xref: /openbmc/qemu/block/qcow.c (revision b45c03f5)
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-common.h"
25 #include "block/block_int.h"
26 #include "qemu/module.h"
27 #include <zlib.h>
28 #include "qapi/qmp/qerror.h"
29 #include "crypto/cipher.h"
30 #include "migration/migration.h"
31 
32 /**************************************************************/
33 /* QEMU COW block driver with compression and encryption support */
34 
35 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
36 #define QCOW_VERSION 1
37 
38 #define QCOW_CRYPT_NONE 0
39 #define QCOW_CRYPT_AES  1
40 
41 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
42 
43 typedef struct QCowHeader {
44     uint32_t magic;
45     uint32_t version;
46     uint64_t backing_file_offset;
47     uint32_t backing_file_size;
48     uint32_t mtime;
49     uint64_t size; /* in bytes */
50     uint8_t cluster_bits;
51     uint8_t l2_bits;
52     uint16_t padding;
53     uint32_t crypt_method;
54     uint64_t l1_table_offset;
55 } QEMU_PACKED QCowHeader;
56 
57 #define L2_CACHE_SIZE 16
58 
59 typedef struct BDRVQcowState {
60     int cluster_bits;
61     int cluster_size;
62     int cluster_sectors;
63     int l2_bits;
64     int l2_size;
65     unsigned int l1_size;
66     uint64_t cluster_offset_mask;
67     uint64_t l1_table_offset;
68     uint64_t *l1_table;
69     uint64_t *l2_cache;
70     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
71     uint32_t l2_cache_counts[L2_CACHE_SIZE];
72     uint8_t *cluster_cache;
73     uint8_t *cluster_data;
74     uint64_t cluster_cache_offset;
75     QCryptoCipher *cipher; /* NULL if no key yet */
76     uint32_t crypt_method_header;
77     CoMutex lock;
78     Error *migration_blocker;
79 } BDRVQcowState;
80 
81 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
82 
83 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
84 {
85     const QCowHeader *cow_header = (const void *)buf;
86 
87     if (buf_size >= sizeof(QCowHeader) &&
88         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
89         be32_to_cpu(cow_header->version) == QCOW_VERSION)
90         return 100;
91     else
92         return 0;
93 }
94 
95 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
96                      Error **errp)
97 {
98     BDRVQcowState *s = bs->opaque;
99     unsigned int len, i, shift;
100     int ret;
101     QCowHeader header;
102 
103     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
104     if (ret < 0) {
105         goto fail;
106     }
107     be32_to_cpus(&header.magic);
108     be32_to_cpus(&header.version);
109     be64_to_cpus(&header.backing_file_offset);
110     be32_to_cpus(&header.backing_file_size);
111     be32_to_cpus(&header.mtime);
112     be64_to_cpus(&header.size);
113     be32_to_cpus(&header.crypt_method);
114     be64_to_cpus(&header.l1_table_offset);
115 
116     if (header.magic != QCOW_MAGIC) {
117         error_setg(errp, "Image not in qcow format");
118         ret = -EINVAL;
119         goto fail;
120     }
121     if (header.version != QCOW_VERSION) {
122         char version[64];
123         snprintf(version, sizeof(version), "QCOW version %" PRIu32,
124                  header.version);
125         error_setg(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
126                    bdrv_get_device_or_node_name(bs), "qcow", version);
127         ret = -ENOTSUP;
128         goto fail;
129     }
130 
131     if (header.size <= 1) {
132         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
133         ret = -EINVAL;
134         goto fail;
135     }
136     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
137         error_setg(errp, "Cluster size must be between 512 and 64k");
138         ret = -EINVAL;
139         goto fail;
140     }
141 
142     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
143      * so bytes = num_entries << 3. */
144     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
145         error_setg(errp, "L2 table size must be between 512 and 64k");
146         ret = -EINVAL;
147         goto fail;
148     }
149 
150     if (header.crypt_method > QCOW_CRYPT_AES) {
151         error_setg(errp, "invalid encryption method in qcow header");
