xref: /openbmc/qemu/block/qcow.c (revision 87776ab7)
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 "crypto/cipher.h"
35 #include "migration/migration.h"
36 
37 /**************************************************************/
38 /* QEMU COW block driver with compression and encryption support */
39 
40 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
41 #define QCOW_VERSION 1
42 
43 #define QCOW_CRYPT_NONE 0
44 #define QCOW_CRYPT_AES  1
45 
46 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
47 
48 typedef struct QCowHeader {
49     uint32_t magic;
50     uint32_t version;
51     uint64_t backing_file_offset;
52     uint32_t backing_file_size;
53     uint32_t mtime;
54     uint64_t size; /* in bytes */
55     uint8_t cluster_bits;
56     uint8_t l2_bits;
57     uint16_t padding;
58     uint32_t crypt_method;
59     uint64_t l1_table_offset;
60 } QEMU_PACKED QCowHeader;
61 
62 #define L2_CACHE_SIZE 16
63 
64 typedef struct BDRVQcowState {
65     int cluster_bits;
66     int cluster_size;
67     int cluster_sectors;
68     int l2_bits;
69     int l2_size;
70     unsigned int l1_size;
71     uint64_t cluster_offset_mask;
72     uint64_t l1_table_offset;
73     uint64_t *l1_table;
74     uint64_t *l2_cache;
75     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
76     uint32_t l2_cache_counts[L2_CACHE_SIZE];
77     uint8_t *cluster_cache;
78     uint8_t *cluster_data;
79     uint64_t cluster_cache_offset;
80     QCryptoCipher *cipher; /* NULL if no key yet */
81     uint32_t crypt_method_header;
82     CoMutex lock;
83     Error *migration_blocker;
84 } BDRVQcowState;
85 
86 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
87 
88 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
89 {
90     const QCowHeader *cow_header = (const void *)buf;
91 
92     if (buf_size >= sizeof(QCowHeader) &&
93         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
94         be32_to_cpu(cow_header->version) == QCOW_VERSION)
95         return 100;
96     else
97         return 0;
98 }
99 
100 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
101                      Error **errp)
102 {
103     BDRVQcowState *s = bs->opaque;
104     unsigned int len, i, shift;
105     int ret;
106     QCowHeader header;
107 
108     ret = bdrv_pread(bs->file->bs, 0, &header, sizeof(header));
109     if (ret < 0) {
110         goto fail;
111     }
112     be32_to_cpus(&header.magic);
113     be32_to_cpus(&header.version);
114     be64_to_cpus(&header.backing_file_offset);
115     be32_to_cpus(&header.backing_file_size);
116     be32_to_cpus(&header.mtime);
117     be64_to_cpus(&header.size);
118     be32_to_cpus(&header.crypt_method);
119     be64_to_cpus(&header.l1_table_offset);
120 
121     if (header.magic != QCOW_MAGIC) {
122         error_setg(errp, "Image not in qcow format");
123         ret = -EINVAL;
124         goto fail;
125     }
126     if (header.version != QCOW_VERSION) {
127         error_setg(errp, "Unsupported qcow version %" PRIu32, header.version);
128         ret = -ENOTSUP;
129         goto fail;
130     }
131 
132     if (header.size <= 1) {
133         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
134         ret = -EINVAL;
135         goto fail;
136     }
137     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
138         error_setg(errp, "Cluster size must be between 512 and 64k");
139         ret = -EINVAL;
140         goto fail;
141     }
142 
143     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
144      * so bytes = num_entries << 3. */
145     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
146         error_setg(errp, "L2 table size must be between 512 and 64k");
147         ret = -EINVAL;
148         goto fail;
149     }
150 
151     if (header.crypt_method > QCOW_CRYPT_AES) {
152         error_setg(errp, "invalid encryption method in qcow header");
153         ret = -EINVAL;
154         goto fail;
155     }
156     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) {
157         error_setg(errp, "AES cipher not available");
158         ret = -EINVAL;
159         goto fail;
160     }
161     s->crypt_method_header = header.crypt_method;
162     if (s->crypt_method_header) {
163         if (bdrv_uses_whitelist() &&
164             s->crypt_method_header == QCOW_CRYPT_AES) {
165             error_report("qcow built-in AES encryption is deprecated");
