xref: /openbmc/qemu/block/qcow.c (revision 8908eb1a)
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.
351  */
352 static uint64_t get_cluster_offset(BlockDriverState *bs,
353                                    uint64_t offset, int allocate,
354                                    int compressed_size,
355                                    int n_start, int n_end)
356 {
357     BDRVQcowState *s = bs->opaque;
358     int min_index, i, j, l1_index, l2_index;
359     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
360     uint32_t min_count;
361     int new_l2_table;
362 
363     l1_index = offset >> (s->l2_bits + s->cluster_bits);
364     l2_offset = s->l1_table[l1_index];
365     new_l2_table = 0;
366     if (!l2_offset) {
367         if (!allocate)
368             return 0;
369         /* allocate a new l2 entry */
370         l2_offset = bdrv_getlength(bs->file->bs);
371         /* round to cluster size */
372         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
373         /* update the L1 entry */
374         s->l1_table[l1_index] = l2_offset;
375         tmp = cpu_to_be64(l2_offset);
376         if (bdrv_pwrite_sync(bs->file,
377                 s->l1_table_offset + l1_index * sizeof(tmp),
378                 &tmp, sizeof(tmp)) < 0)
379             return 0;
380         new_l2_table = 1;
381     }
382     for(i = 0; i < L2_CACHE_SIZE; i++) {
383         if (l2_offset == s->l2_cache_offsets[i]) {
384             /* increment the hit count */
385             if (++s->l2_cache_counts[i] == 0xffffffff) {
386                 for(j = 0; j < L2_CACHE_SIZE; j++) {
387                     s->l2_cache_counts[j] >>= 1;
388                 }
389             }
390             l2_table = s->l2_cache + (i << s->l2_bits);
391             goto found;
392         }
393     }
394     /* not found: load a new entry in the least used one */
395     min_index = 0;
396     min_count = 0xffffffff;
397     for(i = 0; i < L2_CACHE_SIZE; i++) {
398         if (s->l2_cache_counts[i] < min_count) {
399             min_count = s->l2_cache_counts[i];
400             min_index = i;
401         }
402     }
403     l2_table = s->l2_cache + (min_index << s->l2_bits);
404     if (new_l2_table) {
405         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
406         if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
407                 s->l2_size * sizeof(uint64_t)) < 0)
408             return 0;
409     } else {
410         if (bdrv_pread(bs->file, l2_offset, l2_table,
411                        s->l2_size * sizeof(uint64_t)) !=
412             s->l2_size * sizeof(uint64_t))
413             return 0;
414     }
415     s->l2_cache_offsets[min_index] = l2_offset;
416     s->l2_cache_counts[min_index] = 1;
417  found:
418     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
419     cluster_offset = be64_to_cpu(l2_table[l2_index]);
420     if (!cluster_offset ||
421         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
422         if (!allocate)
423             return 0;
424         /* allocate a new cluster */
425         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
426             (n_end - n_start) < s->cluster_sectors) {
427             /* if the cluster is already compressed, we must
428                decompress it in the case it is not completely
429                overwritten */
430             if (decompress_cluster(bs, cluster_offset) < 0)
431                 return 0;
432             cluster_offset = bdrv_getlength(bs->file->bs);
433             cluster_offset = (cluster_offset + s->cluster_size - 1) &
434                 ~(s->cluster_size - 1);
435             /* write the cluster content */
436             if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
437                             s->cluster_size) !=
438                 s->cluster_size)
439                 return -1;
440         } else {
441             cluster_offset = bdrv_getlength(bs->file->bs);
442             if (allocate == 1) {
443                 /* round to cluster size */
444                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
445                     ~(s->cluster_size - 1);
446                 bdrv_truncate(bs->file, cluster_offset + s->cluster_size,
447                               PREALLOC_MODE_OFF, NULL);
448                 /* if encrypted, we must initialize the cluster
449                    content which won't be written */
450                 if (bs->encrypted &&
