xref: /openbmc/qemu/block/qcow2.c (revision 15faf946)
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_int.h"
26 #include "module.h"
27 #include <zlib.h>
28 #include "aes.h"
29 #include "block/qcow2.h"
30 #include "qemu-error.h"
31 #include "qerror.h"
32 #include "trace.h"
33 
34 /*
35   Differences with QCOW:
36 
37   - Support for multiple incremental snapshots.
38   - Memory management by reference counts.
39   - Clusters which have a reference count of one have the bit
40     QCOW_OFLAG_COPIED to optimize write performance.
41   - Size of compressed clusters is stored in sectors to reduce bit usage
42     in the cluster offsets.
43   - Support for storing additional data (such as the VM state) in the
44     snapshots.
45   - If a backing store is used, the cluster size is not constrained
46     (could be backported to QCOW).
47   - L2 tables have always a size of one cluster.
48 */
49 
50 
51 typedef struct {
52     uint32_t magic;
53     uint32_t len;
54 } QCowExtension;
55 
56 #define  QCOW2_EXT_MAGIC_END 0
57 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
58 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
59 
60 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
61 {
62     const QCowHeader *cow_header = (const void *)buf;
63 
64     if (buf_size >= sizeof(QCowHeader) &&
65         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
66         be32_to_cpu(cow_header->version) >= 2)
67         return 100;
68     else
69         return 0;
70 }
71 
72 
73 /*
74  * read qcow2 extension and fill bs
75  * start reading from start_offset
76  * finish reading upon magic of value 0 or when end_offset reached
77  * unknown magic is skipped (future extension this version knows nothing about)
78  * return 0 upon success, non-0 otherwise
79  */
80 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
81                                  uint64_t end_offset, void **p_feature_table)
82 {
83     BDRVQcowState *s = bs->opaque;
84     QCowExtension ext;
85     uint64_t offset;
86     int ret;
87 
88 #ifdef DEBUG_EXT
89     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
90 #endif
91     offset = start_offset;
92     while (offset < end_offset) {
93 
94 #ifdef DEBUG_EXT
95         /* Sanity check */
96         if (offset > s->cluster_size)
97             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
98 
99         printf("attempting to read extended header in offset %lu\n", offset);
100 #endif
101 
102         if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) {
103             fprintf(stderr, "qcow2_read_extension: ERROR: "
104                     "pread fail from offset %" PRIu64 "\n",
105                     offset);
106             return 1;
107         }
108         be32_to_cpus(&ext.magic);
109         be32_to_cpus(&ext.len);
110         offset += sizeof(ext);
111 #ifdef DEBUG_EXT
112         printf("ext.magic = 0x%x\n", ext.magic);
113 #endif
114         if (ext.len > end_offset - offset) {
115             error_report("Header extension too large");
116             return -EINVAL;
117         }
118 
119         switch (ext.magic) {
120         case QCOW2_EXT_MAGIC_END:
121             return 0;
122 
123         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
124             if (ext.len >= sizeof(bs->backing_format)) {
125                 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
126                         " (>=%zu)\n",
127                         ext.len, sizeof(bs->backing_format));
128                 return 2;
129             }
130             if (bdrv_pread(bs->file, offset , bs->backing_format,
131                            ext.len) != ext.len)
132                 return 3;
133             bs->backing_format[ext.len] = '\0';
134 #ifdef DEBUG_EXT
135             printf("Qcow2: Got format extension %s\n", bs->backing_format);
136 #endif
137             break;
138 
139         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
140             if (p_feature_table != NULL) {
141                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
142                 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
143                 if (ret < 0) {
144                     return ret;
145                 }
146 
147                 *p_feature_table = feature_table;
148             }
149             break;
150 
151         default:
152             /* unknown magic - save it in case we need to rewrite the header */
153             {
154                 Qcow2UnknownHeaderExtension *uext;
155 
156                 uext = g_malloc0(sizeof(*uext)  + ext.len);
157                 uext->magic = ext.magic;
158                 uext->len = ext.len;
159                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
160 
161                 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
162                 if (ret < 0) {
163                     return ret;
164                 }
165             }
166             break;
167         }
168 
169         offset += ((ext.len + 7) & ~7);
170     }
171 
172     return 0;
173 }
174 
175 static void cleanup_unknown_header_ext(BlockDriverState *bs)
176 {
177     BDRVQcowState *s = bs->opaque;
178     Qcow2UnknownHeaderExtension *uext, *next;
179 
180     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
181         QLIST_REMOVE(uext, next);
182         g_free(uext);
183     }
184 }
185 
186 static void GCC_FMT_ATTR(2, 3) report_unsupported(BlockDriverState *bs,
187     const char *fmt, ...)
188 {
189     char msg[64];
190     va_list ap;
191 
192     va_start(ap, fmt);
193     vsnprintf(msg, sizeof(msg), fmt, ap);
194     va_end(ap);
195 
196     qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
197         bs->device_name, "qcow2", msg);
198 }
199 
200 static void report_unsupported_feature(BlockDriverState *bs,
201     Qcow2Feature *table, uint64_t mask)
202 {
203     while (table && table->name[0] != '\0') {
204         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
205             if (mask & (1 << table->bit)) {
206                 report_unsupported(bs, "%.46s",table->name);
207                 mask &= ~(1 << table->bit);
208             }
209         }
210         table++;
211     }
212 
213     if (mask) {
214         report_unsupported(bs, "Unknown incompatible feature: %" PRIx64, mask);
215     }
216 }
217 
218 /*
219  * Sets the dirty bit and flushes afterwards if necessary.
220  *
221  * The incompatible_features bit is only set if the image file header was
222  * updated successfully.  Therefore it is not required to check the return
223  * value of this function.
224  */
225 int qcow2_mark_dirty(BlockDriverState *bs)
226 {
227     BDRVQcowState *s = bs->opaque;
228     uint64_t val;
229     int ret;
230 
231     assert(s->qcow_version >= 3);
232 
233     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
234         return 0; /* already dirty */
235     }
236 
237     val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
238     ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
239                       &val, sizeof(val));
240     if (ret < 0) {
241         return ret;
242     }
243     ret = bdrv_flush(bs->file);
244     if (ret < 0) {
245         return ret;
246     }
247 
248     /* Only treat image as dirty if the header was updated successfully */
249     s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
250     return 0;
251 }
252 
253 /*
254  * Clears the dirty bit and flushes before if necessary.  Only call this
255  * function when there are no pending requests, it does not guard against
256  * concurrent requests dirtying the image.
