xref: /openbmc/qemu/block/qcow2.c (revision 795c40b8)
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 #include "qemu/bswap.h"
40 
41 /*
42   Differences with QCOW:
43 
44   - Support for multiple incremental snapshots.
45   - Memory management by reference counts.
46   - Clusters which have a reference count of one have the bit
47     QCOW_OFLAG_COPIED to optimize write performance.
48   - Size of compressed clusters is stored in sectors to reduce bit usage
49     in the cluster offsets.
50   - Support for storing additional data (such as the VM state) in the
51     snapshots.
52   - If a backing store is used, the cluster size is not constrained
53     (could be backported to QCOW).
54   - L2 tables have always a size of one cluster.
55 */
56 
57 
58 typedef struct {
59     uint32_t magic;
60     uint32_t len;
61 } QEMU_PACKED QCowExtension;
62 
63 #define  QCOW2_EXT_MAGIC_END 0
64 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
65 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
66 
67 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
68 {
69     const QCowHeader *cow_header = (const void *)buf;
70 
71     if (buf_size >= sizeof(QCowHeader) &&
72         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
73         be32_to_cpu(cow_header->version) >= 2)
74         return 100;
75     else
76         return 0;
77 }
78 
79 
80 /*
81  * read qcow2 extension and fill bs
82  * start reading from start_offset
83  * finish reading upon magic of value 0 or when end_offset reached
84  * unknown magic is skipped (future extension this version knows nothing about)
85  * return 0 upon success, non-0 otherwise
86  */
87 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
88                                  uint64_t end_offset, void **p_feature_table,
89                                  Error **errp)
90 {
91     BDRVQcow2State *s = bs->opaque;
92     QCowExtension ext;
93     uint64_t offset;
94     int ret;
95 
96 #ifdef DEBUG_EXT
97     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
98 #endif
99     offset = start_offset;
100     while (offset < end_offset) {
101 
102 #ifdef DEBUG_EXT
103         /* Sanity check */
104         if (offset > s->cluster_size)
105             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
106 
107         printf("attempting to read extended header in offset %lu\n", offset);
108 #endif
109 
110         ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
111         if (ret < 0) {
112             error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
113                              "pread fail from offset %" PRIu64, offset);
114             return 1;
115         }
116         be32_to_cpus(&ext.magic);
117         be32_to_cpus(&ext.len);
118         offset += sizeof(ext);
119 #ifdef DEBUG_EXT
120         printf("ext.magic = 0x%x\n", ext.magic);
121 #endif
122         if (offset > end_offset || ext.len > end_offset - offset) {
123             error_setg(errp, "Header extension too large");
124             return -EINVAL;
125         }
126 
127         switch (ext.magic) {
128         case QCOW2_EXT_MAGIC_END:
129             return 0;
130 
131         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
132             if (ext.len >= sizeof(bs->backing_format)) {
133                 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
134                            " too large (>=%zu)", ext.len,
135                            sizeof(bs->backing_format));
136                 return 2;
137             }
138             ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len);
139             if (ret < 0) {
140                 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
141                                  "Could not read format name");
142                 return 3;
143             }
144             bs->backing_format[ext.len] = '\0';
145             s->image_backing_format = g_strdup(bs->backing_format);
146 #ifdef DEBUG_EXT
147             printf("Qcow2: Got format extension %s\n", bs->backing_format);
148 #endif
149             break;
150 
151         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
152             if (p_feature_table != NULL) {
153                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
154                 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
155                 if (ret < 0) {
156                     error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
157                                      "Could not read table");
158                     return ret;
159                 }
160 
161                 *p_feature_table = feature_table;
162             }
163             break;
164 
165         default:
166             /* unknown magic - save it in case we need to rewrite the header */
167             {
168                 Qcow2UnknownHeaderExtension *uext;
169 
170                 uext = g_malloc0(sizeof(*uext)  + ext.len);
171                 uext->magic = ext.magic;
172                 uext->len = ext.len;
173                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
174 
175                 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
176                 if (ret < 0) {
177                     error_setg_errno(errp, -ret, "ERROR: unknown extension: "
178                                      "Could not read data");
179                     return ret;
180                 }
181             }
182             break;
183         }
184 
185         offset += ((ext.len + 7) & ~7);
186     }
187 
188     return 0;
189 }
190 
191 static void cleanup_unknown_header_ext(BlockDriverState *bs)
192 {
193     BDRVQcow2State *s = bs->opaque;
194     Qcow2UnknownHeaderExtension *uext, *next;
195 
196     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
197         QLIST_REMOVE(uext, next);
198         g_free(uext);
199     }
200 }
201 
202 static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
203                                        uint64_t mask)
204 {
205     char *features = g_strdup("");
206     char *old;
207 
208     while (table && table->name[0] != '\0') {
209         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
210             if (mask & (1ULL << table->bit)) {
211                 old = features;
212                 features = g_strdup_printf("%s%s%.46s", old, *old ? ", " : "",
213                                            table->name);
214                 g_free(old);
215                 mask &= ~(1ULL << table->bit);
216             }
217         }
218         table++;
219     }
220 
221     if (mask) {
222         old = features;
223         features = g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64,
224                                    old, *old ? ", " : "", mask);
225         g_free(old);
226     }
227 
228     error_setg(errp, "Unsupported qcow2 feature(s): %s", features);
229     g_free(features);
230 }
231 
232 /*
233  * Sets the dirty bit and flushes afterwards if necessary.
234  *
235  * The incompatible_features bit is only set if the image file header was
236  * updated successfully.  Therefore it is not required to check the return
237  * value of this function.
238  */
239 int qcow2_mark_dirty(BlockDriverState *bs)
240 {
241     BDRVQcow2State *s = bs->opaque;
242     uint64_t val;
243     int ret;
244 
245     assert(s->qcow_version >= 3);
246 
247     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
248         return 0; /* already dirty */
249     }
250 
251     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
252     ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
253                       &val, sizeof(val));
254     if (ret < 0) {
255         return ret;
256     }
257     ret = bdrv_flush(bs->file->bs);
258     if (ret < 0) {
259         return ret;
260     }
261 
262     /* Only treat image as dirty if the header was updated successfully */
263     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
264     return 0;
265 }
266 
267 /*
268  * Clears the dirty bit and flushes before if necessary.  Only call this
269  * function when there are no pending requests, it does not guard against
270  * concurrent requests dirtying the image.
271  */
272 static int qcow2_mark_clean(BlockDriverState *bs)
273 {
274     BDRVQcow2State *s = bs->opaque;
275 
276     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
277         int ret;
278 
279         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
280 
281         ret = bdrv_flush(bs);
282         if (ret < 0) {
283             return ret;
284         }
285 
286         return qcow2_update_header(bs);
287     }
288     return 0;
289 }
290 
291 /*
292  * Marks the image as corrupt.
293  */
294 int qcow2_mark_corrupt(BlockDriverState *bs)
295 {
296     BDRVQcow2State *s = bs->opaque;
297 
298     s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT;
299     return qcow2_update_header(bs);
300 }
301 
302 /*
303  * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
304  * before if necessary.
305  */
306 int qcow2_mark_consistent(BlockDriverState *bs)
307 {
308     BDRVQcow2State *s = bs->opaque;
309 
310     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
311         int ret = bdrv_flush(bs);
312         if (ret < 0) {
313             return ret;
314         }
315 
316         s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT;
317         return qcow2_update_header(bs);
318     }
319     return 0;
320 }
321 
322 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
323                        BdrvCheckMode fix)
324 {
325     int ret = qcow2_check_refcounts(bs, result, fix);
326     if (ret < 0) {
327         return ret;
328     }
329 
330     if (fix && result->check_errors == 0 && result->corruptions == 0) {
331         ret = qcow2_mark_clean(bs);
332         if (ret < 0) {
333             return ret;
334         }
335         return qcow2_mark_consistent(bs);
336     }
337     return ret;
338 }
339 
340 static int validate_table_offset(BlockDriverState *bs, uint64_t offset,
341                                  uint64_t entries, size_t entry_len)
342 {
343     BDRVQcow2State *s = bs->opaque;
344     uint64_t size;
345 
346     /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
347      * because values will be passed to qemu functions taking int64_t. */
348     if (entries > INT64_MAX / entry_len) {
349         return -EINVAL;
350     }
351 
352     size = entries * entry_len;
353 
354     if (INT64_MAX - size < offset) {
355         return -EINVAL;
356     }
357 
358     /* Tables must be cluster aligned */
359     if (offset & (s->cluster_size - 1)) {
360         return -EINVAL;
361     }
362 
363     return 0;
364 }
365 
366 static QemuOptsList qcow2_runtime_opts = {
367     .name = "qcow2",
368     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
369     .desc = {
370         {
371             .name = QCOW2_OPT_LAZY_REFCOUNTS,
372             .type = QEMU_OPT_BOOL,
373             .help = "Postpone refcount updates",
374         },
375         {
376             .name = QCOW2_OPT_DISCARD_REQUEST,
377             .type = QEMU_OPT_BOOL,
378             .help = "Pass guest discard requests to the layer below",
379         },
380         {
381             .name = QCOW2_OPT_DISCARD_SNAPSHOT,
382             .type = QEMU_OPT_BOOL,
383             .help = "Generate discard requests when snapshot related space "
384                     "is freed",
385         },
386         {
387             .name = QCOW2_OPT_DISCARD_OTHER,
388             .type = QEMU_OPT_BOOL,
389             .help = "Generate discard requests when other clusters are freed",
390         },
391         {
392             .name = QCOW2_OPT_OVERLAP,
393             .type = QEMU_OPT_STRING,
394             .help = "Selects which overlap checks to perform from a range of "
395                     "templates (none, constant, cached, all)",
396         },
397         {
398             .name = QCOW2_OPT_OVERLAP_TEMPLATE,
399             .type = QEMU_OPT_STRING,
400             .help = "Selects which overlap checks to perform from a range of "
401                     "templates (none, constant, cached, all)",
402         },
403         {
404             .name = QCOW2_OPT_OVERLAP_MAIN_HEADER,
405             .type = QEMU_OPT_BOOL,
406             .help = "Check for unintended writes into the main qcow2 header",
407         },
408         {
409             .name = QCOW2_OPT_OVERLAP_ACTIVE_L1,
410             .type = QEMU_OPT_BOOL,
411             .help = "Check for unintended writes into the active L1 table",
412         },
413         {
414             .name = QCOW2_OPT_OVERLAP_ACTIVE_L2,
415             .type = QEMU_OPT_BOOL,
416             .help = "Check for unintended writes into an active L2 table",
417         },
418         {
419             .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
420             .type = QEMU_OPT_BOOL,
421             .help = "Check for unintended writes into the refcount table",
422         },
423         {
424             .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
425             .type = QEMU_OPT_BOOL,
426             .help = "Check for unintended writes into a refcount block",
427         },
428         {
429             .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
430             .type = QEMU_OPT_BOOL,
431             .help = "Check for unintended writes into the snapshot table",
432         },
433         {
434             .name = QCOW2_OPT_OVERLAP_INACTIVE_L1,
435             .type = QEMU_OPT_BOOL,
436             .help = "Check for unintended writes into an inactive L1 table",
437         },
438         {
439             .name = QCOW2_OPT_OVERLAP_INACTIVE_L2,
440             .type = QEMU_OPT_BOOL,
441             .help = "Check for unintended writes into an inactive L2 table",
442         },
443         {
444             .name = QCOW2_OPT_CACHE_SIZE,
445             .type = QEMU_OPT_SIZE,
446             .help = "Maximum combined metadata (L2 tables and refcount blocks) "
447                     "cache size",
448         },
449         {
450             .name = QCOW2_OPT_L2_CACHE_SIZE,
451             .type = QEMU_OPT_SIZE,
452             .help = "Maximum L2 table cache size",
453         },
454         {
455             .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
456             .type = QEMU_OPT_SIZE,
457             .help = "Maximum refcount block cache size",
458         },
459         {
460             .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
461             .type = QEMU_OPT_NUMBER,
462             .help = "Clean unused cache entries after this time (in seconds)",
463         },
464         { /* end of list */ }
465     },
466 };
467 
468 static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
469     [QCOW2_OL_MAIN_HEADER_BITNR]    = QCOW2_OPT_OVERLAP_MAIN_HEADER,
470     [QCOW2_OL_ACTIVE_L1_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L1,
471     [QCOW2_OL_ACTIVE_L2_BITNR]      = QCOW2_OPT_OVERLAP_ACTIVE_L2,
472     [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE,
473     [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK,
474     [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE,
475     [QCOW2_OL_INACTIVE_L1_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L1,
476     [QCOW2_OL_INACTIVE_L2_BITNR]    = QCOW2_OPT_OVERLAP_INACTIVE_L2,
477 };
478 
479 static void cache_clean_timer_cb(void *opaque)
480 {
481     BlockDriverState *bs = opaque;
482     BDRVQcow2State *s = bs->opaque;
483     qcow2_cache_clean_unused(bs, s->l2_table_cache);
484     qcow2_cache_clean_unused(bs, s->refcount_block_cache);
485     timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
486               (int64_t) s->cache_clean_interval * 1000);
487 }
488 
489 static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
490 {
491     BDRVQcow2State *s = bs->opaque;
492     if (s->cache_clean_interval > 0) {
493         s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL,
494                                              SCALE_MS, cache_clean_timer_cb,
