xref: /openbmc/qemu/block/qcow2.c (revision a23797ef)
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     BDRVQcowState *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, 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, 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, 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, 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     BDRVQcowState *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     BDRVQcowState *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, offsetof(QCowHeader, incompatible_features),
264                       &val, sizeof(val));
265     if (ret < 0) {
266         return ret;
267     }
268     ret = bdrv_flush(bs->file);
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     BDRVQcowState *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     BDRVQcowState *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     BDRVQcowState *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     BDRVQcowState *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     BDRVQcowState *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     BDRVQcowState *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     BDRVQcowState *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     BDRVQcowState *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 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
593                       Error **errp)
594 {
595     BDRVQcowState *s = bs->opaque;
596     unsigned int len, i;
597     int ret = 0;
598     QCowHeader header;
599     QemuOpts *opts = NULL;
600     Error *local_err = NULL;
601     uint64_t ext_end;
602     uint64_t l1_vm_state_index;
603     const char *opt_overlap_check, *opt_overlap_check_template;
604     int overlap_check_template = 0;
605     uint64_t l2_cache_size, refcount_cache_size;
606     uint64_t cache_clean_interval;
607 
608     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
609     if (ret < 0) {
610         error_setg_errno(errp, -ret, "Could not read qcow2 header");
611         goto fail;
612     }
613     be32_to_cpus(&header.magic);
614     be32_to_cpus(&header.version);
615     be64_to_cpus(&header.backing_file_offset);
616     be32_to_cpus(&header.backing_file_size);
617     be64_to_cpus(&header.size);
618     be32_to_cpus(&header.cluster_bits);
619     be32_to_cpus(&header.crypt_method);
620     be64_to_cpus(&header.l1_table_offset);
621     be32_to_cpus(&header.l1_size);
622     be64_to_cpus(&header.refcount_table_offset);
623     be32_to_cpus(&header.refcount_table_clusters);
624     be64_to_cpus(&header.snapshots_offset);
625     be32_to_cpus(&header.nb_snapshots);
626 
627     if (header.magic != QCOW_MAGIC) {
628         error_setg(errp, "Image is not in qcow2 format");
629         ret = -EINVAL;
630         goto fail;
631     }
632     if (header.version < 2 || header.version > 3) {
633         report_unsupported(bs, errp, "QCOW version %" PRIu32, header.version);
634         ret = -ENOTSUP;
635         goto fail;
636     }
637 
638     s->qcow_version = header.version;
639 
640     /* Initialise cluster size */
641     if (header.cluster_bits < MIN_CLUSTER_BITS ||
642         header.cluster_bits > MAX_CLUSTER_BITS) {
643         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
644                    header.cluster_bits);
645         ret = -EINVAL;
646         goto fail;
647     }
648 
649     s->cluster_bits = header.cluster_bits;
650     s->cluster_size = 1 << s->cluster_bits;
651     s->cluster_sectors = 1 << (s->cluster_bits - 9);
652 
653     /* Initialise version 3 header fields */
654     if (header.version == 2) {
655         header.incompatible_features    = 0;
656         header.compatible_features      = 0;
657         header.autoclear_features       = 0;
658         header.refcount_order           = 4;
659         header.header_length            = 72;
660     } else {
661         be64_to_cpus(&header.incompatible_features);
662         be64_to_cpus(&header.compatible_features);
663         be64_to_cpus(&header.autoclear_features);
664         be32_to_cpus(&header.refcount_order);
665         be32_to_cpus(&header.header_length);
666 
667         if (header.header_length < 104) {
668             error_setg(errp, "qcow2 header too short");
669             ret = -EINVAL;
670             goto fail;
671         }
672     }
673 
674     if (header.header_length > s->cluster_size) {
675         error_setg(errp, "qcow2 header exceeds cluster size");
676         ret = -EINVAL;
677         goto fail;
678     }
679 
680     if (header.header_length > sizeof(header)) {
681         s->unknown_header_fields_size = header.header_length - sizeof(header);
682         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
683         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
684                          s->unknown_header_fields_size);
685         if (ret < 0) {
686             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
687                              "fields");
688             goto fail;
689         }
690     }
691 
692     if (header.backing_file_offset > s->cluster_size) {
693         error_setg(errp, "Invalid backing file offset");
694         ret = -EINVAL;
695         goto fail;
696     }
697 
698     if (header.backing_file_offset) {
699         ext_end = header.backing_file_offset;
700     } else {
701         ext_end = 1 << header.cluster_bits;
702     }
703 
704     /* Handle feature bits */
705     s->incompatible_features    = header.incompatible_features;
706     s->compatible_features      = header.compatible_features;
707     s->autoclear_features       = header.autoclear_features;
708 
709     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
710         void *feature_table = NULL;
711         qcow2_read_extensions(bs, header.header_length, ext_end,
712                               &feature_table, NULL);
713         report_unsupported_feature(bs, errp, feature_table,
714                                    s->incompatible_features &
715                                    ~QCOW2_INCOMPAT_MASK);
716         ret = -ENOTSUP;
717         g_free(feature_table);
718         goto fail;
719     }
720 
721     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
722         /* Corrupt images may not be written to unless they are being repaired
723          */
724         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
725             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
726                        "read/write");
727             ret = -EACCES;
728             goto fail;
729         }
730     }
731 
732     /* Check support for various header values */
733     if (header.refcount_order > 6) {
734         error_setg(errp, "Reference count entry width too large; may not "
735                    "exceed 64 bits");
736         ret = -EINVAL;
737         goto fail;
738     }
739     s->refcount_order = header.refcount_order;
740     s->refcount_bits = 1 << s->refcount_order;
741     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
742     s->refcount_max += s->refcount_max - 1;
743 
744     if (header.crypt_method > QCOW_CRYPT_AES) {
745         error_setg(errp, "Unsupported encryption method: %" PRIu32,
746                    header.crypt_method);
747         ret = -EINVAL;
748         goto fail;
749     }
750     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) {
751         error_setg(errp, "AES cipher not available");
752         ret = -EINVAL;
753         goto fail;
754     }
755     s->crypt_method_header = header.crypt_method;
756     if (s->crypt_method_header) {
757         bs->encrypted = 1;
758     }
759 
760     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
761     s->l2_size = 1 << s->l2_bits;
762     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
763     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
764     s->refcount_block_size = 1 << s->refcount_block_bits;
765     bs->total_sectors = header.size / 512;
766     s->csize_shift = (62 - (s->cluster_bits - 8));
767     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
768     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
769 
770     s->refcount_table_offset = header.refcount_table_offset;
771     s->refcount_table_size =
772         header.refcount_table_clusters << (s->cluster_bits - 3);
773 
774     if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
775         error_setg(errp, "Reference count table too large");
776         ret = -EINVAL;
777         goto fail;
778     }
779 
780     ret = validate_table_offset(bs, s->refcount_table_offset,
781                                 s->refcount_table_size, sizeof(uint64_t));
782     if (ret < 0) {
783         error_setg(errp, "Invalid reference count table offset");
784         goto fail;
785     }
786 
787     /* Snapshot table offset/length */
788     if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
789         error_setg(errp, "Too many snapshots");
790         ret = -EINVAL;
791         goto fail;
792     }
793 
794     ret = validate_table_offset(bs, header.snapshots_offset,
795                                 header.nb_snapshots,
796                                 sizeof(QCowSnapshotHeader));
797     if (ret < 0) {
798         error_setg(errp, "Invalid snapshot table offset");
799         goto fail;
800     }
801 
802     /* read the level 1 table */
803     if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
804         error_setg(errp, "Active L1 table too large");
805         ret = -EFBIG;
806         goto fail;
807     }
808     s->l1_size = header.l1_size;
809 
810     l1_vm_state_index = size_to_l1(s, header.size);
811     if (l1_vm_state_index > INT_MAX) {
812         error_setg(errp, "Image is too big");
813         ret = -EFBIG;
814         goto fail;
815     }
816     s->l1_vm_state_index = l1_vm_state_index;
817 
818     /* the L1 table must contain at least enough entries to put
819        header.size bytes */
820     if (s->l1_size < s->l1_vm_state_index) {
821         error_setg(errp, "L1 table is too small");
822         ret = -EINVAL;
823         goto fail;
824     }
825 
826     ret = validate_table_offset(bs, header.l1_table_offset,
827                                 header.l1_size, sizeof(uint64_t));
828     if (ret < 0) {
829         error_setg(errp, "Invalid L1 table offset");
830         goto fail;
831     }
832     s->l1_table_offset = header.l1_table_offset;
833 
834 
835     if (s->l1_size > 0) {
836         s->l1_table = qemu_try_blockalign(bs->file,
837             align_offset(s->l1_size * sizeof(uint64_t), 512));
838         if (s->l1_table == NULL) {
839             error_setg(errp, "Could not allocate L1 table");
840             ret = -ENOMEM;
841             goto fail;
842         }
843         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
844                          s->l1_size * sizeof(uint64_t));
845         if (ret < 0) {
846             error_setg_errno(errp, -ret, "Could not read L1 table");
847             goto fail;
848         }
849         for(i = 0;i < s->l1_size; i++) {
850             be64_to_cpus(&s->l1_table[i]);
851         }
852     }
853 
854     /* get L2 table/refcount block cache size from command line options */
855     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
856     qemu_opts_absorb_qdict(opts, options, &local_err);
857     if (local_err) {
858         error_propagate(errp, local_err);
859         ret = -EINVAL;
860         goto fail;
861     }
862 
863     read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size,
864                      &local_err);
865     if (local_err) {
866         error_propagate(errp, local_err);
867         ret = -EINVAL;
868         goto fail;
869     }
870 
871     l2_cache_size /= s->cluster_size;
872     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
873         l2_cache_size = MIN_L2_CACHE_SIZE;
874     }
875     if (l2_cache_size > INT_MAX) {
876         error_setg(errp, "L2 cache size too big");
877         ret = -EINVAL;
878         goto fail;
879     }
880 
881     refcount_cache_size /= s->cluster_size;
882     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
883         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
