xref: /openbmc/qemu/block/qcow.c (revision 922a01a013d2270682a188258cbccacfecf8129c)
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 
25 #include "qemu/osdep.h"
26 #include "qapi/error.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/option.h"
32 #include "qemu/bswap.h"
33 #include <zlib.h>
34 #include "qapi/qmp/qdict.h"
35 #include "qapi/qmp/qstring.h"
36 #include "crypto/block.h"
37 #include "migration/blocker.h"
38 #include "block/crypto.h"
39 
40 /**************************************************************/
41 /* QEMU COW block driver with compression and encryption support */
42 
43 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
44 #define QCOW_VERSION 1
45 
46 #define QCOW_CRYPT_NONE 0
47 #define QCOW_CRYPT_AES  1
48 
49 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
50 
51 typedef struct QCowHeader {
52     uint32_t magic;
53     uint32_t version;
54     uint64_t backing_file_offset;
55     uint32_t backing_file_size;
56     uint32_t mtime;
57     uint64_t size; /* in bytes */
58     uint8_t cluster_bits;
59     uint8_t l2_bits;
60     uint16_t padding;
61     uint32_t crypt_method;
62     uint64_t l1_table_offset;
63 } QEMU_PACKED QCowHeader;
64 
65 #define L2_CACHE_SIZE 16
66 
67 typedef struct BDRVQcowState {
68     int cluster_bits;
69     int cluster_size;
70     int cluster_sectors;
71     int l2_bits;
72     int l2_size;
73     unsigned int l1_size;
74     uint64_t cluster_offset_mask;
75     uint64_t l1_table_offset;
76     uint64_t *l1_table;
77     uint64_t *l2_cache;
78     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
79     uint32_t l2_cache_counts[L2_CACHE_SIZE];
80     uint8_t *cluster_cache;
81     uint8_t *cluster_data;
82     uint64_t cluster_cache_offset;
83     QCryptoBlock *crypto; /* Disk encryption format driver */
84     uint32_t crypt_method_header;
85     CoMutex lock;
86     Error *migration_blocker;
87 } BDRVQcowState;
88 
89 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
90 
91 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
92 {
93     const QCowHeader *cow_header = (const void *)buf;
94 
95     if (buf_size >= sizeof(QCowHeader) &&
96         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
97         be32_to_cpu(cow_header->version) == QCOW_VERSION)
98         return 100;
99     else
100         return 0;
101 }
102 
103 static QemuOptsList qcow_runtime_opts = {
104     .name = "qcow",
105     .head = QTAILQ_HEAD_INITIALIZER(qcow_runtime_opts.head),
106     .desc = {
107         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
108         { /* end of list */ }
109     },
110 };
111 
112 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
113                      Error **errp)
114 {
115     BDRVQcowState *s = bs->opaque;
116     unsigned int len, i, shift;
117     int ret;
118     QCowHeader header;
119     Error *local_err = NULL;
120     QCryptoBlockOpenOptions *crypto_opts = NULL;
121     unsigned int cflags = 0;
122     QDict *encryptopts = NULL;
123     const char *encryptfmt;
124 
125     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
126     encryptfmt = qdict_get_try_str(encryptopts, "format");
127 
128     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
129                                false, errp);
130     if (!bs->file) {
131         ret = -EINVAL;
132         goto fail;
133     }
134 
135     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
136     if (ret < 0) {
137         goto fail;
138     }
139     be32_to_cpus(&header.magic);
140     be32_to_cpus(&header.version);
141     be64_to_cpus(&header.backing_file_offset);
142     be32_to_cpus(&header.backing_file_size);
143     be32_to_cpus(&header.mtime);
144     be64_to_cpus(&header.size);
145     be32_to_cpus(&header.crypt_method);
146     be64_to_cpus(&header.l1_table_offset);
147 
148     if (header.magic != QCOW_MAGIC) {
149         error_setg(errp, "Image not in qcow format");
150         ret = -EINVAL;
151         goto fail;
152     }
153     if (header.version != QCOW_VERSION) {
154         error_setg(errp, "Unsupported qcow version %" PRIu32, header.version);
155         ret = -ENOTSUP;
156         goto fail;
157     }
158 
159     if (header.size <= 1) {
160         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
161         ret = -EINVAL;
162         goto fail;
163     }
164     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
165         error_setg(errp, "Cluster size must be between 512 and 64k");
166         ret = -EINVAL;
167         goto fail;
168     }
169 
170     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
171      * so bytes = num_entries << 3. */
172     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
173         error_setg(errp, "L2 table size must be between 512 and 64k");
174         ret = -EINVAL;
175         goto fail;
176     }
177 
178     s->crypt_method_header = header.crypt_method;
179     if (s->crypt_method_header) {
180         if (bdrv_uses_whitelist() &&
181             s->crypt_method_header == QCOW_CRYPT_AES) {
182             error_setg(errp,
183                        "Use of AES-CBC encrypted qcow images is no longer "
184                        "supported in system emulators");
185             error_append_hint(errp,
186                               "You can use 'qemu-img convert' to convert your "
187                               "image to an alternative supported format, such "
188                               "as unencrypted qcow, or raw with the LUKS "
