xref: /openbmc/qemu/block/qcow2.c (revision bf3f363a5b52e6f246bcab74dafffa61d6e4d2a7)
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 #define  QCOW2_EXT_MAGIC_END 0
56 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
57 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
58 
59 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
60 {
61     const QCowHeader *cow_header = (const void *)buf;
62 
63     if (buf_size >= sizeof(QCowHeader) &&
64         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
65         be32_to_cpu(cow_header->version) >= 2)
66         return 100;
67     else
68         return 0;
69 }
70 
71 
72 /*
73  * read qcow2 extension and fill bs
74  * start reading from start_offset
75  * finish reading upon magic of value 0 or when end_offset reached
76  * unknown magic is skipped (future extension this version knows nothing about)
77  * return 0 upon success, non-0 otherwise
78  */
79 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
80                                  uint64_t end_offset, void **p_feature_table)
81 {
82     BDRVQcowState *s = bs->opaque;
83     QCowExtension ext;
84     uint64_t offset;
85     int ret;
86 
87 #ifdef DEBUG_EXT
88     printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
89 #endif
90     offset = start_offset;
91     while (offset < end_offset) {
92 
93 #ifdef DEBUG_EXT
94         /* Sanity check */
95         if (offset > s->cluster_size)
96             printf("qcow2_read_extension: suspicious offset %lu\n", offset);
97 
98         printf("attempting to read extended header in offset %lu\n", offset);
99 #endif
100 
101         if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) {
102             fprintf(stderr, "qcow2_read_extension: ERROR: "
103                     "pread fail from offset %" PRIu64 "\n",
104                     offset);
105             return 1;
106         }
107         be32_to_cpus(&ext.magic);
108         be32_to_cpus(&ext.len);
109         offset += sizeof(ext);
110 #ifdef DEBUG_EXT
111         printf("ext.magic = 0x%x\n", ext.magic);
112 #endif
113         if (ext.len > end_offset - offset) {
114             error_report("Header extension too large");
115             return -EINVAL;
116         }
117 
118         switch (ext.magic) {
119         case QCOW2_EXT_MAGIC_END:
120             return 0;
121 
122         case QCOW2_EXT_MAGIC_BACKING_FORMAT:
123             if (ext.len >= sizeof(bs->backing_format)) {
124                 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
125                         " (>=%zu)\n",
126                         ext.len, sizeof(bs->backing_format));
127                 return 2;
128             }
129             if (bdrv_pread(bs->file, offset , bs->backing_format,
130                            ext.len) != ext.len)
131                 return 3;
132             bs->backing_format[ext.len] = '\0';
133 #ifdef DEBUG_EXT
134             printf("Qcow2: Got format extension %s\n", bs->backing_format);
135 #endif
136             break;
137 
138         case QCOW2_EXT_MAGIC_FEATURE_TABLE:
139             if (p_feature_table != NULL) {
140                 void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
141                 ret = bdrv_pread(bs->file, offset , feature_table, ext.len);
142                 if (ret < 0) {
143                     return ret;
144                 }
145 
146                 *p_feature_table = feature_table;
147             }
148             break;
149 
150         default:
151             /* unknown magic - save it in case we need to rewrite the header */
152             {
153                 Qcow2UnknownHeaderExtension *uext;
154 
155                 uext = g_malloc0(sizeof(*uext)  + ext.len);
156                 uext->magic = ext.magic;
157                 uext->len = ext.len;
158                 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
159 
160                 ret = bdrv_pread(bs->file, offset , uext->data, uext->len);
161                 if (ret < 0) {
162                     return ret;
163                 }
164             }
165             break;
166         }
167 
168         offset += ((ext.len + 7) & ~7);
169     }
170 
171     return 0;
172 }
173 
174 static void cleanup_unknown_header_ext(BlockDriverState *bs)
175 {
176     BDRVQcowState *s = bs->opaque;
177     Qcow2UnknownHeaderExtension *uext, *next;
178 
179     QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
180         QLIST_REMOVE(uext, next);
181         g_free(uext);
182     }
183 }
184 
185 static void GCC_FMT_ATTR(2, 3) report_unsupported(BlockDriverState *bs,
186     const char *fmt, ...)
187 {
188     char msg[64];
189     va_list ap;
190 
191     va_start(ap, fmt);
192     vsnprintf(msg, sizeof(msg), fmt, ap);
193     va_end(ap);
194 
195     qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
196         bs->device_name, "qcow2", msg);
197 }
198 
199 static void report_unsupported_feature(BlockDriverState *bs,
200     Qcow2Feature *table, uint64_t mask)
201 {
202     while (table && table->name[0] != '\0') {
203         if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
204             if (mask & (1 << table->bit)) {
205                 report_unsupported(bs, "%.46s",table->name);
206                 mask &= ~(1 << table->bit);
207             }
208         }
209         table++;
210     }
211 
212     if (mask) {
213         report_unsupported(bs, "Unknown incompatible feature: %" PRIx64, mask);
214     }
215 }
216 
217 static int qcow2_open(BlockDriverState *bs, int flags)
218 {
219     BDRVQcowState *s = bs->opaque;
220     int len, i, ret = 0;
221     QCowHeader header;
222     uint64_t ext_end;
223     bool writethrough;
224 
225     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
226     if (ret < 0) {
227         goto fail;
228     }
229     be32_to_cpus(&header.magic);
230     be32_to_cpus(&header.version);
231     be64_to_cpus(&header.backing_file_offset);
232     be32_to_cpus(&header.backing_file_size);
233     be64_to_cpus(&header.size);
234     be32_to_cpus(&header.cluster_bits);
235     be32_to_cpus(&header.crypt_method);
236     be64_to_cpus(&header.l1_table_offset);
237     be32_to_cpus(&header.l1_size);
238     be64_to_cpus(&header.refcount_table_offset);
