xref: /openbmc/qemu/block/qcow2.c (revision b0d62a3d8e1ea9eccd246829a03e2d23f0982431)
1 /*
2  * Block driver for the QCOW version 2 format
3  *
4  * Copyright (c) 2004-2006 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 #include "qemu-common.h"
25 #include "block/block_int.h"
26 #include "qemu/module.h"
27 #include <zlib.h>
28 #include "block/aes.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 "trace.h"
34 
35 /*
36   Differences with QCOW:
37 
38   - Support for multiple incremental snapshots.
39   - Memory management by reference counts.
40   - Clusters which have a reference count of one have the bit
41     QCOW_OFLAG_COPIED to optimize write performance.
42   - Size of compressed clusters is stored in sectors to reduce bit usage
43     in the cluster offsets.
44   - Support for storing additional data (such as the VM state) in the
45     snapshots.
46   - If a backing store is used, the cluster size is not constrained
47     (could be backported to QCOW).
48   - L2 tables have always a size of one cluster.
49 */
50 
51 
52 typedef struct {
53     uint32_t magic;
54     uint32_t len;
55 } QCowExtension;
56 
57 #define  QCOW2_EXT_MAGIC_END 0
58 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
59 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
60 
61 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
62 {
63     const QCowHeader *cow_header = (const void *)buf;
64 
65     if (buf_size >= sizeof(QCowHeader) &&
66         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
67         be32_to_cpu(cow_header->version) >= 2)
68         return 100;
69     else
70         return 0;
71 }
72 
73 
74 /*
75  * read qcow2 extension and fill bs
76  * start reading from start_offset
77  * finish reading upon magic of value 0 or when end_offset reached
78  * unknown magic is skipped (future extension this version knows nothing about)
79  * return 0 upon success, non-0 otherwise
80  */
81 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
82                                  uint64_t end_offset, void **p_feature_table)
83 {
84     BDRVQcowState *s = bs->opaque;
85     QCowExtension ext;
86     uint64_t offset;
87     int ret;
88 
89 #ifdef DEBUG_EXT
90     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
91 #endif
92     offset = start_offset;
93     while (offset < end_offset) {
94 
95 #ifdef DEBUG_EXT
96         /* Sanity check */
97         if (offset > s->cluster_size)
98             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
99 
100         printf("attempting to read extended header in offset %lu\n", offset);
101 #endif
102 
103         if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) {
104             fprintf(stderr, "qcow2_read_extension: ERROR: "
105                     "pread fail from offset %" PRIu64 "\n",
106                     offset);
107             return 1;
108         }
109         be32_to_cpus(&ext.magic);
110         be32_to_cpus(&ext.len);
111         offset += sizeof(ext);
112 #ifdef DEBUG_EXT
113         printf("ext.magic = 0x%x\n", ext.magic);
114 #endif
115         if (ext.len > end_offset - offset) {
116             error_report("Header extension too large");
117             return -EINVAL;
118         }
119 
120         switch (ext.magic) {
121         case QCOW2_EXT_MAGIC_END:
122             return 0;
123 
124         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
125             if (ext.len >= sizeof(bs->backing_format)) {
126                 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
127                         " (>=%zu)\n",
128                         ext.len, sizeof(bs->backing_format));
129                 return 2;
130             }
131             if (bdrv_pread(bs->file, offset , bs->backing_format,
132                            ext.len) != ext.len)
133                 return 3;
134             bs->backing_format[ext.len] = '\0';
135 #ifdef DEBUG_EXT
136             printf("Qcow2: Got format extension %s\n", bs->backing_format);
137 #endif
138             break;
139 
140         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
141             if (p_feature_table != NULL) {
142                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
143                 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
144                 if (ret < 0) {
145                     return ret;
146                 }
147 
148                 *p_feature_table = feature_table;
149             }
150             break;
151 
152         default:
153             /* unknown magic - save it in case we need to rewrite the header */
154             {
155                 Qcow2UnknownHeaderExtension *uext;
156 
157                 uext = g_malloc0(sizeof(*uext)  + ext.len);
158                 uext->magic = ext.magic;
159                 uext->len = ext.len;
160                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
161 
162                 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
163                 if (ret < 0) {
164                     return ret;
165                 }
166             }
167             break;
168         }
169 
170         offset += ((ext.len + 7) & ~7);
171     }
172 
173     return 0;
174 }
175 
176 static void cleanup_unknown_header_ext(BlockDriverState *bs)
177 {
178     BDRVQcowState *s = bs->opaque;
179     Qcow2UnknownHeaderExtension *uext, *next;
180 
181     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
182         QLIST_REMOVE(uext, next);
183         g_free(uext);
184     }
185 }
186 
187 static void GCC_FMT_ATTR(2, 3) report_unsupported(BlockDriverState *bs,
188     const char *fmt, ...)
189 {
190     char msg[64];
191     va_list ap;
192 
193     va_start(ap, fmt);
194     vsnprintf(msg, sizeof(msg), fmt, ap);
195     va_end(ap);
196 
197     qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
198         bs->device_name, "qcow2", msg);
199 }
200 
201 static void report_unsupported_feature(BlockDriverState *bs,
202     Qcow2Feature *table, uint64_t mask)
203 {
204     while (table && table->name[0] != '\0') {
205         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
206             if (mask & (1 << table->bit)) {
207                 report_unsupported(bs, "%.46s",table->name);
208                 mask &= ~(1 << table->bit);
209             }
210         }
211         table++;
212     }
213 
214     if (mask) {
215         report_unsupported(bs, "Unknown incompatible feature: %" PRIx64, mask);
216     }
217 }
218 
219 /*
220  * Sets the dirty bit and flushes afterwards if necessary.
221  *
222  * The incompatible_features bit is only set if the image file header was
223  * updated successfully.  Therefore it is not required to check the return
224  * value of this function.
225  */
226 int qcow2_mark_dirty(BlockDriverState *bs)
227 {
228     BDRVQcowState *s = bs->opaque;
229     uint64_t val;
230     int ret;
231 
232     assert(s->qcow_version >= 3);
233 
234     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
235         return 0; /* already dirty */
236     }
237 
238     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
239     ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
240                       &val, sizeof(val));
241     if (ret < 0) {
242         return ret;
243     }
244     ret = bdrv_flush(bs->file);
245     if (ret < 0) {
246         return ret;
247     }
248 
249     /* Only treat image as dirty if the header was updated successfully */
250     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
251     return 0;
252 }
253 
254 /*
255  * Clears the dirty bit and flushes before if necessary.  Only call this
256  * function when there are no pending requests, it does not guard against
257  * concurrent requests dirtying the image.
