xref: /openbmc/qemu/block/qcow2.c (revision dd2bf9eb)
1 /*
2  * Block driver for the QCOW version 2 format
3  *
4  * Copyright (c) 2004-2006 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 #include "qemu-common.h"
25 #include "block/block_int.h"
26 #include "qemu/module.h"
27 #include <zlib.h>
28 #include "block/qcow2.h"
29 #include "qemu/error-report.h"
30 #include "qapi/qmp/qerror.h"
31 #include "qapi/qmp/qbool.h"
32 #include "qapi/util.h"
33 #include "qapi/qmp/types.h"
34 #include "qapi-event.h"
35 #include "trace.h"
36 #include "qemu/option_int.h"
37 
38 /*
39   Differences with QCOW:
40 
41   - Support for multiple incremental snapshots.
42   - Memory management by reference counts.
43   - Clusters which have a reference count of one have the bit
44     QCOW_OFLAG_COPIED to optimize write performance.
45   - Size of compressed clusters is stored in sectors to reduce bit usage
46     in the cluster offsets.
47   - Support for storing additional data (such as the VM state) in the
48     snapshots.
49   - If a backing store is used, the cluster size is not constrained
50     (could be backported to QCOW).
51   - L2 tables have always a size of one cluster.
52 */
53 
54 
55 typedef struct {
56     uint32_t magic;
57     uint32_t len;
58 } QEMU_PACKED QCowExtension;
59 
60 #define  QCOW2_EXT_MAGIC_END 0
61 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
62 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
63 
64 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
65 {
66     const QCowHeader *cow_header = (const void *)buf;
67 
68     if (buf_size >= sizeof(QCowHeader) &&
69         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
70         be32_to_cpu(cow_header->version) >= 2)
71         return 100;
72     else
73         return 0;
74 }
75 
76 
77 /*
78  * read qcow2 extension and fill bs
79  * start reading from start_offset
80  * finish reading upon magic of value 0 or when end_offset reached
81  * unknown magic is skipped (future extension this version knows nothing about)
82  * return 0 upon success, non-0 otherwise
83  */
84 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
85                                  uint64_t end_offset, void **p_feature_table,
86                                  Error **errp)
87 {
88     BDRVQcow2State *s = bs->opaque;
89     QCowExtension ext;
90     uint64_t offset;
91     int ret;
92 
93 #ifdef DEBUG_EXT
94     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
95 #endif
96     offset = start_offset;
97     while (offset < end_offset) {
98 
99 #ifdef DEBUG_EXT
100         /* Sanity check */
101         if (offset > s->cluster_size)
102             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
103 
104         printf("attempting to read extended header in offset %lu\n", offset);
105 #endif
106 
107         ret = bdrv_pread(bs->file->bs, offset, &ext, sizeof(ext));
108         if (ret < 0) {
109             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
110                              "pread fail from offset %" PRIu64, offset);
111             return 1;
112         }
113         be32_to_cpus(&ext.magic);
114         be32_to_cpus(&ext.len);
115         offset += sizeof(ext);
116 #ifdef DEBUG_EXT
117         printf("ext.magic = 0x%x\n", ext.magic);
118 #endif
119         if (offset > end_offset || ext.len > end_offset - offset) {
120             error_setg(errp, "Header extension too large");
121             return -EINVAL;
122         }
123 
124         switch (ext.magic) {
125         case QCOW2_EXT_MAGIC_END:
126             return 0;
127 
128         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
129             if (ext.len >= sizeof(bs->backing_format)) {
130                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
131                            " too large (>=%zu)", ext.len,
132                            sizeof(bs->backing_format));
133                 return 2;
134             }
135             ret = bdrv_pread(bs->file->bs, offset, bs->backing_format, ext.len);
136             if (ret < 0) {
137                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
138                                  "Could not read format name");
139                 return 3;
140             }
141             bs->backing_format[ext.len] = '\0';
142             s->image_backing_format = g_strdup(bs->backing_format);
143 #ifdef DEBUG_EXT
144             printf("Qcow2: Got format extension %s\n", bs->backing_format);
145 #endif
146             break;
147 
148         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
149             if (p_feature_table != NULL) {
150                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
151                 ret = bdrv_pread(bs->file->bs, offset , feature_table, ext.len);
152                 if (ret < 0) {
153                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
154                                      "Could not read table");
155                     return ret;
156                 }
157 
158                 *p_feature_table = feature_table;
159             }
160             break;
161 
162         default:
163             /* unknown magic - save it in case we need to rewrite the header */
164             {
165                 Qcow2UnknownHeaderExtension *uext;
166 
167                 uext = g_malloc0(sizeof(*uext)  + ext.len);
168                 uext->magic = ext.magic;
169                 uext->len = ext.len;
170                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
171 
172                 ret = bdrv_pread(bs->file->bs, offset , uext->data, uext->len);
173                 if (ret < 0) {
174                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
175                                      "Could not read data");
176                     return ret;
177                 }
178             }
179             break;
180         }
181 
182         offset += ((ext.len + 7) & ~7);
183     }
184 
185     return 0;
186 }
187 
188 static void cleanup_unknown_header_ext(BlockDriverState *bs)
189 {
190     BDRVQcow2State *s = bs->opaque;
191     Qcow2UnknownHeaderExtension *uext, *next;
192 
193     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
194         QLIST_REMOVE(uext, next);
195         g_free(uext);
196     }
197 }
198 
199 static void GCC_FMT_ATTR(3, 4) report_unsupported(BlockDriverState *bs,
200     Error **errp, const char *fmt, ...)
201 {
202     char msg[64];
203     va_list ap;
204 
205     va_start(ap, fmt);
206     vsnprintf(msg, sizeof(msg), fmt, ap);
207     va_end(ap);
208 
209     error_setg(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
210                bdrv_get_device_or_node_name(bs), "qcow2", msg);
211 }
212 
213 static void report_unsupported_feature(BlockDriverState *bs,
214     Error **errp, Qcow2Feature *table, uint64_t mask)
215 {
216     char *features = g_strdup("");
217     char *old;
218 
219     while (table && table->name[0] != '\0') {
220         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
221             if (mask & (1ULL << table->bit)) {
222                 old = features;
223                 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
224                                            table->name);
225                 g_free(old);
226                 mask &= ~(1ULL << table->bit);
227             }
228         }
229         table++;
230     }
231 
232     if (mask) {
233         old = features;
234         features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
235                                    old, *old ? ", " : "", mask);
236         g_free(old);
237     }
238 
239     report_unsupported(bs, errp, "%s", features);
240     g_free(features);
241 }
242 
243 /*
244  * Sets the dirty bit and flushes afterwards if necessary.
245  *
246  * The incompatible_features bit is only set if the image file header was
247  * updated successfully.  Therefore it is not required to check the return
248  * value of this function.
249  */
250 int qcow2_mark_dirty(BlockDriverState *bs)
251 {
252     BDRVQcow2State *s = bs->opaque;
253     uint64_t val;
254     int ret;
255 
256     assert(s->qcow_version >= 3);
257 
258     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
259         return 0; /* already dirty */
260     }
261 
262     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
263     ret = bdrv_pwrite(bs->file->bs, offsetof(QCowHeader, incompatible_features),
264                       &val, sizeof(val));
265     if (ret < 0) {
266         return ret;
267     }
268     ret = bdrv_flush(bs->file->bs);
269     if (ret < 0) {
270         return ret;
271     }
272 
273     /* Only treat image as dirty if the header was updated successfully */
274     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
275     return 0;
276 }
277 
278 /*
279  * Clears the dirty bit and flushes before if necessary.  Only call this
280  * function when there are no pending requests, it does not guard against
281  * concurrent requests dirtying the image.
282  */
283 static int qcow2_mark_clean(BlockDriverState *bs)
284 {
285     BDRVQcow2State *s = bs->opaque;
286 
287     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
288         int ret;
289 
290         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
291 
292         ret = bdrv_flush(bs);
293         if (ret < 0) {
294             return ret;
295         }
296 
297         return qcow2_update_header(bs);
298     }
299     return 0;
300 }
301 
302 /*
303  * Marks the image as corrupt.
304  */
305 int qcow2_mark_corrupt(BlockDriverState *bs)
306 {
307     BDRVQcow2State *s = bs->opaque;
308 
309     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
310     return qcow2_update_header(bs);
311 }
312 
313 /*
314  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
315  * before if necessary.
316  */
317 int qcow2_mark_consistent(BlockDriverState *bs)
318 {
319     BDRVQcow2State *s = bs->opaque;
320 
321     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
322         int ret = bdrv_flush(bs);
323         if (ret < 0) {
324             return ret;
325         }
326 
327         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
328         return qcow2_update_header(bs);
329     }
330     return 0;
331 }
332 
333 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
334                        BdrvCheckMode fix)
335 {
336     int ret = qcow2_check_refcounts(bs, result, fix);
337     if (ret < 0) {
338         return ret;
339     }
340 
341     if (fix && result->check_errors == 0 && result->corruptions == 0) {
342         ret = qcow2_mark_clean(bs);
343         if (ret < 0) {
344             return ret;
345         }
346         return qcow2_mark_consistent(bs);
347     }
348     return ret;
349 }
350 
351 static int validate_table_offset(BlockDriverState *bs, uint64_t offset,
352                                  uint64_t entries, size_t entry_len)
353 {
354     BDRVQcow2State *s = bs->opaque;
355     uint64_t size;
356 
357     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
358      * because values will be passed to qemu functions taking int64_t. */
359     if (entries > INT64_MAX / entry_len) {
360         return -EINVAL;
361     }
362 
363     size = entries * entry_len;
364 
365     if (INT64_MAX - size < offset) {
366         return -EINVAL;
367     }
368 
369     /* Tables must be cluster aligned */
370     if (offset & (s->cluster_size - 1)) {
371         return -EINVAL;
372     }
373 
374     return 0;
375 }
376 
377 static QemuOptsList qcow2_runtime_opts = {
378     .name = "qcow2",
379     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
380     .desc = {
381         {
382             .name = QCOW2_OPT_LAZY_REFCOUNTS,
383             .type = QEMU_OPT_BOOL,
384             .help = "Postpone refcount updates",
385         },
386         {
387             .name = QCOW2_OPT_DISCARD_REQUEST,
388             .type = QEMU_OPT_BOOL,
389             .help = "Pass guest discard requests to the layer below",
390         },
391         {
392             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
393             .type = QEMU_OPT_BOOL,
394             .help = "Generate discard requests when snapshot related space "
395                     "is freed",
396         },
397         {
398             .name = QCOW2_OPT_DISCARD_OTHER,
399             .type = QEMU_OPT_BOOL,
400             .help = "Generate discard requests when other clusters are freed",
401         },
402         {
403             .name = QCOW2_OPT_OVERLAP,
404             .type = QEMU_OPT_STRING,
405             .help = "Selects which overlap checks to perform from a range of "
406                     "templates (none, constant, cached, all)",
407         },
408         {
409             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
410             .type = QEMU_OPT_STRING,
411             .help = "Selects which overlap checks to perform from a range of "
412                     "templates (none, constant, cached, all)",
413         },
414         {
415             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
416             .type = QEMU_OPT_BOOL,
417             .help = "Check for unintended writes into the main qcow2 header",
418         },
419         {
420             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
421             .type = QEMU_OPT_BOOL,
422             .help = "Check for unintended writes into the active L1 table",
423         },
424         {
425             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
426             .type = QEMU_OPT_BOOL,
427             .help = "Check for unintended writes into an active L2 table",
428         },
429         {
430             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
431             .type = QEMU_OPT_BOOL,
432             .help = "Check for unintended writes into the refcount table",
433         },
434         {
435             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
436             .type = QEMU_OPT_BOOL,
437             .help = "Check for unintended writes into a refcount block",
438         },
439         {
440             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
441             .type = QEMU_OPT_BOOL,
442             .help = "Check for unintended writes into the snapshot table",
443         },
444         {
445             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
446             .type = QEMU_OPT_BOOL,
447             .help = "Check for unintended writes into an inactive L1 table",
448         },
449         {
450             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
451             .type = QEMU_OPT_BOOL,
452             .help = "Check for unintended writes into an inactive L2 table",
453         },
454         {
455             .name = QCOW2_OPT_CACHE_SIZE,
456             .type = QEMU_OPT_SIZE,
457             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
458                     "cache size",
459         },
460         {
461             .name = QCOW2_OPT_L2_CACHE_SIZE,
462             .type = QEMU_OPT_SIZE,
463             .help = "Maximum L2 table cache size",
464         },
465         {
466             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
467             .type = QEMU_OPT_SIZE,
468             .help = "Maximum refcount block cache size",
469         },
470         {
471             .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
472             .type = QEMU_OPT_NUMBER,
473             .help = "Clean unused cache entries after this time (in seconds)",
474         },
475         { /* end of list */ }
476     },
477 };
478 
479 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
480     [QCOW2_OL_MAIN_HEADER_BITNR]    = QCOW2_OPT_OVERLAP_MAIN_HEADER,
