xref: /openbmc/qemu/block/qcow.c (revision d8e39b70)
1 /*
2  * Block driver for the QCOW format
3  *
4  * Copyright (c) 2004-2006 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 #include "qemu/osdep.h"
25 #include "qapi/error.h"
26 #include "qemu-common.h"
27 #include "qemu/error-report.h"
28 #include "block/block_int.h"
29 #include "sysemu/block-backend.h"
30 #include "qemu/module.h"
31 #include "qemu/bswap.h"
32 #include <zlib.h>
33 #include "qapi/qmp/qerror.h"
34 #include "qapi/qmp/qstring.h"
35 #include "crypto/block.h"
36 #include "migration/blocker.h"
37 #include "block/crypto.h"
38 
39 /**************************************************************/
40 /* QEMU COW block driver with compression and encryption support */
41 
42 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
43 #define QCOW_VERSION 1
44 
45 #define QCOW_CRYPT_NONE 0
46 #define QCOW_CRYPT_AES  1
47 
48 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
49 
50 typedef struct QCowHeader {
51     uint32_t magic;
52     uint32_t version;
53     uint64_t backing_file_offset;
54     uint32_t backing_file_size;
55     uint32_t mtime;
56     uint64_t size; /* in bytes */
57     uint8_t cluster_bits;
58     uint8_t l2_bits;
59     uint16_t padding;
60     uint32_t crypt_method;
61     uint64_t l1_table_offset;
62 } QEMU_PACKED QCowHeader;
63 
64 #define L2_CACHE_SIZE 16
65 
66 typedef struct BDRVQcowState {
67     int cluster_bits;
68     int cluster_size;
69     int cluster_sectors;
70     int l2_bits;
71     int l2_size;
72     unsigned int l1_size;
73     uint64_t cluster_offset_mask;
74     uint64_t l1_table_offset;
75     uint64_t *l1_table;
76     uint64_t *l2_cache;
77     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
78     uint32_t l2_cache_counts[L2_CACHE_SIZE];
79     uint8_t *cluster_cache;
80     uint8_t *cluster_data;
81     uint64_t cluster_cache_offset;
82     QCryptoBlock *crypto; /* Disk encryption format driver */
83     uint32_t crypt_method_header;
84     CoMutex lock;
85     Error *migration_blocker;
86 } BDRVQcowState;
87 
88 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
89 
90 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
91 {
92     const QCowHeader *cow_header = (const void *)buf;
93 
94     if (buf_size >= sizeof(QCowHeader) &&
95         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
96         be32_to_cpu(cow_header->version) == QCOW_VERSION)
97         return 100;
98     else
99         return 0;
100 }
101 
102 static QemuOptsList qcow_runtime_opts = {
103     .name = "qcow",
104     .head = QTAILQ_HEAD_INITIALIZER(qcow_runtime_opts.head),
105     .desc = {
106         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
107         { /* end of list */ }
108     },
109 };
110 
111 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
112                      Error **errp)
113 {
114     BDRVQcowState *s = bs->opaque;
115     unsigned int len, i, shift;
116     int ret;
117     QCowHeader header;
118     Error *local_err = NULL;
119     QCryptoBlockOpenOptions *crypto_opts = NULL;
120     unsigned int cflags = 0;
121     QDict *encryptopts = NULL;
122     const char *encryptfmt;
123 
124     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
125     encryptfmt = qdict_get_try_str(encryptopts, "format");
126 
127     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
128                                false, errp);
129     if (!bs->file) {
130         ret = -EINVAL;
131         goto fail;
132     }
133 
134     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
135     if (ret < 0) {
136         goto fail;
137     }
138     be32_to_cpus(&header.magic);
139     be32_to_cpus(&header.version);
140     be64_to_cpus(&header.backing_file_offset);
141     be32_to_cpus(&header.backing_file_size);
142     be32_to_cpus(&header.mtime);
143     be64_to_cpus(&header.size);
144     be32_to_cpus(&header.crypt_method);
145     be64_to_cpus(&header.l1_table_offset);
146 
147     if (header.magic != QCOW_MAGIC) {
148         error_setg(errp, "Image not in qcow format");
149         ret = -EINVAL;
150         goto fail;
151     }
152     if (header.version != QCOW_VERSION) {
153         error_setg(errp, "Unsupported qcow version %" PRIu32, header.version);
154         ret = -ENOTSUP;
155         goto fail;
156     }
157 
158     if (header.size <= 1) {
159         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
160         ret = -EINVAL;
161         goto fail;
162     }
163     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
164         error_setg(errp, "Cluster size must be between 512 and 64k");
165         ret = -EINVAL;
166         goto fail;
167     }
168 
169     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
170      * so bytes = num_entries << 3. */
171     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
172         error_setg(errp, "L2 table size must be between 512 and 64k");
173         ret = -EINVAL;
174         goto fail;
175     }
176 
177     s->crypt_method_header = header.crypt_method;