257  */
258 static int qcow2_mark_clean(BlockDriverState *bs)
259 {
260     BDRVQcowState *s = bs->opaque;
261 
262     if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
263         int ret = bdrv_flush(bs);
264         if (ret < 0) {
265             return ret;
266         }
267 
268         s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
269         return qcow2_update_header(bs);
270     }
271     return 0;
272 }
273 
274 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
275                        BdrvCheckMode fix)
276 {
277     int ret = qcow2_check_refcounts(bs, result, fix);
278     if (ret < 0) {
279         return ret;
280     }
281 
282     if (fix && result->check_errors == 0 && result->corruptions == 0) {
283         return qcow2_mark_clean(bs);
284     }
285     return ret;
286 }
287 
288 static int qcow2_open(BlockDriverState *bs, int flags)
289 {
290     BDRVQcowState *s = bs->opaque;
291     int len, i, ret = 0;
292     QCowHeader header;
293     uint64_t ext_end;
294 
295     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
296     if (ret < 0) {
297         goto fail;
298     }
299     be32_to_cpus(&header.magic);
300     be32_to_cpus(&header.version);
301     be64_to_cpus(&header.backing_file_offset);
302     be32_to_cpus(&header.backing_file_size);
303     be64_to_cpus(&header.size);
304     be32_to_cpus(&header.cluster_bits);
305     be32_to_cpus(&header.crypt_method);
306     be64_to_cpus(&header.l1_table_offset);
307     be32_to_cpus(&header.l1_size);
308     be64_to_cpus(&header.refcount_table_offset);
309     be32_to_cpus(&header.refcount_table_clusters);
310     be64_to_cpus(&header.snapshots_offset);
311     be32_to_cpus(&header.nb_snapshots);
312 
313     if (header.magic != QCOW_MAGIC) {
314         ret = -EINVAL;
315         goto fail;
316     }
317     if (header.version < 2 || header.version > 3) {
318         report_unsupported(bs, "QCOW version %d", header.version);
319         ret = -ENOTSUP;
320         goto fail;
321     }
322 
323     s->qcow_version = header.version;
324 
325     /* Initialise version 3 header fields */
326     if (header.version == 2) {
327         header.incompatible_features    = 0;
328         header.compatible_features      = 0;
329         header.autoclear_features       = 0;
330         header.refcount_order           = 4;
331         header.header_length            = 72;
332     } else {
333         be64_to_cpus(&header.incompatible_features);
334         be64_to_cpus(&header.compatible_features);
335         be64_to_cpus(&header.autoclear_features);
336         be32_to_cpus(&header.refcount_order);
337         be32_to_cpus(&header.header_length);
338     }
339 
340     if (header.header_length > sizeof(header)) {
341         s->unknown_header_fields_size = header.header_length - sizeof(header);
342         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
343         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
344                          s->unknown_header_fields_size);
345         if (ret < 0) {
346             goto fail;
347         }
348     }
349 
350     if (header.backing_file_offset) {
351         ext_end = header.backing_file_offset;
352     } else {
353         ext_end = 1 << header.cluster_bits;
354     }
355 
356     /* Handle feature bits */
357     s->incompatible_features    = header.incompatible_features;
358     s->compatible_features      = header.compatible_features;
359     s->autoclear_features       = header.autoclear_features;
360 
361     if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
362         void *feature_table = NULL;
363         qcow2_read_extensions(bs, header.header_length, ext_end,
364                               &feature_table);
365         report_unsupported_feature(bs, feature_table,
366                                    s->incompatible_features &
367                                    ~QCOW2_INCOMPAT_MASK);
368         ret = -ENOTSUP;
369         goto fail;
370     }
371 
372     /* Check support for various header values */
373     if (header.refcount_order != 4) {
374         report_unsupported(bs, "%d bit reference counts",
375                            1 << header.refcount_order);
376         ret = -ENOTSUP;
377         goto fail;
378     }
379 
380     if (header.cluster_bits < MIN_CLUSTER_BITS ||
381         header.cluster_bits > MAX_CLUSTER_BITS) {
382         ret = -EINVAL;
383         goto fail;
384     }
385     if (header.crypt_method > QCOW_CRYPT_AES) {
386         ret = -EINVAL;
387         goto fail;
388     }
389     s->crypt_method_header = header.crypt_method;
390     if (s->crypt_method_header) {
391         bs->encrypted = 1;
392     }
393     s->cluster_bits = header.cluster_bits;
394     s->cluster_size = 1 << s->cluster_bits;
395     s->cluster_sectors = 1 << (s->cluster_bits - 9);
396     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
397     s->l2_size = 1 << s->l2_bits;
398     bs->total_sectors = header.size / 512;
399     s->csize_shift = (62 - (s->cluster_bits - 8));
400     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
401     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
402     s->refcount_table_offset = header.refcount_table_offset;
403     s->refcount_table_size =
404         header.refcount_table_clusters << (s->cluster_bits - 3);
405 
406     s->snapshots_offset = header.snapshots_offset;
407     s->nb_snapshots = header.nb_snapshots;
408 
409     /* read the level 1 table */
410     s->l1_size = header.l1_size;
411     s->l1_vm_state_index = size_to_l1(s, header.size);
412     /* the L1 table must contain at least enough entries to put
413        header.size bytes */
414     if (s->l1_size < s->l1_vm_state_index) {
415         ret = -EINVAL;
416         goto fail;
417     }
418     s->l1_table_offset = header.l1_table_offset;
419     if (s->l1_size > 0) {
420         s->l1_table = g_malloc0(
421             align_offset(s->l1_size * sizeof(uint64_t), 512));
422         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
423                          s->l1_size * sizeof(uint64_t));
424         if (ret < 0) {
425             goto fail;
426         }
427         for(i = 0;i < s->l1_size; i++) {
428             be64_to_cpus(&s->l1_table[i]);
429         }
430     }
431 
432     /* alloc L2 table/refcount block cache */
433     s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE);
434     s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE);
435 
436     s->cluster_cache = g_malloc(s->cluster_size);
437     /* one more sector for decompressed data alignment */
438     s->cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
439                                   + 512);
440     s->cluster_cache_offset = -1;
441     s->flags = flags;
442 
443     ret = qcow2_refcount_init(bs);
444     if (ret != 0) {
445         goto fail;
446     }
447 
448     QLIST_INIT(&s->cluster_allocs);
449 
450     /* read qcow2 extensions */
451     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL)) {
452         ret = -EINVAL;
453         goto fail;
454     }
455 
456     /* read the backing file name */
457     if (header.backing_file_offset != 0) {
458         len = header.backing_file_size;
459         if (len > 1023) {
460             len = 1023;
461         }
462         ret = bdrv_pread(bs->file, header.backing_file_offset,
463                          bs->backing_file, len);
464         if (ret < 0) {