495                                              bs);
496         timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
497                   (int64_t) s->cache_clean_interval * 1000);
498     }
499 }
500 
501 static void cache_clean_timer_del(BlockDriverState *bs)
502 {
503     BDRVQcow2State *s = bs->opaque;
504     if (s->cache_clean_timer) {
505         timer_del(s->cache_clean_timer);
506         timer_free(s->cache_clean_timer);
507         s->cache_clean_timer = NULL;
508     }
509 }
510 
511 static void qcow2_detach_aio_context(BlockDriverState *bs)
512 {
513     cache_clean_timer_del(bs);
514 }
515 
516 static void qcow2_attach_aio_context(BlockDriverState *bs,
517                                      AioContext *new_context)
518 {
519     cache_clean_timer_init(bs, new_context);
520 }
521 
522 static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
523                              uint64_t *l2_cache_size,
524                              uint64_t *refcount_cache_size, Error **errp)
525 {
526     BDRVQcow2State *s = bs->opaque;
527     uint64_t combined_cache_size;
528     bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
529 
530     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
531     l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
532     refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
533 
534     combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
535     *l2_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, 0);
536     *refcount_cache_size = qemu_opt_get_size(opts,
537                                              QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
538 
539     if (combined_cache_size_set) {
540         if (l2_cache_size_set && refcount_cache_size_set) {
541             error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
542                        " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
543                        "the same time");
544             return;
545         } else if (*l2_cache_size > combined_cache_size) {
546             error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
547                        QCOW2_OPT_CACHE_SIZE);
548             return;
549         } else if (*refcount_cache_size > combined_cache_size) {
550             error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
551                        QCOW2_OPT_CACHE_SIZE);
552             return;
553         }
554 
555         if (l2_cache_size_set) {
556             *refcount_cache_size = combined_cache_size - *l2_cache_size;
557         } else if (refcount_cache_size_set) {
558             *l2_cache_size = combined_cache_size - *refcount_cache_size;
559         } else {
560             *refcount_cache_size = combined_cache_size
561                                  / (DEFAULT_L2_REFCOUNT_SIZE_RATIO + 1);
562             *l2_cache_size = combined_cache_size - *refcount_cache_size;
563         }
564     } else {
565         if (!l2_cache_size_set && !refcount_cache_size_set) {
566             *l2_cache_size = MAX(DEFAULT_L2_CACHE_BYTE_SIZE,
567                                  (uint64_t)DEFAULT_L2_CACHE_CLUSTERS
568                                  * s->cluster_size);
569             *refcount_cache_size = *l2_cache_size
570                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
571         } else if (!l2_cache_size_set) {
572             *l2_cache_size = *refcount_cache_size
573                            * DEFAULT_L2_REFCOUNT_SIZE_RATIO;
574         } else if (!refcount_cache_size_set) {
575             *refcount_cache_size = *l2_cache_size
576                                  / DEFAULT_L2_REFCOUNT_SIZE_RATIO;
577         }
578     }
579 }
580 
581 typedef struct Qcow2ReopenState {
582     Qcow2Cache *l2_table_cache;
583     Qcow2Cache *refcount_block_cache;
584     bool use_lazy_refcounts;
585     int overlap_check;
586     bool discard_passthrough[QCOW2_DISCARD_MAX];
587     uint64_t cache_clean_interval;
588 } Qcow2ReopenState;
589 
590 static int qcow2_update_options_prepare(BlockDriverState *bs,
591                                         Qcow2ReopenState *r,
592                                         QDict *options, int flags,
593                                         Error **errp)
594 {
595     BDRVQcow2State *s = bs->opaque;
596     QemuOpts *opts = NULL;
597     const char *opt_overlap_check, *opt_overlap_check_template;
598     int overlap_check_template = 0;
599     uint64_t l2_cache_size, refcount_cache_size;
600     int i;
601     Error *local_err = NULL;
602     int ret;
603 
604     opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
605     qemu_opts_absorb_qdict(opts, options, &local_err);
606     if (local_err) {
607         error_propagate(errp, local_err);
608         ret = -EINVAL;
609         goto fail;
610     }
611 
612     /* get L2 table/refcount block cache size from command line options */
613     read_cache_sizes(bs, opts, &l2_cache_size, &refcount_cache_size,
614                      &local_err);
615     if (local_err) {
616         error_propagate(errp, local_err);
617         ret = -EINVAL;
618         goto fail;
619     }
620 
621     l2_cache_size /= s->cluster_size;
622     if (l2_cache_size < MIN_L2_CACHE_SIZE) {
623         l2_cache_size = MIN_L2_CACHE_SIZE;
624     }
625     if (l2_cache_size > INT_MAX) {
626         error_setg(errp, "L2 cache size too big");
627         ret = -EINVAL;
628         goto fail;
629     }
630 
631     refcount_cache_size /= s->cluster_size;
632     if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) {
633         refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE;
634     }
635     if (refcount_cache_size > INT_MAX) {
636         error_setg(errp, "Refcount cache size too big");
637         ret = -EINVAL;
638         goto fail;
639     }
640 
641     /* alloc new L2 table/refcount block cache, flush old one */
642     if (s->l2_table_cache) {
643         ret = qcow2_cache_flush(bs, s->l2_table_cache);
644         if (ret) {
645             error_setg_errno(errp, -ret, "Failed to flush the L2 table cache");
646             goto fail;
647         }
648     }
649 
650     if (s->refcount_block_cache) {
651         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
652         if (ret) {
653             error_setg_errno(errp, -ret,
654                              "Failed to flush the refcount block cache");
655             goto fail;
656         }
657     }
658 
659     r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size);
660     r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size);
661     if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
662         error_setg(errp, "Could not allocate metadata caches");
663         ret = -ENOMEM;
664         goto fail;
665     }
666 
667     /* New interval for cache cleanup timer */
668     r->cache_clean_interval =
669         qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
670                             s->cache_clean_interval);
671 #ifndef CONFIG_LINUX
672     if (r->cache_clean_interval != 0) {
673         error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL
674                    " not supported on this host");
675         ret = -EINVAL;
676         goto fail;
677     }
678 #endif
679     if (r->cache_clean_interval > UINT_MAX) {
680         error_setg(errp, "Cache clean interval too big");
681         ret = -EINVAL;
682         goto fail;
683     }
684 
685     /* lazy-refcounts; flush if going from enabled to disabled */
686     r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
687         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
688     if (r->use_lazy_refcounts && s->qcow_version < 3) {
689         error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
690                    "qemu 1.1 compatibility level");
691         ret = -EINVAL;
692         goto fail;
693     }
694 
695     if (s->use_lazy_refcounts && !r->use_lazy_refcounts) {
696         ret = qcow2_mark_clean(bs);
697         if (ret < 0) {
698             error_setg_errno(errp, -ret, "Failed to disable lazy refcounts");
699             goto fail;
700         }
701     }
702 
703     /* Overlap check options */
704     opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP);
705     opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE);
706     if (opt_overlap_check_template && opt_overlap_check &&
707         strcmp(opt_overlap_check_template, opt_overlap_check))
708     {
709         error_setg(errp, "Conflicting values for qcow2 options '"
710                    QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
711                    "' ('%s')", opt_overlap_check, opt_overlap_check_template);
712         ret = -EINVAL;
713         goto fail;
714     }
715     if (!opt_overlap_check) {
716         opt_overlap_check = opt_overlap_check_template ?: "cached";
717     }
718 
719     if (!strcmp(opt_overlap_check, "none")) {
720         overlap_check_template = 0;
721     } else if (!strcmp(opt_overlap_check, "constant")) {
722         overlap_check_template = QCOW2_OL_CONSTANT;
723     } else if (!strcmp(opt_overlap_check, "cached")) {
724         overlap_check_template = QCOW2_OL_CACHED;
725     } else if (!strcmp(opt_overlap_check, "all")) {
726         overlap_check_template = QCOW2_OL_ALL;
727     } else {
728         error_setg(errp, "Unsupported value '%s' for qcow2 option "
729                    "'overlap-check'. Allowed are any of the following: "
730                    "none, constant, cached, all", opt_overlap_check);
731         ret = -EINVAL;
732         goto fail;
733     }
734 
735     r->overlap_check = 0;
736     for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) {
737         /* overlap-check defines a template bitmask, but every flag may be
738          * overwritten through the associated boolean option */
739         r->overlap_check |=
740             qemu_opt_get_bool(opts, overlap_bool_option_names[i],
741                               overlap_check_template & (1 << i)) << i;
742     }
743 
744     r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
745     r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
746     r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
747         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
748                           flags & BDRV_O_UNMAP);
749     r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
750         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
751     r->discard_passthrough[QCOW2_DISCARD_OTHER] =
752         qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
753 
754     ret = 0;
755 fail:
756     qemu_opts_del(opts);
757     opts = NULL;
758     return ret;
759 }
760 
761 static void qcow2_update_options_commit(BlockDriverState *bs,
762                                         Qcow2ReopenState *r)
763 {
764     BDRVQcow2State *s = bs->opaque;
765     int i;
766 
767     if (s->l2_table_cache) {
768         qcow2_cache_destroy(bs, s->l2_table_cache);
769     }
770     if (s->refcount_block_cache) {
771         qcow2_cache_destroy(bs, s->refcount_block_cache);
772     }
773     s->l2_table_cache = r->l2_table_cache;
774     s->refcount_block_cache = r->refcount_block_cache;
775 
776     s->overlap_check = r->overlap_check;
777     s->use_lazy_refcounts = r->use_lazy_refcounts;
778 
779     for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
780         s->discard_passthrough[i] = r->discard_passthrough[i];
781     }
782 
783     if (s->cache_clean_interval != r->cache_clean_interval) {
784         cache_clean_timer_del(bs);
785         s->cache_clean_interval = r->cache_clean_interval;
786         cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
787     }
788 }
789 
790 static void qcow2_update_options_abort(BlockDriverState *bs,
791                                        Qcow2ReopenState *r)
792 {
793     if (r->l2_table_cache) {
794         qcow2_cache_destroy(bs, r->l2_table_cache);
795     }
796     if (r->refcount_block_cache) {
797         qcow2_cache_destroy(bs, r->refcount_block_cache);
798     }
799 }
800 
801 static int qcow2_update_options(BlockDriverState *bs, QDict *options,
802                                 int flags, Error **errp)
803 {
804     Qcow2ReopenState r = {};
805     int ret;
806 
807     ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
808     if (ret >= 0) {
809         qcow2_update_options_commit(bs, &r);
810     } else {
811         qcow2_update_options_abort(bs, &r);
812     }
813 
814     return ret;
815 }
816 
817 static int qcow2_do_open(BlockDriverState *bs, QDict *options, int flags,
818                          Error **errp)
819 {
820     BDRVQcow2State *s = bs->opaque;
821     unsigned int len, i;
822     int ret = 0;
823     QCowHeader header;
824     Error *local_err = NULL;
825     uint64_t ext_end;
826     uint64_t l1_vm_state_index;
827 
828     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
829     if (ret < 0) {
830         error_setg_errno(errp, -ret, "Could not read qcow2 header");
831         goto fail;
832     }
833     be32_to_cpus(&header.magic);
834     be32_to_cpus(&header.version);
835     be64_to_cpus(&header.backing_file_offset);
836     be32_to_cpus(&header.backing_file_size);
837     be64_to_cpus(&header.size);
838     be32_to_cpus(&header.cluster_bits);
839     be32_to_cpus(&header.crypt_method);
840     be64_to_cpus(&header.l1_table_offset);
841     be32_to_cpus(&header.l1_size);
842     be64_to_cpus(&header.refcount_table_offset);
843     be32_to_cpus(&header.refcount_table_clusters);
844     be64_to_cpus(&header.snapshots_offset);
845     be32_to_cpus(&header.nb_snapshots);
846 
847     if (header.magic != QCOW_MAGIC) {
848         error_setg(errp, "Image is not in qcow2 format");
849         ret = -EINVAL;
850         goto fail;
851     }
852     if (header.version < 2 || header.version > 3) {
853         error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
854         ret = -ENOTSUP;
855         goto fail;
856     }
857 
858     s->qcow_version = header.version;
859 
860     /* Initialise cluster size */
861     if (header.cluster_bits < MIN_CLUSTER_BITS ||
862         header.cluster_bits > MAX_CLUSTER_BITS) {
863         error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
864                    header.cluster_bits);
865         ret = -EINVAL;
866         goto fail;
867     }
868 
869     s->cluster_bits = header.cluster_bits;
870     s->cluster_size = 1 << s->cluster_bits;
871     s->cluster_sectors = 1 << (s->cluster_bits - 9);
872 
873     /* Initialise version 3 header fields */
874     if (header.version == 2) {
875         header.incompatible_features    = 0;
876         header.compatible_features      = 0;
877         header.autoclear_features       = 0;
878         header.refcount_order           = 4;
879         header.header_length            = 72;
880     } else {
881         be64_to_cpus(&header.incompatible_features);
882         be64_to_cpus(&header.compatible_features);
883         be64_to_cpus(&header.autoclear_features);
884         be32_to_cpus(&header.refcount_order);
885         be32_to_cpus(&header.header_length);
886 
887         if (header.header_length < 104) {
888             error_setg(errp, "qcow2 header too short");
889             ret = -EINVAL;
890             goto fail;
891         }
892     }
893 
894     if (header.header_length > s->cluster_size) {
895         error_setg(errp, "qcow2 header exceeds cluster size");
896         ret = -EINVAL;
897         goto fail;
898     }
899 
900     if (header.header_length > sizeof(header)) {
901         s->unknown_header_fields_size = header.header_length - sizeof(header);
902         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
903         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
904                          s->unknown_header_fields_size);
905         if (ret < 0) {
906             error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
907                              "fields");
908             goto fail;
909         }
910     }
911 
