error **errp)
4471 {
4472     BDRVQcow2State *s = bs->opaque;
4473     int old_version = s->qcow_version, new_version = old_version;
4474     uint64_t new_size = 0;
4475     const char *backing_file = NULL, *backing_format = NULL;
4476     bool lazy_refcounts = s->use_lazy_refcounts;
4477     const char *compat = NULL;
4478     uint64_t cluster_size = s->cluster_size;
4479     bool encrypt;
4480     int encformat;
4481     int refcount_bits = s->refcount_bits;
4482     int ret;
4483     QemuOptDesc *desc = opts->list->desc;
4484     Qcow2AmendHelperCBInfo helper_cb_info;
4485 
4486     while (desc && desc->name) {
4487         if (!qemu_opt_find(opts, desc->name)) {
4488             /* only change explicitly defined options */
4489             desc++;
4490             continue;
4491         }
4492 
4493         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
4494             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
4495             if (!compat) {
4496                 /* preserve default */
4497             } else if (!strcmp(compat, "0.10")) {
4498                 new_version = 2;
4499             } else if (!strcmp(compat, "1.1")) {
4500                 new_version = 3;
4501             } else {
4502                 error_setg(errp, "Unknown compatibility level %s", compat);
4503                 return -EINVAL;
4504             }
4505         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
4506             error_setg(errp, "Cannot change preallocation mode");
4507             return -ENOTSUP;
4508         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
4509             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
4510         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
4511             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4512         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
4513             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4514         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
4515             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
4516                                         !!s->crypto);
4517 
4518             if (encrypt != !!s->crypto) {
4519                 error_setg(errp,
4520                            "Changing the encryption flag is not supported");
4521                 return -ENOTSUP;
4522             }
4523         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT_FORMAT)) {
4524             encformat = qcow2_crypt_method_from_format(
4525                 qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT));
4526 
4527             if (encformat != s->crypt_method_header) {
4528                 error_setg(errp,
4529                            "Changing the encryption format is not supported");
4530                 return -ENOTSUP;
4531             }
4532         } else if (g_str_has_prefix(desc->name, "encrypt.")) {
4533             error_setg(errp,
4534                        "Changing the encryption parameters is not supported");
4535             return -ENOTSUP;
4536         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
4537             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
4538                                              cluster_size);
4539             if (cluster_size != s->cluster_size) {
4540                 error_setg(errp, "Changing the cluster size is not supported");
4541                 return -ENOTSUP;
4542             }
4543         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
4544             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
4545                                                lazy_refcounts);
4546         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
4547             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
4548                                                 refcount_bits);
4549 
4550             if (refcount_bits <= 0 || refcount_bits > 64 ||
4551                 !is_power_of_2(refcount_bits))
4552             {
4553                 error_setg(errp, "Refcount width must be a power of two and "
4554                            "may not exceed 64 bits");
4555                 return -EINVAL;
4556             }
4557         } else {
4558             /* if this point is reached, this probably means a new option was
4559              * added without having it covered here */
4560             abort();
4561         }
4562 
4563         desc++;
4564     }
4565 
4566     helper_cb_info = (Qcow2AmendHelperCBInfo){
4567         .original_status_cb = status_cb,
4568         .original_cb_opaque = cb_opaque,
4569         .total_operations = (new_version < old_version)
4570                           + (s->refcount_bits != refcount_bits)
4571     };
4572 
4573     /* Upgrade first (some features may require compat=1.1) */
4574     if (new_version > old_version) {
4575         s->qcow_version = new_version;
4576         ret = qcow2_update_header(bs);
4577         if (ret < 0) {
4578             s->qcow_version = old_version;
4579             error_setg_errno(errp, -ret, "Failed to update the image header");
4580             return ret;
4581         }
4582     }
4583 
4584     if (s->refcount_bits != refcount_bits) {
4585         int refcount_order = ctz32(refcount_bits);
4586 
4587         if (new_version < 3 && refcount_bits != 16) {
4588             error_setg(errp, "Refcount widths other than 16 bits require "