465             goto fail;
466         }
467         bs->backing_file[len] = '\0';
468     }
469 
470     ret = qcow2_read_snapshots(bs);
471     if (ret < 0) {
472         goto fail;
473     }
474 
475     /* Clear unknown autoclear feature bits */
476     if (!bs->read_only && s->autoclear_features != 0) {
477         s->autoclear_features = 0;
478         ret = qcow2_update_header(bs);
479         if (ret < 0) {
480             goto fail;
481         }
482     }
483 
484     /* Initialise locks */
485     qemu_co_mutex_init(&s->lock);
486 
487     /* Repair image if dirty */
488     if (!(flags & BDRV_O_CHECK) && !bs->read_only &&
489         (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
490         BdrvCheckResult result = {0};
491 
492         ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS);
493         if (ret < 0) {
494             goto fail;
495         }
496     }
497 
498 #ifdef DEBUG_ALLOC
499     {
500         BdrvCheckResult result = {0};
501         qcow2_check_refcounts(bs, &result, 0);
502     }
503 #endif
504     return ret;
505 
506  fail:
507     g_free(s->unknown_header_fields);
508     cleanup_unknown_header_ext(bs);
509     qcow2_free_snapshots(bs);
510     qcow2_refcount_close(bs);
511     g_free(s->l1_table);
512     if (s->l2_table_cache) {
513         qcow2_cache_destroy(bs, s->l2_table_cache);
514     }
515     g_free(s->cluster_cache);
516     qemu_vfree(s->cluster_data);
517     return ret;
518 }
519 
520 static int qcow2_set_key(BlockDriverState *bs, const char *key)
521 {
522     BDRVQcowState *s = bs->opaque;
523     uint8_t keybuf[16];
524     int len, i;
525 
526     memset(keybuf, 0, 16);
527     len = strlen(key);
528     if (len > 16)
529         len = 16;
530     /* XXX: we could compress the chars to 7 bits to increase
531        entropy */
532     for(i = 0;i < len;i++) {
533         keybuf[i] = key[i];
534     }
535     s->crypt_method = s->crypt_method_header;
536 
537     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
538         return -1;
539     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
540         return -1;
541 #if 0
542     /* test */
543     {
544         uint8_t in[16];
545         uint8_t out[16];
546         uint8_t tmp[16];
547         for(i=0;i<16;i++)
548             in[i] = i;
549         AES_encrypt(in, tmp, &s->aes_encrypt_key);
550         AES_decrypt(tmp, out, &s->aes_decrypt_key);
551         for(i = 0; i < 16; i++)
552             printf(" %02x", tmp[i]);
553         printf("\n");
554         for(i = 0; i < 16; i++)
555             printf(" %02x", out[i]);
556         printf("\n");
557     }
558 #endif
559     return 0;
560 }
561 
562 /* We have nothing to do for QCOW2 reopen, stubs just return
563  * success */
564 static int qcow2_reopen_prepare(BDRVReopenState *state,
565                                 BlockReopenQueue *queue, Error **errp)
566 {
567     return 0;
568 }
569 
570 static int coroutine_fn qcow2_co_is_allocated(BlockDriverState *bs,
571         int64_t sector_num, int nb_sectors, int *pnum)
572 {
573     BDRVQcowState *s = bs->opaque;
574     uint64_t cluster_offset;
575     int ret;
576 
577     *pnum = nb_sectors;
578     /* FIXME We can get errors here, but the bdrv_co_is_allocated interface
579      * can't pass them on today */
580     qemu_co_mutex_lock(&s->lock);
581     ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
582     qemu_co_mutex_unlock(&s->lock);
583     if (ret < 0) {
584         *pnum = 0;
585     }
586 
587     return (cluster_offset != 0);
588 }
589 
590 /* handle reading after the end of the backing file */
591 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
592                   int64_t sector_num, int nb_sectors)
593 {
594     int n1;
595     if ((sector_num + nb_sectors) <= bs->total_sectors)
596         return nb_sectors;
597     if (sector_num >= bs->total_sectors)
598         n1 = 0;
599     else
600         n1 = bs->total_sectors - sector_num;
601 
602     qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1));
603 
604     return n1;
605 }
606 
607 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
608                           int remaining_sectors, QEMUIOVector *qiov)
609 {
610     BDRVQcowState *s = bs->opaque;
611     int index_in_cluster, n1;
612     int ret;
613     int cur_nr_sectors; /* number of sectors in current iteration */
614     uint64_t cluster_offset = 0;
615     uint64_t bytes_done = 0;
616     QEMUIOVector hd_qiov;
617     uint8_t *cluster_data = NULL;
618 
619     qemu_iovec_init(&hd_qiov, qiov->niov);
620 
621     qemu_co_mutex_lock(&s->lock);
622 
623     while (remaining_sectors != 0) {
624 
625         /* prepare next request */
626         cur_nr_sectors = remaining_sectors;
627         if (s->crypt_method) {
628             cur_nr_sectors = MIN(cur_nr_sectors,
629                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
630         }
631 
632         ret = qcow2_get_cluster_offset(bs, sector_num << 9,
633             &cur_nr_sectors, &cluster_offset);
634         if (ret < 0) {
635             goto fail;
636         }
637 
638         index_in_cluster = sector_num & (s->cluster_sectors - 1);
639 
640         qemu_iovec_reset(&hd_qiov);
641         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
642             cur_nr_sectors * 512);
643 
644         switch (ret) {
645         case QCOW2_CLUSTER_UNALLOCATED:
646 
647             if (bs->backing_hd) {
648                 /* read from the base image */
649                 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov,
650                     sector_num, cur_nr_sectors);
651                 if (n1 > 0) {
652                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
653                     qemu_co_mutex_unlock(&s->lock);
654                     ret = bdrv_co_readv(bs->backing_hd, sector_num,
655                                         n1, &hd_qiov);
656                     qemu_co_mutex_lock(&s->lock);
657                     if (ret < 0) {
658                         goto fail;
659                     }
660                 }
661             } else {
662                 /* Note: in this case, no need to wait */
663                 qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
664             }
665             break;
666 
667         case QCOW2_CLUSTER_ZERO:
668             if (s->qcow_version < 3) {
669                 ret = -EIO;
670                 goto fail;
671             }
672             qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
673             break;
674 
675         case QCOW2_CLUSTER_COMPRESSED:
676             /* add AIO support for compressed blocks ? */
677             ret = qcow2_decompress_cluster(bs, cluster_offset);
678             if (ret < 0) {
679                 goto fail;
680             }
681 
682             qemu_iovec_from_buf(&hd_qiov, 0,
683                 s->cluster_cache + index_in_cluster * 512,
684                 512 * cur_nr_sectors);
685             break;
686 
687         case QCOW2_CLUSTER_NORMAL:
688             if ((cluster_offset & 511) != 0) {
689                 ret = -EIO;
690                 goto fail;
691             }
692 
693             if (s->crypt_method) {
694                 /*
695                  * For encrypted images, read everything into a temporary
696                  * contiguous buffer on which the AES functions can work.