178     if (s->crypt_method_header) {
179         if (bdrv_uses_whitelist() &&
180             s->crypt_method_header == QCOW_CRYPT_AES) {
181             error_setg(errp,
182                        "Use of AES-CBC encrypted qcow images is no longer "
183                        "supported in system emulators");
184             error_append_hint(errp,
185                               "You can use 'qemu-img convert' to convert your "
186                               "image to an alternative supported format, such "
187                               "as unencrypted qcow, or raw with the LUKS "
188                               "format instead.\n");
189             ret = -ENOSYS;
190             goto fail;
191         }
192         if (s->crypt_method_header == QCOW_CRYPT_AES) {
193             if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
194                 error_setg(errp,
195                            "Header reported 'aes' encryption format but "
196                            "options specify '%s'", encryptfmt);
197                 ret = -EINVAL;
198                 goto fail;
199             }
200             qdict_del(encryptopts, "format");
201             crypto_opts = block_crypto_open_opts_init(
202                 Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp);
203             if (!crypto_opts) {
204                 ret = -EINVAL;
205                 goto fail;
206             }
207 
208             if (flags & BDRV_O_NO_IO) {
209                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
210             }
211             s->crypto = qcrypto_block_open(crypto_opts, "encrypt.",
212                                            NULL, NULL, cflags, errp);
213             if (!s->crypto) {
214                 ret = -EINVAL;
215                 goto fail;
216             }
217         } else {
218             error_setg(errp, "invalid encryption method in qcow header");
219             ret = -EINVAL;
220             goto fail;
221         }
222         bs->encrypted = true;
223     } else {
224         if (encryptfmt) {
225             error_setg(errp, "No encryption in image header, but options "
226                        "specified format '%s'", encryptfmt);
227             ret = -EINVAL;
228             goto fail;
229         }
230     }
231     s->cluster_bits = header.cluster_bits;
232     s->cluster_size = 1 << s->cluster_bits;
233     s->cluster_sectors = 1 << (s->cluster_bits - 9);
234     s->l2_bits = header.l2_bits;
235     s->l2_size = 1 << s->l2_bits;
236     bs->total_sectors = header.size / 512;
237     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
238 
239     /* read the level 1 table */
240     shift = s->cluster_bits + s->l2_bits;
241     if (header.size > UINT64_MAX - (1LL << shift)) {
242         error_setg(errp, "Image too large");
243         ret = -EINVAL;
244         goto fail;
245     } else {
246         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
247         if (l1_size > INT_MAX / sizeof(uint64_t)) {
248             error_setg(errp, "Image too large");
249             ret = -EINVAL;
250             goto fail;
251         }
252         s->l1_size = l1_size;
253     }
254 
255     s->l1_table_offset = header.l1_table_offset;
256     s->l1_table = g_try_new(uint64_t, s->l1_size);
257     if (s->l1_table == NULL) {
258         error_setg(errp, "Could not allocate memory for L1 table");
259         ret = -ENOMEM;
260         goto fail;
261     }
262 
263     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
264                s->l1_size * sizeof(uint64_t));
265     if (ret < 0) {
266         goto fail;
267     }
268 
269     for(i = 0;i < s->l1_size; i++) {
270         be64_to_cpus(&s->l1_table[i]);
271     }
272 
273     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
274     s->l2_cache =
275         qemu_try_blockalign(bs->file->bs,
276                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
277     if (s->l2_cache == NULL) {
278         error_setg(errp, "Could not allocate L2 table cache");
279         ret = -ENOMEM;
280         goto fail;
281     }
282     s->cluster_cache = g_malloc(s->cluster_size);
283     s->cluster_data = g_malloc(s->cluster_size);
284     s->cluster_cache_offset = -1;
285 
286     /* read the backing file name */
287     if (header.backing_file_offset != 0) {
288         len = header.backing_file_size;
289         if (len > 1023 || len >= sizeof(bs->backing_file)) {
290             error_setg(errp, "Backing file name too long");
291             ret = -EINVAL;
292             goto fail;
293         }
294         ret = bdrv_pread(bs->file, header.backing_file_offset,
295                    bs->backing_file, len);
296         if (ret < 0) {
297             goto fail;
298         }
299         bs->backing_file[len] = '\0';
300     }
301 
302     /* Disable migration when qcow images are used */
303     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
304                "does not support live migration",