152         ret = -EINVAL;
153         goto fail;
154     }
155     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) {
156         error_setg(errp, "AES cipher not available");
157         ret = -EINVAL;
158         goto fail;
159     }
160     s->crypt_method_header = header.crypt_method;
161     if (s->crypt_method_header) {
162         bs->encrypted = 1;
163     }
164     s->cluster_bits = header.cluster_bits;
165     s->cluster_size = 1 << s->cluster_bits;
166     s->cluster_sectors = 1 << (s->cluster_bits - 9);
167     s->l2_bits = header.l2_bits;
168     s->l2_size = 1 << s->l2_bits;
169     bs->total_sectors = header.size / 512;
170     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
171 
172     /* read the level 1 table */
173     shift = s->cluster_bits + s->l2_bits;
174     if (header.size > UINT64_MAX - (1LL << shift)) {
175         error_setg(errp, "Image too large");
176         ret = -EINVAL;
177         goto fail;
178     } else {
179         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
180         if (l1_size > INT_MAX / sizeof(uint64_t)) {
181             error_setg(errp, "Image too large");
182             ret = -EINVAL;
183             goto fail;
184         }
185         s->l1_size = l1_size;
186     }
187 
188     s->l1_table_offset = header.l1_table_offset;
189     s->l1_table = g_try_new(uint64_t, s->l1_size);
190     if (s->l1_table == NULL) {
191         error_setg(errp, "Could not allocate memory for L1 table");
192         ret = -ENOMEM;
193         goto fail;
194     }
195 
196     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
197                s->l1_size * sizeof(uint64_t));
198     if (ret < 0) {
199         goto fail;
200     }
201 
202     for(i = 0;i < s->l1_size; i++) {
203         be64_to_cpus(&s->l1_table[i]);
204     }
205 
206     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
207     s->l2_cache =
208         qemu_try_blockalign(bs->file,
209                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
210     if (s->l2_cache == NULL) {
211         error_setg(errp, "Could not allocate L2 table cache");
212         ret = -ENOMEM;
213         goto fail;
214     }
215     s->cluster_cache = g_malloc(s->cluster_size);
216     s->cluster_data = g_malloc(s->cluster_size);
217     s->cluster_cache_offset = -1;
218 
219     /* read the backing file name */
220     if (header.backing_file_offset != 0) {
221         len = header.backing_file_size;
222         if (len > 1023 || len >= sizeof(bs->backing_file)) {
223             error_setg(errp, "Backing file name too long");
224             ret = -EINVAL;
225             goto fail;
226         }
227         ret = bdrv_pread(bs->file, header.backing_file_offset,
228                    bs->backing_file, len);
229         if (ret < 0) {
230             goto fail;
231         }
232         bs->backing_file[len] = '\0';
233     }
234 
235     /* Disable migration when qcow images are used */
236     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
237                "does not support live migration",
238                bdrv_get_device_or_node_name(bs));
239     migrate_add_blocker(s->migration_blocker);
240 
241     qemu_co_mutex_init(&s->lock);
242     return 0;
243 
244  fail:
245     g_free(s->l1_table);
246     qemu_vfree(s->l2_cache);
247     g_free(s->cluster_cache);
248     g_free(s->cluster_data);
249     return ret;
250 }
251 
252 
253 /* We have nothing to do for QCOW reopen, stubs just return
254  * success */
255 static int qcow_reopen_prepare(BDRVReopenState *state,
256                                BlockReopenQueue *queue, Error **errp)
257 {
258     return 0;
259 }
260 
261 static int qcow_set_key(BlockDriverState *bs, const char *key)
262 {
263     BDRVQcowState *s = bs->opaque;
264     uint8_t keybuf[16];
265     int len, i;
266     Error *err;
267 
268     memset(keybuf, 0, 16);
269     len = strlen(key);
270     if (len > 16)
271         len = 16;
272     /* XXX: we could compress the chars to 7 bits to increase
273        entropy */
274     for(i = 0;i < len;i++) {
275         keybuf[i] = key[i];
276     }
277     assert(bs->encrypted);
278 
279     qcrypto_cipher_free(s->cipher);
