xref: /openbmc/qemu/block/qcow2.c (revision 4b4629d9)
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/osdep.h"
25 #include "block/block_int.h"
26 #include "sysemu/block-backend.h"
27 #include "qemu/module.h"
28 #include <zlib.h>
29 #include "block/qcow2.h"
30 #include "qemu/error-report.h"
31 #include "qapi/qmp/qerror.h"
32 #include "qapi/qmp/qbool.h"
33 #include "qapi/util.h"
34 #include "qapi/qmp/types.h"
35 #include "qapi-event.h"
36 #include "trace.h"
37 #include "qemu/option_int.h"
38 #include "qemu/cutils.h"
39 
40 /*
41   Differences with QCOW:
42 
43   - Support for multiple incremental snapshots.
44   - Memory management by reference counts.
45   - Clusters which have a reference count of one have the bit
46     QCOW_OFLAG_COPIED to optimize write performance.
47   - Size of compressed clusters is stored in sectors to reduce bit usage
48     in the cluster offsets.
49   - Support for storing additional data (such as the VM state) in the
50     snapshots.
51   - If a backing store is used, the cluster size is not constrained
52     (could be backported to QCOW).
53   - L2 tables have always a size of one cluster.
54 */
55 
56 
57 typedef struct {
58     uint32_t magic;
59     uint32_t len;
60 } QEMU_PACKED QCowExtension;
61 
62 #define  QCOW2_EXT_MAGIC_END 0
63 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
64 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
65 
66 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
67 {
68     const QCowHeader *cow_header = (const void *)buf;
69 
70     if (buf_size >= sizeof(QCowHeader) &&
71         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
72         be32_to_cpu(cow_header->version) >= 2)
73         return 100;
74     else
75         return 0;
76 }
77 
78 
79 /*
80  * read qcow2 extension and fill bs
81  * start reading from start_offset
82  * finish reading upon magic of value 0 or when end_offset reached
83  * unknown magic is skipped (future extension this version knows nothing about)
84  * return 0 upon success, non-0 otherwise
85  */
86 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
87                                  uint64_t end_offset, void **p_feature_table,
88                                  Error **errp)
89 {
90     BDRVQcow2State *s = bs->opaque;
91     QCowExtension ext;
92     uint64_t offset;
93     int ret;
94 
95 #ifdef DEBUG_EXT
96     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
97 #endif
98     offset = start_offset;
99     while (offset < end_offset) {
100 
101 #ifdef DEBUG_EXT
102         /* Sanity check */
103         if (offset > s->cluster_size)
104             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
105 
106         printf("attempting to read extended header in offset %lu\n", offset);
107 #endif
108 
109         ret = bdrv_pread(bs->file->bs, offset, &ext, sizeof(ext));
110         if (ret < 0) {
111             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
112                              "pread fail from offset %" PRIu64, offset);
113             return 1;
114         }
115         be32_to_cpus(&ext.magic);
116         be32_to_cpus(&ext.len);
117         offset += sizeof(ext);
118 #ifdef DEBUG_EXT
119         printf("ext.magic = 0x%x\n", ext.magic);
120 #endif
121         if (offset > end_offset || ext.len > end_offset - offset) {
122             error_setg(errp, "Header extension too large");
123             return -EINVAL;
124         }
125 
126         switch (ext.magic) {
127         case QCOW2_EXT_MAGIC_END:
128             return 0;
129 
130         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
131             if (ext.len >= sizeof(bs->backing_format)) {
132                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
133                            " too large (>=%zu)", ext.len,
134                            sizeof(bs->backing_format));
135                 return 2;
136             }
137             ret = bdrv_pread(bs->file->bs, offset, bs->backing_format, ext.len);
138             if (ret < 0) {
139                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
140                                  "Could not read format name");
141                 return 3;
142             }
143             bs->backing_format[ext.len] = '\0';
144             s->image_backing_format = g_strdup(bs->backing_format);
145 #ifdef DEBUG_EXT
146             printf("Qcow2: Got format extension %s\n", bs->backing_format);
147 #endif
148             break;
149 
150         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
151             if (p_feature_table != NULL) {
152                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
153                 ret = bdrv_pread(bs->file->bs, offset , feature_table, ext.len);
154                 if (ret < 0) {
155                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
156                                      "Could not read table");
157                     return ret;
158                 }
159 
160                 *p_feature_table = feature_table;
161             }
162             break;
163 
164         default:
165             /* unknown magic - save it in case we need to rewrite the header */
166             {
167                 Qcow2UnknownHeaderExtension *uext;
168 
169                 uext = g_malloc0(sizeof(*uext)  + ext.len);
170                 uext->magic = ext.magic;
171                 uext->len = ext.len;
172                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
173 
174                 ret = bdrv_pread(bs->file->bs, offset , uext->data, uext->len);
175                 if (ret < 0) {
176                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
177                                      "Could not read data");
178                     return ret;
179                 }
180             }
181             break;
182         }
183 
184         offset += ((ext.len + 7) & ~7);
185     }
186 
187     return 0;
188 }
189 
190 static void cleanup_unknown_header_ext(BlockDriverState *bs)
191 {
192     BDRVQcow2State *s = bs->opaque;
193     Qcow2UnknownHeaderExtension *uext, *next;
194 
195     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
196         QLIST_REMOVE(uext, next);
197         g_free(uext);
198     }
199 }
200 
201 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
202                                        uint64_t mask)
203 {
204     char *features = g_strdup("");
205     char *old;
206 
207     while (table && table->name[0] != '\0') {
208         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
209             if (mask & (1ULL << table->bit)) {
210                 old = features;
211                 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
212                                            table->name);
213                 g_free(old);
214                 mask &= ~(1ULL << table->bit);
215             }
216         }
217         table++;
218     }
219 
220     if (mask) {
221         old = features;
222         features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
223                                    old, *old ? ", " : "", mask);
224         g_free(old);
225     }
226 
227     error_setg(errp, "Unsupported qcow2 feature(s): %s", features);
228     g_free(features);
229 }
230 
231 /*
232  * Sets the dirty bit and flushes afterwards if necessary.
233  *
234  * The incompatible_features bit is only set if the image file header was
235  * updated successfully.  Therefore it is not required to check the return
236  * value of this function.
237  */
238 int qcow2_mark_dirty(BlockDriverState *bs)
239 {
240     BDRVQcow2State *s = bs->opaque;
241     uint64_t val;
242     int ret;
243 
244     assert(s->qcow_version >= 3);
245 
246     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
247         return 0; /* already dirty */
248     }
249 
250     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
251     ret = bdrv_pwrite(bs->file->bs, offsetof(QCowHeader, incompatible_features),
252                       &val, sizeof(val));
253     if (ret < 0) {
254         return ret;
255     }
256     ret = bdrv_flush(bs->file->bs);
257     if (ret < 0) {
258         return ret;
259     }
260 
261     /* Only treat image as dirty if the header was updated successfully */
262     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
263     return 0;
264 }
265 
266 /*
267  * Clears the dirty bit and flushes before if necessary.  Only call this
268  * function when there are no pending requests, it does not guard against
269  * concurrent requests dirtying the image.
270  */
271 static int qcow2_mark_clean(BlockDriverState *bs)
272 {
273     BDRVQcow2State *s = bs->opaque;
274 
275     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
276         int ret;
277 
278         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
279 
280         ret = bdrv_flush(bs);
281         if (ret < 0) {
282             return ret;
283         }
284 
285         return qcow2_update_header(bs);
286     }
287     return 0;
288 }
289 
290 /*
291  * Marks the image as corrupt.
292  */
293 int qcow2_mark_corrupt(BlockDriverState *bs)
294 {
295     BDRVQcow2State *s = bs->opaque;
296 
297     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
298     return qcow2_update_header(bs);
299 }
300 
301 /*
302  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
303  * before if necessary.
304  */
305 int qcow2_mark_consistent(BlockDriverState *bs)
306 {
307     BDRVQcow2State *s = bs->opaque;
308 
309     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
310         int ret = bdrv_flush(bs);
311         if (ret < 0) {
312             return ret;
313         }
314 
315         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
316         return qcow2_update_header(bs);
317     }
318     return 0;
319 }
320 
321 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
322                        BdrvCheckMode fix)
323 {
324     int ret = qcow2_check_refcounts(bs, result, fix);
325     if (ret < 0) {
326         return ret;
327     }
328 
329     if (fix && result->check_errors == 0 && result->corruptions == 0) {
330         ret = qcow2_mark_clean(bs);
331         if (ret < 0) {
332             return ret;
333         }
334         return qcow2_mark_consistent(bs);
335     }
336     return ret;
337 }
338 
339 static int validate_table_offset(BlockDriverState *bs, uint64_t offset,
340                                  uint64_t entries, size_t entry_len)
341 {
342     BDRVQcow2State *s = bs->opaque;
343     uint64_t size;
344 
345     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
346      * because values will be passed to qemu functions taking int64_t. */
347     if (entries > INT64_MAX / entry_len) {
348         return -EINVAL;
349     }
350 
351     size = entries * entry_len;
352 
353     if (INT64_MAX - size < offset) {
354         return -EINVAL;
355     }
356 
357     /* Tables must be cluster aligned */
358     if (offset & (s->cluster_size - 1)) {
359         return -EINVAL;
360     }
361 
362     return 0;
363 }
364 
365 static QemuOptsList qcow2_runtime_opts = {
366     .name = "qcow2",
367     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
368     .desc = {
369         {
370             .name = QCOW2_OPT_LAZY_REFCOUNTS,
371             .type = QEMU_OPT_BOOL,
372             .help = "Postpone refcount updates",
373         },
374         {
375             .name = QCOW2_OPT_DISCARD_REQUEST,
376             .type = QEMU_OPT_BOOL,
377             .help = "Pass guest discard requests to the layer below",
378         },
379         {
380             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
381             .type = QEMU_OPT_BOOL,
382             .help = "Generate discard requests when snapshot related space "
383                     "is freed",
384         },
385         {
386             .name = QCOW2_OPT_DISCARD_OTHER,
387             .type = QEMU_OPT_BOOL,
388             .help = "Generate discard requests when other clusters are freed",
389         },
390         {
391             .name = QCOW2_OPT_OVERLAP,
392             .type = QEMU_OPT_STRING,
393             .help = "Selects which overlap checks to perform from a range of "
394                     "templates (none, constant, cached, all)",
395         },
396         {
397             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
398             .type = QEMU_OPT_STRING,
399             .help = "Selects which overlap checks to perform from a range of "
400                     "templates (none, constant, cached, all)",
401         },
402         {
403             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
404             .type = QEMU_OPT_BOOL,
405             .help = "Check for unintended writes into the main qcow2 header",
406         },
407         {
408             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
409             .type = QEMU_OPT_BOOL,
410             .help = "Check for unintended writes into the active L1 table",
411         },
412         {
413             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
414             .type = QEMU_OPT_BOOL,
415             .help = "Check for unintended writes into an active L2 table",
416         },
417         {
418             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
419             .type = QEMU_OPT_BOOL,
420             .help = "Check for unintended writes into the refcount table",
421         },
422         {
423             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
424             .type = QEMU_OPT_BOOL,
425             .help = "Check for unintended writes into a refcount block",
426         },
427         {
428             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
429             .type = QEMU_OPT_BOOL,
430             .help = "Check for unintended writes into the snapshot table",
431         },
432         {
433             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
434             .type = QEMU_OPT_BOOL,
435             .help = "Check for unintended writes into an inactive L1 table",
436         },
437         {
438             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
439             .type = QEMU_OPT_BOOL,
440             .help = "Check for unintended writes into an inactive L2 table",
441         },
442         {
443             .name = QCOW2_OPT_CACHE_SIZE,
444             .type = QEMU_OPT_SIZE,
445             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
446                     "cache size",
447         },
448         {
449             .name = QCOW2_OPT_L2_CACHE_SIZE,
450             .type = QEMU_OPT_SIZE,
451             .help = "Maximum L2 table cache size",
452         },
453         {
454             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
455             .type = QEMU_OPT_SIZE,
456             .help = "Maximum refcount block cache size",
457         },
458         {
459             .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
460             .type = QEMU_OPT_NUMBER,
461             .help = "Clean unused cache entries after this time (in seconds)",
462         },
463         { /* end of list */ }
464     },
465 };
466 
467 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
468     [QCOW2_OL_MAIN_HEADER_BITNR]    = QCOW2_OPT_OVERLAP_MAIN_HEADER,
469     [QCOW2_OL_ACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L1,
470     [QCOW2_OL_ACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L2,
471     [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
472     [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
473     [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
474     [QCOW2_OL_INACTIVE_L1_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L1,