305                bdrv_get_device_or_node_name(bs));
306     ret = migrate_add_blocker(s->migration_blocker, &local_err);
307     if (local_err) {
308         error_propagate(errp, local_err);
309         error_free(s->migration_blocker);
310         goto fail;
311     }
312 
313     QDECREF(encryptopts);
314     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
315     qemu_co_mutex_init(&s->lock);
316     return 0;
317 
318  fail:
319     g_free(s->l1_table);
320     qemu_vfree(s->l2_cache);
321     g_free(s->cluster_cache);
322     g_free(s->cluster_data);
323     qcrypto_block_free(s->crypto);
324     QDECREF(encryptopts);
325     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
326     return ret;
327 }
328 
329 
330 /* We have nothing to do for QCOW reopen, stubs just return
331  * success */
332 static int qcow_reopen_prepare(BDRVReopenState *state,
333                                BlockReopenQueue *queue, Error **errp)
334 {
335     return 0;
336 }
337 
338 
339 /* 'allocate' is:
340  *
341  * 0 to not allocate.
342  *
343  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
344  * 'n_end')
345  *
346  * 2 to allocate a compressed cluster of size
347  * 'compressed_size'. 'compressed_size' must be > 0 and <
348  * cluster_size
349  *
350  * return 0 if not allocated, 1 if *result is assigned, and negative
351  * errno on failure.
352  */
353 static int get_cluster_offset(BlockDriverState *bs,
354                               uint64_t offset, int allocate,
355                               int compressed_size,
356                               int n_start, int n_end, uint64_t *result)
357 {
358     BDRVQcowState *s = bs->opaque;
359     int min_index, i, j, l1_index, l2_index, ret;
360     int64_t l2_offset;
361     uint64_t *l2_table, cluster_offset, tmp;
362     uint32_t min_count;
363     int new_l2_table;
364 
365     *result = 0;
366     l1_index = offset >> (s->l2_bits + s->cluster_bits);
367     l2_offset = s->l1_table[l1_index];
368     new_l2_table = 0;
369     if (!l2_offset) {
370         if (!allocate)
371             return 0;
372         /* allocate a new l2 entry */
373         l2_offset = bdrv_getlength(bs->file->bs);
374         if (l2_offset < 0) {
375             return l2_offset;
376         }
377         /* round to cluster size */
378         l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size);
379         /* update the L1 entry */
380         s->l1_table[l1_index] = l2_offset;
381         tmp = cpu_to_be64(l2_offset);
382         BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
383         ret = bdrv_pwrite_sync(bs->file,
384                                s->l1_table_offset + l1_index * sizeof(tmp),
385                                &tmp, sizeof(tmp));
386         if (ret < 0) {
387             return ret;
388         }
389         new_l2_table = 1;
390     }
391     for(i = 0; i < L2_CACHE_SIZE; i++) {
392         if (l2_offset == s->l2_cache_offsets[i]) {
393             /* increment the hit count */
394             if (++s->l2_cache_counts[i] == 0xffffffff) {
395                 for(j = 0; j < L2_CACHE_SIZE; j++) {
396                     s->l2_cache_counts[j] >>= 1;
397                 }
398             }
399             l2_table = s->l2_cache + (i << s->l2_bits);
400             goto found;
401         }
402     }
403     /* not found: load a new entry in the least used one */
404     min_index = 0;
405     min_count = 0xffffffff;
406     for(i = 0; i < L2_CACHE_SIZE; i++) {
407         if (s->l2_cache_counts[i] < min_count) {
408             min_count = s->l2_cache_counts[i];
409             min_index = i;
410         }
411     }
412     l2_table = s->l2_cache + (min_index << s->l2_bits);
413     BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
414     if (new_l2_table) {
415         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
416         ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
417                                s->l2_size * sizeof(uint64_t));
418         if (ret < 0) {
419             return ret;
420         }
421     } else {
422         ret = bdrv_pread(bs->file, l2_offset, l2_table,
423                          s->l2_size * sizeof(uint64_t));
424         if (ret < 0) {
425             return ret;
426         }
427     }
428     s->l2_cache_offsets[min_index] = l2_offset;
429     s->l2_cache_counts[min_index] = 1;
430  found:
431     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
432     cluster_offset = be64_to_cpu(l2_table[l2_index]);
433     if (!cluster_offset ||
434         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
435         if (!allocate)
436             return 0;
437         BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
438         /* allocate a new cluster */
439         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
440             (n_end - n_start) < s->cluster_sectors) {
441             /* if the cluster is already compressed, we must
442                decompress it in the case it is not completely