189                               "format instead.\n");
190             ret = -ENOSYS;
191             goto fail;
192         }
193         if (s->crypt_method_header == QCOW_CRYPT_AES) {
194             if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
195                 error_setg(errp,
196                            "Header reported 'aes' encryption format but "
197                            "options specify '%s'", encryptfmt);
198                 ret = -EINVAL;
199                 goto fail;
200             }
201             qdict_del(encryptopts, "format");
202             crypto_opts = block_crypto_open_opts_init(
203                 Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp);
204             if (!crypto_opts) {
205                 ret = -EINVAL;
206                 goto fail;
207             }
208 
209             if (flags & BDRV_O_NO_IO) {
210                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
211             }
212             s->crypto = qcrypto_block_open(crypto_opts, "encrypt.",
213                                            NULL, NULL, cflags, errp);
214             if (!s->crypto) {
215                 ret = -EINVAL;
216                 goto fail;
217             }
218         } else {
219             error_setg(errp, "invalid encryption method in qcow header");
220             ret = -EINVAL;
221             goto fail;
222         }
223         bs->encrypted = true;
224     } else {
225         if (encryptfmt) {
226             error_setg(errp, "No encryption in image header, but options "
227                        "specified format '%s'", encryptfmt);
228             ret = -EINVAL;
229             goto fail;
230         }
231     }
232     s->cluster_bits = header.cluster_bits;
233     s->cluster_size = 1 << s->cluster_bits;
234     s->cluster_sectors = 1 << (s->cluster_bits - 9);
235     s->l2_bits = header.l2_bits;
236     s->l2_size = 1 << s->l2_bits;
237     bs->total_sectors = header.size / 512;
238     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
239 
240     /* read the level 1 table */
241     shift = s->cluster_bits + s->l2_bits;
242     if (header.size > UINT64_MAX - (1LL << shift)) {
243         error_setg(errp, "Image too large");
244         ret = -EINVAL;
245         goto fail;
246     } else {
247         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
248         if (l1_size > INT_MAX / sizeof(uint64_t)) {
249             error_setg(errp, "Image too large");
250             ret = -EINVAL;
251             goto fail;
252         }
253         s->l1_size = l1_size;
254     }
255 
256     s->l1_table_offset = header.l1_table_offset;
257     s->l1_table = g_try_new(uint64_t, s->l1_size);
258     if (s->l1_table == NULL) {
259         error_setg(errp, "Could not allocate memory for L1 table");
260         ret = -ENOMEM;
261         goto fail;
262     }
263 
264     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
265                s->l1_size * sizeof(uint64_t));
266     if (ret < 0) {
267         goto fail;
268     }
269 
270     for(i = 0;i < s->l1_size; i++) {
271         be64_to_cpus(&s->l1_table[i]);
272     }
273 
274     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
275     s->l2_cache =
276         qemu_try_blockalign(bs->file->bs,
277                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
278     if (s->l2_cache == NULL) {
279         error_setg(errp, "Could not allocate L2 table cache");
280         ret = -ENOMEM;
281         goto fail;
282     }
283     s->cluster_cache = g_malloc(s->cluster_size);
284     s->cluster_data = g_malloc(s->cluster_size);
285     s->cluster_cache_offset = -1;
286 
287     /* read the backing file name */
288     if (header.backing_file_offset != 0) {
289         len = header.backing_file_size;
290         if (len > 1023 || len >= sizeof(bs->backing_file)) {
291             error_setg(errp, "Backing file name too long");
292             ret = -EINVAL;
293             goto fail;
294         }
295         ret = bdrv_pread(bs->file, header.backing_file_offset,
296                    bs->backing_file, len);
297         if (ret < 0) {
298             goto fail;
299         }
300         bs->backing_file[len] = '\0';
301     }
302 
303     /* Disable migration when qcow images are used */
304     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
305                "does not support live migration",
306                bdrv_get_device_or_node_name(bs));
307     ret = migrate_add_blocker(s->migration_blocker, &local_err);
308     if (local_err) {
309         error_propagate(errp, local_err);
310         error_free(s->migration_blocker);
311         goto fail;
312     }
313 
314     QDECREF(encryptopts);
315     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
316     qemu_co_mutex_init(&s->lock);
317     return 0;
318 
319  fail:
320     g_free(s->l1_table);
321     qemu_vfree(s->l2_cache);
322     g_free(s->cluster_cache);
323     g_free(s->cluster_data);
324     qcrypto_block_free(s->crypto);
325     QDECREF(encryptopts);
326     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
327     return ret;
328 }
329 
330 
331 /* We have nothing to do for QCOW reopen, stubs just return
332  * success */
333 static int qcow_reopen_prepare(BDRVReopenState *state,
334                                BlockReopenQueue *queue, Error **errp)
335 {
336     return 0;
337 }
338 
339 
340 /* 'allocate' is:
341  *
342  * 0 to not allocate.