166             error_printf("Support for it will be removed in a future release.\n"
167                          "You can use 'qemu-img convert' to switch to an\n"
168                          "unencrypted qcow image, or a LUKS raw image.\n");
169         }
170 
171         bs->encrypted = 1;
172     }
173     s->cluster_bits = header.cluster_bits;
174     s->cluster_size = 1 << s->cluster_bits;
175     s->cluster_sectors = 1 << (s->cluster_bits - 9);
176     s->l2_bits = header.l2_bits;
177     s->l2_size = 1 << s->l2_bits;
178     bs->total_sectors = header.size / 512;
179     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
180 
181     /* read the level 1 table */
182     shift = s->cluster_bits + s->l2_bits;
183     if (header.size > UINT64_MAX - (1LL << shift)) {
184         error_setg(errp, "Image too large");
185         ret = -EINVAL;
186         goto fail;
187     } else {
188         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
189         if (l1_size > INT_MAX / sizeof(uint64_t)) {
190             error_setg(errp, "Image too large");
191             ret = -EINVAL;
192             goto fail;
193         }
194         s->l1_size = l1_size;
195     }
196 
197     s->l1_table_offset = header.l1_table_offset;
198     s->l1_table = g_try_new(uint64_t, s->l1_size);
199     if (s->l1_table == NULL) {
200         error_setg(errp, "Could not allocate memory for L1 table");
201         ret = -ENOMEM;
202         goto fail;
203     }
204 
205     ret = bdrv_pread(bs->file->bs, s->l1_table_offset, s->l1_table,
206                s->l1_size * sizeof(uint64_t));
207     if (ret < 0) {
208         goto fail;
209     }
210 
211     for(i = 0;i < s->l1_size; i++) {
212         be64_to_cpus(&s->l1_table[i]);
213     }
214 
215     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
216     s->l2_cache =
217         qemu_try_blockalign(bs->file->bs,
218                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
219     if (s->l2_cache == NULL) {
220         error_setg(errp, "Could not allocate L2 table cache");
221         ret = -ENOMEM;
222         goto fail;
223     }
224     s->cluster_cache = g_malloc(s->cluster_size);
225     s->cluster_data = g_malloc(s->cluster_size);
226     s->cluster_cache_offset = -1;
227 
228     /* read the backing file name */
229     if (header.backing_file_offset != 0) {
230         len = header.backing_file_size;
231         if (len > 1023 || len >= sizeof(bs->backing_file)) {
232             error_setg(errp, "Backing file name too long");
233             ret = -EINVAL;
234             goto fail;
235         }
236         ret = bdrv_pread(bs->file->bs, header.backing_file_offset,
237                    bs->backing_file, len);
238         if (ret < 0) {
239             goto fail;
240         }
241         bs->backing_file[len] = '\0';
242     }
243 
244     /* Disable migration when qcow images are used */
245     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
246                "does not support live migration",
247                bdrv_get_device_or_node_name(bs));
248     migrate_add_blocker(s->migration_blocker);
249 
250     qemu_co_mutex_init(&s->lock);
251     return 0;
252 
253  fail:
254     g_free(s->l1_table);
255     qemu_vfree(s->l2_cache);
256     g_free(s->cluster_cache);
257     g_free(s->cluster_data);
258     return ret;
259 }
260 
261 
262 /* We have nothing to do for QCOW reopen, stubs just return
263  * success */
264 static int qcow_reopen_prepare(BDRVReopenState *state,
265                                BlockReopenQueue *queue, Error **errp)
266 {
267     return 0;
268 }
269 
270 static int qcow_set_key(BlockDriverState *bs, const char *key)
271 {
272     BDRVQcowState *s = bs->opaque;
273     uint8_t keybuf[16];
274     int len, i;
275     Error *err;
276 
277     memset(keybuf, 0, 16);
278     len = strlen(key);
279     if (len > 16)
280         len = 16;
281     /* XXX: we could compress the chars to 7 bits to increase
282        entropy */
283     for(i = 0;i < len;i++) {
284         keybuf[i] = key[i];
285     }
286     assert(bs->encrypted);
287 
288     qcrypto_cipher_free(s->cipher);
289     s->cipher = qcrypto_cipher_new(
290         QCRYPTO_CIPHER_ALG_AES_128,
291         QCRYPTO_CIPHER_MODE_CBC,
292         keybuf, G_N_ELEMENTS(keybuf),
293         &err);
294 
295     if (!s->cipher) {
