xref: /openbmc/qemu/block/qcow.c (revision c8ca2a23)
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 int coroutine_fn qcow_co_block_status(BlockDriverState *bs,
528                                              bool want_zero,
529                                              int64_t offset, int64_t bytes,
530                                              int64_t *pnum, int64_t *map,
531                                              BlockDriverState **file)
532 {
533     BDRVQcowState *s = bs->opaque;
534     int index_in_cluster, ret;
535     int64_t n;
536     uint64_t cluster_offset;
537 
538     qemu_co_mutex_lock(&s->lock);
539     ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
540     qemu_co_mutex_unlock(&s->lock);
541     if (ret < 0) {
542         return ret;
543     }
544     index_in_cluster = offset & (s->cluster_size - 1);
545     n = s->cluster_size - index_in_cluster;
546     if (n > bytes) {
547         n = bytes;
548     }
549     *pnum = n;
550     if (!cluster_offset) {
551         return 0;
552     }
553     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) {
554         return BDRV_BLOCK_DATA;
555     }
556     *map = cluster_offset | index_in_cluster;
557     *file = bs->file->bs;
558     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
559 }
560 
561 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
562                              const uint8_t *buf, int buf_size)
563 {
564     z_stream strm1, *strm = &strm1;
565     int ret, out_len;
566 
567     memset(strm, 0, sizeof(*strm));
568 
569     strm->next_in = (uint8_t *)buf;
570     strm->avail_in = buf_size;
571     strm->next_out = out_buf;
572     strm->avail_out = out_buf_size;
573 
574     ret = inflateInit2(strm, -12);
575     if (ret != Z_OK)
576         return -1;
577     ret = inflate(strm, Z_FINISH);
578     out_len = strm->next_out - out_buf;
579     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
580         out_len != out_buf_size) {
581         inflateEnd(strm);
582         return -1;
583     }
584     inflateEnd(strm);
585     return 0;
586 }
587 
588 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
589 {
590     BDRVQcowState *s = bs->opaque;
591     int ret, csize;
592     uint64_t coffset;
593 
594     coffset = cluster_offset & s->cluster_offset_mask;
595     if (s->cluster_cache_offset != coffset) {
596         csize = cluster_offset >> (63 - s->cluster_bits);
597         csize &= (s->cluster_size - 1);
598         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
599         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
600         if (ret != csize)
601             return -1;
602         if (decompress_buffer(s->cluster_cache, s->cluster_size,
603                               s->cluster_data, csize) < 0) {
604             return -1;
605         }
606         s->cluster_cache_offset = coffset;
607     }
608     return 0;
609 }
610 
611 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
612                          int nb_sectors, QEMUIOVector *qiov)
613 {
614     BDRVQcowState *s = bs->opaque;
615     int index_in_cluster;
616     int ret = 0, n;
617     uint64_t cluster_offset;
618     struct iovec hd_iov;
619     QEMUIOVector hd_qiov;
620     uint8_t *buf;
621     void *orig_buf;
622 
623     if (qiov->niov > 1) {
624         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
625         if (buf == NULL) {
626             return -ENOMEM;
627         }
628     } else {
629         orig_buf = NULL;
630         buf = (uint8_t *)qiov->iov->iov_base;
631     }
632 
633     qemu_co_mutex_lock(&s->lock);
634 
635     while (nb_sectors != 0) {
636         /* prepare next request */
637         ret = get_cluster_offset(bs, sector_num << 9,
638                                  0, 0, 0, 0, &cluster_offset);
639         if (ret < 0) {
640             break;
641         }
642         index_in_cluster = sector_num & (s->cluster_sectors - 1);
643         n = s->cluster_sectors - index_in_cluster;
644         if (n > nb_sectors) {
645             n = nb_sectors;
646         }
647 
648         if (!cluster_offset) {
649             if (bs->backing) {
650                 /* read from the base image */
651                 hd_iov.iov_base = (void *)buf;
652                 hd_iov.iov_len = n * 512;
653                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