343  *
344  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
345  * 'n_end')
346  *
347  * 2 to allocate a compressed cluster of size
348  * 'compressed_size'. 'compressed_size' must be > 0 and <
349  * cluster_size
350  *
351  * return 0 if not allocated, 1 if *result is assigned, and negative
352  * errno on failure.
353  */
354 static int get_cluster_offset(BlockDriverState *bs,
355                               uint64_t offset, int allocate,
356                               int compressed_size,
357                               int n_start, int n_end, uint64_t *result)
358 {
359     BDRVQcowState *s = bs->opaque;
360     int min_index, i, j, l1_index, l2_index, ret;
361     int64_t l2_offset;
362     uint64_t *l2_table, cluster_offset, tmp;
363     uint32_t min_count;
364     int new_l2_table;
365 
366     *result = 0;
367     l1_index = offset >> (s->l2_bits + s->cluster_bits);
368     l2_offset = s->l1_table[l1_index];
369     new_l2_table = 0;
370     if (!l2_offset) {
371         if (!allocate)
372             return 0;
373         /* allocate a new l2 entry */
374         l2_offset = bdrv_getlength(bs->file->bs);
375         if (l2_offset < 0) {
376             return l2_offset;
377         }
378         /* round to cluster size */
379         l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size);
380         /* update the L1 entry */
381         s->l1_table[l1_index] = l2_offset;
382         tmp = cpu_to_be64(l2_offset);
383         BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
384         ret = bdrv_pwrite_sync(bs->file,
385                                s->l1_table_offset + l1_index * sizeof(tmp),
386                                &tmp, sizeof(tmp));
387         if (ret < 0) {
388             return ret;
389         }
390         new_l2_table = 1;
391     }
392     for(i = 0; i < L2_CACHE_SIZE; i++) {
393         if (l2_offset == s->l2_cache_offsets[i]) {
394             /* increment the hit count */
395             if (++s->l2_cache_counts[i] == 0xffffffff) {
396                 for(j = 0; j < L2_CACHE_SIZE; j++) {
397                     s->l2_cache_counts[j] >>= 1;
398                 }
399             }
400             l2_table = s->l2_cache + (i << s->l2_bits);
401             goto found;
402         }
403     }
404     /* not found: load a new entry in the least used one */
405     min_index = 0;
406     min_count = 0xffffffff;
407     for(i = 0; i < L2_CACHE_SIZE; i++) {
408         if (s->l2_cache_counts[i] < min_count) {
409             min_count = s->l2_cache_counts[i];
410             min_index = i;
411         }
412     }
413     l2_table = s->l2_cache + (min_index << s->l2_bits);
414     BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
415     if (new_l2_table) {
416         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
417         ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
418                                s->l2_size * sizeof(uint64_t));
419         if (ret < 0) {
420             return ret;
421         }
422     } else {
423         ret = bdrv_pread(bs->file, l2_offset, l2_table,
424                          s->l2_size * sizeof(uint64_t));
425         if (ret < 0) {
426             return ret;
427         }
428     }
429     s->l2_cache_offsets[min_index] = l2_offset;
430     s->l2_cache_counts[min_index] = 1;
431  found:
432     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
433     cluster_offset = be64_to_cpu(l2_table[l2_index]);
434     if (!cluster_offset ||
435         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
436         if (!allocate)
437             return 0;
438         BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
439         /* allocate a new cluster */
440         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
441             (n_end - n_start) < s->cluster_sectors) {
442             /* if the cluster is already compressed, we must
443                decompress it in the case it is not completely
444                overwritten */
445             if (decompress_cluster(bs, cluster_offset) < 0) {
446                 return -EIO;
447             }
448             cluster_offset = bdrv_getlength(bs->file->bs);
449             if ((int64_t) cluster_offset < 0) {
450                 return cluster_offset;
451             }
452             cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
453             /* write the cluster content */
454             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
455             ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
456                               s->cluster_size);
457             if (ret < 0) {
458                 return ret;
459             }
460         } else {
461             cluster_offset = bdrv_getlength(bs->file->bs);
462             if ((int64_t) cluster_offset < 0) {