475     [QCOW2_OL_INACTIVE_L2_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L2,
476 };
477 
478 static void cache_clean_timer_cb(void *opaque)
479 {
480     BlockDriverState *bs = opaque;
481     BDRVQcow2State *s = bs->opaque;
482     qcow2_cache_clean_unused(bs, s->l2_table_cache);
483     qcow2_cache_clean_unused(bs, s->refcount_block_cache);
484     timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
485               (int64_t) s->cache_clean_interval * 1000);
486 }
487 
488 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
489 {
490     BDRVQcow2State *s = bs->opaque;
491     if (s->cache_clean_interval > 0) {
492         s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
493                                              SCALE_MS, cache_clean_timer_cb,
494                                              bs);
495         timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
496                   (int64_t) s->cache_clean_interval * 1000);
497     }
498 }
499 
500 static void cache_clean_timer_del(BlockDriverState *bs)
501 {
502     BDRVQcow2State *s = bs->opaque;
503     if (s->cache_clean_timer) {
504         timer_del(s->cache_clean_timer);
505         timer_free(s->cache_clean_timer);
506         s->cache_clean_timer = NULL;
507     }
508 }
509 
510 static void qcow2_detach_aio_context(BlockDriverState *bs)
511 {
512     cache_clean_timer_del(bs);
513 }
514 
515 static void qcow2_attach_aio_context(BlockDriverState *bs,
516                                      AioContext *new_context)
517 {
518     cache_clean_timer_init(bs, new_context);
519 }
520 
521 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
522                              uint64_t *l2_cache_size,
523                              uint64_t *refcount_cache_size, Error **errp)
524 {
525     BDRVQcow2State *s = bs->opaque;
526     uint64_t combined_cache_size;
527     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
528 
529     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
530     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
531     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
532 
533     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
534     *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0);
535     *refcount_cache_size = qemu_opt_get_size(opts,
536                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
537 
538     if (combined_cache_size_set) {
539         if (l2_cache_size_set && refcount_cache_size_set) {
540             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
541                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
542                        "the same time");
543             return;
544         } else if (*l2_cache_size > combined_cache_size) {
545             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
546                        QCOW2_OPT_CACHE_SIZE);
547             return;
548         } else if (*refcount_cache_size > combined_cache_size) {
549             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
550                        QCOW2_OPT_CACHE_SIZE);
551             return;
552         }
553 
554         if (l2_cache_size_set) {
555             *refcount_cache_size = combined_cache_size - *l2_cache_size;
556         } else if (refcount_cache_size_set) {
557             *l2_cache_size = combined_cache_size - *refcount_cache_size;
558         } else {
559             *refcount_cache_size = combined_cache_size
560                                  / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1);
561             *l2_cache_size = combined_cache_size - *refcount_cache_size;
562         }
563     } else {
564         if (!l2_cache_size_set && !refcount_cache_size_set) {
565             *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE,
566                                  (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
567                                  * s->cluster_size);
568             *refcount_cache_size = *l2_cache_size
569                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
570         } else if (!l2_cache_size_set) {
571             *l2_cache_size = *refcount_cache_size
572                            * DEFAULT_L2_REFCOUNT_SIZE_RATIO;
573         } else if (!refcount_cache_size_set) {
574             *refcount_cache_size = *l2_cache_size
575                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
576         }
577     }
578 }
579 
580 typedef struct Qcow2ReopenState {
581     Qcow2Cache *l2_table_cache;
582     Qcow2Cache *refcount_block_cache;
583     bool use_lazy_refcounts;
584     int overlap_check;
585     bool discard_passthrough[QCOW2_DISCARD_MAX];
586     uint64_t cache_clean_interval;
587 } Qcow2ReopenState;
588 
589 static int qcow2_update_options_prepare(BlockDriverState *bs,
590                                         Qcow2ReopenState *r,
591                                         QDict *options, int flags,
592                                         Error **errp)
593 {
594     BDRVQcow2State *s = bs->opaque;
595     QemuOpts *opts = NULL;
596     const char *opt_overlap_check, *opt_overlap_check_template;
597     int overlap_check_template = 0;
598     uint64_t l2_cache_size, refcount_cache_size;
599     int i;
600     Error *local_err = NULL;
601     int ret;
602 
603     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
604     qemu_opts_absorb_qdict(opts, options, &local_err);
605     if (local_err) {
606         error_propagate(errp, local_err);
607         ret = -EINVAL;
608         goto fail;
609     }
610 
611     /* get L2 table/refcount block cache size from command line options */
612     read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size,
613                      &local_err);
614     if (local_err) {
615         error_propagate(errp, local_err);
616         ret = -EINVAL;
617         goto fail;
618     }
619 
620     l2_cache_size /= s->cluster_size;
621     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
622         l2_cache_size = MIN_L2_CACHE_SIZE;
623     }
624     if (l2_cache_size > INT_MAX) {
625         error_setg(errp, "L2 cache size too big");
626         ret = -EINVAL;
627         goto fail;
628     }
629 
630     refcount_cache_size /= s->cluster_size;
631     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
632         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
633     }
634     if (refcount_cache_size > INT_MAX) {
635         error_setg(errp, "Refcount cache size too big");
636         ret = -EINVAL;
637         goto fail;
638     }
639 
640     /* alloc new L2 table/refcount block cache, flush old one */
641     if (s->l2_table_cache) {
642         ret = qcow2_cache_flush(bs, s->l2_table_cache);
643         if (ret) {
644             error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
645             goto fail;
646         }
647     }
648 
649     if (s->refcount_block_cache) {
650         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
651         if (ret) {
652             error_setg_errno(errp, -ret,
653                              "Failed to flush the refcount block cache");
654             goto fail;
655         }
656     }
657 
658     r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size);
659     r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size);
660     if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
661         error_setg(errp, "Could not allocate metadata caches");
662         ret = -ENOMEM;
663         goto fail;
664     }
665 
666     /* New interval for cache cleanup timer */
667     r->cache_clean_interval =
668         qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
669                             s->cache_clean_interval);
670     if (r->cache_clean_interval > UINT_MAX) {
671         error_setg(errp, "Cache clean interval too big");
672         ret = -EINVAL;
673         goto fail;
674     }
675 
676     /* lazy-refcounts; flush if going from enabled to disabled */
677     r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
678         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
679     if (r->use_lazy_refcounts && s->qcow_version < 3) {
680         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
681                    "qemu 1.1 compatibility level");
682         ret = -EINVAL;
683         goto fail;
684     }
685 
686     if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
687         ret = qcow2_mark_clean(bs);
688         if (ret < 0) {
689             error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
690             goto fail;
691         }
692     }
693 
694     /* Overlap check options */
695     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
696     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
697     if (opt_overlap_check_template && opt_overlap_check &&
698         strcmp(opt_overlap_check_template, opt_overlap_check))
699     {
700         error_setg(errp, "Conflicting values for qcow2 options '"
701                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
702                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
703         ret = -EINVAL;
704         goto fail;
705     }
706     if (!opt_overlap_check) {
707         opt_overlap_check = opt_overlap_check_template ?: "cached";
708     }
709 
710     if (!strcmp(opt_overlap_check, "none")) {
711         overlap_check_template = 0;
712     } else if (!strcmp(opt_overlap_check, "constant")) {
713         overlap_check_template = QCOW2_OL_CONSTANT;
714     } else if (!strcmp(opt_overlap_check, "cached")) {
715         overlap_check_template = QCOW2_OL_CACHED;
716     } else if (!strcmp(opt_overlap_check, "all")) {
717         overlap_check_template = QCOW2_OL_ALL;
718     } else {
719         error_setg(errp, "Unsupported value '%s' for qcow2 option "
720                    "'overlap-check'. Allowed are any of the following: "
721                    "none, constant, cached, all", opt_overlap_check);
722         ret = -EINVAL;
723         goto fail;
724     }
725 
726     r->overlap_check = 0;
727     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
728         /* overlap-check defines a template bitmask, but every flag may be
729          * overwritten through the associated boolean option */
730         r->overlap_check |=
731             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
732                               overlap_check_template & (1 << i)) << i;
733     }
734 
735     r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
736     r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
737     r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
738         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
739                           flags & BDRV_O_UNMAP);
740     r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
741         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
742     r->discard_passthrough[QCOW2_DISCARD_OTHER] =
743         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
744 
745     ret = 0;
746 fail:
747     qemu_opts_del(opts);
748     opts = NULL;
749     return ret;
750 }
751 
752 static void qcow2_update_options_commit(BlockDriverState *bs,
753                                         Qcow2ReopenState *r)
754 {
755     BDRVQcow2State *s = bs->opaque;
756     int i;
757 
758     if (s->l2_table_cache) {
759         qcow2_cache_destroy(bs, s->l2_table_cache);
760     }
761     if (s->refcount_block_cache) {
762         qcow2_cache_destroy(bs, s->refcount_block_cache);
763     }
764     s->l2_table_cache = r->l2_table_cache;
765     s->refcount_block_cache = r->refcount_block_cache;
766 
767     s->overlap_check = r->overlap_check;
768     s->use_lazy_refcounts = r->use_lazy_refcounts;
769 
770     for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
771         s->discard_passthrough[i] = r->discard_passthrough[i];
772     }
773 
774     if (s->cache_clean_interval != r->cache_clean_interval) {
775         cache_clean_timer_del(bs);
776         s->cache_clean_interval = r->cache_clean_interval;
777         cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
778     }
779 }
780 
781 static void qcow2_update_options_abort(BlockDriverState *bs,
782                                        Qcow2ReopenState *r)
783 {
784     if (r->l2_table_cache) {
785         qcow2_cache_destroy(bs, r->l2_table_cache);
786     }
787     if (r->refcount_block_cache) {
788         qcow2_cache_destroy(bs, r->refcount_block_cache);
789     }
790 }
791 
792 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
793                                 int flags, Error **errp)
794 {
795     Qcow2ReopenState r = {};
796     int ret;
797 
798     ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
799     if (ret >= 0) {
800         qcow2_update_options_commit(bs, &r);
801     } else {
802         qcow2_update_options_abort(bs, &r);
803     }
804 
805     return ret;
806 }
807 
808 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
809                       Error **errp)
810 {
811     BDRVQcow2State *s = bs->opaque;
812     unsigned int len, i;
813     int ret = 0;
814     QCowHeader header;
815     Error *local_err = NULL;
816     uint64_t ext_end;
817     uint64_t l1_vm_state_index;
818 
819     ret = bdrv_pread(bs->file->bs, 0, &header, sizeof(header));
820     if (ret < 0) {
821         error_setg_errno(errp, -ret, "Could not read qcow2 header");
822         goto fail;
823     }
824     be32_to_cpus(&header.magic);
825     be32_to_cpus(&header.version);
826     be64_to_cpus(&header.backing_file_offset);
827     be32_to_cpus(&header.backing_file_size);
828     be64_to_cpus(&header.size);
829     be32_to_cpus(&header.cluster_bits);
830     be32_to_cpus(&header.crypt_method);
831     be64_to_cpus(&header.l1_table_offset);
832     be32_to_cpus(&header.l1_size);
833     be64_to_cpus(&header.refcount_table_offset);
834     be32_to_cpus(&header.refcount_table_clusters);
835     be64_to_cpus(&header.snapshots_offset);
836     be32_to_cpus(&header.nb_snapshots);
837 
838     if (header.magic != QCOW_MAGIC) {
839         error_setg(errp, "Image is not in qcow2 format");
840         ret = -EINVAL;
841         goto fail;
842     }
843     if (header.version < 2 || header.version > 3) {
844         error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
845         ret = -ENOTSUP;
846         goto fail;
847     }
848 
849     s->qcow_version = header.version;
850 
851     /* Initialise cluster size */
852     if (header.cluster_bits < MIN_CLUSTER_BITS ||
853         header.cluster_bits > MAX_CLUSTER_BITS) {
854         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
855                    header.cluster_bits);
856         ret = -EINVAL;
857         goto fail;
858     }
859 
860     s->cluster_bits = header.cluster_bits;
861     s->cluster_size = 1 << s->cluster_bits;
862     s->cluster_sectors = 1 << (s->cluster_bits - 9);
863 
864     /* Initialise version 3 header fields */
865     if (header.version == 2) {
866         header.incompatible_features    = 0;
867         header.compatible_features      = 0;
868         header.autoclear_features       = 0;
869         header.refcount_order           = 4;
870         header.header_length            = 72;
871     } else {
872         be64_to_cpus(&header.incompatible_features);
873         be64_to_cpus(&header.compatible_features);
874         be64_to_cpus(&header.autoclear_features);
875         be32_to_cpus(&header.refcount_order);
876         be32_to_cpus(&header.header_length);
877 
878         if (header.header_length < 104) {
879             error_setg(errp, "qcow2 header too short");
880             ret = -EINVAL;
881             goto fail;
882         }
883     }
