xref: /openbmc/qemu/block/qcow2.c (revision 6f2945cd)
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         { /* end of list */ }
471     },
472 };
473 
474 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
475     [QCOW2_OL_MAIN_HEADER_BITNR]    = QCOW2_OPT_OVERLAP_MAIN_HEADER,
476     [QCOW2_OL_ACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L1,
477     [QCOW2_OL_ACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L2,
478     [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
479     [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
480     [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
481     [QCOW2_OL_INACTIVE_L1_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L1,
482     [QCOW2_OL_INACTIVE_L2_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L2,
483 };
484 
485 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
486                              uint64_t *l2_cache_size,
487                              uint64_t *refcount_cache_size, Error **errp)
488 {
489     BDRVQcowState *s = bs->opaque;
490     uint64_t combined_cache_size;
491     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
492 
493     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
494     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
495     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
496 
497     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
498     *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0);
499     *refcount_cache_size = qemu_opt_get_size(opts,
500                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
501 
502     if (combined_cache_size_set) {
503         if (l2_cache_size_set && refcount_cache_size_set) {
504             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
505                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
506                        "the same time");
507             return;
508         } else if (*l2_cache_size > combined_cache_size) {
509             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
510                        QCOW2_OPT_CACHE_SIZE);
511             return;
512         } else if (*refcount_cache_size > combined_cache_size) {
513             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
514                        QCOW2_OPT_CACHE_SIZE);
515             return;
516         }
517 
518         if (l2_cache_size_set) {
519             *refcount_cache_size = combined_cache_size - *l2_cache_size;
520         } else if (refcount_cache_size_set) {
521             *l2_cache_size = combined_cache_size - *refcount_cache_size;
522         } else {
523             *refcount_cache_size = combined_cache_size
524                                  / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1);
525             *l2_cache_size = combined_cache_size - *refcount_cache_size;
526         }
527     } else {
528         if (!l2_cache_size_set && !refcount_cache_size_set) {
529             *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE,
530                                  (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
531                                  * s->cluster_size);
532             *refcount_cache_size = *l2_cache_size
533                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
534         } else if (!l2_cache_size_set) {
535             *l2_cache_size = *refcount_cache_size
536                            * DEFAULT_L2_REFCOUNT_SIZE_RATIO;
537         } else if (!refcount_cache_size_set) {
538             *refcount_cache_size = *l2_cache_size
539                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
540         }
541     }
542 }
543 
544 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
545                       Error **errp)
546 {
547     BDRVQcowState *s = bs->opaque;
548     unsigned int len, i;
549     int ret = 0;
550     QCowHeader header;
551     QemuOpts *opts = NULL;
552     Error *local_err = NULL;
553     uint64_t ext_end;
554     uint64_t l1_vm_state_index;
555     const char *opt_overlap_check, *opt_overlap_check_template;
556     int overlap_check_template = 0;
557     uint64_t l2_cache_size, refcount_cache_size;
558 
559     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
560     if (ret < 0) {
561         error_setg_errno(errp, -ret, "Could not read qcow2 header");
562         goto fail;
563     }
564     be32_to_cpus(&header.magic);
565     be32_to_cpus(&header.version);
566     be64_to_cpus(&header.backing_file_offset);
567     be32_to_cpus(&header.backing_file_size);
568     be64_to_cpus(&header.size);
569     be32_to_cpus(&header.cluster_bits);
570     be32_to_cpus(&header.crypt_method);
571     be64_to_cpus(&header.l1_table_offset);
572     be32_to_cpus(&header.l1_size);
573     be64_to_cpus(&header.refcount_table_offset);
574     be32_to_cpus(&header.refcount_table_clusters);
575     be64_to_cpus(&header.snapshots_offset);
576     be32_to_cpus(&header.nb_snapshots);
577 
578     if (header.magic != QCOW_MAGIC) {
579         error_setg(errp, "Image is not in qcow2 format");
580         ret = -EINVAL;
581         goto fail;
582     }
583     if (header.version < 2 || header.version > 3) {
584         report_unsupported(bs, errp, "QCOW version %" PRIu32, header.version);
585         ret = -ENOTSUP;
586         goto fail;
587     }
588 
589     s->qcow_version = header.version;
590 
591     /* Initialise cluster size */
592     if (header.cluster_bits < MIN_CLUSTER_BITS ||
593         header.cluster_bits > MAX_CLUSTER_BITS) {
594         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
595                    header.cluster_bits);
596         ret = -EINVAL;
597         goto fail;
598     }
599 
600     s->cluster_bits = header.cluster_bits;
601     s->cluster_size = 1 << s->cluster_bits;
602     s->cluster_sectors = 1 << (s->cluster_bits - 9);
603 
604     /* Initialise version 3 header fields */
605     if (header.version == 2) {
606         header.incompatible_features    = 0;
607         header.compatible_features      = 0;
608         header.autoclear_features       = 0;
609         header.refcount_order           = 4;
610         header.header_length            = 72;
611     } else {
612         be64_to_cpus(&header.incompatible_features);
613         be64_to_cpus(&header.compatible_features);
614         be64_to_cpus(&header.autoclear_features);
615         be32_to_cpus(&header.refcount_order);
616         be32_to_cpus(&header.header_length);
617 
618         if (header.header_length < 104) {
619             error_setg(errp, "qcow2 header too short");
620             ret = -EINVAL;
621             goto fail;
622         }
623     }
624 
625     if (header.header_length > s->cluster_size) {
626         error_setg(errp, "qcow2 header exceeds cluster size");
627         ret = -EINVAL;
628         goto fail;
629     }
630 
631     if (header.header_length > sizeof(header)) {
632         s->unknown_header_fields_size = header.header_length - sizeof(header);
633         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
634         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
635                          s->unknown_header_fields_size);
636         if (ret < 0) {
637             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
638                              "fields");
639             goto fail;
640         }
641     }
642 
643     if (header.backing_file_offset > s->cluster_size) {
644         error_setg(errp, "Invalid backing file offset");
645         ret = -EINVAL;
646         goto fail;
647     }
648 
649     if (header.backing_file_offset) {
650         ext_end = header.backing_file_offset;
651     } else {
652         ext_end = 1 << header.cluster_bits;
653     }
654 
655     /* Handle feature bits */
656     s->incompatible_features    = header.incompatible_features;
657     s->compatible_features      = header.compatible_features;
658     s->autoclear_features       = header.autoclear_features;
659 
660     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
661         void *feature_table = NULL;
662         qcow2_read_extensions(bs, header.header_length, ext_end,
663                               &feature_table, NULL);
664         report_unsupported_feature(bs, errp, feature_table,
665                                    s->incompatible_features &
666                                    ~QCOW2_INCOMPAT_MASK);
667         ret = -ENOTSUP;
668         g_free(feature_table);
669         goto fail;
670     }
671 
672     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
673         /* Corrupt images may not be written to unless they are being repaired
674          */
675         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
676             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
677                        "read/write");
678             ret = -EACCES;
679             goto fail;
680         }
681     }
682 
683     /* Check support for various header values */
684     if (header.refcount_order > 6) {
685         error_setg(errp, "Reference count entry width too large; may not "
686                    "exceed 64 bits");
687         ret = -EINVAL;
688         goto fail;
689     }
690     s->refcount_order = header.refcount_order;
691     s->refcount_bits = 1 << s->refcount_order;
692     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
693     s->refcount_max += s->refcount_max - 1;
694 
695     if (header.crypt_method > QCOW_CRYPT_AES) {
696         error_setg(errp, "Unsupported encryption method: %" PRIu32,
697                    header.crypt_method);
698         ret = -EINVAL;
699         goto fail;
700     }
701     s->crypt_method_header = header.crypt_method;
702     if (s->crypt_method_header) {
703         bs->encrypted = 1;
704     }
705 
706     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
707     s->l2_size = 1 << s->l2_bits;
708     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
709     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
710     s->refcount_block_size = 1 << s->refcount_block_bits;
711     bs->total_sectors = header.size / 512;
712     s->csize_shift = (62 - (s->cluster_bits - 8));
713     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
714     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
715 
716     s->refcount_table_offset = header.refcount_table_offset;
717     s->refcount_table_size =
718         header.refcount_table_clusters << (s->cluster_bits - 3);
719 
720     if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
721         error_setg(errp, "Reference count table too large");
722         ret = -EINVAL;
723         goto fail;
724     }
725 
726     ret = validate_table_offset(bs, s->refcount_table_offset,
727                                 s->refcount_table_size, sizeof(uint64_t));
728     if (ret < 0) {
729         error_setg(errp, "Invalid reference count table offset");
730         goto fail;
731     }
732 
733     /* Snapshot table offset/length */
734     if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
735         error_setg(errp, "Too many snapshots");
736         ret = -EINVAL;
737         goto fail;
738     }
739 
740     ret = validate_table_offset(bs, header.snapshots_offset,
741                                 header.nb_snapshots,
742                                 sizeof(QCowSnapshotHeader));
743     if (ret < 0) {
744         error_setg(errp, "Invalid snapshot table offset");
745         goto fail;
746     }
747 
748     /* read the level 1 table */
749     if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
750         error_setg(errp, "Active L1 table too large");
751         ret = -EFBIG;
752         goto fail;
753     }
754     s->l1_size = header.l1_size;
755 
756     l1_vm_state_index = size_to_l1(s, header.size);
757     if (l1_vm_state_index > INT_MAX) {
758         error_setg(errp, "Image is too big");
759         ret = -EFBIG;
760         goto fail;
761     }
762     s->l1_vm_state_index = l1_vm_state_index;
763 
764     /* the L1 table must contain at least enough entries to put
765        header.size bytes */
766     if (s->l1_size < s->l1_vm_state_index) {
767         error_setg(errp, "L1 table is too small");
768         ret = -EINVAL;
769         goto fail;
770     }
771 
772     ret = validate_table_offset(bs, header.l1_table_offset,
773                                 header.l1_size, sizeof(uint64_t));
774     if (ret < 0) {
775         error_setg(errp, "Invalid L1 table offset");
776         goto fail;
777     }
778     s->l1_table_offset = header.l1_table_offset;
779 
780 
781     if (s->l1_size > 0) {
782         s->l1_table = qemu_try_blockalign(bs->file,
783             align_offset(s->l1_size * sizeof(uint64_t), 512));