451                     (n_end - n_start) < s->cluster_sectors) {
452                     uint64_t start_sect;
453                     assert(s->crypto);
454                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
455                     for(i = 0; i < s->cluster_sectors; i++) {
456                         if (i < n_start || i >= n_end) {
457                             Error *err = NULL;
458                             memset(s->cluster_data, 0x00, 512);
459                             if (qcrypto_block_encrypt(s->crypto, start_sect + i,
460                                                       s->cluster_data,
461                                                       BDRV_SECTOR_SIZE,
462                                                       &err) < 0) {
463                                 error_free(err);
464                                 errno = EIO;
465                                 return -1;
466                             }
467                             if (bdrv_pwrite(bs->file,
468                                             cluster_offset + i * 512,
469                                             s->cluster_data, 512) != 512)
470                                 return -1;
471                         }
472                     }
473                 }
474             } else if (allocate == 2) {
475                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
476                     (uint64_t)compressed_size << (63 - s->cluster_bits);
477             }
478         }
479         /* update L2 table */
480         tmp = cpu_to_be64(cluster_offset);
481         l2_table[l2_index] = tmp;
482         if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
483                 &tmp, sizeof(tmp)) < 0)
484             return 0;
485     }
486     return cluster_offset;
487 }
488 
489 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
490         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
491 {
492     BDRVQcowState *s = bs->opaque;
493     int index_in_cluster, n;
494     uint64_t cluster_offset;
495 
496     qemu_co_mutex_lock(&s->lock);
497     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
498     qemu_co_mutex_unlock(&s->lock);
499     index_in_cluster = sector_num & (s->cluster_sectors - 1);
500     n = s->cluster_sectors - index_in_cluster;
501     if (n > nb_sectors)
502         n = nb_sectors;
503     *pnum = n;
504     if (!cluster_offset) {
505         return 0;
506     }
507     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) {
508         return BDRV_BLOCK_DATA;
509     }
510     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
511     *file = bs->file->bs;
512     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
513 }
514 
515 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
516                              const uint8_t *buf, int buf_size)
517 {
518     z_stream strm1, *strm = &strm1;
519     int ret, out_len;
520 
521     memset(strm, 0, sizeof(*strm));
522 
523     strm->next_in = (uint8_t *)buf;
524     strm->avail_in = buf_size;
525     strm->next_out = out_buf;
526     strm->avail_out = out_buf_size;
527 
528     ret = inflateInit2(strm, -12);
529     if (ret != Z_OK)
530         return -1;
531     ret = inflate(strm, Z_FINISH);
532     out_len = strm->next_out - out_buf;
533     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
534         out_len != out_buf_size) {
535         inflateEnd(strm);
536         return -1;
537     }
538     inflateEnd(strm);
539     return 0;
540 }
541 
542 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
543 {
544     BDRVQcowState *s = bs->opaque;
545     int ret, csize;
546     uint64_t coffset;
547 
548     coffset = cluster_offset & s->cluster_offset_mask;
549     if (s->cluster_cache_offset != coffset) {
550         csize = cluster_offset >> (63 - s->cluster_bits);
551         csize &= (s->cluster_size - 1);
552         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
553         if (ret != csize)
554             return -1;
555         if (decompress_buffer(s->cluster_cache, s->cluster_size,
556                               s->cluster_data, csize) < 0) {
557             return -1;
558         }
559         s->cluster_cache_offset = coffset;
560     }
561     return 0;
562 }
563 
564 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
565                          int nb_sectors, QEMUIOVector *qiov)
566 {
567     BDRVQcowState *s = bs->opaque;
568     int index_in_cluster;
569     int ret = 0, n;