258  */
259 static int qcow2_mark_clean(BlockDriverState *bs)
260 {
261     BDRVQcowState *s = bs->opaque;
262 
263     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
264         int ret = bdrv_flush(bs);
265         if (ret < 0) {
266             return ret;
267         }
268 
269         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
270         return qcow2_update_header(bs);
271     }
272     return 0;
273 }
274 
275 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
276                        BdrvCheckMode fix)
277 {
278     int ret = qcow2_check_refcounts(bs, result, fix);
279     if (ret < 0) {
280         return ret;
281     }
282 
283     if (fix && result->check_errors == 0 && result->corruptions == 0) {
284         return qcow2_mark_clean(bs);
285     }
286     return ret;
287 }
288 
289 static QemuOptsList qcow2_runtime_opts = {
290     .name = "qcow2",
291     .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
292     .desc = {
293         {
294             .name = "lazy_refcounts",
295             .type = QEMU_OPT_BOOL,
296             .help = "Postpone refcount updates",
297         },
298         { /* end of list */ }
299     },
300 };
301 
302 static int qcow2_open(BlockDriverState *bs, QDict *options, int flags)
303 {
304     BDRVQcowState *s = bs->opaque;
305     int len, i, ret = 0;
306     QCowHeader header;
307     QemuOpts *opts;
308     Error *local_err = NULL;
309     uint64_t ext_end;
310 
311     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
312     if (ret < 0) {
313         goto fail;
314     }
315     be32_to_cpus(&header.magic);
316     be32_to_cpus(&header.version);
317     be64_to_cpus(&header.backing_file_offset);
318     be32_to_cpus(&header.backing_file_size);
319     be64_to_cpus(&header.size);
320     be32_to_cpus(&header.cluster_bits);
321     be32_to_cpus(&header.crypt_method);
322     be64_to_cpus(&header.l1_table_offset);
323     be32_to_cpus(&header.l1_size);
324     be64_to_cpus(&header.refcount_table_offset);
325     be32_to_cpus(&header.refcount_table_clusters);
326     be64_to_cpus(&header.snapshots_offset);
327     be32_to_cpus(&header.nb_snapshots);
328 
329     if (header.magic != QCOW_MAGIC) {
330         ret = -EMEDIUMTYPE;
331         goto fail;
332     }
333     if (header.version < 2 || header.version > 3) {
334         report_unsupported(bs, "QCOW version %d", header.version);
335         ret = -ENOTSUP;
336         goto fail;
337     }
338 
339     s->qcow_version = header.version;
340 
341     /* Initialise version 3 header fields */
342     if (header.version == 2) {
343         header.incompatible_features    = 0;
344         header.compatible_features      = 0;
345         header.autoclear_features       = 0;
346         header.refcount_order           = 4;
347         header.header_length            = 72;
348     } else {
349         be64_to_cpus(&header.incompatible_features);
350         be64_to_cpus(&header.compatible_features);
351         be64_to_cpus(&header.autoclear_features);
352         be32_to_cpus(&header.refcount_order);
353         be32_to_cpus(&header.header_length);
354     }
355 
356     if (header.header_length > sizeof(header)) {
357         s->unknown_header_fields_size = header.header_length - sizeof(header);
358         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
359         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
360                          s->unknown_header_fields_size);
361         if (ret < 0) {
362             goto fail;
363         }
364     }
365 
366     if (header.backing_file_offset) {
367         ext_end = header.backing_file_offset;
368     } else {
369         ext_end = 1 << header.cluster_bits;
370     }
371 
372     /* Handle feature bits */
373     s->incompatible_features    = header.incompatible_features;
374     s->compatible_features      = header.compatible_features;
375     s->autoclear_features       = header.autoclear_features;
376 
377     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
378         void *feature_table = NULL;
379         qcow2_read_extensions(bs, header.header_length, ext_end,
380                               &feature_table);
381         report_unsupported_feature(bs, feature_table,
382                                    s->incompatible_features &
383                                    ~QCOW2_INCOMPAT_MASK);
384         ret = -ENOTSUP;
385         goto fail;
386     }
387 
388     /* Check support for various header values */
389     if (header.refcount_order != 4) {
390         report_unsupported(bs, "%d bit reference counts",
391                            1 << header.refcount_order);
392         ret = -ENOTSUP;
393         goto fail;
394     }
395 
396     if (header.cluster_bits < MIN_CLUSTER_BITS ||
397         header.cluster_bits > MAX_CLUSTER_BITS) {
398         ret = -EINVAL;
399         goto fail;
400     }
401     if (header.crypt_method > QCOW_CRYPT_AES) {
402         ret = -EINVAL;
403         goto fail;
404     }
405     s->crypt_method_header = header.crypt_method;
406     if (s->crypt_method_header) {
407         bs->encrypted = 1;
408     }
409     s->cluster_bits = header.cluster_bits;
410     s->cluster_size = 1 << s->cluster_bits;
411     s->cluster_sectors = 1 << (s->cluster_bits - 9);
412     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
413     s->l2_size = 1 << s->l2_bits;
414     bs->total_sectors = header.size / 512;
415     s->csize_shift = (62 - (s->cluster_bits - 8));
416     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
417     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
418     s->refcount_table_offset = header.refcount_table_offset;
419     s->refcount_table_size =
420         header.refcount_table_clusters << (s->cluster_bits - 3);
421 
422     s->snapshots_offset = header.snapshots_offset;
423     s->nb_snapshots = header.nb_snapshots;
424 
425     /* read the level 1 table */
426     s->l1_size = header.l1_size;
427     s->l1_vm_state_index = size_to_l1(s, header.size);
428     /* the L1 table must contain at least enough entries to put
429        header.size bytes */
430     if (s->l1_size < s->l1_vm_state_index) {
431         ret = -EINVAL;
432         goto fail;
433     }
434     s->l1_table_offset = header.l1_table_offset;
435     if (s->l1_size > 0) {
436         s->l1_table = g_malloc0(
437             align_offset(s->l1_size * sizeof(uint64_t), 512));
438         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
439                          s->l1_size * sizeof(uint64_t));
440         if (ret < 0) {
441             goto fail;
442         }
443         for(i = 0;i < s->l1_size; i++) {
444             be64_to_cpus(&s->l1_table[i]);
445         }
446     }
447 
448     /* alloc L2 table/refcount block cache */
449     s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE);
450     s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE);
451 
452     s->cluster_cache = g_malloc(s->cluster_size);
453     /* one more sector for decompressed data alignment */
454     s->cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
455                                   + 512);
456     s->cluster_cache_offset = -1;
457     s->flags = flags;
458 
459     ret = qcow2_refcount_init(bs);
460     if (ret != 0) {
461         goto fail;
462     }
463 
464     QLIST_INIT(&s->cluster_allocs);
465 
466     /* read qcow2 extensions */
467     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL)) {
468         ret = -EINVAL;
469         goto fail;
470     }
471 
472     /* read the backing file name */
473     if (header.backing_file_offset != 0) {
474         len = header.backing_file_size;
475         if (len > 1023) {
476             len = 1023;
477         }
478         ret = bdrv_pread(bs->file, header.backing_file_offset,
479                          bs->backing_file, len);
480         if (ret < 0) {
481             goto fail;
482         }
483         bs->backing_file[len] = '\0';
484     }
485 
486     ret = qcow2_read_snapshots(bs);