784         if (s->l1_table == NULL) {
785             error_setg(errp, "Could not allocate L1 table");
786             ret = -ENOMEM;
787             goto fail;
788         }
789         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
790                          s->l1_size * sizeof(uint64_t));
791         if (ret < 0) {
792             error_setg_errno(errp, -ret, "Could not read L1 table");
793             goto fail;
794         }
795         for(i = 0;i < s->l1_size; i++) {
796             be64_to_cpus(&s->l1_table[i]);
797         }
798     }
799 
800     /* get L2 table/refcount block cache size from command line options */
801     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
802     qemu_opts_absorb_qdict(opts, options, &local_err);
803     if (local_err) {
804         error_propagate(errp, local_err);
805         ret = -EINVAL;
806         goto fail;
807     }
808 
809     read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size,
810                      &local_err);
811     if (local_err) {
812         error_propagate(errp, local_err);
813         ret = -EINVAL;
814         goto fail;
815     }
816 
817     l2_cache_size /= s->cluster_size;
818     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
819         l2_cache_size = MIN_L2_CACHE_SIZE;
820     }
821     if (l2_cache_size > INT_MAX) {
822         error_setg(errp, "L2 cache size too big");
823         ret = -EINVAL;
824         goto fail;
825     }
826 
827     refcount_cache_size /= s->cluster_size;
828     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
829         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
830     }
831     if (refcount_cache_size > INT_MAX) {
832         error_setg(errp, "Refcount cache size too big");
833         ret = -EINVAL;
834         goto fail;
835     }
836 
837     /* alloc L2 table/refcount block cache */
838     s->l2_table_cache = qcow2_cache_create(bs, l2_cache_size);
839     s->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size);
840     if (s->l2_table_cache == NULL || s->refcount_block_cache == NULL) {
841         error_setg(errp, "Could not allocate metadata caches");
842         ret = -ENOMEM;
843         goto fail;
844     }
845 
846     s->cluster_cache = g_malloc(s->cluster_size);
847     /* one more sector for decompressed data alignment */
848     s->cluster_data = qemu_try_blockalign(bs->file, QCOW_MAX_CRYPT_CLUSTERS
849                                                     * s->cluster_size + 512);
850     if (s->cluster_data == NULL) {
851         error_setg(errp, "Could not allocate temporary cluster buffer");
852         ret = -ENOMEM;
853         goto fail;
854     }
855 
856     s->cluster_cache_offset = -1;
857     s->flags = flags;
858 
859     ret = qcow2_refcount_init(bs);
860     if (ret != 0) {
861         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
862         goto fail;
863     }
864 
865     QLIST_INIT(&s->cluster_allocs);
866     QTAILQ_INIT(&s->discards);
867 
868     /* read qcow2 extensions */
869     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
870         &local_err)) {
871         error_propagate(errp, local_err);
872         ret = -EINVAL;
873         goto fail;
874     }
875 
876     /* read the backing file name */
877     if (header.backing_file_offset != 0) {
878         len = header.backing_file_size;
879         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
880             len >= sizeof(bs->backing_file)) {
881             error_setg(errp, "Backing file name too long");
882             ret = -EINVAL;
883             goto fail;
884         }
885         ret = bdrv_pread(bs->file, header.backing_file_offset,
886                          bs->backing_file, len);
887         if (ret < 0) {
888             error_setg_errno(errp, -ret, "Could not read backing file name");
889             goto fail;
890         }
891         bs->backing_file[len] = '\0';
892         s->image_backing_file = g_strdup(bs->backing_file);
893     }
894 
895     /* Internal snapshots */
896     s->snapshots_offset = header.snapshots_offset;
897     s->nb_snapshots = header.nb_snapshots;
898 
899     ret = qcow2_read_snapshots(bs);
900     if (ret < 0) {
901         error_setg_errno(errp, -ret, "Could not read snapshots");
902         goto fail;
903     }
904 
905     /* Clear unknown autoclear feature bits */
906     if (!bs->read_only && !(flags & BDRV_O_INCOMING) && s->autoclear_features) {
907         s->autoclear_features = 0;
908         ret = qcow2_update_header(bs);
909         if (ret < 0) {
910             error_setg_errno(errp, -ret, "Could not update qcow2 header");
911             goto fail;
912         }
913     }
914 
915     /* Initialise locks */
916     qemu_co_mutex_init(&s->lock);
917 
918     /* Repair image if dirty */
919     if (!(flags & (BDRV_O_CHECK | BDRV_O_INCOMING)) && !bs->read_only &&
920         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
921         BdrvCheckResult result = {0};
922 
923         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
924         if (ret < 0) {
925             error_setg_errno(errp, -ret, "Could not repair dirty image");
926             goto fail;
927         }
928     }
929 
930     /* Enable lazy_refcounts according to image and command line options */
931     s->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
932         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
933 
934     s->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
935     s->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
936     s->discard_passthrough[QCOW2_DISCARD_REQUEST] =
937         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
938                           flags & BDRV_O_UNMAP);
939     s->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
940         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
941     s->discard_passthrough[QCOW2_DISCARD_OTHER] =
942         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
943 
944     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
945     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
946     if (opt_overlap_check_template && opt_overlap_check &&
947         strcmp(opt_overlap_check_template, opt_overlap_check))
948     {
949         error_setg(errp, "Conflicting values for qcow2 options '"
950                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
951                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
952         ret = -EINVAL;
953         goto fail;
954     }
955     if (!opt_overlap_check) {
956         opt_overlap_check = opt_overlap_check_template ?: "cached";
957     }
958 
959     if (!strcmp(opt_overlap_check, "none")) {
960         overlap_check_template = 0;
961     } else if (!strcmp(opt_overlap_check, "constant")) {
962         overlap_check_template = QCOW2_OL_CONSTANT;
963     } else if (!strcmp(opt_overlap_check, "cached")) {
964         overlap_check_template = QCOW2_OL_CACHED;
965     } else if (!strcmp(opt_overlap_check, "all")) {
966         overlap_check_template = QCOW2_OL_ALL;
967     } else {
968         error_setg(errp, "Unsupported value '%s' for qcow2 option "
969                    "'overlap-check'. Allowed are either of the following: "
970                    "none, constant, cached, all", opt_overlap_check);
971         ret = -EINVAL;
972         goto fail;
973     }
974 
975     s->overlap_check = 0;
976     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
977         /* overlap-check defines a template bitmask, but every flag may be
978          * overwritten through the associated boolean option */
979         s->overlap_check |=
980             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
981                               overlap_check_template & (1 << i)) << i;
982     }
983 
984     qemu_opts_del(opts);
985     opts = NULL;
986 
987     if (s->use_lazy_refcounts && s->qcow_version < 3) {
988         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
989                    "qemu 1.1 compatibility level");
990         ret = -EINVAL;
991         goto fail;
992     }
993 
994 #ifdef DEBUG_ALLOC
995     {
996         BdrvCheckResult result = {0};
997         qcow2_check_refcounts(bs, &result, 0);
998     }
999 #endif
1000     return ret;
1001 
1002  fail:
1003     qemu_opts_del(opts);
1004     g_free(s->unknown_header_fields);
1005     cleanup_unknown_header_ext(bs);
1006     qcow2_free_snapshots(bs);
1007     qcow2_refcount_close(bs);
1008     qemu_vfree(s->l1_table);
1009     /* else pre-write overlap checks in cache_destroy may crash */
1010     s->l1_table = NULL;
1011     if (s->l2_table_cache) {
1012         qcow2_cache_destroy(bs, s->l2_table_cache);
1013     }
1014     if (s->refcount_block_cache) {
1015         qcow2_cache_destroy(bs, s->refcount_block_cache);
1016     }
1017     g_free(s->cluster_cache);
1018     qemu_vfree(s->cluster_data);
1019     return ret;
1020 }
1021 
1022 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1023 {
1024     BDRVQcowState *s = bs->opaque;
1025 
1026     bs->bl.write_zeroes_alignment = s->cluster_sectors;
1027 }
1028 
1029 static int qcow2_set_key(BlockDriverState *bs, const char *key)
1030 {
1031     BDRVQcowState *s = bs->opaque;
1032     uint8_t keybuf[16];
1033     int len, i;
1034 
1035     memset(keybuf, 0, 16);
1036     len = strlen(key);
1037     if (len > 16)
1038         len = 16;
1039     /* XXX: we could compress the chars to 7 bits to increase
1040        entropy */
1041     for(i = 0;i < len;i++) {
1042         keybuf[i] = key[i];
1043     }
1044     assert(bs->encrypted);
1045     s->crypt_method = s->crypt_method_header;
1046 
1047     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
1048         return -1;
1049     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
1050         return -1;
1051 #if 0
1052     /* test */
1053     {
1054         uint8_t in[16];
1055         uint8_t out[16];
1056         uint8_t tmp[16];
1057         for(i=0;i<16;i++)
1058             in[i] = i;
1059         AES_encrypt(in, tmp, &s->aes_encrypt_key);
1060         AES_decrypt(tmp, out, &s->aes_decrypt_key);
1061         for(i = 0; i < 16; i++)
1062             printf(" %02x", tmp[i]);
1063         printf("\n");
1064         for(i = 0; i < 16; i++)
1065             printf(" %02x", out[i]);
1066         printf("\n");
1067     }
1068 #endif
1069     return 0;
1070 }
1071 
1072 /* We have no actual commit/abort logic for qcow2, but we need to write out any
1073  * unwritten data if we reopen read-only. */
1074 static int qcow2_reopen_prepare(BDRVReopenState *state,
1075                                 BlockReopenQueue *queue, Error **errp)
1076 {
1077     int ret;
1078 
1079     if ((state->flags & BDRV_O_RDWR) == 0) {
1080         ret = bdrv_flush(state->bs);
1081         if (ret < 0) {
1082             return ret;
1083         }
1084 
1085         ret = qcow2_mark_clean(state->bs);
1086         if (ret < 0) {
1087             return ret;
1088         }
1089     }
1090 
1091     return 0;
1092 }
1093 
1094 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1095         int64_t sector_num, int nb_sectors, int *pnum)
1096 {
1097     BDRVQcowState *s = bs->opaque;
1098     uint64_t cluster_offset;
1099     int index_in_cluster, ret;
1100     int64_t status = 0;
1101 
1102     *pnum = nb_sectors;
1103     qemu_co_mutex_lock(&s->lock);
1104     ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
1105     qemu_co_mutex_unlock(&s->lock);
1106     if (ret < 0) {
1107         return ret;
1108     }
1109 
1110     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1111         !s->crypt_method) {
1112         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1113         cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1114         status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1115     }
1116     if (ret == QCOW2_CLUSTER_ZERO) {
1117         status |= BDRV_BLOCK_ZERO;
1118     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1119         status |= BDRV_BLOCK_DATA;
1120     }
1121     return status;
1122 }
1123 
1124 /* handle reading after the end of the backing file */
1125 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
1126                   int64_t sector_num, int nb_sectors)
1127 {
1128     int n1;
1129     if ((sector_num + nb_sectors) <= bs->total_sectors)
1130         return nb_sectors;
1131     if (sector_num >= bs->total_sectors)
1132         n1 = 0;
1133     else
1134         n1 = bs->total_sectors - sector_num;
1135 
1136     qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1));
1137 
1138     return n1;
1139 }
1140 
1141 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