570     uint64_t cluster_offset;
571     struct iovec hd_iov;
572     QEMUIOVector hd_qiov;
573     uint8_t *buf;
574     void *orig_buf;
575     Error *err = NULL;
576 
577     if (qiov->niov > 1) {
578         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
579         if (buf == NULL) {
580             return -ENOMEM;
581         }
582     } else {
583         orig_buf = NULL;
584         buf = (uint8_t *)qiov->iov->iov_base;
585     }
586 
587     qemu_co_mutex_lock(&s->lock);
588 
589     while (nb_sectors != 0) {
590         /* prepare next request */
591         cluster_offset = get_cluster_offset(bs, sector_num << 9,
592                                                  0, 0, 0, 0);
593         index_in_cluster = sector_num & (s->cluster_sectors - 1);
594         n = s->cluster_sectors - index_in_cluster;
595         if (n > nb_sectors) {
596             n = nb_sectors;
597         }
598 
599         if (!cluster_offset) {
600             if (bs->backing) {
601                 /* read from the base image */
602                 hd_iov.iov_base = (void *)buf;
603                 hd_iov.iov_len = n * 512;
604                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
605                 qemu_co_mutex_unlock(&s->lock);
606                 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov);
607                 qemu_co_mutex_lock(&s->lock);
608                 if (ret < 0) {
609                     goto fail;
610                 }
611             } else {
612                 /* Note: in this case, no need to wait */
613                 memset(buf, 0, 512 * n);
614             }
615         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
616             /* add AIO support for compressed blocks ? */
617             if (decompress_cluster(bs, cluster_offset) < 0) {
618                 goto fail;
619             }
620             memcpy(buf,
621                    s->cluster_cache + index_in_cluster * 512, 512 * n);
622         } else {
623             if ((cluster_offset & 511) != 0) {
624                 goto fail;
625             }
626             hd_iov.iov_base = (void *)buf;
627             hd_iov.iov_len = n * 512;
628             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
629             qemu_co_mutex_unlock(&s->lock);
630             ret = bdrv_co_readv(bs->file,
631                                 (cluster_offset >> 9) + index_in_cluster,
632                                 n, &hd_qiov);
633             qemu_co_mutex_lock(&s->lock);
634             if (ret < 0) {
635                 break;
636             }
637             if (bs->encrypted) {
638                 assert(s->crypto);
639                 if (qcrypto_block_decrypt(s->crypto, sector_num, buf,
640                                           n * BDRV_SECTOR_SIZE, &err) < 0) {
641                     goto fail;
642                 }
643             }
644         }
645         ret = 0;
646 
647         nb_sectors -= n;
648         sector_num += n;
649         buf += n * 512;
650     }
651 
652 done:
653     qemu_co_mutex_unlock(&s->lock);
654 
655     if (qiov->niov > 1) {
656         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
657         qemu_vfree(orig_buf);
658     }
659 
660     return ret;
661 
662 fail:
663     error_free(err);
664     ret = -EIO;
665     goto done;
666 }
667 
668 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
669                           int nb_sectors, QEMUIOVector *qiov)
670 {
671     BDRVQcowState *s = bs->opaque;
672     int index_in_cluster;
673     uint64_t cluster_offset;
674     int ret = 0, n;
675     struct iovec hd_iov;
676     QEMUIOVector hd_qiov;
677     uint8_t *buf;
678     void *orig_buf;
679 
680     s->cluster_cache_offset = -1; /* disable compressed cache */
681 
682     /* We must always copy the iov when encrypting, so we
683      * don't modify the original data buffer during encryption */
684     if (bs->encrypted || qiov->niov > 1) {
685         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
686         if (buf == NULL) {
687             return -ENOMEM;
688         }
689         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
690     } else {
691         orig_buf = NULL;
692         buf = (uint8_t *)qiov->iov->iov_base;
693     }
694 
695     qemu_co_mutex_lock(&s->lock);
696 
697     while (nb_sectors != 0) {
698 
699         index_in_cluster = sector_num & (s->cluster_sectors - 1);
700         n = s->cluster_sectors - index_in_cluster;