481     [QCOW2_OL_ACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L1,
482     [QCOW2_OL_ACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L2,
483     [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
484     [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
485     [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
486     [QCOW2_OL_INACTIVE_L1_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L1,
487     [QCOW2_OL_INACTIVE_L2_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L2,
488 };
489 
490 static void cache_clean_timer_cb(void *opaque)
491 {
492     BlockDriverState *bs = opaque;
493     BDRVQcow2State *s = bs->opaque;
494     qcow2_cache_clean_unused(bs, s->l2_table_cache);
495     qcow2_cache_clean_unused(bs, s->refcount_block_cache);
496     timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
497               (int64_t) s->cache_clean_interval * 1000);
498 }
499 
500 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
501 {
502     BDRVQcow2State *s = bs->opaque;
503     if (s->cache_clean_interval > 0) {
504         s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
505                                              SCALE_MS, cache_clean_timer_cb,
506                                              bs);
507         timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
508                   (int64_t) s->cache_clean_interval * 1000);
509     }
510 }
511 
512 static void cache_clean_timer_del(BlockDriverState *bs)
513 {
514     BDRVQcow2State *s = bs->opaque;
515     if (s->cache_clean_timer) {
516         timer_del(s->cache_clean_timer);
517         timer_free(s->cache_clean_timer);
518         s->cache_clean_timer = NULL;
519     }
520 }
521 
522 static void qcow2_detach_aio_context(BlockDriverState *bs)
523 {
524     cache_clean_timer_del(bs);
525 }
526 
527 static void qcow2_attach_aio_context(BlockDriverState *bs,
528                                      AioContext *new_context)
529 {
530     cache_clean_timer_init(bs, new_context);
531 }
532 
533 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
534                              uint64_t *l2_cache_size,
535                              uint64_t *refcount_cache_size, Error **errp)
536 {
537     BDRVQcow2State *s = bs->opaque;
538     uint64_t combined_cache_size;
539     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
540 
541     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
542     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
543     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
544 
545     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
546     *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0);
547     *refcount_cache_size = qemu_opt_get_size(opts,
548                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
549 
550     if (combined_cache_size_set) {
551         if (l2_cache_size_set && refcount_cache_size_set) {
552             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
553                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
554                        "the same time");
555             return;
556         } else if (*l2_cache_size > combined_cache_size) {
557             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
558                        QCOW2_OPT_CACHE_SIZE);
559             return;
560         } else if (*refcount_cache_size > combined_cache_size) {
561             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
562                        QCOW2_OPT_CACHE_SIZE);
563             return;
564         }
565 
566         if (l2_cache_size_set) {
567             *refcount_cache_size = combined_cache_size - *l2_cache_size;
568         } else if (refcount_cache_size_set) {
569             *l2_cache_size = combined_cache_size - *refcount_cache_size;
570         } else {
571             *refcount_cache_size = combined_cache_size
572                                  / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1);
573             *l2_cache_size = combined_cache_size - *refcount_cache_size;
574         }
575     } else {
576         if (!l2_cache_size_set && !refcount_cache_size_set) {
577             *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE,
578                                  (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
579                                  * s->cluster_size);
580             *refcount_cache_size = *l2_cache_size
581                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
582         } else if (!l2_cache_size_set) {
583             *l2_cache_size = *refcount_cache_size
584                            * DEFAULT_L2_REFCOUNT_SIZE_RATIO;
585         } else if (!refcount_cache_size_set) {
586             *refcount_cache_size = *l2_cache_size
587                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
588         }
589     }
590 }
591 
592 typedef struct Qcow2ReopenState {
593     Qcow2Cache *l2_table_cache;
594     Qcow2Cache *refcount_block_cache;
595     bool use_lazy_refcounts;
596     int overlap_check;
597     bool discard_passthrough[QCOW2_DISCARD_MAX];
598     uint64_t cache_clean_interval;
599 } Qcow2ReopenState;
600 
601 static int qcow2_update_options_prepare(BlockDriverState *bs,
602                                         Qcow2ReopenState *r,
603                                         QDict *options, int flags,
604                                         Error **errp)
605 {
606     BDRVQcow2State *s = bs->opaque;
607     QemuOpts *opts = NULL;
608     const char *opt_overlap_check, *opt_overlap_check_template;
609     int overlap_check_template = 0;
610     uint64_t l2_cache_size, refcount_cache_size;
611     int i;
612     Error *local_err = NULL;
613     int ret;
614 
615     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
616     qemu_opts_absorb_qdict(opts, options, &local_err);
617     if (local_err) {
618         error_propagate(errp, local_err);
619         ret = -EINVAL;
620         goto fail;
621     }
622 
623     /* get L2 table/refcount block cache size from command line options */
624     read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size,
625                      &local_err);
626     if (local_err) {
627         error_propagate(errp, local_err);
628         ret = -EINVAL;
629         goto fail;
630     }
631 
632     l2_cache_size /= s->cluster_size;
633     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
634         l2_cache_size = MIN_L2_CACHE_SIZE;
635     }
636     if (l2_cache_size > INT_MAX) {
637         error_setg(errp, "L2 cache size too big");
638         ret = -EINVAL;
639         goto fail;
640     }
641 
642     refcount_cache_size /= s->cluster_size;
643     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
644         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
645     }
646     if (refcount_cache_size > INT_MAX) {
647         error_setg(errp, "Refcount cache size too big");
648         ret = -EINVAL;
649         goto fail;
650     }
651 
652     /* alloc new L2 table/refcount block cache, flush old one */
653     if (s->l2_table_cache) {
654         ret = qcow2_cache_flush(bs, s->l2_table_cache);
655         if (ret) {
656             error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
657             goto fail;
658         }
659     }
660 
661     if (s->refcount_block_cache) {
662         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
663         if (ret) {
664             error_setg_errno(errp, -ret,
665                              "Failed to flush the refcount block cache");
666             goto fail;
667         }
668     }
669 
670     r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size);
671     r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size);
672     if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
673         error_setg(errp, "Could not allocate metadata caches");
674         ret = -ENOMEM;
675         goto fail;
676     }
677 
678     /* New interval for cache cleanup timer */
679     r->cache_clean_interval =
680         qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
681                             s->cache_clean_interval);
682     if (r->cache_clean_interval > UINT_MAX) {
683         error_setg(errp, "Cache clean interval too big");
684         ret = -EINVAL;
685         goto fail;
686     }
687 
688     /* lazy-refcounts; flush if going from enabled to disabled */
689     r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
690         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
691     if (r->use_lazy_refcounts && s->qcow_version < 3) {
692         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
693                    "qemu 1.1 compatibility level");
694         ret = -EINVAL;
695         goto fail;
696     }
697 
698     if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
699         ret = qcow2_mark_clean(bs);
700         if (ret < 0) {
701             error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
702             goto fail;
703         }
704     }
705 
706     /* Overlap check options */
707     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
708     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
709     if (opt_overlap_check_template && opt_overlap_check &&
710         strcmp(opt_overlap_check_template, opt_overlap_check))
711     {
712         error_setg(errp, "Conflicting values for qcow2 options '"
713                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
714                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
715         ret = -EINVAL;
716         goto fail;
717     }
718     if (!opt_overlap_check) {
719         opt_overlap_check = opt_overlap_check_template ?: "cached";
720     }
721 
722     if (!strcmp(opt_overlap_check, "none")) {
723         overlap_check_template = 0;
724     } else if (!strcmp(opt_overlap_check, "constant")) {
725         overlap_check_template = QCOW2_OL_CONSTANT;
726     } else if (!strcmp(opt_overlap_check, "cached")) {
727         overlap_check_template = QCOW2_OL_CACHED;
728     } else if (!strcmp(opt_overlap_check, "all")) {
729         overlap_check_template = QCOW2_OL_ALL;
730     } else {
731         error_setg(errp, "Unsupported value '%s' for qcow2 option "
732                    "'overlap-check'. Allowed are any of the following: "
733                    "none, constant, cached, all", opt_overlap_check);
734         ret = -EINVAL;
735         goto fail;
736     }
737 
738     r->overlap_check = 0;
739     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
740         /* overlap-check defines a template bitmask, but every flag may be
741          * overwritten through the associated boolean option */
742         r->overlap_check |=
743             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
744                               overlap_check_template & (1 << i)) << i;
745     }
746 
747     r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
748     r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
749     r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
750         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
751                           flags & BDRV_O_UNMAP);
752     r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
753         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
754     r->discard_passthrough[QCOW2_DISCARD_OTHER] =
755         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
756 
757     ret = 0;
758 fail:
759     qemu_opts_del(opts);
760     opts = NULL;
761     return ret;
762 }
763 
764 static void qcow2_update_options_commit(BlockDriverState *bs,
765                                         Qcow2ReopenState *r)
766 {
767     BDRVQcow2State *s = bs->opaque;
768     int i;
769 
770     if (s->l2_table_cache) {
771         qcow2_cache_destroy(bs, s->l2_table_cache);
772     }
773     if (s->refcount_block_cache) {
774         qcow2_cache_destroy(bs, s->refcount_block_cache);
775     }
776     s->l2_table_cache = r->l2_table_cache;
777     s->refcount_block_cache = r->refcount_block_cache;
778 
779     s->overlap_check = r->overlap_check;
780     s->use_lazy_refcounts = r->use_lazy_refcounts;
781 
782     for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
783         s->discard_passthrough[i] = r->discard_passthrough[i];
784     }
785 
786     if (s->cache_clean_interval != r->cache_clean_interval) {
787         cache_clean_timer_del(bs);
788         s->cache_clean_interval = r->cache_clean_interval;
789         cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
790     }
791 }
792 
793 static void qcow2_update_options_abort(BlockDriverState *bs,
794                                        Qcow2ReopenState *r)
795 {
796     if (r->l2_table_cache) {
797         qcow2_cache_destroy(bs, r->l2_table_cache);
798     }
799     if (r->refcount_block_cache) {
800         qcow2_cache_destroy(bs, r->refcount_block_cache);
801     }
802 }
803 
804 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
805                                 int flags, Error **errp)
806 {
807     Qcow2ReopenState r = {};
808     int ret;
809 
810     ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
811     if (ret >= 0) {
812         qcow2_update_options_commit(bs, &r);
813     } else {
814         qcow2_update_options_abort(bs, &r);
815     }
816 
817     return ret;
818 }
819 
820 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
821                       Error **errp)
822 {
823     BDRVQcow2State *s = bs->opaque;
824     unsigned int len, i;
825     int ret = 0;
826     QCowHeader header;
827     Error *local_err = NULL;
828     uint64_t ext_end;
829     uint64_t l1_vm_state_index;
830 
831     ret = bdrv_pread(bs->file->bs, 0, &header, sizeof(header));
832     if (ret < 0) {
833         error_setg_errno(errp, -ret, "Could not read qcow2 header");
834         goto fail;
835     }
836     be32_to_cpus(&header.magic);
837     be32_to_cpus(&header.version);
838     be64_to_cpus(&header.backing_file_offset);
839     be32_to_cpus(&header.backing_file_size);
840     be64_to_cpus(&header.size);
841     be32_to_cpus(&header.cluster_bits);
842     be32_to_cpus(&header.crypt_method);
843     be64_to_cpus(&header.l1_table_offset);
844     be32_to_cpus(&header.l1_size);
845     be64_to_cpus(&header.refcount_table_offset);
846     be32_to_cpus(&header.refcount_table_clusters);
847     be64_to_cpus(&header.snapshots_offset);
848     be32_to_cpus(&header.nb_snapshots);
849 
850     if (header.magic != QCOW_MAGIC) {
851         error_setg(errp, "Image is not in qcow2 format");
852         ret = -EINVAL;
853         goto fail;
854     }
855     if (header.version < 2 || header.version > 3) {
856         report_unsupported(bs, errp, "QCOW version %" PRIu32, header.version);
857         ret = -ENOTSUP;
858         goto fail;
859     }
860 
861     s->qcow_version = header.version;
862 
863     /* Initialise cluster size */
864     if (header.cluster_bits < MIN_CLUSTER_BITS ||
865         header.cluster_bits > MAX_CLUSTER_BITS) {
866         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
867                    header.cluster_bits);
868         ret = -EINVAL;
869         goto fail;
870     }
871 
872     s->cluster_bits = header.cluster_bits;
873     s->cluster_size = 1 << s->cluster_bits;
874     s->cluster_sectors = 1 << (s->cluster_bits - 9);
875 
876     /* Initialise version 3 header fields */
877     if (header.version == 2) {
878         header.incompatible_features    = 0;
879         header.compatible_features      = 0;
880         header.autoclear_features       = 0;
881         header.refcount_order           = 4;
882         header.header_length            = 72;
883     } else {