697                  */
698                 if (!cluster_data) {
699                     cluster_data =
700                         qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
701                 }
702 
703                 assert(cur_nr_sectors <=
704                     QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
705                 qemu_iovec_reset(&hd_qiov);
706                 qemu_iovec_add(&hd_qiov, cluster_data,
707                     512 * cur_nr_sectors);
708             }
709 
710             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
711             qemu_co_mutex_unlock(&s->lock);
712             ret = bdrv_co_readv(bs->file,
713                                 (cluster_offset >> 9) + index_in_cluster,
714                                 cur_nr_sectors, &hd_qiov);
715             qemu_co_mutex_lock(&s->lock);
716             if (ret < 0) {
717                 goto fail;
718             }
719             if (s->crypt_method) {
720                 qcow2_encrypt_sectors(s, sector_num,  cluster_data,
721                     cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
722                 qemu_iovec_from_buf(qiov, bytes_done,
723                     cluster_data, 512 * cur_nr_sectors);
724             }
725             break;
726 
727         default:
728             g_assert_not_reached();
729             ret = -EIO;
730             goto fail;
731         }
732 
733         remaining_sectors -= cur_nr_sectors;
734         sector_num += cur_nr_sectors;
735         bytes_done += cur_nr_sectors * 512;
736     }
737     ret = 0;
738 
739 fail:
740     qemu_co_mutex_unlock(&s->lock);
741 
742     qemu_iovec_destroy(&hd_qiov);
743     qemu_vfree(cluster_data);
744 
745     return ret;
746 }
747 
748 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
749                            int64_t sector_num,
750                            int remaining_sectors,
751                            QEMUIOVector *qiov)
752 {
753     BDRVQcowState *s = bs->opaque;
754     int index_in_cluster;
755     int n_end;
756     int ret;
757     int cur_nr_sectors; /* number of sectors in current iteration */
758     uint64_t cluster_offset;
759     QEMUIOVector hd_qiov;
760     uint64_t bytes_done = 0;
761     uint8_t *cluster_data = NULL;
762     QCowL2Meta *l2meta;
763 
764     trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
765                                  remaining_sectors);
766 
767     qemu_iovec_init(&hd_qiov, qiov->niov);
768 
769     s->cluster_cache_offset = -1; /* disable compressed cache */
770 
771     qemu_co_mutex_lock(&s->lock);
772 
773     while (remaining_sectors != 0) {
774 
775         l2meta = NULL;
776 
777         trace_qcow2_writev_start_part(qemu_coroutine_self());
778         index_in_cluster = sector_num & (s->cluster_sectors - 1);
779         n_end = index_in_cluster + remaining_sectors;
780         if (s->crypt_method &&
781             n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) {
782             n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
783         }
784 
785         ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
786             index_in_cluster, n_end, &cur_nr_sectors, &cluster_offset, &l2meta);
787         if (ret < 0) {
788             goto fail;
789         }
790 
791         assert((cluster_offset & 511) == 0);
792 
793         qemu_iovec_reset(&hd_qiov);
794         qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
795             cur_nr_sectors * 512);
796 
797         if (s->crypt_method) {
798             if (!cluster_data) {
799                 cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS *
800                                                  s->cluster_size);
801             }
802 
803             assert(hd_qiov.size <=
804                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
805             qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
806 
807             qcow2_encrypt_sectors(s, sector_num, cluster_data,
808                 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
809 
810             qemu_iovec_reset(&hd_qiov);
811             qemu_iovec_add(&hd_qiov, cluster_data,
812                 cur_nr_sectors * 512);
813         }
814 
815         qemu_co_mutex_unlock(&s->lock);
816         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
817         trace_qcow2_writev_data(qemu_coroutine_self(),
818                                 (cluster_offset >> 9) + index_in_cluster);
819         ret = bdrv_co_writev(bs->file,
820                              (cluster_offset >> 9) + index_in_cluster,
821                              cur_nr_sectors, &hd_qiov);
822         qemu_co_mutex_lock(&s->lock);
823         if (ret < 0) {
824             goto fail;
825         }
826 
827         if (l2meta != NULL) {
828             ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
829             if (ret < 0) {
830                 goto fail;
831             }
832 
833             /* Take the request off the list of running requests */
834             if (l2meta->nb_clusters != 0) {
835                 QLIST_REMOVE(l2meta, next_in_flight);
836             }
837 
838             qemu_co_mutex_unlock(&s->lock);
839             qemu_co_queue_restart_all(&l2meta->dependent_requests);
840             qemu_co_mutex_lock(&s->lock);
841 
842             g_free(l2meta);
843             l2meta = NULL;
844         }
845 
846         remaining_sectors -= cur_nr_sectors;
847         sector_num += cur_nr_sectors;
848         bytes_done += cur_nr_sectors * 512;
849         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
850     }
851     ret = 0;
852 
853 fail:
854     qemu_co_mutex_unlock(&s->lock);
855 
856     if (l2meta != NULL) {
857         if (l2meta->nb_clusters != 0) {
858             QLIST_REMOVE(l2meta, next_in_flight);
859         }
860         qemu_co_queue_restart_all(&l2meta->dependent_requests);
861         g_free(l2meta);
862     }
863 
864     qemu_iovec_destroy(&hd_qiov);
865     qemu_vfree(cluster_data);
866     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
867 
868     return ret;
869 }
870 
871 static void qcow2_close(BlockDriverState *bs)
872 {
873     BDRVQcowState *s = bs->opaque;
874     g_free(s->l1_table);
875 
876     qcow2_cache_flush(bs, s->l2_table_cache);
877     qcow2_cache_flush(bs, s->refcount_block_cache);
878 
879     qcow2_mark_clean(bs);
880 
881     qcow2_cache_destroy(bs, s->l2_table_cache);
882     qcow2_cache_destroy(bs, s->refcount_block_cache);
883 
884     g_free(s->unknown_header_fields);
885     cleanup_unknown_header_ext(bs);
886 
887     g_free(s->cluster_cache);
888     qemu_vfree(s->cluster_data);
889     qcow2_refcount_close(bs);
890     qcow2_free_snapshots(bs);
891 }
892 
893 static void qcow2_invalidate_cache(BlockDriverState *bs)
894 {
895     BDRVQcowState *s = bs->opaque;
896     int flags = s->flags;
897     AES_KEY aes_encrypt_key;
898     AES_KEY aes_decrypt_key;
899     uint32_t crypt_method = 0;
900 
901     /*
902      * Backing files are read-only which makes all of their metadata immutable,
903      * that means we don't have to worry about reopening them here.