443                overwritten */
444             if (decompress_cluster(bs, cluster_offset) < 0) {
445                 return -EIO;
446             }
447             cluster_offset = bdrv_getlength(bs->file->bs);
448             if ((int64_t) cluster_offset < 0) {
449                 return cluster_offset;
450             }
451             cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
452             /* write the cluster content */
453             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
454             ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
455                               s->cluster_size);
456             if (ret < 0) {
457                 return ret;
458             }
459         } else {
460             cluster_offset = bdrv_getlength(bs->file->bs);
461             if ((int64_t) cluster_offset < 0) {
462                 return cluster_offset;
463             }
464             if (allocate == 1) {
465                 /* round to cluster size */
466                 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
467                 if (cluster_offset + s->cluster_size > INT64_MAX) {
468                     return -E2BIG;
469                 }
470                 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size,
471                                     PREALLOC_MODE_OFF, NULL);
472                 if (ret < 0) {
473                     return ret;
474                 }
475                 /* if encrypted, we must initialize the cluster
476                    content which won't be written */
477                 if (bs->encrypted &&
478                     (n_end - n_start) < s->cluster_sectors) {
479                     uint64_t start_sect;
480                     assert(s->crypto);
481                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
482                     for(i = 0; i < s->cluster_sectors; i++) {
483                         if (i < n_start || i >= n_end) {
484                             memset(s->cluster_data, 0x00, 512);
485                             if (qcrypto_block_encrypt(s->crypto,
486                                                       (start_sect + i) *
487                                                       BDRV_SECTOR_SIZE,
488                                                       s->cluster_data,
489                                                       BDRV_SECTOR_SIZE,
490                                                       NULL) < 0) {
491                                 return -EIO;
492                             }
493                             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
494                             ret = bdrv_pwrite(bs->file,
495                                               cluster_offset + i * 512,
496                                               s->cluster_data, 512);
497                             if (ret < 0) {
498                                 return ret;
499                             }
500                         }
501                     }
502                 }
503             } else if (allocate == 2) {
504                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
505                     (uint64_t)compressed_size << (63 - s->cluster_bits);
506             }
507         }
508         /* update L2 table */
509         tmp = cpu_to_be64(cluster_offset);
510         l2_table[l2_index] = tmp;
511         if (allocate == 2) {
512             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
513         } else {
514             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
515         }
516         ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
517                                &tmp, sizeof(tmp));
518         if (ret < 0) {
519             return ret;
520         }
521     }
522     *result = cluster_offset;
523     return 1;
524 }
525 
526 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
527         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
528 {
529     BDRVQcowState *s = bs->opaque;
530     int index_in_cluster, n, ret;
531     uint64_t cluster_offset;
532 
533     qemu_co_mutex_lock(&s->lock);
534     ret = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0, &cluster_offset);
535     qemu_co_mutex_unlock(&s->lock);
536     if (ret < 0) {
537         return ret;
538     }
539     index_in_cluster = sector_num & (s->cluster_sectors - 1);
540     n = s->cluster_sectors - index_in_cluster;
541     if (n > nb_sectors)
542         n = nb_sectors;
543     *pnum = n;
544     if (!cluster_offset) {
545         return 0;
546     }
547     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) {
548         return BDRV_BLOCK_DATA;
549     }
550     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
551     *file = bs->file->bs;
552     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
553 }
554 
555 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
556                              const uint8_t *buf, int buf_size)
557 {
558     z_stream strm1, *strm = &strm1;