912     if (header.backing_file_offset > s->cluster_size) {
913         error_setg(errp, "Invalid backing file offset");
914         ret = -EINVAL;
915         goto fail;
916     }
917 
918     if (header.backing_file_offset) {
919         ext_end = header.backing_file_offset;
920     } else {
921         ext_end = 1 << header.cluster_bits;
922     }
923 
924     /* Handle feature bits */
925     s->incompatible_features    = header.incompatible_features;
926     s->compatible_features      = header.compatible_features;
927     s->autoclear_features       = header.autoclear_features;
928 
929     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
930         void *feature_table = NULL;
931         qcow2_read_extensions(bs, header.header_length, ext_end,
932                               &feature_table, NULL);
933         report_unsupported_feature(errp, feature_table,
934                                    s->incompatible_features &
935                                    ~QCOW2_INCOMPAT_MASK);
936         ret = -ENOTSUP;
937         g_free(feature_table);
938         goto fail;
939     }
940 
941     if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
942         /* Corrupt images may not be written to unless they are being repaired
943          */
944         if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
945             error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
946                        "read/write");
947             ret = -EACCES;
948             goto fail;
949         }
950     }
951 
952     /* Check support for various header values */
953     if (header.refcount_order > 6) {
954         error_setg(errp, "Reference count entry width too large; may not "
955                    "exceed 64 bits");
956         ret = -EINVAL;
957         goto fail;
958     }
959     s->refcount_order = header.refcount_order;
960     s->refcount_bits = 1 << s->refcount_order;
961     s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
962     s->refcount_max += s->refcount_max - 1;
963 
964     if (header.crypt_method > QCOW_CRYPT_AES) {
965         error_setg(errp, "Unsupported encryption method: %" PRIu32,
966                    header.crypt_method);
967         ret = -EINVAL;
968         goto fail;
969     }
970     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128,
971                                  QCRYPTO_CIPHER_MODE_CBC)) {
972         error_setg(errp, "AES cipher not available");
973         ret = -EINVAL;
974         goto fail;
975     }
976     s->crypt_method_header = header.crypt_method;
977     if (s->crypt_method_header) {
978         if (bdrv_uses_whitelist() &&
979             s->crypt_method_header == QCOW_CRYPT_AES) {
980             error_setg(errp,
981                        "Use of AES-CBC encrypted qcow2 images is no longer "
982                        "supported in system emulators");
983             error_append_hint(errp,
984                               "You can use 'qemu-img convert' to convert your "
985                               "image to an alternative supported format, such "
986                               "as unencrypted qcow2, or raw with the LUKS "
987                               "format instead.\n");
988             ret = -ENOSYS;
989             goto fail;
990         }
991 
992         bs->encrypted = true;
993     }
994 
995     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
996     s->l2_size = 1 << s->l2_bits;
997     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
998     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
999     s->refcount_block_size = 1 << s->refcount_block_bits;
1000     bs->total_sectors = header.size / 512;
1001     s->csize_shift = (62 - (s->cluster_bits - 8));
1002     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1003     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
1004 
1005     s->refcount_table_offset = header.refcount_table_offset;
1006     s->refcount_table_size =
1007         header.refcount_table_clusters << (s->cluster_bits - 3);
1008 
1009     if (header.refcount_table_clusters > qcow2_max_refcount_clusters(s)) {
1010         error_setg(errp, "Reference count table too large");
1011         ret = -EINVAL;
1012         goto fail;
1013     }
1014 
1015     ret = validate_table_offset(bs, s->refcount_table_offset,
1016                                 s->refcount_table_size, sizeof(uint64_t));
1017     if (ret < 0) {
1018         error_setg(errp, "Invalid reference count table offset");
1019         goto fail;
1020     }
1021 
1022     /* Snapshot table offset/length */
1023     if (header.nb_snapshots > QCOW_MAX_SNAPSHOTS) {
1024         error_setg(errp, "Too many snapshots");
1025         ret = -EINVAL;
1026         goto fail;
1027     }
1028 
1029     ret = validate_table_offset(bs, header.snapshots_offset,
1030                                 header.nb_snapshots,
1031                                 sizeof(QCowSnapshotHeader));
1032     if (ret < 0) {
1033         error_setg(errp, "Invalid snapshot table offset");
1034         goto fail;
1035     }
1036 
1037     /* read the level 1 table */
1038     if (header.l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
1039         error_setg(errp, "Active L1 table too large");
1040         ret = -EFBIG;
1041         goto fail;
1042     }
1043     s->l1_size = header.l1_size;
1044 
1045     l1_vm_state_index = size_to_l1(s, header.size);
1046     if (l1_vm_state_index > INT_MAX) {
1047         error_setg(errp, "Image is too big");
1048         ret = -EFBIG;
1049         goto fail;
1050     }
1051     s->l1_vm_state_index = l1_vm_state_index;
1052 
1053     /* the L1 table must contain at least enough entries to put
1054        header.size bytes */
1055     if (s->l1_size < s->l1_vm_state_index) {
1056         error_setg(errp, "L1 table is too small");
1057         ret = -EINVAL;
1058         goto fail;
1059     }
1060 
1061     ret = validate_table_offset(bs, header.l1_table_offset,
1062                                 header.l1_size, sizeof(uint64_t));
1063     if (ret < 0) {
1064         error_setg(errp, "Invalid L1 table offset");
1065         goto fail;
1066     }
1067     s->l1_table_offset = header.l1_table_offset;
1068 
1069 
1070     if (s->l1_size > 0) {
1071         s->l1_table = qemu_try_blockalign(bs->file->bs,
1072             align_offset(s->l1_size * sizeof(uint64_t), 512));
1073         if (s->l1_table == NULL) {
1074             error_setg(errp, "Could not allocate L1 table");
1075             ret = -ENOMEM;
1076             goto fail;
1077         }
1078         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
1079                          s->l1_size * sizeof(uint64_t));
1080         if (ret < 0) {
1081             error_setg_errno(errp, -ret, "Could not read L1 table");
1082             goto fail;
1083         }
1084         for(i = 0;i < s->l1_size; i++) {
1085             be64_to_cpus(&s->l1_table[i]);
1086         }
1087     }
1088 
1089     /* Parse driver-specific options */
1090     ret = qcow2_update_options(bs, options, flags, errp);
1091     if (ret < 0) {
1092         goto fail;
1093     }
1094 
1095     s->cluster_cache = g_malloc(s->cluster_size);
1096     /* one more sector for decompressed data alignment */
1097     s->cluster_data = qemu_try_blockalign(bs->file->bs, QCOW_MAX_CRYPT_CLUSTERS
1098                                                     * s->cluster_size + 512);
1099     if (s->cluster_data == NULL) {
1100         error_setg(errp, "Could not allocate temporary cluster buffer");
1101         ret = -ENOMEM;
1102         goto fail;
1103     }
1104 
1105     s->cluster_cache_offset = -1;
1106     s->flags = flags;
1107 
1108     ret = qcow2_refcount_init(bs);
1109     if (ret != 0) {
1110         error_setg_errno(errp, -ret, "Could not initialize refcount handling");
1111         goto fail;
1112     }
1113 
1114     QLIST_INIT(&s->cluster_allocs);
1115     QTAILQ_INIT(&s->discards);
1116 
1117     /* read qcow2 extensions */
1118     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
1119         &local_err)) {
1120         error_propagate(errp, local_err);
1121         ret = -EINVAL;
1122         goto fail;
1123     }
1124 
1125     /* read the backing file name */
1126     if (header.backing_file_offset != 0) {
1127         len = header.backing_file_size;
1128         if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
1129             len >= sizeof(bs->backing_file)) {
1130             error_setg(errp, "Backing file name too long");
1131             ret = -EINVAL;
1132             goto fail;
1133         }
1134         ret = bdrv_pread(bs->file, header.backing_file_offset,
1135                          bs->backing_file, len);
1136         if (ret < 0) {
1137             error_setg_errno(errp, -ret, "Could not read backing file name");
1138             goto fail;
1139         }
1140         bs->backing_file[len] = '\0';
1141         s->image_backing_file = g_strdup(bs->backing_file);
1142     }
1143 
1144     /* Internal snapshots */
1145     s->snapshots_offset = header.snapshots_offset;
1146     s->nb_snapshots = header.nb_snapshots;
1147 
1148     ret = qcow2_read_snapshots(bs);
1149     if (ret < 0) {
1150         error_setg_errno(errp, -ret, "Could not read snapshots");
1151         goto fail;
1152     }
1153 
1154     /* Clear unknown autoclear feature bits */
1155     if (!bs->read_only && !(flags & BDRV_O_INACTIVE) && s->autoclear_features) {
1156         s->autoclear_features = 0;
1157         ret = qcow2_update_header(bs);
1158         if (ret < 0) {
1159             error_setg_errno(errp, -ret, "Could not update qcow2 header");
1160             goto fail;
1161         }
1162     }
1163 
1164     /* Initialise locks */
1165     qemu_co_mutex_init(&s->lock);
1166     bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
1167 
1168     /* Repair image if dirty */
1169     if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
1170         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
1171         BdrvCheckResult result = {0};
1172 
1173         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
1174         if (ret < 0) {
1175             error_setg_errno(errp, -ret, "Could not repair dirty image");
1176             goto fail;
1177         }
1178     }
1179 
1180 #ifdef DEBUG_ALLOC
1181     {
1182         BdrvCheckResult result = {0};
1183         qcow2_check_refcounts(bs, &result, 0);
1184     }
1185 #endif
1186     return ret;
1187 
1188  fail:
1189     g_free(s->unknown_header_fields);
1190     cleanup_unknown_header_ext(bs);
1191     qcow2_free_snapshots(bs);
1192     qcow2_refcount_close(bs);
1193     qemu_vfree(s->l1_table);
1194     /* else pre-write overlap checks in cache_destroy may crash */
1195     s->l1_table = NULL;
1196     cache_clean_timer_del(bs);
1197     if (s->l2_table_cache) {
1198         qcow2_cache_destroy(bs, s->l2_table_cache);
1199     }
1200     if (s->refcount_block_cache) {
1201         qcow2_cache_destroy(bs, s->refcount_block_cache);
1202     }
1203     g_free(s->cluster_cache);
1204     qemu_vfree(s->cluster_data);
1205     return ret;
1206 }
1207 
1208 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1209                       Error **errp)
1210 {
1211     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
1212                                false, errp);
1213     if (!bs->file) {
1214         return -EINVAL;
1215     }
1216 
1217     return qcow2_do_open(bs, options, flags, errp);
1218 }
1219 
1220 static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
1221 {
1222     BDRVQcow2State *s = bs->opaque;
1223 
1224     if (bs->encrypted) {
1225         /* Encryption works on a sector granularity */
1226         bs->bl.request_alignment = BDRV_SECTOR_SIZE;
1227     }
1228     bs->bl.pwrite_zeroes_alignment = s->cluster_size;
1229     bs->bl.pdiscard_alignment = s->cluster_size;
1230 }
1231 
1232 static int qcow2_set_key(BlockDriverState *bs, const char *key)
1233 {
1234     BDRVQcow2State *s = bs->opaque;
1235     uint8_t keybuf[16];
1236     int len, i;
1237     Error *err = NULL;
1238 
1239     memset(keybuf, 0, 16);
1240     len = strlen(key);
1241     if (len > 16)
1242         len = 16;
1243     /* XXX: we could compress the chars to 7 bits to increase
1244        entropy */
1245     for(i = 0;i < len;i++) {
1246         keybuf[i] = key[i];
1247     }
1248     assert(bs->encrypted);
1249 
1250     qcrypto_cipher_free(s->cipher);
1251     s->cipher = qcrypto_cipher_new(
1252         QCRYPTO_CIPHER_ALG_AES_128,
1253         QCRYPTO_CIPHER_MODE_CBC,
1254         keybuf, G_N_ELEMENTS(keybuf),
1255         &err);
1256 
1257     if (!s->cipher) {
1258         /* XXX would be nice if errors in this method could
1259          * be properly propagate to the caller. Would need
1260          * the bdrv_set_key() API signature to be fixed. */
1261         error_free(err);
1262         return -1;
1263     }
1264     return 0;
1265 }
1266 
1267 static int qcow2_reopen_prepare(BDRVReopenState *state,
1268                                 BlockReopenQueue *queue, Error **errp)
1269 {
1270     Qcow2ReopenState *r;
1271     int ret;
1272 
1273     r = g_new0(Qcow2ReopenState, 1);
1274     state->opaque = r;
1275 
1276     ret = qcow2_update_options_prepare(state->bs, r, state->options,
1277                                        state->flags, errp);
1278     if (ret < 0) {
1279         goto fail;
1280     }
1281 
1282     /* We need to write out any unwritten data if we reopen read-only. */
1283     if ((state->flags & BDRV_O_RDWR) == 0) {
1284         ret = bdrv_flush(state->bs);
1285         if (ret < 0) {
1286             goto fail;
1287         }
1288 
1289         ret = qcow2_mark_clean(state->bs);
1290         if (ret < 0) {
1291             goto fail;
1292         }
1293     }
1294 
1295     return 0;
1296 
1297 fail:
1298     qcow2_update_options_abort(state->bs, r);
1299     g_free(r);
1300     return ret;
1301 }
1302 
1303 static void qcow2_reopen_commit(BDRVReopenState *state)
1304 {
1305     qcow2_update_options_commit(state->bs, state->opaque);
1306     g_free(state->opaque);
1307 }
1308 
1309 static void qcow2_reopen_abort(BDRVReopenState *state)
1310 {
1311     qcow2_update_options_abort(state->bs, state->opaque);
1312     g_free(state->opaque);
1313 }
1314 
1315 static void qcow2_join_options(QDict *options, QDict *old_options)
1316 {
1317     bool has_new_overlap_template =
1318         qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
1319         qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
1320     bool has_new_total_cache_size =
1321         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
1322     bool has_all_cache_options;
1323 
1324     /* New overlap template overrides all old overlap options */
1325     if (has_new_overlap_template) {
1326         qdict_del(old_options, QCOW2_OPT_OVERLAP);
1327         qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
1328         qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
1329         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
1330         qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
1331         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
1332         qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
1333         qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
1334         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
1335         qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
1336     }
1337 
1338     /* New total cache size overrides all old options */
1339     if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
1340         qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
1341         qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1342     }
1343 
1344     qdict_join(options, old_options, false);
1345 
1346     /*
1347      * If after merging all cache size options are set, an old total size is
1348      * overwritten. Do keep all options, however, if all three are new. The
1349      * resulting error message is what we want to happen.