239     be32_to_cpus(&header.refcount_table_clusters);
240     be64_to_cpus(&header.snapshots_offset);
241     be32_to_cpus(&header.nb_snapshots);
242 
243     if (header.magic != QCOW_MAGIC) {
244         ret = -EINVAL;
245         goto fail;
246     }
247     if (header.version < 2 || header.version > 3) {
248         report_unsupported(bs, "QCOW version %d", header.version);
249         ret = -ENOTSUP;
250         goto fail;
251     }
252 
253     s->qcow_version = header.version;
254 
255     /* Initialise version 3 header fields */
256     if (header.version == 2) {
257         header.incompatible_features    = 0;
258         header.compatible_features      = 0;
259         header.autoclear_features       = 0;
260         header.refcount_order           = 4;
261         header.header_length            = 72;
262     } else {
263         be64_to_cpus(&header.incompatible_features);
264         be64_to_cpus(&header.compatible_features);
265         be64_to_cpus(&header.autoclear_features);
266         be32_to_cpus(&header.refcount_order);
267         be32_to_cpus(&header.header_length);
268     }
269 
270     if (header.header_length > sizeof(header)) {
271         s->unknown_header_fields_size = header.header_length - sizeof(header);
272         s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
273         ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
274                          s->unknown_header_fields_size);
275         if (ret < 0) {
276             goto fail;
277         }
278     }
279 
280     if (header.backing_file_offset) {
281         ext_end = header.backing_file_offset;
282     } else {
283         ext_end = 1 << header.cluster_bits;
284     }
285 
286     /* Handle feature bits */
287     s->incompatible_features    = header.incompatible_features;
288     s->compatible_features      = header.compatible_features;
289     s->autoclear_features       = header.autoclear_features;
290 
291     if (s->incompatible_features != 0) {
292         void *feature_table = NULL;
293         qcow2_read_extensions(bs, header.header_length, ext_end,
294                               &feature_table);
295         report_unsupported_feature(bs, feature_table,
296                                    s->incompatible_features);
297         ret = -ENOTSUP;
298         goto fail;
299     }
300 
301     if (!bs->read_only && s->autoclear_features != 0) {
302         s->autoclear_features = 0;
303         qcow2_update_header(bs);
304     }
305 
306     /* Check support for various header values */
307     if (header.refcount_order != 4) {
308         report_unsupported(bs, "%d bit reference counts",
309                            1 << header.refcount_order);
310         ret = -ENOTSUP;
311         goto fail;
312     }
313 
314     if (header.cluster_bits < MIN_CLUSTER_BITS ||
315         header.cluster_bits > MAX_CLUSTER_BITS) {
316         ret = -EINVAL;
317         goto fail;
318     }
319     if (header.crypt_method > QCOW_CRYPT_AES) {
320         ret = -EINVAL;
321         goto fail;
322     }
323     s->crypt_method_header = header.crypt_method;
324     if (s->crypt_method_header) {
325         bs->encrypted = 1;
326     }
327     s->cluster_bits = header.cluster_bits;
328     s->cluster_size = 1 << s->cluster_bits;
329     s->cluster_sectors = 1 << (s->cluster_bits - 9);
330     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
331     s->l2_size = 1 << s->l2_bits;
332     bs->total_sectors = header.size / 512;
333     s->csize_shift = (62 - (s->cluster_bits - 8));
334     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
335     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
336     s->refcount_table_offset = header.refcount_table_offset;
337     s->refcount_table_size =
338         header.refcount_table_clusters << (s->cluster_bits - 3);
339 
340     s->snapshots_offset = header.snapshots_offset;
341     s->nb_snapshots = header.nb_snapshots;
342 
343     /* read the level 1 table */
344     s->l1_size = header.l1_size;
345     s->l1_vm_state_index = size_to_l1(s, header.size);
346     /* the L1 table must contain at least enough entries to put
347        header.size bytes */
348     if (s->l1_size < s->l1_vm_state_index) {
349         ret = -EINVAL;
350         goto fail;
351     }
352     s->l1_table_offset = header.l1_table_offset;
353     if (s->l1_size > 0) {
354         s->l1_table = g_malloc0(
355             align_offset(s->l1_size * sizeof(uint64_t), 512));
356         ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
357                          s->l1_size * sizeof(uint64_t));
358         if (ret < 0) {
359             goto fail;
360         }
361         for(i = 0;i < s->l1_size; i++) {
362             be64_to_cpus(&s->l1_table[i]);
363         }
364     }
365 
366     /* alloc L2 table/refcount block cache */
367     writethrough = ((flags & BDRV_O_CACHE_WB) == 0);
368     s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE, writethrough);
369     s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE,
370         writethrough);
371 
372     s->cluster_cache = g_malloc(s->cluster_size);
373     /* one more sector for decompressed data alignment */
374     s->cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
375                                   + 512);
376     s->cluster_cache_offset = -1;
377     s->flags = flags;
378 
379     ret = qcow2_refcount_init(bs);
380     if (ret != 0) {
381         goto fail;
382     }
383 
384     QLIST_INIT(&s->cluster_allocs);
385 
386     /* read qcow2 extensions */
387     if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL)) {
388         ret = -EINVAL;
389         goto fail;
390     }
391 
392     /* read the backing file name */
393     if (header.backing_file_offset != 0) {
394         len = header.backing_file_size;
395         if (len > 1023) {
396             len = 1023;
397         }
398         ret = bdrv_pread(bs->file, header.backing_file_offset,
399                          bs->backing_file, len);
400         if (ret < 0) {
401             goto fail;
402         }
403         bs->backing_file[len] = '\0';
404     }
405 
406     ret = qcow2_read_snapshots(bs);
407     if (ret < 0) {
408         goto fail;
409     }
410 
411     /* Initialise locks */
412     qemu_co_mutex_init(&s->lock);