904      */
905 
906     if (s->crypt_method) {
907         crypt_method = s->crypt_method;
908         memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key));
909         memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key));
910     }
911 
912     qcow2_close(bs);
913 
914     memset(s, 0, sizeof(BDRVQcowState));
915     qcow2_open(bs, flags);
916 
917     if (crypt_method) {
918         s->crypt_method = crypt_method;
919         memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key));
920         memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key));
921     }
922 }
923 
924 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
925     size_t len, size_t buflen)
926 {
927     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
928     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
929 
930     if (buflen < ext_len) {
931         return -ENOSPC;
932     }
933 
934     *ext_backing_fmt = (QCowExtension) {
935         .magic  = cpu_to_be32(magic),
936         .len    = cpu_to_be32(len),
937     };
938     memcpy(buf + sizeof(QCowExtension), s, len);
939 
940     return ext_len;
941 }
942 
943 /*
944  * Updates the qcow2 header, including the variable length parts of it, i.e.
945  * the backing file name and all extensions. qcow2 was not designed to allow
946  * such changes, so if we run out of space (we can only use the first cluster)
947  * this function may fail.
948  *
949  * Returns 0 on success, -errno in error cases.
950  */
951 int qcow2_update_header(BlockDriverState *bs)
952 {
953     BDRVQcowState *s = bs->opaque;
954     QCowHeader *header;
955     char *buf;
956     size_t buflen = s->cluster_size;
957     int ret;
958     uint64_t total_size;
959     uint32_t refcount_table_clusters;
960     size_t header_length;
961     Qcow2UnknownHeaderExtension *uext;
962 
963     buf = qemu_blockalign(bs, buflen);
964 
965     /* Header structure */
966     header = (QCowHeader*) buf;
967 
968     if (buflen < sizeof(*header)) {
969         ret = -ENOSPC;
970         goto fail;
971     }
972 
973     header_length = sizeof(*header) + s->unknown_header_fields_size;
974     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
975     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
976 
977     *header = (QCowHeader) {
978         /* Version 2 fields */
979         .magic                  = cpu_to_be32(QCOW_MAGIC),
980         .version                = cpu_to_be32(s->qcow_version),
981         .backing_file_offset    = 0,
982         .backing_file_size      = 0,
983         .cluster_bits           = cpu_to_be32(s->cluster_bits),
984         .size                   = cpu_to_be64(total_size),
985         .crypt_method           = cpu_to_be32(s->crypt_method_header),
986         .l1_size                = cpu_to_be32(s->l1_size),
987         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
988         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
989         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
990         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
991         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
992 
993         /* Version 3 fields */
994         .incompatible_features  = cpu_to_be64(s->incompatible_features),
995         .compatible_features    = cpu_to_be64(s->compatible_features),
996         .autoclear_features     = cpu_to_be64(s->autoclear_features),
997         .refcount_order         = cpu_to_be32(3 + REFCOUNT_SHIFT),
998         .header_length          = cpu_to_be32(header_length),
999     };
1000 
1001     /* For older versions, write a shorter header */
1002     switch (s->qcow_version) {
1003     case 2:
1004         ret = offsetof(QCowHeader, incompatible_features);
1005         break;
1006     case 3:
1007         ret = sizeof(*header);
1008         break;
1009     default:
1010         ret = -EINVAL;
1011         goto fail;
1012     }
1013 
1014     buf += ret;
1015     buflen -= ret;
1016     memset(buf, 0, buflen);
1017 
1018     /* Preserve any unknown field in the header */
1019     if (s->unknown_header_fields_size) {
1020         if (buflen < s->unknown_header_fields_size) {
1021             ret = -ENOSPC;
1022             goto fail;
1023         }
1024 
1025         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
1026         buf += s->unknown_header_fields_size;
1027         buflen -= s->unknown_header_fields_size;
1028     }
1029 
1030     /* Backing file format header extension */
1031     if (*bs->backing_format) {
1032         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
1033                              bs->backing_format, strlen(bs->backing_format),
1034                              buflen);
1035         if (ret < 0) {
1036             goto fail;
1037         }
1038 
1039         buf += ret;
1040         buflen -= ret;
1041     }
1042 
1043     /* Feature table */
1044     Qcow2Feature features[] = {
1045         {
1046             .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
1047             .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
1048             .name = "dirty bit",
1049         },
1050         {
1051             .type = QCOW2_FEAT_TYPE_COMPATIBLE,
1052             .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
1053             .name = "lazy refcounts",
1054         },
1055     };
1056 
1057     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
1058                          features, sizeof(features), buflen);
1059     if (ret < 0) {
1060         goto fail;
1061     }
1062     buf += ret;
1063     buflen -= ret;
1064 
1065     /* Keep unknown header extensions */
1066     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
1067         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
1068         if (ret < 0) {
1069             goto fail;
1070         }
1071 
1072         buf += ret;
1073         buflen -= ret;
1074     }
1075 
1076     /* End of header extensions */
1077     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
1078     if (ret < 0) {
1079         goto fail;
1080     }
1081 
1082     buf += ret;
1083     buflen -= ret;
1084 
1085     /* Backing file name */
1086     if (*bs->backing_file) {
1087         size_t backing_file_len = strlen(bs->backing_file);
1088 
1089         if (buflen < backing_file_len) {
1090             ret = -ENOSPC;
1091             goto fail;
1092         }
1093 
1094         /* Using strncpy is ok here, since buf is not NUL-terminated. */
1095         strncpy(buf, bs->backing_file, buflen);
1096 
1097         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
1098         header->backing_file_size   = cpu_to_be32(backing_file_len);
1099     }
1100 
1101     /* Write the new header */
1102     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
1103     if (ret < 0) {
1104         goto fail;
1105     }
1106 
1107     ret = 0;
1108 fail:
1109     qemu_vfree(header);
1110     return ret;
1111 }
1112 
1113 static int qcow2_change_backing_file(BlockDriverState *bs,
1114     const char *backing_file, const char *backing_fmt)
1115 {
1116     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1117     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1118 
1119     return qcow2_update_header(bs);
1120 }
1121 
1122 static int preallocate(BlockDriverState *bs)
1123 {
1124     uint64_t nb_sectors;
1125     uint64_t offset;
1126     uint64_t host_offset = 0;
1127     int num;
1128     int ret;
1129     QCowL2Meta *meta;
1130 
1131     nb_sectors = bdrv_getlength(bs) >> 9;
1132     offset = 0;
1133 
1134     while (nb_sectors) {
1135         num = MIN(nb_sectors, INT_MAX >> 9);
1136         ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num,
1137                                          &host_offset, &meta);
1138         if (ret < 0) {
1139             return ret;
1140         }
1141 
1142         ret = qcow2_alloc_cluster_link_l2(bs, meta);
1143         if (ret < 0) {
1144             qcow2_free_any_clusters(bs, meta->alloc_offset, meta->nb_clusters);
1145             return ret;
1146         }
1147 
1148         /* There are no dependent requests, but we need to remove our request
1149          * from the list of in-flight requests */
1150         if (meta != NULL) {
1151             QLIST_REMOVE(meta, next_in_flight);
1152         }
1153 
1154         /* TODO Preallocate data if requested */
1155 
1156         nb_sectors -= num;
1157         offset += num << 9;
1158     }
1159 
1160     /*
1161      * It is expected that the image file is large enough to actually contain
1162      * all of the allocated clusters (otherwise we get failing reads after
1163      * EOF). Extend the image to the last allocated sector.