1142                           int remaining_sectors, QEMUIOVector *qiov)
1143 {
1144     BDRVQcowState *s = bs->opaque;
1145     int index_in_cluster, n1;
1146     int ret;
1147     int cur_nr_sectors; /* number of sectors in current iteration */
1148     uint64_t cluster_offset = 0;
1149     uint64_t bytes_done = 0;
1150     QEMUIOVector hd_qiov;
1151     uint8_t *cluster_data = NULL;
1152 
1153     qemu_iovec_init(&hd_qiov, qiov->niov);
1154 
1155     qemu_co_mutex_lock(&s->lock);
1156 
1157     while (remaining_sectors != 0) {
1158 
1159         /* prepare next request */
1160         cur_nr_sectors = remaining_sectors;
1161         if (s->crypt_method) {
1162             cur_nr_sectors = MIN(cur_nr_sectors,
1163                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1164         }
1165 
1166         ret = qcow2_get_cluster_offset(bs, sector_num << 9,
1167             &cur_nr_sectors, &cluster_offset);
1168         if (ret < 0) {
1169             goto fail;
1170         }
1171 
1172         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1173 
1174         qemu_iovec_reset(&hd_qiov);
1175         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1176             cur_nr_sectors * 512);
1177 
1178         switch (ret) {
1179         case QCOW2_CLUSTER_UNALLOCATED:
1180 
1181             if (bs->backing_hd) {
1182                 /* read from the base image */
1183                 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov,
1184                     sector_num, cur_nr_sectors);
1185                 if (n1 > 0) {
1186                     QEMUIOVector local_qiov;
1187 
1188                     qemu_iovec_init(&local_qiov, hd_qiov.niov);
1189                     qemu_iovec_concat(&local_qiov, &hd_qiov, 0,
1190                                       n1 * BDRV_SECTOR_SIZE);
1191 
1192                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1193                     qemu_co_mutex_unlock(&s->lock);
1194                     ret = bdrv_co_readv(bs->backing_hd, sector_num,
1195                                         n1, &local_qiov);
1196                     qemu_co_mutex_lock(&s->lock);
1197 
1198                     qemu_iovec_destroy(&local_qiov);
1199 
1200                     if (ret < 0) {
1201                         goto fail;
1202                     }
1203                 }
1204             } else {
1205                 /* Note: in this case, no need to wait */
1206                 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1207             }
1208             break;
1209 
1210         case QCOW2_CLUSTER_ZERO:
1211             qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1212             break;
1213 
1214         case QCOW2_CLUSTER_COMPRESSED:
1215             /* add AIO support for compressed blocks ? */
1216             ret = qcow2_decompress_cluster(bs, cluster_offset);
1217             if (ret < 0) {
1218                 goto fail;
1219             }
1220 
1221             qemu_iovec_from_buf(&hd_qiov, 0,
1222                 s->cluster_cache + index_in_cluster * 512,
1223                 512 * cur_nr_sectors);
1224             break;
1225 
1226         case QCOW2_CLUSTER_NORMAL:
1227             if ((cluster_offset & 511) != 0) {
1228                 ret = -EIO;
1229                 goto fail;
1230             }
1231 
1232             if (bs->encrypted) {
1233                 assert(s->crypt_method);
1234 
1235                 /*
1236                  * For encrypted images, read everything into a temporary
1237                  * contiguous buffer on which the AES functions can work.
1238                  */
1239                 if (!cluster_data) {
1240                     cluster_data =
1241                         qemu_try_blockalign(bs->file, QCOW_MAX_CRYPT_CLUSTERS
1242                                                       * s->cluster_size);
1243                     if (cluster_data == NULL) {
1244                         ret = -ENOMEM;
1245                         goto fail;
1246                     }
1247                 }
1248 
1249                 assert(cur_nr_sectors <=
1250                     QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1251                 qemu_iovec_reset(&hd_qiov);
1252                 qemu_iovec_add(&hd_qiov, cluster_data,
1253                     512 * cur_nr_sectors);
1254             }
1255 
1256             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1257             qemu_co_mutex_unlock(&s->lock);
1258             ret = bdrv_co_readv(bs->file,
1259                                 (cluster_offset >> 9) + index_in_cluster,
1260                                 cur_nr_sectors, &hd_qiov);
1261             qemu_co_mutex_lock(&s->lock);
1262             if (ret < 0) {
1263                 goto fail;
1264             }
1265             if (bs->encrypted) {
1266                 assert(s->crypt_method);
1267                 qcow2_encrypt_sectors(s, sector_num,  cluster_data,
1268                     cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
1269                 qemu_iovec_from_buf(qiov, bytes_done,
1270                     cluster_data, 512 * cur_nr_sectors);
1271             }
1272             break;
1273 
1274         default:
1275             g_assert_not_reached();
1276             ret = -EIO;
1277             goto fail;
1278         }
1279 
1280         remaining_sectors -= cur_nr_sectors;
1281         sector_num += cur_nr_sectors;
1282         bytes_done += cur_nr_sectors * 512;
1283     }
1284     ret = 0;
1285 
1286 fail:
1287     qemu_co_mutex_unlock(&s->lock);
1288 
1289     qemu_iovec_destroy(&hd_qiov);
1290     qemu_vfree(cluster_data);
1291 
1292     return ret;
1293 }
1294 
1295 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
1296                            int64_t sector_num,
1297                            int remaining_sectors,
1298                            QEMUIOVector *qiov)
1299 {
1300     BDRVQcowState *s = bs->opaque;
1301     int index_in_cluster;
1302     int ret;
1303     int cur_nr_sectors; /* number of sectors in current iteration */
1304     uint64_t cluster_offset;
1305     QEMUIOVector hd_qiov;
1306     uint64_t bytes_done = 0;
1307     uint8_t *cluster_data = NULL;
1308     QCowL2Meta *l2meta = NULL;
1309 
1310     trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
1311                                  remaining_sectors);
1312 
1313     qemu_iovec_init(&hd_qiov, qiov->niov);
1314 
1315     s->cluster_cache_offset = -1; /* disable compressed cache */
1316 
1317     qemu_co_mutex_lock(&s->lock);
1318 
1319     while (remaining_sectors != 0) {
1320 
1321         l2meta = NULL;
1322 
1323         trace_qcow2_writev_start_part(qemu_coroutine_self());
1324         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1325         cur_nr_sectors = remaining_sectors;
1326         if (bs->encrypted &&
1327             cur_nr_sectors >
1328             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster) {
1329             cur_nr_sectors =
1330                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster;
1331         }
1332 
1333         ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
1334             &cur_nr_sectors, &cluster_offset, &l2meta);
1335         if (ret < 0) {
1336             goto fail;
1337         }
1338 
1339         assert((cluster_offset & 511) == 0);
1340 
1341         qemu_iovec_reset(&hd_qiov);
1342         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1343             cur_nr_sectors * 512);
1344 
1345         if (bs->encrypted) {
1346             assert(s->crypt_method);
1347             if (!cluster_data) {
1348                 cluster_data = qemu_try_blockalign(bs->file,
1349                                                    QCOW_MAX_CRYPT_CLUSTERS
1350                                                    * s->cluster_size);
1351                 if (cluster_data == NULL) {
1352                     ret = -ENOMEM;
1353                     goto fail;
1354                 }
1355             }
1356 
1357             assert(hd_qiov.size <=
1358                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1359             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1360 
1361             qcow2_encrypt_sectors(s, sector_num, cluster_data,
1362                 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
1363 
1364             qemu_iovec_reset(&hd_qiov);
1365             qemu_iovec_add(&hd_qiov, cluster_data,
1366                 cur_nr_sectors * 512);
1367         }
1368 
1369         ret = qcow2_pre_write_overlap_check(bs, 0,
1370                 cluster_offset + index_in_cluster * BDRV_SECTOR_SIZE,
1371                 cur_nr_sectors * BDRV_SECTOR_SIZE);
1372         if (ret < 0) {
1373             goto fail;
1374         }
1375 
1376         qemu_co_mutex_unlock(&s->lock);
1377         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1378         trace_qcow2_writev_data(qemu_coroutine_self(),
1379                                 (cluster_offset >> 9) + index_in_cluster);
1380         ret = bdrv_co_writev(bs->file,
1381                              (cluster_offset >> 9) + index_in_cluster,
1382                              cur_nr_sectors, &hd_qiov);
1383         qemu_co_mutex_lock(&s->lock);
1384         if (ret < 0) {
1385             goto fail;
1386         }
1387 
1388         while (l2meta != NULL) {
1389             QCowL2Meta *next;
1390 
1391             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1392             if (ret < 0) {
1393                 goto fail;
1394             }
1395 
1396             /* Take the request off the list of running requests */
1397             if (l2meta->nb_clusters != 0) {
1398                 QLIST_REMOVE(l2meta, next_in_flight);
1399             }
1400 
1401             qemu_co_queue_restart_all(&l2meta->dependent_requests);
1402 
1403             next = l2meta->next;
1404             g_free(l2meta);
1405             l2meta = next;
1406         }
1407 
1408         remaining_sectors -= cur_nr_sectors;
1409         sector_num += cur_nr_sectors;
1410         bytes_done += cur_nr_sectors * 512;
1411         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
1412     }
1413     ret = 0;
1414 
1415 fail:
1416     qemu_co_mutex_unlock(&s->lock);
1417 
1418     while (l2meta != NULL) {
1419         QCowL2Meta *next;
1420 
1421         if (l2meta->nb_clusters != 0) {
1422             QLIST_REMOVE(l2meta, next_in_flight);
1423         }
1424         qemu_co_queue_restart_all(&l2meta->dependent_requests);
1425 
1426         next = l2meta->next;
1427         g_free(l2meta);
1428         l2meta = next;
1429     }
1430 
1431     qemu_iovec_destroy(&hd_qiov);
1432     qemu_vfree(cluster_data);
1433     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
1434 
1435     return ret;
1436 }
1437 
1438 static void qcow2_close(BlockDriverState *bs)
1439 {
1440     BDRVQcowState *s = bs->opaque;
1441     qemu_vfree(s->l1_table);
1442     /* else pre-write overlap checks in cache_destroy may crash */
1443     s->l1_table = NULL;
1444 
1445     if (!(bs->open_flags & BDRV_O_INCOMING)) {
1446         int ret1, ret2;
1447 
1448         ret1 = qcow2_cache_flush(bs, s->l2_table_cache);
1449         ret2 = qcow2_cache_flush(bs, s->refcount_block_cache);
1450 
1451         if (ret1) {
1452             error_report("Failed to flush the L2 table cache: %s",
1453                          strerror(-ret1));
1454         }
1455         if (ret2) {
1456             error_report("Failed to flush the refcount block cache: %s",
1457                          strerror(-ret2));
1458         }
1459 
1460         if (!ret1 && !ret2) {
1461             qcow2_mark_clean(bs);
1462         }
1463     }
1464 
1465     qcow2_cache_destroy(bs, s->l2_table_cache);
1466     qcow2_cache_destroy(bs, s->refcount_block_cache);
1467 
1468     g_free(s->unknown_header_fields);
1469     cleanup_unknown_header_ext(bs);
1470 
1471     g_free(s->image_backing_file);
1472     g_free(s->image_backing_format);
1473 
1474     g_free(s->cluster_cache);
1475     qemu_vfree(s->cluster_data);
1476     qcow2_refcount_close(bs);
1477     qcow2_free_snapshots(bs);
1478 }
1479 
1480 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
1481 {
1482     BDRVQcowState *s = bs->opaque;
1483     int flags = s->flags;
1484     AES_KEY aes_encrypt_key;
1485     AES_KEY aes_decrypt_key;
1486     uint32_t crypt_method = 0;
1487     QDict *options;
1488     Error *local_err = NULL;
1489     int ret;
1490 
1491     /*
1492      * Backing files are read-only which makes all of their metadata immutable,
1493      * that means we don't have to worry about reopening them here.