884     }
885     if (refcount_cache_size > INT_MAX) {
886         error_setg(errp, "Refcount cache size too big");
887         ret = -EINVAL;
888         goto fail;
889     }
890 
891     /* alloc L2 table/refcount block cache */
892     s->l2_table_cache = qcow2_cache_create(bs, l2_cache_size);
893     s->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size);
894     if (s->l2_table_cache == NULL || s->refcount_block_cache == NULL) {
895         error_setg(errp, "Could not allocate metadata caches");
896         ret = -ENOMEM;
897         goto fail;
898     }
899 
900     cache_clean_interval =
901         qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL, 0);
902     if (cache_clean_interval > UINT_MAX) {
903         error_setg(errp, "Cache clean interval too big");
904         ret = -EINVAL;
905         goto fail;
906     }
907     s->cache_clean_interval = cache_clean_interval;
908     cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
909 
910     s->cluster_cache = g_malloc(s->cluster_size);
911     /* one more sector for decompressed data alignment */
912     s->cluster_data = qemu_try_blockalign(bs->file, QCOW_MAX_CRYPT_CLUSTERS
913                                                     * s->cluster_size + 512);
914     if (s->cluster_data == NULL) {
915         error_setg(errp, "Could not allocate temporary cluster buffer");
916         ret = -ENOMEM;
917         goto fail;
918     }
919 
920     s->cluster_cache_offset = -1;
921     s->flags = flags;
922 
923     ret = qcow2_refcount_init(bs);
924     if (ret != 0) {
925         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
926         goto fail;
927     }
928 
929     QLIST_INIT(&s->cluster_allocs);
930     QTAILQ_INIT(&s->discards);
931 
932     /* read qcow2 extensions */
933     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
934         &local_err)) {
935         error_propagate(errp, local_err);
936         ret = -EINVAL;
937         goto fail;
938     }
939 
940     /* read the backing file name */
941     if (header.backing_file_offset != 0) {
942         len = header.backing_file_size;
943         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
944             len >= sizeof(bs->backing_file)) {
945             error_setg(errp, "Backing file name too long");
946             ret = -EINVAL;
947             goto fail;
948         }
949         ret = bdrv_pread(bs->file, header.backing_file_offset,
950                          bs->backing_file, len);
951         if (ret < 0) {
952             error_setg_errno(errp, -ret, "Could not read backing file name");
953             goto fail;
954         }
955         bs->backing_file[len] = '\0';
956         s->image_backing_file = g_strdup(bs->backing_file);
957     }
958 
959     /* Internal snapshots */
960     s->snapshots_offset = header.snapshots_offset;
961     s->nb_snapshots = header.nb_snapshots;
962 
963     ret = qcow2_read_snapshots(bs);
964     if (ret < 0) {
965         error_setg_errno(errp, -ret, "Could not read snapshots");
966         goto fail;
967     }
968 
969     /* Clear unknown autoclear feature bits */
970     if (!bs->read_only && !(flags & BDRV_O_INCOMING) && s->autoclear_features) {
971         s->autoclear_features = 0;
972         ret = qcow2_update_header(bs);
973         if (ret < 0) {
974             error_setg_errno(errp, -ret, "Could not update qcow2 header");
975             goto fail;
976         }
977     }
978 
979     /* Initialise locks */
980     qemu_co_mutex_init(&s->lock);
981 
982     /* Repair image if dirty */
983     if (!(flags & (BDRV_O_CHECK | BDRV_O_INCOMING)) && !bs->read_only &&
984         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
985         BdrvCheckResult result = {0};
986 
987         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
988         if (ret < 0) {
989             error_setg_errno(errp, -ret, "Could not repair dirty image");
990             goto fail;
991         }
992     }
993 
994     /* Enable lazy_refcounts according to image and command line options */
995     s->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
996         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
997 
998     s->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
999     s->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
1000     s->discard_passthrough[QCOW2_DISCARD_REQUEST] =
1001         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
1002                           flags & BDRV_O_UNMAP);
1003     s->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
1004         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
1005     s->discard_passthrough[QCOW2_DISCARD_OTHER] =
1006         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
1007 
1008     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
1009     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
1010     if (opt_overlap_check_template && opt_overlap_check &&
1011         strcmp(opt_overlap_check_template, opt_overlap_check))
1012     {
1013         error_setg(errp, "Conflicting values for qcow2 options '"
1014                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1015                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
1016         ret = -EINVAL;
1017         goto fail;
1018     }
1019     if (!opt_overlap_check) {
1020         opt_overlap_check = opt_overlap_check_template ?: "cached";
1021     }
1022 
1023     if (!strcmp(opt_overlap_check, "none")) {
1024         overlap_check_template = 0;
1025     } else if (!strcmp(opt_overlap_check, "constant")) {
1026         overlap_check_template = QCOW2_OL_CONSTANT;
1027     } else if (!strcmp(opt_overlap_check, "cached")) {
1028         overlap_check_template = QCOW2_OL_CACHED;
1029     } else if (!strcmp(opt_overlap_check, "all")) {
1030         overlap_check_template = QCOW2_OL_ALL;
1031     } else {
1032         error_setg(errp, "Unsupported value '%s' for qcow2 option "
1033                    "'overlap-check'. Allowed are either of the following: "
1034                    "none, constant, cached, all", opt_overlap_check);
1035         ret = -EINVAL;
1036         goto fail;
1037     }
1038 
1039     s->overlap_check = 0;
1040     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
1041         /* overlap-check defines a template bitmask, but every flag may be
1042          * overwritten through the associated boolean option */
1043         s->overlap_check |=
1044             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
1045                               overlap_check_template & (1 << i)) << i;
1046     }
1047 
1048     qemu_opts_del(opts);
1049     opts = NULL;
1050 
1051     if (s->use_lazy_refcounts && s->qcow_version < 3) {
1052         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
1053                    "qemu 1.1 compatibility level");
1054         ret = -EINVAL;
1055         goto fail;
1056     }
1057 
1058 #ifdef DEBUG_ALLOC
1059     {
1060         BdrvCheckResult result = {0};
1061         qcow2_check_refcounts(bs, &result, 0);
1062     }
1063 #endif
1064     return ret;
1065 
1066  fail:
1067     qemu_opts_del(opts);
1068     g_free(s->unknown_header_fields);
1069     cleanup_unknown_header_ext(bs);
1070     qcow2_free_snapshots(bs);
1071     qcow2_refcount_close(bs);
1072     qemu_vfree(s->l1_table);
1073     /* else pre-write overlap checks in cache_destroy may crash */
1074     s->l1_table = NULL;
1075     cache_clean_timer_del(bs);
1076     if (s->l2_table_cache) {
1077         qcow2_cache_destroy(bs, s->l2_table_cache);
1078     }
1079     if (s->refcount_block_cache) {
1080         qcow2_cache_destroy(bs, s->refcount_block_cache);
1081     }
1082     g_free(s->cluster_cache);
1083     qemu_vfree(s->cluster_data);
1084     return ret;
1085 }
1086 
1087 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1088 {
1089     BDRVQcowState *s = bs->opaque;
1090 
1091     bs->bl.write_zeroes_alignment = s->cluster_sectors;
1092 }
1093 
1094 static int qcow2_set_key(BlockDriverState *bs, const char *key)
1095 {
1096     BDRVQcowState *s = bs->opaque;
1097     uint8_t keybuf[16];
1098     int len, i;
1099     Error *err = NULL;
1100 
1101     memset(keybuf, 0, 16);
1102     len = strlen(key);
1103     if (len > 16)
1104         len = 16;
1105     /* XXX: we could compress the chars to 7 bits to increase
1106        entropy */
1107     for(i = 0;i < len;i++) {
1108         keybuf[i] = key[i];
1109     }
1110     assert(bs->encrypted);
1111 
1112     qcrypto_cipher_free(s->cipher);
1113     s->cipher = qcrypto_cipher_new(
1114         QCRYPTO_CIPHER_ALG_AES_128,
1115         QCRYPTO_CIPHER_MODE_CBC,
1116         keybuf, G_N_ELEMENTS(keybuf),
1117         &err);
1118 
1119     if (!s->cipher) {
1120         /* XXX would be nice if errors in this method could
1121          * be properly propagate to the caller. Would need
1122          * the bdrv_set_key() API signature to be fixed. */
1123         error_free(err);
1124         return -1;
1125     }
1126     return 0;
1127 }
1128 
1129 /* We have no actual commit/abort logic for qcow2, but we need to write out any
1130  * unwritten data if we reopen read-only. */
1131 static int qcow2_reopen_prepare(BDRVReopenState *state,
1132                                 BlockReopenQueue *queue, Error **errp)
1133 {
1134     int ret;
1135 
1136     if ((state->flags & BDRV_O_RDWR) == 0) {
1137         ret = bdrv_flush(state->bs);
1138         if (ret < 0) {
1139             return ret;
1140         }
1141 
1142         ret = qcow2_mark_clean(state->bs);
1143         if (ret < 0) {
1144             return ret;
1145         }
1146     }
1147 
1148     return 0;
1149 }
1150 
1151 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1152         int64_t sector_num, int nb_sectors, int *pnum)
1153 {
1154     BDRVQcowState *s = bs->opaque;
1155     uint64_t cluster_offset;
1156     int index_in_cluster, ret;
1157     int64_t status = 0;
1158 
1159     *pnum = nb_sectors;
1160     qemu_co_mutex_lock(&s->lock);
1161     ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
1162     qemu_co_mutex_unlock(&s->lock);
1163     if (ret < 0) {
1164         return ret;
1165     }
1166 
1167     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1168         !s->cipher) {
1169         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1170         cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1171         status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1172     }
1173     if (ret == QCOW2_CLUSTER_ZERO) {
1174         status |= BDRV_BLOCK_ZERO;
1175     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1176         status |= BDRV_BLOCK_DATA;
1177     }
1178     return status;
1179 }
1180 
1181 /* handle reading after the end of the backing file */
1182 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
1183                   int64_t sector_num, int nb_sectors)
1184 {
1185     int n1;
1186     if ((sector_num + nb_sectors) <= bs->total_sectors)
1187         return nb_sectors;
1188     if (sector_num >= bs->total_sectors)
1189         n1 = 0;
1190     else
1191         n1 = bs->total_sectors - sector_num;
1192 
1193     qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1));
1194 
1195     return n1;
1196 }
1197 
1198 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
1199                           int remaining_sectors, QEMUIOVector *qiov)
1200 {
1201     BDRVQcowState *s = bs->opaque;
1202     int index_in_cluster, n1;
1203     int ret;
1204     int cur_nr_sectors; /* number of sectors in current iteration */
1205     uint64_t cluster_offset = 0;
1206     uint64_t bytes_done = 0;
1207     QEMUIOVector hd_qiov;
1208     uint8_t *cluster_data = NULL;
1209 
1210     qemu_iovec_init(&hd_qiov, qiov->niov);
1211 
1212     qemu_co_mutex_lock(&s->lock);
1213 
1214     while (remaining_sectors != 0) {
1215 
1216         /* prepare next request */
1217         cur_nr_sectors = remaining_sectors;
1218         if (s->cipher) {
1219             cur_nr_sectors = MIN(cur_nr_sectors,
1220                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1221         }
1222 
1223         ret = qcow2_get_cluster_offset(bs, sector_num << 9,
1224             &cur_nr_sectors, &cluster_offset);
