xref: /openbmc/qemu/block/qcow.c (revision ab938ae4)
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         ret = bdrv_pwrite_sync(bs->file,
383                                s->l1_table_offset + l1_index * sizeof(tmp),
384                                &tmp, sizeof(tmp));
385         if (ret < 0) {
386             return ret;
387         }
388         new_l2_table = 1;
389     }
390     for(i = 0; i < L2_CACHE_SIZE; i++) {
391         if (l2_offset == s->l2_cache_offsets[i]) {
392             /* increment the hit count */
393             if (++s->l2_cache_counts[i] == 0xffffffff) {
394                 for(j = 0; j < L2_CACHE_SIZE; j++) {
395                     s->l2_cache_counts[j] >>= 1;
396                 }
397             }
398             l2_table = s->l2_cache + (i << s->l2_bits);
399             goto found;
400         }
401     }
402     /* not found: load a new entry in the least used one */
403     min_index = 0;
404     min_count = 0xffffffff;
405     for(i = 0; i < L2_CACHE_SIZE; i++) {
406         if (s->l2_cache_counts[i] < min_count) {
407             min_count = s->l2_cache_counts[i];
408             min_index = i;
409         }
410     }
411     l2_table = s->l2_cache + (min_index << s->l2_bits);
412     if (new_l2_table) {
413         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
414         ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
415                                s->l2_size * sizeof(uint64_t));
416         if (ret < 0) {
417             return ret;
418         }
419     } else {
420         ret = bdrv_pread(bs->file, l2_offset, l2_table,
421                          s->l2_size * sizeof(uint64_t));
422         if (ret < 0) {
423             return ret;
424         }
425     }
426     s->l2_cache_offsets[min_index] = l2_offset;
427     s->l2_cache_counts[min_index] = 1;
428  found:
429     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
430     cluster_offset = be64_to_cpu(l2_table[l2_index]);
431     if (!cluster_offset ||
432         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
433         if (!allocate)
434             return 0;
435         /* allocate a new cluster */
436         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
437             (n_end - n_start) < s->cluster_sectors) {
438             /* if the cluster is already compressed, we must
439                decompress it in the case it is not completely
440                overwritten */
441             if (decompress_cluster(bs, cluster_offset) < 0) {
442                 return -EIO;
443             }
444             cluster_offset = bdrv_getlength(bs->file->bs);
445             if ((int64_t) cluster_offset < 0) {
446                 return cluster_offset;
447             }
448             cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
449             /* write the cluster content */
450             ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
451                               s->cluster_size);
452             if (ret < 0) {
453                 return ret;
454             }
455         } else {
456             cluster_offset = bdrv_getlength(bs->file->bs);
457             if ((int64_t) cluster_offset < 0) {
458                 return cluster_offset;
459             }
460             if (allocate == 1) {
461                 /* round to cluster size */
462                 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
463                 if (cluster_offset + s->cluster_size > INT64_MAX) {
464                     return -E2BIG;
465                 }
466                 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size,
467                                     PREALLOC_MODE_OFF, NULL);
468                 if (ret < 0) {
469                     return ret;
470                 }
471                 /* if encrypted, we must initialize the cluster
472                    content which won't be written */
473                 if (bs->encrypted &&
474                     (n_end - n_start) < s->cluster_sectors) {
475                     uint64_t start_sect;
476                     assert(s->crypto);
477                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
478                     for(i = 0; i < s->cluster_sectors; i++) {
479                         if (i < n_start || i >= n_end) {
480                             memset(s->cluster_data, 0x00, 512);