1494      */
1495 
1496     if (bs->encrypted) {
1497         assert(s->crypt_method);
1498         crypt_method = s->crypt_method;
1499         memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key));
1500         memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key));
1501     }
1502 
1503     qcow2_close(bs);
1504 
1505     bdrv_invalidate_cache(bs->file, &local_err);
1506     if (local_err) {
1507         error_propagate(errp, local_err);
1508         return;
1509     }
1510 
1511     memset(s, 0, sizeof(BDRVQcowState));
1512     options = qdict_clone_shallow(bs->options);
1513 
1514     ret = qcow2_open(bs, options, flags, &local_err);
1515     QDECREF(options);
1516     if (local_err) {
1517         error_setg(errp, "Could not reopen qcow2 layer: %s",
1518                    error_get_pretty(local_err));
1519         error_free(local_err);
1520         return;
1521     } else if (ret < 0) {
1522         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
1523         return;
1524     }
1525 
1526     if (bs->encrypted) {
1527         s->crypt_method = crypt_method;
1528         memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key));
1529         memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key));
1530     }
1531 }
1532 
1533 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
1534     size_t len, size_t buflen)
1535 {
1536     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
1537     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
1538 
1539     if (buflen < ext_len) {
1540         return -ENOSPC;
1541     }
1542 
1543     *ext_backing_fmt = (QCowExtension) {
1544         .magic  = cpu_to_be32(magic),
1545         .len    = cpu_to_be32(len),
1546     };
1547     memcpy(buf + sizeof(QCowExtension), s, len);
1548 
1549     return ext_len;
1550 }
1551 
1552 /*
1553  * Updates the qcow2 header, including the variable length parts of it, i.e.
1554  * the backing file name and all extensions. qcow2 was not designed to allow
1555  * such changes, so if we run out of space (we can only use the first cluster)
1556  * this function may fail.
1557  *
1558  * Returns 0 on success, -errno in error cases.
1559  */
1560 int qcow2_update_header(BlockDriverState *bs)
1561 {
1562     BDRVQcowState *s = bs->opaque;
1563     QCowHeader *header;
1564     char *buf;
1565     size_t buflen = s->cluster_size;
1566     int ret;
1567     uint64_t total_size;
1568     uint32_t refcount_table_clusters;
1569     size_t header_length;
1570     Qcow2UnknownHeaderExtension *uext;
1571 
1572     buf = qemu_blockalign(bs, buflen);
1573 
1574     /* Header structure */
1575     header = (QCowHeader*) buf;
1576 
1577     if (buflen < sizeof(*header)) {
1578         ret = -ENOSPC;
1579         goto fail;
1580     }
1581 
1582     header_length = sizeof(*header) + s->unknown_header_fields_size;
1583     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1584     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1585 
1586     *header = (QCowHeader) {
1587         /* Version 2 fields */
1588         .magic                  = cpu_to_be32(QCOW_MAGIC),
1589         .version                = cpu_to_be32(s->qcow_version),
1590         .backing_file_offset    = 0,
1591         .backing_file_size      = 0,
1592         .cluster_bits           = cpu_to_be32(s->cluster_bits),
1593         .size                   = cpu_to_be64(total_size),
1594         .crypt_method           = cpu_to_be32(s->crypt_method_header),
1595         .l1_size                = cpu_to_be32(s->l1_size),
1596         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
1597         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
1598         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1599         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
1600         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
1601 
1602         /* Version 3 fields */
1603         .incompatible_features  = cpu_to_be64(s->incompatible_features),
1604         .compatible_features    = cpu_to_be64(s->compatible_features),
1605         .autoclear_features     = cpu_to_be64(s->autoclear_features),
1606         .refcount_order         = cpu_to_be32(s->refcount_order),
1607         .header_length          = cpu_to_be32(header_length),
1608     };
1609 
1610     /* For older versions, write a shorter header */
1611     switch (s->qcow_version) {
1612     case 2:
1613         ret = offsetof(QCowHeader, incompatible_features);
1614         break;
1615     case 3:
1616         ret = sizeof(*header);
1617         break;
1618     default:
1619         ret = -EINVAL;
1620         goto fail;
1621     }
1622 
1623     buf += ret;
1624     buflen -= ret;
1625     memset(buf, 0, buflen);
1626 
1627     /* Preserve any unknown field in the header */
1628     if (s->unknown_header_fields_size) {
1629         if (buflen < s->unknown_header_fields_size) {
1630             ret = -ENOSPC;
1631             goto fail;
1632         }
1633 
1634         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1635         buf += s->unknown_header_fields_size;
1636         buflen -= s->unknown_header_fields_size;
1637     }
1638 
1639     /* Backing file format header extension */
1640     if (s->image_backing_format) {
1641         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1642                              s->image_backing_format,
1643                              strlen(s->image_backing_format),
1644                              buflen);
1645         if (ret < 0) {
1646             goto fail;
1647         }
1648 
1649         buf += ret;
1650         buflen -= ret;
1651     }
1652 
1653     /* Feature table */
1654     Qcow2Feature features[] = {
1655         {
1656             .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1657             .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
1658             .name = "dirty bit",
1659         },
1660         {
1661             .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1662             .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
1663             .name = "corrupt bit",
1664         },
1665         {
1666             .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1667             .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1668             .name = "lazy refcounts",
1669         },
1670     };
1671 
1672     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1673                          features, sizeof(features), buflen);
1674     if (ret < 0) {
1675         goto fail;
1676     }
1677     buf += ret;
1678     buflen -= ret;
1679 
1680     /* Keep unknown header extensions */
1681     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1682         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1683         if (ret < 0) {
1684             goto fail;
1685         }
1686 
1687         buf += ret;
1688         buflen -= ret;
1689     }
1690 
1691     /* End of header extensions */
1692     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1693     if (ret < 0) {
1694         goto fail;
1695     }
1696 
1697     buf += ret;
1698     buflen -= ret;
1699 
1700     /* Backing file name */
1701     if (s->image_backing_file) {
1702         size_t backing_file_len = strlen(s->image_backing_file);
1703 
1704         if (buflen < backing_file_len) {
1705             ret = -ENOSPC;
1706             goto fail;
1707         }
1708 
1709         /* Using strncpy is ok here, since buf is not NUL-terminated. */
1710         strncpy(buf, s->image_backing_file, buflen);
1711 
1712         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1713         header->backing_file_size   = cpu_to_be32(backing_file_len);
1714     }
1715 
1716     /* Write the new header */
1717     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
1718     if (ret < 0) {
1719         goto fail;
1720     }
1721 
1722     ret = 0;
1723 fail:
1724     qemu_vfree(header);
1725     return ret;
1726 }
1727 
1728 static int qcow2_change_backing_file(BlockDriverState *bs,
1729     const char *backing_file, const char *backing_fmt)
1730 {
1731     BDRVQcowState *s = bs->opaque;
1732 
1733     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1734     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1735 
1736     g_free(s->image_backing_file);
1737     g_free(s->image_backing_format);
1738 
1739     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
1740     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
1741 
1742     return qcow2_update_header(bs);
1743 }
1744 
1745 static int preallocate(BlockDriverState *bs)
1746 {
1747     uint64_t nb_sectors;
1748     uint64_t offset;
1749     uint64_t host_offset = 0;
1750     int num;
1751     int ret;
1752     QCowL2Meta *meta;
1753 
1754     nb_sectors = bdrv_nb_sectors(bs);
1755     offset = 0;
1756 
1757     while (nb_sectors) {
1758         num = MIN(nb_sectors, INT_MAX >> BDRV_SECTOR_BITS);
1759         ret = qcow2_alloc_cluster_offset(bs, offset, &num,
1760                                          &host_offset, &meta);
1761         if (ret < 0) {
1762             return ret;
1763         }
1764 
1765         while (meta) {
1766             QCowL2Meta *next = meta->next;
1767 
1768             ret = qcow2_alloc_cluster_link_l2(bs, meta);
1769             if (ret < 0) {
1770                 qcow2_free_any_clusters(bs, meta->alloc_offset,
1771                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
1772                 return ret;
1773             }
1774 
1775             /* There are no dependent requests, but we need to remove our
1776              * request from the list of in-flight requests */
1777             QLIST_REMOVE(meta, next_in_flight);
1778 
1779             g_free(meta);
1780             meta = next;
1781         }
1782 
1783         /* TODO Preallocate data if requested */
1784 
1785         nb_sectors -= num;
1786         offset += num << BDRV_SECTOR_BITS;
1787     }
1788 
1789     /*
1790      * It is expected that the image file is large enough to actually contain
1791      * all of the allocated clusters (otherwise we get failing reads after
1792      * EOF). Extend the image to the last allocated sector.