413 
414 #ifdef DEBUG_ALLOC
415     {
416         BdrvCheckResult result = {0};
417         qcow2_check_refcounts(bs, &result);
418     }
419 #endif
420     return ret;
421 
422  fail:
423     g_free(s->unknown_header_fields);
424     cleanup_unknown_header_ext(bs);
425     qcow2_free_snapshots(bs);
426     qcow2_refcount_close(bs);
427     g_free(s->l1_table);
428     if (s->l2_table_cache) {
429         qcow2_cache_destroy(bs, s->l2_table_cache);
430     }
431     g_free(s->cluster_cache);
432     qemu_vfree(s->cluster_data);
433     return ret;
434 }
435 
436 static int qcow2_set_key(BlockDriverState *bs, const char *key)
437 {
438     BDRVQcowState *s = bs->opaque;
439     uint8_t keybuf[16];
440     int len, i;
441 
442     memset(keybuf, 0, 16);
443     len = strlen(key);
444     if (len > 16)
445         len = 16;
446     /* XXX: we could compress the chars to 7 bits to increase
447        entropy */
448     for(i = 0;i < len;i++) {
449         keybuf[i] = key[i];
450     }
451     s->crypt_method = s->crypt_method_header;
452 
453     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
454         return -1;
455     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
456         return -1;
457 #if 0
458     /* test */
459     {
460         uint8_t in[16];
461         uint8_t out[16];
462         uint8_t tmp[16];
463         for(i=0;i<16;i++)
464             in[i] = i;
465         AES_encrypt(in, tmp, &s->aes_encrypt_key);
466         AES_decrypt(tmp, out, &s->aes_decrypt_key);
467         for(i = 0; i < 16; i++)
468             printf(" %02x", tmp[i]);
469         printf("\n");
470         for(i = 0; i < 16; i++)
471             printf(" %02x", out[i]);
472         printf("\n");
473     }
474 #endif
475     return 0;
476 }
477 
478 static int coroutine_fn qcow2_co_is_allocated(BlockDriverState *bs,
479         int64_t sector_num, int nb_sectors, int *pnum)
480 {
481     BDRVQcowState *s = bs->opaque;
482     uint64_t cluster_offset;
483     int ret;
484 
485     *pnum = nb_sectors;
486     /* FIXME We can get errors here, but the bdrv_co_is_allocated interface
487      * can't pass them on today */
488     qemu_co_mutex_lock(&s->lock);
489     ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
490     qemu_co_mutex_unlock(&s->lock);
491     if (ret < 0) {
492         *pnum = 0;
493     }
494 
495     return (cluster_offset != 0);
496 }
497 
498 /* handle reading after the end of the backing file */
499 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
500                   int64_t sector_num, int nb_sectors)
501 {
502     int n1;
503     if ((sector_num + nb_sectors) <= bs->total_sectors)
504         return nb_sectors;
505     if (sector_num >= bs->total_sectors)
506         n1 = 0;
507     else
508         n1 = bs->total_sectors - sector_num;
509 
510     qemu_iovec_memset_skip(qiov, 0, 512 * (nb_sectors - n1), 512 * n1);
511 
512     return n1;
513 }
514 
515 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
516                           int remaining_sectors, QEMUIOVector *qiov)
517 {
518     BDRVQcowState *s = bs->opaque;
519     int index_in_cluster, n1;
520     int ret;
521     int cur_nr_sectors; /* number of sectors in current iteration */
522     uint64_t cluster_offset = 0;
523     uint64_t bytes_done = 0;
524     QEMUIOVector hd_qiov;
525     uint8_t *cluster_data = NULL;
526 
527     qemu_iovec_init(&hd_qiov, qiov->niov);
528 
529     qemu_co_mutex_lock(&s->lock);
530 
531     while (remaining_sectors != 0) {
532 
533         /* prepare next request */
534         cur_nr_sectors = remaining_sectors;
535         if (s->crypt_method) {
536             cur_nr_sectors = MIN(cur_nr_sectors,
537                 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
538         }
539 
540         ret = qcow2_get_cluster_offset(bs, sector_num << 9,
541             &cur_nr_sectors, &cluster_offset);
542         if (ret < 0) {
543             goto fail;
544         }
545 
546         index_in_cluster = sector_num & (s->cluster_sectors - 1);
547 
548         qemu_iovec_reset(&hd_qiov);
549         qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
550             cur_nr_sectors * 512);
551 
552         switch (ret) {
553         case QCOW2_CLUSTER_UNALLOCATED:
554 
555             if (bs->backing_hd) {
556                 /* read from the base image */
557                 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov,
558                     sector_num, cur_nr_sectors);
559                 if (n1 > 0) {
560                     BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
561                     qemu_co_mutex_unlock(&s->lock);
562                     ret = bdrv_co_readv(bs->backing_hd, sector_num,
563                                         n1, &hd_qiov);
564                     qemu_co_mutex_lock(&s->lock);
565                     if (ret < 0) {
566                         goto fail;
567                     }
568                 }
569             } else {
570                 /* Note: in this case, no need to wait */
571                 qemu_iovec_memset(&hd_qiov, 0, 512 * cur_nr_sectors);
572             }
573             break;
574 
575         case QCOW2_CLUSTER_ZERO:
576             if (s->qcow_version < 3) {
577                 ret = -EIO;
578                 goto fail;
579             }
580             qemu_iovec_memset(&hd_qiov, 0, 512 * cur_nr_sectors);
581             break;
582 
583         case QCOW2_CLUSTER_COMPRESSED:
584             /* add AIO support for compressed blocks ? */
585             ret = qcow2_decompress_cluster(bs, cluster_offset);
586             if (ret < 0) {
587                 goto fail;
588             }
589 
590             qemu_iovec_from_buffer(&hd_qiov,
591                 s->cluster_cache + index_in_cluster * 512,
592                 512 * cur_nr_sectors);
593             break;
594 
595         case QCOW2_CLUSTER_NORMAL:
596             if ((cluster_offset & 511) != 0) {
597                 ret = -EIO;
598                 goto fail;
599             }
600 
601             if (s->crypt_method) {
602                 /*
603                  * For encrypted images, read everything into a temporary
604                  * contiguous buffer on which the AES functions can work.