463                 return cluster_offset;
464             }
465             if (allocate == 1) {
466                 /* round to cluster size */
467                 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
468                 if (cluster_offset + s->cluster_size > INT64_MAX) {
469                     return -E2BIG;
470                 }
471                 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size,
472                                     PREALLOC_MODE_OFF, NULL);
473                 if (ret < 0) {
474                     return ret;
475                 }
476                 /* if encrypted, we must initialize the cluster
477                    content which won't be written */
478                 if (bs->encrypted &&
479                     (n_end - n_start) < s->cluster_sectors) {
480                     uint64_t start_sect;
481                     assert(s->crypto);
482                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
483                     for(i = 0; i < s->cluster_sectors; i++) {
484                         if (i < n_start || i >= n_end) {
485                             memset(s->cluster_data, 0x00, 512);
486                             if (qcrypto_block_encrypt(s->crypto,
487                                                       (start_sect + i) *
488                                                       BDRV_SECTOR_SIZE,
489                                                       s->cluster_data,
490                                                       BDRV_SECTOR_SIZE,
491                                                       NULL) < 0) {
492                                 return -EIO;
493                             }
494                             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
495                             ret = bdrv_pwrite(bs->file,
496                                               cluster_offset + i * 512,
497                                               s->cluster_data, 512);
498                             if (ret < 0) {
499                                 return ret;
500                             }
501                         }
502                     }
503                 }
504             } else if (allocate == 2) {
505                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
506                     (uint64_t)compressed_size << (63 - s->cluster_bits);
507             }
508         }
509         /* update L2 table */
510         tmp = cpu_to_be64(cluster_offset);
511         l2_table[l2_index] = tmp;
512         if (allocate == 2) {
513             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
514         } else {
515             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
516         }
517         ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
518                                &tmp, sizeof(tmp));
519         if (ret < 0) {
520             return ret;
521         }
522     }
523     *result = cluster_offset;
524     return 1;
525 }
526 
527 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
528         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
529 {
530     BDRVQcowState *s = bs->opaque;
531     int index_in_cluster, n, ret;
532     uint64_t cluster_offset;
533 
534     qemu_co_mutex_lock(&s->lock);
535     ret = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0, &cluster_offset);
536     qemu_co_mutex_unlock(&s->lock);
537     if (ret < 0) {
538         return ret;
539     }
540     index_in_cluster = sector_num & (s->cluster_sectors - 1);
541     n = s->cluster_sectors - index_in_cluster;
542     if (n > nb_sectors)
543         n = nb_sectors;
544     *pnum = n;
545     if (!cluster_offset) {
546         return 0;
547     }
548     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) {
549         return BDRV_BLOCK_DATA;
550     }
551     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
552     *file = bs->file->bs;
553     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
554 }
555 
556 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
557                              const uint8_t *buf, int buf_size)
558 {
559     z_stream strm1, *strm = &strm1;
560     int ret, out_len;
561 
562     memset(strm, 0, sizeof(*strm));
563 
564     strm->next_in = (uint8_t *)buf;
565     strm->avail_in = buf_size;
566     strm->next_out = out_buf;
567     strm->avail_out = out_buf_size;
568 
569     ret = inflateInit2(strm, -12);
570     if (ret != Z_OK)
571         return -1;
572     ret = inflate(strm, Z_FINISH);
573     out_len = strm->next_out - out_buf;
574     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
575         out_len != out_buf_size) {
576         inflateEnd(strm);
577         return -1;
578     }
579     inflateEnd(strm);
580     return 0;
581 }
582 
583 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
584 {