481                             if (qcrypto_block_encrypt(s->crypto, start_sect + i,
482                                                       s->cluster_data,
483                                                       BDRV_SECTOR_SIZE,
484                                                       NULL) < 0) {
485                                 return -EIO;
486                             }
487                             ret = bdrv_pwrite(bs->file,
488                                               cluster_offset + i * 512,
489                                               s->cluster_data, 512);
490                             if (ret < 0) {
491                                 return ret;
492                             }
493                         }
494                     }
495                 }
496             } else if (allocate == 2) {
497                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
498                     (uint64_t)compressed_size << (63 - s->cluster_bits);
499             }
500         }
501         /* update L2 table */
502         tmp = cpu_to_be64(cluster_offset);
503         l2_table[l2_index] = tmp;
504         ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
505                                &tmp, sizeof(tmp));
506         if (ret < 0) {
507             return ret;
508         }
509     }
510     *result = cluster_offset;
511     return 1;
512 }
513 
514 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
515         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
516 {
517     BDRVQcowState *s = bs->opaque;
518     int index_in_cluster, n, ret;
519     uint64_t cluster_offset;
520 
521     qemu_co_mutex_lock(&s->lock);
522     ret = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0, &cluster_offset);
523     qemu_co_mutex_unlock(&s->lock);
524     if (ret < 0) {
525         return ret;
526     }
527     index_in_cluster = sector_num & (s->cluster_sectors - 1);
528     n = s->cluster_sectors - index_in_cluster;
529     if (n > nb_sectors)
530         n = nb_sectors;
531     *pnum = n;
532     if (!cluster_offset) {
533         return 0;
534     }
535     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) {
536         return BDRV_BLOCK_DATA;
537     }
538     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
539     *file = bs->file->bs;
540     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
541 }
542 
543 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
544                              const uint8_t *buf, int buf_size)
545 {
546     z_stream strm1, *strm = &strm1;
547     int ret, out_len;
548 
549     memset(strm, 0, sizeof(*strm));
550 
551     strm->next_in = (uint8_t *)buf;
552     strm->avail_in = buf_size;
553     strm->next_out = out_buf;
554     strm->avail_out = out_buf_size;
555 
556     ret = inflateInit2(strm, -12);
557     if (ret != Z_OK)
558         return -1;
559     ret = inflate(strm, Z_FINISH);
560     out_len = strm->next_out - out_buf;
561     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
562         out_len != out_buf_size) {
563         inflateEnd(strm);
564         return -1;
565     }
566     inflateEnd(strm);
567     return 0;
568 }
569 
570 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
571 {
572     BDRVQcowState *s = bs->opaque;
573     int ret, csize;
574     uint64_t coffset;
575 
576     coffset = cluster_offset & s->cluster_offset_mask;
577     if (s->cluster_cache_offset != coffset) {
578         csize = cluster_offset >> (63 - s->cluster_bits);
579         csize &= (s->cluster_size - 1);
580         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
581         if (ret != csize)
582             return -1;
583         if (decompress_buffer(s->cluster_cache, s->cluster_size,
584                               s->cluster_data, csize) < 0) {
585             return -1;
586         }
587         s->cluster_cache_offset = coffset;
588     }
589     return 0;
590 }
591 
592 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
593                          int nb_sectors, QEMUIOVector *qiov)
594 {
595     BDRVQcowState *s = bs->opaque;
596     int index_in_cluster;
597     int ret = 0, n;
598     uint64_t cluster_offset;
599     struct iovec hd_iov;
600     QEMUIOVector hd_qiov;
601     uint8_t *buf;
602     void *orig_buf;
603 
604     if (qiov->niov > 1) {
605         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
606         if (buf == NULL) {
607             return -ENOMEM;