605                  */
606                 if (!cluster_data) {
607                     cluster_data =
608                         qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
609                 }
610 
611                 assert(cur_nr_sectors <=
612                     QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
613                 qemu_iovec_reset(&hd_qiov);
614                 qemu_iovec_add(&hd_qiov, cluster_data,
615                     512 * cur_nr_sectors);
616             }
617 
618             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
619             qemu_co_mutex_unlock(&s->lock);
620             ret = bdrv_co_readv(bs->file,
621                                 (cluster_offset >> 9) + index_in_cluster,
622                                 cur_nr_sectors, &hd_qiov);
623             qemu_co_mutex_lock(&s->lock);
624             if (ret < 0) {
625                 goto fail;
626             }
627             if (s->crypt_method) {
628                 qcow2_encrypt_sectors(s, sector_num,  cluster_data,
629                     cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
630                 qemu_iovec_reset(&hd_qiov);
631                 qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
632                     cur_nr_sectors * 512);
633                 qemu_iovec_from_buffer(&hd_qiov, cluster_data,
634                     512 * cur_nr_sectors);
635             }
636             break;
637 
638         default:
639             g_assert_not_reached();
640             ret = -EIO;
641             goto fail;
642         }
643 
644         remaining_sectors -= cur_nr_sectors;
645         sector_num += cur_nr_sectors;
646         bytes_done += cur_nr_sectors * 512;
647     }
648     ret = 0;
649 
650 fail:
651     qemu_co_mutex_unlock(&s->lock);
652 
653     qemu_iovec_destroy(&hd_qiov);
654     qemu_vfree(cluster_data);
655 
656     return ret;
657 }
658 
659 static void run_dependent_requests(BDRVQcowState *s, QCowL2Meta *m)
660 {
661     /* Take the request off the list of running requests */
662     if (m->nb_clusters != 0) {
663         QLIST_REMOVE(m, next_in_flight);
664     }
665 
666     /* Restart all dependent requests */
667     if (!qemu_co_queue_empty(&m->dependent_requests)) {
668         qemu_co_mutex_unlock(&s->lock);
669         qemu_co_queue_restart_all(&m->dependent_requests);
670         qemu_co_mutex_lock(&s->lock);
671     }
672 }
673 
674 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
675                            int64_t sector_num,
676                            int remaining_sectors,
677                            QEMUIOVector *qiov)
678 {
679     BDRVQcowState *s = bs->opaque;
680     int index_in_cluster;
681     int n_end;
682     int ret;
683     int cur_nr_sectors; /* number of sectors in current iteration */
684     uint64_t cluster_offset;
685     QEMUIOVector hd_qiov;
686     uint64_t bytes_done = 0;
687     uint8_t *cluster_data = NULL;
688     QCowL2Meta l2meta = {
689         .nb_clusters = 0,
690     };
691 
692     trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
693                                  remaining_sectors);
694 
695     qemu_co_queue_init(&l2meta.dependent_requests);
696 
697     qemu_iovec_init(&hd_qiov, qiov->niov);
698 
699     s->cluster_cache_offset = -1; /* disable compressed cache */
700 
701     qemu_co_mutex_lock(&s->lock);
702 
703     while (remaining_sectors != 0) {
704 
705         trace_qcow2_writev_start_part(qemu_coroutine_self());
706         index_in_cluster = sector_num & (s->cluster_sectors - 1);
707         n_end = index_in_cluster + remaining_sectors;
708         if (s->crypt_method &&
709             n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) {
710             n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
711         }
712 
713         ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
714             index_in_cluster, n_end, &cur_nr_sectors, &l2meta);
715         if (ret < 0) {
716             goto fail;
717         }
718 
719         cluster_offset = l2meta.cluster_offset;
720         assert((cluster_offset & 511) == 0);
721 
722         qemu_iovec_reset(&hd_qiov);
723         qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
724             cur_nr_sectors * 512);
725 
726         if (s->crypt_method) {
727             if (!cluster_data) {
728                 cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS *
729                                                  s->cluster_size);
730             }
731 
732             assert(hd_qiov.size <=
733                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
734             qemu_iovec_to_buffer(&hd_qiov, cluster_data);
735 
736             qcow2_encrypt_sectors(s, sector_num, cluster_data,
737                 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
738 
739             qemu_iovec_reset(&hd_qiov);
740             qemu_iovec_add(&hd_qiov, cluster_data,
741                 cur_nr_sectors * 512);
742         }
743 
744         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
745         qemu_co_mutex_unlock(&s->lock);
746         trace_qcow2_writev_data(qemu_coroutine_self(),
747                                 (cluster_offset >> 9) + index_in_cluster);
748         ret = bdrv_co_writev(bs->file,
749                              (cluster_offset >> 9) + index_in_cluster,
750                              cur_nr_sectors, &hd_qiov);
751         qemu_co_mutex_lock(&s->lock);
752         if (ret < 0) {
753             goto fail;
754         }
755 
756         ret = qcow2_alloc_cluster_link_l2(bs, &l2meta);
757         if (ret < 0) {
758             goto fail;
759         }
760 
761         run_dependent_requests(s, &l2meta);
762 
763         remaining_sectors -= cur_nr_sectors;
764         sector_num += cur_nr_sectors;
765         bytes_done += cur_nr_sectors * 512;
766         trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
767     }
768     ret = 0;
769 
770 fail:
771     run_dependent_requests(s, &l2meta);
772 
773     qemu_co_mutex_unlock(&s->lock);
774 
775     qemu_iovec_destroy(&hd_qiov);
776     qemu_vfree(cluster_data);
777     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
778 
779     return ret;
780 }
781 
782 static void qcow2_close(BlockDriverState *bs)
783 {
784     BDRVQcowState *s = bs->opaque;
785     g_free(s->l1_table);
786 
787     qcow2_cache_flush(bs, s->l2_table_cache);
788     qcow2_cache_flush(bs, s->refcount_block_cache);
789 
790     qcow2_cache_destroy(bs, s->l2_table_cache);
791     qcow2_cache_destroy(bs, s->refcount_block_cache);
792 
793     g_free(s->unknown_header_fields);
794     cleanup_unknown_header_ext(bs);
795 
796     g_free(s->cluster_cache);
797     qemu_vfree(s->cluster_data);
798     qcow2_refcount_close(bs);
799     qcow2_free_snapshots(bs);
800 }
801 
802 static void qcow2_invalidate_cache(BlockDriverState *bs)
803 {
804     BDRVQcowState *s = bs->opaque;
805     int flags = s->flags;
806     AES_KEY aes_encrypt_key;
807     AES_KEY aes_decrypt_key;
808     uint32_t crypt_method = 0;
809 
810     /*
811      * Backing files are read-only which makes all of their metadata immutable,
812      * that means we don't have to worry about reopening them here.