280     s->cipher = qcrypto_cipher_new(
281         QCRYPTO_CIPHER_ALG_AES_128,
282         QCRYPTO_CIPHER_MODE_CBC,
283         keybuf, G_N_ELEMENTS(keybuf),
284         &err);
285 
286     if (!s->cipher) {
287         /* XXX would be nice if errors in this method could
288          * be properly propagate to the caller. Would need
289          * the bdrv_set_key() API signature to be fixed. */
290         error_free(err);
291         return -1;
292     }
293     return 0;
294 }
295 
296 /* The crypt function is compatible with the linux cryptoloop
297    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
298    supported */
299 static int encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
300                            uint8_t *out_buf, const uint8_t *in_buf,
301                            int nb_sectors, bool enc, Error **errp)
302 {
303     union {
304         uint64_t ll[2];
305         uint8_t b[16];
306     } ivec;
307     int i;
308     int ret;
309 
310     for(i = 0; i < nb_sectors; i++) {
311         ivec.ll[0] = cpu_to_le64(sector_num);
312         ivec.ll[1] = 0;
313         if (qcrypto_cipher_setiv(s->cipher,
314                                  ivec.b, G_N_ELEMENTS(ivec.b),
315                                  errp) < 0) {
316             return -1;
317         }
318         if (enc) {
319             ret = qcrypto_cipher_encrypt(s->cipher,
320                                          in_buf,
321                                          out_buf,
322                                          512,
323                                          errp);
324         } else {
325             ret = qcrypto_cipher_decrypt(s->cipher,
326                                          in_buf,
327                                          out_buf,
328                                          512,
329                                          errp);
330         }
331         if (ret < 0) {
332             return -1;
333         }
334         sector_num++;
335         in_buf += 512;
336         out_buf += 512;
337     }
338     return 0;
339 }
340 
341 /* 'allocate' is:
342  *
343  * 0 to not allocate.
344  *
345  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
346  * 'n_end')
347  *
348  * 2 to allocate a compressed cluster of size
349  * 'compressed_size'. 'compressed_size' must be > 0 and <
350  * cluster_size
351  *
352  * return 0 if not allocated.
353  */
354 static uint64_t get_cluster_offset(BlockDriverState *bs,
355                                    uint64_t offset, int allocate,
356                                    int compressed_size,
357                                    int n_start, int n_end)
358 {
359     BDRVQcowState *s = bs->opaque;
360     int min_index, i, j, l1_index, l2_index;
361     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
362     uint32_t min_count;
363     int new_l2_table;
364 
365     l1_index = offset >> (s->l2_bits + s->cluster_bits);
366     l2_offset = s->l1_table[l1_index];
367     new_l2_table = 0;
368     if (!l2_offset) {
369         if (!allocate)
370             return 0;
371         /* allocate a new l2 entry */
372         l2_offset = bdrv_getlength(bs->file);
373         /* round to cluster size */
374         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
375         /* update the L1 entry */
376         s->l1_table[l1_index] = l2_offset;
377         tmp = cpu_to_be64(l2_offset);
378         if (bdrv_pwrite_sync(bs->file,
379                 s->l1_table_offset + l1_index * sizeof(tmp),
380                 &tmp, sizeof(tmp)) < 0)
381             return 0;
382         new_l2_table = 1;
383     }
384     for(i = 0; i < L2_CACHE_SIZE; i++) {
385         if (l2_offset == s->l2_cache_offsets[i]) {
386             /* increment the hit count */
387             if (++s->l2_cache_counts[i] == 0xffffffff) {
388                 for(j = 0; j < L2_CACHE_SIZE; j++) {
389                     s->l2_cache_counts[j] >>= 1;
390                 }
391             }
392             l2_table = s->l2_cache + (i << s->l2_bits);
393             goto found;
394         }
395     }
396     /* not found: load a new entry in the least used one */
397     min_index = 0;
398     min_count = 0xffffffff;