1164      */
1165     if (host_offset != 0) {
1166         uint8_t buf[512];
1167         memset(buf, 0, 512);
1168         ret = bdrv_write(bs->file, (host_offset >> 9) + num - 1, buf, 1);
1169         if (ret < 0) {
1170             return ret;
1171         }
1172     }
1173 
1174     return 0;
1175 }
1176 
1177 static int qcow2_create2(const char *filename, int64_t total_size,
1178                          const char *backing_file, const char *backing_format,
1179                          int flags, size_t cluster_size, int prealloc,
1180                          QEMUOptionParameter *options, int version)
1181 {
1182     /* Calculate cluster_bits */
1183     int cluster_bits;
1184     cluster_bits = ffs(cluster_size) - 1;
1185     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
1186         (1 << cluster_bits) != cluster_size)
1187     {
1188         error_report(
1189             "Cluster size must be a power of two between %d and %dk",
1190             1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
1191         return -EINVAL;
1192     }
1193 
1194     /*
1195      * Open the image file and write a minimal qcow2 header.
1196      *
1197      * We keep things simple and start with a zero-sized image. We also
1198      * do without refcount blocks or a L1 table for now. We'll fix the
1199      * inconsistency later.
1200      *
1201      * We do need a refcount table because growing the refcount table means
1202      * allocating two new refcount blocks - the seconds of which would be at
1203      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
1204      * size for any qcow2 image.
1205      */
1206     BlockDriverState* bs;
1207     QCowHeader header;
1208     uint8_t* refcount_table;
1209     int ret;
1210 
1211     ret = bdrv_create_file(filename, options);
1212     if (ret < 0) {
1213         return ret;
1214     }
1215 
1216     ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR);
1217     if (ret < 0) {
1218         return ret;
1219     }
1220 
1221     /* Write the header */
1222     memset(&header, 0, sizeof(header));
1223     header.magic = cpu_to_be32(QCOW_MAGIC);
1224     header.version = cpu_to_be32(version);
1225     header.cluster_bits = cpu_to_be32(cluster_bits);
1226     header.size = cpu_to_be64(0);
1227     header.l1_table_offset = cpu_to_be64(0);
1228     header.l1_size = cpu_to_be32(0);
1229     header.refcount_table_offset = cpu_to_be64(cluster_size);
1230     header.refcount_table_clusters = cpu_to_be32(1);
1231     header.refcount_order = cpu_to_be32(3 + REFCOUNT_SHIFT);
1232     header.header_length = cpu_to_be32(sizeof(header));
1233 
1234     if (flags & BLOCK_FLAG_ENCRYPT) {
1235         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1236     } else {
1237         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1238     }
1239 
1240     if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
1241         header.compatible_features |=
1242             cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
1243     }
1244 
1245     ret = bdrv_pwrite(bs, 0, &header, sizeof(header));
1246     if (ret < 0) {
1247         goto out;
1248     }
1249 
1250     /* Write an empty refcount table */
1251     refcount_table = g_malloc0(cluster_size);
1252     ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size);
1253     g_free(refcount_table);
1254 
1255     if (ret < 0) {
1256         goto out;
1257     }
1258 
1259     bdrv_close(bs);
1260 
1261     /*
1262      * And now open the image and make it consistent first (i.e. increase the
1263      * refcount of the cluster that is occupied by the header and the refcount
1264      * table)
1265      */
1266     BlockDriver* drv = bdrv_find_format("qcow2");
1267     assert(drv != NULL);
1268     ret = bdrv_open(bs, filename,
1269         BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv);
1270     if (ret < 0) {
1271         goto out;
1272     }
1273 
1274     ret = qcow2_alloc_clusters(bs, 2 * cluster_size);
1275     if (ret < 0) {
1276         goto out;
1277 
1278     } else if (ret != 0) {
1279         error_report("Huh, first cluster in empty image is already in use?");
1280         abort();
1281     }
1282 
1283     /* Okay, now that we have a valid image, let's give it the right size */
1284     ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE);
1285     if (ret < 0) {
1286         goto out;
1287     }
1288 
1289     /* Want a backing file? There you go.*/
1290     if (backing_file) {
1291         ret = bdrv_change_backing_file(bs, backing_file, backing_format);
1292         if (ret < 0) {
1293             goto out;
1294         }
1295     }
1296 
1297     /* And if we're supposed to preallocate metadata, do that now */
1298     if (prealloc) {
1299         BDRVQcowState *s = bs->opaque;
1300         qemu_co_mutex_lock(&s->lock);
1301         ret = preallocate(bs);
1302         qemu_co_mutex_unlock(&s->lock);
1303         if (ret < 0) {
1304             goto out;
1305         }
1306     }
1307 
1308     ret = 0;
1309 out:
1310     bdrv_delete(bs);
1311     return ret;
1312 }
1313 
1314 static int qcow2_create(const char *filename, QEMUOptionParameter *options)
1315 {
1316     const char *backing_file = NULL;
1317     const char *backing_fmt = NULL;
1318     uint64_t sectors = 0;
1319     int flags = 0;
1320     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
1321     int prealloc = 0;
1322     int version = 2;
1323 
1324     /* Read out options */
1325     while (options && options->name) {
1326         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1327             sectors = options->value.n / 512;
1328         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1329             backing_file = options->value.s;
1330         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
1331             backing_fmt = options->value.s;
1332         } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
1333             flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
1334         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
1335             if (options->value.n) {
1336                 cluster_size = options->value.n;
1337             }
1338         } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
1339             if (!options->value.s || !strcmp(options->value.s, "off")) {
1340                 prealloc = 0;