1225         if (ret < 0) {
1226             goto fail;
1227         }
1228 
1229         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1230 
1231         qemu_iovec_reset(&hd_qiov);
1232         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1233             cur_nr_sectors * 512);
1234 
1235         switch (ret) {
1236         case QCOW2_CLUSTER_UNALLOCATED:
1237 
1238             if (bs->backing_hd) {
1239                 /* read from the base image */
1240                 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov,
1241                     sector_num, cur_nr_sectors);
1242                 if (n1 > 0) {
1243                     QEMUIOVector local_qiov;
1244 
1245                     qemu_iovec_init(&local_qiov, hd_qiov.niov);
1246                     qemu_iovec_concat(&local_qiov, &hd_qiov, 0,
1247                                       n1 * BDRV_SECTOR_SIZE);
1248 
1249                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1250                     qemu_co_mutex_unlock(&s->lock);
1251                     ret = bdrv_co_readv(bs->backing_hd, sector_num,
1252                                         n1, &local_qiov);
1253                     qemu_co_mutex_lock(&s->lock);
1254 
1255                     qemu_iovec_destroy(&local_qiov);
1256 
1257                     if (ret < 0) {
1258                         goto fail;
1259                     }
1260                 }
1261             } else {
1262                 /* Note: in this case, no need to wait */
1263                 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1264             }
1265             break;
1266 
1267         case QCOW2_CLUSTER_ZERO:
1268             qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1269             break;
1270 
1271         case QCOW2_CLUSTER_COMPRESSED:
1272             /* add AIO support for compressed blocks ? */
1273             ret = qcow2_decompress_cluster(bs, cluster_offset);
1274             if (ret < 0) {
1275                 goto fail;
1276             }
1277 
1278             qemu_iovec_from_buf(&hd_qiov, 0,
1279                 s->cluster_cache + index_in_cluster * 512,
1280                 512 * cur_nr_sectors);
1281             break;
1282 
1283         case QCOW2_CLUSTER_NORMAL:
1284             if ((cluster_offset & 511) != 0) {
1285                 ret = -EIO;
1286                 goto fail;
1287             }
1288 
1289             if (bs->encrypted) {
1290                 assert(s->cipher);
1291 
1292                 /*
1293                  * For encrypted images, read everything into a temporary
1294                  * contiguous buffer on which the AES functions can work.
1295                  */
1296                 if (!cluster_data) {
1297                     cluster_data =
1298                         qemu_try_blockalign(bs->file, QCOW_MAX_CRYPT_CLUSTERS
1299                                                       * s->cluster_size);
1300                     if (cluster_data == NULL) {
1301                         ret = -ENOMEM;
1302                         goto fail;
1303                     }
1304                 }
1305 
1306                 assert(cur_nr_sectors <=
1307                     QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1308                 qemu_iovec_reset(&hd_qiov);
1309                 qemu_iovec_add(&hd_qiov, cluster_data,
1310                     512 * cur_nr_sectors);
1311             }
1312 
1313             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1314             qemu_co_mutex_unlock(&s->lock);
1315             ret = bdrv_co_readv(bs->file,
1316                                 (cluster_offset >> 9) + index_in_cluster,
1317                                 cur_nr_sectors, &hd_qiov);
1318             qemu_co_mutex_lock(&s->lock);
1319             if (ret < 0) {
1320                 goto fail;
1321             }
1322             if (bs->encrypted) {
1323                 assert(s->cipher);
1324                 Error *err = NULL;
1325                 if (qcow2_encrypt_sectors(s, sector_num,  cluster_data,
1326                                           cluster_data, cur_nr_sectors, false,
1327                                           &err) < 0) {
1328                     error_free(err);
1329                     ret = -EIO;
1330                     goto fail;
1331                 }
1332                 qemu_iovec_from_buf(qiov, bytes_done,
1333                     cluster_data, 512 * cur_nr_sectors);
1334             }
1335             break;
1336 
1337         default:
1338             g_assert_not_reached();
1339             ret = -EIO;
1340             goto fail;
1341         }
1342 
1343         remaining_sectors -= cur_nr_sectors;
1344         sector_num += cur_nr_sectors;
1345         bytes_done += cur_nr_sectors * 512;
1346     }
1347     ret = 0;
1348 
1349 fail:
1350     qemu_co_mutex_unlock(&s->lock);
1351 
1352     qemu_iovec_destroy(&hd_qiov);
1353     qemu_vfree(cluster_data);
1354 
1355     return ret;
1356 }
1357 
1358 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
1359                            int64_t sector_num,
1360                            int remaining_sectors,
1361                            QEMUIOVector *qiov)
1362 {
1363     BDRVQcowState *s = bs->opaque;
1364     int index_in_cluster;
1365     int ret;
1366     int cur_nr_sectors; /* number of sectors in current iteration */
1367     uint64_t cluster_offset;
1368     QEMUIOVector hd_qiov;
1369     uint64_t bytes_done = 0;
1370     uint8_t *cluster_data = NULL;
1371     QCowL2Meta *l2meta = NULL;
1372 
1373     trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
1374                                  remaining_sectors);
1375 
1376     qemu_iovec_init(&hd_qiov, qiov->niov);
1377 
1378     s->cluster_cache_offset = -1; /* disable compressed cache */
1379 
1380     qemu_co_mutex_lock(&s->lock);
1381 
1382     while (remaining_sectors != 0) {
1383 
1384         l2meta = NULL;
1385 
1386         trace_qcow2_writev_start_part(qemu_coroutine_self());
1387         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1388         cur_nr_sectors = remaining_sectors;
1389         if (bs->encrypted &&
1390             cur_nr_sectors >
1391             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster) {
1392             cur_nr_sectors =
1393                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster;
1394         }
1395 
1396         ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
1397             &cur_nr_sectors, &cluster_offset, &l2meta);
1398         if (ret < 0) {
1399             goto fail;
1400         }
1401 
1402         assert((cluster_offset & 511) == 0);
1403 
1404         qemu_iovec_reset(&hd_qiov);
1405         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1406             cur_nr_sectors * 512);
1407 
1408         if (bs->encrypted) {
1409             Error *err = NULL;
1410             assert(s->cipher);
1411             if (!cluster_data) {
1412                 cluster_data = qemu_try_blockalign(bs->file,
1413                                                    QCOW_MAX_CRYPT_CLUSTERS
1414                                                    * s->cluster_size);
1415                 if (cluster_data == NULL) {
1416                     ret = -ENOMEM;
1417                     goto fail;
1418                 }
1419             }
1420 
1421             assert(hd_qiov.size <=
1422                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1423             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1424 
1425             if (qcow2_encrypt_sectors(s, sector_num, cluster_data,
1426                                       cluster_data, cur_nr_sectors,
1427                                       true, &err) < 0) {
1428                 error_free(err);
1429                 ret = -EIO;
1430                 goto fail;
1431             }
1432 
1433             qemu_iovec_reset(&hd_qiov);
1434             qemu_iovec_add(&hd_qiov, cluster_data,
1435                 cur_nr_sectors * 512);
1436         }
1437 
1438         ret = qcow2_pre_write_overlap_check(bs, 0,
1439                 cluster_offset + index_in_cluster * BDRV_SECTOR_SIZE,
1440                 cur_nr_sectors * BDRV_SECTOR_SIZE);
1441         if (ret < 0) {
1442             goto fail;
1443         }
1444 
1445         qemu_co_mutex_unlock(&s->lock);
1446         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1447         trace_qcow2_writev_data(qemu_coroutine_self(),
1448                                 (cluster_offset >> 9) + index_in_cluster);
1449         ret = bdrv_co_writev(bs->file,
1450                              (cluster_offset >> 9) + index_in_cluster,
1451                              cur_nr_sectors, &hd_qiov);
1452         qemu_co_mutex_lock(&s->lock);
1453         if (ret < 0) {
1454             goto fail;
1455         }
1456 
1457         while (l2meta != NULL) {
1458             QCowL2Meta *next;
1459 
1460             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1461             if (ret < 0) {
1462                 goto fail;
1463             }
1464 
1465             /* Take the request off the list of running requests */
1466             if (l2meta->nb_clusters != 0) {
1467                 QLIST_REMOVE(l2meta, next_in_flight);
1468             }
1469 
1470             qemu_co_queue_restart_all(&l2meta->dependent_requests);
1471 
1472             next = l2meta->next;
1473             g_free(l2meta);
1474             l2meta = next;
1475         }
1476 
1477         remaining_sectors -= cur_nr_sectors;
1478         sector_num += cur_nr_sectors;
1479         bytes_done += cur_nr_sectors * 512;
1480         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
1481     }
1482     ret = 0;
1483 
1484 fail:
1485     qemu_co_mutex_unlock(&s->lock);
1486 
1487     while (l2meta != NULL) {
1488         QCowL2Meta *next;
1489 
1490         if (l2meta->nb_clusters != 0) {
1491             QLIST_REMOVE(l2meta, next_in_flight);
1492         }
1493         qemu_co_queue_restart_all(&l2meta->dependent_requests);
1494 
1495         next = l2meta->next;
1496         g_free(l2meta);
1497         l2meta = next;
1498     }
1499 
1500     qemu_iovec_destroy(&hd_qiov);
1501     qemu_vfree(cluster_data);
1502     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
1503 
1504     return ret;
1505 }
1506 
1507 static void qcow2_close(BlockDriverState *bs)
1508 {
1509     BDRVQcowState *s = bs->opaque;
1510     qemu_vfree(s->l1_table);
1511     /* else pre-write overlap checks in cache_destroy may crash */
1512     s->l1_table = NULL;
1513 
1514     if (!(bs->open_flags & BDRV_O_INCOMING)) {
1515         int ret1, ret2;
1516 
1517         ret1 = qcow2_cache_flush(bs, s->l2_table_cache);
1518         ret2 = qcow2_cache_flush(bs, s->refcount_block_cache);
1519 
1520         if (ret1) {
1521             error_report("Failed to flush the L2 table cache: %s",
1522                          strerror(-ret1));
1523         }
1524         if (ret2) {
1525             error_report("Failed to flush the refcount block cache: %s",
1526                          strerror(-ret2));
1527         }
1528 
1529         if (!ret1 && !ret2) {
1530             qcow2_mark_clean(bs);
1531         }
1532     }
1533 
1534     cache_clean_timer_del(bs);
1535     qcow2_cache_destroy(bs, s->l2_table_cache);
1536     qcow2_cache_destroy(bs, s->refcount_block_cache);
1537 
1538     qcrypto_cipher_free(s->cipher);
1539     s->cipher = NULL;
1540 
1541     g_free(s->unknown_header_fields);
1542     cleanup_unknown_header_ext(bs);
1543 
1544     g_free(s->image_backing_file);
1545     g_free(s->image_backing_format);
1546 
1547     g_free(s->cluster_cache);
1548     qemu_vfree(s->cluster_data);
1549     qcow2_refcount_close(bs);
1550     qcow2_free_snapshots(bs);
1551 }
1552 
1553 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
1554 {
1555     BDRVQcowState *s = bs->opaque;
1556     int flags = s->flags;
1557     QCryptoCipher *cipher = NULL;
1558     QDict *options;
1559     Error *local_err = NULL;
1560     int ret;
1561 
1562     /*
1563      * Backing files are read-only which makes all of their metadata immutable,
1564      * that means we don't have to worry about reopening them here.