884 
885     if (header.header_length > s->cluster_size) {
886         error_setg(errp, "qcow2 header exceeds cluster size");
887         ret = -EINVAL;
888         goto fail;
889     }
890 
891     if (header.header_length > sizeof(header)) {
892         s->unknown_header_fields_size = header.header_length - sizeof(header);
893         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
894         ret = bdrv_pread(bs->file->bs, sizeof(header), s->unknown_header_fields,
895                          s->unknown_header_fields_size);
896         if (ret < 0) {
897             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
898                              "fields");
899             goto fail;
900         }
901     }
902 
903     if (header.backing_file_offset > s->cluster_size) {
904         error_setg(errp, "Invalid backing file offset");
905         ret = -EINVAL;
906         goto fail;
907     }
908 
909     if (header.backing_file_offset) {
910         ext_end = header.backing_file_offset;
911     } else {
912         ext_end = 1 << header.cluster_bits;
913     }
914 
915     /* Handle feature bits */
916     s->incompatible_features    = header.incompatible_features;
917     s->compatible_features      = header.compatible_features;
918     s->autoclear_features       = header.autoclear_features;
919 
920     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
921         void *feature_table = NULL;
922         qcow2_read_extensions(bs, header.header_length, ext_end,
923                               &feature_table, NULL);
924         report_unsupported_feature(errp, feature_table,
925                                    s->incompatible_features &
926                                    ~QCOW2_INCOMPAT_MASK);
927         ret = -ENOTSUP;
928         g_free(feature_table);
929         goto fail;
930     }
931 
932     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
933         /* Corrupt images may not be written to unless they are being repaired
934          */
935         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
936             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
937                        "read/write");
938             ret = -EACCES;
939             goto fail;
940         }
941     }
942 
943     /* Check support for various header values */
944     if (header.refcount_order > 6) {
945         error_setg(errp, "Reference count entry width too large; may not "
946                    "exceed 64 bits");
947         ret = -EINVAL;
948         goto fail;
949     }
950     s->refcount_order = header.refcount_order;
951     s->refcount_bits = 1 << s->refcount_order;
952     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
953     s->refcount_max += s->refcount_max - 1;
954 
955     if (header.crypt_method > QCOW_CRYPT_AES) {
956         error_setg(errp, "Unsupported encryption method: %" PRIu32,
957                    header.crypt_method);
958         ret = -EINVAL;
959         goto fail;
960     }
961     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) {
962         error_setg(errp, "AES cipher not available");
963         ret = -EINVAL;
964         goto fail;
965     }
966     s->crypt_method_header = header.crypt_method;
967     if (s->crypt_method_header) {
968         if (bdrv_uses_whitelist() &&
969             s->crypt_method_header == QCOW_CRYPT_AES) {
970             error_report("qcow2 built-in AES encryption is deprecated");
971             error_printf("Support for it will be removed in a future release.\n"
972                          "You can use 'qemu-img convert' to switch to an\n"
973                          "unencrypted qcow2 image, or a LUKS raw image.\n");
974         }
975 
976         bs->encrypted = 1;
977     }
978 
979     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
980     s->l2_size = 1 << s->l2_bits;
981     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
982     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
983     s->refcount_block_size = 1 << s->refcount_block_bits;
984     bs->total_sectors = header.size / 512;
985     s->csize_shift = (62 - (s->cluster_bits - 8));
986     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
987     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
988 
989     s->refcount_table_offset = header.refcount_table_offset;
990     s->refcount_table_size =
991         header.refcount_table_clusters << (s->cluster_bits - 3);
992 
993     if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
994         error_setg(errp, "Reference count table too large");
995         ret = -EINVAL;
996         goto fail;
997     }
998 
999     ret = validate_table_offset(bs, s->refcount_table_offset,
1000                                 s->refcount_table_size, sizeof(uint64_t));
1001     if (ret < 0) {
1002         error_setg(errp, "Invalid reference count table offset");
1003         goto fail;
1004     }
1005 
1006     /* Snapshot table offset/length */
1007     if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
1008         error_setg(errp, "Too many snapshots");
1009         ret = -EINVAL;
1010         goto fail;
1011     }
1012 
1013     ret = validate_table_offset(bs, header.snapshots_offset,
1014                                 header.nb_snapshots,
1015                                 sizeof(QCowSnapshotHeader));
1016     if (ret < 0) {
1017         error_setg(errp, "Invalid snapshot table offset");
1018         goto fail;
1019     }
1020 
1021     /* read the level 1 table */
1022     if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
1023         error_setg(errp, "Active L1 table too large");
1024         ret = -EFBIG;
1025         goto fail;
1026     }
1027     s->l1_size = header.l1_size;
1028 
1029     l1_vm_state_index = size_to_l1(s, header.size);
1030     if (l1_vm_state_index > INT_MAX) {
1031         error_setg(errp, "Image is too big");
1032         ret = -EFBIG;
1033         goto fail;
1034     }
1035     s->l1_vm_state_index = l1_vm_state_index;
1036 
1037     /* the L1 table must contain at least enough entries to put
1038        header.size bytes */
1039     if (s->l1_size < s->l1_vm_state_index) {
1040         error_setg(errp, "L1 table is too small");
1041         ret = -EINVAL;
1042         goto fail;
1043     }
1044 
1045     ret = validate_table_offset(bs, header.l1_table_offset,
1046                                 header.l1_size, sizeof(uint64_t));
1047     if (ret < 0) {
1048         error_setg(errp, "Invalid L1 table offset");
1049         goto fail;
1050     }
1051     s->l1_table_offset = header.l1_table_offset;
1052 
1053 
1054     if (s->l1_size > 0) {
1055         s->l1_table = qemu_try_blockalign(bs->file->bs,
1056             align_offset(s->l1_size * sizeof(uint64_t), 512));
1057         if (s->l1_table == NULL) {
1058             error_setg(errp, "Could not allocate L1 table");
1059             ret = -ENOMEM;
1060             goto fail;
1061         }
1062         ret = bdrv_pread(bs->file->bs, s->l1_table_offset, s->l1_table,
1063                          s->l1_size * sizeof(uint64_t));
1064         if (ret < 0) {
1065             error_setg_errno(errp, -ret, "Could not read L1 table");
1066             goto fail;
1067         }
1068         for(i = 0;i < s->l1_size; i++) {
1069             be64_to_cpus(&s->l1_table[i]);
1070         }
1071     }
1072 
1073     /* Parse driver-specific options */
1074     ret = qcow2_update_options(bs, options, flags, errp);
1075     if (ret < 0) {
1076         goto fail;
1077     }
1078 
1079     s->cluster_cache = g_malloc(s->cluster_size);
1080     /* one more sector for decompressed data alignment */
1081     s->cluster_data = qemu_try_blockalign(bs->file->bs, QCOW_MAX_CRYPT_CLUSTERS
1082                                                     * s->cluster_size + 512);
1083     if (s->cluster_data == NULL) {
1084         error_setg(errp, "Could not allocate temporary cluster buffer");
1085         ret = -ENOMEM;
1086         goto fail;
1087     }
1088 
1089     s->cluster_cache_offset = -1;
1090     s->flags = flags;
1091 
1092     ret = qcow2_refcount_init(bs);
1093     if (ret != 0) {
1094         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1095         goto fail;
1096     }
1097 
1098     QLIST_INIT(&s->cluster_allocs);
1099     QTAILQ_INIT(&s->discards);
1100 
1101     /* read qcow2 extensions */
1102     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1103         &local_err)) {
1104         error_propagate(errp, local_err);
1105         ret = -EINVAL;
1106         goto fail;
1107     }
1108 
1109     /* read the backing file name */
1110     if (header.backing_file_offset != 0) {
1111         len = header.backing_file_size;
1112         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1113             len >= sizeof(bs->backing_file)) {
1114             error_setg(errp, "Backing file name too long");
1115             ret = -EINVAL;
1116             goto fail;
1117         }
1118         ret = bdrv_pread(bs->file->bs, header.backing_file_offset,
1119                          bs->backing_file, len);
1120         if (ret < 0) {
1121             error_setg_errno(errp, -ret, "Could not read backing file name");
1122             goto fail;
1123         }
1124         bs->backing_file[len] = '\0';
1125         s->image_backing_file = g_strdup(bs->backing_file);
1126     }
1127 
1128     /* Internal snapshots */
1129     s->snapshots_offset = header.snapshots_offset;
1130     s->nb_snapshots = header.nb_snapshots;
1131 
1132     ret = qcow2_read_snapshots(bs);
1133     if (ret < 0) {
1134         error_setg_errno(errp, -ret, "Could not read snapshots");
1135         goto fail;
1136     }
1137 
1138     /* Clear unknown autoclear feature bits */
1139     if (!bs->read_only && !(flags & BDRV_O_INACTIVE) && s->autoclear_features) {
1140         s->autoclear_features = 0;
1141         ret = qcow2_update_header(bs);
1142         if (ret < 0) {
1143             error_setg_errno(errp, -ret, "Could not update qcow2 header");
1144             goto fail;
1145         }
1146     }
1147 
1148     /* Initialise locks */
1149     qemu_co_mutex_init(&s->lock);
1150 
1151     /* Repair image if dirty */
1152     if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1153         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1154         BdrvCheckResult result = {0};
1155 
1156         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1157         if (ret < 0) {
1158             error_setg_errno(errp, -ret, "Could not repair dirty image");
1159             goto fail;
1160         }
1161     }
1162 
1163 #ifdef DEBUG_ALLOC
1164     {
1165         BdrvCheckResult result = {0};
1166         qcow2_check_refcounts(bs, &result, 0);
1167     }
1168 #endif
1169     return ret;
1170 
1171  fail:
1172     g_free(s->unknown_header_fields);
1173     cleanup_unknown_header_ext(bs);
1174     qcow2_free_snapshots(bs);
1175     qcow2_refcount_close(bs);
1176     qemu_vfree(s->l1_table);
1177     /* else pre-write overlap checks in cache_destroy may crash */
1178     s->l1_table = NULL;
1179     cache_clean_timer_del(bs);
1180     if (s->l2_table_cache) {
1181         qcow2_cache_destroy(bs, s->l2_table_cache);
1182     }
1183     if (s->refcount_block_cache) {
1184         qcow2_cache_destroy(bs, s->refcount_block_cache);
1185     }
1186     g_free(s->cluster_cache);
1187     qemu_vfree(s->cluster_data);
1188     return ret;
1189 }
1190 
1191 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1192 {
1193     BDRVQcow2State *s = bs->opaque;
1194 
1195     bs->bl.write_zeroes_alignment = s->cluster_sectors;
1196 }
1197 
1198 static int qcow2_set_key(BlockDriverState *bs, const char *key)
1199 {
1200     BDRVQcow2State *s = bs->opaque;
1201     uint8_t keybuf[16];
1202     int len, i;
1203     Error *err = NULL;
1204 
1205     memset(keybuf, 0, 16);
1206     len = strlen(key);
1207     if (len > 16)
1208         len = 16;
1209     /* XXX: we could compress the chars to 7 bits to increase
1210        entropy */
1211     for(i = 0;i < len;i++) {
1212         keybuf[i] = key[i];
1213     }
1214     assert(bs->encrypted);
1215 
1216     qcrypto_cipher_free(s->cipher);
1217     s->cipher = qcrypto_cipher_new(
1218         QCRYPTO_CIPHER_ALG_AES_128,
1219         QCRYPTO_CIPHER_MODE_CBC,
1220         keybuf, G_N_ELEMENTS(keybuf),
1221         &err);
1222 
1223     if (!s->cipher) {
1224         /* XXX would be nice if errors in this method could
1225          * be properly propagate to the caller. Would need
1226          * the bdrv_set_key() API signature to be fixed. */
1227         error_free(err);
1228         return -1;
1229     }
1230     return 0;
1231 }
1232 
1233 static int qcow2_reopen_prepare(BDRVReopenState *state,
1234                                 BlockReopenQueue *queue, Error **errp)
1235 {
1236     Qcow2ReopenState *r;
1237     int ret;
1238 
1239     r = g_new0(Qcow2ReopenState, 1);
1240     state->opaque = r;
1241 
1242     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1243                                        state->flags, errp);
1244     if (ret < 0) {
1245         goto fail;
1246     }
1247 
1248     /* We need to write out any unwritten data if we reopen read-only. */
1249     if ((state->flags & BDRV_O_RDWR) == 0) {
1250         ret = bdrv_flush(state->bs);
1251         if (ret < 0) {
1252             goto fail;
1253         }
1254 
1255         ret = qcow2_mark_clean(state->bs);
1256         if (ret < 0) {
1257             goto fail;
1258         }
1259     }
1260 
1261     return 0;
1262 
1263 fail:
1264     qcow2_update_options_abort(state->bs, r);
1265     g_free(r);
1266     return ret;
1267 }
1268 
1269 static void qcow2_reopen_commit(BDRVReopenState *state)
1270 {
1271     qcow2_update_options_commit(state->bs, state->opaque);
1272     g_free(state->opaque);
1273 }
1274 
1275 static void qcow2_reopen_abort(BDRVReopenState *state)
1276 {
1277     qcow2_update_options_abort(state->bs, state->opaque);
1278     g_free(state->opaque);
1279 }
1280 
1281 static void qcow2_join_options(QDict *options, QDict *old_options)
1282 {
1283     bool has_new_overlap_template =
1284         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1285         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1286     bool has_new_total_cache_size =
1287         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1288     bool has_all_cache_options;
1289 
1290     /* New overlap template overrides all old overlap options */
1291     if (has_new_overlap_template) {
1292         qdict_del(old_options, QCOW2_OPT_OVERLAP);
1293         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1294         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1295         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1296         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1297         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1298         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1299         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1300         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1301         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1302     }
1303 
1304     /* New total cache size overrides all old options */
1305     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1306         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1307         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1308     }
1309 
1310     qdict_join(options, old_options, false);
1311 
1312     /*
1313      * If after merging all cache size options are set, an old total size is
1314      * overwritten. Do keep all options, however, if all three are new. The
1315      * resulting error message is what we want to happen.