813      */
814 
815     if (s->crypt_method) {
816         crypt_method = s->crypt_method;
817         memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key));
818         memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key));
819     }
820 
821     qcow2_close(bs);
822 
823     memset(s, 0, sizeof(BDRVQcowState));
824     qcow2_open(bs, flags);
825 
826     if (crypt_method) {
827         s->crypt_method = crypt_method;
828         memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key));
829         memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key));
830     }
831 }
832 
833 static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
834     size_t len, size_t buflen)
835 {
836     QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
837     size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
838 
839     if (buflen < ext_len) {
840         return -ENOSPC;
841     }
842 
843     *ext_backing_fmt = (QCowExtension) {
844         .magic  = cpu_to_be32(magic),
845         .len    = cpu_to_be32(len),
846     };
847     memcpy(buf + sizeof(QCowExtension), s, len);
848 
849     return ext_len;
850 }
851 
852 /*
853  * Updates the qcow2 header, including the variable length parts of it, i.e.
854  * the backing file name and all extensions. qcow2 was not designed to allow
855  * such changes, so if we run out of space (we can only use the first cluster)
856  * this function may fail.
857  *
858  * Returns 0 on success, -errno in error cases.
859  */
860 int qcow2_update_header(BlockDriverState *bs)
861 {
862     BDRVQcowState *s = bs->opaque;
863     QCowHeader *header;
864     char *buf;
865     size_t buflen = s->cluster_size;
866     int ret;
867     uint64_t total_size;
868     uint32_t refcount_table_clusters;
869     size_t header_length;
870     Qcow2UnknownHeaderExtension *uext;
871 
872     buf = qemu_blockalign(bs, buflen);
873 
874     /* Header structure */
875     header = (QCowHeader*) buf;
876 
877     if (buflen < sizeof(*header)) {
878         ret = -ENOSPC;
879         goto fail;
880     }
881 
882     header_length = sizeof(*header) + s->unknown_header_fields_size;
883     total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
884     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
885 
886     *header = (QCowHeader) {
887         /* Version 2 fields */
888         .magic                  = cpu_to_be32(QCOW_MAGIC),
889         .version                = cpu_to_be32(s->qcow_version),
890         .backing_file_offset    = 0,
891         .backing_file_size      = 0,
892         .cluster_bits           = cpu_to_be32(s->cluster_bits),
893         .size                   = cpu_to_be64(total_size),
894         .crypt_method           = cpu_to_be32(s->crypt_method_header),
895         .l1_size                = cpu_to_be32(s->l1_size),
896         .l1_table_offset        = cpu_to_be64(s->l1_table_offset),
897         .refcount_table_offset  = cpu_to_be64(s->refcount_table_offset),
898         .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
899         .nb_snapshots           = cpu_to_be32(s->nb_snapshots),
900         .snapshots_offset       = cpu_to_be64(s->snapshots_offset),
901 
902         /* Version 3 fields */
903         .incompatible_features  = cpu_to_be64(s->incompatible_features),
904         .compatible_features    = cpu_to_be64(s->compatible_features),
905         .autoclear_features     = cpu_to_be64(s->autoclear_features),
906         .refcount_order         = cpu_to_be32(3 + REFCOUNT_SHIFT),
907         .header_length          = cpu_to_be32(header_length),
908     };
909 
910     /* For older versions, write a shorter header */
911     switch (s->qcow_version) {
912     case 2:
913         ret = offsetof(QCowHeader, incompatible_features);
914         break;
915     case 3:
916         ret = sizeof(*header);
917         break;
918     default:
919         return -EINVAL;
920     }
921 
922     buf += ret;
923     buflen -= ret;
924     memset(buf, 0, buflen);
925 
926     /* Preserve any unknown field in the header */
927     if (s->unknown_header_fields_size) {
928         if (buflen < s->unknown_header_fields_size) {
929             ret = -ENOSPC;
930             goto fail;
931         }
932 
933         memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
934         buf += s->unknown_header_fields_size;
935         buflen -= s->unknown_header_fields_size;
936     }
937 
938     /* Backing file format header extension */
939     if (*bs->backing_format) {
940         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
941                              bs->backing_format, strlen(bs->backing_format),
942                              buflen);
943         if (ret < 0) {
944             goto fail;
945         }
946 
947         buf += ret;
948         buflen -= ret;
949     }
950 
951     /* Feature table */
952     Qcow2Feature features[] = {
953         /* no feature defined yet */
954     };
955 
956     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
957                          features, sizeof(features), buflen);
958     if (ret < 0) {
959         goto fail;
960     }
961     buf += ret;
962     buflen -= ret;
963 
964     /* Keep unknown header extensions */
965     QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
966         ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
967         if (ret < 0) {
968             goto fail;
969         }
970 
971         buf += ret;
972         buflen -= ret;
973     }
974 
975     /* End of header extensions */
976     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
977     if (ret < 0) {
978         goto fail;
979     }
980 
981     buf += ret;
982     buflen -= ret;
983 
984     /* Backing file name */
985     if (*bs->backing_file) {
986         size_t backing_file_len = strlen(bs->backing_file);
987 
988         if (buflen < backing_file_len) {
989             ret = -ENOSPC;
990             goto fail;
991         }
992 
993         strncpy(buf, bs->backing_file, buflen);
994 
995         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
996         header->backing_file_size   = cpu_to_be32(backing_file_len);
997     }
998 
999     /* Write the new header */
1000     ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
1001     if (ret < 0) {
1002         goto fail;
1003     }
1004 
1005     ret = 0;
1006 fail:
1007     qemu_vfree(header);
1008     return ret;
1009 }
1010 
1011 static int qcow2_change_backing_file(BlockDriverState *bs,
1012     const char *backing_file, const char *backing_fmt)
1013 {
1014     /* Backing file format doesn't make sense without a backing file */
1015     if (backing_fmt && !backing_file) {
1016         return -EINVAL;
1017     }
1018 
1019     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
1020     pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
1021 
1022     return qcow2_update_header(bs);
1023 }
1024 
1025 static int preallocate(BlockDriverState *bs)
1026 {
1027     uint64_t nb_sectors;
1028     uint64_t offset;
1029     int num;
1030     int ret;
1031     QCowL2Meta meta;
1032 
1033     nb_sectors = bdrv_getlength(bs) >> 9;
1034     offset = 0;
1035     qemu_co_queue_init(&meta.dependent_requests);
1036     meta.cluster_offset = 0;
1037 
1038     while (nb_sectors) {
1039         num = MIN(nb_sectors, INT_MAX >> 9);
1040         ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta);
1041         if (ret < 0) {
1042             return ret;
1043         }
1044 
1045         ret = qcow2_alloc_cluster_link_l2(bs, &meta);
1046         if (ret < 0) {
1047             qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters);
1048             return ret;
1049         }
1050 
1051         /* There are no dependent requests, but we need to remove our request
1052          * from the list of in-flight requests */
1053         run_dependent_requests(bs->opaque, &meta);
1054 
1055         /* TODO Preallocate data if requested */
1056 
1057         nb_sectors -= num;
1058         offset += num << 9;
1059     }
1060 
1061     /*
1062      * It is expected that the image file is large enough to actually contain
1063      * all of the allocated clusters (otherwise we get failing reads after
1064      * EOF). Extend the image to the last allocated sector.