654                 qemu_co_mutex_unlock(&s->lock);
655                 /* qcow2 emits this on bs->file instead of bs->backing */
656                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
657                 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov);
658                 qemu_co_mutex_lock(&s->lock);
659                 if (ret < 0) {
660                     break;
661                 }
662             } else {
663                 /* Note: in this case, no need to wait */
664                 memset(buf, 0, 512 * n);
665             }
666         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
667             /* add AIO support for compressed blocks ? */
668             if (decompress_cluster(bs, cluster_offset) < 0) {
669                 ret = -EIO;
670                 break;
671             }
672             memcpy(buf,
673                    s->cluster_cache + index_in_cluster * 512, 512 * n);
674         } else {
675             if ((cluster_offset & 511) != 0) {
676                 ret = -EIO;
677                 break;
678             }
679             hd_iov.iov_base = (void *)buf;
680             hd_iov.iov_len = n * 512;
681             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
682             qemu_co_mutex_unlock(&s->lock);
683             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
684             ret = bdrv_co_readv(bs->file,
685                                 (cluster_offset >> 9) + index_in_cluster,
686                                 n, &hd_qiov);
687             qemu_co_mutex_lock(&s->lock);
688             if (ret < 0) {
689                 break;
690             }
691             if (bs->encrypted) {
692                 assert(s->crypto);
693                 if (qcrypto_block_decrypt(s->crypto,
694                                           sector_num * BDRV_SECTOR_SIZE, buf,
695                                           n * BDRV_SECTOR_SIZE, NULL) < 0) {
696                     ret = -EIO;
697                     break;
698                 }
699             }
700         }
701         ret = 0;
702 
703         nb_sectors -= n;
704         sector_num += n;
705         buf += n * 512;
706     }
707 
708     qemu_co_mutex_unlock(&s->lock);
709 
710     if (qiov->niov > 1) {
711         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
712         qemu_vfree(orig_buf);
713     }
714 
715     return ret;
716 }
717 
718 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
719                           int nb_sectors, QEMUIOVector *qiov)
720 {
721     BDRVQcowState *s = bs->opaque;
722     int index_in_cluster;
723     uint64_t cluster_offset;
724     int ret = 0, n;
725     struct iovec hd_iov;
726     QEMUIOVector hd_qiov;
727     uint8_t *buf;
728     void *orig_buf;
729 
730     s->cluster_cache_offset = -1; /* disable compressed cache */
731 
732     /* We must always copy the iov when encrypting, so we
733      * don't modify the original data buffer during encryption */
734     if (bs->encrypted || qiov->niov > 1) {
735         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
736         if (buf == NULL) {
737             return -ENOMEM;
738         }
739         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
740     } else {
741         orig_buf = NULL;
742         buf = (uint8_t *)qiov->iov->iov_base;
743     }
744 
745     qemu_co_mutex_lock(&s->lock);
746 
747     while (nb_sectors != 0) {
748 
749         index_in_cluster = sector_num & (s->cluster_sectors - 1);
750         n = s->cluster_sectors - index_in_cluster;
751         if (n > nb_sectors) {
752             n = nb_sectors;
753         }
754         ret = get_cluster_offset(bs, sector_num << 9, 1, 0,
755                                  index_in_cluster,
756                                  index_in_cluster + n, &cluster_offset);
757         if (ret < 0) {
758             break;
759         }
760         if (!cluster_offset || (cluster_offset & 511) != 0) {
761             ret = -EIO;
762             break;
763         }
764         if (bs->encrypted) {
765             assert(s->crypto);
766             if (qcrypto_block_encrypt(s->crypto, sector_num * BDRV_SECTOR_SIZE,
767                                       buf, n * BDRV_SECTOR_SIZE, NULL) < 0) {
768                 ret = -EIO;