1350      */
1351     has_all_cache_options =
1352         qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
1353         qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
1354         qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
1355 
1356     if (has_all_cache_options && !has_new_total_cache_size) {
1357         qdict_del(options, QCOW2_OPT_CACHE_SIZE);
1358     }
1359 }
1360 
1361 static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
1362         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1363 {
1364     BDRVQcow2State *s = bs->opaque;
1365     uint64_t cluster_offset;
1366     int index_in_cluster, ret;
1367     unsigned int bytes;
1368     int64_t status = 0;
1369 
1370     bytes = MIN(INT_MAX, nb_sectors * BDRV_SECTOR_SIZE);
1371     qemu_co_mutex_lock(&s->lock);
1372     ret = qcow2_get_cluster_offset(bs, sector_num << 9, &bytes,
1373                                    &cluster_offset);
1374     qemu_co_mutex_unlock(&s->lock);
1375     if (ret < 0) {
1376         return ret;
1377     }
1378 
1379     *pnum = bytes >> BDRV_SECTOR_BITS;
1380 
1381     if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED &&
1382         !s->cipher) {
1383         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1384         cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
1385         *file = bs->file->bs;
1386         status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset;
1387     }
1388     if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) {
1389         status |= BDRV_BLOCK_ZERO;
1390     } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
1391         status |= BDRV_BLOCK_DATA;
1392     }
1393     return status;
1394 }
1395 
1396 /* handle reading after the end of the backing file */
1397 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
1398                         int64_t offset, int bytes)
1399 {
1400     uint64_t bs_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1401     int n1;
1402 
1403     if ((offset + bytes) <= bs_size) {
1404         return bytes;
1405     }
1406 
1407     if (offset >= bs_size) {
1408         n1 = 0;
1409     } else {
1410         n1 = bs_size - offset;
1411     }
1412 
1413     qemu_iovec_memset(qiov, n1, 0, bytes - n1);
1414 
1415     return n1;
1416 }
1417 
1418 static coroutine_fn int qcow2_co_preadv(BlockDriverState *bs, uint64_t offset,
1419                                         uint64_t bytes, QEMUIOVector *qiov,
1420                                         int flags)
1421 {
1422     BDRVQcow2State *s = bs->opaque;
1423     int offset_in_cluster, n1;
1424     int ret;
1425     unsigned int cur_bytes; /* number of bytes in current iteration */
1426     uint64_t cluster_offset = 0;
1427     uint64_t bytes_done = 0;
1428     QEMUIOVector hd_qiov;
1429     uint8_t *cluster_data = NULL;
1430 
1431     qemu_iovec_init(&hd_qiov, qiov->niov);
1432 
1433     qemu_co_mutex_lock(&s->lock);
1434 
1435     while (bytes != 0) {
1436 
1437         /* prepare next request */
1438         cur_bytes = MIN(bytes, INT_MAX);
1439         if (s->cipher) {
1440             cur_bytes = MIN(cur_bytes,
1441                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1442         }
1443 
1444         ret = qcow2_get_cluster_offset(bs, offset, &cur_bytes, &cluster_offset);
1445         if (ret < 0) {
1446             goto fail;
1447         }
1448 
1449         offset_in_cluster = offset_into_cluster(s, offset);
1450 
1451         qemu_iovec_reset(&hd_qiov);
1452         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
1453 
1454         switch (ret) {
1455         case QCOW2_CLUSTER_UNALLOCATED:
1456 
1457             if (bs->backing) {
1458                 /* read from the base image */
1459                 n1 = qcow2_backing_read1(bs->backing->bs, &hd_qiov,
1460                                          offset, cur_bytes);
1461                 if (n1 > 0) {
1462                     QEMUIOVector local_qiov;
1463 
1464                     qemu_iovec_init(&local_qiov, hd_qiov.niov);
1465                     qemu_iovec_concat(&local_qiov, &hd_qiov, 0, n1);
1466 
1467                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
1468                     qemu_co_mutex_unlock(&s->lock);
1469                     ret = bdrv_co_preadv(bs->backing, offset, n1,
1470                                          &local_qiov, 0);
1471                     qemu_co_mutex_lock(&s->lock);
1472 
1473                     qemu_iovec_destroy(&local_qiov);
1474 
1475                     if (ret < 0) {
1476                         goto fail;
1477                     }
1478                 }
1479             } else {
1480                 /* Note: in this case, no need to wait */
1481                 qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1482             }
1483             break;
1484 
1485         case QCOW2_CLUSTER_ZERO_PLAIN:
1486         case QCOW2_CLUSTER_ZERO_ALLOC:
1487             qemu_iovec_memset(&hd_qiov, 0, 0, cur_bytes);
1488             break;
1489 
1490         case QCOW2_CLUSTER_COMPRESSED:
1491             /* add AIO support for compressed blocks ? */
1492             ret = qcow2_decompress_cluster(bs, cluster_offset);
1493             if (ret < 0) {
1494                 goto fail;
1495             }
1496 
1497             qemu_iovec_from_buf(&hd_qiov, 0,
1498                                 s->cluster_cache + offset_in_cluster,
1499                                 cur_bytes);
1500             break;
1501 
1502         case QCOW2_CLUSTER_NORMAL:
1503             if ((cluster_offset & 511) != 0) {
1504                 ret = -EIO;
1505                 goto fail;
1506             }
1507 
1508             if (bs->encrypted) {
1509                 assert(s->cipher);
1510 
1511                 /*
1512                  * For encrypted images, read everything into a temporary
1513                  * contiguous buffer on which the AES functions can work.
1514                  */
1515                 if (!cluster_data) {
1516                     cluster_data =
1517                         qemu_try_blockalign(bs->file->bs,
1518                                             QCOW_MAX_CRYPT_CLUSTERS
1519                                             * s->cluster_size);
1520                     if (cluster_data == NULL) {
1521                         ret = -ENOMEM;
1522                         goto fail;
1523                     }
1524                 }
1525 
1526                 assert(cur_bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1527                 qemu_iovec_reset(&hd_qiov);
1528                 qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
1529             }
1530 
1531             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1532             qemu_co_mutex_unlock(&s->lock);
1533             ret = bdrv_co_preadv(bs->file,
1534                                  cluster_offset + offset_in_cluster,
1535                                  cur_bytes, &hd_qiov, 0);
1536             qemu_co_mutex_lock(&s->lock);
1537             if (ret < 0) {
1538                 goto fail;
1539             }
1540             if (bs->encrypted) {
1541                 assert(s->cipher);
1542                 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
1543                 assert((cur_bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
1544                 Error *err = NULL;
1545                 if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS,
1546                                           cluster_data, cluster_data,
1547                                           cur_bytes >> BDRV_SECTOR_BITS,
1548                                           false, &err) < 0) {
1549                     error_free(err);
1550                     ret = -EIO;
1551                     goto fail;
1552                 }
1553                 qemu_iovec_from_buf(qiov, bytes_done, cluster_data, cur_bytes);
1554             }
1555             break;
1556 
1557         default:
1558             g_assert_not_reached();
1559             ret = -EIO;
1560             goto fail;
1561         }
1562 
1563         bytes -= cur_bytes;
1564         offset += cur_bytes;
1565         bytes_done += cur_bytes;
1566     }
1567     ret = 0;
1568 
1569 fail:
1570     qemu_co_mutex_unlock(&s->lock);
1571 
1572     qemu_iovec_destroy(&hd_qiov);
1573     qemu_vfree(cluster_data);
1574 
1575     return ret;
1576 }
1577 
1578 static coroutine_fn int qcow2_co_pwritev(BlockDriverState *bs, uint64_t offset,
1579                                          uint64_t bytes, QEMUIOVector *qiov,
1580                                          int flags)
1581 {
1582     BDRVQcow2State *s = bs->opaque;
1583     int offset_in_cluster;
1584     int ret;
1585     unsigned int cur_bytes; /* number of sectors in current iteration */
1586     uint64_t cluster_offset;
1587     QEMUIOVector hd_qiov;
1588     uint64_t bytes_done = 0;
1589     uint8_t *cluster_data = NULL;
1590     QCowL2Meta *l2meta = NULL;
1591 
1592     trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
1593 
1594     qemu_iovec_init(&hd_qiov, qiov->niov);
1595 
1596     s->cluster_cache_offset = -1; /* disable compressed cache */
1597 
1598     qemu_co_mutex_lock(&s->lock);
1599 
1600     while (bytes != 0) {
1601 
1602         l2meta = NULL;
1603 
1604         trace_qcow2_writev_start_part(qemu_coroutine_self());
1605         offset_in_cluster = offset_into_cluster(s, offset);
1606         cur_bytes = MIN(bytes, INT_MAX);
1607         if (bs->encrypted) {
1608             cur_bytes = MIN(cur_bytes,
1609                             QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
1610                             - offset_in_cluster);
1611         }
1612 
1613         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
1614                                          &cluster_offset, &l2meta);
1615         if (ret < 0) {
1616             goto fail;
1617         }
1618 
1619         assert((cluster_offset & 511) == 0);
1620 
1621         qemu_iovec_reset(&hd_qiov);
1622         qemu_iovec_concat(&hd_qiov, qiov, bytes_done, cur_bytes);
1623 
1624         if (bs->encrypted) {
1625             Error *err = NULL;
1626             assert(s->cipher);
1627             if (!cluster_data) {
1628                 cluster_data = qemu_try_blockalign(bs->file->bs,
1629                                                    QCOW_MAX_CRYPT_CLUSTERS
1630                                                    * s->cluster_size);
1631                 if (cluster_data == NULL) {
1632                     ret = -ENOMEM;
1633                     goto fail;
1634                 }
1635             }
1636 
1637             assert(hd_qiov.size <=
1638                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
1639             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
1640 
1641             if (qcow2_encrypt_sectors(s, offset >> BDRV_SECTOR_BITS,
1642                                       cluster_data, cluster_data,
1643                                       cur_bytes >>BDRV_SECTOR_BITS,
1644                                       true, &err) < 0) {
1645                 error_free(err);
1646                 ret = -EIO;
1647                 goto fail;
1648             }
1649 
1650             qemu_iovec_reset(&hd_qiov);
1651             qemu_iovec_add(&hd_qiov, cluster_data, cur_bytes);
1652         }
1653 
1654         ret = qcow2_pre_write_overlap_check(bs, 0,
1655                 cluster_offset + offset_in_cluster, cur_bytes);
1656         if (ret < 0) {
1657             goto fail;
1658         }
1659 
1660         qemu_co_mutex_unlock(&s->lock);
1661         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
1662         trace_qcow2_writev_data(qemu_coroutine_self(),
1663                                 cluster_offset + offset_in_cluster);
1664         ret = bdrv_co_pwritev(bs->file,
1665                               cluster_offset + offset_in_cluster,
1666                               cur_bytes, &hd_qiov, 0);
1667         qemu_co_mutex_lock(&s->lock);
1668         if (ret < 0) {
1669             goto fail;
1670         }
1671 
1672         while (l2meta != NULL) {
1673             QCowL2Meta *next;
1674 
1675             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
1676             if (ret < 0) {
1677                 goto fail;
1678             }
1679 
1680             /* Take the request off the list of running requests */
1681             if (l2meta->nb_clusters != 0) {
1682                 QLIST_REMOVE(l2meta, next_in_flight);
1683             }
1684 
1685             qemu_co_queue_restart_all(&l2meta->dependent_requests);
1686 
1687             next = l2meta->next;
1688             g_free(l2meta);
1689             l2meta = next;
1690         }
1691 
1692         bytes -= cur_bytes;
1693         offset += cur_bytes;
1694         bytes_done += cur_bytes;
1695         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
1696     }
1697     ret = 0;
1698 
1699 fail:
1700     qemu_co_mutex_unlock(&s->lock);
1701 
1702     while (l2meta != NULL) {
1703         QCowL2Meta *next;
1704 
1705         if (l2meta->nb_clusters != 0) {
1706             QLIST_REMOVE(l2meta, next_in_flight);
1707         }
1708         qemu_co_queue_restart_all(&l2meta->dependent_requests);
1709 
1710         next = l2meta->next;
1711         g_free(l2meta);
1712         l2meta = next;
1713     }
1714 
1715     qemu_iovec_destroy(&hd_qiov);
1716     qemu_vfree(cluster_data);
1717     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
1718 
1719     return ret;
1720 }
1721 
1722 static int qcow2_inactivate(BlockDriverState *bs)
1723 {
1724     BDRVQcow2State *s = bs->opaque;
1725     int ret, result = 0;
1726 
1727     ret = qcow2_cache_flush(bs, s->l2_table_cache);
1728     if (ret) {
1729         result = ret;
1730         error_report("Failed to flush the L2 table cache: %s",
1731                      strerror(-ret));
1732     }
1733 
1734     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1735     if (ret) {
1736         result = ret;
1737         error_report("Failed to flush the refcount block cache: %s",
1738                      strerror(-ret));
1739     }
1740 
1741     if (result == 0) {
1742         qcow2_mark_clean(bs);
1743     }
1744 
1745     return result;
1746 }
1747 
1748 static void qcow2_close(BlockDriverState *bs)
1749 {
1750     BDRVQcow2State *s = bs->opaque;
1751     qemu_vfree(s->l1_table);
1752     /* else pre-write overlap checks in cache_destroy may crash */
1753     s->l1_table = NULL;
1754 
1755     if (!(s->flags & BDRV_O_INACTIVE)) {
1756         qcow2_inactivate(bs);
1757     }
1758 
1759     cache_clean_timer_del(bs);
1760     qcow2_cache_destroy(bs, s->l2_table_cache);
1761     qcow2_cache_destroy(bs, s->refcount_block_cache);
1762 
1763     qcrypto_cipher_free(s->cipher);
1764     s->cipher = NULL;
1765 
1766     g_free(s->unknown_header_fields);
1767     cleanup_unknown_header_ext(bs);
1768 
1769     g_free(s->image_backing_file);
1770     g_free(s->image_backing_format);
1771 
1772     g_free(s->cluster_cache);
1773     qemu_vfree(s->cluster_data);
1774     qcow2_refcount_close(bs);
1775     qcow2_free_snapshots(bs);
1776 }
1777 
1778 static void qcow2_invalidate_cache(BlockDriverState *bs, Error **errp)
1779 {
1780     BDRVQcow2State *s = bs->opaque;
1781     int flags = s->flags;
1782     QCryptoCipher *cipher = NULL;
1783     QDict *options;
1784     Error *local_err = NULL;
1785     int ret;
1786 
1787     /*
1788      * Backing files are read-only which makes all of their metadata immutable,
1789      * that means we don't have to worry about reopening them here.