1316      */
1317     has_all_cache_options =
1318         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1319         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1320         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1321 
1322     if (has_all_cache_options && !has_new_total_cache_size) {
1323         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1324     }
1325 }
1326 
1327 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1328         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1329 {
1330     BDRVQcow2State *s = bs->opaque;
1331     uint64_t cluster_offset;
1332     int index_in_cluster, ret;
1333     int64_t status = 0;
1334 
1335     *pnum = nb_sectors;
1336     qemu_co_mutex_lock(&s->lock);
1337     ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
1338     qemu_co_mutex_unlock(&s->lock);
1339     if (ret < 0) {
1340         return ret;
1341     }
1342 
1343     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1344         !s->cipher) {
1345         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1346         cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1347         *file = bs->file->bs;
1348         status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1349     }
1350     if (ret == QCOW2_CLUSTER_ZERO) {
1351         status |= BDRV_BLOCK_ZERO;
1352     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1353         status |= BDRV_BLOCK_DATA;
1354     }
1355     return status;
1356 }
1357 
1358 /* handle reading after the end of the backing file */
1359 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
1360                   int64_t sector_num, int nb_sectors)
1361 {
1362     int n1;
1363     if ((sector_num + nb_sectors) <= bs->total_sectors)
1364         return nb_sectors;
1365     if (sector_num >= bs->total_sectors)
1366         n1 = 0;
1367     else
1368         n1 = bs->total_sectors - sector_num;
1369 
1370     qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1));
1371 
1372     return n1;
1373 }
1374 
1375 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
1376                           int remaining_sectors, QEMUIOVector *qiov)
1377 {
1378     BDRVQcow2State *s = bs->opaque;
1379     int index_in_cluster, n1;
1380     int ret;
1381     int cur_nr_sectors; /* number of sectors in current iteration */
1382     uint64_t cluster_offset = 0;
1383     uint64_t bytes_done = 0;
1384     QEMUIOVector hd_qiov;
1385     uint8_t *cluster_data = NULL;
1386 
1387     qemu_iovec_init(&hd_qiov, qiov->niov);
1388 
1389     qemu_co_mutex_lock(&s->lock);
1390 
1391     while (remaining_sectors != 0) {
1392 
1393         /* prepare next request */
1394         cur_nr_sectors = remaining_sectors;
1395         if (s->cipher) {
1396             cur_nr_sectors = MIN(cur_nr_sectors,
1397                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1398         }
1399 
1400         ret = qcow2_get_cluster_offset(bs, sector_num << 9,
1401             &cur_nr_sectors, &cluster_offset);
1402         if (ret < 0) {
1403             goto fail;
1404         }
1405 
1406         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1407 
1408         qemu_iovec_reset(&hd_qiov);
1409         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1410             cur_nr_sectors * 512);
1411 
1412         switch (ret) {
1413         case QCOW2_CLUSTER_UNALLOCATED:
1414 
1415             if (bs->backing) {
1416                 /* read from the base image */
1417                 n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov,
1418                     sector_num, cur_nr_sectors);
1419                 if (n1 > 0) {
1420                     QEMUIOVector local_qiov;
1421 
1422                     qemu_iovec_init(&local_qiov, hd_qiov.niov);
1423                     qemu_iovec_concat(&local_qiov, &hd_qiov, 0,
1424                                       n1 * BDRV_SECTOR_SIZE);
1425 
1426                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1427                     qemu_co_mutex_unlock(&s->lock);
1428                     ret = bdrv_co_readv(bs->backing->bs, sector_num,
1429                                         n1, &local_qiov);
1430                     qemu_co_mutex_lock(&s->lock);
1431 
1432                     qemu_iovec_destroy(&local_qiov);
1433 
1434                     if (ret < 0) {
1435                         goto fail;
1436                     }
1437                 }
1438             } else {
1439                 /* Note: in this case, no need to wait */
1440                 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1441             }
1442             break;
1443 
1444         case QCOW2_CLUSTER_ZERO:
1445             qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
1446             break;
1447 
1448         case QCOW2_CLUSTER_COMPRESSED:
1449             /* add AIO support for compressed blocks ? */
1450             ret = qcow2_decompress_cluster(bs, cluster_offset);
1451             if (ret < 0) {
1452                 goto fail;
1453             }
1454 
1455             qemu_iovec_from_buf(&hd_qiov, 0,
1456                 s->cluster_cache + index_in_cluster * 512,
1457                 512 * cur_nr_sectors);
1458             break;
1459 
1460         case QCOW2_CLUSTER_NORMAL:
1461             if ((cluster_offset & 511) != 0) {
1462                 ret = -EIO;
1463                 goto fail;
1464             }
1465 
1466             if (bs->encrypted) {
1467                 assert(s->cipher);
1468 
1469                 /*
1470                  * For encrypted images, read everything into a temporary
1471                  * contiguous buffer on which the AES functions can work.
1472                  */
1473                 if (!cluster_data) {
1474                     cluster_data =
1475                         qemu_try_blockalign(bs->file->bs,
1476                                             QCOW_MAX_CRYPT_CLUSTERS
1477                                             * s->cluster_size);
1478                     if (cluster_data == NULL) {
1479                         ret = -ENOMEM;
1480                         goto fail;
1481                     }
1482                 }
1483 
1484                 assert(cur_nr_sectors <=
1485                     QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
1486                 qemu_iovec_reset(&hd_qiov);
1487                 qemu_iovec_add(&hd_qiov, cluster_data,
1488                     512 * cur_nr_sectors);
1489             }
1490 
1491             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1492             qemu_co_mutex_unlock(&s->lock);
1493             ret = bdrv_co_readv(bs->file->bs,
1494                                 (cluster_offset >> 9) + index_in_cluster,
1495                                 cur_nr_sectors, &hd_qiov);
1496             qemu_co_mutex_lock(&s->lock);
1497             if (ret < 0) {
1498                 goto fail;
1499             }
1500             if (bs->encrypted) {
1501                 assert(s->cipher);
1502                 Error *err = NULL;
1503                 if (qcow2_encrypt_sectors(s, sector_num,  cluster_data,
1504                                           cluster_data, cur_nr_sectors, false,
1505                                           &err) < 0) {
1506                     error_free(err);
1507                     ret = -EIO;
1508                     goto fail;
1509                 }
1510                 qemu_iovec_from_buf(qiov, bytes_done,
1511                     cluster_data, 512 * cur_nr_sectors);
1512             }
1513             break;
1514 
1515         default:
1516             g_assert_not_reached();
1517             ret = -EIO;
1518             goto fail;
1519         }
1520 
1521         remaining_sectors -= cur_nr_sectors;
1522         sector_num += cur_nr_sectors;
1523         bytes_done += cur_nr_sectors * 512;
1524     }
1525     ret = 0;
1526 
1527 fail:
1528     qemu_co_mutex_unlock(&s->lock);
1529 
1530     qemu_iovec_destroy(&hd_qiov);
1531     qemu_vfree(cluster_data);
1532 
1533     return ret;
1534 }
1535 
1536 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
1537                            int64_t sector_num,
1538                            int remaining_sectors,
1539                            QEMUIOVector *qiov)
1540 {
1541     BDRVQcow2State *s = bs->opaque;
1542     int index_in_cluster;
1543     int ret;
1544     int cur_nr_sectors; /* number of sectors in current iteration */
1545     uint64_t cluster_offset;
1546     QEMUIOVector hd_qiov;
1547     uint64_t bytes_done = 0;
1548     uint8_t *cluster_data = NULL;
1549     QCowL2Meta *l2meta = NULL;
1550 
1551     trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
1552                                  remaining_sectors);
1553 
1554     qemu_iovec_init(&hd_qiov, qiov->niov);
1555 
1556     s->cluster_cache_offset = -1; /* disable compressed cache */
1557 
1558     qemu_co_mutex_lock(&s->lock);
1559 
1560     while (remaining_sectors != 0) {
1561 
1562         l2meta = NULL;
1563 
1564         trace_qcow2_writev_start_part(qemu_coroutine_self());
1565         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1566         cur_nr_sectors = remaining_sectors;
1567         if (bs->encrypted &&
1568             cur_nr_sectors >
1569             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster) {
1570             cur_nr_sectors =
1571                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors - index_in_cluster;
1572         }
1573 
1574         ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
1575             &cur_nr_sectors, &cluster_offset, &l2meta);
1576         if (ret < 0) {
1577             goto fail;
1578         }
1579 
1580         assert((cluster_offset & 511) == 0);
1581 
1582         qemu_iovec_reset(&hd_qiov);
1583         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
1584             cur_nr_sectors * 512);
1585 
1586         if (bs->encrypted) {
1587             Error *err = NULL;
1588             assert(s->cipher);
1589             if (!cluster_data) {
1590                 cluster_data = qemu_try_blockalign(bs->file->bs,
1591                                                    QCOW_MAX_CRYPT_CLUSTERS
1592                                                    * s->cluster_size);
1593                 if (cluster_data == NULL) {
1594                     ret = -ENOMEM;
1595                     goto fail;
1596                 }
1597             }
1598 
1599             assert(hd_qiov.size <=
1600                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1601             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1602 
1603             if (qcow2_encrypt_sectors(s, sector_num, cluster_data,
1604                                       cluster_data, cur_nr_sectors,
1605                                       true, &err) < 0) {
1606                 error_free(err);
1607                 ret = -EIO;
1608                 goto fail;
1609             }
1610 
1611             qemu_iovec_reset(&hd_qiov);
1612             qemu_iovec_add(&hd_qiov, cluster_data,
1613                 cur_nr_sectors * 512);
1614         }
1615 
1616         ret = qcow2_pre_write_overlap_check(bs, 0,
1617                 cluster_offset + index_in_cluster * BDRV_SECTOR_SIZE,
1618                 cur_nr_sectors * BDRV_SECTOR_SIZE);
1619         if (ret < 0) {
1620             goto fail;
1621         }
1622 
1623         qemu_co_mutex_unlock(&s->lock);
1624         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1625         trace_qcow2_writev_data(qemu_coroutine_self(),
1626                                 (cluster_offset >> 9) + index_in_cluster);
1627         ret = bdrv_co_writev(bs->file->bs,
1628                              (cluster_offset >> 9) + index_in_cluster,
1629                              cur_nr_sectors, &hd_qiov);
1630         qemu_co_mutex_lock(&s->lock);
1631         if (ret < 0) {
1632             goto fail;
1633         }
1634 
1635         while (l2meta != NULL) {
1636             QCowL2Meta *next;
1637 
1638             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1639             if (ret < 0) {
1640                 goto fail;
1641             }
1642 
1643             /* Take the request off the list of running requests */
1644             if (l2meta->nb_clusters != 0) {
1645                 QLIST_REMOVE(l2meta, next_in_flight);
1646             }
1647 
1648             qemu_co_queue_restart_all(&l2meta->dependent_requests);
1649 
1650             next = l2meta->next;
1651             g_free(l2meta);
1652             l2meta = next;
1653         }
1654 
1655         remaining_sectors -= cur_nr_sectors;
1656         sector_num += cur_nr_sectors;
1657         bytes_done += cur_nr_sectors * 512;
1658         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
1659     }
1660     ret = 0;
1661 
1662 fail:
1663     qemu_co_mutex_unlock(&s->lock);
1664 
1665     while (l2meta != NULL) {
1666         QCowL2Meta *next;
1667 
1668         if (l2meta->nb_clusters != 0) {
1669             QLIST_REMOVE(l2meta, next_in_flight);
1670         }
1671         qemu_co_queue_restart_all(&l2meta->dependent_requests);
1672 
1673         next = l2meta->next;
1674         g_free(l2meta);
1675         l2meta = next;
1676     }
1677 
1678     qemu_iovec_destroy(&hd_qiov);
1679     qemu_vfree(cluster_data);
1680     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
1681 
1682     return ret;
1683 }
1684 
1685 static int qcow2_inactivate(BlockDriverState *bs)
1686 {
1687     BDRVQcow2State *s = bs->opaque;
1688     int ret, result = 0;
1689 
1690     ret = qcow2_cache_flush(bs, s->l2_table_cache);
1691     if (ret) {
1692         result = ret;
1693         error_report("Failed to flush the L2 table cache: %s",
1694                      strerror(-ret));
1695     }
1696 
1697     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1698     if (ret) {
1699         result = ret;
1700         error_report("Failed to flush the refcount block cache: %s",
1701                      strerror(-ret));
1702     }
1703 
1704     if (result == 0) {
1705         qcow2_mark_clean(bs);
1706     }
1707 
1708     return result;
1709 }
1710 
1711 static void qcow2_close(BlockDriverState *bs)
1712 {
1713     BDRVQcow2State *s = bs->opaque;
1714     qemu_vfree(s->l1_table);
1715     /* else pre-write overlap checks in cache_destroy may crash */
1716     s->l1_table = NULL;
1717 
1718     if (!(s->flags & BDRV_O_INACTIVE)) {
1719         qcow2_inactivate(bs);
1720     }
1721 
1722     cache_clean_timer_del(bs);
1723     qcow2_cache_destroy(bs, s->l2_table_cache);
1724     qcow2_cache_destroy(bs, s->refcount_block_cache);
1725 
1726     qcrypto_cipher_free(s->cipher);
1727     s->cipher = NULL;
1728 
1729     g_free(s->unknown_header_fields);
1730     cleanup_unknown_header_ext(bs);
1731 
1732     g_free(s->image_backing_file);
1733     g_free(s->image_backing_format);
1734 
1735     g_free(s->cluster_cache);
1736     qemu_vfree(s->cluster_data);
1737     qcow2_refcount_close(bs);
1738     qcow2_free_snapshots(bs);
1739 }
1740 
1741 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
1742 {
1743     BDRVQcow2State *s = bs->opaque;
1744     int flags = s->flags;
1745     QCryptoCipher *cipher = NULL;
1746     QDict *options;
1747     Error *local_err = NULL;
1748     int ret;
1749 
1750     /*
1751      * Backing files are read-only which makes all of their metadata immutable,
1752      * that means we don't have to worry about reopening them here.