608         }
609     } else {
610         orig_buf = NULL;
611         buf = (uint8_t *)qiov->iov->iov_base;
612     }
613 
614     qemu_co_mutex_lock(&s->lock);
615 
616     while (nb_sectors != 0) {
617         /* prepare next request */
618         ret = get_cluster_offset(bs, sector_num << 9,
619                                  0, 0, 0, 0, &cluster_offset);
620         if (ret < 0) {
621             break;
622         }
623         index_in_cluster = sector_num & (s->cluster_sectors - 1);
624         n = s->cluster_sectors - index_in_cluster;
625         if (n > nb_sectors) {
626             n = nb_sectors;
627         }
628 
629         if (!cluster_offset) {
630             if (bs->backing) {
631                 /* read from the base image */
632                 hd_iov.iov_base = (void *)buf;
633                 hd_iov.iov_len = n * 512;
634                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
635                 qemu_co_mutex_unlock(&s->lock);
636                 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov);
637                 qemu_co_mutex_lock(&s->lock);
638                 if (ret < 0) {
639                     break;
640                 }
641             } else {
642                 /* Note: in this case, no need to wait */
643                 memset(buf, 0, 512 * n);
644             }
645         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
646             /* add AIO support for compressed blocks ? */
647             if (decompress_cluster(bs, cluster_offset) < 0) {
648                 ret = -EIO;
649                 break;
650             }
651             memcpy(buf,
652                    s->cluster_cache + index_in_cluster * 512, 512 * n);
653         } else {
654             if ((cluster_offset & 511) != 0) {
655                 ret = -EIO;
656                 break;
657             }
658             hd_iov.iov_base = (void *)buf;
659             hd_iov.iov_len = n * 512;
660             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
661             qemu_co_mutex_unlock(&s->lock);
662             ret = bdrv_co_readv(bs->file,
663                                 (cluster_offset >> 9) + index_in_cluster,
664                                 n, &hd_qiov);
665             qemu_co_mutex_lock(&s->lock);
666             if (ret < 0) {
667                 break;
668             }
669             if (bs->encrypted) {
670                 assert(s->crypto);
671                 if (qcrypto_block_decrypt(s->crypto, sector_num, buf,
672                                           n * BDRV_SECTOR_SIZE, NULL) < 0) {
673                     ret = -EIO;
674                     break;
675                 }
676             }
677         }
678         ret = 0;
679 
680         nb_sectors -= n;
681         sector_num += n;
682         buf += n * 512;
683     }
684 
685     qemu_co_mutex_unlock(&s->lock);
686 
687     if (qiov->niov > 1) {
688         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
689         qemu_vfree(orig_buf);
690     }
691 
692     return ret;
693 }
694 
695 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
696                           int nb_sectors, QEMUIOVector *qiov)
697 {
698     BDRVQcowState *s = bs->opaque;
699     int index_in_cluster;
700     uint64_t cluster_offset;
701     int ret = 0, n;
702     struct iovec hd_iov;
703     QEMUIOVector hd_qiov;
704     uint8_t *buf;
705     void *orig_buf;
706 
707     s->cluster_cache_offset = -1; /* disable compressed cache */
708 
709     /* We must always copy the iov when encrypting, so we
710      * don't modify the original data buffer during encryption */
711     if (bs->encrypted || qiov->niov > 1) {
712         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
713         if (buf == NULL) {
714             return -ENOMEM;
715         }
716         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
717     } else {
718         orig_buf = NULL;
719         buf = (uint8_t *)qiov->iov->iov_base;
720     }
721 
722     qemu_co_mutex_lock(&s->lock);
723 
724     while (nb_sectors != 0) {
725 
726         index_in_cluster = sector_num & (s->cluster_sectors - 1);
727         n = s->cluster_sectors - index_in_cluster;
728         if (n > nb_sectors) {
729             n = nb_sectors;
730         }
731         ret = get_cluster_offset(bs, sector_num << 9, 1, 0,
732                                  index_in_cluster,