884         be64_to_cpus(&header.incompatible_features);
885         be64_to_cpus(&header.compatible_features);
886         be64_to_cpus(&header.autoclear_features);
887         be32_to_cpus(&header.refcount_order);
888         be32_to_cpus(&header.header_length);
889 
890         if (header.header_length < 104) {
891             error_setg(errp, "qcow2 header too short");
892             ret = -EINVAL;
893             goto fail;
894         }
895     }
896 
897     if (header.header_length > s->cluster_size) {
898         error_setg(errp, "qcow2 header exceeds cluster size");
899         ret = -EINVAL;
900         goto fail;
901     }
902 
903     if (header.header_length > sizeof(header)) {
904         s->unknown_header_fields_size = header.header_length - sizeof(header);
905         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
906         ret = bdrv_pread(bs->file->bs, sizeof(header), s->unknown_header_fields,
907                          s->unknown_header_fields_size);
908         if (ret < 0) {
909             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
910                              "fields");
911             goto fail;
912         }
913     }
914 
915     if (header.backing_file_offset > s->cluster_size) {
916         error_setg(errp, "Invalid backing file offset");
917         ret = -EINVAL;
918         goto fail;
919     }
920 
921     if (header.backing_file_offset) {
922         ext_end = header.backing_file_offset;
923     } else {
924         ext_end = 1 << header.cluster_bits;
925     }
926 
927     /* Handle feature bits */
928     s->incompatible_features    = header.incompatible_features;
929     s->compatible_features      = header.compatible_features;
930     s->autoclear_features       = header.autoclear_features;
931 
932     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
933         void *feature_table = NULL;
934         qcow2_read_extensions(bs, header.header_length, ext_end,
935                               &feature_table, NULL);
936         report_unsupported_feature(bs, errp, feature_table,
937                                    s->incompatible_features &
938                                    ~QCOW2_INCOMPAT_MASK);
939         ret = -ENOTSUP;
940         g_free(feature_table);
941         goto fail;
942     }
943 
944     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
945         /* Corrupt images may not be written to unless they are being repaired
946          */
947         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
948             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
949                        "read/write");
950             ret = -EACCES;
951             goto fail;
952         }
953     }
954 
955     /* Check support for various header values */
956     if (header.refcount_order > 6) {
957         error_setg(errp, "Reference count entry width too large; may not "
958                    "exceed 64 bits");
959         ret = -EINVAL;
960         goto fail;
961     }
962     s->refcount_order = header.refcount_order;
963     s->refcount_bits = 1 << s->refcount_order;
964     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
965     s->refcount_max += s->refcount_max - 1;
966 
967     if (header.crypt_method > QCOW_CRYPT_AES) {
968         error_setg(errp, "Unsupported encryption method: %" PRIu32,
969                    header.crypt_method);
970         ret = -EINVAL;
971         goto fail;
972     }
973     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) {
974         error_setg(errp, "AES cipher not available");
975         ret = -EINVAL;
976         goto fail;
977     }
978     s->crypt_method_header = header.crypt_method;
979     if (s->crypt_method_header) {
980         bs->encrypted = 1;
981     }
982 
983     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
984     s->l2_size = 1 << s->l2_bits;
985     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
986     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
987     s->refcount_block_size = 1 << s->refcount_block_bits;
988     bs->total_sectors = header.size / 512;
989     s->csize_shift = (62 - (s->cluster_bits - 8));
990     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
991     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
992 
993     s->refcount_table_offset = header.refcount_table_offset;
994     s->refcount_table_size =
995         header.refcount_table_clusters << (s->cluster_bits - 3);
996 
997     if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
998         error_setg(errp, "Reference count table too large");
999         ret = -EINVAL;
1000         goto fail;
1001     }
1002 
1003     ret = validate_table_offset(bs, s->refcount_table_offset,
1004                                 s->refcount_table_size, sizeof(uint64_t));
1005     if (ret < 0) {
1006         error_setg(errp, "Invalid reference count table offset");
1007         goto fail;
1008     }
1009 
1010     /* Snapshot table offset/length */
1011     if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
1012         error_setg(errp, "Too many snapshots");
1013         ret = -EINVAL;
1014         goto fail;
1015     }
1016 
1017     ret = validate_table_offset(bs, header.snapshots_offset,
1018                                 header.nb_snapshots,
1019                                 sizeof(QCowSnapshotHeader));
1020     if (ret < 0) {
1021         error_setg(errp, "Invalid snapshot table offset");
1022         goto fail;
1023     }
1024 
1025     /* read the level 1 table */
1026     if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
1027         error_setg(errp, "Active L1 table too large");
1028         ret = -EFBIG;
1029         goto fail;
1030     }
1031     s->l1_size = header.l1_size;
1032 
1033     l1_vm_state_index = size_to_l1(s, header.size);
1034     if (l1_vm_state_index > INT_MAX) {
1035         error_setg(errp, "Image is too big");
1036         ret = -EFBIG;
1037         goto fail;
1038     }
1039     s->l1_vm_state_index = l1_vm_state_index;
1040 
1041     /* the L1 table must contain at least enough entries to put
1042        header.size bytes */
1043     if (s->l1_size < s->l1_vm_state_index) {
1044         error_setg(errp, "L1 table is too small");
1045         ret = -EINVAL;
1046         goto fail;
1047     }
1048 
1049     ret = validate_table_offset(bs, header.l1_table_offset,
1050                                 header.l1_size, sizeof(uint64_t));
1051     if (ret < 0) {
1052         error_setg(errp, "Invalid L1 table offset");
1053         goto fail;
1054     }
1055     s->l1_table_offset = header.l1_table_offset;
1056 
1057 
1058     if (s->l1_size > 0) {
1059         s->l1_table = qemu_try_blockalign(bs->file->bs,
1060             align_offset(s->l1_size * sizeof(uint64_t), 512));
1061         if (s->l1_table == NULL) {
1062             error_setg(errp, "Could not allocate L1 table");
1063             ret = -ENOMEM;
1064             goto fail;
1065         }
1066         ret = bdrv_pread(bs->file->bs, s->l1_table_offset, s->l1_table,
1067                          s->l1_size * sizeof(uint64_t));
1068         if (ret < 0) {
1069             error_setg_errno(errp, -ret, "Could not read L1 table");
1070             goto fail;
1071         }
1072         for(i = 0;i < s->l1_size; i++) {
1073             be64_to_cpus(&s->l1_table[i]);
1074         }
1075     }
1076 
1077     /* Parse driver-specific options */
1078     ret = qcow2_update_options(bs, options, flags, errp);
1079     if (ret < 0) {
1080         goto fail;
1081     }
1082 
1083     s->cluster_cache = g_malloc(s->cluster_size);
1084     /* one more sector for decompressed data alignment */
1085     s->cluster_data = qemu_try_blockalign(bs->file->bs, QCOW_MAX_CRYPT_CLUSTERS
1086                                                     * s->cluster_size + 512);
1087     if (s->cluster_data == NULL) {
1088         error_setg(errp, "Could not allocate temporary cluster buffer");
1089         ret = -ENOMEM;
1090         goto fail;
1091     }
1092 
1093     s->cluster_cache_offset = -1;
1094     s->flags = flags;
1095 
1096     ret = qcow2_refcount_init(bs);
1097     if (ret != 0) {
1098         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1099         goto fail;
1100     }
1101 
1102     QLIST_INIT(&s->cluster_allocs);
1103     QTAILQ_INIT(&s->discards);
1104 
1105     /* read qcow2 extensions */
1106     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1107         &local_err)) {
1108         error_propagate(errp, local_err);
1109         ret = -EINVAL;
1110         goto fail;
1111     }
1112 
1113     /* read the backing file name */
1114     if (header.backing_file_offset != 0) {
1115         len = header.backing_file_size;
1116         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1117             len >= sizeof(bs->backing_file)) {
1118             error_setg(errp, "Backing file name too long");
1119             ret = -EINVAL;
1120             goto fail;
1121         }
1122         ret = bdrv_pread(bs->file->bs, header.backing_file_offset,
1123                          bs->backing_file, len);
1124         if (ret < 0) {
1125             error_setg_errno(errp, -ret, "Could not read backing file name");
1126             goto fail;
1127         }
1128         bs->backing_file[len] = '\0';
1129         s->image_backing_file = g_strdup(bs->backing_file);
1130     }
1131 
1132     /* Internal snapshots */
1133     s->snapshots_offset = header.snapshots_offset;
1134     s->nb_snapshots = header.nb_snapshots;
1135 
1136     ret = qcow2_read_snapshots(bs);
1137     if (ret < 0) {
1138         error_setg_errno(errp, -ret, "Could not read snapshots");
1139         goto fail;
1140     }
1141 
1142     /* Clear unknown autoclear feature bits */
1143     if (!bs->read_only && !(flags & BDRV_O_INCOMING) && s->autoclear_features) {
1144         s->autoclear_features = 0;
1145         ret = qcow2_update_header(bs);
1146         if (ret < 0) {
1147             error_setg_errno(errp, -ret, "Could not update qcow2 header");
1148             goto fail;
1149         }
1150     }
1151 
1152     /* Initialise locks */
1153     qemu_co_mutex_init(&s->lock);
1154 
1155     /* Repair image if dirty */
1156     if (!(flags & (BDRV_O_CHECK | BDRV_O_INCOMING)) && !bs->read_only &&
1157         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1158         BdrvCheckResult result = {0};
1159 
1160         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1161         if (ret < 0) {
1162             error_setg_errno(errp, -ret, "Could not repair dirty image");
1163             goto fail;
1164         }
1165     }
1166 
1167 #ifdef DEBUG_ALLOC
1168     {
1169         BdrvCheckResult result = {0};
1170         qcow2_check_refcounts(bs, &result, 0);
1171     }
1172 #endif
1173     return ret;
1174 
1175  fail:
1176     g_free(s->unknown_header_fields);
1177     cleanup_unknown_header_ext(bs);
1178     qcow2_free_snapshots(bs);
1179     qcow2_refcount_close(bs);
1180     qemu_vfree(s->l1_table);
1181     /* else pre-write overlap checks in cache_destroy may crash */
1182     s->l1_table = NULL;
1183     cache_clean_timer_del(bs);
1184     if (s->l2_table_cache) {
1185         qcow2_cache_destroy(bs, s->l2_table_cache);
1186     }
1187     if (s->refcount_block_cache) {
1188         qcow2_cache_destroy(bs, s->refcount_block_cache);
1189     }
1190     g_free(s->cluster_cache);
1191     qemu_vfree(s->cluster_data);
1192     return ret;
1193 }
1194 
1195 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1196 {
1197     BDRVQcow2State *s = bs->opaque;
1198 
1199     bs->bl.write_zeroes_alignment = s->cluster_sectors;
1200 }
1201 
1202 static int qcow2_set_key(BlockDriverState *bs, const char *key)
1203 {
1204     BDRVQcow2State *s = bs->opaque;
1205     uint8_t keybuf[16];
1206     int len, i;
1207     Error *err = NULL;
1208 
1209     memset(keybuf, 0, 16);
1210     len = strlen(key);
1211     if (len > 16)
1212         len = 16;
1213     /* XXX: we could compress the chars to 7 bits to increase
1214        entropy */
1215     for(i = 0;i < len;i++) {
1216         keybuf[i] = key[i];
1217     }
1218     assert(bs->encrypted);
1219 
1220     qcrypto_cipher_free(s->cipher);
1221     s->cipher = qcrypto_cipher_new(
1222         QCRYPTO_CIPHER_ALG_AES_128,
1223         QCRYPTO_CIPHER_MODE_CBC,
1224         keybuf, G_N_ELEMENTS(keybuf),
1225         &err);
1226 
1227     if (!s->cipher) {
1228         /* XXX would be nice if errors in this method could
1229          * be properly propagate to the caller. Would need
1230          * the bdrv_set_key() API signature to be fixed. */
1231         error_free(err);
1232         return -1;
1233     }
1234     return 0;
1235 }
1236 
1237 static int qcow2_reopen_prepare(BDRVReopenState *state,
1238                                 BlockReopenQueue *queue, Error **errp)
1239 {
1240     Qcow2ReopenState *r;
1241     int ret;
1242 
1243     r = g_new0(Qcow2ReopenState, 1);
1244     state->opaque = r;
1245 
1246     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1247                                        state->flags, errp);
1248     if (ret < 0) {
1249         goto fail;
1250     }
1251 
1252     /* We need to write out any unwritten data if we reopen read-only. */
1253     if ((state->flags & BDRV_O_RDWR) == 0) {
1254         ret = bdrv_flush(state->bs);
1255         if (ret < 0) {
1256             goto fail;
1257         }
1258 
1259         ret = qcow2_mark_clean(state->bs);
1260         if (ret < 0) {
1261             goto fail;
1262         }
1263     }
1264 
1265     return 0;
1266 
1267 fail:
1268     qcow2_update_options_abort(state->bs, r);
1269     g_free(r);
1270     return ret;
1271 }
1272 
1273 static void qcow2_reopen_commit(BDRVReopenState *state)
1274 {
1275     qcow2_update_options_commit(state->bs, state->opaque);
1276     g_free(state->opaque);
1277 }
1278 
1279 static void qcow2_reopen_abort(BDRVReopenState *state)
1280 {
1281     qcow2_update_options_abort(state->bs, state->opaque);
1282     g_free(state->opaque);
1283 }
1284 
1285 static void qcow2_join_options(QDict *options, QDict *old_options)
1286 {
1287     bool has_new_overlap_template =
1288         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1289         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1290     bool has_new_total_cache_size =
1291         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1292     bool has_all_cache_options;
1293 
1294     /* New overlap template overrides all old overlap options */
1295     if (has_new_overlap_template) {
1296         qdict_del(old_options, QCOW2_OPT_OVERLAP);
1297         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1298         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1299         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1300         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1301         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1302         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1303         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1304         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1305         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1306     }
1307 
1308     /* New total cache size overrides all old options */
1309     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1310         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1311         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1312     }
1313 
1314     qdict_join(options, old_options, false);
1315 
1316     /*
1317      * If after merging all cache size options are set, an old total size is
1318      * overwritten. Do keep all options, however, if all three are new. The
1319      * resulting error message is what we want to happen.