1753      */
1754 
1755     cipher = s->cipher;
1756     s->cipher = NULL;
1757 
1758     qcow2_close(bs);
1759 
1760     memset(s, 0, sizeof(BDRVQcow2State));
1761     options = qdict_clone_shallow(bs->options);
1762 
1763     flags &= ~BDRV_O_INACTIVE;
1764     ret = qcow2_open(bs, options, flags, &local_err);
1765     QDECREF(options);
1766     if (local_err) {
1767         error_propagate(errp, local_err);
1768         error_prepend(errp, "Could not reopen qcow2 layer: ");
1769         bs->drv = NULL;
1770         return;
1771     } else if (ret < 0) {
1772         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
1773         bs->drv = NULL;
1774         return;
1775     }
1776 
1777     s->cipher = cipher;
1778 }
1779 
1780 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
1781     size_t len, size_t buflen)
1782 {
1783     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
1784     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
1785 
1786     if (buflen < ext_len) {
1787         return -ENOSPC;
1788     }
1789 
1790     *ext_backing_fmt = (QCowExtension) {
1791         .magic  = cpu_to_be32(magic),
1792         .len    = cpu_to_be32(len),
1793     };
1794     memcpy(buf + sizeof(QCowExtension), s, len);
1795 
1796     return ext_len;
1797 }
1798 
1799 /*
1800  * Updates the qcow2 header, including the variable length parts of it, i.e.
1801  * the backing file name and all extensions. qcow2 was not designed to allow
1802  * such changes, so if we run out of space (we can only use the first cluster)
1803  * this function may fail.
1804  *
1805  * Returns 0 on success, -errno in error cases.
1806  */
1807 int qcow2_update_header(BlockDriverState *bs)
1808 {
1809     BDRVQcow2State *s = bs->opaque;
1810     QCowHeader *header;
1811     char *buf;
1812     size_t buflen = s->cluster_size;
1813     int ret;
1814     uint64_t total_size;
1815     uint32_t refcount_table_clusters;
1816     size_t header_length;
1817     Qcow2UnknownHeaderExtension *uext;
1818 
1819     buf = qemu_blockalign(bs, buflen);
1820 
1821     /* Header structure */
1822     header = (QCowHeader*) buf;
1823 
1824     if (buflen < sizeof(*header)) {
1825         ret = -ENOSPC;
1826         goto fail;
1827     }
1828 
1829     header_length = sizeof(*header) + s->unknown_header_fields_size;
1830     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1831     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1832 
1833     *header = (QCowHeader) {
1834         /* Version 2 fields */
1835         .magic                  = cpu_to_be32(QCOW_MAGIC),
1836         .version                = cpu_to_be32(s->qcow_version),
1837         .backing_file_offset    = 0,
1838         .backing_file_size      = 0,
1839         .cluster_bits           = cpu_to_be32(s->cluster_bits),
1840         .size                   = cpu_to_be64(total_size),
1841         .crypt_method           = cpu_to_be32(s->crypt_method_header),
1842         .l1_size                = cpu_to_be32(s->l1_size),
1843         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
1844         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
1845         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1846         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
1847         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
1848 
1849         /* Version 3 fields */
1850         .incompatible_features  = cpu_to_be64(s->incompatible_features),
1851         .compatible_features    = cpu_to_be64(s->compatible_features),
1852         .autoclear_features     = cpu_to_be64(s->autoclear_features),
1853         .refcount_order         = cpu_to_be32(s->refcount_order),
1854         .header_length          = cpu_to_be32(header_length),
1855     };
1856 
1857     /* For older versions, write a shorter header */
1858     switch (s->qcow_version) {
1859     case 2:
1860         ret = offsetof(QCowHeader, incompatible_features);
1861         break;
1862     case 3:
1863         ret = sizeof(*header);
1864         break;
1865     default:
1866         ret = -EINVAL;
1867         goto fail;
1868     }
1869 
1870     buf += ret;
1871     buflen -= ret;
1872     memset(buf, 0, buflen);
1873 
1874     /* Preserve any unknown field in the header */
1875     if (s->unknown_header_fields_size) {
1876         if (buflen < s->unknown_header_fields_size) {
1877             ret = -ENOSPC;
1878             goto fail;
1879         }
1880 
1881         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1882         buf += s->unknown_header_fields_size;
1883         buflen -= s->unknown_header_fields_size;
1884     }
1885 
1886     /* Backing file format header extension */
1887     if (s->image_backing_format) {
1888         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1889                              s->image_backing_format,
1890                              strlen(s->image_backing_format),
1891                              buflen);
1892         if (ret < 0) {
1893             goto fail;
1894         }
1895 
1896         buf += ret;
1897         buflen -= ret;
1898     }
1899 
1900     /* Feature table */
1901     if (s->qcow_version >= 3) {
1902         Qcow2Feature features[] = {
1903             {
1904                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1905                 .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
1906                 .name = "dirty bit",
1907             },
1908             {
1909                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1910                 .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
1911                 .name = "corrupt bit",
1912             },
1913             {
1914                 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1915                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1916                 .name = "lazy refcounts",
1917             },
1918         };
1919 
1920         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1921                              features, sizeof(features), buflen);
1922         if (ret < 0) {
1923             goto fail;
1924         }
1925         buf += ret;
1926         buflen -= ret;
1927     }
1928 
1929     /* Keep unknown header extensions */
1930     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1931         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1932         if (ret < 0) {
1933             goto fail;
1934         }
1935 
1936         buf += ret;
1937         buflen -= ret;
1938     }
1939 
1940     /* End of header extensions */
1941     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1942     if (ret < 0) {
1943         goto fail;
1944     }
1945 
1946     buf += ret;
1947     buflen -= ret;
1948 
1949     /* Backing file name */
1950     if (s->image_backing_file) {
1951         size_t backing_file_len = strlen(s->image_backing_file);
1952 
1953         if (buflen < backing_file_len) {
1954             ret = -ENOSPC;
1955             goto fail;
1956         }
1957 
1958         /* Using strncpy is ok here, since buf is not NUL-terminated. */
1959         strncpy(buf, s->image_backing_file, buflen);
1960 
1961         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1962         header->backing_file_size   = cpu_to_be32(backing_file_len);
1963     }
1964 
1965     /* Write the new header */
1966     ret = bdrv_pwrite(bs->file->bs, 0, header, s->cluster_size);
1967     if (ret < 0) {
1968         goto fail;
1969     }
1970 
1971     ret = 0;
1972 fail:
1973     qemu_vfree(header);
1974     return ret;
1975 }
1976 
1977 static int qcow2_change_backing_file(BlockDriverState *bs,
1978     const char *backing_file, const char *backing_fmt)
1979 {
1980     BDRVQcow2State *s = bs->opaque;
1981 
1982     if (backing_file && strlen(backing_file) > 1023) {
1983         return -EINVAL;
1984     }
1985 
1986     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1987     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1988 
1989     g_free(s->image_backing_file);
1990     g_free(s->image_backing_format);
1991 
1992     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
1993     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
1994 
1995     return qcow2_update_header(bs);
1996 }
1997 
1998 static int preallocate(BlockDriverState *bs)
1999 {
2000     uint64_t nb_sectors;
2001     uint64_t offset;
2002     uint64_t host_offset = 0;
2003     int num;
2004     int ret;
2005     QCowL2Meta *meta;
2006 
2007     nb_sectors = bdrv_nb_sectors(bs);
2008     offset = 0;
2009 
2010     while (nb_sectors) {
2011         num = MIN(nb_sectors, INT_MAX >> BDRV_SECTOR_BITS);
2012         ret = qcow2_alloc_cluster_offset(bs, offset, &num,
2013                                          &host_offset, &meta);
2014         if (ret < 0) {
2015             return ret;
2016         }
2017 
2018         while (meta) {
2019             QCowL2Meta *next = meta->next;
2020 
2021             ret = qcow2_alloc_cluster_link_l2(bs, meta);
2022             if (ret < 0) {
2023                 qcow2_free_any_clusters(bs, meta->alloc_offset,
2024                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
2025                 return ret;
2026             }
2027 
2028             /* There are no dependent requests, but we need to remove our
2029              * request from the list of in-flight requests */
2030             QLIST_REMOVE(meta, next_in_flight);
2031 
2032             g_free(meta);
2033             meta = next;
2034         }
2035 
2036         /* TODO Preallocate data if requested */
2037 
2038         nb_sectors -= num;
2039         offset += num << BDRV_SECTOR_BITS;
2040     }
2041 
2042     /*
2043      * It is expected that the image file is large enough to actually contain
2044      * all of the allocated clusters (otherwise we get failing reads after
2045      * EOF). Extend the image to the last allocated sector.
2046      */
2047     if (host_offset != 0) {
2048         uint8_t buf[BDRV_SECTOR_SIZE];
2049         memset(buf, 0, BDRV_SECTOR_SIZE);
2050         ret = bdrv_write(bs->file->bs,
2051                          (host_offset >> BDRV_SECTOR_BITS) + num - 1,
2052                          buf, 1);
2053         if (ret < 0) {
2054             return ret;
2055         }
2056     }
2057 
2058     return 0;
2059 }
2060 
2061 static int qcow2_create2(const char *filename, int64_t total_size,
2062                          const char *backing_file, const char *backing_format,
2063                          int flags, size_t cluster_size, PreallocMode prealloc,
2064                          QemuOpts *opts, int version, int refcount_order,
2065                          Error **errp)
2066 {
2067     int cluster_bits;
2068     QDict *options;
2069 
2070     /* Calculate cluster_bits */
2071     cluster_bits = ctz32(cluster_size);
2072     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
2073         (1 << cluster_bits) != cluster_size)
2074     {
2075         error_setg(errp, "Cluster size must be a power of two between %d and "
2076                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
2077         return -EINVAL;
2078     }
2079 
2080     /*
2081      * Open the image file and write a minimal qcow2 header.
2082      *
2083      * We keep things simple and start with a zero-sized image. We also
2084      * do without refcount blocks or a L1 table for now. We'll fix the
2085      * inconsistency later.
2086      *
2087      * We do need a refcount table because growing the refcount table means
2088      * allocating two new refcount blocks - the seconds of which would be at
2089      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2090      * size for any qcow2 image.
2091      */
2092     BlockBackend *blk;
2093     QCowHeader *header;
2094     uint64_t* refcount_table;
2095     Error *local_err = NULL;
2096     int ret;
2097 
2098     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
2099         /* Note: The following calculation does not need to be exact; if it is a
2100          * bit off, either some bytes will be "leaked" (which is fine) or we
2101          * will need to increase the file size by some bytes (which is fine,
2102          * too, as long as the bulk is allocated here). Therefore, using
2103          * floating point arithmetic is fine. */
2104         int64_t meta_size = 0;
2105         uint64_t nreftablee, nrefblocke, nl1e, nl2e;
2106         int64_t aligned_total_size = align_offset(total_size, cluster_size);
2107         int refblock_bits, refblock_size;
2108         /* refcount entry size in bytes */
2109         double rces = (1 << refcount_order) / 8.;
2110 
2111         /* see qcow2_open() */
2112         refblock_bits = cluster_bits - (refcount_order - 3);
2113         refblock_size = 1 << refblock_bits;
2114 
2115         /* header: 1 cluster */
2116         meta_size += cluster_size;
2117 
2118         /* total size of L2 tables */
2119         nl2e = aligned_total_size / cluster_size;
2120         nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
2121         meta_size += nl2e * sizeof(uint64_t);
2122 
2123         /* total size of L1 tables */
2124         nl1e = nl2e * sizeof(uint64_t) / cluster_size;
2125         nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
2126         meta_size += nl1e * sizeof(uint64_t);
2127 
2128         /* total size of refcount blocks
2129          *
2130          * note: every host cluster is reference-counted, including metadata
2131          * (even refcount blocks are recursively included).