733                                  index_in_cluster + n, &cluster_offset);
734         if (ret < 0) {
735             break;
736         }
737         if (!cluster_offset || (cluster_offset & 511) != 0) {
738             ret = -EIO;
739             break;
740         }
741         if (bs->encrypted) {
742             assert(s->crypto);
743             if (qcrypto_block_encrypt(s->crypto, sector_num, buf,
744                                       n * BDRV_SECTOR_SIZE, NULL) < 0) {
745                 ret = -EIO;
746                 break;
747             }
748         }
749 
750         hd_iov.iov_base = (void *)buf;
751         hd_iov.iov_len = n * 512;
752         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
753         qemu_co_mutex_unlock(&s->lock);
754         ret = bdrv_co_writev(bs->file,
755                              (cluster_offset >> 9) + index_in_cluster,
756                              n, &hd_qiov);
757         qemu_co_mutex_lock(&s->lock);
758         if (ret < 0) {
759             break;
760         }
761         ret = 0;
762 
763         nb_sectors -= n;
764         sector_num += n;
765         buf += n * 512;
766     }
767     qemu_co_mutex_unlock(&s->lock);
768 
769     qemu_vfree(orig_buf);
770 
771     return ret;
772 }
773 
774 static void qcow_close(BlockDriverState *bs)
775 {
776     BDRVQcowState *s = bs->opaque;
777 
778     qcrypto_block_free(s->crypto);
779     s->crypto = NULL;
780     g_free(s->l1_table);
781     qemu_vfree(s->l2_cache);
782     g_free(s->cluster_cache);
783     g_free(s->cluster_data);
784 
785     migrate_del_blocker(s->migration_blocker);
786     error_free(s->migration_blocker);
787 }
788 
789 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
790 {
791     int header_size, backing_filename_len, l1_size, shift, i;
792     QCowHeader header;
793     uint8_t *tmp;
794     int64_t total_size = 0;
795     char *backing_file = NULL;
796     Error *local_err = NULL;
797     int ret;
798     BlockBackend *qcow_blk;
799     char *encryptfmt = NULL;
800     QDict *options;
801     QDict *encryptopts = NULL;
802     QCryptoBlockCreateOptions *crypto_opts = NULL;
803     QCryptoBlock *crypto = NULL;
804 
805     /* Read out options */
806     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
807                           BDRV_SECTOR_SIZE);
808     if (total_size == 0) {
809         error_setg(errp, "Image size is too small, cannot be zero length");
810         ret = -EINVAL;
811         goto cleanup;
812     }
813 
814     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
815     encryptfmt = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
816     if (encryptfmt) {
817         if (qemu_opt_get(opts, BLOCK_OPT_ENCRYPT)) {
818             error_setg(errp, "Options " BLOCK_OPT_ENCRYPT " and "
819                        BLOCK_OPT_ENCRYPT_FORMAT " are mutually exclusive");
820             ret = -EINVAL;
821             goto cleanup;
822         }
823     } else if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
824         encryptfmt = g_strdup("aes");
825     }
826 
827     ret = bdrv_create_file(filename, opts, &local_err);
828     if (ret < 0) {
829         error_propagate(errp, local_err);
830         goto cleanup;
831     }
832 
833     qcow_blk = blk_new_open(filename, NULL, NULL,
834                             BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
835                             &local_err);
836     if (qcow_blk == NULL) {
837         error_propagate(errp, local_err);
838         ret = -EIO;
839         goto cleanup;
840     }
841 
842     blk_set_allow_write_beyond_eof(qcow_blk, true);
843 
844     ret = blk_truncate(qcow_blk, 0, PREALLOC_MODE_OFF, errp);
845     if (ret < 0) {
846         goto exit;
847     }
848 
849     memset(&header, 0, sizeof(header));
850     header.magic = cpu_to_be32(QCOW_MAGIC);
851     header.version = cpu_to_be32(QCOW_VERSION);
852     header.size = cpu_to_be64(total_size);
853     header_size = sizeof(header);
854     backing_filename_len = 0;
855     if (backing_file) {
856         if (strcmp(backing_file, "fat:")) {
857             header.backing_file_offset = cpu_to_be64(header_size);
858             backing_filename_len = strlen(backing_file);
859             header.backing_file_size = cpu_to_be32(backing_filename_len);
860             header_size += backing_filename_len;
861         } else {
862             /* special backing file for vvfat */