585     BDRVQcowState *s = bs->opaque;
586     int ret, csize;
587     uint64_t coffset;
588 
589     coffset = cluster_offset & s->cluster_offset_mask;
590     if (s->cluster_cache_offset != coffset) {
591         csize = cluster_offset >> (63 - s->cluster_bits);
592         csize &= (s->cluster_size - 1);
593         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
594         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
595         if (ret != csize)
596             return -1;
597         if (decompress_buffer(s->cluster_cache, s->cluster_size,
598                               s->cluster_data, csize) < 0) {
599             return -1;
600         }
601         s->cluster_cache_offset = coffset;
602     }
603     return 0;
604 }
605 
606 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
607                          int nb_sectors, QEMUIOVector *qiov)
608 {
609     BDRVQcowState *s = bs->opaque;
610     int index_in_cluster;
611     int ret = 0, n;
612     uint64_t cluster_offset;
613     struct iovec hd_iov;
614     QEMUIOVector hd_qiov;
615     uint8_t *buf;
616     void *orig_buf;
617 
618     if (qiov->niov > 1) {
619         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
620         if (buf == NULL) {
621             return -ENOMEM;
622         }
623     } else {
624         orig_buf = NULL;
625         buf = (uint8_t *)qiov->iov->iov_base;
626     }
627 
628     qemu_co_mutex_lock(&s->lock);
629 
630     while (nb_sectors != 0) {
631         /* prepare next request */
632         ret = get_cluster_offset(bs, sector_num << 9,
633                                  0, 0, 0, 0, &cluster_offset);
634         if (ret < 0) {
635             break;
636         }
637         index_in_cluster = sector_num & (s->cluster_sectors - 1);
638         n = s->cluster_sectors - index_in_cluster;
639         if (n > nb_sectors) {
640             n = nb_sectors;
641         }
642 
643         if (!cluster_offset) {
644             if (bs->backing) {
645                 /* read from the base image */
646                 hd_iov.iov_base = (void *)buf;
647                 hd_iov.iov_len = n * 512;
648                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
649                 qemu_co_mutex_unlock(&s->lock);
650                 /* qcow2 emits this on bs->file instead of bs->backing */
651                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
652                 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov);
653                 qemu_co_mutex_lock(&s->lock);
654                 if (ret < 0) {
655                     break;
656                 }
657             } else {
658                 /* Note: in this case, no need to wait */
659                 memset(buf, 0, 512 * n);
660             }
661         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
662             /* add AIO support for compressed blocks ? */
663             if (decompress_cluster(bs, cluster_offset) < 0) {
664                 ret = -EIO;
665                 break;
666             }
667             memcpy(buf,
668                    s->cluster_cache + index_in_cluster * 512, 512 * n);
669         } else {
670             if ((cluster_offset & 511) != 0) {
671                 ret = -EIO;
672                 break;
673             }
674             hd_iov.iov_base = (void *)buf;
675             hd_iov.iov_len = n * 512;
676             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
677             qemu_co_mutex_unlock(&s->lock);
678             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
679             ret = bdrv_co_readv(bs->file,
680                                 (cluster_offset >> 9) + index_in_cluster,
681                                 n, &hd_qiov);
682             qemu_co_mutex_lock(&s->lock);
683             if (ret < 0) {
684                 break;
685             }
686             if (bs->encrypted) {
687                 assert(s->crypto);
688                 if (qcrypto_block_decrypt(s->crypto,
689                                           sector_num * BDRV_SECTOR_SIZE, buf,
690                                           n * BDRV_SECTOR_SIZE, NULL) < 0) {
691                     ret = -EIO;
692                     break;
693                 }
694             }
695         }
696         ret = 0;
697 
698         nb_sectors -= n;
699         sector_num += n;
700         buf += n * 512;
701     }
702 
703     qemu_co_mutex_unlock(&s->lock);
704 
705     if (qiov->niov > 1) {
706         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
707         qemu_vfree(orig_buf);
708     }
709 
710     return ret;
711 }
712 
713 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
714                           int nb_sectors, QEMUIOVector *qiov)
715 {
716     BDRVQcowState *s = bs->opaque;
717     int index_in_cluster;
718     uint64_t cluster_offset;