1790      */
1791 
1792     cipher = s->cipher;
1793     s->cipher = NULL;
1794 
1795     qcow2_close(bs);
1796 
1797     memset(s, 0, sizeof(BDRVQcow2State));
1798     options = qdict_clone_shallow(bs->options);
1799 
1800     flags &= ~BDRV_O_INACTIVE;
1801     ret = qcow2_do_open(bs, options, flags, &local_err);
1802     QDECREF(options);
1803     if (local_err) {
1804         error_propagate(errp, local_err);
1805         error_prepend(errp, "Could not reopen qcow2 layer: ");
1806         bs->drv = NULL;
1807         return;
1808     } else if (ret < 0) {
1809         error_setg_errno(errp, -ret, "Could not reopen qcow2 layer");
1810         bs->drv = NULL;
1811         return;
1812     }
1813 
1814     s->cipher = cipher;
1815 }
1816 
1817 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
1818     size_t len, size_t buflen)
1819 {
1820     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
1821     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
1822 
1823     if (buflen < ext_len) {
1824         return -ENOSPC;
1825     }
1826 
1827     *ext_backing_fmt = (QCowExtension) {
1828         .magic  = cpu_to_be32(magic),
1829         .len    = cpu_to_be32(len),
1830     };
1831 
1832     if (len) {
1833         memcpy(buf + sizeof(QCowExtension), s, len);
1834     }
1835 
1836     return ext_len;
1837 }
1838 
1839 /*
1840  * Updates the qcow2 header, including the variable length parts of it, i.e.
1841  * the backing file name and all extensions. qcow2 was not designed to allow
1842  * such changes, so if we run out of space (we can only use the first cluster)
1843  * this function may fail.
1844  *
1845  * Returns 0 on success, -errno in error cases.
1846  */
1847 int qcow2_update_header(BlockDriverState *bs)
1848 {
1849     BDRVQcow2State *s = bs->opaque;
1850     QCowHeader *header;
1851     char *buf;
1852     size_t buflen = s->cluster_size;
1853     int ret;
1854     uint64_t total_size;
1855     uint32_t refcount_table_clusters;
1856     size_t header_length;
1857     Qcow2UnknownHeaderExtension *uext;
1858 
1859     buf = qemu_blockalign(bs, buflen);
1860 
1861     /* Header structure */
1862     header = (QCowHeader*) buf;
1863 
1864     if (buflen < sizeof(*header)) {
1865         ret = -ENOSPC;
1866         goto fail;
1867     }
1868 
1869     header_length = sizeof(*header) + s->unknown_header_fields_size;
1870     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1871     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1872 
1873     *header = (QCowHeader) {
1874         /* Version 2 fields */
1875         .magic                  = cpu_to_be32(QCOW_MAGIC),
1876         .version                = cpu_to_be32(s->qcow_version),
1877         .backing_file_offset    = 0,
1878         .backing_file_size      = 0,
1879         .cluster_bits           = cpu_to_be32(s->cluster_bits),
1880         .size                   = cpu_to_be64(total_size),
1881         .crypt_method           = cpu_to_be32(s->crypt_method_header),
1882         .l1_size                = cpu_to_be32(s->l1_size),
1883         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
1884         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
1885         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1886         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
1887         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
1888 
1889         /* Version 3 fields */
1890         .incompatible_features  = cpu_to_be64(s->incompatible_features),
1891         .compatible_features    = cpu_to_be64(s->compatible_features),
1892         .autoclear_features     = cpu_to_be64(s->autoclear_features),
1893         .refcount_order         = cpu_to_be32(s->refcount_order),
1894         .header_length          = cpu_to_be32(header_length),
1895     };
1896 
1897     /* For older versions, write a shorter header */
1898     switch (s->qcow_version) {
1899     case 2:
1900         ret = offsetof(QCowHeader, incompatible_features);
1901         break;
1902     case 3:
1903         ret = sizeof(*header);
1904         break;
1905     default:
1906         ret = -EINVAL;
1907         goto fail;
1908     }
1909 
1910     buf += ret;
1911     buflen -= ret;
1912     memset(buf, 0, buflen);
1913 
1914     /* Preserve any unknown field in the header */
1915     if (s->unknown_header_fields_size) {
1916         if (buflen < s->unknown_header_fields_size) {
1917             ret = -ENOSPC;
1918             goto fail;
1919         }
1920 
1921         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1922         buf += s->unknown_header_fields_size;
1923         buflen -= s->unknown_header_fields_size;
1924     }
1925 
1926     /* Backing file format header extension */
1927     if (s->image_backing_format) {
1928         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1929                              s->image_backing_format,
1930                              strlen(s->image_backing_format),
1931                              buflen);
1932         if (ret < 0) {
1933             goto fail;
1934         }
1935 
1936         buf += ret;
1937         buflen -= ret;
1938     }
1939 
1940     /* Feature table */
1941     if (s->qcow_version >= 3) {
1942         Qcow2Feature features[] = {
1943             {
1944                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1945                 .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
1946                 .name = "dirty bit",
1947             },
1948             {
1949                 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1950                 .bit  = QCOW2_INCOMPAT_CORRUPT_BITNR,
1951                 .name = "corrupt bit",
1952             },
1953             {
1954                 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1955                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1956                 .name = "lazy refcounts",
1957             },
1958         };
1959 
1960         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1961                              features, sizeof(features), buflen);
1962         if (ret < 0) {
1963             goto fail;
1964         }
1965         buf += ret;
1966         buflen -= ret;
1967     }
1968 
1969     /* Keep unknown header extensions */
1970     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1971         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1972         if (ret < 0) {
1973             goto fail;
1974         }
1975 
1976         buf += ret;
1977         buflen -= ret;
1978     }
1979 
1980     /* End of header extensions */
1981     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1982     if (ret < 0) {
1983         goto fail;
1984     }
1985 
1986     buf += ret;
1987     buflen -= ret;
1988 
1989     /* Backing file name */
1990     if (s->image_backing_file) {
1991         size_t backing_file_len = strlen(s->image_backing_file);
1992 
1993         if (buflen < backing_file_len) {
1994             ret = -ENOSPC;
1995             goto fail;
1996         }
1997 
1998         /* Using strncpy is ok here, since buf is not NUL-terminated. */
1999         strncpy(buf, s->image_backing_file, buflen);
2000 
2001         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
2002         header->backing_file_size   = cpu_to_be32(backing_file_len);
2003     }
2004 
2005     /* Write the new header */
2006     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
2007     if (ret < 0) {
2008         goto fail;
2009     }
2010 
2011     ret = 0;
2012 fail:
2013     qemu_vfree(header);
2014     return ret;
2015 }
2016 
2017 static int qcow2_change_backing_file(BlockDriverState *bs,
2018     const char *backing_file, const char *backing_fmt)
2019 {
2020     BDRVQcow2State *s = bs->opaque;
2021 
2022     if (backing_file && strlen(backing_file) > 1023) {
2023         return -EINVAL;
2024     }
2025 
2026     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2027     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2028 
2029     g_free(s->image_backing_file);
2030     g_free(s->image_backing_format);
2031 
2032     s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
2033     s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
2034 
2035     return qcow2_update_header(bs);
2036 }
2037 
2038 static int preallocate(BlockDriverState *bs)
2039 {
2040     uint64_t bytes;
2041     uint64_t offset;
2042     uint64_t host_offset = 0;
2043     unsigned int cur_bytes;
2044     int ret;
2045     QCowL2Meta *meta;
2046 
2047     bytes = bdrv_getlength(bs);
2048     offset = 0;
2049 
2050     while (bytes) {
2051         cur_bytes = MIN(bytes, INT_MAX);
2052         ret = qcow2_alloc_cluster_offset(bs, offset, &cur_bytes,
2053                                          &host_offset, &meta);
2054         if (ret < 0) {
2055             return ret;
2056         }
2057 
2058         while (meta) {
2059             QCowL2Meta *next = meta->next;
2060 
2061             ret = qcow2_alloc_cluster_link_l2(bs, meta);
2062             if (ret < 0) {
2063                 qcow2_free_any_clusters(bs, meta->alloc_offset,
2064                                         meta->nb_clusters, QCOW2_DISCARD_NEVER);
2065                 return ret;
2066             }
2067 
2068             /* There are no dependent requests, but we need to remove our
2069              * request from the list of in-flight requests */
2070             QLIST_REMOVE(meta, next_in_flight);
2071 
2072             g_free(meta);
2073             meta = next;
2074         }
2075 
2076         /* TODO Preallocate data if requested */
2077 
2078         bytes -= cur_bytes;
2079         offset += cur_bytes;
2080     }
2081 
2082     /*
2083      * It is expected that the image file is large enough to actually contain
2084      * all of the allocated clusters (otherwise we get failing reads after
2085      * EOF). Extend the image to the last allocated sector.
2086      */
2087     if (host_offset != 0) {
2088         uint8_t data = 0;
2089         ret = bdrv_pwrite(bs->file, (host_offset + cur_bytes) - 1,
2090                           &data, 1);
2091         if (ret < 0) {
2092             return ret;
2093         }
2094     }
2095 
2096     return 0;
2097 }
2098 
2099 static int qcow2_create2(const char *filename, int64_t total_size,
2100                          const char *backing_file, const char *backing_format,
2101                          int flags, size_t cluster_size, PreallocMode prealloc,
2102                          QemuOpts *opts, int version, int refcount_order,
2103                          Error **errp)
2104 {
2105     int cluster_bits;
2106     QDict *options;
2107 
2108     /* Calculate cluster_bits */
2109     cluster_bits = ctz32(cluster_size);
2110     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
2111         (1 << cluster_bits) != cluster_size)
2112     {
2113         error_setg(errp, "Cluster size must be a power of two between %d and "
2114                    "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
2115         return -EINVAL;
2116     }
2117 
2118     /*
2119      * Open the image file and write a minimal qcow2 header.
2120      *
2121      * We keep things simple and start with a zero-sized image. We also
2122      * do without refcount blocks or a L1 table for now. We'll fix the
2123      * inconsistency later.
2124      *
2125      * We do need a refcount table because growing the refcount table means
2126      * allocating two new refcount blocks - the seconds of which would be at
2127      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2128      * size for any qcow2 image.
2129      */
2130     BlockBackend *blk;
2131     QCowHeader *header;
2132     uint64_t* refcount_table;
2133     Error *local_err = NULL;
2134     int ret;
2135 
2136     if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
2137         /* Note: The following calculation does not need to be exact; if it is a
2138          * bit off, either some bytes will be "leaked" (which is fine) or we
2139          * will need to increase the file size by some bytes (which is fine,
2140          * too, as long as the bulk is allocated here). Therefore, using
2141          * floating point arithmetic is fine. */
2142         int64_t meta_size = 0;
2143         uint64_t nreftablee, nrefblocke, nl1e, nl2e, refblock_count;
2144         int64_t aligned_total_size = align_offset(total_size, cluster_size);
2145         int refblock_bits, refblock_size;
2146         /* refcount entry size in bytes */
2147         double rces = (1 << refcount_order) / 8.;
2148 
2149         /* see qcow2_open() */
2150         refblock_bits = cluster_bits - (refcount_order - 3);
2151         refblock_size = 1 << refblock_bits;
2152 
2153         /* header: 1 cluster */
2154         meta_size += cluster_size;
2155 
2156         /* total size of L2 tables */
2157         nl2e = aligned_total_size / cluster_size;
2158         nl2e = align_offset(nl2e, cluster_size / sizeof(uint64_t));
2159         meta_size += nl2e * sizeof(uint64_t);
2160 
2161         /* total size of L1 tables */
2162         nl1e = nl2e * sizeof(uint64_t) / cluster_size;
2163         nl1e = align_offset(nl1e, cluster_size / sizeof(uint64_t));
2164         meta_size += nl1e * sizeof(uint64_t);
2165 
2166         /* total size of refcount blocks
2167          *
2168          * note: every host cluster is reference-counted, including metadata
2169          * (even refcount blocks are recursively included).