1320      */
1321     has_all_cache_options =
1322         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1323         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1324         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1325 
1326     if (has_all_cache_options && !has_new_total_cache_size) {
1327         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1328     }
1329 }
1330 
1331 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1332         int64_t sector_num, int nb_sectors, int *pnum)
1333 {
1334     BDRVQcow2State *s = bs->opaque;
1335     uint64_t cluster_offset;
1336     int index_in_cluster, ret;
1337     int64_t status = 0;
1338 
1339     *pnum = nb_sectors;
1340     qemu_co_mutex_lock(&s->lock);
1341     ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
1342     qemu_co_mutex_unlock(&s->lock);
1343     if (ret < 0) {
1344         return ret;
1345     }
1346 
1347     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1348         !s->cipher) {
1349         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1350         cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1351         status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1352     }
1353     if (ret == QCOW2_CLUSTER_ZERO) {
1354         status |= BDRV_BLOCK_ZERO;
1355     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1356         status |= BDRV_BLOCK_DATA;
1357     }
1358     return status;
1359 }
1360 
1361 /* handle reading after the end of the backing file */
1362 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
1363                   int64_t sector_num, int nb_sectors)
1364 {
1365     int n1;
1366     if ((sector_num + nb_sectors) <= bs->total_sectors)
1367         return nb_sectors;
1368     if (sector_num >= bs->total_sectors)
1369         n1 = 0;
1370     else
1371         n1 = bs->total_sectors - sector_num;
1372 
1373     qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1));
1374 
1375     return n1;
1376 }
1377 
1378 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
1379                           int remaining_sectors, QEMUIOVector *qiov)
1380 {
1381     BDRVQcow2State *s = bs->opaque;
1382     int index_in_cluster, n1;
1383     int ret;
1384     int cur_nr_sectors; /* number of sectors in current iteration */
1385     uint64_t cluster_offset = 0;
1386     uint64_t bytes_done = 0;
1387     QEMUIOVector hd_qiov;
1388     uint8_t *cluster_data = NULL;
1389 
1390     qemu_iovec_init(&hd_qiov, qiov->niov);
1391 
1392     qemu_co_mutex_lock(&s->lock);
1393 
1394     while (remaining_sectors != 0) {
1395 
1396         /* prepare next request */
1397         cur_nr_sectors = remaining_sectors;
1398         if (s->cipher) {
1399             cur_nr_sectors = MIN(cur_nr_sectors,
1400                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1401         }
1402 
1403         ret = qcow2_get_cluster_offset(bs, sector_num << 9,
1404             &cur_nr_sectors, &cluster_offset);
1405         if (ret < 0) {
1406             goto fail;
1407         }
1408 
1409         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1410 
1411         qemu_iovec_reset(&hd_qiov);
1412         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1413             cur_nr_sectors * 512);
1414 
1415         switch (ret) {
1416         case QCOW2_CLUSTER_UNALLOCATED:
1417 
1418             if (bs->backing) {
1419                 /* read from the base image */
1420                 n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov,
1421                     sector_num, cur_nr_sectors);
1422                 if (n1 > 0) {
1423                     QEMUIOVector local_qiov;
1424 
1425                     qemu_iovec_init(&local_qiov, hd_qiov.niov);
1426                     qemu_iovec_concat(&local_qiov, &hd_qiov, 0,
1427                                       n1 * BDRV_SECTOR_SIZE);
1428 
1429                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1430                     qemu_co_mutex_unlock(&s->lock);
1431                     ret = bdrv_co_readv(bs->backing->bs, sector_num,
1432                                         n1, &local_qiov);
1433                     qemu_co_mutex_lock(&s->lock);
1434 
1435                     qemu_iovec_destroy(&local_qiov);
1436 
1437                     if (ret < 0) {
1438                         goto fail;
1439                     }
1440                 }
1441             } else {
1442                 /* Note: in this case, no need to wait */
1443                 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1444             }
1445             break;
1446 
1447         case QCOW2_CLUSTER_ZERO:
1448             qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1449             break;
1450 
1451         case QCOW2_CLUSTER_COMPRESSED:
1452             /* add AIO support for compressed blocks ? */
1453             ret = qcow2_decompress_cluster(bs, cluster_offset);
1454             if (ret < 0) {
1455                 goto fail;
1456             }
1457 
1458             qemu_iovec_from_buf(&hd_qiov, 0,
1459                 s->cluster_cache + index_in_cluster * 512,
1460                 512 * cur_nr_sectors);
1461             break;
1462 
1463         case QCOW2_CLUSTER_NORMAL:
1464             if ((cluster_offset & 511) != 0) {
1465                 ret = -EIO;
1466                 goto fail;
1467             }
1468 
1469             if (bs->encrypted) {
1470                 assert(s->cipher);
1471 
1472                 /*
1473                  * For encrypted images, read everything into a temporary
1474                  * contiguous buffer on which the AES functions can work.
1475                  */
1476                 if (!cluster_data) {
1477                     cluster_data =
1478                         qemu_try_blockalign(bs->file->bs,
1479                                             QCOW_MAX_CRYPT_CLUSTERS
1480                                             * s->cluster_size);
1481                     if (cluster_data == NULL) {
1482                         ret = -ENOMEM;
1483                         goto fail;
1484                     }
1485                 }
1486 
1487                 assert(cur_nr_sectors <=
1488                     QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1489                 qemu_iovec_reset(&hd_qiov);
1490                 qemu_iovec_add(&hd_qiov, cluster_data,
1491                     512 * cur_nr_sectors);
1492             }
1493 
1494             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1495             qemu_co_mutex_unlock(&s->lock);
1496             ret = bdrv_co_readv(bs->file->bs,
1497                                 (cluster_offset >> 9) + index_in_cluster,
1498                                 cur_nr_sectors, &hd_qiov);
1499             qemu_co_mutex_lock(&s->lock);
1500             if (ret < 0) {
1501                 goto fail;
1502             }
1503             if (bs->encrypted) {
1504                 assert(s->cipher);
1505                 Error *err = NULL;
1506                 if (qcow2_encrypt_sectors(s, sector_num,  cluster_data,
1507                                           cluster_data, cur_nr_sectors, false,
1508                                           &err) < 0) {
1509                     error_free(err);
1510                     ret = -EIO;
1511                     goto fail;
1512                 }
1513                 qemu_iovec_from_buf(qiov, bytes_done,
1514                     cluster_data, 512 * cur_nr_sectors);
1515             }
1516             break;
1517 
1518         default:
1519             g_assert_not_reached();
1520             ret = -EIO;
1521             goto fail;
1522         }
1523 
1524         remaining_sectors -= cur_nr_sectors;
1525         sector_num += cur_nr_sectors;
1526         bytes_done += cur_nr_sectors * 512;
1527     }
1528     ret = 0;
1529 
1530 fail:
1531     qemu_co_mutex_unlock(&s->lock);
1532 
1533     qemu_iovec_destroy(&hd_qiov);
1534     qemu_vfree(cluster_data);
1535 
1536     return ret;
1537 }
1538 
1539 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
1540                            int64_t sector_num,
1541                            int remaining_sectors,
1542                            QEMUIOVector *qiov)
1543 {
1544     BDRVQcow2State *s = bs->opaque;
1545     int index_in_cluster;
1546     int ret;
1547     int cur_nr_sectors; /* number of sectors in current iteration */
1548     uint64_t cluster_offset;
1549     QEMUIOVector hd_qiov;
1550     uint64_t bytes_done = 0;
1551     uint8_t *cluster_data = NULL;
1552     QCowL2Meta *l2meta = NULL;
1553 
1554     trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
1555                                  remaining_sectors);
1556 
1557     qemu_iovec_init(&hd_qiov, qiov->niov);
1558 
1559     s->cluster_cache_offset = -1; /* disable compressed cache */
1560 
1561     qemu_co_mutex_lock(&s->lock);
1562 
1563     while (remaining_sectors != 0) {
1564 
1565         l2meta = NULL;
1566 
1567         trace_qcow2_writev_start_part(qemu_coroutine_self());
1568         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1569         cur_nr_sectors = remaining_sectors;
1570         if (bs->encrypted &&
1571             cur_nr_sectors >
1572             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster) {
1573             cur_nr_sectors =
1574                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster;
1575         }
1576 
1577         ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
1578             &cur_nr_sectors, &cluster_offset, &l2meta);
1579         if (ret < 0) {
1580             goto fail;
1581         }
1582 
1583         assert((cluster_offset & 511) == 0);
1584 
1585         qemu_iovec_reset(&hd_qiov);
1586         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1587             cur_nr_sectors * 512);
1588 
1589         if (bs->encrypted) {
1590             Error *err = NULL;
1591             assert(s->cipher);
1592             if (!cluster_data) {
1593                 cluster_data = qemu_try_blockalign(bs->file->bs,
1594                                                    QCOW_MAX_CRYPT_CLUSTERS
1595                                                    * s->cluster_size);
1596                 if (cluster_data == NULL) {
1597                     ret = -ENOMEM;
1598                     goto fail;
1599                 }
1600             }
1601 
1602             assert(hd_qiov.size <=
1603                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1604             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1605 
1606             if (qcow2_encrypt_sectors(s, sector_num, cluster_data,
1607                                       cluster_data, cur_nr_sectors,
1608                                       true, &err) < 0) {
1609                 error_free(err);
1610                 ret = -EIO;
1611                 goto fail;
1612             }
1613 
1614             qemu_iovec_reset(&hd_qiov);
1615             qemu_iovec_add(&hd_qiov, cluster_data,
1616                 cur_nr_sectors * 512);
1617         }
1618 
1619         ret = qcow2_pre_write_overlap_check(bs, 0,
1620                 cluster_offset + index_in_cluster * BDRV_SECTOR_SIZE,
1621                 cur_nr_sectors * BDRV_SECTOR_SIZE);
1622         if (ret < 0) {
1623             goto fail;
1624         }
1625 
1626         qemu_co_mutex_unlock(&s->lock);
1627         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1628         trace_qcow2_writev_data(qemu_coroutine_self(),
1629                                 (cluster_offset >> 9) + index_in_cluster);
1630         ret = bdrv_co_writev(bs->file->bs,
1631                              (cluster_offset >> 9) + index_in_cluster,
1632                              cur_nr_sectors, &hd_qiov);
1633         qemu_co_mutex_lock(&s->lock);
1634         if (ret < 0) {
1635             goto fail;
1636         }
1637 
1638         while (l2meta != NULL) {
1639             QCowL2Meta *next;
1640 
1641             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1642             if (ret < 0) {
1643                 goto fail;
1644             }
1645 
1646             /* Take the request off the list of running requests */
1647             if (l2meta->nb_clusters != 0) {
1648                 QLIST_REMOVE(l2meta, next_in_flight);
1649             }
1650 
1651             qemu_co_queue_restart_all(&l2meta->dependent_requests);
1652 
1653             next = l2meta->next;
1654             g_free(l2meta);
1655             l2meta = next;
1656         }
1657 
1658         remaining_sectors -= cur_nr_sectors;
1659         sector_num += cur_nr_sectors;
1660         bytes_done += cur_nr_sectors * 512;
1661         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
1662     }
1663     ret = 0;
1664 
1665 fail:
1666     qemu_co_mutex_unlock(&s->lock);
1667 
1668     while (l2meta != NULL) {
1669         QCowL2Meta *next;
1670 
1671         if (l2meta->nb_clusters != 0) {
1672             QLIST_REMOVE(l2meta, next_in_flight);
1673         }
1674         qemu_co_queue_restart_all(&l2meta->dependent_requests);
1675 
1676         next = l2meta->next;
1677         g_free(l2meta);
1678         l2meta = next;
1679     }
1680 
1681     qemu_iovec_destroy(&hd_qiov);
1682     qemu_vfree(cluster_data);
1683     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
1684 
1685     return ret;
1686 }
1687 
1688 static void qcow2_close(BlockDriverState *bs)
1689 {
1690     BDRVQcow2State *s = bs->opaque;
1691     qemu_vfree(s->l1_table);
1692     /* else pre-write overlap checks in cache_destroy may crash */
1693     s->l1_table = NULL;
1694 
1695     if (!(bs->open_flags & BDRV_O_INCOMING)) {
1696         int ret1, ret2;
1697 
1698         ret1 = qcow2_cache_flush(bs, s->l2_table_cache);
1699         ret2 = qcow2_cache_flush(bs, s->refcount_block_cache);
1700 
1701         if (ret1) {
1702             error_report("Failed to flush the L2 table cache: %s",
1703                          strerror(-ret1));
1704         }
1705         if (ret2) {
1706             error_report("Failed to flush the refcount block cache: %s",
1707                          strerror(-ret2));
1708         }
1709 
1710         if (!ret1 && !ret2) {
1711             qcow2_mark_clean(bs);
1712         }
1713     }
1714 
1715     cache_clean_timer_del(bs);
1716     qcow2_cache_destroy(bs, s->l2_table_cache);
1717     qcow2_cache_destroy(bs, s->refcount_block_cache);
1718 
1719     qcrypto_cipher_free(s->cipher);
1720     s->cipher = NULL;
1721 
1722     g_free(s->unknown_header_fields);
1723     cleanup_unknown_header_ext(bs);
1724 
1725     g_free(s->image_backing_file);
1726     g_free(s->image_backing_format);
1727 
1728     g_free(s->cluster_cache);
1729     qemu_vfree(s->cluster_data);
1730     qcow2_refcount_close(bs);
1731     qcow2_free_snapshots(bs);
1732 }
1733 
1734 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
1735 {
1736     BDRVQcow2State *s = bs->opaque;
1737     int flags = s->flags;
1738     QCryptoCipher *cipher = NULL;
1739     QDict *options;
1740     Error *local_err = NULL;
1741     int ret;
1742 
1743     /*
1744      * Backing files are read-only which makes all of their metadata immutable,
1745      * that means we don't have to worry about reopening them here.