1065      */
1066     if (meta.cluster_offset != 0) {
1067         uint8_t buf[512];
1068         memset(buf, 0, 512);
1069         ret = bdrv_write(bs->file, (meta.cluster_offset >> 9) + num - 1, buf, 1);
1070         if (ret < 0) {
1071             return ret;
1072         }
1073     }
1074 
1075     return 0;
1076 }
1077 
1078 static int qcow2_create2(const char *filename, int64_t total_size,
1079                          const char *backing_file, const char *backing_format,
1080                          int flags, size_t cluster_size, int prealloc,
1081                          QEMUOptionParameter *options, int version)
1082 {
1083     /* Calculate cluster_bits */
1084     int cluster_bits;
1085     cluster_bits = ffs(cluster_size) - 1;
1086     if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
1087         (1 << cluster_bits) != cluster_size)
1088     {
1089         error_report(
1090             "Cluster size must be a power of two between %d and %dk",
1091             1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
1092         return -EINVAL;
1093     }
1094 
1095     /*
1096      * Open the image file and write a minimal qcow2 header.
1097      *
1098      * We keep things simple and start with a zero-sized image. We also
1099      * do without refcount blocks or a L1 table for now. We'll fix the
1100      * inconsistency later.
1101      *
1102      * We do need a refcount table because growing the refcount table means
1103      * allocating two new refcount blocks - the seconds of which would be at
1104      * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
1105      * size for any qcow2 image.
1106      */
1107     BlockDriverState* bs;
1108     QCowHeader header;
1109     uint8_t* refcount_table;
1110     int ret;
1111 
1112     ret = bdrv_create_file(filename, options);
1113     if (ret < 0) {
1114         return ret;
1115     }
1116 
1117     ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR);
1118     if (ret < 0) {
1119         return ret;
1120     }
1121 
1122     /* Write the header */
1123     memset(&header, 0, sizeof(header));
1124     header.magic = cpu_to_be32(QCOW_MAGIC);
1125     header.version = cpu_to_be32(version);
1126     header.cluster_bits = cpu_to_be32(cluster_bits);
1127     header.size = cpu_to_be64(0);
1128     header.l1_table_offset = cpu_to_be64(0);
1129     header.l1_size = cpu_to_be32(0);
1130     header.refcount_table_offset = cpu_to_be64(cluster_size);
1131     header.refcount_table_clusters = cpu_to_be32(1);
1132     header.refcount_order = cpu_to_be32(3 + REFCOUNT_SHIFT);
1133     header.header_length = cpu_to_be32(sizeof(header));
1134 
1135     if (flags & BLOCK_FLAG_ENCRYPT) {
1136         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1137     } else {
1138         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1139     }
1140 
1141     ret = bdrv_pwrite(bs, 0, &header, sizeof(header));
1142     if (ret < 0) {
1143         goto out;
1144     }
1145 
1146     /* Write an empty refcount table */
1147     refcount_table = g_malloc0(cluster_size);
1148     ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size);
1149     g_free(refcount_table);
1150 
1151     if (ret < 0) {
1152         goto out;
1153     }
1154 
1155     bdrv_close(bs);
1156 
1157     /*
1158      * And now open the image and make it consistent first (i.e. increase the
1159      * refcount of the cluster that is occupied by the header and the refcount
1160      * table)
1161      */
1162     BlockDriver* drv = bdrv_find_format("qcow2");
1163     assert(drv != NULL);
1164     ret = bdrv_open(bs, filename,
1165         BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv);
1166     if (ret < 0) {
1167         goto out;
1168     }
1169 
1170     ret = qcow2_alloc_clusters(bs, 2 * cluster_size);
1171     if (ret < 0) {
1172         goto out;
1173 
1174     } else if (ret != 0) {
1175         error_report("Huh, first cluster in empty image is already in use?");
1176         abort();
1177     }
1178 
1179     /* Okay, now that we have a valid image, let's give it the right size */
1180     ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE);
1181     if (ret < 0) {
1182         goto out;
1183     }
1184 
1185     /* Want a backing file? There you go.*/
1186     if (backing_file) {
1187         ret = bdrv_change_backing_file(bs, backing_file, backing_format);
1188         if (ret < 0) {
1189             goto out;
1190         }
1191     }
1192 
1193     /* And if we're supposed to preallocate metadata, do that now */
1194     if (prealloc) {
1195         ret = preallocate(bs);
1196         if (ret < 0) {
1197             goto out;
1198         }
1199     }
1200 
1201     ret = 0;
1202 out:
1203     bdrv_delete(bs);
1204     return ret;
1205 }
1206 
1207 static int qcow2_create(const char *filename, QEMUOptionParameter *options)
1208 {
1209     const char *backing_file = NULL;
1210     const char *backing_fmt = NULL;
1211     uint64_t sectors = 0;
1212     int flags = 0;
1213     size_t cluster_size = DEFAULT_CLUSTER_SIZE;
1214     int prealloc = 0;
1215     int version = 2;
1216 
1217     /* Read out options */
1218     while (options && options->name) {
1219         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1220             sectors = options->value.n / 512;
1221         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1222             backing_file = options->value.s;
1223         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
1224             backing_fmt = options->value.s;
1225         } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
1226             flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
1227         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
1228             if (options->value.n) {
1229                 cluster_size = options->value.n;
1230             }
1231         } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
1232             if (!options->value.s || !strcmp(options->value.s, "off")) {
1233                 prealloc = 0;