296         /* XXX would be nice if errors in this method could
297          * be properly propagate to the caller. Would need
298          * the bdrv_set_key() API signature to be fixed. */
299         error_free(err);
300         return -1;
301     }
302     return 0;
303 }
304 
305 /* The crypt function is compatible with the linux cryptoloop
306    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
307    supported */
308 static int encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
309                            uint8_t *out_buf, const uint8_t *in_buf,
310                            int nb_sectors, bool enc, Error **errp)
311 {
312     union {
313         uint64_t ll[2];
314         uint8_t b[16];
315     } ivec;
316     int i;
317     int ret;
318 
319     for(i = 0; i < nb_sectors; i++) {
320         ivec.ll[0] = cpu_to_le64(sector_num);
321         ivec.ll[1] = 0;
322         if (qcrypto_cipher_setiv(s->cipher,
323                                  ivec.b, G_N_ELEMENTS(ivec.b),
324                                  errp) < 0) {
325             return -1;
326         }
327         if (enc) {
328             ret = qcrypto_cipher_encrypt(s->cipher,
329                                          in_buf,
330                                          out_buf,
331                                          512,
332                                          errp);
333         } else {
334             ret = qcrypto_cipher_decrypt(s->cipher,
335                                          in_buf,
336                                          out_buf,
337                                          512,
338                                          errp);
339         }
340         if (ret < 0) {
341             return -1;
342         }
343         sector_num++;
344         in_buf += 512;
345         out_buf += 512;
346     }
347     return 0;
348 }
349 
350 /* 'allocate' is:
351  *
352  * 0 to not allocate.
353  *
354  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
355  * 'n_end')
356  *
357  * 2 to allocate a compressed cluster of size
358  * 'compressed_size'. 'compressed_size' must be > 0 and <
359  * cluster_size
360  *
361  * return 0 if not allocated.
362  */
363 static uint64_t get_cluster_offset(BlockDriverState *bs,
364                                    uint64_t offset, int allocate,
365                                    int compressed_size,
366                                    int n_start, int n_end)
367 {
368     BDRVQcowState *s = bs->opaque;
369     int min_index, i, j, l1_index, l2_index;
370     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
371     uint32_t min_count;
372     int new_l2_table;
373 
374     l1_index = offset >> (s->l2_bits + s->cluster_bits);
375     l2_offset = s->l1_table[l1_index];
376     new_l2_table = 0;
377     if (!l2_offset) {
378         if (!allocate)
379             return 0;
380         /* allocate a new l2 entry */
381         l2_offset = bdrv_getlength(bs->file->bs);
382         /* round to cluster size */
383         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
384         /* update the L1 entry */
385         s->l1_table[l1_index] = l2_offset;
386         tmp = cpu_to_be64(l2_offset);
387         if (bdrv_pwrite_sync(bs->file->bs,
388                 s->l1_table_offset + l1_index * sizeof(tmp),
389                 &tmp, sizeof(tmp)) < 0)
390             return 0;
391         new_l2_table = 1;
392     }
393     for(i = 0; i < L2_CACHE_SIZE; i++) {
394         if (l2_offset == s->l2_cache_offsets[i]) {
395             /* increment the hit count */
396             if (++s->l2_cache_counts[i] == 0xffffffff) {
397                 for(j = 0; j < L2_CACHE_SIZE; j++) {
398                     s->l2_cache_counts[j] >>= 1;
399                 }
400             }
401             l2_table = s->l2_cache + (i << s->l2_bits);
402             goto found;
403         }
404     }
405     /* not found: load a new entry in the least used one */
406     min_index = 0;
407     min_count = 0xffffffff;
408     for(i = 0; i < L2_CACHE_SIZE; i++) {
409         if (s->l2_cache_counts[i] < min_count) {
410             min_count = s->l2_cache_counts[i];
411             min_index = i;
412         }
413     }
414     l2_table = s->l2_cache + (min_index << s->l2_bits);
415     if (new_l2_table) {
416         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