863             g_free(backing_file);
864             backing_file = NULL;
865         }
866         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
867                                     unmodified sectors */
868         header.l2_bits = 12; /* 32 KB L2 tables */
869     } else {
870         header.cluster_bits = 12; /* 4 KB clusters */
871         header.l2_bits = 9; /* 4 KB L2 tables */
872     }
873     header_size = (header_size + 7) & ~7;
874     shift = header.cluster_bits + header.l2_bits;
875     l1_size = (total_size + (1LL << shift) - 1) >> shift;
876 
877     header.l1_table_offset = cpu_to_be64(header_size);
878 
879     options = qemu_opts_to_qdict(opts, NULL);
880     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
881     QDECREF(options);
882     if (encryptfmt) {
883         if (!g_str_equal(encryptfmt, "aes")) {
884             error_setg(errp, "Unknown encryption format '%s', expected 'aes'",
885                        encryptfmt);
886             ret = -EINVAL;
887             goto exit;
888         }
889         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
890 
891         crypto_opts = block_crypto_create_opts_init(
892             Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp);
893         if (!crypto_opts) {
894             ret = -EINVAL;
895             goto exit;
896         }
897 
898         crypto = qcrypto_block_create(crypto_opts, "encrypt.",
899                                       NULL, NULL, NULL, errp);
900         if (!crypto) {
901             ret = -EINVAL;
902             goto exit;
903         }
904     } else {
905         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
906     }
907 
908     /* write all the data */
909     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
910     if (ret != sizeof(header)) {
911         goto exit;
912     }
913 
914     if (backing_file) {
915         ret = blk_pwrite(qcow_blk, sizeof(header),
916                          backing_file, backing_filename_len, 0);
917         if (ret != backing_filename_len) {
918             goto exit;
919         }
920     }
921 
922     tmp = g_malloc0(BDRV_SECTOR_SIZE);
923     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
924          i++) {
925         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
926                          tmp, BDRV_SECTOR_SIZE, 0);
927         if (ret != BDRV_SECTOR_SIZE) {
928             g_free(tmp);
929             goto exit;
930         }
931     }
932 
933     g_free(tmp);
934     ret = 0;
935 exit:
936     blk_unref(qcow_blk);
937 cleanup:
938     QDECREF(encryptopts);
939     g_free(encryptfmt);
940     qcrypto_block_free(crypto);
941     qapi_free_QCryptoBlockCreateOptions(crypto_opts);
942     g_free(backing_file);
943     return ret;
944 }
945 
946 static int qcow_make_empty(BlockDriverState *bs)
947 {
948     BDRVQcowState *s = bs->opaque;
949     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
950     int ret;
951 
952     memset(s->l1_table, 0, l1_length);
953     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
954             l1_length) < 0)
955         return -1;
956     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length,
957                         PREALLOC_MODE_OFF, NULL);
958     if (ret < 0)
959         return ret;
960 
961     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
962     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
963     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
964 
965     return 0;
966 }
967 
968 /* XXX: put compressed sectors first, then all the cluster aligned
969    tables to avoid losing bytes in alignment */
970 static coroutine_fn int
971 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
972                            uint64_t bytes, QEMUIOVector *qiov)
973 {
974     BDRVQcowState *s = bs->opaque;
975     QEMUIOVector hd_qiov;
976     struct iovec iov;
977     z_stream strm;
978     int ret, out_len;
979     uint8_t *buf, *out_buf;
980     uint64_t cluster_offset;
981 
982     buf = qemu_blockalign(bs, s->cluster_size);
983     if (bytes != s->cluster_size) {
984         if (bytes > s->cluster_size ||
985             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
986         {
987             qemu_vfree(buf);
988             return -EINVAL;
989         }
990         /* Zero-pad last write if image size is not cluster aligned */