1746      */
1747 
1748     cipher = s->cipher;
1749     s->cipher = NULL;
1750 
1751     qcow2_close(bs);
1752 
1753     bdrv_invalidate_cache(bs->file->bs, &local_err);
1754     if (local_err) {
1755         error_propagate(errp, local_err);
1756         return;
1757     }
1758 
1759     memset(s, 0, sizeof(BDRVQcow2State));
1760     options = qdict_clone_shallow(bs->options);
1761 
1762     ret = qcow2_open(bs, options, flags, &local_err);
1763     QDECREF(options);
1764     if (local_err) {
1765         error_setg(errp, "Could not reopen qcow2 layer: %s",
1766                    error_get_pretty(local_err));
1767         error_free(local_err);
1768         return;
1769     } else if (ret < 0) {
1770         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
1771         return;
1772     }
1773 
1774     s->cipher = cipher;
1775 }
1776 
1777 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
1778     size_t len, size_t buflen)
1779 {
1780     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
1781     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
1782 
1783     if (buflen < ext_len) {
1784         return -ENOSPC;
1785     }
1786 
1787     *ext_backing_fmt = (QCowExtension) {
1788         .magic  = cpu_to_be32(magic),
1789         .len    = cpu_to_be32(len),
1790     };
1791     memcpy(buf + sizeof(QCowExtension), s, len);
1792 
1793     return ext_len;
1794 }
1795 
1796 /*
1797  * Updates the qcow2 header, including the variable length parts of it, i.e.
1798  * the backing file name and all extensions. qcow2 was not designed to allow
1799  * such changes, so if we run out of space (we can only use the first cluster)
1800  * this function may fail.
1801  *
1802  * Returns 0 on success, -errno in error cases.
1803  */
1804 int qcow2_update_header(BlockDriverState *bs)
1805 {
1806     BDRVQcow2State *s = bs->opaque;
1807     QCowHeader *header;
1808     char *buf;
1809     size_t buflen = s->cluster_size;
1810     int ret;
1811     uint64_t total_size;
1812     uint32_t refcount_table_clusters;
1813     size_t header_length;
1814     Qcow2UnknownHeaderExtension *uext;
1815 
1816     buf = qemu_blockalign(bs, buflen);
1817 
1818     /* Header structure */
1819     header = (QCowHeader*) buf;
1820 
1821     if (buflen < sizeof(*header)) {
1822         ret = -ENOSPC;
1823         goto fail;
1824     }
1825 
1826     header_length = sizeof(*header) + s->unknown_header_fields_size;
1827     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1828     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1829 
1830     *header = (QCowHeader) {
1831         /* Version 2 fields */
1832         .magic                  = cpu_to_be32(QCOW_MAGIC),
1833         .version                = cpu_to_be32(s->qcow_version),
1834         .backing_file_offset    = 0,
1835         .backing_file_size      = 0,
1836         .cluster_bits           = cpu_to_be32(s->cluster_bits),
1837         .size                   = cpu_to_be64(total_size),
1838         .crypt_method           = cpu_to_be32(s->crypt_method_header),
1839         .l1_size                = cpu_to_be32(s->l1_size),
1840         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
1841         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
1842         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1843         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
1844         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
1845 
1846         /* Version 3 fields */
1847         .incompatible_features  = cpu_to_be64(s->incompatible_features),
1848         .compatible_features    = cpu_to_be64(s->compatible_features),
1849         .autoclear_features     = cpu_to_be64(s->autoclear_features),
1850         .refcount_order         = cpu_to_be32(s->refcount_order),
1851         .header_length          = cpu_to_be32(header_length),
1852     };
1853 
1854     /* For older versions, write a shorter header */
1855     switch (s->qcow_version) {
1856     case 2:
1857         ret = offsetof(QCowHeader, incompatible_features);
1858         break;
1859     case 3:
1860         ret = sizeof(*header);
1861         break;
1862     default:
1863         ret = -EINVAL;
1864         goto fail;
1865     }
1866 
1867     buf += ret;
1868     buflen -= ret;
1869     memset(buf, 0, buflen);
1870 
1871     /* Preserve any unknown field in the header */
1872     if (s->unknown_header_fields_size) {
1873         if (buflen < s->unknown_header_fields_size) {
1874             ret = -ENOSPC;
1875             goto fail;
1876         }
1877 
1878         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1879         buf += s->unknown_header_fields_size;
1880         buflen -= s->unknown_header_fields_size;
1881     }
1882 
1883     /* Backing file format header extension */
1884     if (s->image_backing_format) {
1885         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1886                              s->image_backing_format,
1887                              strlen(s->image_backing_format),
1888                              buflen);
1889         if (ret < 0) {
1890             goto fail;
1891         }
1892 
1893         buf += ret;
1894         buflen -= ret;
1895     }
1896 
1897     /* Feature table */
1898     Qcow2Feature features[] = {
1899         {
1900             .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1901             .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
1902             .name = "dirty bit",
1903         },
1904         {
1905             .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1906             .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
1907             .name = "corrupt bit",
1908         },
1909         {
1910             .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1911             .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1912             .name = "lazy refcounts",
1913         },
1914     };
1915 
1916     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1917                          features, sizeof(features), buflen);
1918     if (ret < 0) {
1919         goto fail;
1920     }
1921     buf += ret;
1922     buflen -= ret;
1923 
1924     /* Keep unknown header extensions */
1925     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1926         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1927         if (ret < 0) {
1928             goto fail;
1929         }
1930 
1931         buf += ret;
1932         buflen -= ret;
1933     }
1934 
1935     /* End of header extensions */
1936     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1937     if (ret < 0) {
1938         goto fail;
1939     }
1940 
1941     buf += ret;
1942     buflen -= ret;
1943 
1944     /* Backing file name */
1945     if (s->image_backing_file) {
1946         size_t backing_file_len = strlen(s->image_backing_file);
1947 
1948         if (buflen < backing_file_len) {
1949             ret = -ENOSPC;
1950             goto fail;
1951         }
1952 
1953         /* Using strncpy is ok here, since buf is not NUL-terminated. */
1954         strncpy(buf, s->image_backing_file, buflen);
1955 
1956         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1957         header->backing_file_size   = cpu_to_be32(backing_file_len);
1958     }
1959 
1960     /* Write the new header */
1961     ret = bdrv_pwrite(bs->file->bs, 0, header, s->cluster_size);
1962     if (ret < 0) {
1963         goto fail;
1964     }
1965 
1966     ret = 0;
1967 fail:
1968     qemu_vfree(header);
1969     return ret;
1970 }
1971 
1972 static int qcow2_change_backing_file(BlockDriverState *bs,
1973     const char *backing_file, const char *backing_fmt)
1974 {
1975     BDRVQcow2State *s = bs->opaque;
1976 
1977     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1978     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1979 
1980     g_free(s->image_backing_file);
1981     g_free(s->image_backing_format);
1982 
1983     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
1984     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
1985 
1986     return qcow2_update_header(bs);
1987 }
1988 
1989 static int preallocate(BlockDriverState *bs)
1990 {
1991     uint64_t nb_sectors;
1992     uint64_t offset;
1993     uint64_t host_offset = 0;
1994     int num;
1995     int ret;
1996     QCowL2Meta *meta;
1997 
1998     nb_sectors = bdrv_nb_sectors(bs);
1999     offset = 0;
2000 
2001     while (nb_sectors) {
2002         num = MIN(nb_sectors, INT_MAX >> BDRV_SECTOR_BITS);
2003         ret = qcow2_alloc_cluster_offset(bs, offset, &num,
2004                                          &host_offset, &meta);
2005         if (ret < 0) {
2006             return ret;
2007         }
2008 
2009         while (meta) {
2010             QCowL2Meta *next = meta->next;
2011 
2012             ret = qcow2_alloc_cluster_link_l2(bs, meta);
2013             if (ret < 0) {
2014                 qcow2_free_any_clusters(bs, meta->alloc_offset,
2015                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
2016                 return ret;
2017             }
2018 
2019             /* There are no dependent requests, but we need to remove our
2020              * request from the list of in-flight requests */
2021             QLIST_REMOVE(meta, next_in_flight);
2022 
2023             g_free(meta);
2024             meta = next;
2025         }
2026 
2027         /* TODO Preallocate data if requested */
2028 
2029         nb_sectors -= num;
2030         offset += num << BDRV_SECTOR_BITS;
2031     }
2032 
2033     /*
2034      * It is expected that the image file is large enough to actually contain
2035      * all of the allocated clusters (otherwise we get failing reads after
2036      * EOF). Extend the image to the last allocated sector.
2037      */
2038     if (host_offset != 0) {
2039         uint8_t buf[BDRV_SECTOR_SIZE];
2040         memset(buf, 0, BDRV_SECTOR_SIZE);
2041         ret = bdrv_write(bs->file->bs,
2042                          (host_offset >> BDRV_SECTOR_BITS) + num - 1,
2043                          buf, 1);
2044         if (ret < 0) {
2045             return ret;
2046         }
2047     }
2048 
2049     return 0;
2050 }
2051 
2052 static int qcow2_create2(const char *filename, int64_t total_size,
2053                          const char *backing_file, const char *backing_format,
2054                          int flags, size_t cluster_size, PreallocMode prealloc,
2055                          QemuOpts *opts, int version, int refcount_order,
2056                          Error **errp)
2057 {
2058     int cluster_bits;
2059     QDict *options;
2060 
2061     /* Calculate cluster_bits */
2062     cluster_bits = ctz32(cluster_size);
2063     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
2064         (1 << cluster_bits) != cluster_size)
2065     {
2066         error_setg(errp, "Cluster size must be a power of two between %d and "
2067                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
2068         return -EINVAL;
2069     }
2070 
2071     /*
2072      * Open the image file and write a minimal qcow2 header.
2073      *
2074      * We keep things simple and start with a zero-sized image. We also
2075      * do without refcount blocks or a L1 table for now. We'll fix the
2076      * inconsistency later.
2077      *
2078      * We do need a refcount table because growing the refcount table means
2079      * allocating two new refcount blocks - the seconds of which would be at
2080      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2081      * size for any qcow2 image.
2082      */
2083     BlockDriverState* bs;
2084     QCowHeader *header;
2085     uint64_t* refcount_table;
2086     Error *local_err = NULL;
2087     int ret;
2088 
2089     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
2090         /* Note: The following calculation does not need to be exact; if it is a
2091          * bit off, either some bytes will be "leaked" (which is fine) or we
2092          * will need to increase the file size by some bytes (which is fine,
2093          * too, as long as the bulk is allocated here). Therefore, using
2094          * floating point arithmetic is fine. */
2095         int64_t meta_size = 0;
2096         uint64_t nreftablee, nrefblocke, nl1e, nl2e;
2097         int64_t aligned_total_size = align_offset(total_size, cluster_size);
2098         int refblock_bits, refblock_size;
2099         /* refcount entry size in bytes */
2100         double rces = (1 << refcount_order) / 8.;
2101 
2102         /* see qcow2_open() */
2103         refblock_bits = cluster_bits - (refcount_order - 3);
2104         refblock_size = 1 << refblock_bits;
2105 
2106         /* header: 1 cluster */
2107         meta_size += cluster_size;
2108 
2109         /* total size of L2 tables */
2110         nl2e = aligned_total_size / cluster_size;
2111         nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
2112         meta_size += nl2e * sizeof(uint64_t);
2113 
2114         /* total size of L1 tables */
2115         nl1e = nl2e * sizeof(uint64_t) / cluster_size;
2116         nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
2117         meta_size += nl1e * sizeof(uint64_t);
2118 
2119         /* total size of refcount blocks
2120          *
2121          * note: every host cluster is reference-counted, including metadata
2122          * (even refcount blocks are recursively included).