417         if (bdrv_pwrite_sync(bs->file->bs, l2_offset, l2_table,
418                 s->l2_size * sizeof(uint64_t)) < 0)
419             return 0;
420     } else {
421         if (bdrv_pread(bs->file->bs, l2_offset, l2_table,
422                        s->l2_size * sizeof(uint64_t)) !=
423             s->l2_size * sizeof(uint64_t))
424             return 0;
425     }
426     s->l2_cache_offsets[min_index] = l2_offset;
427     s->l2_cache_counts[min_index] = 1;
428  found:
429     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
430     cluster_offset = be64_to_cpu(l2_table[l2_index]);
431     if (!cluster_offset ||
432         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
433         if (!allocate)
434             return 0;
435         /* allocate a new cluster */
436         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
437             (n_end - n_start) < s->cluster_sectors) {
438             /* if the cluster is already compressed, we must
439                decompress it in the case it is not completely
440                overwritten */
441             if (decompress_cluster(bs, cluster_offset) < 0)
442                 return 0;
443             cluster_offset = bdrv_getlength(bs->file->bs);
444             cluster_offset = (cluster_offset + s->cluster_size - 1) &
445                 ~(s->cluster_size - 1);
446             /* write the cluster content */
447             if (bdrv_pwrite(bs->file->bs, cluster_offset, s->cluster_cache,
448                             s->cluster_size) !=
449                 s->cluster_size)
450                 return -1;
451         } else {
452             cluster_offset = bdrv_getlength(bs->file->bs);
453             if (allocate == 1) {
454                 /* round to cluster size */
455                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
456                     ~(s->cluster_size - 1);
457                 bdrv_truncate(bs->file->bs, cluster_offset + s->cluster_size);
458                 /* if encrypted, we must initialize the cluster
459                    content which won't be written */
460                 if (bs->encrypted &&
461                     (n_end - n_start) < s->cluster_sectors) {
462                     uint64_t start_sect;
463                     assert(s->cipher);
464                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
465                     memset(s->cluster_data + 512, 0x00, 512);
466                     for(i = 0; i < s->cluster_sectors; i++) {
467                         if (i < n_start || i >= n_end) {
468                             Error *err = NULL;
469                             if (encrypt_sectors(s, start_sect + i,
470                                                 s->cluster_data,
471                                                 s->cluster_data + 512, 1,
472                                                 true, &err) < 0) {
473                                 error_free(err);
474                                 errno = EIO;
475                                 return -1;
476                             }
477                             if (bdrv_pwrite(bs->file->bs,
478                                             cluster_offset + i * 512,
479                                             s->cluster_data, 512) != 512)
480                                 return -1;
481                         }
482                     }
483                 }
484             } else if (allocate == 2) {
485                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
486                     (uint64_t)compressed_size << (63 - s->cluster_bits);
487             }
488         }
489         /* update L2 table */
490         tmp = cpu_to_be64(cluster_offset);
491         l2_table[l2_index] = tmp;
492         if (bdrv_pwrite_sync(bs->file->bs, l2_offset + l2_index * sizeof(tmp),
493                 &tmp, sizeof(tmp)) < 0)
494             return 0;
495     }
496     return cluster_offset;
497 }
498 
499 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
500         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
501 {
502     BDRVQcowState *s = bs->opaque;
503     int index_in_cluster, n;
504     uint64_t cluster_offset;
505 
506     qemu_co_mutex_lock(&s->lock);
507     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
508     qemu_co_mutex_unlock(&s->lock);
509     index_in_cluster = sector_num & (s->cluster_sectors - 1);