4589                        "compatibility level 1.1 or above (use compat=1.1 or "
4590                        "greater)");
4591             return -EINVAL;
4592         }
4593 
4594         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
4595         ret = qcow2_change_refcount_order(bs, refcount_order,
4596                                           &qcow2_amend_helper_cb,
4597                                           &helper_cb_info, errp);
4598         if (ret < 0) {
4599             return ret;
4600         }
4601     }
4602 
4603     if (backing_file || backing_format) {
4604         ret = qcow2_change_backing_file(bs,
4605                     backing_file ?: s->image_backing_file,
4606                     backing_format ?: s->image_backing_format);
4607         if (ret < 0) {
4608             error_setg_errno(errp, -ret, "Failed to change the backing file");
4609             return ret;
4610         }
4611     }
4612 
4613     if (s->use_lazy_refcounts != lazy_refcounts) {
4614         if (lazy_refcounts) {
4615             if (new_version < 3) {
4616                 error_setg(errp, "Lazy refcounts only supported with "
4617                            "compatibility level 1.1 and above (use compat=1.1 "
4618                            "or greater)");
4619                 return -EINVAL;
4620             }
4621             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
4622             ret = qcow2_update_header(bs);
4623             if (ret < 0) {
4624                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
4625                 error_setg_errno(errp, -ret, "Failed to update the image header");
4626                 return ret;
4627             }
4628             s->use_lazy_refcounts = true;
4629         } else {
4630             /* make image clean first */
4631             ret = qcow2_mark_clean(bs);
4632             if (ret < 0) {
4633                 error_setg_errno(errp, -ret, "Failed to make the image clean");
4634                 return ret;
4635             }
4636             /* now disallow lazy refcounts */
4637             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
4638             ret = qcow2_update_header(bs);
4639             if (ret < 0) {
4640                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
4641                 error_setg_errno(errp, -ret, "Failed to update the image header");
4642                 return ret;
4643             }
4644             s->use_lazy_refcounts = false;
4645         }
4646     }
4647 
4648     if (new_size) {
4649         BlockBackend *blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
4650         ret = blk_insert_bs(blk, bs, errp);
4651         if (ret < 0) {
4652             blk_unref(blk);
4653             return ret;
4654         }
4655 
4656         ret = blk_truncate(blk, new_size, PREALLOC_MODE_OFF, errp);
4657         blk_unref(blk);
4658         if (ret < 0) {
4659             return ret;
4660         }
4661     }
4662 
4663     /* Downgrade last (so unsupported features can be removed before) */
4664     if (new_version < old_version) {
4665         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
4666         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
4667                               &helper_cb_info, errp);
4668         if (ret < 0) {
4669             return ret;
4670         }
4671     }
4672 
4673     return 0;
4674 }
4675 
4676 /*
4677  * If offset or size are negative, respectively, they will not be included in
4678  * the BLOCK_IMAGE_CORRUPTED event emitted.
4679  * fatal will be ignored for read-only BDS; corruptions found there will always
4680  * be considered non-fatal.
4681  */
4682 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
4683                              int64_t size, const char *message_format, ...)
4684 {
4685     BDRVQcow2State *s = bs->opaque;
4686     const char *node_name;
4687     char *message;
4688     va_list ap;
4689 
4690     fatal = fatal && bdrv_is_writable(bs);
4691 
4692     if (s->signaled_corruption &&
4693         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
4694     {
4695         return;
4696     }
4697 
4698     va_start(ap, message_format);
4699     message = g_strdup_vprintf(message_format, ap);
4700     va_end(ap);
4701 
4702     if (fatal) {
4703         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
4704                 "corruption events will be suppressed\n", message);
4705     } else {
4706         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
4707                 "corruption events will be suppressed\n", message);
4708     }
4709 
4710     node_name = bdrv_get_node_name(bs);
4711     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
4712                                           *node_name != '\0', node_name,
4713                                           message, offset >= 0, offset,
4714                                           size >= 0, size,
4715                                           fatal);
4716     g_free(message);
4717 
4718     if (fatal) {
4719         qcow2_mark_corrupt(bs);
4720         bs->drv = NULL; /* make BDS unusable */
4721     }
4722 
4723     s->signaled_corruption = true;
4724 }
4725 
4726 static QemuOptsList qcow2_create_opts = {
4727     .name = "qcow2-create-opts",
4728     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
4729     .desc = {
4730         {
4731             .name = BLOCK_OPT_SIZE,
4732             .type = QEMU_OPT_SIZE,
4733             .help = "Virtual disk size"