487     if (ret < 0) {
488         goto fail;
489     }
490 
491     /* Clear unknown autoclear feature bits */
492     if (!bs->read_only && s->autoclear_features != 0) {
493         s->autoclear_features = 0;
494         ret = qcow2_update_header(bs);
495         if (ret < 0) {
496             goto fail;
497         }
498     }
499 
500     /* Initialise locks */
501     qemu_co_mutex_init(&s->lock);
502 
503     /* Repair image if dirty */
504     if (!(flags & BDRV_O_CHECK) && !bs->read_only &&
505         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
506         BdrvCheckResult result = {0};
507 
508         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS);
509         if (ret < 0) {
510             goto fail;
511         }
512     }
513 
514     /* Enable lazy_refcounts according to image and command line options */
515     opts = qemu_opts_create_nofail(&qcow2_runtime_opts);
516     qemu_opts_absorb_qdict(opts, options, &local_err);
517     if (error_is_set(&local_err)) {
518         qerror_report_err(local_err);
519         error_free(local_err);
520         ret = -EINVAL;
521         goto fail;
522     }
523 
524     s->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
525         (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
526 
527     qemu_opts_del(opts);
528 
529     if (s->use_lazy_refcounts && s->qcow_version < 3) {
530         qerror_report(ERROR_CLASS_GENERIC_ERROR, "Lazy refcounts require "
531             "a qcow2 image with at least qemu 1.1 compatibility level");
532         ret = -EINVAL;
533         goto fail;
534     }
535 
536 #ifdef DEBUG_ALLOC
537     {
538         BdrvCheckResult result = {0};
539         qcow2_check_refcounts(bs, &result, 0);
540     }
541 #endif
542     return ret;
543 
544  fail:
545     g_free(s->unknown_header_fields);
546     cleanup_unknown_header_ext(bs);
547     qcow2_free_snapshots(bs);
548     qcow2_refcount_close(bs);
549     g_free(s->l1_table);
550     if (s->l2_table_cache) {
551         qcow2_cache_destroy(bs, s->l2_table_cache);
552     }
553     g_free(s->cluster_cache);
554     qemu_vfree(s->cluster_data);
555     return ret;
556 }
557 
558 static int qcow2_set_key(BlockDriverState *bs, const char *key)
559 {
560     BDRVQcowState *s = bs->opaque;
561     uint8_t keybuf[16];
562     int len, i;
563 
564     memset(keybuf, 0, 16);
565     len = strlen(key);
566     if (len > 16)
567         len = 16;
568     /* XXX: we could compress the chars to 7 bits to increase
569        entropy */
570     for(i = 0;i < len;i++) {
571         keybuf[i] = key[i];
572     }
573     s->crypt_method = s->crypt_method_header;
574 
575     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
576         return -1;
577     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
578         return -1;
579 #if 0
580     /* test */
581     {
582         uint8_t in[16];
583         uint8_t out[16];
584         uint8_t tmp[16];
585         for(i=0;i<16;i++)
586             in[i] = i;
587         AES_encrypt(in, tmp, &s->aes_encrypt_key);
588         AES_decrypt(tmp, out, &s->aes_decrypt_key);
589         for(i = 0; i < 16; i++)
590             printf(" %02x", tmp[i]);
591         printf("\n");
592         for(i = 0; i < 16; i++)
593             printf(" %02x", out[i]);
594         printf("\n");
595     }
596 #endif
597     return 0;
598 }
599 
600 /* We have nothing to do for QCOW2 reopen, stubs just return
601  * success */
602 static int qcow2_reopen_prepare(BDRVReopenState *state,
603                                 BlockReopenQueue *queue, Error **errp)
604 {
605     return 0;
606 }
607 
608 static int coroutine_fn qcow2_co_is_allocated(BlockDriverState *bs,
609         int64_t sector_num, int nb_sectors, int *pnum)
610 {
611     BDRVQcowState *s = bs->opaque;
612     uint64_t cluster_offset;
613     int ret;
614 
615     *pnum = nb_sectors;
616     /* FIXME We can get errors here, but the bdrv_co_is_allocated interface
617      * can't pass them on today */
618     qemu_co_mutex_lock(&s->lock);
619     ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
620     qemu_co_mutex_unlock(&s->lock);
621     if (ret < 0) {
622         *pnum = 0;
623     }
624 
625     return (cluster_offset != 0) || (ret == QCOW2_CLUSTER_ZERO);
626 }
627 
628 /* handle reading after the end of the backing file */
629 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
630                   int64_t sector_num, int nb_sectors)
631 {
632     int n1;
633     if ((sector_num + nb_sectors) <= bs->total_sectors)
634         return nb_sectors;
635     if (sector_num >= bs->total_sectors)
636         n1 = 0;
637     else
638         n1 = bs->total_sectors - sector_num;
639 
640     qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1));
641 
642     return n1;
643 }
644 
645 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
646                           int remaining_sectors, QEMUIOVector *qiov)
647 {
648     BDRVQcowState *s = bs->opaque;
649     int index_in_cluster, n1;
650     int ret;
651     int cur_nr_sectors; /* number of sectors in current iteration */
652     uint64_t cluster_offset = 0;
653     uint64_t bytes_done = 0;
654     QEMUIOVector hd_qiov;
655     uint8_t *cluster_data = NULL;
656 
657     qemu_iovec_init(&hd_qiov, qiov->niov);
658 
659     qemu_co_mutex_lock(&s->lock);
660 
661     while (remaining_sectors != 0) {
662 
663         /* prepare next request */
664         cur_nr_sectors = remaining_sectors;
665         if (s->crypt_method) {
666             cur_nr_sectors = MIN(cur_nr_sectors,
667                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
668         }
669 
670         ret = qcow2_get_cluster_offset(bs, sector_num << 9,
671             &cur_nr_sectors, &cluster_offset);
672         if (ret < 0) {
673             goto fail;
674         }
675 
676         index_in_cluster = sector_num & (s->cluster_sectors - 1);
677 
678         qemu_iovec_reset(&hd_qiov);
679         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
680             cur_nr_sectors * 512);
681 
682         switch (ret) {
683         case QCOW2_CLUSTER_UNALLOCATED:
684 
685             if (bs->backing_hd) {
686                 /* read from the base image */
687                 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov,
688                     sector_num, cur_nr_sectors);
689                 if (n1 > 0) {
690                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
691                     qemu_co_mutex_unlock(&s->lock);
692                     ret = bdrv_co_readv(bs->backing_hd, sector_num,
693                                         n1, &hd_qiov);
694                     qemu_co_mutex_lock(&s->lock);
695                     if (ret < 0) {
696                         goto fail;
697                     }
698                 }
699             } else {
700                 /* Note: in this case, no need to wait */
701                 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
702             }
703             break;
704 
705         case QCOW2_CLUSTER_ZERO:
706             qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
707             break;
708 
709         case QCOW2_CLUSTER_COMPRESSED:
710             /* add AIO support for compressed blocks ? */
711             ret = qcow2_decompress_cluster(bs, cluster_offset);
712             if (ret < 0) {
713                 goto fail;
714             }
715 
716             qemu_iovec_from_buf(&hd_qiov, 0,
717                 s->cluster_cache + index_in_cluster * 512,
718                 512 * cur_nr_sectors);
719             break;
720 
721         case QCOW2_CLUSTER_NORMAL:
722             if ((cluster_offset & 511) != 0) {
723                 ret = -EIO;
724                 goto fail;
725             }
726 
727             if (s->crypt_method) {
728                 /*
729                  * For encrypted images, read everything into a temporary
730                  * contiguous buffer on which the AES functions can work.