2170          * Let:
2171          *   a = total_size (this is the guest disk size)
2172          *   m = meta size not including refcount blocks and refcount tables
2173          *   c = cluster size
2174          *   y1 = number of refcount blocks entries
2175          *   y2 = meta size including everything
2176          *   rces = refcount entry size in bytes
2177          * then,
2178          *   y1 = (y2 + a)/c
2179          *   y2 = y1 * rces + y1 * rces * sizeof(u64) / c + m
2180          * we can get y1:
2181          *   y1 = (a + m) / (c - rces - rces * sizeof(u64) / c)
2182          */
2183         nrefblocke = (aligned_total_size + meta_size + cluster_size)
2184                    / (cluster_size - rces - rces * sizeof(uint64_t)
2185                                                  / cluster_size);
2186         refblock_count = DIV_ROUND_UP(nrefblocke, refblock_size);
2187         meta_size += refblock_count * cluster_size;
2188 
2189         /* total size of refcount tables */
2190         nreftablee = align_offset(refblock_count,
2191                                   cluster_size / sizeof(uint64_t));
2192         meta_size += nreftablee * sizeof(uint64_t);
2193 
2194         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
2195                             aligned_total_size + meta_size, &error_abort);
2196         qemu_opt_set(opts, BLOCK_OPT_PREALLOC, PreallocMode_lookup[prealloc],
2197                      &error_abort);
2198     }
2199 
2200     ret = bdrv_create_file(filename, opts, &local_err);
2201     if (ret < 0) {
2202         error_propagate(errp, local_err);
2203         return ret;
2204     }
2205 
2206     blk = blk_new_open(filename, NULL, NULL,
2207                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
2208                        &local_err);
2209     if (blk == NULL) {
2210         error_propagate(errp, local_err);
2211         return -EIO;
2212     }
2213 
2214     blk_set_allow_write_beyond_eof(blk, true);
2215 
2216     /* Write the header */
2217     QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
2218     header = g_malloc0(cluster_size);
2219     *header = (QCowHeader) {
2220         .magic                      = cpu_to_be32(QCOW_MAGIC),
2221         .version                    = cpu_to_be32(version),
2222         .cluster_bits               = cpu_to_be32(cluster_bits),
2223         .size                       = cpu_to_be64(0),
2224         .l1_table_offset            = cpu_to_be64(0),
2225         .l1_size                    = cpu_to_be32(0),
2226         .refcount_table_offset      = cpu_to_be64(cluster_size),
2227         .refcount_table_clusters    = cpu_to_be32(1),
2228         .refcount_order             = cpu_to_be32(refcount_order),
2229         .header_length              = cpu_to_be32(sizeof(*header)),
2230     };
2231 
2232     if (flags & BLOCK_FLAG_ENCRYPT) {
2233         header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
2234     } else {
2235         header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
2236     }
2237 
2238     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
2239         header->compatible_features |=
2240             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
2241     }
2242 
2243     ret = blk_pwrite(blk, 0, header, cluster_size, 0);
2244     g_free(header);
2245     if (ret < 0) {
2246         error_setg_errno(errp, -ret, "Could not write qcow2 header");
2247         goto out;
2248     }
2249 
2250     /* Write a refcount table with one refcount block */
2251     refcount_table = g_malloc0(2 * cluster_size);
2252     refcount_table[0] = cpu_to_be64(2 * cluster_size);
2253     ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0);
2254     g_free(refcount_table);
2255 
2256     if (ret < 0) {
2257         error_setg_errno(errp, -ret, "Could not write refcount table");
2258         goto out;
2259     }
2260 
2261     blk_unref(blk);
2262     blk = NULL;
2263 
2264     /*
2265      * And now open the image and make it consistent first (i.e. increase the
2266      * refcount of the cluster that is occupied by the header and the refcount
2267      * table)
2268      */
2269     options = qdict_new();
2270     qdict_put_str(options, "driver", "qcow2");
2271     blk = blk_new_open(filename, NULL, options,
2272                        BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
2273                        &local_err);
2274     if (blk == NULL) {
2275         error_propagate(errp, local_err);
2276         ret = -EIO;
2277         goto out;
2278     }
2279 
2280     ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
2281     if (ret < 0) {
2282         error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
2283                          "header and refcount table");
2284         goto out;
2285 
2286     } else if (ret != 0) {
2287         error_report("Huh, first cluster in empty image is already in use?");
2288         abort();
2289     }
2290 
2291     /* Create a full header (including things like feature table) */
2292     ret = qcow2_update_header(blk_bs(blk));
2293     if (ret < 0) {
2294         error_setg_errno(errp, -ret, "Could not update qcow2 header");
2295         goto out;
2296     }
2297 
2298     /* Okay, now that we have a valid image, let's give it the right size */
2299     ret = blk_truncate(blk, total_size, errp);
2300     if (ret < 0) {
2301         error_prepend(errp, "Could not resize image: ");
2302         goto out;
2303     }
2304 
2305     /* Want a backing file? There you go.*/
2306     if (backing_file) {
2307         ret = bdrv_change_backing_file(blk_bs(blk), backing_file, backing_format);
2308         if (ret < 0) {
2309             error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
2310                              "with format '%s'", backing_file, backing_format);
2311             goto out;
2312         }
2313     }
2314 
2315     /* And if we're supposed to preallocate metadata, do that now */
2316     if (prealloc != PREALLOC_MODE_OFF) {
2317         BDRVQcow2State *s = blk_bs(blk)->opaque;
2318         qemu_co_mutex_lock(&s->lock);
2319         ret = preallocate(blk_bs(blk));
2320         qemu_co_mutex_unlock(&s->lock);
2321         if (ret < 0) {
2322             error_setg_errno(errp, -ret, "Could not preallocate metadata");
2323             goto out;
2324         }
2325     }
2326 
2327     blk_unref(blk);
2328     blk = NULL;
2329 
2330     /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
2331     options = qdict_new();
2332     qdict_put_str(options, "driver", "qcow2");
2333     blk = blk_new_open(filename, NULL, options,
2334                        BDRV_O_RDWR | BDRV_O_NO_BACKING, &local_err);
2335     if (blk == NULL) {
2336         error_propagate(errp, local_err);
2337         ret = -EIO;
2338         goto out;
2339     }
2340 
2341     ret = 0;
2342 out:
2343     if (blk) {
2344         blk_unref(blk);
2345     }
2346     return ret;
2347 }
2348 
2349 static int qcow2_create(const char *filename, QemuOpts *opts, Error **errp)
2350 {
2351     char *backing_file = NULL;
2352     char *backing_fmt = NULL;
2353     char *buf = NULL;
2354     uint64_t size = 0;
2355     int flags = 0;
2356     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
2357     PreallocMode prealloc;
2358     int version = 3;
2359     uint64_t refcount_bits = 16;
2360     int refcount_order;
2361     Error *local_err = NULL;
2362     int ret;
2363 
2364     /* Read out options */
2365     size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2366                     BDRV_SECTOR_SIZE);
2367     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
2368     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
2369     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
2370         flags |= BLOCK_FLAG_ENCRYPT;
2371     }
2372     cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
2373                                          DEFAULT_CLUSTER_SIZE);
2374     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2375     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
2376                                PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
2377                                &local_err);
2378     if (local_err) {
2379         error_propagate(errp, local_err);
2380         ret = -EINVAL;
2381         goto finish;
2382     }
2383     g_free(buf);
2384     buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
2385     if (!buf) {
2386         /* keep the default */
2387     } else if (!strcmp(buf, "0.10")) {
2388         version = 2;
2389     } else if (!strcmp(buf, "1.1")) {
2390         version = 3;
2391     } else {
2392         error_setg(errp, "Invalid compatibility level: '%s'", buf);
2393         ret = -EINVAL;
2394         goto finish;
2395     }
2396 
2397     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_LAZY_REFCOUNTS, false)) {
2398         flags |= BLOCK_FLAG_LAZY_REFCOUNTS;
2399     }
2400 
2401     if (backing_file && prealloc != PREALLOC_MODE_OFF) {
2402         error_setg(errp, "Backing file and preallocation cannot be used at "
2403                    "the same time");
2404         ret = -EINVAL;
2405         goto finish;
2406     }
2407 
2408     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
2409         error_setg(errp, "Lazy refcounts only supported with compatibility "
2410                    "level 1.1 and above (use compat=1.1 or greater)");
2411         ret = -EINVAL;
2412         goto finish;
2413     }
2414 
2415     refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS,
2416                                             refcount_bits);
2417     if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
2418         error_setg(errp, "Refcount width must be a power of two and may not "
2419                    "exceed 64 bits");
2420         ret = -EINVAL;
2421         goto finish;
2422     }
2423 
2424     if (version < 3 && refcount_bits != 16) {
2425         error_setg(errp, "Different refcount widths than 16 bits require "
2426                    "compatibility level 1.1 or above (use compat=1.1 or "
2427                    "greater)");
2428         ret = -EINVAL;
2429         goto finish;
2430     }
2431 
2432     refcount_order = ctz32(refcount_bits);
2433 
2434     ret = qcow2_create2(filename, size, backing_file, backing_fmt, flags,
2435                         cluster_size, prealloc, opts, version, refcount_order,
2436                         &local_err);
2437     error_propagate(errp, local_err);
2438 
2439 finish:
2440     g_free(backing_file);
2441     g_free(backing_fmt);
2442     g_free(buf);
2443     return ret;
2444 }
2445 
2446 
2447 static bool is_zero_sectors(BlockDriverState *bs, int64_t start,
2448                             uint32_t count)
2449 {
2450     int nr;
2451     BlockDriverState *file;
2452     int64_t res;
2453 
2454     if (start + count > bs->total_sectors) {
2455         count = bs->total_sectors - start;
2456     }
2457 
2458     if (!count) {
2459         return true;
2460     }
2461     res = bdrv_get_block_status_above(bs, NULL, start, count,
2462                                       &nr, &file);
2463     return res >= 0 && (res & BDRV_BLOCK_ZERO) && nr == count;
2464 }
2465 
2466 static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
2467     int64_t offset, int count, BdrvRequestFlags flags)
2468 {
2469     int ret;
2470     BDRVQcow2State *s = bs->opaque;
2471 
2472     uint32_t head = offset % s->cluster_size;
2473     uint32_t tail = (offset + count) % s->cluster_size;
2474 
2475     trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, count);
2476     if (offset + count == bs->total_sectors * BDRV_SECTOR_SIZE) {
2477         tail = 0;
2478     }
2479 
2480     if (head || tail) {
2481         int64_t cl_start = (offset - head) >> BDRV_SECTOR_BITS;
2482         uint64_t off;
2483         unsigned int nr;
2484 
2485         assert(head + count <= s->cluster_size);
2486 
2487         /* check whether remainder of cluster already reads as zero */
2488         if (!(is_zero_sectors(bs, cl_start,
2489                               DIV_ROUND_UP(head, BDRV_SECTOR_SIZE)) &&
2490               is_zero_sectors(bs, (offset + count) >> BDRV_SECTOR_BITS,
2491                               DIV_ROUND_UP(-tail & (s->cluster_size - 1),
2492                                            BDRV_SECTOR_SIZE)))) {
2493             return -ENOTSUP;
2494         }
2495 
2496         qemu_co_mutex_lock(&s->lock);
2497         /* We can have new write after previous check */
2498         offset = cl_start << BDRV_SECTOR_BITS;
2499         count = s->cluster_size;
2500         nr = s->cluster_size;
2501         ret = qcow2_get_cluster_offset(bs, offset, &nr, &off);
2502         if (ret != QCOW2_CLUSTER_UNALLOCATED &&
2503             ret != QCOW2_CLUSTER_ZERO_PLAIN &&
2504             ret != QCOW2_CLUSTER_ZERO_ALLOC) {
2505             qemu_co_mutex_unlock(&s->lock);
2506             return -ENOTSUP;
2507         }
2508     } else {
2509         qemu_co_mutex_lock(&s->lock);
2510     }
2511 
2512     trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, count);
2513 
2514     /* Whatever is left can use real zero clusters */
2515     ret = qcow2_cluster_zeroize(bs, offset, count, flags);
2516     qemu_co_mutex_unlock(&s->lock);
2517 
2518     return ret;
2519 }
2520 
2521 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
2522                                           int64_t offset, int count)
2523 {
2524     int ret;
2525     BDRVQcow2State *s = bs->opaque;
2526 
2527     if (!QEMU_IS_ALIGNED(offset | count, s->cluster_size)) {
2528         assert(count < s->cluster_size);
2529         /* Ignore partial clusters, except for the special case of the
2530          * complete partial cluster at the end of an unaligned file */
2531         if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
2532             offset + count != bs->total_sectors * BDRV_SECTOR_SIZE) {
2533             return -ENOTSUP;
2534         }
2535     }
2536 
2537     qemu_co_mutex_lock(&s->lock);
2538     ret = qcow2_cluster_discard(bs, offset, count, QCOW2_DISCARD_REQUEST,
2539                                 false);
2540     qemu_co_mutex_unlock(&s->lock);
2541     return ret;
2542 }
2543 
2544 static int qcow2_truncate(BlockDriverState *bs, int64_t offset, Error **errp)
2545 {
2546     BDRVQcow2State *s = bs->opaque;
2547     int64_t new_l1_size;
2548     int ret;
2549 
2550     if (offset & 511) {
2551         error_setg(errp, "The new size must be a multiple of 512");
2552         return -EINVAL;
2553     }
2554 
2555     /* cannot proceed if image has snapshots */
2556     if (s->nb_snapshots) {
2557         error_setg(errp, "Can't resize an image which has snapshots");
2558         return -ENOTSUP;
2559     }
2560 
2561     /* shrinking is currently not supported */
2562     if (offset < bs->total_sectors * 512) {
2563         error_setg(errp, "qcow2 doesn't support shrinking images yet");
2564         return -ENOTSUP;
2565     }
2566 
2567     new_l1_size = size_to_l1(s, offset);
2568     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
2569     if (ret < 0) {
2570         error_setg_errno(errp, -ret, "Failed to grow the L1 table");
2571         return ret;
2572     }
2573 
2574     /* write updated header.size */
2575     offset = cpu_to_be64(offset);
2576     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