1234             } else if (!strcmp(options->value.s, "metadata")) {
1235                 prealloc = 1;
1236             } else {
1237                 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
1238                     options->value.s);
1239                 return -EINVAL;
1240             }
1241         } else if (!strcmp(options->name, BLOCK_OPT_COMPAT_LEVEL)) {
1242             if (!options->value.s || !strcmp(options->value.s, "0.10")) {
1243                 version = 2;
1244             } else if (!strcmp(options->value.s, "1.1")) {
1245                 version = 3;
1246             } else {
1247                 fprintf(stderr, "Invalid compatibility level: '%s'\n",
1248                     options->value.s);
1249                 return -EINVAL;
1250             }
1251         }
1252         options++;
1253     }
1254 
1255     if (backing_file && prealloc) {
1256         fprintf(stderr, "Backing file and preallocation cannot be used at "
1257             "the same time\n");
1258         return -EINVAL;
1259     }
1260 
1261     return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags,
1262                          cluster_size, prealloc, options, version);
1263 }
1264 
1265 static int qcow2_make_empty(BlockDriverState *bs)
1266 {
1267 #if 0
1268     /* XXX: not correct */
1269     BDRVQcowState *s = bs->opaque;
1270     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1271     int ret;
1272 
1273     memset(s->l1_table, 0, l1_length);
1274     if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
1275         return -1;
1276     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
1277     if (ret < 0)
1278         return ret;
1279 
1280     l2_cache_reset(bs);
1281 #endif
1282     return 0;
1283 }
1284 
1285 static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
1286     int64_t sector_num, int nb_sectors)
1287 {
1288     int ret;
1289     BDRVQcowState *s = bs->opaque;
1290 
1291     /* Emulate misaligned zero writes */
1292     if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
1293         return -ENOTSUP;
1294     }
1295 
1296     /* Whatever is left can use real zero clusters */
1297     qemu_co_mutex_lock(&s->lock);
1298     ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1299         nb_sectors);
1300     qemu_co_mutex_unlock(&s->lock);
1301 
1302     return ret;
1303 }
1304 
1305 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
1306     int64_t sector_num, int nb_sectors)
1307 {
1308     int ret;
1309     BDRVQcowState *s = bs->opaque;
1310 
1311     qemu_co_mutex_lock(&s->lock);
1312     ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1313         nb_sectors);
1314     qemu_co_mutex_unlock(&s->lock);
1315     return ret;
1316 }
1317 
1318 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
1319 {
1320     BDRVQcowState *s = bs->opaque;
1321     int ret, new_l1_size;
1322 
1323     if (offset & 511) {
1324         error_report("The new size must be a multiple of 512");
1325         return -EINVAL;
1326     }
1327 
1328     /* cannot proceed if image has snapshots */
1329     if (s->nb_snapshots) {
1330         error_report("Can't resize an image which has snapshots");
1331         return -ENOTSUP;
1332     }
1333 
1334     /* shrinking is currently not supported */
1335     if (offset < bs->total_sectors * 512) {
1336         error_report("qcow2 doesn't support shrinking images yet");
1337         return -ENOTSUP;
1338     }
1339 
1340     new_l1_size = size_to_l1(s, offset);
1341     ret = qcow2_grow_l1_table(bs, new_l1_size, true);
1342     if (ret < 0) {
1343         return ret;
1344     }
1345 
1346     /* write updated header.size */
1347     offset = cpu_to_be64(offset);
1348     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
1349                            &offset, sizeof(uint64_t));
1350     if (ret < 0) {
1351         return ret;
1352     }
1353 
1354     s->l1_vm_state_index = new_l1_size;
1355     return 0;
1356 }
1357 
1358 /* XXX: put compressed sectors first, then all the cluster aligned
1359    tables to avoid losing bytes in alignment */
1360 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
1361                                   const uint8_t *buf, int nb_sectors)
1362 {
1363     BDRVQcowState *s = bs->opaque;
1364     z_stream strm;
1365     int ret, out_len;
1366     uint8_t *out_buf;
1367     uint64_t cluster_offset;
1368 
1369     if (nb_sectors == 0) {
1370         /* align end of file to a sector boundary to ease reading with
1371            sector based I/Os */
1372         cluster_offset = bdrv_getlength(bs->file);
1373         cluster_offset = (cluster_offset + 511) & ~511;
1374         bdrv_truncate(bs->file, cluster_offset);
1375         return 0;
1376     }
1377 
1378     if (nb_sectors != s->cluster_sectors)
1379         return -EINVAL;
1380 
1381     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1382 
1383     /* best compression, small window, no zlib header */
1384     memset(&strm, 0, sizeof(strm));
1385     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1386                        Z_DEFLATED, -12,
1387                        9, Z_DEFAULT_STRATEGY);
1388     if (ret != 0) {
1389         ret = -EINVAL;
1390         goto fail;
1391     }
1392 
1393     strm.avail_in = s->cluster_size;
1394     strm.next_in = (uint8_t *)buf;
1395     strm.avail_out = s->cluster_size;
1396     strm.next_out = out_buf;
1397 
1398     ret = deflate(&strm, Z_FINISH);
1399     if (ret != Z_STREAM_END && ret != Z_OK) {
1400         deflateEnd(&strm);
1401         ret = -EINVAL;
1402         goto fail;
1403     }
1404     out_len = strm.next_out - out_buf;
1405 
1406     deflateEnd(&strm);
1407 
1408     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1409         /* could not compress: write normal cluster */
1410         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1411         if (ret < 0) {
1412             goto fail;
1413         }
1414     } else {
1415         cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1416             sector_num << 9, out_len);