399     for(i = 0; i < L2_CACHE_SIZE; i++) {
400         if (s->l2_cache_counts[i] < min_count) {
401             min_count = s->l2_cache_counts[i];
402             min_index = i;
403         }
404     }
405     l2_table = s->l2_cache + (min_index << s->l2_bits);
406     if (new_l2_table) {
407         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
408         if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
409                 s->l2_size * sizeof(uint64_t)) < 0)
410             return 0;
411     } else {
412         if (bdrv_pread(bs->file, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
413             s->l2_size * sizeof(uint64_t))
414             return 0;
415     }
416     s->l2_cache_offsets[min_index] = l2_offset;
417     s->l2_cache_counts[min_index] = 1;
418  found:
419     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
420     cluster_offset = be64_to_cpu(l2_table[l2_index]);
421     if (!cluster_offset ||
422         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
423         if (!allocate)
424             return 0;
425         /* allocate a new cluster */
426         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
427             (n_end - n_start) < s->cluster_sectors) {
428             /* if the cluster is already compressed, we must
429                decompress it in the case it is not completely
430                overwritten */
431             if (decompress_cluster(bs, cluster_offset) < 0)
432                 return 0;
433             cluster_offset = bdrv_getlength(bs->file);
434             cluster_offset = (cluster_offset + s->cluster_size - 1) &
435                 ~(s->cluster_size - 1);
436             /* write the cluster content */
437             if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache, s->cluster_size) !=
438                 s->cluster_size)
439                 return -1;
440         } else {
441             cluster_offset = bdrv_getlength(bs->file);
442             if (allocate == 1) {
443                 /* round to cluster size */
444                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
445                     ~(s->cluster_size - 1);
446                 bdrv_truncate(bs->file, cluster_offset + s->cluster_size);
447                 /* if encrypted, we must initialize the cluster
448                    content which won't be written */
449                 if (bs->encrypted &&
450                     (n_end - n_start) < s->cluster_sectors) {
451                     uint64_t start_sect;
452                     assert(s->cipher);
453                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
454                     memset(s->cluster_data + 512, 0x00, 512);
455                     for(i = 0; i < s->cluster_sectors; i++) {
456                         if (i < n_start || i >= n_end) {
457                             Error *err = NULL;
458                             if (encrypt_sectors(s, start_sect + i,
459                                                 s->cluster_data,
460                                                 s->cluster_data + 512, 1,
461                                                 true, &err) < 0) {
462                                 error_free(err);
463                                 errno = EIO;
464                                 return -1;
465                             }
466                             if (bdrv_pwrite(bs->file, cluster_offset + i * 512,
467                                             s->cluster_data, 512) != 512)
468                                 return -1;
469                         }
470                     }
471                 }
472             } else if (allocate == 2) {
473                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
474                     (uint64_t)compressed_size << (63 - s->cluster_bits);
475             }
476         }
477         /* update L2 table */
478         tmp = cpu_to_be64(cluster_offset);
479         l2_table[l2_index] = tmp;
480         if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
481                 &tmp, sizeof(tmp)) < 0)
482             return 0;
483     }
484     return cluster_offset;
485 }
486 
487 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