559     int ret, out_len;
560 
561     memset(strm, 0, sizeof(*strm));
562 
563     strm->next_in = (uint8_t *)buf;
564     strm->avail_in = buf_size;
565     strm->next_out = out_buf;
566     strm->avail_out = out_buf_size;
567 
568     ret = inflateInit2(strm, -12);
569     if (ret != Z_OK)
570         return -1;
571     ret = inflate(strm, Z_FINISH);
572     out_len = strm->next_out - out_buf;
573     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
574         out_len != out_buf_size) {
575         inflateEnd(strm);
576         return -1;
577     }
578     inflateEnd(strm);
579     return 0;
580 }
581 
582 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
583 {
584     BDRVQcowState *s = bs->opaque;
585     int ret, csize;
586     uint64_t coffset;
587 
588     coffset = cluster_offset & s->cluster_offset_mask;
589     if (s->cluster_cache_offset != coffset) {
590         csize = cluster_offset >> (63 - s->cluster_bits);
591         csize &= (s->cluster_size - 1);
592         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
593         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
594         if (ret != csize)
595             return -1;
596         if (decompress_buffer(s->cluster_cache, s->cluster_size,
597                               s->cluster_data, csize) < 0) {
598             return -1;
599         }
600         s->cluster_cache_offset = coffset;
601     }
602     return 0;
603 }
604 
605 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
606                          int nb_sectors, QEMUIOVector *qiov)
607 {
608     BDRVQcowState *s = bs->opaque;
609     int index_in_cluster;
610     int ret = 0, n;
611     uint64_t cluster_offset;
612     struct iovec hd_iov;
613     QEMUIOVector hd_qiov;
614     uint8_t *buf;
615     void *orig_buf;
616 
617     if (qiov->niov > 1) {
618         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
619         if (buf == NULL) {
620             return -ENOMEM;
621         }
622     } else {
623         orig_buf = NULL;
624         buf = (uint8_t *)qiov->iov->iov_base;
625     }
626 
627     qemu_co_mutex_lock(&s->lock);
628 
629     while (nb_sectors != 0) {
630         /* prepare next request */
631         ret = get_cluster_offset(bs, sector_num << 9,
632                                  0, 0, 0, 0, &cluster_offset);
633         if (ret < 0) {
634             break;
635         }
636         index_in_cluster = sector_num & (s->cluster_sectors - 1);
637         n = s->cluster_sectors - index_in_cluster;
638         if (n > nb_sectors) {
639             n = nb_sectors;
640         }
641 
642         if (!cluster_offset) {
643             if (bs->backing) {
644                 /* read from the base image */
645                 hd_iov.iov_base = (void *)buf;
646                 hd_iov.iov_len = n * 512;
647                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
648                 qemu_co_mutex_unlock(&s->lock);
649                 /* qcow2 emits this on bs->file instead of bs->backing */
650                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
651                 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov);
652                 qemu_co_mutex_lock(&s->lock);
653                 if (ret < 0) {
654                     break;
655                 }
656             } else {
657                 /* Note: in this case, no need to wait */
658                 memset(buf, 0, 512 * n);
659             }
660         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
661             /* add AIO support for compressed blocks ? */
662             if (decompress_cluster(bs, cluster_offset) < 0) {
663                 ret = -EIO;
664                 break;
665             }
666             memcpy(buf,
667                    s->cluster_cache + index_in_cluster * 512, 512 * n);
668         } else {
669             if ((cluster_offset & 511) != 0) {
670                 ret = -EIO;
671                 break;
672             }
673             hd_iov.iov_base = (void *)buf;
674             hd_iov.iov_len = n * 512;
675             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
676             qemu_co_mutex_unlock(&s->lock);
677             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
678             ret = bdrv_co_readv(bs->file,
679                                 (cluster_offset >> 9) + index_in_cluster,
680                                 n, &hd_qiov);
681             qemu_co_mutex_lock(&s->lock);
682             if (ret < 0) {
683                 break;
684             }
685             if (bs->encrypted) {
686                 assert(s->crypto);
687                 if (qcrypto_block_decrypt(s->crypto,
688                                           sector_num * BDRV_SECTOR_SIZE, buf,
689                                           n * BDRV_SECTOR_SIZE, NULL) < 0) {
690                     ret = -EIO;
691                     break;
692                 }
693             }
694         }
695         ret = 0;
696 
697         nb_sectors -= n;