731                  */
732                 if (!cluster_data) {
733                     cluster_data =
734                         qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
735                 }
736 
737                 assert(cur_nr_sectors <=
738                     QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
739                 qemu_iovec_reset(&hd_qiov);
740                 qemu_iovec_add(&hd_qiov, cluster_data,
741                     512 * cur_nr_sectors);
742             }
743 
744             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
745             qemu_co_mutex_unlock(&s->lock);
746             ret = bdrv_co_readv(bs->file,
747                                 (cluster_offset >> 9) + index_in_cluster,
748                                 cur_nr_sectors, &hd_qiov);
749             qemu_co_mutex_lock(&s->lock);
750             if (ret < 0) {
751                 goto fail;
752             }
753             if (s->crypt_method) {
754                 qcow2_encrypt_sectors(s, sector_num,  cluster_data,
755                     cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
756                 qemu_iovec_from_buf(qiov, bytes_done,
757                     cluster_data, 512 * cur_nr_sectors);
758             }
759             break;
760 
761         default:
762             g_assert_not_reached();
763             ret = -EIO;
764             goto fail;
765         }
766 
767         remaining_sectors -= cur_nr_sectors;
768         sector_num += cur_nr_sectors;
769         bytes_done += cur_nr_sectors * 512;
770     }
771     ret = 0;
772 
773 fail:
774     qemu_co_mutex_unlock(&s->lock);
775 
776     qemu_iovec_destroy(&hd_qiov);
777     qemu_vfree(cluster_data);
778 
779     return ret;
780 }
781 
782 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
783                            int64_t sector_num,
784                            int remaining_sectors,
785                            QEMUIOVector *qiov)
786 {
787     BDRVQcowState *s = bs->opaque;
788     int index_in_cluster;
789     int n_end;
790     int ret;
791     int cur_nr_sectors; /* number of sectors in current iteration */
792     uint64_t cluster_offset;
793     QEMUIOVector hd_qiov;
794     uint64_t bytes_done = 0;
795     uint8_t *cluster_data = NULL;
796     QCowL2Meta *l2meta = NULL;
797 
798     trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
799                                  remaining_sectors);
800 
801     qemu_iovec_init(&hd_qiov, qiov->niov);
802 
803     s->cluster_cache_offset = -1; /* disable compressed cache */
804 
805     qemu_co_mutex_lock(&s->lock);
806 
807     while (remaining_sectors != 0) {
808 
809         l2meta = NULL;
810 
811         trace_qcow2_writev_start_part(qemu_coroutine_self());
812         index_in_cluster = sector_num & (s->cluster_sectors - 1);
813         n_end = index_in_cluster + remaining_sectors;
814         if (s->crypt_method &&
815             n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) {
816             n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
817         }
818 
819         ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
820             index_in_cluster, n_end, &cur_nr_sectors, &cluster_offset, &l2meta);
821         if (ret < 0) {
822             goto fail;
823         }
824 
825         assert((cluster_offset & 511) == 0);
826 
827         qemu_iovec_reset(&hd_qiov);
828         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
829             cur_nr_sectors * 512);
830 
831         if (s->crypt_method) {
832             if (!cluster_data) {
833                 cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS *
834                                                  s->cluster_size);
835             }
836 
837             assert(hd_qiov.size <=
838                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
839             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
840 
841             qcow2_encrypt_sectors(s, sector_num, cluster_data,
842                 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
843 
844             qemu_iovec_reset(&hd_qiov);
845             qemu_iovec_add(&hd_qiov, cluster_data,
846                 cur_nr_sectors * 512);
847         }
848 
849         qemu_co_mutex_unlock(&s->lock);
850         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
851         trace_qcow2_writev_data(qemu_coroutine_self(),
852                                 (cluster_offset >> 9) + index_in_cluster);
853         ret = bdrv_co_writev(bs->file,
854                              (cluster_offset >> 9) + index_in_cluster,
855                              cur_nr_sectors, &hd_qiov);
856         qemu_co_mutex_lock(&s->lock);
857         if (ret < 0) {
858             goto fail;
859         }
860 
861         if (l2meta != NULL) {
862             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
863             if (ret < 0) {
864                 goto fail;
865             }
866 
867             /* Take the request off the list of running requests */
868             if (l2meta->nb_clusters != 0) {
869                 QLIST_REMOVE(l2meta, next_in_flight);
870             }
871 
872             qemu_co_mutex_unlock(&s->lock);
873             qemu_co_queue_restart_all(&l2meta->dependent_requests);
874             qemu_co_mutex_lock(&s->lock);
875 
876             g_free(l2meta);
877             l2meta = NULL;
878         }
879 
880         remaining_sectors -= cur_nr_sectors;
881         sector_num += cur_nr_sectors;
882         bytes_done += cur_nr_sectors * 512;
883         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
884     }
885     ret = 0;
886 
887 fail:
888     qemu_co_mutex_unlock(&s->lock);
889 
890     if (l2meta != NULL) {
891         if (l2meta->nb_clusters != 0) {
892             QLIST_REMOVE(l2meta, next_in_flight);
893         }
894         qemu_co_queue_restart_all(&l2meta->dependent_requests);
895         g_free(l2meta);
896     }
897 
898     qemu_iovec_destroy(&hd_qiov);
899     qemu_vfree(cluster_data);
900     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
901 
902     return ret;
903 }
904 
905 static void qcow2_close(BlockDriverState *bs)
906 {
907     BDRVQcowState *s = bs->opaque;
908     g_free(s->l1_table);
909 
910     qcow2_cache_flush(bs, s->l2_table_cache);
911     qcow2_cache_flush(bs, s->refcount_block_cache);
912 
913     qcow2_mark_clean(bs);
914 
915     qcow2_cache_destroy(bs, s->l2_table_cache);
916     qcow2_cache_destroy(bs, s->refcount_block_cache);
917 
918     g_free(s->unknown_header_fields);
919     cleanup_unknown_header_ext(bs);
920 
921     g_free(s->cluster_cache);
922     qemu_vfree(s->cluster_data);
923     qcow2_refcount_close(bs);
924     qcow2_free_snapshots(bs);
925 }
926 
927 static void qcow2_invalidate_cache(BlockDriverState *bs)
928 {
929     BDRVQcowState *s = bs->opaque;
930     int flags = s->flags;
931     AES_KEY aes_encrypt_key;
932     AES_KEY aes_decrypt_key;
933     uint32_t crypt_method = 0;
934     QDict *options;
935 
936     /*
937      * Backing files are read-only which makes all of their metadata immutable,
938      * that means we don't have to worry about reopening them here.