701         if (n > nb_sectors) {
702             n = nb_sectors;
703         }
704         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
705                                             index_in_cluster,
706                                             index_in_cluster + n);
707         if (!cluster_offset || (cluster_offset & 511) != 0) {
708             ret = -EIO;
709             break;
710         }
711         if (bs->encrypted) {
712             Error *err = NULL;
713             assert(s->crypto);
714             if (qcrypto_block_encrypt(s->crypto, sector_num, buf,
715                                       n * BDRV_SECTOR_SIZE, &err) < 0) {
716                 error_free(err);
717                 ret = -EIO;
718                 break;
719             }
720         }
721 
722         hd_iov.iov_base = (void *)buf;
723         hd_iov.iov_len = n * 512;
724         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
725         qemu_co_mutex_unlock(&s->lock);
726         ret = bdrv_co_writev(bs->file,
727                              (cluster_offset >> 9) + index_in_cluster,
728                              n, &hd_qiov);
729         qemu_co_mutex_lock(&s->lock);
730         if (ret < 0) {
731             break;
732         }
733         ret = 0;
734 
735         nb_sectors -= n;
736         sector_num += n;
737         buf += n * 512;
738     }
739     qemu_co_mutex_unlock(&s->lock);
740 
741     qemu_vfree(orig_buf);
742 
743     return ret;
744 }
745 
746 static void qcow_close(BlockDriverState *bs)
747 {
748     BDRVQcowState *s = bs->opaque;
749 
750     qcrypto_block_free(s->crypto);
751     s->crypto = NULL;
752     g_free(s->l1_table);
753     qemu_vfree(s->l2_cache);
754     g_free(s->cluster_cache);
755     g_free(s->cluster_data);
756 
757     migrate_del_blocker(s->migration_blocker);
758     error_free(s->migration_blocker);
759 }
760 
761 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
762 {
763     int header_size, backing_filename_len, l1_size, shift, i;
764     QCowHeader header;
765     uint8_t *tmp;
766     int64_t total_size = 0;
767     char *backing_file = NULL;
768     Error *local_err = NULL;
769     int ret;
770     BlockBackend *qcow_blk;
771     char *encryptfmt = NULL;
772     QDict *options;
773     QDict *encryptopts = NULL;
774     QCryptoBlockCreateOptions *crypto_opts = NULL;
775     QCryptoBlock *crypto = NULL;
776 
777     /* Read out options */
778     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
779                           BDRV_SECTOR_SIZE);
780     if (total_size == 0) {
781         error_setg(errp, "Image size is too small, cannot be zero length");
782         ret = -EINVAL;
783         goto cleanup;
784     }
785 
786     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
787     encryptfmt = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
788     if (encryptfmt) {
789         if (qemu_opt_get(opts, BLOCK_OPT_ENCRYPT)) {
790             error_setg(errp, "Options " BLOCK_OPT_ENCRYPT " and "
791                        BLOCK_OPT_ENCRYPT_FORMAT " are mutually exclusive");
792             ret = -EINVAL;
793             goto cleanup;
794         }
795     } else if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
796         encryptfmt = g_strdup("aes");
797     }
798 
799     ret = bdrv_create_file(filename, opts, &local_err);
800     if (ret < 0) {
801         error_propagate(errp, local_err);
802         goto cleanup;
803     }
804 
805     qcow_blk = blk_new_open(filename, NULL, NULL,
806                             BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
807                             &local_err);
808     if (qcow_blk == NULL) {
809         error_propagate(errp, local_err);
810         ret = -EIO;
811         goto cleanup;
812     }
813 
814     blk_set_allow_write_beyond_eof(qcow_blk, true);
815 
816     ret = blk_truncate(qcow_blk, 0, PREALLOC_MODE_OFF, errp);
817     if (ret < 0) {
818         goto exit;
819     }
820 
821     memset(&header, 0, sizeof(header));
822     header.magic = cpu_to_be32(QCOW_MAGIC);
823     header.version = cpu_to_be32(QCOW_VERSION);
824     header.size = cpu_to_be64(total_size);
825     header_size = sizeof(header);
826     backing_filename_len = 0;
827     if (backing_file) {
828         if (strcmp(backing_file, "fat:")) {
829             header.backing_file_offset = cpu_to_be64(header_size);
830             backing_filename_len = strlen(backing_file);