2132          * Let:
2133          *   a = total_size (this is the guest disk size)
2134          *   m = meta size not including refcount blocks and refcount tables
2135          *   c = cluster size
2136          *   y1 = number of refcount blocks entries
2137          *   y2 = meta size including everything
2138          *   rces = refcount entry size in bytes
2139          * then,
2140          *   y1 = (y2 + a)/c
2141          *   y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m
2142          * we can get y1:
2143          *   y1 = (a + m) / (c - rces - rces * sizeof(u64) / c)
2144          */
2145         nrefblocke = (aligned_total_size + meta_size + cluster_size)
2146                    / (cluster_size - rces - rces * sizeof(uint64_t)
2147                                                  / cluster_size);
2148         meta_size += DIV_ROUND_UP(nrefblocke, refblock_size) * cluster_size;
2149 
2150         /* total size of refcount tables */
2151         nreftablee = nrefblocke / refblock_size;
2152         nreftablee = align_offset(nreftablee, cluster_size / sizeof(uint64_t));
2153         meta_size += nreftablee * sizeof(uint64_t);
2154 
2155         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
2156                             aligned_total_size + meta_size, &error_abort);
2157         qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc],
2158                      &error_abort);
2159     }
2160 
2161     ret = bdrv_create_file(filename, opts, &local_err);
2162     if (ret < 0) {
2163         error_propagate(errp, local_err);
2164         return ret;
2165     }
2166 
2167     blk = blk_new_open(filename, NULL, NULL,
2168                        BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
2169     if (blk == NULL) {
2170         error_propagate(errp, local_err);
2171         return -EIO;
2172     }
2173 
2174     blk_set_allow_write_beyond_eof(blk, true);
2175 
2176     /* Write the header */
2177     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
2178     header = g_malloc0(cluster_size);
2179     *header = (QCowHeader) {
2180         .magic                      = cpu_to_be32(QCOW_MAGIC),
2181         .version                    = cpu_to_be32(version),
2182         .cluster_bits               = cpu_to_be32(cluster_bits),
2183         .size                       = cpu_to_be64(0),
2184         .l1_table_offset            = cpu_to_be64(0),
2185         .l1_size                    = cpu_to_be32(0),
2186         .refcount_table_offset      = cpu_to_be64(cluster_size),
2187         .refcount_table_clusters    = cpu_to_be32(1),
2188         .refcount_order             = cpu_to_be32(refcount_order),
2189         .header_length              = cpu_to_be32(sizeof(*header)),
2190     };
2191 
2192     if (flags & BLOCK_FLAG_ENCRYPT) {
2193         header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
2194     } else {
2195         header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
2196     }
2197 
2198     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
2199         header->compatible_features |=
2200             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
2201     }
2202 
2203     ret = blk_pwrite(blk, 0, header, cluster_size, 0);
2204     g_free(header);
2205     if (ret < 0) {
2206         error_setg_errno(errp, -ret, "Could not write qcow2 header");
2207         goto out;
2208     }
2209 
2210     /* Write a refcount table with one refcount block */
2211     refcount_table = g_malloc0(2 * cluster_size);
2212     refcount_table[0] = cpu_to_be64(2 * cluster_size);
2213     ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
2214     g_free(refcount_table);
2215 
2216     if (ret < 0) {
2217         error_setg_errno(errp, -ret, "Could not write refcount table");
2218         goto out;
2219     }
2220 
2221     blk_unref(blk);
2222     blk = NULL;
2223 
2224     /*
2225      * And now open the image and make it consistent first (i.e. increase the
2226      * refcount of the cluster that is occupied by the header and the refcount
2227      * table)
2228      */
2229     options = qdict_new();
2230     qdict_put(options, "driver", qstring_from_str("qcow2"));
2231     blk = blk_new_open(filename, NULL, options,
2232                        BDRV_O_RDWR | BDRV_O_NO_FLUSH, &local_err);
2233     if (blk == NULL) {
2234         error_propagate(errp, local_err);
2235         ret = -EIO;
2236         goto out;
2237     }
2238 
2239     ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
2240     if (ret < 0) {
2241         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
2242                          "header and refcount table");
2243         goto out;
2244 
2245     } else if (ret != 0) {
2246         error_report("Huh, first cluster in empty image is already in use?");
2247         abort();
2248     }
2249 
2250     /* Create a full header (including things like feature table) */
2251     ret = qcow2_update_header(blk_bs(blk));
2252     if (ret < 0) {
2253         error_setg_errno(errp, -ret, "Could not update qcow2 header");
2254         goto out;
2255     }
2256 
2257     /* Okay, now that we have a valid image, let's give it the right size */
2258     ret = blk_truncate(blk, total_size);
2259     if (ret < 0) {
2260         error_setg_errno(errp, -ret, "Could not resize image");
2261         goto out;
2262     }
2263 
2264     /* Want a backing file? There you go.*/
2265     if (backing_file) {
2266         ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format);
2267         if (ret < 0) {
2268             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
2269                              "with format '%s'", backing_file, backing_format);
2270             goto out;
2271         }
2272     }
2273 
2274     /* And if we're supposed to preallocate metadata, do that now */
2275     if (prealloc != PREALLOC_MODE_OFF) {
2276         BDRVQcow2State *s = blk_bs(blk)->opaque;
2277         qemu_co_mutex_lock(&s->lock);
2278         ret = preallocate(blk_bs(blk));
2279         qemu_co_mutex_unlock(&s->lock);
2280         if (ret < 0) {
2281             error_setg_errno(errp, -ret, "Could not preallocate metadata");
2282             goto out;
2283         }
2284     }
2285 
2286     blk_unref(blk);
2287     blk = NULL;
2288 
2289     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
2290     options = qdict_new();
2291     qdict_put(options, "driver", qstring_from_str("qcow2"));
2292     blk = blk_new_open(filename, NULL, options,
2293                        BDRV_O_RDWR | BDRV_O_NO_BACKING, &local_err);
2294     if (blk == NULL) {
2295         error_propagate(errp, local_err);
2296         ret = -EIO;
2297         goto out;
2298     }
2299 
2300     ret = 0;
2301 out:
2302     if (blk) {
2303         blk_unref(blk);
2304     }
2305     return ret;
2306 }
2307 
2308 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
2309 {
2310     char *backing_file = NULL;
2311     char *backing_fmt = NULL;
2312     char *buf = NULL;
2313     uint64_t size = 0;
2314     int flags = 0;
2315     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
2316     PreallocMode prealloc;
2317     int version = 3;
2318     uint64_t refcount_bits = 16;
2319     int refcount_order;
2320     Error *local_err = NULL;
2321     int ret;
2322 
2323     /* Read out options */
2324     size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2325                     BDRV_SECTOR_SIZE);
2326     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2327     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2328     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2329         flags |= BLOCK_FLAG_ENCRYPT;
2330     }
2331     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2332                                          DEFAULT_CLUSTER_SIZE);
2333     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2334     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
2335                                PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
2336                                &local_err);
2337     if (local_err) {
2338         error_propagate(errp, local_err);
2339         ret = -EINVAL;
2340         goto finish;
2341     }
2342     g_free(buf);
2343     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2344     if (!buf) {
2345         /* keep the default */
2346     } else if (!strcmp(buf, "0.10")) {
2347         version = 2;
2348     } else if (!strcmp(buf, "1.1")) {
2349         version = 3;
2350     } else {
2351         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2352         ret = -EINVAL;
2353         goto finish;
2354     }
2355 
2356     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2357         flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2358     }
2359 
2360     if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2361         error_setg(errp, "Backing file and preallocation cannot be used at "
2362                    "the same time");
2363         ret = -EINVAL;
2364         goto finish;
2365     }
2366 
2367     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2368         error_setg(errp, "Lazy refcounts only supported with compatibility "
2369                    "level 1.1 and above (use compat=1.1 or greater)");
2370         ret = -EINVAL;
2371         goto finish;
2372     }
2373 
2374     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS,
2375                                             refcount_bits);
2376     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2377         error_setg(errp, "Refcount width must be a power of two and may not "
2378                    "exceed 64 bits");
2379         ret = -EINVAL;
2380         goto finish;
2381     }
2382 
2383     if (version < 3 && refcount_bits != 16) {
2384         error_setg(errp, "Different refcount widths than 16 bits require "
2385                    "compatibility level 1.1 or above (use compat=1.1 or "
2386                    "greater)");
2387         ret = -EINVAL;
2388         goto finish;
2389     }
2390 
2391     refcount_order = ctz32(refcount_bits);
2392 
2393     ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2394                         cluster_size, prealloc, opts, version, refcount_order,
2395                         &local_err);
2396     if (local_err) {
2397         error_propagate(errp, local_err);
2398     }
2399 
2400 finish:
2401     g_free(backing_file);
2402     g_free(backing_fmt);
2403     g_free(buf);
2404     return ret;
2405 }
2406 
2407 
2408 static bool is_zero_cluster(BlockDriverState *bs, int64_t start)
2409 {
2410     BDRVQcow2State *s = bs->opaque;
2411     int nr;
2412     BlockDriverState *file;
2413     int64_t res = bdrv_get_block_status_above(bs, NULL, start,
2414                                               s->cluster_sectors, &nr, &file);
2415     return res >= 0 && ((res & BDRV_BLOCK_ZERO) || !(res & BDRV_BLOCK_DATA));
2416 }
2417 
2418 static bool is_zero_cluster_top_locked(BlockDriverState *bs, int64_t start)
2419 {
2420     BDRVQcow2State *s = bs->opaque;
2421     int nr = s->cluster_sectors;
2422     uint64_t off;
2423     int ret;
2424 
2425     ret = qcow2_get_cluster_offset(bs, start << BDRV_SECTOR_BITS, &nr, &off);
2426     return ret == QCOW2_CLUSTER_UNALLOCATED || ret == QCOW2_CLUSTER_ZERO;
2427 }
2428 
2429 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
2430     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2431 {
2432     int ret;
2433     BDRVQcow2State *s = bs->opaque;
2434 
2435     int head = sector_num % s->cluster_sectors;
2436     int tail = (sector_num + nb_sectors) % s->cluster_sectors;
2437 
2438     if (head != 0 || tail != 0) {
2439         int64_t cl_end = -1;
2440 
2441         sector_num -= head;
2442         nb_sectors += head;
2443 
2444         if (tail != 0) {
2445             nb_sectors += s->cluster_sectors - tail;
2446         }
2447 
2448         if (!is_zero_cluster(bs, sector_num)) {
2449             return -ENOTSUP;
2450         }
2451 
2452         if (nb_sectors > s->cluster_sectors) {
2453             /* Technically the request can cover 2 clusters, f.e. 4k write
2454                at s->cluster_sectors - 2k offset. One of these cluster can
2455                be zeroed, one unallocated */
2456             cl_end = sector_num + nb_sectors - s->cluster_sectors;
2457             if (!is_zero_cluster(bs, cl_end)) {
2458                 return -ENOTSUP;
2459             }
2460         }
2461 
2462         qemu_co_mutex_lock(&s->lock);
2463         /* We can have new write after previous check */
2464         if (!is_zero_cluster_top_locked(bs, sector_num) ||
2465                 (cl_end > 0 && !is_zero_cluster_top_locked(bs, cl_end))) {
2466             qemu_co_mutex_unlock(&s->lock);
2467             return -ENOTSUP;
2468         }
2469     } else {
2470         qemu_co_mutex_lock(&s->lock);
2471     }
2472 
2473     /* Whatever is left can use real zero clusters */
2474     ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS, nb_sectors);
2475     qemu_co_mutex_unlock(&s->lock);
2476 
2477     return ret;
2478 }
2479 
2480 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
2481     int64_t sector_num, int nb_sectors)
2482 {
2483     int ret;
2484     BDRVQcow2State *s = bs->opaque;
2485 
2486     qemu_co_mutex_lock(&s->lock);
2487     ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
2488         nb_sectors, QCOW2_DISCARD_REQUEST, false);
2489     qemu_co_mutex_unlock(&s->lock);
2490     return ret;
2491 }
2492 
2493 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
2494 {
2495     BDRVQcow2State *s = bs->opaque;
2496     int64_t new_l1_size;
2497     int ret;
2498 
2499     if (offset & 511) {
2500         error_report("The new size must be a multiple of 512");
2501         return -EINVAL;
2502     }
2503 
2504     /* cannot proceed if image has snapshots */
2505     if (s->nb_snapshots) {
2506         error_report("Can't resize an image which has snapshots");
2507         return -ENOTSUP;
2508     }
2509 
2510     /* shrinking is currently not supported */
2511     if (offset < bs->total_sectors * 512) {
2512         error_report("qcow2 doesn't support shrinking images yet");
2513         return -ENOTSUP;
2514     }
2515 
2516     new_l1_size = size_to_l1(s, offset);
2517     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
2518     if (ret < 0) {
2519         return ret;
2520     }
2521 
2522     /* write updated header.size */
2523     offset = cpu_to_be64(offset);
2524     ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, size),
2525                            &offset, sizeof(uint64_t));
2526     if (ret < 0) {
2527         return ret;
2528     }
2529 
2530     s->l1_vm_state_index = new_l1_size;
2531     return 0;
2532 }
2533 
2534 /* XXX: put compressed sectors first, then all the cluster aligned
2535    tables to avoid losing bytes in alignment */
2536 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
2537                                   const uint8_t *buf, int nb_sectors)
2538 {
2539     BDRVQcow2State *s = bs->opaque;
2540     z_stream strm;
2541     int ret, out_len;
2542     uint8_t *out_buf;
2543     uint64_t cluster_offset;
2544 
2545     if (nb_sectors == 0) {