939      */
940 
941     if (s->crypt_method) {
942         crypt_method = s->crypt_method;
943         memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key));
944         memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key));
945     }
946 
947     qcow2_close(bs);
948 
949     options = qdict_new();
950     qdict_put(options, QCOW2_OPT_LAZY_REFCOUNTS,
951               qbool_from_int(s->use_lazy_refcounts));
952 
953     memset(s, 0, sizeof(BDRVQcowState));
954     qcow2_open(bs, options, flags);
955 
956     QDECREF(options);
957 
958     if (crypt_method) {
959         s->crypt_method = crypt_method;
960         memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key));
961         memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key));
962     }
963 }
964 
965 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
966     size_t len, size_t buflen)
967 {
968     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
969     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
970 
971     if (buflen < ext_len) {
972         return -ENOSPC;
973     }
974 
975     *ext_backing_fmt = (QCowExtension) {
976         .magic  = cpu_to_be32(magic),
977         .len    = cpu_to_be32(len),
978     };
979     memcpy(buf + sizeof(QCowExtension), s, len);
980 
981     return ext_len;
982 }
983 
984 /*
985  * Updates the qcow2 header, including the variable length parts of it, i.e.
986  * the backing file name and all extensions. qcow2 was not designed to allow
987  * such changes, so if we run out of space (we can only use the first cluster)
988  * this function may fail.
989  *
990  * Returns 0 on success, -errno in error cases.
991  */
992 int qcow2_update_header(BlockDriverState *bs)
993 {
994     BDRVQcowState *s = bs->opaque;
995     QCowHeader *header;
996     char *buf;
997     size_t buflen = s->cluster_size;
998     int ret;
999     uint64_t total_size;
1000     uint32_t refcount_table_clusters;
1001     size_t header_length;
1002     Qcow2UnknownHeaderExtension *uext;
1003 
1004     buf = qemu_blockalign(bs, buflen);
1005 
1006     /* Header structure */
1007     header = (QCowHeader*) buf;
1008 
1009     if (buflen < sizeof(*header)) {
1010         ret = -ENOSPC;
1011         goto fail;
1012     }
1013 
1014     header_length = sizeof(*header) + s->unknown_header_fields_size;
1015     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
1016     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
1017 
1018     *header = (QCowHeader) {
1019         /* Version 2 fields */
1020         .magic                  = cpu_to_be32(QCOW_MAGIC),
1021         .version                = cpu_to_be32(s->qcow_version),
1022         .backing_file_offset    = 0,
1023         .backing_file_size      = 0,
1024         .cluster_bits           = cpu_to_be32(s->cluster_bits),
1025         .size                   = cpu_to_be64(total_size),
1026         .crypt_method           = cpu_to_be32(s->crypt_method_header),
1027         .l1_size                = cpu_to_be32(s->l1_size),
1028         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
1029         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
1030         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
1031         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
1032         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
1033 
1034         /* Version 3 fields */
1035         .incompatible_features  = cpu_to_be64(s->incompatible_features),
1036         .compatible_features    = cpu_to_be64(s->compatible_features),
1037         .autoclear_features     = cpu_to_be64(s->autoclear_features),
1038         .refcount_order         = cpu_to_be32(3 + REFCOUNT_SHIFT),
1039         .header_length          = cpu_to_be32(header_length),
1040     };
1041 
1042     /* For older versions, write a shorter header */
1043     switch (s->qcow_version) {
1044     case 2:
1045         ret = offsetof(QCowHeader, incompatible_features);
1046         break;
1047     case 3:
1048         ret = sizeof(*header);
1049         break;
1050     default:
1051         ret = -EINVAL;
1052         goto fail;
1053     }
1054 
1055     buf += ret;
1056     buflen -= ret;
1057     memset(buf, 0, buflen);
1058 
1059     /* Preserve any unknown field in the header */
1060     if (s->unknown_header_fields_size) {
1061         if (buflen < s->unknown_header_fields_size) {
1062             ret = -ENOSPC;
1063             goto fail;
1064         }
1065 
1066         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1067         buf += s->unknown_header_fields_size;
1068         buflen -= s->unknown_header_fields_size;
1069     }
1070 
1071     /* Backing file format header extension */
1072     if (*bs->backing_format) {
1073         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1074                              bs->backing_format, strlen(bs->backing_format),
1075                              buflen);
1076         if (ret < 0) {
1077             goto fail;
1078         }
1079 
1080         buf += ret;
1081         buflen -= ret;
1082     }
1083 
1084     /* Feature table */
1085     Qcow2Feature features[] = {
1086         {
1087             .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1088             .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
1089             .name = "dirty bit",
1090         },
1091         {
1092             .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1093             .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1094             .name = "lazy refcounts",
1095         },
1096     };
1097 
1098     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1099                          features, sizeof(features), buflen);
1100     if (ret < 0) {
1101         goto fail;
1102     }
1103     buf += ret;
1104     buflen -= ret;
1105 
1106     /* Keep unknown header extensions */
1107     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1108         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1109         if (ret < 0) {
1110             goto fail;
1111         }
1112 
1113         buf += ret;
1114         buflen -= ret;
1115     }
1116 
1117     /* End of header extensions */
1118     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1119     if (ret < 0) {
1120         goto fail;
1121     }
1122 
1123     buf += ret;
1124     buflen -= ret;
1125 
1126     /* Backing file name */
1127     if (*bs->backing_file) {
1128         size_t backing_file_len = strlen(bs->backing_file);
1129 
1130         if (buflen < backing_file_len) {
1131             ret = -ENOSPC;
1132             goto fail;
1133         }
1134 
1135         /* Using strncpy is ok here, since buf is not NUL-terminated. */
1136         strncpy(buf, bs->backing_file, buflen);
1137 
1138         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1139         header->backing_file_size   = cpu_to_be32(backing_file_len);
1140     }
1141 
1142     /* Write the new header */