510     n = s->cluster_sectors - index_in_cluster;
511     if (n > nb_sectors)
512         n = nb_sectors;
513     *pnum = n;
514     if (!cluster_offset) {
515         return 0;
516     }
517     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->cipher) {
518         return BDRV_BLOCK_DATA;
519     }
520     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
521     *file = bs->file->bs;
522     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
523 }
524 
525 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
526                              const uint8_t *buf, int buf_size)
527 {
528     z_stream strm1, *strm = &strm1;
529     int ret, out_len;
530 
531     memset(strm, 0, sizeof(*strm));
532 
533     strm->next_in = (uint8_t *)buf;
534     strm->avail_in = buf_size;
535     strm->next_out = out_buf;
536     strm->avail_out = out_buf_size;
537 
538     ret = inflateInit2(strm, -12);
539     if (ret != Z_OK)
540         return -1;
541     ret = inflate(strm, Z_FINISH);
542     out_len = strm->next_out - out_buf;
543     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
544         out_len != out_buf_size) {
545         inflateEnd(strm);
546         return -1;
547     }
548     inflateEnd(strm);
549     return 0;
550 }
551 
552 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
553 {
554     BDRVQcowState *s = bs->opaque;
555     int ret, csize;
556     uint64_t coffset;
557 
558     coffset = cluster_offset & s->cluster_offset_mask;
559     if (s->cluster_cache_offset != coffset) {
560         csize = cluster_offset >> (63 - s->cluster_bits);
561         csize &= (s->cluster_size - 1);
562         ret = bdrv_pread(bs->file->bs, coffset, s->cluster_data, csize);
563         if (ret != csize)
564             return -1;
565         if (decompress_buffer(s->cluster_cache, s->cluster_size,
566                               s->cluster_data, csize) < 0) {
567             return -1;
568         }
569         s->cluster_cache_offset = coffset;
570     }
571     return 0;
572 }
573 
574 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
575                          int nb_sectors, QEMUIOVector *qiov)
576 {
577     BDRVQcowState *s = bs->opaque;
578     int index_in_cluster;
579     int ret = 0, n;
580     uint64_t cluster_offset;
581     struct iovec hd_iov;
582     QEMUIOVector hd_qiov;
583     uint8_t *buf;
584     void *orig_buf;
585     Error *err = NULL;
586 
587     if (qiov->niov > 1) {
588         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
589         if (buf == NULL) {
590             return -ENOMEM;
591         }
592     } else {
593         orig_buf = NULL;
594         buf = (uint8_t *)qiov->iov->iov_base;
595     }
596 
597     qemu_co_mutex_lock(&s->lock);
598 
599     while (nb_sectors != 0) {
600         /* prepare next request */
601         cluster_offset = get_cluster_offset(bs, sector_num << 9,
602                                                  0, 0, 0, 0);
603         index_in_cluster = sector_num & (s->cluster_sectors - 1);
604         n = s->cluster_sectors - index_in_cluster;
605         if (n > nb_sectors) {
606             n = nb_sectors;
607         }
608 
609         if (!cluster_offset) {
610             if (bs->backing) {
611                 /* read from the base image */
612                 hd_iov.iov_base = (void *)buf;
613                 hd_iov.iov_len = n * 512;
614                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
615                 qemu_co_mutex_unlock(&s->lock);
616                 ret = bdrv_co_readv(bs->backing->bs, sector_num,
617                                     n, &hd_qiov);
618                 qemu_co_mutex_lock(&s->lock);
619                 if (ret < 0) {
620                     goto fail;
621                 }
622             } else {
623                 /* Note: in this case, no need to wait */
624                 memset(buf, 0, 512 * n);
625             }
626         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
627             /* add AIO support for compressed blocks ? */
628             if (decompress_cluster(bs, cluster_offset) < 0) {
629                 goto fail;
630             }
631             memcpy(buf,
632                    s->cluster_cache + index_in_cluster * 512, 512 * n);
633         } else {
634             if ((cluster_offset & 511) != 0) {