488         int64_t sector_num, int nb_sectors, int *pnum)
489 {
490     BDRVQcowState *s = bs->opaque;
491     int index_in_cluster, n;
492     uint64_t cluster_offset;
493 
494     qemu_co_mutex_lock(&s->lock);
495     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
496     qemu_co_mutex_unlock(&s->lock);
497     index_in_cluster = sector_num & (s->cluster_sectors - 1);
498     n = s->cluster_sectors - index_in_cluster;
499     if (n > nb_sectors)
500         n = nb_sectors;
501     *pnum = n;
502     if (!cluster_offset) {
503         return 0;
504     }
505     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->cipher) {
506         return BDRV_BLOCK_DATA;
507     }
508     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
509     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
510 }
511 
512 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
513                              const uint8_t *buf, int buf_size)
514 {
515     z_stream strm1, *strm = &strm1;
516     int ret, out_len;
517 
518     memset(strm, 0, sizeof(*strm));
519 
520     strm->next_in = (uint8_t *)buf;
521     strm->avail_in = buf_size;
522     strm->next_out = out_buf;
523     strm->avail_out = out_buf_size;
524 
525     ret = inflateInit2(strm, -12);
526     if (ret != Z_OK)
527         return -1;
528     ret = inflate(strm, Z_FINISH);
529     out_len = strm->next_out - out_buf;
530     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
531         out_len != out_buf_size) {
532         inflateEnd(strm);
533         return -1;
534     }
535     inflateEnd(strm);
536     return 0;
537 }
538 
539 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
540 {
541     BDRVQcowState *s = bs->opaque;
542     int ret, csize;
543     uint64_t coffset;
544 
545     coffset = cluster_offset & s->cluster_offset_mask;
546     if (s->cluster_cache_offset != coffset) {
547         csize = cluster_offset >> (63 - s->cluster_bits);
548         csize &= (s->cluster_size - 1);
549         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
550         if (ret != csize)
551             return -1;
552         if (decompress_buffer(s->cluster_cache, s->cluster_size,
553                               s->cluster_data, csize) < 0) {
554             return -1;
555         }
556         s->cluster_cache_offset = coffset;
557     }
558     return 0;
559 }
560 
561 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
562                          int nb_sectors, QEMUIOVector *qiov)
563 {
564     BDRVQcowState *s = bs->opaque;
565     int index_in_cluster;
566     int ret = 0, n;
567     uint64_t cluster_offset;
568     struct iovec hd_iov;
569     QEMUIOVector hd_qiov;
570     uint8_t *buf;
571     void *orig_buf;
572     Error *err = NULL;
573 
574     if (qiov->niov > 1) {
575         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
576         if (buf == NULL) {
577             return -ENOMEM;
578         }
579     } else {
580         orig_buf = NULL;
581         buf = (uint8_t *)qiov->iov->iov_base;
582     }
583 
584     qemu_co_mutex_lock(&s->lock);
585 
586     while (nb_sectors != 0) {
587         /* prepare next request */
588         cluster_offset = get_cluster_offset(bs, sector_num << 9,
589                                                  0, 0, 0, 0);
590         index_in_cluster = sector_num & (s->cluster_sectors - 1);
591         n = s->cluster_sectors - index_in_cluster;
592         if (n > nb_sectors) {
593             n = nb_sectors;
594         }
595 
596         if (!cluster_offset) {
597             if (bs->backing_hd) {
598                 /* read from the base image */
599                 hd_iov.iov_base = (void *)buf;
600                 hd_iov.iov_len = n * 512;
601                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
602                 qemu_co_mutex_unlock(&s->lock);
603                 ret = bdrv_co_readv(bs->backing_hd, sector_num,
604                                     n, &hd_qiov);
605                 qemu_co_mutex_lock(&s->lock);
606                 if (ret < 0) {
607                     goto fail;
608                 }
609             } else {
610                 /* Note: in this case, no need to wait */