1143     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
1144     if (ret < 0) {
1145         goto fail;
1146     }
1147 
1148     ret = 0;
1149 fail:
1150     qemu_vfree(header);
1151     return ret;
1152 }
1153 
1154 static int qcow2_change_backing_file(BlockDriverState *bs,
1155     const char *backing_file, const char *backing_fmt)
1156 {
1157     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1158     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1159 
1160     return qcow2_update_header(bs);
1161 }
1162 
1163 static int preallocate(BlockDriverState *bs)
1164 {
1165     uint64_t nb_sectors;
1166     uint64_t offset;
1167     uint64_t host_offset = 0;
1168     int num;
1169     int ret;
1170     QCowL2Meta *meta;
1171 
1172     nb_sectors = bdrv_getlength(bs) >> 9;
1173     offset = 0;
1174 
1175     while (nb_sectors) {
1176         num = MIN(nb_sectors, INT_MAX >> 9);
1177         ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num,
1178                                          &host_offset, &meta);
1179         if (ret < 0) {
1180             return ret;
1181         }
1182 
1183         ret = qcow2_alloc_cluster_link_l2(bs, meta);
1184         if (ret < 0) {
1185             qcow2_free_any_clusters(bs, meta->alloc_offset, meta->nb_clusters);
1186             return ret;
1187         }
1188 
1189         /* There are no dependent requests, but we need to remove our request
1190          * from the list of in-flight requests */
1191         if (meta != NULL) {
1192             QLIST_REMOVE(meta, next_in_flight);
1193         }
1194 
1195         /* TODO Preallocate data if requested */
1196 
1197         nb_sectors -= num;
1198         offset += num << 9;
1199     }
1200 
1201     /*
1202      * It is expected that the image file is large enough to actually contain
1203      * all of the allocated clusters (otherwise we get failing reads after
1204      * EOF). Extend the image to the last allocated sector.
1205      */
1206     if (host_offset != 0) {
1207         uint8_t buf[512];
1208         memset(buf, 0, 512);
1209         ret = bdrv_write(bs->file, (host_offset >> 9) + num - 1, buf, 1);
1210         if (ret < 0) {
1211             return ret;
1212         }
1213     }
1214 
1215     return 0;
1216 }
1217 
1218 static int qcow2_create2(const char *filename, int64_t total_size,
1219                          const char *backing_file, const char *backing_format,
1220                          int flags, size_t cluster_size, int prealloc,
1221                          QEMUOptionParameter *options, int version)
1222 {
1223     /* Calculate cluster_bits */
1224     int cluster_bits;
1225     cluster_bits = ffs(cluster_size) - 1;
1226     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
1227         (1 << cluster_bits) != cluster_size)
1228     {
1229         error_report(
1230             "Cluster size must be a power of two between %d and %dk",
1231             1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
1232         return -EINVAL;
1233     }
1234 
1235     /*
1236      * Open the image file and write a minimal qcow2 header.
1237      *
1238      * We keep things simple and start with a zero-sized image. We also
1239      * do without refcount blocks or a L1 table for now. We'll fix the
1240      * inconsistency later.
1241      *
1242      * We do need a refcount table because growing the refcount table means
1243      * allocating two new refcount blocks - the seconds of which would be at
1244      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
1245      * size for any qcow2 image.
1246      */
1247     BlockDriverState* bs;
1248     QCowHeader header;
1249     uint8_t* refcount_table;
1250     int ret;
1251 
1252     ret = bdrv_create_file(filename, options);
1253     if (ret < 0) {
1254         return ret;
1255     }
1256 
1257     ret = bdrv_file_open(&bs, filename, NULL, BDRV_O_RDWR);
1258     if (ret < 0) {
1259         return ret;
1260     }
1261 
1262     /* Write the header */
1263     memset(&header, 0, sizeof(header));
1264     header.magic = cpu_to_be32(QCOW_MAGIC);
1265     header.version = cpu_to_be32(version);
1266     header.cluster_bits = cpu_to_be32(cluster_bits);
1267     header.size = cpu_to_be64(0);
1268     header.l1_table_offset = cpu_to_be64(0);
1269     header.l1_size = cpu_to_be32(0);
1270     header.refcount_table_offset = cpu_to_be64(cluster_size);
1271     header.refcount_table_clusters = cpu_to_be32(1);
1272     header.refcount_order = cpu_to_be32(3 + REFCOUNT_SHIFT);
1273     header.header_length = cpu_to_be32(sizeof(header));
1274 
1275     if (flags & BLOCK_FLAG_ENCRYPT) {
1276         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1277     } else {
1278         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1279     }
1280 
1281     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
1282         header.compatible_features |=
1283             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
1284     }
1285 
1286     ret = bdrv_pwrite(bs, 0, &header, sizeof(header));
1287     if (ret < 0) {
1288         goto out;
1289     }
1290 
1291     /* Write an empty refcount table */
1292     refcount_table = g_malloc0(cluster_size);
1293     ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size);
1294     g_free(refcount_table);
1295 
1296     if (ret < 0) {
1297         goto out;
1298     }
1299 
1300     bdrv_close(bs);
1301 
1302     /*
1303      * And now open the image and make it consistent first (i.e. increase the
1304      * refcount of the cluster that is occupied by the header and the refcount
1305      * table)
1306      */
1307     BlockDriver* drv = bdrv_find_format("qcow2");
1308     assert(drv != NULL);
1309     ret = bdrv_open(bs, filename, NULL,
1310         BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv);
1311     if (ret < 0) {
1312         goto out;
1313     }
1314 
1315     ret = qcow2_alloc_clusters(bs, 2 * cluster_size);
1316     if (ret < 0) {
1317         goto out;
1318 
1319     } else if (ret != 0) {
1320         error_report("Huh, first cluster in empty image is already in use?");
1321         abort();
1322     }
1323 
1324     /* Okay, now that we have a valid image, let's give it the right size */
1325     ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE);
1326     if (ret < 0) {
1327         goto out;
1328     }
1329 
1330     /* Want a backing file? There you go.*/
1331     if (backing_file) {
1332         ret = bdrv_change_backing_file(bs, backing_file, backing_format);
1333         if (ret < 0) {
1334             goto out;
1335         }
1336     }
1337 
1338     /* And if we're supposed to preallocate metadata, do that now */
1339     if (prealloc) {
1340         BDRVQcowState *s = bs->opaque;
1341         qemu_co_mutex_lock(&s->lock);
1342         ret = preallocate(bs);
1343         qemu_co_mutex_unlock(&s->lock);
1344         if (ret < 0) {
1345             goto out;
1346         }
1347     }
1348 
1349     ret = 0;
1350 out:
1351     bdrv_delete(bs);