4734         },
4735         {
4736             .name = BLOCK_OPT_COMPAT_LEVEL,
4737             .type = QEMU_OPT_STRING,
4738             .help = "Compatibility level (0.10 or 1.1)"
4739         },
4740         {
4741             .name = BLOCK_OPT_BACKING_FILE,
4742             .type = QEMU_OPT_STRING,
4743             .help = "File name of a base image"
4744         },
4745         {
4746             .name = BLOCK_OPT_BACKING_FMT,
4747             .type = QEMU_OPT_STRING,
4748             .help = "Image format of the base image"
4749         },
4750         {
4751             .name = BLOCK_OPT_ENCRYPT,
4752             .type = QEMU_OPT_BOOL,
4753             .help = "Encrypt the image with format 'aes'. (Deprecated "
4754                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
4755         },
4756         {
4757             .name = BLOCK_OPT_ENCRYPT_FORMAT,
4758             .type = QEMU_OPT_STRING,
4759             .help = "Encrypt the image, format choices: 'aes', 'luks'",
4760         },
4761         BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
4762             "ID of secret providing qcow AES key or LUKS passphrase"),
4763         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
4764         BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
4765         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
4766         BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
4767         BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
4768         BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
4769         {
4770             .name = BLOCK_OPT_CLUSTER_SIZE,
4771             .type = QEMU_OPT_SIZE,
4772             .help = "qcow2 cluster size",
4773             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
4774         },
4775         {
4776             .name = BLOCK_OPT_PREALLOC,
4777             .type = QEMU_OPT_STRING,
4778             .help = "Preallocation mode (allowed values: off, metadata, "
4779                     "falloc, full)"
4780         },
4781         {
4782             .name = BLOCK_OPT_LAZY_REFCOUNTS,
4783             .type = QEMU_OPT_BOOL,
4784             .help = "Postpone refcount updates",
4785             .def_value_str = "off"
4786         },
4787         {
4788             .name = BLOCK_OPT_REFCOUNT_BITS,
4789             .type = QEMU_OPT_NUMBER,
4790             .help = "Width of a reference count entry in bits",
4791             .def_value_str = "16"
4792         },
4793         { /* end of list */ }
4794     }
4795 };
4796 
4797 BlockDriver bdrv_qcow2 = {
4798     .format_name        = "qcow2",
4799     .instance_size      = sizeof(BDRVQcow2State),
4800     .bdrv_probe         = qcow2_probe,
4801     .bdrv_open          = qcow2_open,
4802     .bdrv_close         = qcow2_close,
4803     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
4804     .bdrv_reopen_commit   = qcow2_reopen_commit,
4805     .bdrv_reopen_abort    = qcow2_reopen_abort,
4806     .bdrv_join_options    = qcow2_join_options,
4807     .bdrv_child_perm      = bdrv_format_default_perms,
4808     .bdrv_co_create_opts  = qcow2_co_create_opts,
4809     .bdrv_co_create       = qcow2_co_create,
4810     .bdrv_has_zero_init = bdrv_has_zero_init_1,
4811     .bdrv_co_block_status = qcow2_co_block_status,
4812 
4813     .bdrv_co_preadv         = qcow2_co_preadv,
4814     .bdrv_co_pwritev        = qcow2_co_pwritev,
4815     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
4816 
4817     .bdrv_co_pwrite_zeroes  = qcow2_co_pwrite_zeroes,
4818     .bdrv_co_pdiscard       = qcow2_co_pdiscard,
4819     .bdrv_co_copy_range_from = qcow2_co_copy_range_from,
4820     .bdrv_co_copy_range_to  = qcow2_co_copy_range_to,
4821     .bdrv_co_truncate       = qcow2_co_truncate,
4822     .bdrv_co_pwritev_compressed = qcow2_co_pwritev_compressed,
4823     .bdrv_make_empty        = qcow2_make_empty,
4824 
4825     .bdrv_snapshot_create   = qcow2_snapshot_create,
4826     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
4827     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
4828     .bdrv_snapshot_list     = qcow2_snapshot_list,
4829     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
4830     .bdrv_measure           = qcow2_measure,
4831     .bdrv_get_info          = qcow2_get_info,
4832     .bdrv_get_specific_info = qcow2_get_specific_info,
4833 
4834     .bdrv_save_vmstate    = qcow2_save_vmstate,
4835     .bdrv_load_vmstate    = qcow2_load_vmstate,
4836 
4837     .supports_backing           = true,
4838     .bdrv_change_backing_file   = qcow2_change_backing_file,
4839 
4840     .bdrv_refresh_limits        = qcow2_refresh_limits,
4841     .bdrv_co_invalidate_cache   = qcow2_co_invalidate_cache,
4842     .bdrv_inactivate            = qcow2_inactivate,
4843 
4844     .create_opts         = &qcow2_create_opts,
4845     .bdrv_co_check       = qcow2_co_check,
4846     .bdrv_amend_options  = qcow2_amend_options,
4847 
4848     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
4849     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
4850 
4851     .bdrv_reopen_bitmaps_rw = qcow2_reopen_bitmaps_rw,
4852     .bdrv_can_store_new_dirty_bitmap = qcow2_can_store_new_dirty_bitmap,
4853     .bdrv_remove_persistent_dirty_bitmap = qcow2_remove_persistent_dirty_bitmap,
4854 };
4855 
4856 static void bdrv_qcow2_init(void)
4857 {
4858     bdrv_register(&bdrv_qcow2);
4859 }
4860 
4861 block_init(bdrv_qcow2_init);
4862