719     int ret = 0, n;
720     struct iovec hd_iov;
721     QEMUIOVector hd_qiov;
722     uint8_t *buf;
723     void *orig_buf;
724 
725     s->cluster_cache_offset = -1; /* disable compressed cache */
726 
727     /* We must always copy the iov when encrypting, so we
728      * don't modify the original data buffer during encryption */
729     if (bs->encrypted || qiov->niov > 1) {
730         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
731         if (buf == NULL) {
732             return -ENOMEM;
733         }
734         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
735     } else {
736         orig_buf = NULL;
737         buf = (uint8_t *)qiov->iov->iov_base;
738     }
739 
740     qemu_co_mutex_lock(&s->lock);
741 
742     while (nb_sectors != 0) {
743 
744         index_in_cluster = sector_num & (s->cluster_sectors - 1);
745         n = s->cluster_sectors - index_in_cluster;
746         if (n > nb_sectors) {
747             n = nb_sectors;
748         }
749         ret = get_cluster_offset(bs, sector_num << 9, 1, 0,
750                                  index_in_cluster,
751                                  index_in_cluster + n, &cluster_offset);
752         if (ret < 0) {
753             break;
754         }
755         if (!cluster_offset || (cluster_offset & 511) != 0) {
756             ret = -EIO;
757             break;
758         }
759         if (bs->encrypted) {
760             assert(s->crypto);
761             if (qcrypto_block_encrypt(s->crypto, sector_num * BDRV_SECTOR_SIZE,
762                                       buf, n * BDRV_SECTOR_SIZE, NULL) < 0) {
763                 ret = -EIO;
764                 break;
765             }
766         }
767 
768         hd_iov.iov_base = (void *)buf;
769         hd_iov.iov_len = n * 512;
770         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
771         qemu_co_mutex_unlock(&s->lock);
772         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
773         ret = bdrv_co_writev(bs->file,
774                              (cluster_offset >> 9) + index_in_cluster,
775                              n, &hd_qiov);
776         qemu_co_mutex_lock(&s->lock);
777         if (ret < 0) {
778             break;
779         }
780         ret = 0;
781 
782         nb_sectors -= n;
783         sector_num += n;
784         buf += n * 512;
785     }
786     qemu_co_mutex_unlock(&s->lock);
787 
788     qemu_vfree(orig_buf);
789 
790     return ret;
791 }
792 
793 static void qcow_close(BlockDriverState *bs)
794 {
795     BDRVQcowState *s = bs->opaque;
796 
797     qcrypto_block_free(s->crypto);
798     s->crypto = NULL;
799     g_free(s->l1_table);
800     qemu_vfree(s->l2_cache);
801     g_free(s->cluster_cache);
802     g_free(s->cluster_data);
803 
804     migrate_del_blocker(s->migration_blocker);
805     error_free(s->migration_blocker);
806 }
807 
808 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
809 {
810     int header_size, backing_filename_len, l1_size, shift, i;
811     QCowHeader header;
812     uint8_t *tmp;
813     int64_t total_size = 0;
814     char *backing_file = NULL;
815     Error *local_err = NULL;
816     int ret;
817     BlockBackend *qcow_blk;
818     char *encryptfmt = NULL;
819     QDict *options;
820     QDict *encryptopts = NULL;
821     QCryptoBlockCreateOptions *crypto_opts = NULL;
822     QCryptoBlock *crypto = NULL;
823 
824     /* Read out options */
825     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
826                           BDRV_SECTOR_SIZE);
827     if (total_size == 0) {
828         error_setg(errp, "Image size is too small, cannot be zero length");
829         ret = -EINVAL;
830         goto cleanup;
831     }
832 
833     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
834     encryptfmt = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
835     if (encryptfmt) {
836         if (qemu_opt_get(opts, BLOCK_OPT_ENCRYPT)) {
837             error_setg(errp, "Options " BLOCK_OPT_ENCRYPT " and "
838                        BLOCK_OPT_ENCRYPT_FORMAT " are mutually exclusive");
839             ret = -EINVAL;
840             goto cleanup;
841         }
842     } else if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
843         encryptfmt = g_strdup("aes");
844     }
845 
846     ret = bdrv_create_file(filename, opts, &local_err);
847     if (ret < 0) {
848         error_propagate(errp, local_err);
849         goto cleanup;
850     }
851 
852     qcow_blk = blk_new_open(filename, NULL, NULL,
853                             BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
854                             &local_err);
855     if (qcow_blk == NULL) {
856         error_propagate(errp, local_err);
857         ret = -EIO;
858         goto cleanup;
859     }
860 
861     blk_set_allow_write_beyond_eof(qcow_blk, true);
862 