611                 memset(buf, 0, 512 * n);
612             }
613         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
614             /* add AIO support for compressed blocks ? */
615             if (decompress_cluster(bs, cluster_offset) < 0) {
616                 goto fail;
617             }
618             memcpy(buf,
619                    s->cluster_cache + index_in_cluster * 512, 512 * n);
620         } else {
621             if ((cluster_offset & 511) != 0) {
622                 goto fail;
623             }
624             hd_iov.iov_base = (void *)buf;
625             hd_iov.iov_len = n * 512;
626             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
627             qemu_co_mutex_unlock(&s->lock);
628             ret = bdrv_co_readv(bs->file,
629                                 (cluster_offset >> 9) + index_in_cluster,
630                                 n, &hd_qiov);
631             qemu_co_mutex_lock(&s->lock);
632             if (ret < 0) {
633                 break;
634             }
635             if (bs->encrypted) {
636                 assert(s->cipher);
637                 if (encrypt_sectors(s, sector_num, buf, buf,
638                                     n, false, &err) < 0) {
639                     goto fail;
640                 }
641             }
642         }
643         ret = 0;
644 
645         nb_sectors -= n;
646         sector_num += n;
647         buf += n * 512;
648     }
649 
650 done:
651     qemu_co_mutex_unlock(&s->lock);
652 
653     if (qiov->niov > 1) {
654         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
655         qemu_vfree(orig_buf);
656     }
657 
658     return ret;
659 
660 fail:
661     error_free(err);
662     ret = -EIO;
663     goto done;
664 }
665 
666 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
667                           int nb_sectors, QEMUIOVector *qiov)
668 {
669     BDRVQcowState *s = bs->opaque;
670     int index_in_cluster;
671     uint64_t cluster_offset;
672     const uint8_t *src_buf;
673     int ret = 0, n;
674     uint8_t *cluster_data = NULL;
675     struct iovec hd_iov;
676     QEMUIOVector hd_qiov;
677     uint8_t *buf;
678     void *orig_buf;
679 
680     s->cluster_cache_offset = -1; /* disable compressed cache */
681 
682     if (qiov->niov > 1) {
683         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
684         if (buf == NULL) {
685             return -ENOMEM;
686         }
687         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
688     } else {
689         orig_buf = NULL;
690         buf = (uint8_t *)qiov->iov->iov_base;
691     }
692 
693     qemu_co_mutex_lock(&s->lock);
694 
695     while (nb_sectors != 0) {
696 
697         index_in_cluster = sector_num & (s->cluster_sectors - 1);
698         n = s->cluster_sectors - index_in_cluster;
699         if (n > nb_sectors) {
700             n = nb_sectors;
701         }
702         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
703                                             index_in_cluster,
704                                             index_in_cluster + n);
705         if (!cluster_offset || (cluster_offset & 511) != 0) {
706             ret = -EIO;
707             break;
708         }
709         if (bs->encrypted) {
710             Error *err = NULL;
711             assert(s->cipher);
712             if (!cluster_data) {
713                 cluster_data = g_malloc0(s->cluster_size);
714             }
715             if (encrypt_sectors(s, sector_num, cluster_data, buf,
716                                 n, true, &err) < 0) {
717                 error_free(err);
718                 ret = -EIO;
719                 break;
720             }
721             src_buf = cluster_data;
722         } else {
723             src_buf = buf;
724         }
725 
726         hd_iov.iov_base = (void *)src_buf;
727         hd_iov.iov_len = n * 512;
728         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
729         qemu_co_mutex_unlock(&s->lock);
730         ret = bdrv_co_writev(bs->file,
731                              (cluster_offset >> 9) + index_in_cluster,
732                              n, &hd_qiov);