698         sector_num += n;
699         buf += n * 512;
700     }
701 
702     qemu_co_mutex_unlock(&s->lock);
703 
704     if (qiov->niov > 1) {
705         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
706         qemu_vfree(orig_buf);
707     }
708 
709     return ret;
710 }
711 
712 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
713                           int nb_sectors, QEMUIOVector *qiov)
714 {
715     BDRVQcowState *s = bs->opaque;
716     int index_in_cluster;
717     uint64_t cluster_offset;
718     int ret = 0, n;
719     struct iovec hd_iov;
720     QEMUIOVector hd_qiov;
721     uint8_t *buf;
722     void *orig_buf;
723 
724     s->cluster_cache_offset = -1; /* disable compressed cache */
725 
726     /* We must always copy the iov when encrypting, so we
727      * don't modify the original data buffer during encryption */
728     if (bs->encrypted || qiov->niov > 1) {
729         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
730         if (buf == NULL) {
731             return -ENOMEM;
732         }
733         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
734     } else {
735         orig_buf = NULL;
736         buf = (uint8_t *)qiov->iov->iov_base;
737     }
738 
739     qemu_co_mutex_lock(&s->lock);
740 
741     while (nb_sectors != 0) {
742 
743         index_in_cluster = sector_num & (s->cluster_sectors - 1);
744         n = s->cluster_sectors - index_in_cluster;
745         if (n > nb_sectors) {
746             n = nb_sectors;
747         }
748         ret = get_cluster_offset(bs, sector_num << 9, 1, 0,
749                                  index_in_cluster,
750                                  index_in_cluster + n, &cluster_offset);
751         if (ret < 0) {
752             break;
753         }
754         if (!cluster_offset || (cluster_offset & 511) != 0) {
755             ret = -EIO;
756             break;
757         }
758         if (bs->encrypted) {
759             assert(s->crypto);
760             if (qcrypto_block_encrypt(s->crypto, sector_num * BDRV_SECTOR_SIZE,
761                                       buf, n * BDRV_SECTOR_SIZE, NULL) < 0) {
762                 ret = -EIO;
763                 break;
764             }
765         }
766 
767         hd_iov.iov_base = (void *)buf;
768         hd_iov.iov_len = n * 512;
769         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
770         qemu_co_mutex_unlock(&s->lock);
771         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
772         ret = bdrv_co_writev(bs->file,
773                              (cluster_offset >> 9) + index_in_cluster,
774                              n, &hd_qiov);
775         qemu_co_mutex_lock(&s->lock);
776         if (ret < 0) {
777             break;
778         }
779         ret = 0;
780 
781         nb_sectors -= n;
782         sector_num += n;
783         buf += n * 512;
784     }
785     qemu_co_mutex_unlock(&s->lock);
786 
787     qemu_vfree(orig_buf);
788 
789     return ret;
790 }
791 
792 static void qcow_close(BlockDriverState *bs)
793 {
794     BDRVQcowState *s = bs->opaque;
795 
796     qcrypto_block_free(s->crypto);
797     s->crypto = NULL;
798     g_free(s->l1_table);
799     qemu_vfree(s->l2_cache);
800     g_free(s->cluster_cache);
801     g_free(s->cluster_data);
802 
803     migrate_del_blocker(s->migration_blocker);
804     error_free(s->migration_blocker);
805 }
806 
807 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
808 {
809     int header_size, backing_filename_len, l1_size, shift, i;
810     QCowHeader header;
811     uint8_t *tmp;
812     int64_t total_size = 0;
813     char *backing_file = NULL;
814     Error *local_err = NULL;
815     int ret;
816     BlockBackend *qcow_blk;
817     char *encryptfmt = NULL;
818     QDict *options;
819     QDict *encryptopts = NULL;
820     QCryptoBlockCreateOptions *crypto_opts = NULL;
821     QCryptoBlock *crypto = NULL;
822 
823     /* Read out options */
824     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
825                           BDRV_SECTOR_SIZE);
826     if (total_size == 0) {
827         error_setg(errp, "Image size is too small, cannot be zero length");
828         ret = -EINVAL;
829         goto cleanup;
830     }
831 
832     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
833     encryptfmt = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
834     if (encryptfmt) {
835         if (qemu_opt_get(opts, BLOCK_OPT_ENCRYPT)) {
836             error_setg(errp, "Options " BLOCK_OPT_ENCRYPT " and "
837                        BLOCK_OPT_ENCRYPT_FORMAT " are mutually exclusive");
838             ret = -EINVAL;
839             goto cleanup;
840         }
841     } else if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
842         encryptfmt = g_strdup("aes");
843     }
844 
845     ret = bdrv_create_file(filename, opts, &local_err);
846     if (ret < 0) {
847         error_propagate(errp, local_err);
848         goto cleanup;
849     }
850 
851     qcow_blk = blk_new_open(filename, NULL, NULL,