863     ret = blk_truncate(qcow_blk, 0, PREALLOC_MODE_OFF, errp);
864     if (ret < 0) {
865         goto exit;
866     }
867 
868     memset(&header, 0, sizeof(header));
869     header.magic = cpu_to_be32(QCOW_MAGIC);
870     header.version = cpu_to_be32(QCOW_VERSION);
871     header.size = cpu_to_be64(total_size);
872     header_size = sizeof(header);
873     backing_filename_len = 0;
874     if (backing_file) {
875         if (strcmp(backing_file, "fat:")) {
876             header.backing_file_offset = cpu_to_be64(header_size);
877             backing_filename_len = strlen(backing_file);
878             header.backing_file_size = cpu_to_be32(backing_filename_len);
879             header_size += backing_filename_len;
880         } else {
881             /* special backing file for vvfat */
882             g_free(backing_file);
883             backing_file = NULL;
884         }
885         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
886                                     unmodified sectors */
887         header.l2_bits = 12; /* 32 KB L2 tables */
888     } else {
889         header.cluster_bits = 12; /* 4 KB clusters */
890         header.l2_bits = 9; /* 4 KB L2 tables */
891     }
892     header_size = (header_size + 7) & ~7;
893     shift = header.cluster_bits + header.l2_bits;
894     l1_size = (total_size + (1LL << shift) - 1) >> shift;
895 
896     header.l1_table_offset = cpu_to_be64(header_size);
897 
898     options = qemu_opts_to_qdict(opts, NULL);
899     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
900     QDECREF(options);
901     if (encryptfmt) {
902         if (!g_str_equal(encryptfmt, "aes")) {
903             error_setg(errp, "Unknown encryption format '%s', expected 'aes'",
904                        encryptfmt);
905             ret = -EINVAL;
906             goto exit;
907         }
908         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
909 
910         crypto_opts = block_crypto_create_opts_init(
911             Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp);
912         if (!crypto_opts) {
913             ret = -EINVAL;
914             goto exit;
915         }
916 
917         crypto = qcrypto_block_create(crypto_opts, "encrypt.",
918                                       NULL, NULL, NULL, errp);
919         if (!crypto) {
920             ret = -EINVAL;
921             goto exit;
922         }
923     } else {
924         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
925     }
926 
927     /* write all the data */
928     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
929     if (ret != sizeof(header)) {
930         goto exit;
931     }
932 
933     if (backing_file) {
934         ret = blk_pwrite(qcow_blk, sizeof(header),
935                          backing_file, backing_filename_len, 0);
936         if (ret != backing_filename_len) {
937             goto exit;
938         }
939     }
940 
941     tmp = g_malloc0(BDRV_SECTOR_SIZE);
942     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
943          i++) {
944         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
945                          tmp, BDRV_SECTOR_SIZE, 0);
946         if (ret != BDRV_SECTOR_SIZE) {
947             g_free(tmp);
948             goto exit;
949         }
950     }
951 
952     g_free(tmp);
953     ret = 0;
954 exit:
955     blk_unref(qcow_blk);
956 cleanup:
957     QDECREF(encryptopts);
958     g_free(encryptfmt);
959     qcrypto_block_free(crypto);
960     qapi_free_QCryptoBlockCreateOptions(crypto_opts);
961     g_free(backing_file);
962     return ret;
963 }
964 
965 static int qcow_make_empty(BlockDriverState *bs)
966 {
967     BDRVQcowState *s = bs->opaque;
968     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
969     int ret;
970 
971     memset(s->l1_table, 0, l1_length);
972     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
973             l1_length) < 0)
974         return -1;
975     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length,
976                         PREALLOC_MODE_OFF, NULL);
977     if (ret < 0)
978         return ret;
979 
980     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
981     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
982     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
983 
984     return 0;
985 }
986 
987 /* XXX: put compressed sectors first, then all the cluster aligned
988    tables to avoid losing bytes in alignment */
989 static coroutine_fn int
990 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
991                            uint64_t bytes, QEMUIOVector *qiov)
992 {
993     BDRVQcowState *s = bs->opaque;
994     QEMUIOVector hd_qiov;
995     struct iovec iov;
996     z_stream strm;
997     int ret, out_len;
998     uint8_t *buf, *out_buf;
999     uint64_t cluster_offset;
1000 
1001     buf = qemu_blockalign(bs, s->cluster_size);