1417         if (!cluster_offset) {
1418             ret = -EIO;
1419             goto fail;
1420         }
1421         cluster_offset &= s->cluster_offset_mask;
1422         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1423         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
1424         if (ret < 0) {
1425             goto fail;
1426         }
1427     }
1428 
1429     ret = 0;
1430 fail:
1431     g_free(out_buf);
1432     return ret;
1433 }
1434 
1435 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
1436 {
1437     BDRVQcowState *s = bs->opaque;
1438     int ret;
1439 
1440     qemu_co_mutex_lock(&s->lock);
1441     ret = qcow2_cache_flush(bs, s->l2_table_cache);
1442     if (ret < 0) {
1443         qemu_co_mutex_unlock(&s->lock);
1444         return ret;
1445     }
1446 
1447     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1448     if (ret < 0) {
1449         qemu_co_mutex_unlock(&s->lock);
1450         return ret;
1451     }
1452     qemu_co_mutex_unlock(&s->lock);
1453 
1454     return 0;
1455 }
1456 
1457 static int64_t qcow2_vm_state_offset(BDRVQcowState *s)
1458 {
1459 	return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1460 }
1461 
1462 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1463 {
1464     BDRVQcowState *s = bs->opaque;
1465     bdi->cluster_size = s->cluster_size;
1466     bdi->vm_state_offset = qcow2_vm_state_offset(s);
1467     return 0;
1468 }
1469 
1470 
1471 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result)
1472 {
1473     return qcow2_check_refcounts(bs, result);
1474 }
1475 
1476 #if 0
1477 static void dump_refcounts(BlockDriverState *bs)
1478 {
1479     BDRVQcowState *s = bs->opaque;
1480     int64_t nb_clusters, k, k1, size;
1481     int refcount;
1482 
1483     size = bdrv_getlength(bs->file);
1484     nb_clusters = size_to_clusters(s, size);
1485     for(k = 0; k < nb_clusters;) {
1486         k1 = k;
1487         refcount = get_refcount(bs, k);
1488         k++;
1489         while (k < nb_clusters && get_refcount(bs, k) == refcount)
1490             k++;
1491         printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
1492                k - k1);
1493     }
1494 }
1495 #endif
1496 
1497 static int qcow2_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1498                               int64_t pos, int size)
1499 {
1500     BDRVQcowState *s = bs->opaque;
1501     int growable = bs->growable;
1502     int ret;
1503 
1504     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
1505     bs->growable = 1;
1506     ret = bdrv_pwrite(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1507     bs->growable = growable;
1508 
1509     return ret;
1510 }
1511 
1512 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1513                               int64_t pos, int size)
1514 {
1515     BDRVQcowState *s = bs->opaque;
1516     int growable = bs->growable;
1517     int ret;
1518 
1519     BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
1520     bs->growable = 1;
1521     ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1522     bs->growable = growable;
1523 
1524     return ret;
1525 }
1526 
1527 static QEMUOptionParameter qcow2_create_options[] = {
1528     {
1529         .name = BLOCK_OPT_SIZE,
1530         .type = OPT_SIZE,
1531         .help = "Virtual disk size"
1532     },
1533     {
1534         .name = BLOCK_OPT_COMPAT_LEVEL,
1535         .type = OPT_STRING,
1536         .help = "Compatibility level (0.10 or 1.1)"
1537     },
1538     {
1539         .name = BLOCK_OPT_BACKING_FILE,
1540         .type = OPT_STRING,
1541         .help = "File name of a base image"
1542     },
1543     {
1544         .name = BLOCK_OPT_BACKING_FMT,
1545         .type = OPT_STRING,
1546         .help = "Image format of the base image"
1547     },
1548     {
1549         .name = BLOCK_OPT_ENCRYPT,
1550         .type = OPT_FLAG,
1551         .help = "Encrypt the image"
1552     },
1553     {
1554         .name = BLOCK_OPT_CLUSTER_SIZE,
1555         .type = OPT_SIZE,
1556         .help = "qcow2 cluster size",
1557         .value = { .n = DEFAULT_CLUSTER_SIZE },
1558     },
1559     {
1560         .name = BLOCK_OPT_PREALLOC,
1561         .type = OPT_STRING,
1562         .help = "Preallocation mode (allowed values: off, metadata)"
1563     },
1564     { NULL }
1565 };
1566 
1567 static BlockDriver bdrv_qcow2 = {
1568     .format_name        = "qcow2",
1569     .instance_size      = sizeof(BDRVQcowState),
1570     .bdrv_probe         = qcow2_probe,
1571     .bdrv_open          = qcow2_open,
1572     .bdrv_close         = qcow2_close,
1573     .bdrv_create        = qcow2_create,
1574     .bdrv_co_is_allocated = qcow2_co_is_allocated,
1575     .bdrv_set_key       = qcow2_set_key,
1576     .bdrv_make_empty    = qcow2_make_empty,
1577 
1578     .bdrv_co_readv          = qcow2_co_readv,
1579     .bdrv_co_writev         = qcow2_co_writev,
1580     .bdrv_co_flush_to_os    = qcow2_co_flush_to_os,
1581 
1582     .bdrv_co_write_zeroes   = qcow2_co_write_zeroes,
1583     .bdrv_co_discard        = qcow2_co_discard,
1584     .bdrv_truncate          = qcow2_truncate,
1585     .bdrv_write_compressed  = qcow2_write_compressed,
1586 
1587     .bdrv_snapshot_create   = qcow2_snapshot_create,
1588     .bdrv_snapshot_goto     = qcow2_snapshot_goto,
1589     .bdrv_snapshot_delete   = qcow2_snapshot_delete,
1590     .bdrv_snapshot_list     = qcow2_snapshot_list,
1591     .bdrv_snapshot_load_tmp     = qcow2_snapshot_load_tmp,
1592     .bdrv_get_info      = qcow2_get_info,
1593 
1594     .bdrv_save_vmstate    = qcow2_save_vmstate,
1595     .bdrv_load_vmstate    = qcow2_load_vmstate,
1596 
1597     .bdrv_change_backing_file   = qcow2_change_backing_file,
1598 
1599     .bdrv_invalidate_cache      = qcow2_invalidate_cache,
1600 
1601     .create_options = qcow2_create_options,
1602     .bdrv_check = qcow2_check,
1603 };
1604 
1605 static void bdrv_qcow2_init(void)
1606 {
1607     bdrv_register(&bdrv_qcow2);
1608 }
1609 
1610 block_init(bdrv_qcow2_init);
1611