2123          * Let:
2124          *   a = total_size (this is the guest disk size)
2125          *   m = meta size not including refcount blocks and refcount tables
2126          *   c = cluster size
2127          *   y1 = number of refcount blocks entries
2128          *   y2 = meta size including everything
2129          *   rces = refcount entry size in bytes
2130          * then,
2131          *   y1 = (y2 + a)/c
2132          *   y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m
2133          * we can get y1:
2134          *   y1 = (a + m) / (c - rces - rces * sizeof(u64) / c)
2135          */
2136         nrefblocke = (aligned_total_size + meta_size + cluster_size)
2137                    / (cluster_size - rces - rces * sizeof(uint64_t)
2138                                                  / cluster_size);
2139         meta_size += DIV_ROUND_UP(nrefblocke, refblock_size) * cluster_size;
2140 
2141         /* total size of refcount tables */
2142         nreftablee = nrefblocke / refblock_size;
2143         nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t));
2144         meta_size += nreftablee * sizeof(uint64_t);
2145 
2146         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
2147                             aligned_total_size + meta_size, &error_abort);
2148         qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc],
2149                      &error_abort);
2150     }
2151 
2152     ret = bdrv_create_file(filename, opts, &local_err);
2153     if (ret < 0) {
2154         error_propagate(errp, local_err);
2155         return ret;
2156     }
2157 
2158     bs = NULL;
2159     ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
2160                     &local_err);
2161     if (ret < 0) {
2162         error_propagate(errp, local_err);
2163         return ret;
2164     }
2165 
2166     /* Write the header */
2167     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
2168     header = g_malloc0(cluster_size);
2169     *header = (QCowHeader) {
2170         .magic                      = cpu_to_be32(QCOW_MAGIC),
2171         .version                    = cpu_to_be32(version),
2172         .cluster_bits               = cpu_to_be32(cluster_bits),
2173         .size                       = cpu_to_be64(0),
2174         .l1_table_offset            = cpu_to_be64(0),
2175         .l1_size                    = cpu_to_be32(0),
2176         .refcount_table_offset      = cpu_to_be64(cluster_size),
2177         .refcount_table_clusters    = cpu_to_be32(1),
2178         .refcount_order             = cpu_to_be32(refcount_order),
2179         .header_length              = cpu_to_be32(sizeof(*header)),
2180     };
2181 
2182     if (flags & BLOCK_FLAG_ENCRYPT) {
2183         header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
2184     } else {
2185         header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
2186     }
2187 
2188     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
2189         header->compatible_features |=
2190             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
2191     }
2192 
2193     ret = bdrv_pwrite(bs, 0, header, cluster_size);
2194     g_free(header);
2195     if (ret < 0) {
2196         error_setg_errno(errp, -ret, "Could not write qcow2 header");
2197         goto out;
2198     }
2199 
2200     /* Write a refcount table with one refcount block */
2201     refcount_table = g_malloc0(2 * cluster_size);
2202     refcount_table[0] = cpu_to_be64(2 * cluster_size);
2203     ret = bdrv_pwrite(bs, cluster_size, refcount_table, 2 * cluster_size);
2204     g_free(refcount_table);
2205 
2206     if (ret < 0) {
2207         error_setg_errno(errp, -ret, "Could not write refcount table");
2208         goto out;
2209     }
2210 
2211     bdrv_unref(bs);
2212     bs = NULL;
2213 
2214     /*
2215      * And now open the image and make it consistent first (i.e. increase the
2216      * refcount of the cluster that is occupied by the header and the refcount
2217      * table)
2218      */
2219     options = qdict_new();
2220     qdict_put(options, "driver", qstring_from_str("qcow2"));
2221     ret = bdrv_open(&bs, filename, NULL, options,
2222                     BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH,
2223                     &local_err);
2224     if (ret < 0) {
2225         error_propagate(errp, local_err);
2226         goto out;
2227     }
2228 
2229     ret = qcow2_alloc_clusters(bs, 3 * cluster_size);
2230     if (ret < 0) {
2231         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
2232                          "header and refcount table");
2233         goto out;
2234 
2235     } else if (ret != 0) {
2236         error_report("Huh, first cluster in empty image is already in use?");
2237         abort();
2238     }
2239 
2240     /* Okay, now that we have a valid image, let's give it the right size */
2241     ret = bdrv_truncate(bs, total_size);
2242     if (ret < 0) {
2243         error_setg_errno(errp, -ret, "Could not resize image");
2244         goto out;
2245     }
2246 
2247     /* Want a backing file? There you go.*/
2248     if (backing_file) {
2249         ret = bdrv_change_backing_file(bs, backing_file, backing_format);
2250         if (ret < 0) {
2251             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
2252                              "with format '%s'", backing_file, backing_format);
2253             goto out;
2254         }
2255     }
2256 
2257     /* And if we're supposed to preallocate metadata, do that now */
2258     if (prealloc != PREALLOC_MODE_OFF) {
2259         BDRVQcow2State *s = bs->opaque;
2260         qemu_co_mutex_lock(&s->lock);
2261         ret = preallocate(bs);
2262         qemu_co_mutex_unlock(&s->lock);
2263         if (ret < 0) {
2264             error_setg_errno(errp, -ret, "Could not preallocate metadata");
2265             goto out;
2266         }
2267     }
2268 
2269     bdrv_unref(bs);
2270     bs = NULL;
2271 
2272     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
2273     options = qdict_new();
2274     qdict_put(options, "driver", qstring_from_str("qcow2"));
2275     ret = bdrv_open(&bs, filename, NULL, options,
2276                     BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_BACKING,
2277                     &local_err);
2278     if (local_err) {
2279         error_propagate(errp, local_err);
2280         goto out;
2281     }
2282 
2283     ret = 0;
2284 out:
2285     if (bs) {
2286         bdrv_unref(bs);
2287     }
2288     return ret;
2289 }
2290 
2291 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
2292 {
2293     char *backing_file = NULL;
2294     char *backing_fmt = NULL;
2295     char *buf = NULL;
2296     uint64_t size = 0;
2297     int flags = 0;
2298     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
2299     PreallocMode prealloc;
2300     int version = 3;
2301     uint64_t refcount_bits = 16;
2302     int refcount_order;
2303     Error *local_err = NULL;
2304     int ret;
2305 
2306     /* Read out options */
2307     size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2308                     BDRV_SECTOR_SIZE);
2309     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2310     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2311     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2312         flags |= BLOCK_FLAG_ENCRYPT;
2313     }
2314     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2315                                          DEFAULT_CLUSTER_SIZE);
2316     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2317     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
2318                                PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
2319                                &local_err);
2320     if (local_err) {
2321         error_propagate(errp, local_err);
2322         ret = -EINVAL;
2323         goto finish;
2324     }
2325     g_free(buf);
2326     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2327     if (!buf) {
2328         /* keep the default */
2329     } else if (!strcmp(buf, "0.10")) {
2330         version = 2;
2331     } else if (!strcmp(buf, "1.1")) {
2332         version = 3;
2333     } else {
2334         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2335         ret = -EINVAL;
2336         goto finish;
2337     }
2338 
2339     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2340         flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2341     }
2342 
2343     if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2344         error_setg(errp, "Backing file and preallocation cannot be used at "
2345                    "the same time");
2346         ret = -EINVAL;
2347         goto finish;
2348     }
2349 
2350     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2351         error_setg(errp, "Lazy refcounts only supported with compatibility "
2352                    "level 1.1 and above (use compat=1.1 or greater)");
2353         ret = -EINVAL;
2354         goto finish;
2355     }
2356 
2357     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS,
2358                                             refcount_bits);
2359     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2360         error_setg(errp, "Refcount width must be a power of two and may not "
2361                    "exceed 64 bits");
2362         ret = -EINVAL;
2363         goto finish;
2364     }
2365 
2366     if (version < 3 && refcount_bits != 16) {
2367         error_setg(errp, "Different refcount widths than 16 bits require "
2368                    "compatibility level 1.1 or above (use compat=1.1 or "
2369                    "greater)");
2370         ret = -EINVAL;
2371         goto finish;
2372     }
2373 
2374     refcount_order = ctz32(refcount_bits);
2375 
2376     ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2377                         cluster_size, prealloc, opts, version, refcount_order,
2378                         &local_err);
2379     if (local_err) {
2380         error_propagate(errp, local_err);
2381     }
2382 
2383 finish:
2384     g_free(backing_file);
2385     g_free(backing_fmt);
2386     g_free(buf);
2387     return ret;
2388 }
2389 
2390 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
2391     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2392 {
2393     int ret;
2394     BDRVQcow2State *s = bs->opaque;
2395 
2396     /* Emulate misaligned zero writes */
2397     if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
2398         return -ENOTSUP;
2399     }
2400 
2401     /* Whatever is left can use real zero clusters */
2402     qemu_co_mutex_lock(&s->lock);
2403     ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2404         nb_sectors);
2405     qemu_co_mutex_unlock(&s->lock);
2406 
2407     return ret;
2408 }
2409 
2410 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
2411     int64_t sector_num, int nb_sectors)
2412 {
2413     int ret;
2414     BDRVQcow2State *s = bs->opaque;
2415 
2416     qemu_co_mutex_lock(&s->lock);
2417     ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2418         nb_sectors, QCOW2_DISCARD_REQUEST, false);
2419     qemu_co_mutex_unlock(&s->lock);
2420     return ret;
2421 }
2422 
2423 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
2424 {
2425     BDRVQcow2State *s = bs->opaque;
2426     int64_t new_l1_size;
2427     int ret;
2428 
2429     if (offset & 511) {
2430         error_report("The new size must be a multiple of 512");
2431         return -EINVAL;
2432     }
2433 
2434     /* cannot proceed if image has snapshots */
2435     if (s->nb_snapshots) {
2436         error_report("Can't resize an image which has snapshots");
2437         return -ENOTSUP;
2438     }
2439 
2440     /* shrinking is currently not supported */
2441     if (offset < bs->total_sectors * 512) {
2442         error_report("qcow2 doesn't support shrinking images yet");
2443         return -ENOTSUP;
2444     }
2445 
2446     new_l1_size = size_to_l1(s, offset);
2447     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
2448     if (ret < 0) {
2449         return ret;
2450     }
2451 
2452     /* write updated header.size */
2453     offset = cpu_to_be64(offset);
2454     ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, size),
2455                            &offset, sizeof(uint64_t));
2456     if (ret < 0) {
2457         return ret;
2458     }
2459 
2460     s->l1_vm_state_index = new_l1_size;
2461     return 0;
2462 }
2463 
2464 /* XXX: put compressed sectors first, then all the cluster aligned
2465    tables to avoid losing bytes in alignment */
2466 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
2467                                   const uint8_t *buf, int nb_sectors)
2468 {
2469     BDRVQcow2State *s = bs->opaque;
2470     z_stream strm;
2471     int ret, out_len;
2472     uint8_t *out_buf;
2473     uint64_t cluster_offset;
2474 
2475     if (nb_sectors == 0) {
2476         /* align end of file to a sector boundary to ease reading with
2477            sector based I/Os */
2478         cluster_offset = bdrv_getlength(bs->file->bs);
2479         return bdrv_truncate(bs->file->bs, cluster_offset);
2480     }
2481 
2482     if (nb_sectors != s->cluster_sectors) {
2483         ret = -EINVAL;
2484 
2485         /* Zero-pad last write if image size is not cluster aligned */
2486         if (sector_num + nb_sectors == bs->total_sectors &&
2487             nb_sectors < s->cluster_sectors) {
2488             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
2489             memset(pad_buf, 0, s->cluster_size);
2490             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
2491             ret = qcow2_write_compressed(bs, sector_num,
2492                                          pad_buf, s->cluster_sectors);
2493             qemu_vfree(pad_buf);
2494         }
2495         return ret;
2496     }
2497 
2498     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
2499 
2500     /* best compression, small window, no zlib header */
2501     memset(&strm, 0, sizeof(strm));