831             header.backing_file_size = cpu_to_be32(backing_filename_len);
832             header_size += backing_filename_len;
833         } else {
834             /* special backing file for vvfat */
835             g_free(backing_file);
836             backing_file = NULL;
837         }
838         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
839                                     unmodified sectors */
840         header.l2_bits = 12; /* 32 KB L2 tables */
841     } else {
842         header.cluster_bits = 12; /* 4 KB clusters */
843         header.l2_bits = 9; /* 4 KB L2 tables */
844     }
845     header_size = (header_size + 7) & ~7;
846     shift = header.cluster_bits + header.l2_bits;
847     l1_size = (total_size + (1LL << shift) - 1) >> shift;
848 
849     header.l1_table_offset = cpu_to_be64(header_size);
850 
851     options = qemu_opts_to_qdict(opts, NULL);
852     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
853     QDECREF(options);
854     if (encryptfmt) {
855         if (!g_str_equal(encryptfmt, "aes")) {
856             error_setg(errp, "Unknown encryption format '%s', expected 'aes'",
857                        encryptfmt);
858             ret = -EINVAL;
859             goto exit;
860         }
861         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
862 
863         crypto_opts = block_crypto_create_opts_init(
864             Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp);
865         if (!crypto_opts) {
866             ret = -EINVAL;
867             goto exit;
868         }
869 
870         crypto = qcrypto_block_create(crypto_opts, "encrypt.",
871                                       NULL, NULL, NULL, errp);
872         if (!crypto) {
873             ret = -EINVAL;
874             goto exit;
875         }
876     } else {
877         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
878     }
879 
880     /* write all the data */
881     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
882     if (ret != sizeof(header)) {
883         goto exit;
884     }
885 
886     if (backing_file) {
887         ret = blk_pwrite(qcow_blk, sizeof(header),
888                          backing_file, backing_filename_len, 0);
889         if (ret != backing_filename_len) {
890             goto exit;
891         }
892     }
893 
894     tmp = g_malloc0(BDRV_SECTOR_SIZE);
895     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
896          i++) {
897         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
898                          tmp, BDRV_SECTOR_SIZE, 0);
899         if (ret != BDRV_SECTOR_SIZE) {
900             g_free(tmp);
901             goto exit;
902         }
903     }
904 
905     g_free(tmp);
906     ret = 0;
907 exit:
908     blk_unref(qcow_blk);
909 cleanup:
910     QDECREF(encryptopts);
911     g_free(encryptfmt);
912     qcrypto_block_free(crypto);
913     qapi_free_QCryptoBlockCreateOptions(crypto_opts);
914     g_free(backing_file);
915     return ret;
916 }
917 
918 static int qcow_make_empty(BlockDriverState *bs)
919 {
920     BDRVQcowState *s = bs->opaque;
921     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
922     int ret;
923 
924     memset(s->l1_table, 0, l1_length);
925     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
926             l1_length) < 0)
927         return -1;
928     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length,
929                         PREALLOC_MODE_OFF, NULL);
930     if (ret < 0)
931         return ret;
932 
933     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
934     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
935     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
936 
937     return 0;
938 }
939 
940 /* XXX: put compressed sectors first, then all the cluster aligned
941    tables to avoid losing bytes in alignment */
942 static coroutine_fn int
943 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
944                            uint64_t bytes, QEMUIOVector *qiov)
945 {
946     BDRVQcowState *s = bs->opaque;
947     QEMUIOVector hd_qiov;
948     struct iovec iov;
949     z_stream strm;
950     int ret, out_len;
951     uint8_t *buf, *out_buf;
952     uint64_t cluster_offset;
953 
954     buf = qemu_blockalign(bs, s->cluster_size);
955     if (bytes != s->cluster_size) {
956         if (bytes > s->cluster_size ||
957             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
958         {
959             qemu_vfree(buf);