1002     if (bytes != s->cluster_size) {
1003         if (bytes > s->cluster_size ||
1004             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
1005         {
1006             qemu_vfree(buf);
1007             return -EINVAL;
1008         }
1009         /* Zero-pad last write if image size is not cluster aligned */
1010         memset(buf + bytes, 0, s->cluster_size - bytes);
1011     }
1012     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
1013 
1014     out_buf = g_malloc(s->cluster_size);
1015 
1016     /* best compression, small window, no zlib header */
1017     memset(&strm, 0, sizeof(strm));
1018     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1019                        Z_DEFLATED, -12,
1020                        9, Z_DEFAULT_STRATEGY);
1021     if (ret != 0) {
1022         ret = -EINVAL;
1023         goto fail;
1024     }
1025 
1026     strm.avail_in = s->cluster_size;
1027     strm.next_in = (uint8_t *)buf;
1028     strm.avail_out = s->cluster_size;
1029     strm.next_out = out_buf;
1030 
1031     ret = deflate(&strm, Z_FINISH);
1032     if (ret != Z_STREAM_END && ret != Z_OK) {
1033         deflateEnd(&strm);
1034         ret = -EINVAL;
1035         goto fail;
1036     }
1037     out_len = strm.next_out - out_buf;
1038 
1039     deflateEnd(&strm);
1040 
1041     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1042         /* could not compress: write normal cluster */
1043         ret = qcow_co_writev(bs, offset >> BDRV_SECTOR_BITS,
1044                              bytes >> BDRV_SECTOR_BITS, qiov);
1045         if (ret < 0) {
1046             goto fail;
1047         }
1048         goto success;
1049     }
1050     qemu_co_mutex_lock(&s->lock);
1051     ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1052     qemu_co_mutex_unlock(&s->lock);
1053     if (ret < 0) {
1054         goto fail;
1055     }
1056     if (cluster_offset == 0) {
1057         ret = -EIO;
1058         goto fail;
1059     }
1060     cluster_offset &= s->cluster_offset_mask;
1061 
1062     iov = (struct iovec) {
1063         .iov_base   = out_buf,
1064         .iov_len    = out_len,
1065     };
1066     qemu_iovec_init_external(&hd_qiov, &iov, 1);
1067     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1068     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
1069     if (ret < 0) {
1070         goto fail;
1071     }
1072 success:
1073     ret = 0;
1074 fail:
1075     qemu_vfree(buf);
1076     g_free(out_buf);
1077     return ret;
1078 }
1079 
1080 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1081 {
1082     BDRVQcowState *s = bs->opaque;
1083     bdi->cluster_size = s->cluster_size;
1084     return 0;
1085 }
1086 
1087 static QemuOptsList qcow_create_opts = {
1088     .name = "qcow-create-opts",
1089     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1090     .desc = {
1091         {
1092             .name = BLOCK_OPT_SIZE,
1093             .type = QEMU_OPT_SIZE,
1094             .help = "Virtual disk size"
1095         },
1096         {
1097             .name = BLOCK_OPT_BACKING_FILE,
1098             .type = QEMU_OPT_STRING,
1099             .help = "File name of a base image"
1100         },
1101         {
1102             .name = BLOCK_OPT_ENCRYPT,
1103             .type = QEMU_OPT_BOOL,
1104             .help = "Encrypt the image with format 'aes'. (Deprecated "
1105                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1106         },
1107         {
1108             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1109             .type = QEMU_OPT_STRING,
1110             .help = "Encrypt the image, format choices: 'aes'",
1111         },
1112         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1113         { /* end of list */ }
1114     }
1115 };
1116 
1117 static BlockDriver bdrv_qcow = {
1118     .format_name	= "qcow",
1119     .instance_size	= sizeof(BDRVQcowState),
1120     .bdrv_probe		= qcow_probe,
1121     .bdrv_open		= qcow_open,
1122     .bdrv_close		= qcow_close,
1123     .bdrv_child_perm        = bdrv_format_default_perms,
1124     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1125     .bdrv_create            = qcow_create,
1126     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1127     .supports_backing       = true,
1128 
1129     .bdrv_co_readv          = qcow_co_readv,
1130     .bdrv_co_writev         = qcow_co_writev,
1131     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1132 
1133     .bdrv_make_empty        = qcow_make_empty,
1134     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1135     .bdrv_get_info          = qcow_get_info,
1136 
1137     .create_opts            = &qcow_create_opts,
1138 };
1139 
1140 static void bdrv_qcow_init(void)
1141 {
1142     bdrv_register(&bdrv_qcow);
1143 }
1144 
1145 block_init(bdrv_qcow_init);
1146