769                 break;
770             }
771         }
772 
773         hd_iov.iov_base = (void *)buf;
774         hd_iov.iov_len = n * 512;
775         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
776         qemu_co_mutex_unlock(&s->lock);
777         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
778         ret = bdrv_co_writev(bs->file,
779                              (cluster_offset >> 9) + index_in_cluster,
780                              n, &hd_qiov);
781         qemu_co_mutex_lock(&s->lock);
782         if (ret < 0) {
783             break;
784         }
785         ret = 0;
786 
787         nb_sectors -= n;
788         sector_num += n;
789         buf += n * 512;
790     }
791     qemu_co_mutex_unlock(&s->lock);
792 
793     qemu_vfree(orig_buf);
794 
795     return ret;
796 }
797 
798 static void qcow_close(BlockDriverState *bs)
799 {
800     BDRVQcowState *s = bs->opaque;
801 
802     qcrypto_block_free(s->crypto);
803     s->crypto = NULL;
804     g_free(s->l1_table);
805     qemu_vfree(s->l2_cache);
806     g_free(s->cluster_cache);
807     g_free(s->cluster_data);
808 
809     migrate_del_blocker(s->migration_blocker);
810     error_free(s->migration_blocker);
811 }
812 
813 static int coroutine_fn qcow_co_create_opts(const char *filename, QemuOpts *opts,
814                                             Error **errp)
815 {
816     int header_size, backing_filename_len, l1_size, shift, i;
817     QCowHeader header;
818     uint8_t *tmp;
819     int64_t total_size = 0;
820     char *backing_file = NULL;
821     Error *local_err = NULL;
822     int ret;
823     BlockBackend *qcow_blk;
824     char *encryptfmt = NULL;
825     QDict *options;
826     QDict *encryptopts = NULL;
827     QCryptoBlockCreateOptions *crypto_opts = NULL;
828     QCryptoBlock *crypto = NULL;
829 
830     /* Read out options */
831     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
832                           BDRV_SECTOR_SIZE);
833     if (total_size == 0) {
834         error_setg(errp, "Image size is too small, cannot be zero length");
835         ret = -EINVAL;
836         goto cleanup;
837     }
838 
839     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
840     encryptfmt = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
841     if (encryptfmt) {
842         if (qemu_opt_get(opts, BLOCK_OPT_ENCRYPT)) {
843             error_setg(errp, "Options " BLOCK_OPT_ENCRYPT " and "
844                        BLOCK_OPT_ENCRYPT_FORMAT " are mutually exclusive");
845             ret = -EINVAL;
846             goto cleanup;
847         }
848     } else if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
849         encryptfmt = g_strdup("aes");
850     }
851 
852     ret = bdrv_create_file(filename, opts, &local_err);
853     if (ret < 0) {
854         error_propagate(errp, local_err);
855         goto cleanup;
856     }
857 
858     qcow_blk = blk_new_open(filename, NULL, NULL,
859                             BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
860                             &local_err);
861     if (qcow_blk == NULL) {
862         error_propagate(errp, local_err);
863         ret = -EIO;
864         goto cleanup;
865     }
866 
867     blk_set_allow_write_beyond_eof(qcow_blk, true);
868 
869     ret = blk_truncate(qcow_blk, 0, PREALLOC_MODE_OFF, errp);
870     if (ret < 0) {
871         goto exit;
872     }
873 
874     memset(&header, 0, sizeof(header));
875     header.magic = cpu_to_be32(QCOW_MAGIC);
876     header.version = cpu_to_be32(QCOW_VERSION);
877     header.size = cpu_to_be64(total_size);
878     header_size = sizeof(header);
879     backing_filename_len = 0;
880     if (backing_file) {
881         if (strcmp(backing_file, "fat:")) {
882             header.backing_file_offset = cpu_to_be64(header_size);
883             backing_filename_len = strlen(backing_file);
884             header.backing_file_size = cpu_to_be32(backing_filename_len);
885             header_size += backing_filename_len;
886         } else {
887             /* special backing file for vvfat */
888             g_free(backing_file);
889             backing_file = NULL;
890         }
891         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
892                                     unmodified sectors */