635                 goto fail;
636             }
637             hd_iov.iov_base = (void *)buf;
638             hd_iov.iov_len = n * 512;
639             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
640             qemu_co_mutex_unlock(&s->lock);
641             ret = bdrv_co_readv(bs->file->bs,
642                                 (cluster_offset >> 9) + index_in_cluster,
643                                 n, &hd_qiov);
644             qemu_co_mutex_lock(&s->lock);
645             if (ret < 0) {
646                 break;
647             }
648             if (bs->encrypted) {
649                 assert(s->cipher);
650                 if (encrypt_sectors(s, sector_num, buf, buf,
651                                     n, false, &err) < 0) {
652                     goto fail;
653                 }
654             }
655         }
656         ret = 0;
657 
658         nb_sectors -= n;
659         sector_num += n;
660         buf += n * 512;
661     }
662 
663 done:
664     qemu_co_mutex_unlock(&s->lock);
665 
666     if (qiov->niov > 1) {
667         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
668         qemu_vfree(orig_buf);
669     }
670 
671     return ret;
672 
673 fail:
674     error_free(err);
675     ret = -EIO;
676     goto done;
677 }
678 
679 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
680                           int nb_sectors, QEMUIOVector *qiov)
681 {
682     BDRVQcowState *s = bs->opaque;
683     int index_in_cluster;
684     uint64_t cluster_offset;
685     const uint8_t *src_buf;
686     int ret = 0, n;
687     uint8_t *cluster_data = NULL;
688     struct iovec hd_iov;
689     QEMUIOVector hd_qiov;
690     uint8_t *buf;
691     void *orig_buf;
692 
693     s->cluster_cache_offset = -1; /* disable compressed cache */
694 
695     if (qiov->niov > 1) {
696         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
697         if (buf == NULL) {
698             return -ENOMEM;
699         }
700         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
701     } else {
702         orig_buf = NULL;
703         buf = (uint8_t *)qiov->iov->iov_base;
704     }
705 
706     qemu_co_mutex_lock(&s->lock);
707 
708     while (nb_sectors != 0) {
709 
710         index_in_cluster = sector_num & (s->cluster_sectors - 1);
711         n = s->cluster_sectors - index_in_cluster;
712         if (n > nb_sectors) {
713             n = nb_sectors;
714         }
715         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
716                                             index_in_cluster,
717                                             index_in_cluster + n);
718         if (!cluster_offset || (cluster_offset & 511) != 0) {
719             ret = -EIO;
720             break;
721         }
722         if (bs->encrypted) {
723             Error *err = NULL;
724             assert(s->cipher);
725             if (!cluster_data) {
726                 cluster_data = g_malloc0(s->cluster_size);
727             }
728             if (encrypt_sectors(s, sector_num, cluster_data, buf,
729                                 n, true, &err) < 0) {
730                 error_free(err);
731                 ret = -EIO;
732                 break;
733             }
734             src_buf = cluster_data;
735         } else {
736             src_buf = buf;
737         }
738 
739         hd_iov.iov_base = (void *)src_buf;
740         hd_iov.iov_len = n * 512;
741         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
742         qemu_co_mutex_unlock(&s->lock);
743         ret = bdrv_co_writev(bs->file->bs,
744                              (cluster_offset >> 9) + index_in_cluster,
745                              n, &hd_qiov);
746         qemu_co_mutex_lock(&s->lock);
747         if (ret < 0) {
748             break;
749         }
750         ret = 0;
751 
752         nb_sectors -= n;
753         sector_num += n;
754         buf += n * 512;
755     }
756     qemu_co_mutex_unlock(&s->lock);
757 
758     if (qiov->niov > 1) {
759         qemu_vfree(orig_buf);
760     }
761     g_free(cluster_data);
762 
763     return ret;
764 }
765 
766 static void qcow_close(BlockDriverState *bs)
767 {
768     BDRVQcowState *s = bs->opaque;
769 
770     qcrypto_cipher_free(s->cipher);
771     s->cipher = NULL;
772     g_free(s->l1_table);
773     qemu_vfree(s->l2_cache);
774     g_free(s->cluster_cache);