852                             BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
853                             &local_err);
854     if (qcow_blk == NULL) {
855         error_propagate(errp, local_err);
856         ret = -EIO;
857         goto cleanup;
858     }
859 
860     blk_set_allow_write_beyond_eof(qcow_blk, true);
861 
862     ret = blk_truncate(qcow_blk, 0, PREALLOC_MODE_OFF, errp);
863     if (ret < 0) {
864         goto exit;
865     }
866 
867     memset(&header, 0, sizeof(header));
868     header.magic = cpu_to_be32(QCOW_MAGIC);
869     header.version = cpu_to_be32(QCOW_VERSION);
870     header.size = cpu_to_be64(total_size);
871     header_size = sizeof(header);
872     backing_filename_len = 0;
873     if (backing_file) {
874         if (strcmp(backing_file, "fat:")) {
875             header.backing_file_offset = cpu_to_be64(header_size);
876             backing_filename_len = strlen(backing_file);
877             header.backing_file_size = cpu_to_be32(backing_filename_len);
878             header_size += backing_filename_len;
879         } else {
880             /* special backing file for vvfat */
881             g_free(backing_file);
882             backing_file = NULL;
883         }
884         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
885                                     unmodified sectors */
886         header.l2_bits = 12; /* 32 KB L2 tables */
887     } else {
888         header.cluster_bits = 12; /* 4 KB clusters */
889         header.l2_bits = 9; /* 4 KB L2 tables */
890     }
891     header_size = (header_size + 7) & ~7;
892     shift = header.cluster_bits + header.l2_bits;
893     l1_size = (total_size + (1LL << shift) - 1) >> shift;
894 
895     header.l1_table_offset = cpu_to_be64(header_size);
896 
897     options = qemu_opts_to_qdict(opts, NULL);
898     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
899     QDECREF(options);
900     if (encryptfmt) {
901         if (!g_str_equal(encryptfmt, "aes")) {
902             error_setg(errp, "Unknown encryption format '%s', expected 'aes'",
903                        encryptfmt);
904             ret = -EINVAL;
905             goto exit;
906         }
907         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
908 
909         crypto_opts = block_crypto_create_opts_init(
910             Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp);
911         if (!crypto_opts) {
912             ret = -EINVAL;
913             goto exit;
914         }
915 
916         crypto = qcrypto_block_create(crypto_opts, "encrypt.",
917                                       NULL, NULL, NULL, errp);
918         if (!crypto) {
919             ret = -EINVAL;
920             goto exit;
921         }
922     } else {
923         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
924     }
925 
926     /* write all the data */
927     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
928     if (ret != sizeof(header)) {
929         goto exit;
930     }
931 
932     if (backing_file) {
933         ret = blk_pwrite(qcow_blk, sizeof(header),
934                          backing_file, backing_filename_len, 0);
935         if (ret != backing_filename_len) {
936             goto exit;
937         }
938     }
939 
940     tmp = g_malloc0(BDRV_SECTOR_SIZE);
941     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
942          i++) {
943         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
944                          tmp, BDRV_SECTOR_SIZE, 0);
945         if (ret != BDRV_SECTOR_SIZE) {
946             g_free(tmp);
947             goto exit;
948         }
949     }
950 
951     g_free(tmp);
952     ret = 0;
953 exit:
954     blk_unref(qcow_blk);
955 cleanup:
956     QDECREF(encryptopts);
957     g_free(encryptfmt);
958     qcrypto_block_free(crypto);
959     qapi_free_QCryptoBlockCreateOptions(crypto_opts);
960     g_free(backing_file);
961     return ret;
962 }
963 
964 static int qcow_make_empty(BlockDriverState *bs)
965 {
966     BDRVQcowState *s = bs->opaque;
967     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
968     int ret;
969 
970     memset(s->l1_table, 0, l1_length);
971     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
972             l1_length) < 0)
973         return -1;
974     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length,
975                         PREALLOC_MODE_OFF, NULL);
976     if (ret < 0)
977         return ret;
978 
979     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
980     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
981     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
982 
983     return 0;
984 }
985 
986 /* XXX: put compressed sectors first, then all the cluster aligned
987    tables to avoid losing bytes in alignment */
988 static coroutine_fn int
989 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
990                            uint64_t bytes, QEMUIOVector *qiov)
991 {
992     BDRVQcowState *s = bs->opaque;
993     QEMUIOVector hd_qiov;