2502     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
2503                        Z_DEFLATED, -12,
2504                        9, Z_DEFAULT_STRATEGY);
2505     if (ret != 0) {
2506         ret = -EINVAL;
2507         goto fail;
2508     }
2509 
2510     strm.avail_in = s->cluster_size;
2511     strm.next_in = (uint8_t *)buf;
2512     strm.avail_out = s->cluster_size;
2513     strm.next_out = out_buf;
2514 
2515     ret = deflate(&strm, Z_FINISH);
2516     if (ret != Z_STREAM_END && ret != Z_OK) {
2517         deflateEnd(&strm);
2518         ret = -EINVAL;
2519         goto fail;
2520     }
2521     out_len = strm.next_out - out_buf;
2522 
2523     deflateEnd(&strm);
2524 
2525     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
2526         /* could not compress: write normal cluster */
2527         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
2528         if (ret < 0) {
2529             goto fail;
2530         }
2531     } else {
2532         cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
2533             sector_num << 9, out_len);
2534         if (!cluster_offset) {
2535             ret = -EIO;
2536             goto fail;
2537         }
2538         cluster_offset &= s->cluster_offset_mask;
2539 
2540         ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
2541         if (ret < 0) {
2542             goto fail;
2543         }
2544 
2545         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
2546         ret = bdrv_pwrite(bs->file->bs, cluster_offset, out_buf, out_len);
2547         if (ret < 0) {
2548             goto fail;
2549         }
2550     }
2551 
2552     ret = 0;
2553 fail:
2554     g_free(out_buf);
2555     return ret;
2556 }
2557 
2558 static int make_completely_empty(BlockDriverState *bs)
2559 {
2560     BDRVQcow2State *s = bs->opaque;
2561     int ret, l1_clusters;
2562     int64_t offset;
2563     uint64_t *new_reftable = NULL;
2564     uint64_t rt_entry, l1_size2;
2565     struct {
2566         uint64_t l1_offset;
2567         uint64_t reftable_offset;
2568         uint32_t reftable_clusters;
2569     } QEMU_PACKED l1_ofs_rt_ofs_cls;
2570 
2571     ret = qcow2_cache_empty(bs, s->l2_table_cache);
2572     if (ret < 0) {
2573         goto fail;
2574     }
2575 
2576     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
2577     if (ret < 0) {
2578         goto fail;
2579     }
2580 
2581     /* Refcounts will be broken utterly */
2582     ret = qcow2_mark_dirty(bs);
2583     if (ret < 0) {
2584         goto fail;
2585     }
2586 
2587     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2588 
2589     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2590     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
2591 
2592     /* After this call, neither the in-memory nor the on-disk refcount
2593      * information accurately describe the actual references */
2594 
2595     ret = bdrv_write_zeroes(bs->file->bs, s->l1_table_offset / BDRV_SECTOR_SIZE,
2596                             l1_clusters * s->cluster_sectors, 0);
2597     if (ret < 0) {
2598         goto fail_broken_refcounts;
2599     }
2600     memset(s->l1_table, 0, l1_size2);
2601 
2602     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
2603 
2604     /* Overwrite enough clusters at the beginning of the sectors to place
2605      * the refcount table, a refcount block and the L1 table in; this may
2606      * overwrite parts of the existing refcount and L1 table, which is not
2607      * an issue because the dirty flag is set, complete data loss is in fact
2608      * desired and partial data loss is consequently fine as well */
2609     ret = bdrv_write_zeroes(bs->file->bs, s->cluster_size / BDRV_SECTOR_SIZE,
2610                             (2 + l1_clusters) * s->cluster_size /
2611                             BDRV_SECTOR_SIZE, 0);
2612     /* This call (even if it failed overall) may have overwritten on-disk
2613      * refcount structures; in that case, the in-memory refcount information
2614      * will probably differ from the on-disk information which makes the BDS
2615      * unusable */
2616     if (ret < 0) {
2617         goto fail_broken_refcounts;
2618     }
2619 
2620     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2621     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
2622 
2623     /* "Create" an empty reftable (one cluster) directly after the image
2624      * header and an empty L1 table three clusters after the image header;
2625      * the cluster between those two will be used as the first refblock */
2626     cpu_to_be64w(&l1_ofs_rt_ofs_cls.l1_offset, 3 * s->cluster_size);
2627     cpu_to_be64w(&l1_ofs_rt_ofs_cls.reftable_offset, s->cluster_size);
2628     cpu_to_be32w(&l1_ofs_rt_ofs_cls.reftable_clusters, 1);
2629     ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, l1_table_offset),
2630                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
2631     if (ret < 0) {
2632         goto fail_broken_refcounts;
2633     }
2634 
2635     s->l1_table_offset = 3 * s->cluster_size;
2636 
2637     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
2638     if (!new_reftable) {
2639         ret = -ENOMEM;
2640         goto fail_broken_refcounts;
2641     }
2642 
2643     s->refcount_table_offset = s->cluster_size;
2644     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
2645 
2646     g_free(s->refcount_table);
2647     s->refcount_table = new_reftable;
2648     new_reftable = NULL;
2649 
2650     /* Now the in-memory refcount information again corresponds to the on-disk
2651      * information (reftable is empty and no refblocks (the refblock cache is
2652      * empty)); however, this means some clusters (e.g. the image header) are
2653      * referenced, but not refcounted, but the normal qcow2 code assumes that
2654      * the in-memory information is always correct */
2655 
2656     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
2657 
2658     /* Enter the first refblock into the reftable */
2659     rt_entry = cpu_to_be64(2 * s->cluster_size);
2660     ret = bdrv_pwrite_sync(bs->file->bs, s->cluster_size,
2661                            &rt_entry, sizeof(rt_entry));
2662     if (ret < 0) {
2663         goto fail_broken_refcounts;
2664     }
2665     s->refcount_table[0] = 2 * s->cluster_size;
2666 
2667     s->free_cluster_index = 0;
2668     assert(3 + l1_clusters <= s->refcount_block_size);
2669     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
2670     if (offset < 0) {
2671         ret = offset;
2672         goto fail_broken_refcounts;
2673     } else if (offset > 0) {
2674         error_report("First cluster in emptied image is in use");
2675         abort();
2676     }
2677 
2678     /* Now finally the in-memory information corresponds to the on-disk
2679      * structures and is correct */
2680     ret = qcow2_mark_clean(bs);
2681     if (ret < 0) {
2682         goto fail;
2683     }
2684 
2685     ret = bdrv_truncate(bs->file->bs, (3 + l1_clusters) * s->cluster_size);
2686     if (ret < 0) {
2687         goto fail;
2688     }
2689 
2690     return 0;
2691 
2692 fail_broken_refcounts:
2693     /* The BDS is unusable at this point. If we wanted to make it usable, we
2694      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
2695      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
2696      * again. However, because the functions which could have caused this error
2697      * path to be taken are used by those functions as well, it's very likely
2698      * that that sequence will fail as well. Therefore, just eject the BDS. */
2699     bs->drv = NULL;
2700 
2701 fail:
2702     g_free(new_reftable);
2703     return ret;
2704 }
2705 
2706 static int qcow2_make_empty(BlockDriverState *bs)
2707 {
2708     BDRVQcow2State *s = bs->opaque;
2709     uint64_t start_sector;
2710     int sector_step = INT_MAX / BDRV_SECTOR_SIZE;
2711     int l1_clusters, ret = 0;
2712 
2713     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2714 
2715     if (s->qcow_version >= 3 && !s->snapshots &&
2716         3 + l1_clusters <= s->refcount_block_size) {
2717         /* The following function only works for qcow2 v3 images (it requires
2718          * the dirty flag) and only as long as there are no snapshots (because
2719          * it completely empties the image). Furthermore, the L1 table and three
2720          * additional clusters (image header, refcount table, one refcount
2721          * block) have to fit inside one refcount block. */
2722         return make_completely_empty(bs);
2723     }
2724 
2725     /* This fallback code simply discards every active cluster; this is slow,
2726      * but works in all cases */
2727     for (start_sector = 0; start_sector < bs->total_sectors;
2728          start_sector += sector_step)
2729     {
2730         /* As this function is generally used after committing an external
2731          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
2732          * default action for this kind of discard is to pass the discard,
2733          * which will ideally result in an actually smaller image file, as
2734          * is probably desired. */
2735         ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE,
2736                                      MIN(sector_step,
2737                                          bs->total_sectors - start_sector),
2738                                      QCOW2_DISCARD_SNAPSHOT, true);
2739         if (ret < 0) {
2740             break;
2741         }
2742     }
2743 
2744     return ret;
2745 }
2746 
2747 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
2748 {
2749     BDRVQcow2State *s = bs->opaque;
2750     int ret;
2751 
2752     qemu_co_mutex_lock(&s->lock);
2753     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2754     if (ret < 0) {
2755         qemu_co_mutex_unlock(&s->lock);
2756         return ret;
2757     }
2758 
2759     if (qcow2_need_accurate_refcounts(s)) {
2760         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2761         if (ret < 0) {
2762             qemu_co_mutex_unlock(&s->lock);
2763             return ret;
2764         }
2765     }
2766     qemu_co_mutex_unlock(&s->lock);
2767 
2768     return 0;
2769 }
2770 
2771 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2772 {
2773     BDRVQcow2State *s = bs->opaque;
2774     bdi->unallocated_blocks_are_zero = true;
2775     bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
2776     bdi->cluster_size = s->cluster_size;
2777     bdi->vm_state_offset = qcow2_vm_state_offset(s);
2778     return 0;
2779 }
2780 
2781 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
2782 {
2783     BDRVQcow2State *s = bs->opaque;
2784     ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
2785 
2786     *spec_info = (ImageInfoSpecific){
2787         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
2788         .u.qcow2 = g_new(ImageInfoSpecificQCow2, 1),
2789     };
2790     if (s->qcow_version == 2) {
2791         *spec_info->u.qcow2 = (ImageInfoSpecificQCow2){
2792             .compat             = g_strdup("0.10"),
2793             .refcount_bits      = s->refcount_bits,
2794         };
2795     } else if (s->qcow_version == 3) {
2796         *spec_info->u.qcow2 = (ImageInfoSpecificQCow2){
2797             .compat             = g_strdup("1.1"),
2798             .lazy_refcounts     = s->compatible_features &
2799                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
2800             .has_lazy_refcounts = true,
2801             .corrupt            = s->incompatible_features &
2802                                   QCOW2_INCOMPAT_CORRUPT,
2803             .has_corrupt        = true,
2804             .refcount_bits      = s->refcount_bits,
2805         };
2806     } else {
2807         /* if this assertion fails, this probably means a new version was
2808          * added without having it covered here */
2809         assert(false);
2810     }
2811 
2812     return spec_info;
2813 }
2814 
2815 #if 0
2816 static void dump_refcounts(BlockDriverState *bs)
2817 {
2818     BDRVQcow2State *s = bs->opaque;
2819     int64_t nb_clusters, k, k1, size;
2820     int refcount;
2821 
2822     size = bdrv_getlength(bs->file->bs);
2823     nb_clusters = size_to_clusters(s, size);
2824     for(k = 0; k < nb_clusters;) {
2825         k1 = k;
2826         refcount = get_refcount(bs, k);
2827         k++;
2828         while (k < nb_clusters && get_refcount(bs, k) == refcount)
2829             k++;
2830         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
2831                k - k1);
2832     }
2833 }
2834 #endif
2835 
2836 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2837                               int64_t pos)
2838 {
2839     BDRVQcow2State *s = bs->opaque;
2840     int64_t total_sectors = bs->total_sectors;
2841     bool zero_beyond_eof = bs->zero_beyond_eof;
2842     int ret;
2843 
2844     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
2845     bs->zero_beyond_eof = false;
2846     ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov);
2847     bs->zero_beyond_eof = zero_beyond_eof;
2848 
2849     /* bdrv_co_do_writev will have increased the total_sectors value to include
2850      * the VM state - the VM state is however not an actual part of the block
2851      * device, therefore, we need to restore the old value. */
2852     bs->total_sectors = total_sectors;
2853 
2854     return ret;
2855 }
2856 
2857 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2858                               int64_t pos, int size)
2859 {
2860     BDRVQcow2State *s = bs->opaque;
2861     bool zero_beyond_eof = bs->zero_beyond_eof;
2862     int ret;
2863 
2864     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
2865     bs->zero_beyond_eof = false;
2866     ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
2867     bs->zero_beyond_eof = zero_beyond_eof;
2868 
2869     return ret;
2870 }
2871 
2872 /*
2873  * Downgrades an image's version. To achieve this, any incompatible features
2874  * have to be removed.