1352     return ret;
1353 }
1354 
1355 static int qcow2_create(const char *filename, QEMUOptionParameter *options)
1356 {
1357     const char *backing_file = NULL;
1358     const char *backing_fmt = NULL;
1359     uint64_t sectors = 0;
1360     int flags = 0;
1361     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
1362     int prealloc = 0;
1363     int version = 2;
1364 
1365     /* Read out options */
1366     while (options && options->name) {
1367         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1368             sectors = options->value.n / 512;
1369         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1370             backing_file = options->value.s;
1371         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
1372             backing_fmt = options->value.s;
1373         } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
1374             flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
1375         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
1376             if (options->value.n) {
1377                 cluster_size = options->value.n;
1378             }
1379         } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
1380             if (!options->value.s || !strcmp(options->value.s, "off")) {
1381                 prealloc = 0;
1382             } else if (!strcmp(options->value.s, "metadata")) {
1383                 prealloc = 1;
1384             } else {
1385                 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
1386                     options->value.s);
1387                 return -EINVAL;
1388             }
1389         } else if (!strcmp(options->name, BLOCK_OPT_COMPAT_LEVEL)) {
1390             if (!options->value.s || !strcmp(options->value.s, "0.10")) {
1391                 version = 2;
1392             } else if (!strcmp(options->value.s, "1.1")) {
1393                 version = 3;
1394             } else {
1395                 fprintf(stderr, "Invalid compatibility level: '%s'\n",
1396                     options->value.s);
1397                 return -EINVAL;
1398             }
1399         } else if (!strcmp(options->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
1400             flags |= options->value.n ? BLOCK_FLAG_LAZY_REFCOUNTS : 0;
1401         }
1402         options++;
1403     }
1404 
1405     if (backing_file && prealloc) {
1406         fprintf(stderr, "Backing file and preallocation cannot be used at "
1407             "the same time\n");
1408         return -EINVAL;
1409     }
1410 
1411     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
1412         fprintf(stderr, "Lazy refcounts only supported with compatibility "
1413                 "level 1.1 and above (use compat=1.1 or greater)\n");
1414         return -EINVAL;
1415     }
1416 
1417     return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags,
1418                          cluster_size, prealloc, options, version);
1419 }
1420 
1421 static int qcow2_make_empty(BlockDriverState *bs)
1422 {
1423 #if 0
1424     /* XXX: not correct */
1425     BDRVQcowState *s = bs->opaque;
1426     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1427     int ret;
1428 
1429     memset(s->l1_table, 0, l1_length);
1430     if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
1431         return -1;
1432     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
1433     if (ret < 0)
1434         return ret;
1435 
1436     l2_cache_reset(bs);
1437 #endif
1438     return 0;
1439 }
1440 
1441 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
1442     int64_t sector_num, int nb_sectors)
1443 {
1444     int ret;
1445     BDRVQcowState *s = bs->opaque;
1446 
1447     /* Emulate misaligned zero writes */
1448     if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
1449         return -ENOTSUP;
1450     }
1451 
1452     /* Whatever is left can use real zero clusters */
1453     qemu_co_mutex_lock(&s->lock);
1454     ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1455         nb_sectors);
1456     qemu_co_mutex_unlock(&s->lock);
1457 
1458     return ret;
1459 }
1460 
1461 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
1462     int64_t sector_num, int nb_sectors)
1463 {
1464     int ret;
1465     BDRVQcowState *s = bs->opaque;
1466 
1467     qemu_co_mutex_lock(&s->lock);
1468     ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1469         nb_sectors);
1470     qemu_co_mutex_unlock(&s->lock);
1471     return ret;
1472 }
1473 
1474 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
1475 {
1476     BDRVQcowState *s = bs->opaque;
1477     int ret, new_l1_size;
1478 
1479     if (offset & 511) {
1480         error_report("The new size must be a multiple of 512");
1481         return -EINVAL;
1482     }
1483 
1484     /* cannot proceed if image has snapshots */
1485     if (s->nb_snapshots) {
1486         error_report("Can't resize an image which has snapshots");
1487         return -ENOTSUP;
1488     }
1489 
1490     /* shrinking is currently not supported */
1491     if (offset < bs->total_sectors * 512) {
1492         error_report("qcow2 doesn't support shrinking images yet");
1493         return -ENOTSUP;
1494     }
1495 
1496     new_l1_size = size_to_l1(s, offset);
1497     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
1498     if (ret < 0) {
1499         return ret;
1500     }
1501 
1502     /* write updated header.size */
1503     offset = cpu_to_be64(offset);
1504     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
1505                            &offset, sizeof(uint64_t));
1506     if (ret < 0) {
1507         return ret;
1508     }
1509 
1510     s->l1_vm_state_index = new_l1_size;
1511     return 0;
1512 }
1513 
1514 /* XXX: put compressed sectors first, then all the cluster aligned
1515    tables to avoid losing bytes in alignment */
1516 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
1517                                   const uint8_t *buf, int nb_sectors)
1518 {
1519     BDRVQcowState *s = bs->opaque;
1520     z_stream strm;
1521     int ret, out_len;
1522     uint8_t *out_buf;
1523     uint64_t cluster_offset;
1524 
1525     if (nb_sectors == 0) {
1526         /* align end of file to a sector boundary to ease reading with
1527            sector based I/Os */
1528         cluster_offset = bdrv_getlength(bs->file);
1529         cluster_offset = (cluster_offset + 511) & ~511;
1530         bdrv_truncate(bs->file, cluster_offset);
1531         return 0;
1532     }
1533 
1534     if (nb_sectors != s->cluster_sectors)
1535         return -EINVAL;
1536 
1537     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1538 
1539     /* best compression, small window, no zlib header */
1540     memset(&strm, 0, sizeof(strm));
1541     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1542                        Z_DEFLATED, -12,
1543                        9, Z_DEFAULT_STRATEGY);
1544     if (ret != 0) {
1545         ret = -EINVAL;
1546         goto fail;
1547     }
1548 