1341             } else if (!strcmp(options->value.s, "metadata")) {
1342                 prealloc = 1;
1343             } else {
1344                 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
1345                     options->value.s);
1346                 return -EINVAL;
1347             }
1348         } else if (!strcmp(options->name, BLOCK_OPT_COMPAT_LEVEL)) {
1349             if (!options->value.s || !strcmp(options->value.s, "0.10")) {
1350                 version = 2;
1351             } else if (!strcmp(options->value.s, "1.1")) {
1352                 version = 3;
1353             } else {
1354                 fprintf(stderr, "Invalid compatibility level: '%s'\n",
1355                     options->value.s);
1356                 return -EINVAL;
1357             }
1358         } else if (!strcmp(options->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
1359             flags |= options->value.n ? BLOCK_FLAG_LAZY_REFCOUNTS : 0;
1360         }
1361         options++;
1362     }
1363 
1364     if (backing_file && prealloc) {
1365         fprintf(stderr, "Backing file and preallocation cannot be used at "
1366             "the same time\n");
1367         return -EINVAL;
1368     }
1369 
1370     if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
1371         fprintf(stderr, "Lazy refcounts only supported with compatibility "
1372                 "level 1.1 and above (use compat=1.1 or greater)\n");
1373         return -EINVAL;
1374     }
1375 
1376     return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags,
1377                          cluster_size, prealloc, options, version);
1378 }
1379 
1380 static int qcow2_make_empty(BlockDriverState *bs)
1381 {
1382 #if 0
1383     /* XXX: not correct */
1384     BDRVQcowState *s = bs->opaque;
1385     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1386     int ret;
1387 
1388     memset(s->l1_table, 0, l1_length);
1389     if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
1390         return -1;
1391     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
1392     if (ret < 0)
1393         return ret;
1394 
1395     l2_cache_reset(bs);
1396 #endif
1397     return 0;
1398 }
1399 
1400 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
1401     int64_t sector_num, int nb_sectors)
1402 {
1403     int ret;
1404     BDRVQcowState *s = bs->opaque;
1405 
1406     /* Emulate misaligned zero writes */
1407     if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
1408         return -ENOTSUP;
1409     }
1410 
1411     /* Whatever is left can use real zero clusters */
1412     qemu_co_mutex_lock(&s->lock);
1413     ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1414         nb_sectors);
1415     qemu_co_mutex_unlock(&s->lock);
1416 
1417     return ret;
1418 }
1419 
1420 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
1421     int64_t sector_num, int nb_sectors)
1422 {
1423     int ret;
1424     BDRVQcowState *s = bs->opaque;
1425 
1426     qemu_co_mutex_lock(&s->lock);
1427     ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1428         nb_sectors);
1429     qemu_co_mutex_unlock(&s->lock);
1430     return ret;
1431 }
1432 
1433 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
1434 {
1435     BDRVQcowState *s = bs->opaque;
1436     int ret, new_l1_size;
1437 
1438     if (offset & 511) {
1439         error_report("The new size must be a multiple of 512");
1440         return -EINVAL;
1441     }
1442 
1443     /* cannot proceed if image has snapshots */
1444     if (s->nb_snapshots) {
1445         error_report("Can't resize an image which has snapshots");
1446         return -ENOTSUP;
1447     }
1448 
1449     /* shrinking is currently not supported */
1450     if (offset < bs->total_sectors * 512) {
1451         error_report("qcow2 doesn't support shrinking images yet");
1452         return -ENOTSUP;
1453     }
1454 
1455     new_l1_size = size_to_l1(s, offset);
1456     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
1457     if (ret < 0) {
1458         return ret;
1459     }
1460 
1461     /* write updated header.size */
1462     offset = cpu_to_be64(offset);
1463     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
1464                            &offset, sizeof(uint64_t));
1465     if (ret < 0) {
1466         return ret;
1467     }
1468 
1469     s->l1_vm_state_index = new_l1_size;
1470     return 0;
1471 }
1472 
1473 /* XXX: put compressed sectors first, then all the cluster aligned
1474    tables to avoid losing bytes in alignment */
1475 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
1476                                   const uint8_t *buf, int nb_sectors)
1477 {
1478     BDRVQcowState *s = bs->opaque;
1479     z_stream strm;
1480     int ret, out_len;
1481     uint8_t *out_buf;
1482     uint64_t cluster_offset;
1483 
1484     if (nb_sectors == 0) {
1485         /* align end of file to a sector boundary to ease reading with
1486            sector based I/Os */
1487         cluster_offset = bdrv_getlength(bs->file);
1488         cluster_offset = (cluster_offset + 511) & ~511;
1489         bdrv_truncate(bs->file, cluster_offset);
1490         return 0;
1491     }
1492 
1493     if (nb_sectors != s->cluster_sectors)
1494         return -EINVAL;
1495 
1496     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1497 
1498     /* best compression, small window, no zlib header */
1499     memset(&strm, 0, sizeof(strm));
1500     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1501                        Z_DEFLATED, -12,
1502                        9, Z_DEFAULT_STRATEGY);
1503     if (ret != 0) {
1504         ret = -EINVAL;
1505         goto fail;
1506     }
1507 
1508     strm.avail_in = s->cluster_size;
1509     strm.next_in = (uint8_t *)buf;
1510     strm.avail_out = s->cluster_size;
1511     strm.next_out = out_buf;
1512 
1513     ret = deflate(&strm, Z_FINISH);
1514     if (ret != Z_STREAM_END && ret != Z_OK) {
1515         deflateEnd(&strm);
1516         ret = -EINVAL;
1517         goto fail;
1518     }
1519     out_len = strm.next_out - out_buf;
1520 
1521     deflateEnd(&strm);
1522 
1523     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1524         /* could not compress: write normal cluster */
1525         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1526         if (ret < 0) {