2577                            &offset, sizeof(uint64_t));
2578     if (ret < 0) {
2579         error_setg_errno(errp, -ret, "Failed to update the image size");
2580         return ret;
2581     }
2582 
2583     s->l1_vm_state_index = new_l1_size;
2584     return 0;
2585 }
2586 
2587 /* XXX: put compressed sectors first, then all the cluster aligned
2588    tables to avoid losing bytes in alignment */
2589 static coroutine_fn int
2590 qcow2_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
2591                             uint64_t bytes, QEMUIOVector *qiov)
2592 {
2593     BDRVQcow2State *s = bs->opaque;
2594     QEMUIOVector hd_qiov;
2595     struct iovec iov;
2596     z_stream strm;
2597     int ret, out_len;
2598     uint8_t *buf, *out_buf;
2599     uint64_t cluster_offset;
2600 
2601     if (bytes == 0) {
2602         /* align end of file to a sector boundary to ease reading with
2603            sector based I/Os */
2604         cluster_offset = bdrv_getlength(bs->file->bs);
2605         return bdrv_truncate(bs->file, cluster_offset, NULL);
2606     }
2607 
2608     buf = qemu_blockalign(bs, s->cluster_size);
2609     if (bytes != s->cluster_size) {
2610         if (bytes > s->cluster_size ||
2611             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
2612         {
2613             qemu_vfree(buf);
2614             return -EINVAL;
2615         }
2616         /* Zero-pad last write if image size is not cluster aligned */
2617         memset(buf + bytes, 0, s->cluster_size - bytes);
2618     }
2619     qemu_iovec_to_buf(qiov, 0, buf, bytes);
2620 
2621     out_buf = g_malloc(s->cluster_size);
2622 
2623     /* best compression, small window, no zlib header */
2624     memset(&strm, 0, sizeof(strm));
2625     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
2626                        Z_DEFLATED, -12,
2627                        9, Z_DEFAULT_STRATEGY);
2628     if (ret != 0) {
2629         ret = -EINVAL;
2630         goto fail;
2631     }
2632 
2633     strm.avail_in = s->cluster_size;
2634     strm.next_in = (uint8_t *)buf;
2635     strm.avail_out = s->cluster_size;
2636     strm.next_out = out_buf;
2637 
2638     ret = deflate(&strm, Z_FINISH);
2639     if (ret != Z_STREAM_END && ret != Z_OK) {
2640         deflateEnd(&strm);
2641         ret = -EINVAL;
2642         goto fail;
2643     }
2644     out_len = strm.next_out - out_buf;
2645 
2646     deflateEnd(&strm);
2647 
2648     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
2649         /* could not compress: write normal cluster */
2650         ret = qcow2_co_pwritev(bs, offset, bytes, qiov, 0);
2651         if (ret < 0) {
2652             goto fail;
2653         }
2654         goto success;
2655     }
2656 
2657     qemu_co_mutex_lock(&s->lock);
2658     cluster_offset =
2659         qcow2_alloc_compressed_cluster_offset(bs, offset, out_len);
2660     if (!cluster_offset) {
2661         qemu_co_mutex_unlock(&s->lock);
2662         ret = -EIO;
2663         goto fail;
2664     }
2665     cluster_offset &= s->cluster_offset_mask;
2666 
2667     ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
2668     qemu_co_mutex_unlock(&s->lock);
2669     if (ret < 0) {
2670         goto fail;
2671     }
2672 
2673     iov = (struct iovec) {
2674         .iov_base   = out_buf,
2675         .iov_len    = out_len,
2676     };
2677     qemu_iovec_init_external(&hd_qiov, &iov, 1);
2678 
2679     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
2680     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
2681     if (ret < 0) {
2682         goto fail;
2683     }
2684 success:
2685     ret = 0;
2686 fail:
2687     qemu_vfree(buf);
2688     g_free(out_buf);
2689     return ret;
2690 }
2691 
2692 static int make_completely_empty(BlockDriverState *bs)
2693 {
2694     BDRVQcow2State *s = bs->opaque;
2695     Error *local_err = NULL;
2696     int ret, l1_clusters;
2697     int64_t offset;
2698     uint64_t *new_reftable = NULL;
2699     uint64_t rt_entry, l1_size2;
2700     struct {
2701         uint64_t l1_offset;
2702         uint64_t reftable_offset;
2703         uint32_t reftable_clusters;
2704     } QEMU_PACKED l1_ofs_rt_ofs_cls;
2705 
2706     ret = qcow2_cache_empty(bs, s->l2_table_cache);
2707     if (ret < 0) {
2708         goto fail;
2709     }
2710 
2711     ret = qcow2_cache_empty(bs, s->refcount_block_cache);
2712     if (ret < 0) {
2713         goto fail;
2714     }
2715 
2716     /* Refcounts will be broken utterly */
2717     ret = qcow2_mark_dirty(bs);
2718     if (ret < 0) {
2719         goto fail;
2720     }
2721 
2722     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2723 
2724     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2725     l1_size2 = (uint64_t)s->l1_size * sizeof(uint64_t);
2726 
2727     /* After this call, neither the in-memory nor the on-disk refcount
2728      * information accurately describe the actual references */
2729 
2730     ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
2731                              l1_clusters * s->cluster_size, 0);
2732     if (ret < 0) {
2733         goto fail_broken_refcounts;
2734     }
2735     memset(s->l1_table, 0, l1_size2);
2736 
2737     BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
2738 
2739     /* Overwrite enough clusters at the beginning of the sectors to place
2740      * the refcount table, a refcount block and the L1 table in; this may
2741      * overwrite parts of the existing refcount and L1 table, which is not
2742      * an issue because the dirty flag is set, complete data loss is in fact
2743      * desired and partial data loss is consequently fine as well */
2744     ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
2745                              (2 + l1_clusters) * s->cluster_size, 0);
2746     /* This call (even if it failed overall) may have overwritten on-disk
2747      * refcount structures; in that case, the in-memory refcount information
2748      * will probably differ from the on-disk information which makes the BDS
2749      * unusable */
2750     if (ret < 0) {
2751         goto fail_broken_refcounts;
2752     }
2753 
2754     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
2755     BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
2756 
2757     /* "Create" an empty reftable (one cluster) directly after the image
2758      * header and an empty L1 table three clusters after the image header;
2759      * the cluster between those two will be used as the first refblock */
2760     l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
2761     l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
2762     l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
2763     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
2764                            &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls));
2765     if (ret < 0) {
2766         goto fail_broken_refcounts;
2767     }
2768 
2769     s->l1_table_offset = 3 * s->cluster_size;
2770 
2771     new_reftable = g_try_new0(uint64_t, s->cluster_size / sizeof(uint64_t));
2772     if (!new_reftable) {
2773         ret = -ENOMEM;
2774         goto fail_broken_refcounts;
2775     }
2776 
2777     s->refcount_table_offset = s->cluster_size;
2778     s->refcount_table_size   = s->cluster_size / sizeof(uint64_t);
2779     s->max_refcount_table_index = 0;
2780 
2781     g_free(s->refcount_table);
2782     s->refcount_table = new_reftable;
2783     new_reftable = NULL;
2784 
2785     /* Now the in-memory refcount information again corresponds to the on-disk
2786      * information (reftable is empty and no refblocks (the refblock cache is
2787      * empty)); however, this means some clusters (e.g. the image header) are
2788      * referenced, but not refcounted, but the normal qcow2 code assumes that
2789      * the in-memory information is always correct */
2790 
2791     BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
2792 
2793     /* Enter the first refblock into the reftable */
2794     rt_entry = cpu_to_be64(2 * s->cluster_size);
2795     ret = bdrv_pwrite_sync(bs->file, s->cluster_size,
2796                            &rt_entry, sizeof(rt_entry));
2797     if (ret < 0) {
2798         goto fail_broken_refcounts;
2799     }
2800     s->refcount_table[0] = 2 * s->cluster_size;
2801 
2802     s->free_cluster_index = 0;
2803     assert(3 + l1_clusters <= s->refcount_block_size);
2804     offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
2805     if (offset < 0) {
2806         ret = offset;
2807         goto fail_broken_refcounts;
2808     } else if (offset > 0) {
2809         error_report("First cluster in emptied image is in use");
2810         abort();
2811     }
2812 
2813     /* Now finally the in-memory information corresponds to the on-disk
2814      * structures and is correct */
2815     ret = qcow2_mark_clean(bs);
2816     if (ret < 0) {
2817         goto fail;
2818     }
2819 
2820     ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size,
2821                         &local_err);
2822     if (ret < 0) {
2823         error_report_err(local_err);
2824         goto fail;
2825     }
2826 
2827     return 0;
2828 
2829 fail_broken_refcounts:
2830     /* The BDS is unusable at this point. If we wanted to make it usable, we
2831      * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
2832      * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
2833      * again. However, because the functions which could have caused this error
2834      * path to be taken are used by those functions as well, it's very likely
2835      * that that sequence will fail as well. Therefore, just eject the BDS. */
2836     bs->drv = NULL;
2837 
2838 fail:
2839     g_free(new_reftable);
2840     return ret;
2841 }
2842 
2843 static int qcow2_make_empty(BlockDriverState *bs)
2844 {
2845     BDRVQcow2State *s = bs->opaque;
2846     uint64_t offset, end_offset;
2847     int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
2848     int l1_clusters, ret = 0;
2849 
2850     l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / sizeof(uint64_t));
2851 
2852     if (s->qcow_version >= 3 && !s->snapshots &&
2853         3 + l1_clusters <= s->refcount_block_size) {
2854         /* The following function only works for qcow2 v3 images (it requires
2855          * the dirty flag) and only as long as there are no snapshots (because
2856          * it completely empties the image). Furthermore, the L1 table and three
2857          * additional clusters (image header, refcount table, one refcount
2858          * block) have to fit inside one refcount block. */
2859         return make_completely_empty(bs);
2860     }
2861 
2862     /* This fallback code simply discards every active cluster; this is slow,
2863      * but works in all cases */
2864     end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
2865     for (offset = 0; offset < end_offset; offset += step) {
2866         /* As this function is generally used after committing an external
2867          * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
2868          * default action for this kind of discard is to pass the discard,
2869          * which will ideally result in an actually smaller image file, as
2870          * is probably desired. */
2871         ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
2872                                     QCOW2_DISCARD_SNAPSHOT, true);
2873         if (ret < 0) {
2874             break;
2875         }
2876     }
2877 
2878     return ret;
2879 }
2880 
2881 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
2882 {
2883     BDRVQcow2State *s = bs->opaque;
2884     int ret;
2885 
2886     qemu_co_mutex_lock(&s->lock);
2887     ret = qcow2_cache_write(bs, s->l2_table_cache);
2888     if (ret < 0) {
2889         qemu_co_mutex_unlock(&s->lock);
2890         return ret;
2891     }
2892 
2893     if (qcow2_need_accurate_refcounts(s)) {
2894         ret = qcow2_cache_write(bs, s->refcount_block_cache);
2895         if (ret < 0) {
2896             qemu_co_mutex_unlock(&s->lock);
2897             return ret;
2898         }
2899     }
2900     qemu_co_mutex_unlock(&s->lock);
2901 
2902     return 0;
2903 }
2904 
2905 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2906 {
2907     BDRVQcow2State *s = bs->opaque;
2908     bdi->unallocated_blocks_are_zero = true;
2909     bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
2910     bdi->cluster_size = s->cluster_size;
2911     bdi->vm_state_offset = qcow2_vm_state_offset(s);
2912     return 0;
2913 }
2914 
2915 static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
2916 {
2917     BDRVQcow2State *s = bs->opaque;
2918     ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
2919 
2920     *spec_info = (ImageInfoSpecific){
2921         .type  = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
2922         .u.qcow2.data = g_new(ImageInfoSpecificQCow2, 1),
2923     };
2924     if (s->qcow_version == 2) {
2925         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
2926             .compat             = g_strdup("0.10"),
2927             .refcount_bits      = s->refcount_bits,
2928         };
2929     } else if (s->qcow_version == 3) {
2930         *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
2931             .compat             = g_strdup("1.1"),
2932             .lazy_refcounts     = s->compatible_features &
2933                                   QCOW2_COMPAT_LAZY_REFCOUNTS,
2934             .has_lazy_refcounts = true,
2935             .corrupt            = s->incompatible_features &
2936                                   QCOW2_INCOMPAT_CORRUPT,
2937             .has_corrupt        = true,
2938             .refcount_bits      = s->refcount_bits,
2939         };
2940     } else {
2941         /* if this assertion fails, this probably means a new version was
2942          * added without having it covered here */
2943         assert(false);
2944     }
2945 
2946     return spec_info;
2947 }
2948 
2949 #if 0
2950 static void dump_refcounts(BlockDriverState *bs)
2951 {
2952     BDRVQcow2State *s = bs->opaque;
2953     int64_t nb_clusters, k, k1, size;
2954     int refcount;
2955 
2956     size = bdrv_getlength(bs->file->bs);
2957     nb_clusters = size_to_clusters(s, size);
2958     for(k = 0; k < nb_clusters;) {
2959         k1 = k;
2960         refcount = get_refcount(bs, k);
2961         k++;
2962         while (k < nb_clusters && get_refcount(bs, k) == refcount)
2963             k++;
2964         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
2965                k - k1);
2966     }
2967 }
2968 #endif
2969 
2970 static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2971                               int64_t pos)
2972 {
2973     BDRVQcow2State *s = bs->opaque;
2974 
2975     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
2976     return bs->drv->bdrv_co_pwritev(bs, qcow2_vm_state_offset(s) + pos,
2977                                     qiov->size, qiov, 0);
2978 }
2979 
2980 static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
2981                               int64_t pos)
2982 {
2983     BDRVQcow2State *s = bs->opaque;
2984 
2985     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
2986     return bs->drv->bdrv_co_preadv(bs, qcow2_vm_state_offset(s) + pos,
2987                                    qiov->size, qiov, 0);
2988 }
2989 
2990 /*
2991  * Downgrades an image's version. To achieve this, any incompatible features
2992  * have to be removed.