1549     strm.avail_in = s->cluster_size;
1550     strm.next_in = (uint8_t *)buf;
1551     strm.avail_out = s->cluster_size;
1552     strm.next_out = out_buf;
1553 
1554     ret = deflate(&strm, Z_FINISH);
1555     if (ret != Z_STREAM_END && ret != Z_OK) {
1556         deflateEnd(&strm);
1557         ret = -EINVAL;
1558         goto fail;
1559     }
1560     out_len = strm.next_out - out_buf;
1561 
1562     deflateEnd(&strm);
1563 
1564     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1565         /* could not compress: write normal cluster */
1566         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1567         if (ret < 0) {
1568             goto fail;
1569         }
1570     } else {
1571         cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1572             sector_num << 9, out_len);
1573         if (!cluster_offset) {
1574             ret = -EIO;
1575             goto fail;
1576         }
1577         cluster_offset &= s->cluster_offset_mask;
1578         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1579         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
1580         if (ret < 0) {
1581             goto fail;
1582         }
1583     }
1584 
1585     ret = 0;
1586 fail:
1587     g_free(out_buf);
1588     return ret;
1589 }
1590 
1591 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
1592 {
1593     BDRVQcowState *s = bs->opaque;
1594     int ret;
1595 
1596     qemu_co_mutex_lock(&s->lock);
1597     ret = qcow2_cache_flush(bs, s->l2_table_cache);
1598     if (ret < 0) {
1599         qemu_co_mutex_unlock(&s->lock);
1600         return ret;
1601     }
1602 
1603     if (qcow2_need_accurate_refcounts(s)) {
1604         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1605         if (ret < 0) {
1606             qemu_co_mutex_unlock(&s->lock);
1607             return ret;
1608         }
1609     }
1610     qemu_co_mutex_unlock(&s->lock);
1611 
1612     return 0;
1613 }
1614 
1615 static int64_t qcow2_vm_state_offset(BDRVQcowState *s)
1616 {
1617 	return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1618 }
1619 
1620 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1621 {
1622     BDRVQcowState *s = bs->opaque;
1623     bdi->cluster_size = s->cluster_size;
1624     bdi->vm_state_offset = qcow2_vm_state_offset(s);
1625     return 0;
1626 }
1627 
1628 #if 0
1629 static void dump_refcounts(BlockDriverState *bs)
1630 {
1631     BDRVQcowState *s = bs->opaque;
1632     int64_t nb_clusters, k, k1, size;
1633     int refcount;
1634 
1635     size = bdrv_getlength(bs->file);
1636     nb_clusters = size_to_clusters(s, size);
1637     for(k = 0; k < nb_clusters;) {
1638         k1 = k;
1639         refcount = get_refcount(bs, k);
1640         k++;
1641         while (k < nb_clusters && get_refcount(bs, k) == refcount)
1642             k++;
1643         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
1644                k - k1);
1645     }
1646 }
1647 #endif
1648 
1649 static int qcow2_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1650                               int64_t pos, int size)
1651 {
1652     BDRVQcowState *s = bs->opaque;
1653     int growable = bs->growable;
1654     int ret;
1655 
1656     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
1657     bs->growable = 1;
1658     ret = bdrv_pwrite(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1659     bs->growable = growable;
1660 
1661     return ret;
1662 }
1663 
1664 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1665                               int64_t pos, int size)
1666 {
1667     BDRVQcowState *s = bs->opaque;
1668     int growable = bs->growable;
1669     int ret;
1670 
1671     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
1672     bs->growable = 1;
1673     ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1674     bs->growable = growable;
1675 
1676     return ret;
1677 }
1678 
1679 static QEMUOptionParameter qcow2_create_options[] = {
1680     {
1681         .name = BLOCK_OPT_SIZE,
1682         .type = OPT_SIZE,
1683         .help = "Virtual disk size"
1684     },
1685     {
1686         .name = BLOCK_OPT_COMPAT_LEVEL,
1687         .type = OPT_STRING,
1688         .help = "Compatibility level (0.10 or 1.1)"
1689     },
1690     {
1691         .name = BLOCK_OPT_BACKING_FILE,
1692         .type = OPT_STRING,
1693         .help = "File name of a base image"
1694     },
1695     {
1696         .name = BLOCK_OPT_BACKING_FMT,
1697         .type = OPT_STRING,
1698         .help = "Image format of the base image"
1699     },
1700     {
1701         .name = BLOCK_OPT_ENCRYPT,
1702         .type = OPT_FLAG,
1703         .help = "Encrypt the image"
1704     },
1705     {
1706         .name = BLOCK_OPT_CLUSTER_SIZE,
1707         .type = OPT_SIZE,
1708         .help = "qcow2 cluster size",
1709         .value = { .n = DEFAULT_CLUSTER_SIZE },
1710     },
1711     {
1712         .name = BLOCK_OPT_PREALLOC,
1713         .type = OPT_STRING,
1714         .help = "Preallocation mode (allowed values: off, metadata)"
1715     },
1716     {
1717         .name = BLOCK_OPT_LAZY_REFCOUNTS,
1718         .type = OPT_FLAG,
1719         .help = "Postpone refcount updates",
1720     },
1721     { NULL }
1722 };
1723 
1724 static BlockDriver bdrv_qcow2 = {
1725     .format_name        = "qcow2",
1726     .instance_size      = sizeof(BDRVQcowState),
1727     .bdrv_probe         = qcow2_probe,
1728     .bdrv_open          = qcow2_open,
1729     .bdrv_close         = qcow2_close,
1730     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
1731     .bdrv_create        = qcow2_create,
1732     .bdrv_co_is_allocated = qcow2_co_is_allocated,
1733     .bdrv_set_key       = qcow2_set_key,
1734     .bdrv_make_empty    = qcow2_make_empty,
1735 
1736     .bdrv_co_readv          = qcow2_co_readv,
1737     .bdrv_co_writev         = qcow2_co_writev,
1738     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
1739 
1740     .bdrv_co_write_zeroes   = qcow2_co_write_zeroes,
1741     .bdrv_co_discard        = qcow2_co_discard,
1742     .bdrv_truncate          = qcow2_truncate,
1743     .bdrv_write_compressed  = qcow2_write_compressed,
1744 
1745     .bdrv_snapshot_create   = qcow2_snapshot_create,
1746     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
1747     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
1748     .bdrv_snapshot_list     = qcow2_snapshot_list,
1749     .bdrv_snapshot_load_tmp     = qcow2_snapshot_load_tmp,
1750     .bdrv_get_info      = qcow2_get_info,
1751 
1752     .bdrv_save_vmstate    = qcow2_save_vmstate,
1753     .bdrv_load_vmstate    = qcow2_load_vmstate,
1754 
1755     .bdrv_change_backing_file   = qcow2_change_backing_file,
1756 
1757     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
1758 
1759     .create_options = qcow2_create_options,
1760     .bdrv_check = qcow2_check,
1761 };
1762 
1763 static void bdrv_qcow2_init(void)
1764 {
1765     bdrv_register(&bdrv_qcow2);
1766 }
1767 
1768 block_init(bdrv_qcow2_init);
1769