1565      */
1566 
1567     cipher = s->cipher;
1568     s->cipher = NULL;
1569 
1570     qcow2_close(bs);
1571 
1572     bdrv_invalidate_cache(bs->file, &local_err);
1573     if (local_err) {
1574         error_propagate(errp, local_err);
1575         return;
1576     }
1577 
1578     memset(s, 0, sizeof(BDRVQcowState));
1579     options = qdict_clone_shallow(bs->options);
1580 
1581     ret = qcow2_open(bs, options, flags, &local_err);
1582     QDECREF(options);
1583     if (local_err) {
1584         error_setg(errp, "Could not reopen qcow2 layer: %s",
1585                    error_get_pretty(local_err));
1586         error_free(local_err);
1587         return;
1588     } else if (ret < 0) {
1589         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
1590         return;
1591     }
1592 
1593     s->cipher = cipher;
1594 }
1595 
1596 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
1597     size_t len, size_t buflen)
1598 {
1599     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
1600     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
1601 
1602     if (buflen < ext_len) {
1603         return -ENOSPC;
1604     }
1605 
1606     *ext_backing_fmt = (QCowExtension) {
1607         .magic  = cpu_to_be32(magic),
1608         .len    = cpu_to_be32(len),
1609     };
1610     memcpy(buf + sizeof(QCowExtension), s, len);
1611 
1612     return ext_len;
1613 }
1614 
1615 /*
1616  * Updates the qcow2 header, including the variable length parts of it, i.e.
1617  * the backing file name and all extensions. qcow2 was not designed to allow
1618  * such changes, so if we run out of space (we can only use the first cluster)
1619  * this function may fail.
1620  *
1621  * Returns 0 on success, -errno in error cases.
1622  */
1623 int qcow2_update_header(BlockDriverState *bs)
1624 {
1625     BDRVQcowState *s = bs->opaque;
1626     QCowHeader *header;
1627     char *buf;
1628     size_t buflen = s->cluster_size;
1629     int ret;
1630     uint64_t total_size;
1631     uint32_t refcount_table_clusters;
1632     size_t header_length;
1633     Qcow2UnknownHeaderExtension *uext;
1634 
1635     buf = qemu_blockalign(bs, buflen);
1636 
1637     /* Header structure */
1638     header = (QCowHeader*) buf;
1639 
1640     if (buflen < sizeof(*header)) {
1641         ret = -ENOSPC;
1642         goto fail;
1643     }
1644 
1645     header_length = sizeof(*header) + s->unknown_header_fields_size;
1646     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1647     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1648 
1649     *header = (QCowHeader) {
1650         /* Version 2 fields */
1651         .magic                  = cpu_to_be32(QCOW_MAGIC),
1652         .version                = cpu_to_be32(s->qcow_version),
1653         .backing_file_offset    = 0,
1654         .backing_file_size      = 0,
1655         .cluster_bits           = cpu_to_be32(s->cluster_bits),
1656         .size                   = cpu_to_be64(total_size),
1657         .crypt_method           = cpu_to_be32(s->crypt_method_header),
1658         .l1_size                = cpu_to_be32(s->l1_size),
1659         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
1660         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
1661         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1662         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
1663         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
1664 
1665         /* Version 3 fields */
1666         .incompatible_features  = cpu_to_be64(s->incompatible_features),
1667         .compatible_features    = cpu_to_be64(s->compatible_features),
1668         .autoclear_features     = cpu_to_be64(s->autoclear_features),
1669         .refcount_order         = cpu_to_be32(s->refcount_order),
1670         .header_length          = cpu_to_be32(header_length),
1671     };
1672 
1673     /* For older versions, write a shorter header */
1674     switch (s->qcow_version) {
1675     case 2:
1676         ret = offsetof(QCowHeader, incompatible_features);
1677         break;
1678     case 3:
1679         ret = sizeof(*header);
1680         break;
1681     default:
1682         ret = -EINVAL;
1683         goto fail;
1684     }
1685 
1686     buf += ret;
1687     buflen -= ret;
1688     memset(buf, 0, buflen);
1689 
1690     /* Preserve any unknown field in the header */
1691     if (s->unknown_header_fields_size) {
1692         if (buflen < s->unknown_header_fields_size) {
1693             ret = -ENOSPC;
1694             goto fail;
1695         }
1696 
1697         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1698         buf += s->unknown_header_fields_size;
1699         buflen -= s->unknown_header_fields_size;
1700     }
1701 
1702     /* Backing file format header extension */
1703     if (s->image_backing_format) {
1704         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1705                              s->image_backing_format,
1706                              strlen(s->image_backing_format),
1707                              buflen);
1708         if (ret < 0) {
1709             goto fail;
1710         }
1711 
1712         buf += ret;
1713         buflen -= ret;
1714     }
1715 
1716     /* Feature table */
1717     Qcow2Feature features[] = {
1718         {
1719             .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1720             .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
1721             .name = "dirty bit",
1722         },
1723         {
1724             .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1725             .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
1726             .name = "corrupt bit",
1727         },
1728         {
1729             .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1730             .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1731             .name = "lazy refcounts",
1732         },
1733     };
1734 
1735     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1736                          features, sizeof(features), buflen);
1737     if (ret < 0) {
1738         goto fail;
1739     }
1740     buf += ret;
1741     buflen -= ret;
1742 
1743     /* Keep unknown header extensions */
1744     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1745         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1746         if (ret < 0) {
1747             goto fail;
1748         }
1749 
1750         buf += ret;
1751         buflen -= ret;
1752     }
1753 
1754     /* End of header extensions */
1755     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1756     if (ret < 0) {
1757         goto fail;
1758     }
1759 
1760     buf += ret;
1761     buflen -= ret;
1762 
1763     /* Backing file name */
1764     if (s->image_backing_file) {
1765         size_t backing_file_len = strlen(s->image_backing_file);
1766 
1767         if (buflen < backing_file_len) {
1768             ret = -ENOSPC;
1769             goto fail;
1770         }
1771 
1772         /* Using strncpy is ok here, since buf is not NUL-terminated. */
1773         strncpy(buf, s->image_backing_file, buflen);
1774 
1775         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1776         header->backing_file_size   = cpu_to_be32(backing_file_len);
1777     }
1778 
1779     /* Write the new header */
1780     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
1781     if (ret < 0) {
1782         goto fail;
1783     }
1784 
1785     ret = 0;
1786 fail:
1787     qemu_vfree(header);
1788     return ret;
1789 }
1790 
1791 static int qcow2_change_backing_file(BlockDriverState *bs,
1792     const char *backing_file, const char *backing_fmt)
1793 {
1794     BDRVQcowState *s = bs->opaque;
1795 
1796     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1797     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1798 
1799     g_free(s->image_backing_file);
1800     g_free(s->image_backing_format);
1801 
1802     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
1803     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
1804 
1805     return qcow2_update_header(bs);
1806 }
1807 
1808 static int preallocate(BlockDriverState *bs)
1809 {
1810     uint64_t nb_sectors;
1811     uint64_t offset;
1812     uint64_t host_offset = 0;
1813     int num;
1814     int ret;
1815     QCowL2Meta *meta;
1816 
1817     nb_sectors = bdrv_nb_sectors(bs);
1818     offset = 0;
1819 
1820     while (nb_sectors) {
1821         num = MIN(nb_sectors, INT_MAX >> BDRV_SECTOR_BITS);
1822         ret = qcow2_alloc_cluster_offset(bs, offset, &num,
1823                                          &host_offset, &meta);
1824         if (ret < 0) {
1825             return ret;
1826         }
1827 
1828         while (meta) {
1829             QCowL2Meta *next = meta->next;
1830 
1831             ret = qcow2_alloc_cluster_link_l2(bs, meta);
1832             if (ret < 0) {
1833                 qcow2_free_any_clusters(bs, meta->alloc_offset,
1834                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
1835                 return ret;
1836             }
1837 
1838             /* There are no dependent requests, but we need to remove our
1839              * request from the list of in-flight requests */
1840             QLIST_REMOVE(meta, next_in_flight);
1841 
1842             g_free(meta);
1843             meta = next;
1844         }
1845 
1846         /* TODO Preallocate data if requested */
1847 
1848         nb_sectors -= num;
1849         offset += num << BDRV_SECTOR_BITS;
1850     }
1851 
1852     /*
1853      * It is expected that the image file is large enough to actually contain
1854      * all of the allocated clusters (otherwise we get failing reads after
1855      * EOF). Extend the image to the last allocated sector.