991         memset(buf + bytes, 0, s->cluster_size - bytes);
992     }
993     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
994 
995     out_buf = g_malloc(s->cluster_size);
996 
997     /* best compression, small window, no zlib header */
998     memset(&strm, 0, sizeof(strm));
999     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1000                        Z_DEFLATED, -12,
1001                        9, Z_DEFAULT_STRATEGY);
1002     if (ret != 0) {
1003         ret = -EINVAL;
1004         goto fail;
1005     }
1006 
1007     strm.avail_in = s->cluster_size;
1008     strm.next_in = (uint8_t *)buf;
1009     strm.avail_out = s->cluster_size;
1010     strm.next_out = out_buf;
1011 
1012     ret = deflate(&strm, Z_FINISH);
1013     if (ret != Z_STREAM_END && ret != Z_OK) {
1014         deflateEnd(&strm);
1015         ret = -EINVAL;
1016         goto fail;
1017     }
1018     out_len = strm.next_out - out_buf;
1019 
1020     deflateEnd(&strm);
1021 
1022     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1023         /* could not compress: write normal cluster */
1024         ret = qcow_co_writev(bs, offset >> BDRV_SECTOR_BITS,
1025                              bytes >> BDRV_SECTOR_BITS, qiov);
1026         if (ret < 0) {
1027             goto fail;
1028         }
1029         goto success;
1030     }
1031     qemu_co_mutex_lock(&s->lock);
1032     ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1033     qemu_co_mutex_unlock(&s->lock);
1034     if (ret < 0) {
1035         goto fail;
1036     }
1037     if (cluster_offset == 0) {
1038         ret = -EIO;
1039         goto fail;
1040     }
1041     cluster_offset &= s->cluster_offset_mask;
1042 
1043     iov = (struct iovec) {
1044         .iov_base   = out_buf,
1045         .iov_len    = out_len,
1046     };
1047     qemu_iovec_init_external(&hd_qiov, &iov, 1);
1048     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
1049     if (ret < 0) {
1050         goto fail;
1051     }
1052 success:
1053     ret = 0;
1054 fail:
1055     qemu_vfree(buf);
1056     g_free(out_buf);
1057     return ret;
1058 }
1059 
1060 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1061 {
1062     BDRVQcowState *s = bs->opaque;
1063     bdi->cluster_size = s->cluster_size;
1064     return 0;
1065 }
1066 
1067 static QemuOptsList qcow_create_opts = {
1068     .name = "qcow-create-opts",
1069     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1070     .desc = {
1071         {
1072             .name = BLOCK_OPT_SIZE,
1073             .type = QEMU_OPT_SIZE,
1074             .help = "Virtual disk size"
1075         },
1076         {
1077             .name = BLOCK_OPT_BACKING_FILE,
1078             .type = QEMU_OPT_STRING,
1079             .help = "File name of a base image"
1080         },
1081         {
1082             .name = BLOCK_OPT_ENCRYPT,
1083             .type = QEMU_OPT_BOOL,
1084             .help = "Encrypt the image with format 'aes'. (Deprecated "
1085                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1086         },
1087         {
1088             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1089             .type = QEMU_OPT_STRING,
1090             .help = "Encrypt the image, format choices: 'aes'",
1091         },
1092         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1093         { /* end of list */ }
1094     }
1095 };
1096 
1097 static BlockDriver bdrv_qcow = {
1098     .format_name	= "qcow",
1099     .instance_size	= sizeof(BDRVQcowState),
1100     .bdrv_probe		= qcow_probe,
1101     .bdrv_open		= qcow_open,
1102     .bdrv_close		= qcow_close,
1103     .bdrv_child_perm        = bdrv_format_default_perms,
1104     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1105     .bdrv_create            = qcow_create,
1106     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1107     .supports_backing       = true,
1108 
1109     .bdrv_co_readv          = qcow_co_readv,
1110     .bdrv_co_writev         = qcow_co_writev,
1111     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1112 
1113     .bdrv_make_empty        = qcow_make_empty,
1114     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1115     .bdrv_get_info          = qcow_get_info,
1116 
1117     .create_opts            = &qcow_create_opts,
1118 };
1119 
1120 static void bdrv_qcow_init(void)
1121 {
1122     bdrv_register(&bdrv_qcow);
1123 }
1124 
1125 block_init(bdrv_qcow_init);
1126