775     g_free(s->cluster_data);
776 
777     migrate_del_blocker(s->migration_blocker);
778     error_free(s->migration_blocker);
779 }
780 
781 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
782 {
783     int header_size, backing_filename_len, l1_size, shift, i;
784     QCowHeader header;
785     uint8_t *tmp;
786     int64_t total_size = 0;
787     char *backing_file = NULL;
788     int flags = 0;
789     Error *local_err = NULL;
790     int ret;
791     BlockBackend *qcow_blk;
792 
793     /* Read out options */
794     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
795                           BDRV_SECTOR_SIZE);
796     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
797     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
798         flags |= BLOCK_FLAG_ENCRYPT;
799     }
800 
801     ret = bdrv_create_file(filename, opts, &local_err);
802     if (ret < 0) {
803         error_propagate(errp, local_err);
804         goto cleanup;
805     }
806 
807     qcow_blk = blk_new_open(filename, NULL, NULL,
808                             BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
809     if (qcow_blk == NULL) {
810         error_propagate(errp, local_err);
811         ret = -EIO;
812         goto cleanup;
813     }
814 
815     blk_set_allow_write_beyond_eof(qcow_blk, true);
816 
817     ret = blk_truncate(qcow_blk, 0);
818     if (ret < 0) {
819         goto exit;
820     }
821 
822     memset(&header, 0, sizeof(header));
823     header.magic = cpu_to_be32(QCOW_MAGIC);
824     header.version = cpu_to_be32(QCOW_VERSION);
825     header.size = cpu_to_be64(total_size);
826     header_size = sizeof(header);
827     backing_filename_len = 0;
828     if (backing_file) {
829         if (strcmp(backing_file, "fat:")) {
830             header.backing_file_offset = cpu_to_be64(header_size);
831             backing_filename_len = strlen(backing_file);
832             header.backing_file_size = cpu_to_be32(backing_filename_len);
833             header_size += backing_filename_len;
834         } else {
835             /* special backing file for vvfat */
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     if (flags & BLOCK_FLAG_ENCRYPT) {
851         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
852     } else {
853         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
854     }
855 
856     /* write all the data */
857     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
858     if (ret != sizeof(header)) {
859         goto exit;
860     }
861 
862     if (backing_file) {
863         ret = blk_pwrite(qcow_blk, sizeof(header),
864                          backing_file, backing_filename_len, 0);
865         if (ret != backing_filename_len) {
866             goto exit;
867         }
868     }
869 
870     tmp = g_malloc0(BDRV_SECTOR_SIZE);
871     for (i = 0; i < ((sizeof(uint64_t)*l1_size + BDRV_SECTOR_SIZE - 1)/
872         BDRV_SECTOR_SIZE); i++) {
873         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
874                          tmp, BDRV_SECTOR_SIZE, 0);
875         if (ret != BDRV_SECTOR_SIZE) {
876             g_free(tmp);
877             goto exit;
878         }
879     }
880 
881     g_free(tmp);
882     ret = 0;
883 exit:
884     blk_unref(qcow_blk);
885 cleanup:
886     g_free(backing_file);
887     return ret;
888 }
889 
890 static int qcow_make_empty(BlockDriverState *bs)
891 {
892     BDRVQcowState *s = bs->opaque;
893     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
894     int ret;
895 
896     memset(s->l1_table, 0, l1_length);
897     if (bdrv_pwrite_sync(bs->file->bs, s->l1_table_offset, s->l1_table,
898             l1_length) < 0)
899         return -1;
900     ret = bdrv_truncate(bs->file->bs, s->l1_table_offset + l1_length);
901     if (ret < 0)
902         return ret;
903 
904     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
905     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
906     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
907 
908     return 0;
909 }
910 
911 /* XXX: put compressed sectors first, then all the cluster aligned
912    tables to avoid losing bytes in alignment */
913 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
914                                  const uint8_t *buf, int nb_sectors)
915 {
916     BDRVQcowState *s = bs->opaque;
917     z_stream strm;