1856      */
1857     if (host_offset != 0) {
1858         uint8_t buf[BDRV_SECTOR_SIZE];
1859         memset(buf, 0, BDRV_SECTOR_SIZE);
1860         ret = bdrv_write(bs->file, (host_offset >> BDRV_SECTOR_BITS) + num - 1,
1861                          buf, 1);
1862         if (ret < 0) {
1863             return ret;
1864         }
1865     }
1866 
1867     return 0;
1868 }
1869 
1870 static int qcow2_create2(const char *filename, int64_t total_size,
1871                          const char *backing_file, const char *backing_format,
1872                          int flags, size_t cluster_size, PreallocMode prealloc,
1873                          QemuOpts *opts, int version, int refcount_order,
1874                          Error **errp)
1875 {
1876     /* Calculate cluster_bits */
1877     int cluster_bits;
1878     cluster_bits = ctz32(cluster_size);
1879     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
1880         (1 << cluster_bits) != cluster_size)
1881     {
1882         error_setg(errp, "Cluster size must be a power of two between %d and "
1883                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
1884         return -EINVAL;
1885     }
1886 
1887     /*
1888      * Open the image file and write a minimal qcow2 header.
1889      *
1890      * We keep things simple and start with a zero-sized image. We also
1891      * do without refcount blocks or a L1 table for now. We'll fix the
1892      * inconsistency later.
1893      *
1894      * We do need a refcount table because growing the refcount table means
1895      * allocating two new refcount blocks - the seconds of which would be at
1896      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
1897      * size for any qcow2 image.
1898      */
1899     BlockDriverState* bs;
1900     QCowHeader *header;
1901     uint64_t* refcount_table;
1902     Error *local_err = NULL;
1903     int ret;
1904 
1905     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
1906         /* Note: The following calculation does not need to be exact; if it is a
1907          * bit off, either some bytes will be "leaked" (which is fine) or we
1908          * will need to increase the file size by some bytes (which is fine,
1909          * too, as long as the bulk is allocated here). Therefore, using
1910          * floating point arithmetic is fine. */
1911         int64_t meta_size = 0;
1912         uint64_t nreftablee, nrefblocke, nl1e, nl2e;
1913         int64_t aligned_total_size = align_offset(total_size, cluster_size);
1914         int refblock_bits, refblock_size;
1915         /* refcount entry size in bytes */
1916         double rces = (1 << refcount_order) / 8.;
1917 
1918         /* see qcow2_open() */
1919         refblock_bits = cluster_bits - (refcount_order - 3);
1920         refblock_size = 1 << refblock_bits;
1921 
1922         /* header: 1 cluster */
1923         meta_size += cluster_size;
1924 
1925         /* total size of L2 tables */
1926         nl2e = aligned_total_size / cluster_size;
1927         nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
1928         meta_size += nl2e * sizeof(uint64_t);
1929 
1930         /* total size of L1 tables */
1931         nl1e = nl2e * sizeof(uint64_t) / cluster_size;
1932         nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
1933         meta_size += nl1e * sizeof(uint64_t);
1934 
1935         /* total size of refcount blocks
1936          *
1937          * note: every host cluster is reference-counted, including metadata
1938          * (even refcount blocks are recursively included).
1939          * Let:
1940          *   a = total_size (this is the guest disk size)
1941          *   m = meta size not including refcount blocks and refcount tables
1942          *   c = cluster size
1943          *   y1 = number of refcount blocks entries
1944          *   y2 = meta size including everything
1945          *   rces = refcount entry size in bytes
1946          * then,
1947          *   y1 = (y2 + a)/c
1948          *   y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m
1949          * we can get y1:
1950          *   y1 = (a + m) / (c - rces - rces * sizeof(u64) / c)
1951          */
1952         nrefblocke = (aligned_total_size + meta_size + cluster_size)
1953                    / (cluster_size - rces - rces * sizeof(uint64_t)
1954                                                  / cluster_size);
1955         meta_size += DIV_ROUND_UP(nrefblocke, refblock_size) * cluster_size;
1956 
1957         /* total size of refcount tables */
1958         nreftablee = nrefblocke / refblock_size;
1959         nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t));
1960         meta_size += nreftablee * sizeof(uint64_t);
1961 
1962         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
1963                             aligned_total_size + meta_size, &error_abort);
1964         qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc],
1965                      &error_abort);
1966     }
1967 
1968     ret = bdrv_create_file(filename, opts, &local_err);
1969     if (ret < 0) {
1970         error_propagate(errp, local_err);
1971         return ret;
1972     }
1973 
1974     bs = NULL;
1975     ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1976                     NULL, &local_err);
1977     if (ret < 0) {
1978         error_propagate(errp, local_err);
1979         return ret;
1980     }
1981 
1982     /* Write the header */
1983     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
1984     header = g_malloc0(cluster_size);
1985     *header = (QCowHeader) {
1986         .magic                      = cpu_to_be32(QCOW_MAGIC),
1987         .version                    = cpu_to_be32(version),
1988         .cluster_bits               = cpu_to_be32(cluster_bits),
1989         .size                       = cpu_to_be64(0),
1990         .l1_table_offset            = cpu_to_be64(0),
1991         .l1_size                    = cpu_to_be32(0),
1992         .refcount_table_offset      = cpu_to_be64(cluster_size),
1993         .refcount_table_clusters    = cpu_to_be32(1),
1994         .refcount_order             = cpu_to_be32(refcount_order),
1995         .header_length              = cpu_to_be32(sizeof(*header)),
1996     };
1997 
1998     if (flags & BLOCK_FLAG_ENCRYPT) {
1999         header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
2000     } else {
2001         header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
2002     }
2003 
2004     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
2005         header->compatible_features |=
2006             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
2007     }
2008 
2009     ret = bdrv_pwrite(bs, 0, header, cluster_size);
2010     g_free(header);
2011     if (ret < 0) {
2012         error_setg_errno(errp, -ret, "Could not write qcow2 header");
2013         goto out;
2014     }
2015 
2016     /* Write a refcount table with one refcount block */
2017     refcount_table = g_malloc0(2 * cluster_size);
2018     refcount_table[0] = cpu_to_be64(2 * cluster_size);
2019     ret = bdrv_pwrite(bs, cluster_size, refcount_table, 2 * cluster_size);
2020     g_free(refcount_table);
2021 
2022     if (ret < 0) {
2023         error_setg_errno(errp, -ret, "Could not write refcount table");
2024         goto out;
2025     }
2026 
2027     bdrv_unref(bs);
2028     bs = NULL;
2029 
2030     /*
2031      * And now open the image and make it consistent first (i.e. increase the
2032      * refcount of the cluster that is occupied by the header and the refcount
2033      * table)
2034      */
2035     ret = bdrv_open(&bs, filename, NULL, NULL,
2036                     BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH,
2037                     &bdrv_qcow2, &local_err);
2038     if (ret < 0) {
2039         error_propagate(errp, local_err);
2040         goto out;
2041     }
2042 
2043     ret = qcow2_alloc_clusters(bs, 3 * cluster_size);
2044     if (ret < 0) {
2045         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
2046                          "header and refcount table");
2047         goto out;
2048 
2049     } else if (ret != 0) {
2050         error_report("Huh, first cluster in empty image is already in use?");
2051         abort();
2052     }
2053 
2054     /* Okay, now that we have a valid image, let's give it the right size */
2055     ret = bdrv_truncate(bs, total_size);
2056     if (ret < 0) {
2057         error_setg_errno(errp, -ret, "Could not resize image");
2058         goto out;
2059     }
2060 
2061     /* Want a backing file? There you go.*/
2062     if (backing_file) {
2063         ret = bdrv_change_backing_file(bs, backing_file, backing_format);
2064         if (ret < 0) {
2065             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
2066                              "with format '%s'", backing_file, backing_format);
2067             goto out;
2068         }
2069     }
2070 
2071     /* And if we're supposed to preallocate metadata, do that now */
2072     if (prealloc != PREALLOC_MODE_OFF) {
2073         BDRVQcowState *s = bs->opaque;
2074         qemu_co_mutex_lock(&s->lock);
2075         ret = preallocate(bs);
2076         qemu_co_mutex_unlock(&s->lock);
2077         if (ret < 0) {
2078             error_setg_errno(errp, -ret, "Could not preallocate metadata");
2079             goto out;
2080         }
2081     }
2082 
2083     bdrv_unref(bs);
2084     bs = NULL;
2085 
2086     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
2087     ret = bdrv_open(&bs, filename, NULL, NULL,
2088                     BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_BACKING,
2089                     &bdrv_qcow2, &local_err);
2090     if (local_err) {
2091         error_propagate(errp, local_err);
2092         goto out;
2093     }
2094 
2095     ret = 0;
2096 out:
2097     if (bs) {
2098         bdrv_unref(bs);
2099     }
2100     return ret;
2101 }
2102 
2103 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
2104 {
2105     char *backing_file = NULL;
2106     char *backing_fmt = NULL;
2107     char *buf = NULL;
2108     uint64_t size = 0;
2109     int flags = 0;
2110     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
2111     PreallocMode prealloc;
2112     int version = 3;
2113     uint64_t refcount_bits = 16;
2114     int refcount_order;
2115     Error *local_err = NULL;
2116     int ret;
2117 
2118     /* Read out options */