1793      */
1794     if (host_offset != 0) {
1795         uint8_t buf[BDRV_SECTOR_SIZE];
1796         memset(buf, 0, BDRV_SECTOR_SIZE);
1797         ret = bdrv_write(bs->file, (host_offset >> BDRV_SECTOR_BITS) + num - 1,
1798                          buf, 1);
1799         if (ret < 0) {
1800             return ret;
1801         }
1802     }
1803 
1804     return 0;
1805 }
1806 
1807 static int qcow2_create2(const char *filename, int64_t total_size,
1808                          const char *backing_file, const char *backing_format,
1809                          int flags, size_t cluster_size, PreallocMode prealloc,
1810                          QemuOpts *opts, int version, int refcount_order,
1811                          Error **errp)
1812 {
1813     /* Calculate cluster_bits */
1814     int cluster_bits;
1815     cluster_bits = ctz32(cluster_size);
1816     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
1817         (1 << cluster_bits) != cluster_size)
1818     {
1819         error_setg(errp, "Cluster size must be a power of two between %d and "
1820                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
1821         return -EINVAL;
1822     }
1823 
1824     /*
1825      * Open the image file and write a minimal qcow2 header.
1826      *
1827      * We keep things simple and start with a zero-sized image. We also
1828      * do without refcount blocks or a L1 table for now. We'll fix the
1829      * inconsistency later.
1830      *
1831      * We do need a refcount table because growing the refcount table means
1832      * allocating two new refcount blocks - the seconds of which would be at
1833      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
1834      * size for any qcow2 image.
1835      */
1836     BlockDriverState* bs;
1837     QCowHeader *header;
1838     uint64_t* refcount_table;
1839     Error *local_err = NULL;
1840     int ret;
1841 
1842     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
1843         /* Note: The following calculation does not need to be exact; if it is a
1844          * bit off, either some bytes will be "leaked" (which is fine) or we
1845          * will need to increase the file size by some bytes (which is fine,
1846          * too, as long as the bulk is allocated here). Therefore, using
1847          * floating point arithmetic is fine. */
1848         int64_t meta_size = 0;
1849         uint64_t nreftablee, nrefblocke, nl1e, nl2e;
1850         int64_t aligned_total_size = align_offset(total_size, cluster_size);
1851         int refblock_bits, refblock_size;
1852         /* refcount entry size in bytes */
1853         double rces = (1 << refcount_order) / 8.;
1854 
1855         /* see qcow2_open() */
1856         refblock_bits = cluster_bits - (refcount_order - 3);
1857         refblock_size = 1 << refblock_bits;
1858 
1859         /* header: 1 cluster */
1860         meta_size += cluster_size;
1861 
1862         /* total size of L2 tables */
1863         nl2e = aligned_total_size / cluster_size;
1864         nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
1865         meta_size += nl2e * sizeof(uint64_t);
1866 
1867         /* total size of L1 tables */
1868         nl1e = nl2e * sizeof(uint64_t) / cluster_size;
1869         nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
1870         meta_size += nl1e * sizeof(uint64_t);
1871 
1872         /* total size of refcount blocks
1873          *
1874          * note: every host cluster is reference-counted, including metadata
1875          * (even refcount blocks are recursively included).
1876          * Let:
1877          *   a = total_size (this is the guest disk size)
1878          *   m = meta size not including refcount blocks and refcount tables
1879          *   c = cluster size
1880          *   y1 = number of refcount blocks entries
1881          *   y2 = meta size including everything
1882          *   rces = refcount entry size in bytes
1883          * then,
1884          *   y1 = (y2 + a)/c
1885          *   y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m
1886          * we can get y1:
1887          *   y1 = (a + m) / (c - rces - rces * sizeof(u64) / c)
1888          */
1889         nrefblocke = (aligned_total_size + meta_size + cluster_size)
1890                    / (cluster_size - rces - rces * sizeof(uint64_t)
1891                                                  / cluster_size);
1892         meta_size += DIV_ROUND_UP(nrefblocke, refblock_size) * cluster_size;
1893 
1894         /* total size of refcount tables */
1895         nreftablee = nrefblocke / refblock_size;
1896         nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t));
1897         meta_size += nreftablee * sizeof(uint64_t);
1898 
1899         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
1900                             aligned_total_size + meta_size, &error_abort);
1901         qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc],
1902                      &error_abort);
1903     }
1904 
1905     ret = bdrv_create_file(filename, opts, &local_err);
1906     if (ret < 0) {
1907         error_propagate(errp, local_err);
1908         return ret;
1909     }
1910 
1911     bs = NULL;
1912     ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL,
1913                     NULL, &local_err);
1914     if (ret < 0) {
1915         error_propagate(errp, local_err);
1916         return ret;
1917     }
1918 
1919     /* Write the header */
1920     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
1921     header = g_malloc0(cluster_size);
1922     *header = (QCowHeader) {
1923         .magic                      = cpu_to_be32(QCOW_MAGIC),
1924         .version                    = cpu_to_be32(version),
1925         .cluster_bits               = cpu_to_be32(cluster_bits),
1926         .size                       = cpu_to_be64(0),
1927         .l1_table_offset            = cpu_to_be64(0),
1928         .l1_size                    = cpu_to_be32(0),
1929         .refcount_table_offset      = cpu_to_be64(cluster_size),
1930         .refcount_table_clusters    = cpu_to_be32(1),
1931         .refcount_order             = cpu_to_be32(refcount_order),
1932         .header_length              = cpu_to_be32(sizeof(*header)),
1933     };
1934 
1935     if (flags & BLOCK_FLAG_ENCRYPT) {
1936         header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1937     } else {
1938         header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1939     }
1940 
1941     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
1942         header->compatible_features |=
1943             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
1944     }
1945 
1946     ret = bdrv_pwrite(bs, 0, header, cluster_size);
1947     g_free(header);
1948     if (ret < 0) {
1949         error_setg_errno(errp, -ret, "Could not write qcow2 header");
1950         goto out;
1951     }
1952 
1953     /* Write a refcount table with one refcount block */
1954     refcount_table = g_malloc0(2 * cluster_size);
1955     refcount_table[0] = cpu_to_be64(2 * cluster_size);
1956     ret = bdrv_pwrite(bs, cluster_size, refcount_table, 2 * cluster_size);
1957     g_free(refcount_table);
1958 
1959     if (ret < 0) {
1960         error_setg_errno(errp, -ret, "Could not write refcount table");
1961         goto out;
1962     }
1963 
1964     bdrv_unref(bs);
1965     bs = NULL;
1966 
1967     /*
1968      * And now open the image and make it consistent first (i.e. increase the
1969      * refcount of the cluster that is occupied by the header and the refcount
1970      * table)
1971      */
1972     ret = bdrv_open(&bs, filename, NULL, NULL,
1973                     BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH,
1974                     &bdrv_qcow2, &local_err);
1975     if (ret < 0) {
1976         error_propagate(errp, local_err);
1977         goto out;
1978     }
1979 
1980     ret = qcow2_alloc_clusters(bs, 3 * cluster_size);
1981     if (ret < 0) {
1982         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
1983                          "header and refcount table");
1984         goto out;
1985 
1986     } else if (ret != 0) {
1987         error_report("Huh, first cluster in empty image is already in use?");
1988         abort();
1989     }
1990 
1991     /* Okay, now that we have a valid image, let's give it the right size */
1992     ret = bdrv_truncate(bs, total_size);
1993     if (ret < 0) {
1994         error_setg_errno(errp, -ret, "Could not resize image");
1995         goto out;
1996     }
1997 
1998     /* Want a backing file? There you go.*/
1999     if (backing_file) {
2000         ret = bdrv_change_backing_file(bs, backing_file, backing_format);
2001         if (ret < 0) {
2002             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
2003                              "with format '%s'", backing_file, backing_format);
2004             goto out;
2005         }
2006     }
2007 
2008     /* And if we're supposed to preallocate metadata, do that now */
2009     if (prealloc != PREALLOC_MODE_OFF) {
2010         BDRVQcowState *s = bs->opaque;
2011         qemu_co_mutex_lock(&s->lock);
2012         ret = preallocate(bs);
2013         qemu_co_mutex_unlock(&s->lock);
2014         if (ret < 0) {
2015             error_setg_errno(errp, -ret, "Could not preallocate metadata");
2016             goto out;
2017         }
2018     }
2019 
2020     bdrv_unref(bs);
2021     bs = NULL;
2022 
2023     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
2024     ret = bdrv_open(&bs, filename, NULL, NULL,
2025                     BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_BACKING,
2026                     &bdrv_qcow2, &local_err);
2027     if (local_err) {
2028         error_propagate(errp, local_err);
2029         goto out;
2030     }
2031 
2032     ret = 0;
2033 out:
2034     if (bs) {
2035         bdrv_unref(bs);
2036     }
2037     return ret;
2038 }
2039 
2040 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
2041 {
2042     char *backing_file = NULL;
2043     char *backing_fmt = NULL;
2044     char *buf = NULL;
2045     uint64_t size = 0;
2046     int flags = 0;
2047     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
2048     PreallocMode prealloc;
2049     int version = 3;
2050     uint64_t refcount_bits = 16;
2051     int refcount_order;
2052     Error *local_err = NULL;
2053     int ret;
2054 
2055     /* Read out options */