918     int ret, out_len;
919     uint8_t *out_buf;
920     uint64_t cluster_offset;
921 
922     if (nb_sectors != s->cluster_sectors) {
923         ret = -EINVAL;
924 
925         /* Zero-pad last write if image size is not cluster aligned */
926         if (sector_num + nb_sectors == bs->total_sectors &&
927             nb_sectors < s->cluster_sectors) {
928             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
929             memset(pad_buf, 0, s->cluster_size);
930             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
931             ret = qcow_write_compressed(bs, sector_num,
932                                         pad_buf, s->cluster_sectors);
933             qemu_vfree(pad_buf);
934         }
935         return ret;
936     }
937 
938     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
939 
940     /* best compression, small window, no zlib header */
941     memset(&strm, 0, sizeof(strm));
942     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
943                        Z_DEFLATED, -12,
944                        9, Z_DEFAULT_STRATEGY);
945     if (ret != 0) {
946         ret = -EINVAL;
947         goto fail;
948     }
949 
950     strm.avail_in = s->cluster_size;
951     strm.next_in = (uint8_t *)buf;
952     strm.avail_out = s->cluster_size;
953     strm.next_out = out_buf;
954 
955     ret = deflate(&strm, Z_FINISH);
956     if (ret != Z_STREAM_END && ret != Z_OK) {
957         deflateEnd(&strm);
958         ret = -EINVAL;
959         goto fail;
960     }
961     out_len = strm.next_out - out_buf;
962 
963     deflateEnd(&strm);
964 
965     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
966         /* could not compress: write normal cluster */
967         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
968         if (ret < 0) {
969             goto fail;
970         }
971     } else {
972         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
973                                             out_len, 0, 0);
974         if (cluster_offset == 0) {
975             ret = -EIO;
976             goto fail;
977         }
978 
979         cluster_offset &= s->cluster_offset_mask;
980         ret = bdrv_pwrite(bs->file->bs, cluster_offset, out_buf, out_len);
981         if (ret < 0) {
982             goto fail;
983         }
984     }
985 
986     ret = 0;
987 fail:
988     g_free(out_buf);
989     return ret;
990 }
991 
992 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
993 {
994     BDRVQcowState *s = bs->opaque;
995     bdi->cluster_size = s->cluster_size;
996     return 0;
997 }
998 
999 static QemuOptsList qcow_create_opts = {
1000     .name = "qcow-create-opts",
1001     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1002     .desc = {
1003         {
1004             .name = BLOCK_OPT_SIZE,
1005             .type = QEMU_OPT_SIZE,
1006             .help = "Virtual disk size"
1007         },
1008         {
1009             .name = BLOCK_OPT_BACKING_FILE,
1010             .type = QEMU_OPT_STRING,
1011             .help = "File name of a base image"
1012         },
1013         {
1014             .name = BLOCK_OPT_ENCRYPT,
1015             .type = QEMU_OPT_BOOL,
1016             .help = "Encrypt the image",
1017             .def_value_str = "off"
1018         },
1019         { /* end of list */ }
1020     }
1021 };
1022 
1023 static BlockDriver bdrv_qcow = {
1024     .format_name	= "qcow",
1025     .instance_size	= sizeof(BDRVQcowState),
1026     .bdrv_probe		= qcow_probe,
1027     .bdrv_open		= qcow_open,
1028     .bdrv_close		= qcow_close,
1029     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1030     .bdrv_create            = qcow_create,
1031     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1032     .supports_backing       = true,
1033 
1034     .bdrv_co_readv          = qcow_co_readv,
1035     .bdrv_co_writev         = qcow_co_writev,
1036     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1037 
1038     .bdrv_set_key           = qcow_set_key,
1039     .bdrv_make_empty        = qcow_make_empty,
1040     .bdrv_write_compressed  = qcow_write_compressed,
1041     .bdrv_get_info          = qcow_get_info,
1042 
1043     .create_opts            = &qcow_create_opts,
1044 };
1045 
1046 static void bdrv_qcow_init(void)
1047 {
1048     bdrv_register(&bdrv_qcow);
1049 }
1050 
1051 block_init(bdrv_qcow_init);
1052