2119     size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2120                     BDRV_SECTOR_SIZE);
2121     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2122     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2123     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2124         flags |= BLOCK_FLAG_ENCRYPT;
2125     }
2126     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2127                                          DEFAULT_CLUSTER_SIZE);
2128     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2129     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
2130                                PREALLOC_MODE_MAX, PREALLOC_MODE_OFF,
2131                                &local_err);
2132     if (local_err) {
2133         error_propagate(errp, local_err);
2134         ret = -EINVAL;
2135         goto finish;
2136     }
2137     g_free(buf);
2138     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2139     if (!buf) {
2140         /* keep the default */
2141     } else if (!strcmp(buf, "0.10")) {
2142         version = 2;
2143     } else if (!strcmp(buf, "1.1")) {
2144         version = 3;
2145     } else {
2146         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2147         ret = -EINVAL;
2148         goto finish;
2149     }
2150 
2151     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2152         flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2153     }
2154 
2155     if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2156         error_setg(errp, "Backing file and preallocation cannot be used at "
2157                    "the same time");
2158         ret = -EINVAL;
2159         goto finish;
2160     }
2161 
2162     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2163         error_setg(errp, "Lazy refcounts only supported with compatibility "
2164                    "level 1.1 and above (use compat=1.1 or greater)");
2165         ret = -EINVAL;
2166         goto finish;
2167     }
2168 
2169     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS,
2170                                             refcount_bits);
2171     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2172         error_setg(errp, "Refcount width must be a power of two and may not "
2173                    "exceed 64 bits");
2174         ret = -EINVAL;
2175         goto finish;
2176     }
2177 
2178     if (version < 3 && refcount_bits != 16) {
2179         error_setg(errp, "Different refcount widths than 16 bits require "
2180                    "compatibility level 1.1 or above (use compat=1.1 or "
2181                    "greater)");
2182         ret = -EINVAL;
2183         goto finish;
2184     }
2185 
2186     refcount_order = ctz32(refcount_bits);
2187 
2188     ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2189                         cluster_size, prealloc, opts, version, refcount_order,
2190                         &local_err);
2191     if (local_err) {
2192         error_propagate(errp, local_err);
2193     }
2194 
2195 finish:
2196     g_free(backing_file);
2197     g_free(backing_fmt);
2198     g_free(buf);
2199     return ret;
2200 }
2201 
2202 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
2203     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2204 {
2205     int ret;
2206     BDRVQcowState *s = bs->opaque;
2207 
2208     /* Emulate misaligned zero writes */
2209     if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
2210         return -ENOTSUP;
2211     }
2212 
2213     /* Whatever is left can use real zero clusters */
2214     qemu_co_mutex_lock(&s->lock);
2215     ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2216         nb_sectors);
2217     qemu_co_mutex_unlock(&s->lock);
2218 
2219     return ret;
2220 }
2221 
2222 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
2223     int64_t sector_num, int nb_sectors)
2224 {
2225     int ret;
2226     BDRVQcowState *s = bs->opaque;
2227 
2228     qemu_co_mutex_lock(&s->lock);
2229     ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2230         nb_sectors, QCOW2_DISCARD_REQUEST, false);
2231     qemu_co_mutex_unlock(&s->lock);
2232     return ret;
2233 }
2234 
2235 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
2236 {
2237     BDRVQcowState *s = bs->opaque;
2238     int64_t new_l1_size;
2239     int ret;
2240 
2241     if (offset & 511) {
2242         error_report("The new size must be a multiple of 512");
2243         return -EINVAL;
2244     }
2245 
2246     /* cannot proceed if image has snapshots */
2247     if (s->nb_snapshots) {
2248         error_report("Can't resize an image which has snapshots");
2249         return -ENOTSUP;
2250     }
2251 
2252     /* shrinking is currently not supported */
2253     if (offset < bs->total_sectors * 512) {
2254         error_report("qcow2 doesn't support shrinking images yet");
2255         return -ENOTSUP;
2256     }
2257 
2258     new_l1_size = size_to_l1(s, offset);
2259     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
2260     if (ret < 0) {
2261         return ret;
2262     }
2263 
2264     /* write updated header.size */
2265     offset = cpu_to_be64(offset);
2266     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
2267                            &offset, sizeof(uint64_t));
2268     if (ret < 0) {
2269         return ret;
2270     }
2271 
2272     s->l1_vm_state_index = new_l1_size;
2273     return 0;
2274 }
2275 
2276 /* XXX: put compressed sectors first, then all the cluster aligned
2277    tables to avoid losing bytes in alignment */
2278 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
2279                                   const uint8_t *buf, int nb_sectors)
2280 {
2281     BDRVQcowState *s = bs->opaque;
2282     z_stream strm;
2283     int ret, out_len;
2284     uint8_t *out_buf;
2285     uint64_t cluster_offset;
2286 
2287     if (nb_sectors == 0) {
2288         /* align end of file to a sector boundary to ease reading with
2289            sector based I/Os */
2290         cluster_offset = bdrv_getlength(bs->file);
2291         return bdrv_truncate(bs->file, cluster_offset);
2292     }
2293 
2294     if (nb_sectors != s->cluster_sectors) {
2295         ret = -EINVAL;
2296 
2297         /* Zero-pad last write if image size is not cluster aligned */
2298         if (sector_num + nb_sectors == bs->total_sectors &&
2299             nb_sectors < s->cluster_sectors) {
2300             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
2301             memset(pad_buf, 0, s->cluster_size);
2302             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
2303             ret = qcow2_write_compressed(bs, sector_num,
2304                                          pad_buf, s->cluster_sectors);
2305             qemu_vfree(pad_buf);
2306         }
2307         return ret;
2308     }
2309 
2310     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
2311 
2312     /* best compression, small window, no zlib header */
2313     memset(&strm, 0, sizeof(strm));
2314     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
2315                        Z_DEFLATED, -12,
2316                        9, Z_DEFAULT_STRATEGY);
2317     if (ret != 0) {
2318         ret = -EINVAL;
2319         goto fail;
2320     }
2321 
2322     strm.avail_in = s->cluster_size;
2323     strm.next_in = (uint8_t *)buf;
2324     strm.avail_out = s->cluster_size;
2325     strm.next_out = out_buf;
2326 
2327     ret = deflate(&strm, Z_FINISH);
2328     if (ret != Z_STREAM_END && ret != Z_OK) {
2329         deflateEnd(&strm);
2330         ret = -EINVAL;
2331         goto fail;
2332     }
2333     out_len = strm.next_out - out_buf;
2334 
2335     deflateEnd(&strm);
2336 
2337     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
2338         /* could not compress: write normal cluster */
2339         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
2340         if (ret < 0) {
2341             goto fail;
2342         }
2343     } else {
2344         cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
2345             sector_num << 9, out_len);
2346         if (!cluster_offset) {
2347             ret = -EIO;
2348             goto fail;
2349         }
2350         cluster_offset &= s->cluster_offset_mask;
2351 
2352         ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
2353         if (ret < 0) {
2354             goto fail;
2355         }
2356 
2357         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
2358         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
2359         if (ret < 0) {
2360             goto fail;
2361         }
2362     }
2363 
2364     ret = 0;
2365 fail:
2366     g_free(out_buf);
2367     return ret;
2368 }
2369 
2370 static int make_completely_empty(BlockDriverState *bs)
2371 {
2372     BDRVQcowState *s = bs->opaque;
2373     int ret, l1_clusters;
2374     int64_t offset;
2375     uint64_t *new_reftable = NULL;
2376     uint64_t rt_entry, l1_size2;
2377     struct {
2378         uint64_t l1_offset;
2379         uint64_t reftable_offset;
2380         uint32_t reftable_clusters;
2381     } QEMU_PACKED l1_ofs_rt_ofs_cls;
2382 
2383     ret = qcow2_cache_empty(bs, s->l2_table_cache);
2384     if (ret < 0) {
2385         goto fail;
2386     }
2387 
2388     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
2389     if (ret < 0) {
2390         goto fail;
2391     }
2392 
2393     /* Refcounts will be broken utterly */
2394     ret = qcow2_mark_dirty(bs);
2395     if (ret < 0) {
2396         goto fail;
2397     }
2398 
2399     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2400 
2401     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2402     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
2403 
2404     /* After this call, neither the in-memory nor the on-disk refcount
2405      * information accurately describe the actual references */
2406 
2407     ret = bdrv_write_zeroes(bs->file, s->l1_table_offset / BDRV_SECTOR_SIZE,
2408                             l1_clusters * s->cluster_sectors, 0);
2409     if (ret < 0) {
2410         goto fail_broken_refcounts;
2411     }
2412     memset(s->l1_table, 0, l1_size2);
2413 
2414     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
2415 
2416     /* Overwrite enough clusters at the beginning of the sectors to place
2417      * the refcount table, a refcount block and the L1 table in; this may
2418      * overwrite parts of the existing refcount and L1 table, which is not