2056     size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2057                     BDRV_SECTOR_SIZE);
2058     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2059     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2060     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2061         flags |= BLOCK_FLAG_ENCRYPT;
2062     }
2063     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2064                                          DEFAULT_CLUSTER_SIZE);
2065     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2066     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
2067                                PREALLOC_MODE_MAX, PREALLOC_MODE_OFF,
2068                                &local_err);
2069     if (local_err) {
2070         error_propagate(errp, local_err);
2071         ret = -EINVAL;
2072         goto finish;
2073     }
2074     g_free(buf);
2075     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2076     if (!buf) {
2077         /* keep the default */
2078     } else if (!strcmp(buf, "0.10")) {
2079         version = 2;
2080     } else if (!strcmp(buf, "1.1")) {
2081         version = 3;
2082     } else {
2083         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2084         ret = -EINVAL;
2085         goto finish;
2086     }
2087 
2088     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2089         flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2090     }
2091 
2092     if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2093         error_setg(errp, "Backing file and preallocation cannot be used at "
2094                    "the same time");
2095         ret = -EINVAL;
2096         goto finish;
2097     }
2098 
2099     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2100         error_setg(errp, "Lazy refcounts only supported with compatibility "
2101                    "level 1.1 and above (use compat=1.1 or greater)");
2102         ret = -EINVAL;
2103         goto finish;
2104     }
2105 
2106     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS,
2107                                             refcount_bits);
2108     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2109         error_setg(errp, "Refcount width must be a power of two and may not "
2110                    "exceed 64 bits");
2111         ret = -EINVAL;
2112         goto finish;
2113     }
2114 
2115     if (version < 3 && refcount_bits != 16) {
2116         error_setg(errp, "Different refcount widths than 16 bits require "
2117                    "compatibility level 1.1 or above (use compat=1.1 or "
2118                    "greater)");
2119         ret = -EINVAL;
2120         goto finish;
2121     }
2122 
2123     refcount_order = ctz32(refcount_bits);
2124 
2125     ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2126                         cluster_size, prealloc, opts, version, refcount_order,
2127                         &local_err);
2128     if (local_err) {
2129         error_propagate(errp, local_err);
2130     }
2131 
2132 finish:
2133     g_free(backing_file);
2134     g_free(backing_fmt);
2135     g_free(buf);
2136     return ret;
2137 }
2138 
2139 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
2140     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2141 {
2142     int ret;
2143     BDRVQcowState *s = bs->opaque;
2144 
2145     /* Emulate misaligned zero writes */
2146     if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
2147         return -ENOTSUP;
2148     }
2149 
2150     /* Whatever is left can use real zero clusters */
2151     qemu_co_mutex_lock(&s->lock);
2152     ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2153         nb_sectors);
2154     qemu_co_mutex_unlock(&s->lock);
2155 
2156     return ret;
2157 }
2158 
2159 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
2160     int64_t sector_num, int nb_sectors)
2161 {
2162     int ret;
2163     BDRVQcowState *s = bs->opaque;
2164 
2165     qemu_co_mutex_lock(&s->lock);
2166     ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2167         nb_sectors, QCOW2_DISCARD_REQUEST, false);
2168     qemu_co_mutex_unlock(&s->lock);
2169     return ret;
2170 }
2171 
2172 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
2173 {
2174     BDRVQcowState *s = bs->opaque;
2175     int64_t new_l1_size;
2176     int ret;
2177 
2178     if (offset & 511) {
2179         error_report("The new size must be a multiple of 512");
2180         return -EINVAL;
2181     }
2182 
2183     /* cannot proceed if image has snapshots */
2184     if (s->nb_snapshots) {
2185         error_report("Can't resize an image which has snapshots");
2186         return -ENOTSUP;
2187     }
2188 
2189     /* shrinking is currently not supported */
2190     if (offset < bs->total_sectors * 512) {
2191         error_report("qcow2 doesn't support shrinking images yet");
2192         return -ENOTSUP;
2193     }
2194 
2195     new_l1_size = size_to_l1(s, offset);
2196     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
2197     if (ret < 0) {
2198         return ret;
2199     }
2200 
2201     /* write updated header.size */
2202     offset = cpu_to_be64(offset);
2203     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
2204                            &offset, sizeof(uint64_t));
2205     if (ret < 0) {
2206         return ret;
2207     }
2208 
2209     s->l1_vm_state_index = new_l1_size;
2210     return 0;
2211 }
2212 
2213 /* XXX: put compressed sectors first, then all the cluster aligned
2214    tables to avoid losing bytes in alignment */
2215 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
2216                                   const uint8_t *buf, int nb_sectors)
2217 {
2218     BDRVQcowState *s = bs->opaque;
2219     z_stream strm;
2220     int ret, out_len;
2221     uint8_t *out_buf;
2222     uint64_t cluster_offset;
2223 
2224     if (nb_sectors == 0) {
2225         /* align end of file to a sector boundary to ease reading with
2226            sector based I/Os */
2227         cluster_offset = bdrv_getlength(bs->file);
2228         return bdrv_truncate(bs->file, cluster_offset);
2229     }
2230 
2231     if (nb_sectors != s->cluster_sectors) {
2232         ret = -EINVAL;
2233 
2234         /* Zero-pad last write if image size is not cluster aligned */
2235         if (sector_num + nb_sectors == bs->total_sectors &&
2236             nb_sectors < s->cluster_sectors) {
2237             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
2238             memset(pad_buf, 0, s->cluster_size);
2239             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
2240             ret = qcow2_write_compressed(bs, sector_num,
2241                                          pad_buf, s->cluster_sectors);
2242             qemu_vfree(pad_buf);
2243         }
2244         return ret;
2245     }
2246 
2247     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
2248 
2249     /* best compression, small window, no zlib header */
2250     memset(&strm, 0, sizeof(strm));
2251     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
2252                        Z_DEFLATED, -12,
2253                        9, Z_DEFAULT_STRATEGY);
2254     if (ret != 0) {
2255         ret = -EINVAL;
2256         goto fail;
2257     }
2258 
2259     strm.avail_in = s->cluster_size;
2260     strm.next_in = (uint8_t *)buf;
2261     strm.avail_out = s->cluster_size;
2262     strm.next_out = out_buf;
2263 
2264     ret = deflate(&strm, Z_FINISH);
2265     if (ret != Z_STREAM_END && ret != Z_OK) {
2266         deflateEnd(&strm);
2267         ret = -EINVAL;
2268         goto fail;
2269     }
2270     out_len = strm.next_out - out_buf;
2271 
2272     deflateEnd(&strm);
2273 
2274     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
2275         /* could not compress: write normal cluster */
2276         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
2277         if (ret < 0) {
2278             goto fail;
2279         }
2280     } else {
2281         cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
2282             sector_num << 9, out_len);
2283         if (!cluster_offset) {
2284             ret = -EIO;
2285             goto fail;
2286         }
2287         cluster_offset &= s->cluster_offset_mask;
2288 
2289         ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
2290         if (ret < 0) {
2291             goto fail;
2292         }
2293 
2294         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
2295         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
2296         if (ret < 0) {
2297             goto fail;
2298         }
2299     }
2300 
2301     ret = 0;
2302 fail:
2303     g_free(out_buf);
2304     return ret;
2305 }
2306 
2307 static int make_completely_empty(BlockDriverState *bs)
2308 {
2309     BDRVQcowState *s = bs->opaque;
2310     int ret, l1_clusters;
2311     int64_t offset;
2312     uint64_t *new_reftable = NULL;
2313     uint64_t rt_entry, l1_size2;
2314     struct {
2315         uint64_t l1_offset;
2316         uint64_t reftable_offset;
2317         uint32_t reftable_clusters;
2318     } QEMU_PACKED l1_ofs_rt_ofs_cls;
2319 
2320     ret = qcow2_cache_empty(bs, s->l2_table_cache);
2321     if (ret < 0) {
2322         goto fail;
2323     }
2324 
2325     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
2326     if (ret < 0) {
2327         goto fail;
2328     }
2329 
2330     /* Refcounts will be broken utterly */
2331     ret = qcow2_mark_dirty(bs);
2332     if (ret < 0) {
2333         goto fail;
2334     }
2335 
2336     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2337 
2338     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2339     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
2340 
2341     /* After this call, neither the in-memory nor the on-disk refcount
2342      * information accurately describe the actual references */
2343 
2344     ret = bdrv_write_zeroes(bs->file, s->l1_table_offset / BDRV_SECTOR_SIZE,
2345                             l1_clusters * s->cluster_sectors, 0);
2346     if (ret < 0) {
2347         goto fail_broken_refcounts;
2348     }
2349     memset(s->l1_table, 0, l1_size2);
2350 
2351     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
2352 
2353     /* Overwrite enough clusters at the beginning of the sectors to place
2354      * the refcount table, a refcount block and the L1 table in; this may