893         header.l2_bits = 12; /* 32 KB L2 tables */
894     } else {
895         header.cluster_bits = 12; /* 4 KB clusters */
896         header.l2_bits = 9; /* 4 KB L2 tables */
897     }
898     header_size = (header_size + 7) & ~7;
899     shift = header.cluster_bits + header.l2_bits;
900     l1_size = (total_size + (1LL << shift) - 1) >> shift;
901 
902     header.l1_table_offset = cpu_to_be64(header_size);
903 
904     options = qemu_opts_to_qdict(opts, NULL);
905     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
906     QDECREF(options);
907     if (encryptfmt) {
908         if (!g_str_equal(encryptfmt, "aes")) {
909             error_setg(errp, "Unknown encryption format '%s', expected 'aes'",
910                        encryptfmt);
911             ret = -EINVAL;
912             goto exit;
913         }
914         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
915 
916         crypto_opts = block_crypto_create_opts_init(
917             Q_CRYPTO_BLOCK_FORMAT_QCOW, encryptopts, errp);
918         if (!crypto_opts) {
919             ret = -EINVAL;
920             goto exit;
921         }
922 
923         crypto = qcrypto_block_create(crypto_opts, "encrypt.",
924                                       NULL, NULL, NULL, errp);
925         if (!crypto) {
926             ret = -EINVAL;
927             goto exit;
928         }
929     } else {
930         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
931     }
932 
933     /* write all the data */
934     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
935     if (ret != sizeof(header)) {
936         goto exit;
937     }
938 
939     if (backing_file) {
940         ret = blk_pwrite(qcow_blk, sizeof(header),
941                          backing_file, backing_filename_len, 0);
942         if (ret != backing_filename_len) {
943             goto exit;
944         }
945     }
946 
947     tmp = g_malloc0(BDRV_SECTOR_SIZE);
948     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
949          i++) {
950         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
951                          tmp, BDRV_SECTOR_SIZE, 0);
952         if (ret != BDRV_SECTOR_SIZE) {
953             g_free(tmp);
954             goto exit;
955         }
956     }
957 
958     g_free(tmp);
959     ret = 0;
960 exit:
961     blk_unref(qcow_blk);
962 cleanup:
963     QDECREF(encryptopts);
964     g_free(encryptfmt);
965     qcrypto_block_free(crypto);
966     qapi_free_QCryptoBlockCreateOptions(crypto_opts);
967     g_free(backing_file);
968     return ret;
969 }
970 
971 static int qcow_make_empty(BlockDriverState *bs)
972 {
973     BDRVQcowState *s = bs->opaque;
974     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
975     int ret;
976 
977     memset(s->l1_table, 0, l1_length);
978     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
979             l1_length) < 0)
980         return -1;
981     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length,
982                         PREALLOC_MODE_OFF, NULL);
983     if (ret < 0)
984         return ret;
985 
986     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
987     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
988     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
989 
990     return 0;
991 }
992 
993 /* XXX: put compressed sectors first, then all the cluster aligned
994    tables to avoid losing bytes in alignment */
995 static coroutine_fn int
996 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
997                            uint64_t bytes, QEMUIOVector *qiov)
998 {
999     BDRVQcowState *s = bs->opaque;
1000     QEMUIOVector hd_qiov;
1001     struct iovec iov;
1002     z_stream strm;
1003     int ret, out_len;
1004     uint8_t *buf, *out_buf;
1005     uint64_t cluster_offset;
1006 
1007     buf = qemu_blockalign(bs, s->cluster_size);
1008     if (bytes != s->cluster_size) {
1009         if (bytes > s->cluster_size ||
1010             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
1011         {
1012             qemu_vfree(buf);
1013             return -EINVAL;
1014         }
1015         /* Zero-pad last write if image size is not cluster aligned */
1016         memset(buf + bytes, 0, s->cluster_size - bytes);
1017     }
1018     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
1019 
1020     out_buf = g_malloc(s->cluster_size);