2419      * an issue because the dirty flag is set, complete data loss is in fact
2420      * desired and partial data loss is consequently fine as well */
2421     ret = bdrv_write_zeroes(bs->file, s->cluster_size / BDRV_SECTOR_SIZE,
2422                             (2 + l1_clusters) * s->cluster_size /
2423                             BDRV_SECTOR_SIZE, 0);
2424     /* This call (even if it failed overall) may have overwritten on-disk
2425      * refcount structures; in that case, the in-memory refcount information
2426      * will probably differ from the on-disk information which makes the BDS
2427      * unusable */
2428     if (ret < 0) {
2429         goto fail_broken_refcounts;
2430     }
2431 
2432     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2433     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
2434 
2435     /* "Create" an empty reftable (one cluster) directly after the image
2436      * header and an empty L1 table three clusters after the image header;
2437      * the cluster between those two will be used as the first refblock */
2438     cpu_to_be64w(&l1_ofs_rt_ofs_cls.l1_offset, 3 * s->cluster_size);
2439     cpu_to_be64w(&l1_ofs_rt_ofs_cls.reftable_offset, s->cluster_size);
2440     cpu_to_be32w(&l1_ofs_rt_ofs_cls.reftable_clusters, 1);
2441     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
2442                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
2443     if (ret < 0) {
2444         goto fail_broken_refcounts;
2445     }
2446 
2447     s->l1_table_offset = 3 * s->cluster_size;
2448 
2449     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
2450     if (!new_reftable) {
2451         ret = -ENOMEM;
2452         goto fail_broken_refcounts;
2453     }
2454 
2455     s->refcount_table_offset = s->cluster_size;
2456     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
2457 
2458     g_free(s->refcount_table);
2459     s->refcount_table = new_reftable;
2460     new_reftable = NULL;
2461 
2462     /* Now the in-memory refcount information again corresponds to the on-disk
2463      * information (reftable is empty and no refblocks (the refblock cache is
2464      * empty)); however, this means some clusters (e.g. the image header) are
2465      * referenced, but not refcounted, but the normal qcow2 code assumes that
2466      * the in-memory information is always correct */
2467 
2468     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
2469 
2470     /* Enter the first refblock into the reftable */
2471     rt_entry = cpu_to_be64(2 * s->cluster_size);
2472     ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
2473                            &rt_entry, sizeof(rt_entry));
2474     if (ret < 0) {
2475         goto fail_broken_refcounts;
2476     }
2477     s->refcount_table[0] = 2 * s->cluster_size;
2478 
2479     s->free_cluster_index = 0;
2480     assert(3 + l1_clusters <= s->refcount_block_size);
2481     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
2482     if (offset < 0) {
2483         ret = offset;
2484         goto fail_broken_refcounts;
2485     } else if (offset > 0) {
2486         error_report("First cluster in emptied image is in use");
2487         abort();
2488     }
2489 
2490     /* Now finally the in-memory information corresponds to the on-disk
2491      * structures and is correct */
2492     ret = qcow2_mark_clean(bs);
2493     if (ret < 0) {
2494         goto fail;
2495     }
2496 
2497     ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size);
2498     if (ret < 0) {
2499         goto fail;
2500     }
2501 
2502     return 0;
2503 
2504 fail_broken_refcounts:
2505     /* The BDS is unusable at this point. If we wanted to make it usable, we
2506      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
2507      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
2508      * again. However, because the functions which could have caused this error
2509      * path to be taken are used by those functions as well, it's very likely
2510      * that that sequence will fail as well. Therefore, just eject the BDS. */
2511     bs->drv = NULL;
2512 
2513 fail:
2514     g_free(new_reftable);
2515     return ret;
2516 }
2517 
2518 static int qcow2_make_empty(BlockDriverState *bs)
2519 {
2520     BDRVQcowState *s = bs->opaque;
2521     uint64_t start_sector;
2522     int sector_step = INT_MAX / BDRV_SECTOR_SIZE;
2523     int l1_clusters, ret = 0;
2524 
2525     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2526 
2527     if (s->qcow_version >= 3 && !s->snapshots &&
2528         3 + l1_clusters <= s->refcount_block_size) {
2529         /* The following function only works for qcow2 v3 images (it requires
2530          * the dirty flag) and only as long as there are no snapshots (because
2531          * it completely empties the image). Furthermore, the L1 table and three
2532          * additional clusters (image header, refcount table, one refcount
2533          * block) have to fit inside one refcount block. */
2534         return make_completely_empty(bs);
2535     }
2536 
2537     /* This fallback code simply discards every active cluster; this is slow,
2538      * but works in all cases */
2539     for (start_sector = 0; start_sector < bs->total_sectors;
2540          start_sector += sector_step)
2541     {
2542         /* As this function is generally used after committing an external
2543          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
2544          * default action for this kind of discard is to pass the discard,
2545          * which will ideally result in an actually smaller image file, as
2546          * is probably desired. */
2547         ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE,
2548                                      MIN(sector_step,
2549                                          bs->total_sectors - start_sector),
2550                                      QCOW2_DISCARD_SNAPSHOT, true);
2551         if (ret < 0) {
2552             break;
2553         }
2554     }
2555 
2556     return ret;
2557 }
2558 
2559 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
2560 {
2561     BDRVQcowState *s = bs->opaque;
2562     int ret;
2563 
2564     qemu_co_mutex_lock(&s->lock);
2565     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2566     if (ret < 0) {
2567         qemu_co_mutex_unlock(&s->lock);
2568         return ret;
2569     }
2570 
2571     if (qcow2_need_accurate_refcounts(s)) {
2572         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2573         if (ret < 0) {
2574             qemu_co_mutex_unlock(&s->lock);
2575             return ret;
2576         }
2577     }
2578     qemu_co_mutex_unlock(&s->lock);
2579 
2580     return 0;
2581 }
2582 
2583 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2584 {
2585     BDRVQcowState *s = bs->opaque;
2586     bdi->unallocated_blocks_are_zero = true;
2587     bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
2588     bdi->cluster_size = s->cluster_size;
2589     bdi->vm_state_offset = qcow2_vm_state_offset(s);
2590     return 0;
2591 }
2592 
2593 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
2594 {
2595     BDRVQcowState *s = bs->opaque;
2596     ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
2597 
2598     *spec_info = (ImageInfoSpecific){
2599         .kind  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
2600         {
2601             .qcow2 = g_new(ImageInfoSpecificQCow2, 1),
2602         },
2603     };
2604     if (s->qcow_version == 2) {
2605         *spec_info->qcow2 = (ImageInfoSpecificQCow2){
2606             .compat             = g_strdup("0.10"),
2607             .refcount_bits      = s->refcount_bits,
2608         };
2609     } else if (s->qcow_version == 3) {
2610         *spec_info->qcow2 = (ImageInfoSpecificQCow2){
2611             .compat             = g_strdup("1.1"),
2612             .lazy_refcounts     = s->compatible_features &
2613                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
2614             .has_lazy_refcounts = true,
2615             .corrupt            = s->incompatible_features &
2616                                   QCOW2_INCOMPAT_CORRUPT,
2617             .has_corrupt        = true,
2618             .refcount_bits      = s->refcount_bits,
2619         };
2620     }
2621 
2622     return spec_info;
2623 }
2624 
2625 #if 0
2626 static void dump_refcounts(BlockDriverState *bs)
2627 {
2628     BDRVQcowState *s = bs->opaque;
2629     int64_t nb_clusters, k, k1, size;
2630     int refcount;
2631 
2632     size = bdrv_getlength(bs->file);
2633     nb_clusters = size_to_clusters(s, size);
2634     for(k = 0; k < nb_clusters;) {
2635         k1 = k;
2636         refcount = get_refcount(bs, k);
2637         k++;
2638         while (k < nb_clusters && get_refcount(bs, k) == refcount)
2639             k++;
2640         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
2641                k - k1);
2642     }
2643 }
2644 #endif
2645 
2646 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2647                               int64_t pos)
2648 {
2649     BDRVQcowState *s = bs->opaque;
2650     int64_t total_sectors = bs->total_sectors;
2651     bool zero_beyond_eof = bs->zero_beyond_eof;
2652     int ret;
2653 
2654     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
2655     bs->zero_beyond_eof = false;
2656     ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov);
2657     bs->zero_beyond_eof = zero_beyond_eof;
2658 
2659     /* bdrv_co_do_writev will have increased the total_sectors value to include
2660      * the VM state - the VM state is however not an actual part of the block
2661      * device, therefore, we need to restore the old value. */
2662     bs->total_sectors = total_sectors;
2663 
2664     return ret;
2665 }
2666 
2667 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2668                               int64_t pos, int size)
2669 {
2670     BDRVQcowState *s = bs->opaque;
2671     bool zero_beyond_eof = bs->zero_beyond_eof;
2672     int ret;
2673 
2674     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
2675     bs->zero_beyond_eof = false;
2676     ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
2677     bs->zero_beyond_eof = zero_beyond_eof;
2678 
2679     return ret;
2680 }
2681 
2682 /*
2683  * Downgrades an image's version. To achieve this, any incompatible features
2684  * have to be removed.