994     struct iovec iov;
995     z_stream strm;
996     int ret, out_len;
997     uint8_t *buf, *out_buf;
998     uint64_t cluster_offset;
999 
1000     buf = qemu_blockalign(bs, s->cluster_size);
1001     if (bytes != s->cluster_size) {
1002         if (bytes > s->cluster_size ||
1003             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
1004         {
1005             qemu_vfree(buf);
1006             return -EINVAL;
1007         }
1008         /* Zero-pad last write if image size is not cluster aligned */
1009         memset(buf + bytes, 0, s->cluster_size - bytes);
1010     }
1011     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
1012 
1013     out_buf = g_malloc(s->cluster_size);
1014 
1015     /* best compression, small window, no zlib header */
1016     memset(&strm, 0, sizeof(strm));
1017     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1018                        Z_DEFLATED, -12,
1019                        9, Z_DEFAULT_STRATEGY);
1020     if (ret != 0) {
1021         ret = -EINVAL;
1022         goto fail;
1023     }
1024 
1025     strm.avail_in = s->cluster_size;
1026     strm.next_in = (uint8_t *)buf;
1027     strm.avail_out = s->cluster_size;
1028     strm.next_out = out_buf;
1029 
1030     ret = deflate(&strm, Z_FINISH);
1031     if (ret != Z_STREAM_END && ret != Z_OK) {
1032         deflateEnd(&strm);
1033         ret = -EINVAL;
1034         goto fail;
1035     }
1036     out_len = strm.next_out - out_buf;
1037 
1038     deflateEnd(&strm);
1039 
1040     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1041         /* could not compress: write normal cluster */
1042         ret = qcow_co_writev(bs, offset >> BDRV_SECTOR_BITS,
1043                              bytes >> BDRV_SECTOR_BITS, qiov);
1044         if (ret < 0) {
1045             goto fail;
1046         }
1047         goto success;
1048     }
1049     qemu_co_mutex_lock(&s->lock);
1050     ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1051     qemu_co_mutex_unlock(&s->lock);
1052     if (ret < 0) {
1053         goto fail;
1054     }
1055     if (cluster_offset == 0) {
1056         ret = -EIO;
1057         goto fail;
1058     }
1059     cluster_offset &= s->cluster_offset_mask;
1060 
1061     iov = (struct iovec) {
1062         .iov_base   = out_buf,
1063         .iov_len    = out_len,
1064     };
1065     qemu_iovec_init_external(&hd_qiov, &iov, 1);
1066     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1067     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
1068     if (ret < 0) {
1069         goto fail;
1070     }
1071 success:
1072     ret = 0;
1073 fail:
1074     qemu_vfree(buf);
1075     g_free(out_buf);
1076     return ret;
1077 }
1078 
1079 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1080 {
1081     BDRVQcowState *s = bs->opaque;
1082     bdi->cluster_size = s->cluster_size;
1083     return 0;
1084 }
1085 
1086 static QemuOptsList qcow_create_opts = {
1087     .name = "qcow-create-opts",
1088     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1089     .desc = {
1090         {
1091             .name = BLOCK_OPT_SIZE,
1092             .type = QEMU_OPT_SIZE,
1093             .help = "Virtual disk size"
1094         },
1095         {
1096             .name = BLOCK_OPT_BACKING_FILE,
1097             .type = QEMU_OPT_STRING,
1098             .help = "File name of a base image"
1099         },
1100         {
1101             .name = BLOCK_OPT_ENCRYPT,
1102             .type = QEMU_OPT_BOOL,
1103             .help = "Encrypt the image with format 'aes'. (Deprecated "
1104                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1105         },
1106         {
1107             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1108             .type = QEMU_OPT_STRING,
1109             .help = "Encrypt the image, format choices: 'aes'",
1110         },
1111         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1112         { /* end of list */ }
1113     }
1114 };
1115 
1116 static BlockDriver bdrv_qcow = {
1117     .format_name	= "qcow",
1118     .instance_size	= sizeof(BDRVQcowState),
1119     .bdrv_probe		= qcow_probe,
1120     .bdrv_open		= qcow_open,
1121     .bdrv_close		= qcow_close,
1122     .bdrv_child_perm        = bdrv_format_default_perms,
1123     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1124     .bdrv_create            = qcow_create,
1125     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1126     .supports_backing       = true,
1127 
1128     .bdrv_co_readv          = qcow_co_readv,
1129     .bdrv_co_writev         = qcow_co_writev,
1130     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1131 
1132     .bdrv_make_empty        = qcow_make_empty,
1133     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1134     .bdrv_get_info          = qcow_get_info,
1135 
1136     .create_opts            = &qcow_create_opts,
1137 };
1138 
1139 static void bdrv_qcow_init(void)
1140 {
1141     bdrv_register(&bdrv_qcow);
1142 }
1143 
1144 block_init(bdrv_qcow_init);
1145