2875  */
2876 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
2877                            BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
2878 {
2879     BDRVQcow2State *s = bs->opaque;
2880     int current_version = s->qcow_version;
2881     int ret;
2882 
2883     if (target_version == current_version) {
2884         return 0;
2885     } else if (target_version > current_version) {
2886         return -EINVAL;
2887     } else if (target_version != 2) {
2888         return -EINVAL;
2889     }
2890 
2891     if (s->refcount_order != 4) {
2892         error_report("compat=0.10 requires refcount_bits=16");
2893         return -ENOTSUP;
2894     }
2895 
2896     /* clear incompatible features */
2897     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
2898         ret = qcow2_mark_clean(bs);
2899         if (ret < 0) {
2900             return ret;
2901         }
2902     }
2903 
2904     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2905      * the first place; if that happens nonetheless, returning -ENOTSUP is the
2906      * best thing to do anyway */
2907 
2908     if (s->incompatible_features) {
2909         return -ENOTSUP;
2910     }
2911 
2912     /* since we can ignore compatible features, we can set them to 0 as well */
2913     s->compatible_features = 0;
2914     /* if lazy refcounts have been used, they have already been fixed through
2915      * clearing the dirty flag */
2916 
2917     /* clearing autoclear features is trivial */
2918     s->autoclear_features = 0;
2919 
2920     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
2921     if (ret < 0) {
2922         return ret;
2923     }
2924 
2925     s->qcow_version = target_version;
2926     ret = qcow2_update_header(bs);
2927     if (ret < 0) {
2928         s->qcow_version = current_version;
2929         return ret;
2930     }
2931     return 0;
2932 }
2933 
2934 typedef enum Qcow2AmendOperation {
2935     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
2936      * statically initialized to so that the helper CB can discern the first
2937      * invocation from an operation change */
2938     QCOW2_NO_OPERATION = 0,
2939 
2940     QCOW2_CHANGING_REFCOUNT_ORDER,
2941     QCOW2_DOWNGRADING,
2942 } Qcow2AmendOperation;
2943 
2944 typedef struct Qcow2AmendHelperCBInfo {
2945     /* The code coordinating the amend operations should only modify
2946      * these four fields; the rest will be managed by the CB */
2947     BlockDriverAmendStatusCB *original_status_cb;
2948     void *original_cb_opaque;
2949 
2950     Qcow2AmendOperation current_operation;
2951 
2952     /* Total number of operations to perform (only set once) */
2953     int total_operations;
2954 
2955     /* The following fields are managed by the CB */
2956 
2957     /* Number of operations completed */
2958     int operations_completed;
2959 
2960     /* Cumulative offset of all completed operations */
2961     int64_t offset_completed;
2962 
2963     Qcow2AmendOperation last_operation;
2964     int64_t last_work_size;
2965 } Qcow2AmendHelperCBInfo;
2966 
2967 static void qcow2_amend_helper_cb(BlockDriverState *bs,
2968                                   int64_t operation_offset,
2969                                   int64_t operation_work_size, void *opaque)
2970 {
2971     Qcow2AmendHelperCBInfo *info = opaque;
2972     int64_t current_work_size;
2973     int64_t projected_work_size;
2974 
2975     if (info->current_operation != info->last_operation) {
2976         if (info->last_operation != QCOW2_NO_OPERATION) {
2977             info->offset_completed += info->last_work_size;
2978             info->operations_completed++;
2979         }
2980 
2981         info->last_operation = info->current_operation;
2982     }
2983 
2984     assert(info->total_operations > 0);
2985     assert(info->operations_completed < info->total_operations);
2986 
2987     info->last_work_size = operation_work_size;
2988 
2989     current_work_size = info->offset_completed + operation_work_size;
2990 
2991     /* current_work_size is the total work size for (operations_completed + 1)
2992      * operations (which includes this one), so multiply it by the number of
2993      * operations not covered and divide it by the number of operations
2994      * covered to get a projection for the operations not covered */
2995     projected_work_size = current_work_size * (info->total_operations -
2996                                                info->operations_completed - 1)
2997                                             / (info->operations_completed + 1);
2998 
2999     info->original_status_cb(bs, info->offset_completed + operation_offset,
3000                              current_work_size + projected_work_size,
3001                              info->original_cb_opaque);
3002 }
3003 
3004 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
3005                                BlockDriverAmendStatusCB *status_cb,
3006                                void *cb_opaque)
3007 {
3008     BDRVQcow2State *s = bs->opaque;
3009     int old_version = s->qcow_version, new_version = old_version;
3010     uint64_t new_size = 0;
3011     const char *backing_file = NULL, *backing_format = NULL;
3012     bool lazy_refcounts = s->use_lazy_refcounts;
3013     const char *compat = NULL;
3014     uint64_t cluster_size = s->cluster_size;
3015     bool encrypt;
3016     int refcount_bits = s->refcount_bits;
3017     int ret;
3018     QemuOptDesc *desc = opts->list->desc;
3019     Qcow2AmendHelperCBInfo helper_cb_info;
3020 
3021     while (desc && desc->name) {
3022         if (!qemu_opt_find(opts, desc->name)) {
3023             /* only change explicitly defined options */
3024             desc++;
3025             continue;
3026         }
3027 
3028         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
3029             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
3030             if (!compat) {
3031                 /* preserve default */
3032             } else if (!strcmp(compat, "0.10")) {
3033                 new_version = 2;
3034             } else if (!strcmp(compat, "1.1")) {
3035                 new_version = 3;
3036             } else {
3037                 error_report("Unknown compatibility level %s", compat);
3038                 return -EINVAL;
3039             }
3040         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
3041             error_report("Cannot change preallocation mode");
3042             return -ENOTSUP;
3043         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
3044             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
3045         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
3046             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
3047         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
3048             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
3049         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
3050             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
3051                                         !!s->cipher);
3052 
3053             if (encrypt != !!s->cipher) {
3054                 error_report("Changing the encryption flag is not supported");
3055                 return -ENOTSUP;
3056             }
3057         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
3058             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
3059                                              cluster_size);
3060             if (cluster_size != s->cluster_size) {
3061                 error_report("Changing the cluster size is not supported");
3062                 return -ENOTSUP;
3063             }
3064         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
3065             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
3066                                                lazy_refcounts);
3067         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
3068             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
3069                                                 refcount_bits);
3070 
3071             if (refcount_bits <= 0 || refcount_bits > 64 ||
3072                 !is_power_of_2(refcount_bits))
3073             {
3074                 error_report("Refcount width must be a power of two and may "
3075                              "not exceed 64 bits");
3076                 return -EINVAL;
3077             }
3078         } else {
3079             /* if this point is reached, this probably means a new option was
3080              * added without having it covered here */
3081             abort();
3082         }
3083 
3084         desc++;
3085     }
3086 
3087     helper_cb_info = (Qcow2AmendHelperCBInfo){
3088         .original_status_cb = status_cb,
3089         .original_cb_opaque = cb_opaque,
3090         .total_operations = (new_version < old_version)
3091                           + (s->refcount_bits != refcount_bits)
3092     };
3093 
3094     /* Upgrade first (some features may require compat=1.1) */
3095     if (new_version > old_version) {
3096         s->qcow_version = new_version;
3097         ret = qcow2_update_header(bs);
3098         if (ret < 0) {
3099             s->qcow_version = old_version;
3100             return ret;
3101         }
3102     }
3103 
3104     if (s->refcount_bits != refcount_bits) {
3105         int refcount_order = ctz32(refcount_bits);
3106         Error *local_error = NULL;
3107 
3108         if (new_version < 3 && refcount_bits != 16) {
3109             error_report("Different refcount widths than 16 bits require "
3110                          "compatibility level 1.1 or above (use compat=1.1 or "
3111                          "greater)");
3112             return -EINVAL;
3113         }
3114 
3115         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
3116         ret = qcow2_change_refcount_order(bs, refcount_order,
3117                                           &qcow2_amend_helper_cb,
3118                                           &helper_cb_info, &local_error);
3119         if (ret < 0) {
3120             error_report_err(local_error);
3121             return ret;
3122         }
3123     }
3124 
3125     if (backing_file || backing_format) {
3126         ret = qcow2_change_backing_file(bs,
3127                     backing_file ?: s->image_backing_file,
3128                     backing_format ?: s->image_backing_format);
3129         if (ret < 0) {
3130             return ret;
3131         }
3132     }
3133 
3134     if (s->use_lazy_refcounts != lazy_refcounts) {
3135         if (lazy_refcounts) {
3136             if (new_version < 3) {
3137                 error_report("Lazy refcounts only supported with compatibility "
3138                              "level 1.1 and above (use compat=1.1 or greater)");
3139                 return -EINVAL;
3140             }
3141             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3142             ret = qcow2_update_header(bs);
3143             if (ret < 0) {
3144                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3145                 return ret;
3146             }
3147             s->use_lazy_refcounts = true;
3148         } else {
3149             /* make image clean first */
3150             ret = qcow2_mark_clean(bs);
3151             if (ret < 0) {
3152                 return ret;
3153             }
3154             /* now disallow lazy refcounts */
3155             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3156             ret = qcow2_update_header(bs);
3157             if (ret < 0) {
3158                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3159                 return ret;
3160             }
3161             s->use_lazy_refcounts = false;
3162         }
3163     }
3164 
3165     if (new_size) {
3166         ret = bdrv_truncate(bs, new_size);
3167         if (ret < 0) {
3168             return ret;
3169         }
3170     }
3171 
3172     /* Downgrade last (so unsupported features can be removed before) */
3173     if (new_version < old_version) {
3174         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
3175         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
3176                               &helper_cb_info);
3177         if (ret < 0) {
3178             return ret;
3179         }
3180     }
3181 
3182     return 0;
3183 }
3184 
3185 /*
3186  * If offset or size are negative, respectively, they will not be included in
3187  * the BLOCK_IMAGE_CORRUPTED event emitted.
3188  * fatal will be ignored for read-only BDS; corruptions found there will always
3189  * be considered non-fatal.
3190  */
3191 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
3192                              int64_t size, const char *message_format, ...)
3193 {
3194     BDRVQcow2State *s = bs->opaque;
3195     const char *node_name;
3196     char *message;
3197     va_list ap;
3198 
3199     fatal = fatal && !bs->read_only;
3200 
3201     if (s->signaled_corruption &&
3202         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
3203     {
3204         return;
3205     }
3206 
3207     va_start(ap, message_format);
3208     message = g_strdup_vprintf(message_format, ap);
3209     va_end(ap);
3210 
3211     if (fatal) {
3212         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
3213                 "corruption events will be suppressed\n", message);
3214     } else {
3215         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
3216                 "corruption events will be suppressed\n", message);
3217     }
3218 
3219     node_name = bdrv_get_node_name(bs);
3220     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
3221                                           *node_name != '\0', node_name,
3222                                           message, offset >= 0, offset,
3223                                           size >= 0, size,
3224                                           fatal, &error_abort);
3225     g_free(message);
3226 
3227     if (fatal) {
3228         qcow2_mark_corrupt(bs);
3229         bs->drv = NULL; /* make BDS unusable */
3230     }
3231 
3232     s->signaled_corruption = true;
3233 }
3234 
3235 static QemuOptsList qcow2_create_opts = {
3236     .name = "qcow2-create-opts",
3237     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
3238     .desc = {
3239         {
3240             .name = BLOCK_OPT_SIZE,
3241             .type = QEMU_OPT_SIZE,
3242             .help = "Virtual disk size"
3243         },
3244         {
3245             .name = BLOCK_OPT_COMPAT_LEVEL,
3246             .type = QEMU_OPT_STRING,
3247             .help = "Compatibility level (0.10 or 1.1)"
3248         },
3249         {
3250             .name = BLOCK_OPT_BACKING_FILE,
3251             .type = QEMU_OPT_STRING,
3252             .help = "File name of a base image"
3253         },
3254         {
3255             .name = BLOCK_OPT_BACKING_FMT,
3256             .type = QEMU_OPT_STRING,
3257             .help = "Image format of the base image"
3258         },
3259         {
3260             .name = BLOCK_OPT_ENCRYPT,
3261             .type = QEMU_OPT_BOOL,
3262             .help = "Encrypt the image",
3263             .def_value_str = "off"
3264         },
3265         {
3266             .name = BLOCK_OPT_CLUSTER_SIZE,
3267             .type = QEMU_OPT_SIZE,
3268             .help = "qcow2 cluster size",
3269             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
3270         },
3271         {
3272             .name = BLOCK_OPT_PREALLOC,
3273             .type = QEMU_OPT_STRING,
3274             .help = "Preallocation mode (allowed values: off, metadata, "
3275                     "falloc, full)"
3276         },
3277         {
3278             .name = BLOCK_OPT_LAZY_REFCOUNTS,
3279             .type = QEMU_OPT_BOOL,
3280             .help = "Postpone refcount updates",
3281             .def_value_str = "off"
3282         },
3283         {
3284             .name = BLOCK_OPT_REFCOUNT_BITS,
3285             .type = QEMU_OPT_NUMBER,
3286             .help = "Width of a reference count entry in bits",
3287             .def_value_str = "16"
3288         },
3289         { /* end of list */ }
3290     }
3291 };
3292 
3293 BlockDriver bdrv_qcow2 = {
3294     .format_name        = "qcow2",
3295     .instance_size      = sizeof(BDRVQcow2State),
3296     .bdrv_probe         = qcow2_probe,
3297     .bdrv_open          = qcow2_open,
3298     .bdrv_close         = qcow2_close,
3299     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
3300     .bdrv_reopen_commit   = qcow2_reopen_commit,
3301     .bdrv_reopen_abort    = qcow2_reopen_abort,
3302     .bdrv_join_options    = qcow2_join_options,
3303     .bdrv_create        = qcow2_create,
3304     .bdrv_has_zero_init = bdrv_has_zero_init_1,
3305     .bdrv_co_get_block_status = qcow2_co_get_block_status,
3306     .bdrv_set_key       = qcow2_set_key,
3307 
3308     .bdrv_co_readv          = qcow2_co_readv,
3309     .bdrv_co_writev         = qcow2_co_writev,
3310     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
3311 
3312     .bdrv_co_write_zeroes   = qcow2_co_write_zeroes,
3313     .bdrv_co_discard        = qcow2_co_discard,
3314     .bdrv_truncate          = qcow2_truncate,
3315     .bdrv_write_compressed  = qcow2_write_compressed,
3316     .bdrv_make_empty        = qcow2_make_empty,
3317 
3318     .bdrv_snapshot_create   = qcow2_snapshot_create,
3319     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
3320     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
3321     .bdrv_snapshot_list     = qcow2_snapshot_list,
3322     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
3323     .bdrv_get_info          = qcow2_get_info,
3324     .bdrv_get_specific_info = qcow2_get_specific_info,
3325 
3326     .bdrv_save_vmstate    = qcow2_save_vmstate,
3327     .bdrv_load_vmstate    = qcow2_load_vmstate,
3328 
3329     .supports_backing           = true,
3330     .bdrv_change_backing_file   = qcow2_change_backing_file,
3331 
3332     .bdrv_refresh_limits        = qcow2_refresh_limits,
3333     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
3334 
3335     .create_opts         = &qcow2_create_opts,
3336     .bdrv_check          = qcow2_check,
3337     .bdrv_amend_options  = qcow2_amend_options,
3338 
3339     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
3340     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
3341 };
3342 
3343 static void bdrv_qcow2_init(void)
3344 {
3345     bdrv_register(&bdrv_qcow2);
3346 }
3347 
3348 block_init(bdrv_qcow2_init);
3349