1021 
1022     /* best compression, small window, no zlib header */
1023     memset(&strm, 0, sizeof(strm));
1024     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1025                        Z_DEFLATED, -12,
1026                        9, Z_DEFAULT_STRATEGY);
1027     if (ret != 0) {
1028         ret = -EINVAL;
1029         goto fail;
1030     }
1031 
1032     strm.avail_in = s->cluster_size;
1033     strm.next_in = (uint8_t *)buf;
1034     strm.avail_out = s->cluster_size;
1035     strm.next_out = out_buf;
1036 
1037     ret = deflate(&strm, Z_FINISH);
1038     if (ret != Z_STREAM_END && ret != Z_OK) {
1039         deflateEnd(&strm);
1040         ret = -EINVAL;
1041         goto fail;
1042     }
1043     out_len = strm.next_out - out_buf;
1044 
1045     deflateEnd(&strm);
1046 
1047     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1048         /* could not compress: write normal cluster */
1049         ret = qcow_co_writev(bs, offset >> BDRV_SECTOR_BITS,
1050                              bytes >> BDRV_SECTOR_BITS, qiov);
1051         if (ret < 0) {
1052             goto fail;
1053         }
1054         goto success;
1055     }
1056     qemu_co_mutex_lock(&s->lock);
1057     ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1058     qemu_co_mutex_unlock(&s->lock);
1059     if (ret < 0) {
1060         goto fail;
1061     }
1062     if (cluster_offset == 0) {
1063         ret = -EIO;
1064         goto fail;
1065     }
1066     cluster_offset &= s->cluster_offset_mask;
1067 
1068     iov = (struct iovec) {
1069         .iov_base   = out_buf,
1070         .iov_len    = out_len,
1071     };
1072     qemu_iovec_init_external(&hd_qiov, &iov, 1);
1073     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1074     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
1075     if (ret < 0) {
1076         goto fail;
1077     }
1078 success:
1079     ret = 0;
1080 fail:
1081     qemu_vfree(buf);
1082     g_free(out_buf);
1083     return ret;
1084 }
1085 
1086 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1087 {
1088     BDRVQcowState *s = bs->opaque;
1089     bdi->cluster_size = s->cluster_size;
1090     return 0;
1091 }
1092 
1093 static QemuOptsList qcow_create_opts = {
1094     .name = "qcow-create-opts",
1095     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1096     .desc = {
1097         {
1098             .name = BLOCK_OPT_SIZE,
1099             .type = QEMU_OPT_SIZE,
1100             .help = "Virtual disk size"
1101         },
1102         {
1103             .name = BLOCK_OPT_BACKING_FILE,
1104             .type = QEMU_OPT_STRING,
1105             .help = "File name of a base image"
1106         },
1107         {
1108             .name = BLOCK_OPT_ENCRYPT,
1109             .type = QEMU_OPT_BOOL,
1110             .help = "Encrypt the image with format 'aes'. (Deprecated "
1111                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1112         },
1113         {
1114             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1115             .type = QEMU_OPT_STRING,
1116             .help = "Encrypt the image, format choices: 'aes'",
1117         },
1118         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1119         { /* end of list */ }
1120     }
1121 };
1122 
1123 static BlockDriver bdrv_qcow = {
1124     .format_name	= "qcow",
1125     .instance_size	= sizeof(BDRVQcowState),
1126     .bdrv_probe		= qcow_probe,
1127     .bdrv_open		= qcow_open,
1128     .bdrv_close		= qcow_close,
1129     .bdrv_child_perm        = bdrv_format_default_perms,
1130     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1131     .bdrv_co_create_opts    = qcow_co_create_opts,
1132     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1133     .supports_backing       = true,
1134 
1135     .bdrv_co_readv          = qcow_co_readv,
1136     .bdrv_co_writev         = qcow_co_writev,
1137     .bdrv_co_block_status   = qcow_co_block_status,
1138 
1139     .bdrv_make_empty        = qcow_make_empty,
1140     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1141     .bdrv_get_info          = qcow_get_info,
1142 
1143     .create_opts            = &qcow_create_opts,
1144 };
1145 
1146 static void bdrv_qcow_init(void)
1147 {
1148     bdrv_register(&bdrv_qcow);
1149 }
1150 
1151 block_init(bdrv_qcow_init);
1152