2546         /* align end of file to a sector boundary to ease reading with
2547            sector based I/Os */
2548         cluster_offset = bdrv_getlength(bs->file->bs);
2549         return bdrv_truncate(bs->file->bs, cluster_offset);
2550     }
2551 
2552     if (nb_sectors != s->cluster_sectors) {
2553         ret = -EINVAL;
2554 
2555         /* Zero-pad last write if image size is not cluster aligned */
2556         if (sector_num + nb_sectors == bs->total_sectors &&
2557             nb_sectors < s->cluster_sectors) {
2558             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
2559             memset(pad_buf, 0, s->cluster_size);
2560             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
2561             ret = qcow2_write_compressed(bs, sector_num,
2562                                          pad_buf, s->cluster_sectors);
2563             qemu_vfree(pad_buf);
2564         }
2565         return ret;
2566     }
2567 
2568     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
2569 
2570     /* best compression, small window, no zlib header */
2571     memset(&strm, 0, sizeof(strm));
2572     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
2573                        Z_DEFLATED, -12,
2574                        9, Z_DEFAULT_STRATEGY);
2575     if (ret != 0) {
2576         ret = -EINVAL;
2577         goto fail;
2578     }
2579 
2580     strm.avail_in = s->cluster_size;
2581     strm.next_in = (uint8_t *)buf;
2582     strm.avail_out = s->cluster_size;
2583     strm.next_out = out_buf;
2584 
2585     ret = deflate(&strm, Z_FINISH);
2586     if (ret != Z_STREAM_END && ret != Z_OK) {
2587         deflateEnd(&strm);
2588         ret = -EINVAL;
2589         goto fail;
2590     }
2591     out_len = strm.next_out - out_buf;
2592 
2593     deflateEnd(&strm);
2594 
2595     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
2596         /* could not compress: write normal cluster */
2597         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
2598         if (ret < 0) {
2599             goto fail;
2600         }
2601     } else {
2602         cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
2603             sector_num << 9, out_len);
2604         if (!cluster_offset) {
2605             ret = -EIO;
2606             goto fail;
2607         }
2608         cluster_offset &= s->cluster_offset_mask;
2609 
2610         ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
2611         if (ret < 0) {
2612             goto fail;
2613         }
2614 
2615         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
2616         ret = bdrv_pwrite(bs->file->bs, cluster_offset, out_buf, out_len);
2617         if (ret < 0) {
2618             goto fail;
2619         }
2620     }
2621 
2622     ret = 0;
2623 fail:
2624     g_free(out_buf);
2625     return ret;
2626 }
2627 
2628 static int make_completely_empty(BlockDriverState *bs)
2629 {
2630     BDRVQcow2State *s = bs->opaque;
2631     int ret, l1_clusters;
2632     int64_t offset;
2633     uint64_t *new_reftable = NULL;
2634     uint64_t rt_entry, l1_size2;
2635     struct {
2636         uint64_t l1_offset;
2637         uint64_t reftable_offset;
2638         uint32_t reftable_clusters;
2639     } QEMU_PACKED l1_ofs_rt_ofs_cls;
2640 
2641     ret = qcow2_cache_empty(bs, s->l2_table_cache);
2642     if (ret < 0) {
2643         goto fail;
2644     }
2645 
2646     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
2647     if (ret < 0) {
2648         goto fail;
2649     }
2650 
2651     /* Refcounts will be broken utterly */
2652     ret = qcow2_mark_dirty(bs);
2653     if (ret < 0) {
2654         goto fail;
2655     }
2656 
2657     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2658 
2659     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2660     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
2661 
2662     /* After this call, neither the in-memory nor the on-disk refcount
2663      * information accurately describe the actual references */
2664 
2665     ret = bdrv_write_zeroes(bs->file->bs, s->l1_table_offset / BDRV_SECTOR_SIZE,
2666                             l1_clusters * s->cluster_sectors, 0);
2667     if (ret < 0) {
2668         goto fail_broken_refcounts;
2669     }
2670     memset(s->l1_table, 0, l1_size2);
2671 
2672     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
2673 
2674     /* Overwrite enough clusters at the beginning of the sectors to place
2675      * the refcount table, a refcount block and the L1 table in; this may
2676      * overwrite parts of the existing refcount and L1 table, which is not
2677      * an issue because the dirty flag is set, complete data loss is in fact
2678      * desired and partial data loss is consequently fine as well */
2679     ret = bdrv_write_zeroes(bs->file->bs, s->cluster_size / BDRV_SECTOR_SIZE,
2680                             (2 + l1_clusters) * s->cluster_size /
2681                             BDRV_SECTOR_SIZE, 0);
2682     /* This call (even if it failed overall) may have overwritten on-disk
2683      * refcount structures; in that case, the in-memory refcount information
2684      * will probably differ from the on-disk information which makes the BDS
2685      * unusable */
2686     if (ret < 0) {
2687         goto fail_broken_refcounts;
2688     }
2689 
2690     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2691     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
2692 
2693     /* "Create" an empty reftable (one cluster) directly after the image
2694      * header and an empty L1 table three clusters after the image header;
2695      * the cluster between those two will be used as the first refblock */
2696     cpu_to_be64w(&l1_ofs_rt_ofs_cls.l1_offset, 3 * s->cluster_size);
2697     cpu_to_be64w(&l1_ofs_rt_ofs_cls.reftable_offset, s->cluster_size);
2698     cpu_to_be32w(&l1_ofs_rt_ofs_cls.reftable_clusters, 1);
2699     ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, l1_table_offset),
2700                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
2701     if (ret < 0) {
2702         goto fail_broken_refcounts;
2703     }
2704 
2705     s->l1_table_offset = 3 * s->cluster_size;
2706 
2707     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
2708     if (!new_reftable) {
2709         ret = -ENOMEM;
2710         goto fail_broken_refcounts;
2711     }
2712 
2713     s->refcount_table_offset = s->cluster_size;
2714     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
2715 
2716     g_free(s->refcount_table);
2717     s->refcount_table = new_reftable;
2718     new_reftable = NULL;
2719 
2720     /* Now the in-memory refcount information again corresponds to the on-disk
2721      * information (reftable is empty and no refblocks (the refblock cache is
2722      * empty)); however, this means some clusters (e.g. the image header) are
2723      * referenced, but not refcounted, but the normal qcow2 code assumes that
2724      * the in-memory information is always correct */
2725 
2726     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
2727 
2728     /* Enter the first refblock into the reftable */
2729     rt_entry = cpu_to_be64(2 * s->cluster_size);
2730     ret = bdrv_pwrite_sync(bs->file->bs, s->cluster_size,
2731                            &rt_entry, sizeof(rt_entry));
2732     if (ret < 0) {
2733         goto fail_broken_refcounts;
2734     }
2735     s->refcount_table[0] = 2 * s->cluster_size;
2736 
2737     s->free_cluster_index = 0;
2738     assert(3 + l1_clusters <= s->refcount_block_size);
2739     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
2740     if (offset < 0) {
2741         ret = offset;
2742         goto fail_broken_refcounts;
2743     } else if (offset > 0) {
2744         error_report("First cluster in emptied image is in use");
2745         abort();
2746     }
2747 
2748     /* Now finally the in-memory information corresponds to the on-disk
2749      * structures and is correct */
2750     ret = qcow2_mark_clean(bs);
2751     if (ret < 0) {
2752         goto fail;
2753     }
2754 
2755     ret = bdrv_truncate(bs->file->bs, (3 + l1_clusters) * s->cluster_size);
2756     if (ret < 0) {
2757         goto fail;
2758     }
2759 
2760     return 0;
2761 
2762 fail_broken_refcounts:
2763     /* The BDS is unusable at this point. If we wanted to make it usable, we
2764      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
2765      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
2766      * again. However, because the functions which could have caused this error
2767      * path to be taken are used by those functions as well, it's very likely
2768      * that that sequence will fail as well. Therefore, just eject the BDS. */
2769     bs->drv = NULL;
2770 
2771 fail:
2772     g_free(new_reftable);
2773     return ret;
2774 }
2775 
2776 static int qcow2_make_empty(BlockDriverState *bs)
2777 {
2778     BDRVQcow2State *s = bs->opaque;
2779     uint64_t start_sector;
2780     int sector_step = INT_MAX / BDRV_SECTOR_SIZE;
2781     int l1_clusters, ret = 0;
2782 
2783     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2784 
2785     if (s->qcow_version >= 3 && !s->snapshots &&
2786         3 + l1_clusters <= s->refcount_block_size) {
2787         /* The following function only works for qcow2 v3 images (it requires
2788          * the dirty flag) and only as long as there are no snapshots (because
2789          * it completely empties the image). Furthermore, the L1 table and three
2790          * additional clusters (image header, refcount table, one refcount
2791          * block) have to fit inside one refcount block. */
2792         return make_completely_empty(bs);
2793     }
2794 
2795     /* This fallback code simply discards every active cluster; this is slow,
2796      * but works in all cases */
2797     for (start_sector = 0; start_sector < bs->total_sectors;
2798          start_sector += sector_step)
2799     {
2800         /* As this function is generally used after committing an external
2801          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
2802          * default action for this kind of discard is to pass the discard,
2803          * which will ideally result in an actually smaller image file, as
2804          * is probably desired. */
2805         ret = qcow2_discard_clusters(bs, start_sector * BDRV_SECTOR_SIZE,
2806                                      MIN(sector_step,
2807                                          bs->total_sectors - start_sector),
2808                                      QCOW2_DISCARD_SNAPSHOT, true);
2809         if (ret < 0) {
2810             break;
2811         }
2812     }
2813 
2814     return ret;
2815 }
2816 
2817 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
2818 {
2819     BDRVQcow2State *s = bs->opaque;
2820     int ret;
2821 
2822     qemu_co_mutex_lock(&s->lock);
2823     ret = qcow2_cache_flush(bs, s->l2_table_cache);
2824     if (ret < 0) {
2825         qemu_co_mutex_unlock(&s->lock);
2826         return ret;
2827     }
2828 
2829     if (qcow2_need_accurate_refcounts(s)) {
2830         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2831         if (ret < 0) {
2832             qemu_co_mutex_unlock(&s->lock);
2833             return ret;
2834         }
2835     }
2836     qemu_co_mutex_unlock(&s->lock);
2837 
2838     return 0;
2839 }
2840 
2841 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2842 {
2843     BDRVQcow2State *s = bs->opaque;
2844     bdi->unallocated_blocks_are_zero = true;
2845     bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
2846     bdi->cluster_size = s->cluster_size;
2847     bdi->vm_state_offset = qcow2_vm_state_offset(s);
2848     return 0;
2849 }
2850 
2851 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
2852 {
2853     BDRVQcow2State *s = bs->opaque;
2854     ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
2855 
2856     *spec_info = (ImageInfoSpecific){
2857         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
2858         .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1),
2859     };
2860     if (s->qcow_version == 2) {
2861         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
2862             .compat             = g_strdup("0.10"),
2863             .refcount_bits      = s->refcount_bits,
2864         };
2865     } else if (s->qcow_version == 3) {
2866         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
2867             .compat             = g_strdup("1.1"),
2868             .lazy_refcounts     = s->compatible_features &
2869                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
2870             .has_lazy_refcounts = true,
2871             .corrupt            = s->incompatible_features &
2872                                   QCOW2_INCOMPAT_CORRUPT,
2873             .has_corrupt        = true,
2874             .refcount_bits      = s->refcount_bits,
2875         };
2876     } else {
2877         /* if this assertion fails, this probably means a new version was
2878          * added without having it covered here */
2879         assert(false);
2880     }
2881 
2882     return spec_info;
2883 }
2884 
2885 #if 0
2886 static void dump_refcounts(BlockDriverState *bs)
2887 {
2888     BDRVQcow2State *s = bs->opaque;
2889     int64_t nb_clusters, k, k1, size;
2890     int refcount;
2891 
2892     size = bdrv_getlength(bs->file->bs);
2893     nb_clusters = size_to_clusters(s, size);
2894     for(k = 0; k < nb_clusters;) {
2895         k1 = k;
2896         refcount = get_refcount(bs, k);
2897         k++;
2898         while (k < nb_clusters && get_refcount(bs, k) == refcount)
2899             k++;
2900         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
2901                k - k1);
2902     }
2903 }
2904 #endif
2905 
2906 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2907                               int64_t pos)
2908 {
2909     BDRVQcow2State *s = bs->opaque;
2910     int64_t total_sectors = bs->total_sectors;
2911     bool zero_beyond_eof = bs->zero_beyond_eof;
2912     int ret;
2913 
2914     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
2915     bs->zero_beyond_eof = false;
2916     ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov);
2917     bs->zero_beyond_eof = zero_beyond_eof;
2918 
2919     /* bdrv_co_do_writev will have increased the total_sectors value to include
2920      * the VM state - the VM state is however not an actual part of the block
2921      * device, therefore, we need to restore the old value. */
2922     bs->total_sectors = total_sectors;
2923 
2924     return ret;
2925 }
2926 
2927 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2928                               int64_t pos, int size)
2929 {
2930     BDRVQcow2State *s = bs->opaque;
2931     bool zero_beyond_eof = bs->zero_beyond_eof;
2932     int ret;
2933 
2934     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
2935     bs->zero_beyond_eof = false;
2936     ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
2937     bs->zero_beyond_eof = zero_beyond_eof;
2938 
2939     return ret;
2940 }
2941 
2942 /*
2943  * Downgrades an image's version. To achieve this, any incompatible features
2944  * have to be removed.