2355      * overwrite parts of the existing refcount and L1 table, which is not
2356      * an issue because the dirty flag is set, complete data loss is in fact
2357      * desired and partial data loss is consequently fine as well */
2358     ret = bdrv_write_zeroes(bs->file, s->cluster_size / BDRV_SECTOR_SIZE,
2359                             (2 + l1_clusters) * s->cluster_size /
2360                             BDRV_SECTOR_SIZE, 0);
2361     /* This call (even if it failed overall) may have overwritten on-disk
2362      * refcount structures; in that case, the in-memory refcount information
2363      * will probably differ from the on-disk information which makes the BDS
2364      * unusable */
2365     if (ret < 0) {
2366         goto fail_broken_refcounts;
2367     }
2368 
2369     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2370     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
2371 
2372     /* "Create" an empty reftable (one cluster) directly after the image
2373      * header and an empty L1 table three clusters after the image header;
2374      * the cluster between those two will be used as the first refblock */
2375     cpu_to_be64w(&l1_ofs_rt_ofs_cls.l1_offset, 3 * s->cluster_size);
2376     cpu_to_be64w(&l1_ofs_rt_ofs_cls.reftable_offset, s->cluster_size);
2377     cpu_to_be32w(&l1_ofs_rt_ofs_cls.reftable_clusters, 1);
2378     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
2379                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
2380     if (ret < 0) {
2381         goto fail_broken_refcounts;
2382     }
2383 
2384     s->l1_table_offset = 3 * s->cluster_size;
2385 
2386     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
2387     if (!new_reftable) {
2388         ret = -ENOMEM;
2389         goto fail_broken_refcounts;
2390     }
2391 
2392     s->refcount_table_offset = s->cluster_size;
2393     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
2394 
2395     g_free(s->refcount_table);
2396     s->refcount_table = new_reftable;
2397     new_reftable = NULL;
2398 
2399     /* Now the in-memory refcount information again corresponds to the on-disk
2400      * information (reftable is empty and no refblocks (the refblock cache is
2401      * empty)); however, this means some clusters (e.g. the image header) are
2402      * referenced, but not refcounted, but the normal qcow2 code assumes that
2403      * the in-memory information is always correct */
2404 
2405     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
2406 
2407     /* Enter the first refblock into the reftable */
2408     rt_entry = cpu_to_be64(2 * s->cluster_size);
2409     ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
2410                            &rt_entry, sizeof(rt_entry));
2411     if (ret < 0) {
2412         goto fail_broken_refcounts;
2413     }
2414     s->refcount_table[0] = 2 * s->cluster_size;
2415 
2416     s->free_cluster_index = 0;
2417     assert(3 + l1_clusters <= s->refcount_block_size);
2418     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
2419     if (offset < 0) {
2420         ret = offset;
2421         goto fail_broken_refcounts;
2422     } else if (offset > 0) {
2423         error_report("First cluster in emptied image is in use");
2424         abort();
2425     }
2426 
2427     /* Now finally the in-memory information corresponds to the on-disk
2428      * structures and is correct */
2429     ret = qcow2_mark_clean(bs);
2430     if (ret < 0) {
2431         goto fail;
2432     }
2433 
2434     ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size);
2435     if (ret < 0) {
2436         goto fail;
2437     }
2438 
2439     return 0;
2440 
2441 fail_broken_refcounts:
2442     /* The BDS is unusable at this point. If we wanted to make it usable, we
2443      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
2444      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
2445      * again. However, because the functions which could have caused this error
2446      * path to be taken are used by those functions as well, it's very likely
2447      * that that sequence will fail as well. Therefore, just eject the BDS. */
2448     bs->drv = NULL;
2449 
2450 fail:
2451     g_free(new_reftable);
2452     return ret;
2453 }
2454 
2455 static int qcow2_make_empty(BlockDriverState *bs)
2456 {
2457     BDRVQcowState *s = bs->opaque;
2458     uint64_t start_sector;
2459     int sector_step = INT_MAX / BDRV_SECTOR_SIZE;
2460     int l1_clusters, ret = 0;
2461 
2462     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2463 
2464     if (s->qcow_version >= 3 && !s->snapshots &&
2465         3 + l1_clusters <= s->refcount_block_size) {
2466         /* The following function only works for qcow2 v3 images (it requires
2467          * the dirty flag) and only as long as there are no snapshots (because
2468          * it completely empties the image). Furthermore, the L1 table and three
2469          * additional clusters (image header, refcount table, one refcount
2470          * block) have to fit inside one refcount block. */
2471         return make_completely_empty(bs);
2472     }
2473 
2474     /* This fallback code simply discards every active cluster; this is slow,
2475      * but works in all cases */
2476     for (start_sector = 0; start_sector < bs->total_sectors;
2477          start_sector += sector_step)
2478     {
2479         /* As this function is generally used after committing an external
2480          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
2481          * default action for this kind of discard is to pass the discard,
2482          * which will ideally result in an actually smaller image file, as
2483          * is probably desired. */
2484         ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE,
2485                                      MIN(sector_step,
2486                                          bs->total_sectors - start_sector),
2487                                      QCOW2_DISCARD_SNAPSHOT, true);
2488         if (ret < 0) {
2489             break;
2490         }
2491     }
2492 
2493     return ret;
2494 }
2495 
2496 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
2497 {
2498     BDRVQcowState *s = bs->opaque;
2499     int ret;
2500 
2501     qemu_co_mutex_lock(&s->lock);
2502     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2503     if (ret < 0) {
2504         qemu_co_mutex_unlock(&s->lock);
2505         return ret;
2506     }
2507 
2508     if (qcow2_need_accurate_refcounts(s)) {
2509         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2510         if (ret < 0) {
2511             qemu_co_mutex_unlock(&s->lock);
2512             return ret;
2513         }
2514     }
2515     qemu_co_mutex_unlock(&s->lock);
2516 
2517     return 0;
2518 }
2519 
2520 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2521 {
2522     BDRVQcowState *s = bs->opaque;
2523     bdi->unallocated_blocks_are_zero = true;
2524     bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
2525     bdi->cluster_size = s->cluster_size;
2526     bdi->vm_state_offset = qcow2_vm_state_offset(s);
2527     return 0;
2528 }
2529 
2530 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
2531 {
2532     BDRVQcowState *s = bs->opaque;
2533     ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
2534 
2535     *spec_info = (ImageInfoSpecific){
2536         .kind  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
2537         {
2538             .qcow2 = g_new(ImageInfoSpecificQCow2, 1),
2539         },
2540     };
2541     if (s->qcow_version == 2) {
2542         *spec_info->qcow2 = (ImageInfoSpecificQCow2){
2543             .compat             = g_strdup("0.10"),
2544             .refcount_bits      = s->refcount_bits,
2545         };
2546     } else if (s->qcow_version == 3) {
2547         *spec_info->qcow2 = (ImageInfoSpecificQCow2){
2548             .compat             = g_strdup("1.1"),
2549             .lazy_refcounts     = s->compatible_features &
2550                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
2551             .has_lazy_refcounts = true,
2552             .corrupt            = s->incompatible_features &
2553                                   QCOW2_INCOMPAT_CORRUPT,
2554             .has_corrupt        = true,
2555             .refcount_bits      = s->refcount_bits,
2556         };
2557     }
2558 
2559     return spec_info;
2560 }
2561 
2562 #if 0
2563 static void dump_refcounts(BlockDriverState *bs)
2564 {
2565     BDRVQcowState *s = bs->opaque;
2566     int64_t nb_clusters, k, k1, size;
2567     int refcount;
2568 
2569     size = bdrv_getlength(bs->file);
2570     nb_clusters = size_to_clusters(s, size);
2571     for(k = 0; k < nb_clusters;) {
2572         k1 = k;
2573         refcount = get_refcount(bs, k);
2574         k++;
2575         while (k < nb_clusters && get_refcount(bs, k) == refcount)
2576             k++;
2577         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
2578                k - k1);
2579     }
2580 }
2581 #endif
2582 
2583 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2584                               int64_t pos)
2585 {
2586     BDRVQcowState *s = bs->opaque;
2587     int64_t total_sectors = bs->total_sectors;
2588     bool zero_beyond_eof = bs->zero_beyond_eof;
2589     int ret;
2590 
2591     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
2592     bs->zero_beyond_eof = false;
2593     ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov);
2594     bs->zero_beyond_eof = zero_beyond_eof;
2595 
2596     /* bdrv_co_do_writev will have increased the total_sectors value to include
2597      * the VM state - the VM state is however not an actual part of the block
2598      * device, therefore, we need to restore the old value. */
2599     bs->total_sectors = total_sectors;
2600 
2601     return ret;
2602 }
2603 
2604 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2605                               int64_t pos, int size)
2606 {
2607     BDRVQcowState *s = bs->opaque;
2608     bool zero_beyond_eof = bs->zero_beyond_eof;
2609     int ret;
2610 
2611     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
2612     bs->zero_beyond_eof = false;
2613     ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
2614     bs->zero_beyond_eof = zero_beyond_eof;
2615 
2616     return ret;
2617 }
2618 
2619 /*
2620  * Downgrades an image's version. To achieve this, any incompatible features
2621  * have to be removed.