2993  */
2994 static int qcow2_downgrade(BlockDriverState *bs, int target_version,
2995                            BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
2996 {
2997     BDRVQcow2State *s = bs->opaque;
2998     int current_version = s->qcow_version;
2999     int ret;
3000 
3001     if (target_version == current_version) {
3002         return 0;
3003     } else if (target_version > current_version) {
3004         return -EINVAL;
3005     } else if (target_version != 2) {
3006         return -EINVAL;
3007     }
3008 
3009     if (s->refcount_order != 4) {
3010         error_report("compat=0.10 requires refcount_bits=16");
3011         return -ENOTSUP;
3012     }
3013 
3014     /* clear incompatible features */
3015     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
3016         ret = qcow2_mark_clean(bs);
3017         if (ret < 0) {
3018             return ret;
3019         }
3020     }
3021 
3022     /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
3023      * the first place; if that happens nonetheless, returning -ENOTSUP is the
3024      * best thing to do anyway */
3025 
3026     if (s->incompatible_features) {
3027         return -ENOTSUP;
3028     }
3029 
3030     /* since we can ignore compatible features, we can set them to 0 as well */
3031     s->compatible_features = 0;
3032     /* if lazy refcounts have been used, they have already been fixed through
3033      * clearing the dirty flag */
3034 
3035     /* clearing autoclear features is trivial */
3036     s->autoclear_features = 0;
3037 
3038     ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
3039     if (ret < 0) {
3040         return ret;
3041     }
3042 
3043     s->qcow_version = target_version;
3044     ret = qcow2_update_header(bs);
3045     if (ret < 0) {
3046         s->qcow_version = current_version;
3047         return ret;
3048     }
3049     return 0;
3050 }
3051 
3052 typedef enum Qcow2AmendOperation {
3053     /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
3054      * statically initialized to so that the helper CB can discern the first
3055      * invocation from an operation change */
3056     QCOW2_NO_OPERATION = 0,
3057 
3058     QCOW2_CHANGING_REFCOUNT_ORDER,
3059     QCOW2_DOWNGRADING,
3060 } Qcow2AmendOperation;
3061 
3062 typedef struct Qcow2AmendHelperCBInfo {
3063     /* The code coordinating the amend operations should only modify
3064      * these four fields; the rest will be managed by the CB */
3065     BlockDriverAmendStatusCB *original_status_cb;
3066     void *original_cb_opaque;
3067 
3068     Qcow2AmendOperation current_operation;
3069 
3070     /* Total number of operations to perform (only set once) */
3071     int total_operations;
3072 
3073     /* The following fields are managed by the CB */
3074 
3075     /* Number of operations completed */
3076     int operations_completed;
3077 
3078     /* Cumulative offset of all completed operations */
3079     int64_t offset_completed;
3080 
3081     Qcow2AmendOperation last_operation;
3082     int64_t last_work_size;
3083 } Qcow2AmendHelperCBInfo;
3084 
3085 static void qcow2_amend_helper_cb(BlockDriverState *bs,
3086                                   int64_t operation_offset,
3087                                   int64_t operation_work_size, void *opaque)
3088 {
3089     Qcow2AmendHelperCBInfo *info = opaque;
3090     int64_t current_work_size;
3091     int64_t projected_work_size;
3092 
3093     if (info->current_operation != info->last_operation) {
3094         if (info->last_operation != QCOW2_NO_OPERATION) {
3095             info->offset_completed += info->last_work_size;
3096             info->operations_completed++;
3097         }
3098 
3099         info->last_operation = info->current_operation;
3100     }
3101 
3102     assert(info->total_operations > 0);
3103     assert(info->operations_completed < info->total_operations);
3104 
3105     info->last_work_size = operation_work_size;
3106 
3107     current_work_size = info->offset_completed + operation_work_size;
3108 
3109     /* current_work_size is the total work size for (operations_completed + 1)
3110      * operations (which includes this one), so multiply it by the number of
3111      * operations not covered and divide it by the number of operations
3112      * covered to get a projection for the operations not covered */
3113     projected_work_size = current_work_size * (info->total_operations -
3114                                                info->operations_completed - 1)
3115                                             / (info->operations_completed + 1);
3116 
3117     info->original_status_cb(bs, info->offset_completed + operation_offset,
3118                              current_work_size + projected_work_size,
3119                              info->original_cb_opaque);
3120 }
3121 
3122 static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
3123                                BlockDriverAmendStatusCB *status_cb,
3124                                void *cb_opaque)
3125 {
3126     BDRVQcow2State *s = bs->opaque;
3127     int old_version = s->qcow_version, new_version = old_version;
3128     uint64_t new_size = 0;
3129     const char *backing_file = NULL, *backing_format = NULL;
3130     bool lazy_refcounts = s->use_lazy_refcounts;
3131     const char *compat = NULL;
3132     uint64_t cluster_size = s->cluster_size;
3133     bool encrypt;
3134     int refcount_bits = s->refcount_bits;
3135     Error *local_err = NULL;
3136     int ret;
3137     QemuOptDesc *desc = opts->list->desc;
3138     Qcow2AmendHelperCBInfo helper_cb_info;
3139 
3140     while (desc && desc->name) {
3141         if (!qemu_opt_find(opts, desc->name)) {
3142             /* only change explicitly defined options */
3143             desc++;
3144             continue;
3145         }
3146 
3147         if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
3148             compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
3149             if (!compat) {
3150                 /* preserve default */
3151             } else if (!strcmp(compat, "0.10")) {
3152                 new_version = 2;
3153             } else if (!strcmp(compat, "1.1")) {
3154                 new_version = 3;
3155             } else {
3156                 error_report("Unknown compatibility level %s", compat);
3157                 return -EINVAL;
3158             }
3159         } else if (!strcmp(desc->name, BLOCK_OPT_PREALLOC)) {
3160             error_report("Cannot change preallocation mode");
3161             return -ENOTSUP;
3162         } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
3163             new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
3164         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
3165             backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
3166         } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
3167             backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
3168         } else if (!strcmp(desc->name, BLOCK_OPT_ENCRYPT)) {
3169             encrypt = qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT,
3170                                         !!s->cipher);
3171 
3172             if (encrypt != !!s->cipher) {
3173                 error_report("Changing the encryption flag is not supported");
3174                 return -ENOTSUP;
3175             }
3176         } else if (!strcmp(desc->name, BLOCK_OPT_CLUSTER_SIZE)) {
3177             cluster_size = qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE,
3178                                              cluster_size);
3179             if (cluster_size != s->cluster_size) {
3180                 error_report("Changing the cluster size is not supported");
3181                 return -ENOTSUP;
3182             }
3183         } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
3184             lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
3185                                                lazy_refcounts);
3186         } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
3187             refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
3188                                                 refcount_bits);
3189 
3190             if (refcount_bits <= 0 || refcount_bits > 64 ||
3191                 !is_power_of_2(refcount_bits))
3192             {
3193                 error_report("Refcount width must be a power of two and may "
3194                              "not exceed 64 bits");
3195                 return -EINVAL;
3196             }
3197         } else {
3198             /* if this point is reached, this probably means a new option was
3199              * added without having it covered here */
3200             abort();
3201         }
3202 
3203         desc++;
3204     }
3205 
3206     helper_cb_info = (Qcow2AmendHelperCBInfo){
3207         .original_status_cb = status_cb,
3208         .original_cb_opaque = cb_opaque,
3209         .total_operations = (new_version < old_version)
3210                           + (s->refcount_bits != refcount_bits)
3211     };
3212 
3213     /* Upgrade first (some features may require compat=1.1) */
3214     if (new_version > old_version) {
3215         s->qcow_version = new_version;
3216         ret = qcow2_update_header(bs);
3217         if (ret < 0) {
3218             s->qcow_version = old_version;
3219             return ret;
3220         }
3221     }
3222 
3223     if (s->refcount_bits != refcount_bits) {
3224         int refcount_order = ctz32(refcount_bits);
3225         Error *local_error = NULL;
3226 
3227         if (new_version < 3 && refcount_bits != 16) {
3228             error_report("Different refcount widths than 16 bits require "
3229                          "compatibility level 1.1 or above (use compat=1.1 or "
3230                          "greater)");
3231             return -EINVAL;
3232         }
3233 
3234         helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
3235         ret = qcow2_change_refcount_order(bs, refcount_order,
3236                                           &qcow2_amend_helper_cb,
3237                                           &helper_cb_info, &local_error);
3238         if (ret < 0) {
3239             error_report_err(local_error);
3240             return ret;
3241         }
3242     }
3243 
3244     if (backing_file || backing_format) {
3245         ret = qcow2_change_backing_file(bs,
3246                     backing_file ?: s->image_backing_file,
3247                     backing_format ?: s->image_backing_format);
3248         if (ret < 0) {
3249             return ret;
3250         }
3251     }
3252 
3253     if (s->use_lazy_refcounts != lazy_refcounts) {
3254         if (lazy_refcounts) {
3255             if (new_version < 3) {
3256                 error_report("Lazy refcounts only supported with compatibility "
3257                              "level 1.1 and above (use compat=1.1 or greater)");
3258                 return -EINVAL;
3259             }
3260             s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3261             ret = qcow2_update_header(bs);
3262             if (ret < 0) {
3263                 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3264                 return ret;
3265             }
3266             s->use_lazy_refcounts = true;
3267         } else {
3268             /* make image clean first */
3269             ret = qcow2_mark_clean(bs);
3270             if (ret < 0) {
3271                 return ret;
3272             }
3273             /* now disallow lazy refcounts */
3274             s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
3275             ret = qcow2_update_header(bs);
3276             if (ret < 0) {
3277                 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
3278                 return ret;
3279             }
3280             s->use_lazy_refcounts = false;
3281         }
3282     }
3283 
3284     if (new_size) {
3285         BlockBackend *blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
3286         ret = blk_insert_bs(blk, bs, &local_err);
3287         if (ret < 0) {
3288             error_report_err(local_err);
3289             blk_unref(blk);
3290             return ret;
3291         }
3292 
3293         ret = blk_truncate(blk, new_size, &local_err);
3294         blk_unref(blk);
3295         if (ret < 0) {
3296             error_report_err(local_err);
3297             return ret;
3298         }
3299     }
3300 
3301     /* Downgrade last (so unsupported features can be removed before) */
3302     if (new_version < old_version) {
3303         helper_cb_info.current_operation = QCOW2_DOWNGRADING;
3304         ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
3305                               &helper_cb_info);
3306         if (ret < 0) {
3307             return ret;
3308         }
3309     }
3310 
3311     return 0;
3312 }
3313 
3314 /*
3315  * If offset or size are negative, respectively, they will not be included in
3316  * the BLOCK_IMAGE_CORRUPTED event emitted.
3317  * fatal will be ignored for read-only BDS; corruptions found there will always
3318  * be considered non-fatal.
3319  */
3320 void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
3321                              int64_t size, const char *message_format, ...)
3322 {
3323     BDRVQcow2State *s = bs->opaque;
3324     const char *node_name;
3325     char *message;
3326     va_list ap;
3327 
3328     fatal = fatal && !bs->read_only;
3329 
3330     if (s->signaled_corruption &&
3331         (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
3332     {
3333         return;
3334     }
3335 
3336     va_start(ap, message_format);
3337     message = g_strdup_vprintf(message_format, ap);
3338     va_end(ap);
3339 
3340     if (fatal) {
3341         fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
3342                 "corruption events will be suppressed\n", message);
3343     } else {
3344         fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
3345                 "corruption events will be suppressed\n", message);
3346     }
3347 
3348     node_name = bdrv_get_node_name(bs);
3349     qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
3350                                           *node_name != '\0', node_name,
3351                                           message, offset >= 0, offset,
3352                                           size >= 0, size,
3353                                           fatal, &error_abort);
3354     g_free(message);
3355 
3356     if (fatal) {
3357         qcow2_mark_corrupt(bs);
3358         bs->drv = NULL; /* make BDS unusable */
3359     }
3360 
3361     s->signaled_corruption = true;
3362 }
3363 
3364 static QemuOptsList qcow2_create_opts = {
3365     .name = "qcow2-create-opts",
3366     .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
3367     .desc = {
3368         {
3369             .name = BLOCK_OPT_SIZE,
3370             .type = QEMU_OPT_SIZE,
3371             .help = "Virtual disk size"
3372         },
3373         {
3374             .name = BLOCK_OPT_COMPAT_LEVEL,
3375             .type = QEMU_OPT_STRING,
3376             .help = "Compatibility level (0.10 or 1.1)"
3377         },
3378         {
3379             .name = BLOCK_OPT_BACKING_FILE,
3380             .type = QEMU_OPT_STRING,
3381             .help = "File name of a base image"
3382         },
3383         {
3384             .name = BLOCK_OPT_BACKING_FMT,
3385             .type = QEMU_OPT_STRING,
3386             .help = "Image format of the base image"
3387         },
3388         {
3389             .name = BLOCK_OPT_ENCRYPT,
3390             .type = QEMU_OPT_BOOL,
3391             .help = "Encrypt the image",
3392             .def_value_str = "off"
3393         },
3394         {
3395             .name = BLOCK_OPT_CLUSTER_SIZE,
3396             .type = QEMU_OPT_SIZE,
3397             .help = "qcow2 cluster size",
3398             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
3399         },
3400         {
3401             .name = BLOCK_OPT_PREALLOC,
3402             .type = QEMU_OPT_STRING,
3403             .help = "Preallocation mode (allowed values: off, metadata, "
3404                     "falloc, full)"
3405         },
3406         {
3407             .name = BLOCK_OPT_LAZY_REFCOUNTS,
3408             .type = QEMU_OPT_BOOL,
3409             .help = "Postpone refcount updates",
3410             .def_value_str = "off"
3411         },
3412         {
3413             .name = BLOCK_OPT_REFCOUNT_BITS,
3414             .type = QEMU_OPT_NUMBER,
3415             .help = "Width of a reference count entry in bits",
3416             .def_value_str = "16"
3417         },
3418         { /* end of list */ }
3419     }
3420 };
3421 
3422 BlockDriver bdrv_qcow2 = {
3423     .format_name        = "qcow2",
3424     .instance_size      = sizeof(BDRVQcow2State),
3425     .bdrv_probe         = qcow2_probe,
3426     .bdrv_open          = qcow2_open,
3427     .bdrv_close         = qcow2_close,
3428     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
3429     .bdrv_reopen_commit   = qcow2_reopen_commit,
3430     .bdrv_reopen_abort    = qcow2_reopen_abort,
3431     .bdrv_join_options    = qcow2_join_options,
3432     .bdrv_child_perm      = bdrv_format_default_perms,
3433     .bdrv_create        = qcow2_create,
3434     .bdrv_has_zero_init = bdrv_has_zero_init_1,
3435     .bdrv_co_get_block_status = qcow2_co_get_block_status,
3436     .bdrv_set_key       = qcow2_set_key,
3437 
3438     .bdrv_co_preadv         = qcow2_co_preadv,
3439     .bdrv_co_pwritev        = qcow2_co_pwritev,
3440     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
3441 
3442     .bdrv_co_pwrite_zeroes  = qcow2_co_pwrite_zeroes,
3443     .bdrv_co_pdiscard       = qcow2_co_pdiscard,
3444     .bdrv_truncate          = qcow2_truncate,
3445     .bdrv_co_pwritev_compressed = qcow2_co_pwritev_compressed,
3446     .bdrv_make_empty        = qcow2_make_empty,
3447 
3448     .bdrv_snapshot_create   = qcow2_snapshot_create,
3449     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
3450     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
3451     .bdrv_snapshot_list     = qcow2_snapshot_list,
3452     .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
3453     .bdrv_get_info          = qcow2_get_info,
3454     .bdrv_get_specific_info = qcow2_get_specific_info,
3455 
3456     .bdrv_save_vmstate    = qcow2_save_vmstate,
3457     .bdrv_load_vmstate    = qcow2_load_vmstate,
3458 
3459     .supports_backing           = true,
3460     .bdrv_change_backing_file   = qcow2_change_backing_file,
3461 
3462     .bdrv_refresh_limits        = qcow2_refresh_limits,
3463     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
3464     .bdrv_inactivate            = qcow2_inactivate,
3465 
3466     .create_opts         = &qcow2_create_opts,
3467     .bdrv_check          = qcow2_check,
3468     .bdrv_amend_options  = qcow2_amend_options,
3469 
3470     .bdrv_detach_aio_context  = qcow2_detach_aio_context,
3471     .bdrv_attach_aio_context  = qcow2_attach_aio_context,
3472 };
3473 
3474 static void bdrv_qcow2_init(void)
3475 {
3476     bdrv_register(&bdrv_qcow2);
3477 }
3478 
3479 block_init(bdrv_qcow2_init);
3480