2945  */
2946 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
2947                            BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
2948 {
2949     BDRVQcow2State *s = bs->opaque;
2950     int current_version = s->qcow_version;
2951     int ret;
2952 
2953     if (target_version == current_version) {
2954         return 0;
2955     } else if (target_version > current_version) {
2956         return -EINVAL;
2957     } else if (target_version != 2) {
2958         return -EINVAL;
2959     }
2960 
2961     if (s->refcount_order != 4) {
2962         error_report("compat=0.10 requires refcount_bits=16");
2963         return -ENOTSUP;
2964     }
2965 
2966     /* clear incompatible features */
2967     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
2968         ret = qcow2_mark_clean(bs);
2969         if (ret < 0) {
2970             return ret;
2971         }
2972     }
2973 
2974     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
2975      * the first place; if that happens nonetheless, returning -ENOTSUP is the
2976      * best thing to do anyway */
2977 
2978     if (s->incompatible_features) {
2979         return -ENOTSUP;
2980     }
2981 
2982     /* since we can ignore compatible features, we can set them to 0 as well */
2983     s->compatible_features = 0;
2984     /* if lazy refcounts have been used, they have already been fixed through
2985      * clearing the dirty flag */
2986 
2987     /* clearing autoclear features is trivial */
2988     s->autoclear_features = 0;
2989 
2990     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
2991     if (ret < 0) {
2992         return ret;
2993     }
2994 
2995     s->qcow_version = target_version;
2996     ret = qcow2_update_header(bs);
2997     if (ret < 0) {
2998         s->qcow_version = current_version;
2999         return ret;
3000     }
3001     return 0;
3002 }
3003 
3004 typedef enum Qcow2AmendOperation {
3005     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
3006      * statically initialized to so that the helper CB can discern the first
3007      * invocation from an operation change */
3008     QCOW2_NO_OPERATION = 0,
3009 
3010     QCOW2_CHANGING_REFCOUNT_ORDER,
3011     QCOW2_DOWNGRADING,
3012 } Qcow2AmendOperation;
3013 
3014 typedef struct Qcow2AmendHelperCBInfo {
3015     /* The code coordinating the amend operations should only modify
3016      * these four fields; the rest will be managed by the CB */
3017     BlockDriverAmendStatusCB *original_status_cb;
3018     void *original_cb_opaque;
3019 
3020     Qcow2AmendOperation current_operation;
3021 
3022     /* Total number of operations to perform (only set once) */
3023     int total_operations;
3024 
3025     /* The following fields are managed by the CB */
3026 
3027     /* Number of operations completed */
3028     int operations_completed;
3029 
3030     /* Cumulative offset of all completed operations */
3031     int64_t offset_completed;
3032 
3033     Qcow2AmendOperation last_operation;
3034     int64_t last_work_size;
3035 } Qcow2AmendHelperCBInfo;
3036 
3037 static void qcow2_amend_helper_cb(BlockDriverState *bs,
3038                                   int64_t operation_offset,
3039                                   int64_t operation_work_size, void *opaque)
3040 {
3041     Qcow2AmendHelperCBInfo *info = opaque;
3042     int64_t current_work_size;
3043     int64_t projected_work_size;
3044 
3045     if (info->current_operation != info->last_operation) {
3046         if (info->last_operation != QCOW2_NO_OPERATION) {
3047             info->offset_completed += info->last_work_size;
3048             info->operations_completed++;
3049         }
3050 
3051         info->last_operation = info->current_operation;
3052     }
3053 
3054     assert(info->total_operations > 0);
3055     assert(info->operations_completed < info->total_operations);
3056 
3057     info->last_work_size = operation_work_size;
3058 
3059     current_work_size = info->offset_completed + operation_work_size;
3060 
3061     /* current_work_size is the total work size for (operations_completed + 1)
3062      * operations (which includes this one), so multiply it by the number of
3063      * operations not covered and divide it by the number of operations
3064      * covered to get a projection for the operations not covered */
3065     projected_work_size = current_work_size * (info->total_operations -
3066                                                info->operations_completed - 1)
3067                                             / (info->operations_completed + 1);
3068 
3069     info->original_status_cb(bs, info->offset_completed + operation_offset,
3070                              current_work_size + projected_work_size,
3071                              info->original_cb_opaque);
3072 }
3073 
3074 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
3075                                BlockDriverAmendStatusCB *status_cb,
3076                                void *cb_opaque)
3077 {
3078     BDRVQcow2State *s = bs->opaque;
3079     int old_version = s->qcow_version, new_version = old_version;
3080     uint64_t new_size = 0;
3081     const char *backing_file = NULL, *backing_format = NULL;
3082     bool lazy_refcounts = s->use_lazy_refcounts;
3083     const char *compat = NULL;
3084     uint64_t cluster_size = s->cluster_size;
3085     bool encrypt;
3086     int refcount_bits = s->refcount_bits;
3087     int ret;
3088     QemuOptDesc *desc = opts->list->desc;
3089     Qcow2AmendHelperCBInfo helper_cb_info;
3090 
3091     while (desc && desc->name) {
3092         if (!qemu_opt_find(opts, desc->name)) {
3093             /* only change explicitly defined options */
3094             desc++;
3095             continue;
3096         }
3097 
3098         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
3099             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
3100             if (!compat) {
3101                 /* preserve default */
3102             } else if (!strcmp(compat, "0.10")) {
3103                 new_version = 2;
3104             } else if (!strcmp(compat, "1.1")) {
3105                 new_version = 3;
3106             } else {
3107                 error_report("Unknown compatibility level %s", compat);
3108                 return -EINVAL;
3109             }
3110         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
3111             error_report("Cannot change preallocation mode");
3112             return -ENOTSUP;
3113         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
3114             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
3115         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
3116             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
3117         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
3118             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
3119         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
3120             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
3121                                         !!s->cipher);
3122 
3123             if (encrypt != !!s->cipher) {
3124                 error_report("Changing the encryption flag is not supported");
3125                 return -ENOTSUP;
3126             }
3127         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
3128             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
3129                                              cluster_size);
3130             if (cluster_size != s->cluster_size) {
3131                 error_report("Changing the cluster size is not supported");
3132                 return -ENOTSUP;
3133             }
3134         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
3135             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
3136                                                lazy_refcounts);
3137         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
3138             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
3139                                                 refcount_bits);
3140 
3141             if (refcount_bits <= 0 || refcount_bits > 64 ||
3142                 !is_power_of_2(refcount_bits))
3143             {
3144                 error_report("Refcount width must be a power of two and may "
3145                              "not exceed 64 bits");
3146                 return -EINVAL;
3147             }
3148         } else {
3149             /* if this point is reached, this probably means a new option was
3150              * added without having it covered here */
3151             abort();
3152         }
3153 
3154         desc++;
3155     }
3156 
3157     helper_cb_info = (Qcow2AmendHelperCBInfo){
3158         .original_status_cb = status_cb,
3159         .original_cb_opaque = cb_opaque,
3160         .total_operations = (new_version < old_version)
3161                           + (s->refcount_bits != refcount_bits)
3162     };
3163 
3164     /* Upgrade first (some features may require compat=1.1) */
3165     if (new_version > old_version) {
3166         s->qcow_version = new_version;
3167         ret = qcow2_update_header(bs);
3168         if (ret < 0) {
3169             s->qcow_version = old_version;
3170             return ret;
3171         }
3172     }
3173 
3174     if (s->refcount_bits != refcount_bits) {
3175         int refcount_order = ctz32(refcount_bits);
3176         Error *local_error = NULL;
3177 
3178         if (new_version < 3 && refcount_bits != 16) {
3179             error_report("Different refcount widths than 16 bits require "
3180                          "compatibility level 1.1 or above (use compat=1.1 or "
3181                          "greater)");
3182             return -EINVAL;
3183         }
3184 
3185         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
3186         ret = qcow2_change_refcount_order(bs, refcount_order,
3187                                           &qcow2_amend_helper_cb,
3188                                           &helper_cb_info, &local_error);
3189         if (ret < 0) {
3190             error_report_err(local_error);
3191             return ret;
3192         }
3193     }
3194 
3195     if (backing_file || backing_format) {
3196         ret = qcow2_change_backing_file(bs,
3197                     backing_file ?: s->image_backing_file,
3198                     backing_format ?: s->image_backing_format);
3199         if (ret < 0) {
3200             return ret;
3201         }
3202     }
3203 
3204     if (s->use_lazy_refcounts != lazy_refcounts) {
3205         if (lazy_refcounts) {
3206             if (new_version < 3) {
3207                 error_report("Lazy refcounts only supported with compatibility "
3208                              "level 1.1 and above (use compat=1.1 or greater)");
3209                 return -EINVAL;
3210             }
3211             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3212             ret = qcow2_update_header(bs);
3213             if (ret < 0) {
3214                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3215                 return ret;
3216             }
3217             s->use_lazy_refcounts = true;
3218         } else {
3219             /* make image clean first */
3220             ret = qcow2_mark_clean(bs);
3221             if (ret < 0) {
3222                 return ret;
3223             }
3224             /* now disallow lazy refcounts */
3225             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3226             ret = qcow2_update_header(bs);
3227             if (ret < 0) {
3228                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3229                 return ret;
3230             }
3231             s->use_lazy_refcounts = false;
3232         }
3233     }
3234 
3235     if (new_size) {
3236         ret = bdrv_truncate(bs, new_size);
3237         if (ret < 0) {
3238             return ret;
3239         }
3240     }
3241 
3242     /* Downgrade last (so unsupported features can be removed before) */
3243     if (new_version < old_version) {
3244         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
3245         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
3246                               &helper_cb_info);
3247         if (ret < 0) {
3248             return ret;
3249         }
3250     }
3251 
3252     return 0;
3253 }
3254 
3255 /*
3256  * If offset or size are negative, respectively, they will not be included in
3257  * the BLOCK_IMAGE_CORRUPTED event emitted.
3258  * fatal will be ignored for read-only BDS; corruptions found there will always
3259  * be considered non-fatal.
3260  */
3261 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
3262                              int64_t size, const char *message_format, ...)
3263 {
3264     BDRVQcow2State *s = bs->opaque;
3265     const char *node_name;
3266     char *message;
3267     va_list ap;
3268 
3269     fatal = fatal && !bs->read_only;
3270 
3271     if (s->signaled_corruption &&
3272         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
3273     {
3274         return;
3275     }
3276 
3277     va_start(ap, message_format);
3278     message = g_strdup_vprintf(message_format, ap);
3279     va_end(ap);
3280 
3281     if (fatal) {
3282         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
3283                 "corruption events will be suppressed\n", message);
3284     } else {
3285         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
3286                 "corruption events will be suppressed\n", message);
3287     }
3288 
3289     node_name = bdrv_get_node_name(bs);
3290     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
3291                                           *node_name != '\0', node_name,
3292                                           message, offset >= 0, offset,
3293                                           size >= 0, size,
3294                                           fatal, &error_abort);
3295     g_free(message);
3296 
3297     if (fatal) {
3298         qcow2_mark_corrupt(bs);
3299         bs->drv = NULL; /* make BDS unusable */
3300     }
3301 
3302     s->signaled_corruption = true;
3303 }
3304 
3305 static QemuOptsList qcow2_create_opts = {
3306     .name = "qcow2-create-opts",
3307     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
3308     .desc = {
3309         {
3310             .name = BLOCK_OPT_SIZE,
3311             .type = QEMU_OPT_SIZE,
3312             .help = "Virtual disk size"
3313         },
3314         {
3315             .name = BLOCK_OPT_COMPAT_LEVEL,
3316             .type = QEMU_OPT_STRING,
3317             .help = "Compatibility level (0.10 or 1.1)"
3318         },
3319         {
3320             .name = BLOCK_OPT_BACKING_FILE,
3321             .type = QEMU_OPT_STRING,
3322             .help = "File name of a base image"
3323         },
3324         {
3325             .name = BLOCK_OPT_BACKING_FMT,
3326             .type = QEMU_OPT_STRING,
3327             .help = "Image format of the base image"
3328         },
3329         {
3330             .name = BLOCK_OPT_ENCRYPT,
3331             .type = QEMU_OPT_BOOL,
3332             .help = "Encrypt the image",
3333             .def_value_str = "off"
3334         },
3335         {
3336             .name = BLOCK_OPT_CLUSTER_SIZE,
3337             .type = QEMU_OPT_SIZE,
3338             .help = "qcow2 cluster size",
3339             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
3340         },
3341         {
3342             .name = BLOCK_OPT_PREALLOC,
3343             .type = QEMU_OPT_STRING,
3344             .help = "Preallocation mode (allowed values: off, metadata, "
3345                     "falloc, full)"
3346         },
3347         {
3348             .name = BLOCK_OPT_LAZY_REFCOUNTS,
3349             .type = QEMU_OPT_BOOL,
3350             .help = "Postpone refcount updates",
3351             .def_value_str = "off"
3352         },
3353         {
3354             .name = BLOCK_OPT_REFCOUNT_BITS,
3355             .type = QEMU_OPT_NUMBER,
3356             .help = "Width of a reference count entry in bits",
3357             .def_value_str = "16"
3358         },
3359         { /* end of list */ }
3360     }
3361 };
3362 
3363 BlockDriver bdrv_qcow2 = {
3364     .format_name        = "qcow2",
3365     .instance_size      = sizeof(BDRVQcow2State),
3366     .bdrv_probe         = qcow2_probe,
3367     .bdrv_open          = qcow2_open,
3368     .bdrv_close         = qcow2_close,
3369     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
3370     .bdrv_reopen_commit   = qcow2_reopen_commit,
3371     .bdrv_reopen_abort    = qcow2_reopen_abort,
3372     .bdrv_join_options    = qcow2_join_options,
3373     .bdrv_create        = qcow2_create,
3374     .bdrv_has_zero_init = bdrv_has_zero_init_1,
3375     .bdrv_co_get_block_status = qcow2_co_get_block_status,
3376     .bdrv_set_key       = qcow2_set_key,
3377 
3378     .bdrv_co_readv          = qcow2_co_readv,
3379     .bdrv_co_writev         = qcow2_co_writev,
3380     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
3381 
3382     .bdrv_co_write_zeroes   = qcow2_co_write_zeroes,
3383     .bdrv_co_discard        = qcow2_co_discard,
3384     .bdrv_truncate          = qcow2_truncate,
3385     .bdrv_write_compressed  = qcow2_write_compressed,
3386     .bdrv_make_empty        = qcow2_make_empty,
3387 
3388     .bdrv_snapshot_create   = qcow2_snapshot_create,
3389     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
3390     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
3391     .bdrv_snapshot_list     = qcow2_snapshot_list,
3392     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
3393     .bdrv_get_info          = qcow2_get_info,
3394     .bdrv_get_specific_info = qcow2_get_specific_info,
3395 
3396     .bdrv_save_vmstate    = qcow2_save_vmstate,
3397     .bdrv_load_vmstate    = qcow2_load_vmstate,
3398 
3399     .supports_backing           = true,
3400     .bdrv_change_backing_file   = qcow2_change_backing_file,
3401 
3402     .bdrv_refresh_limits        = qcow2_refresh_limits,
3403     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
3404     .bdrv_inactivate            = qcow2_inactivate,
3405 
3406     .create_opts         = &qcow2_create_opts,
3407     .bdrv_check          = qcow2_check,
3408     .bdrv_amend_options  = qcow2_amend_options,
3409 
3410     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
3411     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
3412 };
3413 
3414 static void bdrv_qcow2_init(void)
3415 {
3416     bdrv_register(&bdrv_qcow2);
3417 }
3418 
3419 block_init(bdrv_qcow2_init);
3420