960             return -EINVAL;
961         }
962         /* Zero-pad last write if image size is not cluster aligned */
963         memset(buf + bytes, 0, s->cluster_size - bytes);
964     }
965     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
966 
967     out_buf = g_malloc(s->cluster_size);
968 
969     /* best compression, small window, no zlib header */
970     memset(&strm, 0, sizeof(strm));
971     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
972                        Z_DEFLATED, -12,
973                        9, Z_DEFAULT_STRATEGY);
974     if (ret != 0) {
975         ret = -EINVAL;
976         goto fail;
977     }
978 
979     strm.avail_in = s->cluster_size;
980     strm.next_in = (uint8_t *)buf;
981     strm.avail_out = s->cluster_size;
982     strm.next_out = out_buf;
983 
984     ret = deflate(&strm, Z_FINISH);
985     if (ret != Z_STREAM_END && ret != Z_OK) {
986         deflateEnd(&strm);
987         ret = -EINVAL;
988         goto fail;
989     }
990     out_len = strm.next_out - out_buf;
991 
992     deflateEnd(&strm);
993 
994     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
995         /* could not compress: write normal cluster */
996         ret = qcow_co_writev(bs, offset >> BDRV_SECTOR_BITS,
997                              bytes >> BDRV_SECTOR_BITS, qiov);
998         if (ret < 0) {
999             goto fail;
1000         }
1001         goto success;
1002     }
1003     qemu_co_mutex_lock(&s->lock);
1004     cluster_offset = get_cluster_offset(bs, offset, 2, out_len, 0, 0);
1005     qemu_co_mutex_unlock(&s->lock);
1006     if (cluster_offset == 0) {
1007         ret = -EIO;
1008         goto fail;
1009     }
1010     cluster_offset &= s->cluster_offset_mask;
1011 
1012     iov = (struct iovec) {
1013         .iov_base   = out_buf,
1014         .iov_len    = out_len,
1015     };
1016     qemu_iovec_init_external(&hd_qiov, &iov, 1);
1017     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
1018     if (ret < 0) {
1019         goto fail;
1020     }
1021 success:
1022     ret = 0;
1023 fail:
1024     qemu_vfree(buf);
1025     g_free(out_buf);
1026     return ret;
1027 }
1028 
1029 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1030 {
1031     BDRVQcowState *s = bs->opaque;
1032     bdi->cluster_size = s->cluster_size;
1033     return 0;
1034 }
1035 
1036 static QemuOptsList qcow_create_opts = {
1037     .name = "qcow-create-opts",
1038     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1039     .desc = {
1040         {
1041             .name = BLOCK_OPT_SIZE,
1042             .type = QEMU_OPT_SIZE,
1043             .help = "Virtual disk size"
1044         },
1045         {
1046             .name = BLOCK_OPT_BACKING_FILE,
1047             .type = QEMU_OPT_STRING,
1048             .help = "File name of a base image"
1049         },
1050         {
1051             .name = BLOCK_OPT_ENCRYPT,
1052             .type = QEMU_OPT_BOOL,
1053             .help = "Encrypt the image with format 'aes'. (Deprecated "
1054                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1055         },
1056         {
1057             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1058             .type = QEMU_OPT_STRING,
1059             .help = "Encrypt the image, format choices: 'aes'",
1060         },
1061         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1062         { /* end of list */ }
1063     }
1064 };
1065 
1066 static BlockDriver bdrv_qcow = {
1067     .format_name	= "qcow",
1068     .instance_size	= sizeof(BDRVQcowState),
1069     .bdrv_probe		= qcow_probe,
1070     .bdrv_open		= qcow_open,
1071     .bdrv_close		= qcow_close,
1072     .bdrv_child_perm        = bdrv_format_default_perms,
1073     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1074     .bdrv_create            = qcow_create,
1075     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1076     .supports_backing       = true,
1077 
1078     .bdrv_co_readv          = qcow_co_readv,
1079     .bdrv_co_writev         = qcow_co_writev,
1080     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1081 
1082     .bdrv_make_empty        = qcow_make_empty,
1083     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1084     .bdrv_get_info          = qcow_get_info,
1085 
1086     .create_opts            = &qcow_create_opts,
1087 };
1088 
1089 static void bdrv_qcow_init(void)
1090 {
1091     bdrv_register(&bdrv_qcow);
1092 }
1093 
1094 block_init(bdrv_qcow_init);
1095