2622  */
2623 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
2624                            BlockDriverAmendStatusCB *status_cb)
2625 {
2626     BDRVQcowState *s = bs->opaque;
2627     int current_version = s->qcow_version;
2628     int ret;
2629 
2630     if (target_version == current_version) {
2631         return 0;
2632     } else if (target_version > current_version) {
2633         return -EINVAL;
2634     } else if (target_version != 2) {
2635         return -EINVAL;
2636     }
2637 
2638     if (s->refcount_order != 4) {
2639         /* we would have to convert the image to a refcount_order == 4 image
2640          * here; however, since qemu (at the time of writing this) does not
2641          * support anything different than 4 anyway, there is no point in doing
2642          * so right now; however, we should error out (if qemu supports this in
2643          * the future and this code has not been adapted) */
2644         error_report("qcow2_downgrade: Image refcount orders other than 4 are "
2645                      "currently not supported.");
2646         return -ENOTSUP;
2647     }
2648 
2649     /* clear incompatible features */
2650     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
2651         ret = qcow2_mark_clean(bs);
2652         if (ret < 0) {
2653             return ret;
2654         }
2655     }
2656 
2657     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2658      * the first place; if that happens nonetheless, returning -ENOTSUP is the
2659      * best thing to do anyway */
2660 
2661     if (s->incompatible_features) {
2662         return -ENOTSUP;
2663     }
2664 
2665     /* since we can ignore compatible features, we can set them to 0 as well */
2666     s->compatible_features = 0;
2667     /* if lazy refcounts have been used, they have already been fixed through
2668      * clearing the dirty flag */
2669 
2670     /* clearing autoclear features is trivial */
2671     s->autoclear_features = 0;
2672 
2673     ret = qcow2_expand_zero_clusters(bs, status_cb);
2674     if (ret < 0) {
2675         return ret;
2676     }
2677 
2678     s->qcow_version = target_version;
2679     ret = qcow2_update_header(bs);
2680     if (ret < 0) {
2681         s->qcow_version = current_version;
2682         return ret;
2683     }
2684     return 0;
2685 }
2686 
2687 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
2688                                BlockDriverAmendStatusCB *status_cb)
2689 {
2690     BDRVQcowState *s = bs->opaque;
2691     int old_version = s->qcow_version, new_version = old_version;
2692     uint64_t new_size = 0;
2693     const char *backing_file = NULL, *backing_format = NULL;
2694     bool lazy_refcounts = s->use_lazy_refcounts;
2695     const char *compat = NULL;
2696     uint64_t cluster_size = s->cluster_size;
2697     bool encrypt;
2698     int ret;
2699     QemuOptDesc *desc = opts->list->desc;
2700 
2701     while (desc && desc->name) {
2702         if (!qemu_opt_find(opts, desc->name)) {
2703             /* only change explicitly defined options */
2704             desc++;
2705             continue;
2706         }
2707 
2708         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
2709             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
2710             if (!compat) {
2711                 /* preserve default */
2712             } else if (!strcmp(compat, "0.10")) {
2713                 new_version = 2;
2714             } else if (!strcmp(compat, "1.1")) {
2715                 new_version = 3;
2716             } else {
2717                 fprintf(stderr, "Unknown compatibility level %s.\n", compat);
2718                 return -EINVAL;
2719             }
2720         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
2721             fprintf(stderr, "Cannot change preallocation mode.\n");
2722             return -ENOTSUP;
2723         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
2724             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
2725         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
2726             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
2727         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
2728             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
2729         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
2730             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
2731                                         s->crypt_method);
2732             if (encrypt != !!s->crypt_method) {
2733                 fprintf(stderr, "Changing the encryption flag is not "
2734                         "supported.\n");
2735                 return -ENOTSUP;
2736             }
2737         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
2738             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
2739                                              cluster_size);
2740             if (cluster_size != s->cluster_size) {
2741                 fprintf(stderr, "Changing the cluster size is not "
2742                         "supported.\n");
2743                 return -ENOTSUP;
2744             }
2745         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
2746             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
2747                                                lazy_refcounts);
2748         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
2749             error_report("Cannot change refcount entry width");
2750             return -ENOTSUP;
2751         } else {
2752             /* if this assertion fails, this probably means a new option was
2753              * added without having it covered here */
2754             assert(false);
2755         }
2756 
2757         desc++;
2758     }
2759 
2760     if (new_version != old_version) {
2761         if (new_version > old_version) {
2762             /* Upgrade */
2763             s->qcow_version = new_version;
2764             ret = qcow2_update_header(bs);
2765             if (ret < 0) {
2766                 s->qcow_version = old_version;
2767                 return ret;
2768             }
2769         } else {
2770             ret = qcow2_downgrade(bs, new_version, status_cb);
2771             if (ret < 0) {
2772                 return ret;
2773             }
2774         }
2775     }
2776 
2777     if (backing_file || backing_format) {
2778         ret = qcow2_change_backing_file(bs,
2779                     backing_file ?: s->image_backing_file,
2780                     backing_format ?: s->image_backing_format);
2781         if (ret < 0) {
2782             return ret;
2783         }
2784     }
2785 
2786     if (s->use_lazy_refcounts != lazy_refcounts) {
2787         if (lazy_refcounts) {
2788             if (s->qcow_version < 3) {
2789                 fprintf(stderr, "Lazy refcounts only supported with compatibility "
2790                         "level 1.1 and above (use compat=1.1 or greater)\n");
2791                 return -EINVAL;
2792             }
2793             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
2794             ret = qcow2_update_header(bs);
2795             if (ret < 0) {
2796                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
2797                 return ret;
2798             }
2799             s->use_lazy_refcounts = true;
2800         } else {
2801             /* make image clean first */
2802             ret = qcow2_mark_clean(bs);
2803             if (ret < 0) {
2804                 return ret;
2805             }
2806             /* now disallow lazy refcounts */
2807             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
2808             ret = qcow2_update_header(bs);
2809             if (ret < 0) {
2810                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
2811                 return ret;
2812             }
2813             s->use_lazy_refcounts = false;
2814         }
2815     }
2816 
2817     if (new_size) {
2818         ret = bdrv_truncate(bs, new_size);
2819         if (ret < 0) {
2820             return ret;
2821         }
2822     }
2823 
2824     return 0;
2825 }
2826 
2827 /*
2828  * If offset or size are negative, respectively, they will not be included in
2829  * the BLOCK_IMAGE_CORRUPTED event emitted.
2830  * fatal will be ignored for read-only BDS; corruptions found there will always
2831  * be considered non-fatal.
2832  */
2833 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
2834                              int64_t size, const char *message_format, ...)
2835 {
2836     BDRVQcowState *s = bs->opaque;
2837     const char *node_name;
2838     char *message;
2839     va_list ap;
2840 
2841     fatal = fatal && !bs->read_only;
2842 
2843     if (s->signaled_corruption &&
2844         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
2845     {
2846         return;
2847     }
2848 
2849     va_start(ap, message_format);
2850     message = g_strdup_vprintf(message_format, ap);
2851     va_end(ap);
2852 
2853     if (fatal) {
2854         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
2855                 "corruption events will be suppressed\n", message);
2856     } else {
2857         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
2858                 "corruption events will be suppressed\n", message);
2859     }
2860 
2861     node_name = bdrv_get_node_name(bs);
2862     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
2863                                           *node_name != '\0', node_name,
2864                                           message, offset >= 0, offset,
2865                                           size >= 0, size,
2866                                           fatal, &error_abort);
2867     g_free(message);
2868 
2869     if (fatal) {
2870         qcow2_mark_corrupt(bs);
2871         bs->drv = NULL; /* make BDS unusable */
2872     }
2873 
2874     s->signaled_corruption = true;
2875 }
2876 
2877 static QemuOptsList qcow2_create_opts = {
2878     .name = "qcow2-create-opts",
2879     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
2880     .desc = {
2881         {
2882             .name = BLOCK_OPT_SIZE,
2883             .type = QEMU_OPT_SIZE,
2884             .help = "Virtual disk size"
2885         },
2886         {
2887             .name = BLOCK_OPT_COMPAT_LEVEL,
2888             .type = QEMU_OPT_STRING,
2889             .help = "Compatibility level (0.10 or 1.1)"
2890         },
2891         {
2892             .name = BLOCK_OPT_BACKING_FILE,
2893             .type = QEMU_OPT_STRING,
2894             .help = "File name of a base image"
2895         },
2896         {
2897             .name = BLOCK_OPT_BACKING_FMT,
2898             .type = QEMU_OPT_STRING,
2899             .help = "Image format of the base image"
2900         },
2901         {
2902             .name = BLOCK_OPT_ENCRYPT,
2903             .type = QEMU_OPT_BOOL,
2904             .help = "Encrypt the image",
2905             .def_value_str = "off"
2906         },
2907         {
2908             .name = BLOCK_OPT_CLUSTER_SIZE,
2909             .type = QEMU_OPT_SIZE,
2910             .help = "qcow2 cluster size",
2911             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
2912         },
2913         {
2914             .name = BLOCK_OPT_PREALLOC,
2915             .type = QEMU_OPT_STRING,
2916             .help = "Preallocation mode (allowed values: off, metadata, "
2917                     "falloc, full)"
2918         },
2919         {
2920             .name = BLOCK_OPT_LAZY_REFCOUNTS,
2921             .type = QEMU_OPT_BOOL,
2922             .help = "Postpone refcount updates",
2923             .def_value_str = "off"
2924         },
2925         {
2926             .name = BLOCK_OPT_REFCOUNT_BITS,
2927             .type = QEMU_OPT_NUMBER,
2928             .help = "Width of a reference count entry in bits",
2929             .def_value_str = "16"
2930         },
2931         { /* end of list */ }
2932     }
2933 };
2934 
2935 BlockDriver bdrv_qcow2 = {
2936     .format_name        = "qcow2",
2937     .instance_size      = sizeof(BDRVQcowState),
2938     .bdrv_probe         = qcow2_probe,
2939     .bdrv_open          = qcow2_open,
2940     .bdrv_close         = qcow2_close,
2941     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
2942     .bdrv_create        = qcow2_create,
2943     .bdrv_has_zero_init = bdrv_has_zero_init_1,
2944     .bdrv_co_get_block_status = qcow2_co_get_block_status,
2945     .bdrv_set_key       = qcow2_set_key,
2946 
2947     .bdrv_co_readv          = qcow2_co_readv,
2948     .bdrv_co_writev         = qcow2_co_writev,
2949     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
2950 
2951     .bdrv_co_write_zeroes   = qcow2_co_write_zeroes,
2952     .bdrv_co_discard        = qcow2_co_discard,
2953     .bdrv_truncate          = qcow2_truncate,
2954     .bdrv_write_compressed  = qcow2_write_compressed,
2955     .bdrv_make_empty        = qcow2_make_empty,
2956 
2957     .bdrv_snapshot_create   = qcow2_snapshot_create,
2958     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
2959     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
2960     .bdrv_snapshot_list     = qcow2_snapshot_list,
2961     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
2962     .bdrv_get_info          = qcow2_get_info,
2963     .bdrv_get_specific_info = qcow2_get_specific_info,
2964 
2965     .bdrv_save_vmstate    = qcow2_save_vmstate,
2966     .bdrv_load_vmstate    = qcow2_load_vmstate,
2967 
2968     .supports_backing           = true,
2969     .bdrv_change_backing_file   = qcow2_change_backing_file,
2970 
2971     .bdrv_refresh_limits        = qcow2_refresh_limits,
2972     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
2973 
2974     .create_opts         = &qcow2_create_opts,
2975     .bdrv_check          = qcow2_check,
2976     .bdrv_amend_options  = qcow2_amend_options,
2977 };
2978 
2979 static void bdrv_qcow2_init(void)
2980 {
2981     bdrv_register(&bdrv_qcow2);
2982 }
2983 
2984 block_init(bdrv_qcow2_init);
2985