1527             goto fail;
1528         }
1529     } else {
1530         cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1531             sector_num << 9, out_len);
1532         if (!cluster_offset) {
1533             ret = -EIO;
1534             goto fail;
1535         }
1536         cluster_offset &= s->cluster_offset_mask;
1537         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1538         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
1539         if (ret < 0) {
1540             goto fail;
1541         }
1542     }
1543 
1544     ret = 0;
1545 fail:
1546     g_free(out_buf);
1547     return ret;
1548 }
1549 
1550 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
1551 {
1552     BDRVQcowState *s = bs->opaque;
1553     int ret;
1554 
1555     qemu_co_mutex_lock(&s->lock);
1556     ret = qcow2_cache_flush(bs, s->l2_table_cache);
1557     if (ret < 0) {
1558         qemu_co_mutex_unlock(&s->lock);
1559         return ret;
1560     }
1561 
1562     if (qcow2_need_accurate_refcounts(s)) {
1563         ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1564         if (ret < 0) {
1565             qemu_co_mutex_unlock(&s->lock);
1566             return ret;
1567         }
1568     }
1569     qemu_co_mutex_unlock(&s->lock);
1570 
1571     return 0;
1572 }
1573 
1574 static int64_t qcow2_vm_state_offset(BDRVQcowState *s)
1575 {
1576 	return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1577 }
1578 
1579 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1580 {
1581     BDRVQcowState *s = bs->opaque;
1582     bdi->cluster_size = s->cluster_size;
1583     bdi->vm_state_offset = qcow2_vm_state_offset(s);
1584     return 0;
1585 }
1586 
1587 #if 0
1588 static void dump_refcounts(BlockDriverState *bs)
1589 {
1590     BDRVQcowState *s = bs->opaque;
1591     int64_t nb_clusters, k, k1, size;
1592     int refcount;
1593 
1594     size = bdrv_getlength(bs->file);
1595     nb_clusters = size_to_clusters(s, size);
1596     for(k = 0; k < nb_clusters;) {
1597         k1 = k;
1598         refcount = get_refcount(bs, k);
1599         k++;
1600         while (k < nb_clusters && get_refcount(bs, k) == refcount)
1601             k++;
1602         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
1603                k - k1);
1604     }
1605 }
1606 #endif
1607 
1608 static int qcow2_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1609                               int64_t pos, int size)
1610 {
1611     BDRVQcowState *s = bs->opaque;
1612     int growable = bs->growable;
1613     int ret;
1614 
1615     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
1616     bs->growable = 1;
1617     ret = bdrv_pwrite(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1618     bs->growable = growable;
1619 
1620     return ret;
1621 }
1622 
1623 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1624                               int64_t pos, int size)
1625 {
1626     BDRVQcowState *s = bs->opaque;
1627     int growable = bs->growable;
1628     int ret;
1629 
1630     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
1631     bs->growable = 1;
1632     ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1633     bs->growable = growable;
1634 
1635     return ret;
1636 }
1637 
1638 static QEMUOptionParameter qcow2_create_options[] = {
1639     {
1640         .name = BLOCK_OPT_SIZE,
1641         .type = OPT_SIZE,
1642         .help = "Virtual disk size"
1643     },
1644     {
1645         .name = BLOCK_OPT_COMPAT_LEVEL,
1646         .type = OPT_STRING,
1647         .help = "Compatibility level (0.10 or 1.1)"
1648     },
1649     {
1650         .name = BLOCK_OPT_BACKING_FILE,
1651         .type = OPT_STRING,
1652         .help = "File name of a base image"
1653     },
1654     {
1655         .name = BLOCK_OPT_BACKING_FMT,
1656         .type = OPT_STRING,
1657         .help = "Image format of the base image"
1658     },
1659     {
1660         .name = BLOCK_OPT_ENCRYPT,
1661         .type = OPT_FLAG,
1662         .help = "Encrypt the image"
1663     },
1664     {
1665         .name = BLOCK_OPT_CLUSTER_SIZE,
1666         .type = OPT_SIZE,
1667         .help = "qcow2 cluster size",
1668         .value = { .n = DEFAULT_CLUSTER_SIZE },
1669     },
1670     {
1671         .name = BLOCK_OPT_PREALLOC,
1672         .type = OPT_STRING,
1673         .help = "Preallocation mode (allowed values: off, metadata)"
1674     },
1675     {
1676         .name = BLOCK_OPT_LAZY_REFCOUNTS,
1677         .type = OPT_FLAG,
1678         .help = "Postpone refcount updates",
1679     },
1680     { NULL }
1681 };
1682 
1683 static BlockDriver bdrv_qcow2 = {
1684     .format_name        = "qcow2",
1685     .instance_size      = sizeof(BDRVQcowState),
1686     .bdrv_probe         = qcow2_probe,
1687     .bdrv_open          = qcow2_open,
1688     .bdrv_close         = qcow2_close,
1689     .bdrv_reopen_prepare  = qcow2_reopen_prepare,
1690     .bdrv_create        = qcow2_create,
1691     .bdrv_co_is_allocated = qcow2_co_is_allocated,
1692     .bdrv_set_key       = qcow2_set_key,
1693     .bdrv_make_empty    = qcow2_make_empty,
1694 
1695     .bdrv_co_readv          = qcow2_co_readv,
1696     .bdrv_co_writev         = qcow2_co_writev,
1697     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
1698 
1699     .bdrv_co_write_zeroes   = qcow2_co_write_zeroes,
1700     .bdrv_co_discard        = qcow2_co_discard,
1701     .bdrv_truncate          = qcow2_truncate,
1702     .bdrv_write_compressed  = qcow2_write_compressed,
1703 
1704     .bdrv_snapshot_create   = qcow2_snapshot_create,
1705     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
1706     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
1707     .bdrv_snapshot_list     = qcow2_snapshot_list,
1708     .bdrv_snapshot_load_tmp     = qcow2_snapshot_load_tmp,
1709     .bdrv_get_info      = qcow2_get_info,
1710 
1711     .bdrv_save_vmstate    = qcow2_save_vmstate,
1712     .bdrv_load_vmstate    = qcow2_load_vmstate,
1713 
1714     .bdrv_change_backing_file   = qcow2_change_backing_file,
1715 
1716     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
1717 
1718     .create_options = qcow2_create_options,
1719     .bdrv_check = qcow2_check,
1720 };
1721 
1722 static void bdrv_qcow2_init(void)
1723 {
1724     bdrv_register(&bdrv_qcow2);
1725 }
1726 
1727 block_init(bdrv_qcow2_init);
1728