2685  */
2686 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
2687                            BlockDriverAmendStatusCB *status_cb)
2688 {
2689     BDRVQcowState *s = bs->opaque;
2690     int current_version = s->qcow_version;
2691     int ret;
2692 
2693     if (target_version == current_version) {
2694         return 0;
2695     } else if (target_version > current_version) {
2696         return -EINVAL;
2697     } else if (target_version != 2) {
2698         return -EINVAL;
2699     }
2700 
2701     if (s->refcount_order != 4) {
2702         /* we would have to convert the image to a refcount_order == 4 image
2703          * here; however, since qemu (at the time of writing this) does not
2704          * support anything different than 4 anyway, there is no point in doing
2705          * so right now; however, we should error out (if qemu supports this in
2706          * the future and this code has not been adapted) */
2707         error_report("qcow2_downgrade: Image refcount orders other than 4 are "
2708                      "currently not supported.");
2709         return -ENOTSUP;
2710     }
2711 
2712     /* clear incompatible features */
2713     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
2714         ret = qcow2_mark_clean(bs);
2715         if (ret < 0) {
2716             return ret;
2717         }
2718     }
2719 
2720     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2721      * the first place; if that happens nonetheless, returning -ENOTSUP is the
2722      * best thing to do anyway */
2723 
2724     if (s->incompatible_features) {
2725         return -ENOTSUP;
2726     }
2727 
2728     /* since we can ignore compatible features, we can set them to 0 as well */
2729     s->compatible_features = 0;
2730     /* if lazy refcounts have been used, they have already been fixed through
2731      * clearing the dirty flag */
2732 
2733     /* clearing autoclear features is trivial */
2734     s->autoclear_features = 0;
2735 
2736     ret = qcow2_expand_zero_clusters(bs, status_cb);
2737     if (ret < 0) {
2738         return ret;
2739     }
2740 
2741     s->qcow_version = target_version;
2742     ret = qcow2_update_header(bs);
2743     if (ret < 0) {
2744         s->qcow_version = current_version;
2745         return ret;
2746     }
2747     return 0;
2748 }
2749 
2750 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
2751                                BlockDriverAmendStatusCB *status_cb)
2752 {
2753     BDRVQcowState *s = bs->opaque;
2754     int old_version = s->qcow_version, new_version = old_version;
2755     uint64_t new_size = 0;
2756     const char *backing_file = NULL, *backing_format = NULL;
2757     bool lazy_refcounts = s->use_lazy_refcounts;
2758     const char *compat = NULL;
2759     uint64_t cluster_size = s->cluster_size;
2760     bool encrypt;
2761     int ret;
2762     QemuOptDesc *desc = opts->list->desc;
2763 
2764     while (desc && desc->name) {
2765         if (!qemu_opt_find(opts, desc->name)) {
2766             /* only change explicitly defined options */
2767             desc++;
2768             continue;
2769         }
2770 
2771         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
2772             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
2773             if (!compat) {
2774                 /* preserve default */
2775             } else if (!strcmp(compat, "0.10")) {
2776                 new_version = 2;
2777             } else if (!strcmp(compat, "1.1")) {
2778                 new_version = 3;
2779             } else {
2780                 fprintf(stderr, "Unknown compatibility level %s.\n", compat);
2781                 return -EINVAL;
2782             }
2783         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
2784             fprintf(stderr, "Cannot change preallocation mode.\n");
2785             return -ENOTSUP;
2786         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
2787             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
2788         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
2789             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
2790         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
2791             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
2792         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
2793             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
2794                                         !!s->cipher);
2795 
2796             if (encrypt != !!s->cipher) {
2797                 fprintf(stderr, "Changing the encryption flag is not "
2798                         "supported.\n");
2799                 return -ENOTSUP;
2800             }
2801         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
2802             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
2803                                              cluster_size);
2804             if (cluster_size != s->cluster_size) {
2805                 fprintf(stderr, "Changing the cluster size is not "
2806                         "supported.\n");
2807                 return -ENOTSUP;
2808             }
2809         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
2810             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
2811                                                lazy_refcounts);
2812         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
2813             error_report("Cannot change refcount entry width");
2814             return -ENOTSUP;
2815         } else {
2816             /* if this assertion fails, this probably means a new option was
2817              * added without having it covered here */
2818             assert(false);
2819         }
2820 
2821         desc++;
2822     }
2823 
2824     if (new_version != old_version) {
2825         if (new_version > old_version) {
2826             /* Upgrade */
2827             s->qcow_version = new_version;
2828             ret = qcow2_update_header(bs);
2829             if (ret < 0) {
2830                 s->qcow_version = old_version;
2831                 return ret;
2832             }
2833         } else {
2834             ret = qcow2_downgrade(bs, new_version, status_cb);
2835             if (ret < 0) {
2836                 return ret;
2837             }
2838         }
2839     }
2840 
2841     if (backing_file || backing_format) {
2842         ret = qcow2_change_backing_file(bs,
2843                     backing_file ?: s->image_backing_file,
2844                     backing_format ?: s->image_backing_format);
2845         if (ret < 0) {
2846             return ret;
2847         }
2848     }
2849 
2850     if (s->use_lazy_refcounts != lazy_refcounts) {
2851         if (lazy_refcounts) {
2852             if (s->qcow_version < 3) {
2853                 fprintf(stderr, "Lazy refcounts only supported with compatibility "
2854                         "level 1.1 and above (use compat=1.1 or greater)\n");
2855                 return -EINVAL;
2856             }
2857             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
2858             ret = qcow2_update_header(bs);
2859             if (ret < 0) {
2860                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
2861                 return ret;
2862             }
2863             s->use_lazy_refcounts = true;
2864         } else {
2865             /* make image clean first */
2866             ret = qcow2_mark_clean(bs);
2867             if (ret < 0) {
2868                 return ret;
2869             }
2870             /* now disallow lazy refcounts */
2871             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
2872             ret = qcow2_update_header(bs);
2873             if (ret < 0) {
2874                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
2875                 return ret;
2876             }
2877             s->use_lazy_refcounts = false;
2878         }
2879     }
2880 
2881     if (new_size) {
2882         ret = bdrv_truncate(bs, new_size);
2883         if (ret < 0) {
2884             return ret;
2885         }
2886     }
2887 
2888     return 0;
2889 }
2890 
2891 /*
2892  * If offset or size are negative, respectively, they will not be included in
2893  * the BLOCK_IMAGE_CORRUPTED event emitted.
2894  * fatal will be ignored for read-only BDS; corruptions found there will always
2895  * be considered non-fatal.
2896  */
2897 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
2898                              int64_t size, const char *message_format, ...)
2899 {
2900     BDRVQcowState *s = bs->opaque;
2901     const char *node_name;
2902     char *message;
2903     va_list ap;
2904 
2905     fatal = fatal && !bs->read_only;
2906 
2907     if (s->signaled_corruption &&
2908         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
2909     {
2910         return;
2911     }
2912 
2913     va_start(ap, message_format);
2914     message = g_strdup_vprintf(message_format, ap);
2915     va_end(ap);
2916 
2917     if (fatal) {
2918         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
2919                 "corruption events will be suppressed\n", message);
2920     } else {
2921         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
2922                 "corruption events will be suppressed\n", message);
2923     }
2924 
2925     node_name = bdrv_get_node_name(bs);
2926     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
2927                                           *node_name != '\0', node_name,
2928                                           message, offset >= 0, offset,
2929                                           size >= 0, size,
2930                                           fatal, &error_abort);
2931     g_free(message);
2932 
2933     if (fatal) {
2934         qcow2_mark_corrupt(bs);
2935         bs->drv = NULL; /* make BDS unusable */
2936     }
2937 
2938     s->signaled_corruption = true;
2939 }
2940 
2941 static QemuOptsList qcow2_create_opts = {
2942     .name = "qcow2-create-opts",
2943     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
2944     .desc = {
2945         {
2946             .name = BLOCK_OPT_SIZE,
2947             .type = QEMU_OPT_SIZE,
2948             .help = "Virtual disk size"
2949         },
2950         {
2951             .name = BLOCK_OPT_COMPAT_LEVEL,
2952             .type = QEMU_OPT_STRING,
2953             .help = "Compatibility level (0.10 or 1.1)"
2954         },
2955         {
2956             .name = BLOCK_OPT_BACKING_FILE,
2957             .type = QEMU_OPT_STRING,
2958             .help = "File name of a base image"
2959         },
2960         {
2961             .name = BLOCK_OPT_BACKING_FMT,
2962             .type = QEMU_OPT_STRING,
2963             .help = "Image format of the base image"
2964         },
2965         {
2966             .name = BLOCK_OPT_ENCRYPT,
2967             .type = QEMU_OPT_BOOL,
2968             .help = "Encrypt the image",
2969             .def_value_str = "off"
2970         },
2971         {
2972             .name = BLOCK_OPT_CLUSTER_SIZE,
2973             .type = QEMU_OPT_SIZE,
2974             .help = "qcow2 cluster size",
2975             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
2976         },
2977         {
2978             .name = BLOCK_OPT_PREALLOC,
2979             .type = QEMU_OPT_STRING,
2980             .help = "Preallocation mode (allowed values: off, metadata, "
2981                     "falloc, full)"
2982         },
2983         {
2984             .name = BLOCK_OPT_LAZY_REFCOUNTS,
2985             .type = QEMU_OPT_BOOL,
2986             .help = "Postpone refcount updates",
2987             .def_value_str = "off"
2988         },
2989         {
2990             .name = BLOCK_OPT_REFCOUNT_BITS,
2991             .type = QEMU_OPT_NUMBER,
2992             .help = "Width of a reference count entry in bits",
2993             .def_value_str = "16"
2994         },
2995         { /* end of list */ }
2996     }
2997 };
2998 
2999 BlockDriver bdrv_qcow2 = {
3000     .format_name        = "qcow2",
3001     .instance_size      = sizeof(BDRVQcowState),
3002     .bdrv_probe         = qcow2_probe,
3003     .bdrv_open          = qcow2_open,
3004     .bdrv_close         = qcow2_close,
3005     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
3006     .bdrv_create        = qcow2_create,
3007     .bdrv_has_zero_init = bdrv_has_zero_init_1,
3008     .bdrv_co_get_block_status = qcow2_co_get_block_status,
3009     .bdrv_set_key       = qcow2_set_key,
3010 
3011     .bdrv_co_readv          = qcow2_co_readv,
3012     .bdrv_co_writev         = qcow2_co_writev,
3013     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
3014 
3015     .bdrv_co_write_zeroes   = qcow2_co_write_zeroes,
3016     .bdrv_co_discard        = qcow2_co_discard,
3017     .bdrv_truncate          = qcow2_truncate,
3018     .bdrv_write_compressed  = qcow2_write_compressed,
3019     .bdrv_make_empty        = qcow2_make_empty,
3020 
3021     .bdrv_snapshot_create   = qcow2_snapshot_create,
3022     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
3023     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
3024     .bdrv_snapshot_list     = qcow2_snapshot_list,
3025     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
3026     .bdrv_get_info          = qcow2_get_info,
3027     .bdrv_get_specific_info = qcow2_get_specific_info,
3028 
3029     .bdrv_save_vmstate    = qcow2_save_vmstate,
3030     .bdrv_load_vmstate    = qcow2_load_vmstate,
3031 
3032     .supports_backing           = true,
3033     .bdrv_change_backing_file   = qcow2_change_backing_file,
3034 
3035     .bdrv_refresh_limits        = qcow2_refresh_limits,
3036     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
3037 
3038     .create_opts         = &qcow2_create_opts,
3039     .bdrv_check          = qcow2_check,
3040     .bdrv_amend_options  = qcow2_amend_options,
3041 
3042     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
3043     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
3044 };
3045 
3046 static void bdrv_qcow2_init(void)
3047 {
3048     bdrv_register(&bdrv_qcow2);
3049 }
3050 
3051 block_init(bdrv_qcow2_init);
3052