733         qemu_co_mutex_lock(&s->lock);
734         if (ret < 0) {
735             break;
736         }
737         ret = 0;
738 
739         nb_sectors -= n;
740         sector_num += n;
741         buf += n * 512;
742     }
743     qemu_co_mutex_unlock(&s->lock);
744 
745     if (qiov->niov > 1) {
746         qemu_vfree(orig_buf);
747     }
748     g_free(cluster_data);
749 
750     return ret;
751 }
752 
753 static void qcow_close(BlockDriverState *bs)
754 {
755     BDRVQcowState *s = bs->opaque;
756 
757     qcrypto_cipher_free(s->cipher);
758     s->cipher = NULL;
759     g_free(s->l1_table);
760     qemu_vfree(s->l2_cache);
761     g_free(s->cluster_cache);
762     g_free(s->cluster_data);
763 
764     migrate_del_blocker(s->migration_blocker);
765     error_free(s->migration_blocker);
766 }
767 
768 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
769 {
770     int header_size, backing_filename_len, l1_size, shift, i;
771     QCowHeader header;
772     uint8_t *tmp;
773     int64_t total_size = 0;
774     char *backing_file = NULL;
775     int flags = 0;
776     Error *local_err = NULL;
777     int ret;
778     BlockDriverState *qcow_bs;
779 
780     /* Read out options */
781     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
782                           BDRV_SECTOR_SIZE);
783     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
784     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
785         flags |= BLOCK_FLAG_ENCRYPT;
786     }
787 
788     ret = bdrv_create_file(filename, opts, &local_err);
789     if (ret < 0) {
790         error_propagate(errp, local_err);
791         goto cleanup;
792     }
793 
794     qcow_bs = NULL;
795     ret = bdrv_open(&qcow_bs, filename, NULL, NULL,
796                     BDRV_O_RDWR | BDRV_O_PROTOCOL, NULL, &local_err);
797     if (ret < 0) {
798         error_propagate(errp, local_err);
799         goto cleanup;
800     }
801 
802     ret = bdrv_truncate(qcow_bs, 0);
803     if (ret < 0) {
804         goto exit;
805     }
806 
807     memset(&header, 0, sizeof(header));
808     header.magic = cpu_to_be32(QCOW_MAGIC);
809     header.version = cpu_to_be32(QCOW_VERSION);
810     header.size = cpu_to_be64(total_size);
811     header_size = sizeof(header);
812     backing_filename_len = 0;
813     if (backing_file) {
814         if (strcmp(backing_file, "fat:")) {
815             header.backing_file_offset = cpu_to_be64(header_size);
816             backing_filename_len = strlen(backing_file);
817             header.backing_file_size = cpu_to_be32(backing_filename_len);
818             header_size += backing_filename_len;
819         } else {
820             /* special backing file for vvfat */
821             backing_file = NULL;
822         }
823         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
824                                     unmodified sectors */
825         header.l2_bits = 12; /* 32 KB L2 tables */
826     } else {
827         header.cluster_bits = 12; /* 4 KB clusters */
828         header.l2_bits = 9; /* 4 KB L2 tables */
829     }
830     header_size = (header_size + 7) & ~7;
831     shift = header.cluster_bits + header.l2_bits;
832     l1_size = (total_size + (1LL << shift) - 1) >> shift;
833 
834     header.l1_table_offset = cpu_to_be64(header_size);
835     if (flags & BLOCK_FLAG_ENCRYPT) {
836         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
837     } else {
838         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
839     }
840 
841     /* write all the data */
842     ret = bdrv_pwrite(qcow_bs, 0, &header, sizeof(header));
843     if (ret != sizeof(header)) {
844         goto exit;
845     }
846 
847     if (backing_file) {
848         ret = bdrv_pwrite(qcow_bs, sizeof(header),
849             backing_file, backing_filename_len);
850         if (ret != backing_filename_len) {
851             goto exit;
852         }
853     }
854 
855     tmp = g_malloc0(BDRV_SECTOR_SIZE);
856     for (i = 0; i < ((sizeof(uint64_t)*l1_size + BDRV_SECTOR_SIZE - 1)/
857         BDRV_SECTOR_SIZE); i++) {
858         ret = bdrv_pwrite(qcow_bs, header_size +
859             BDRV_SECTOR_SIZE*i, tmp, BDRV_SECTOR_SIZE);
860         if (ret != BDRV_SECTOR_SIZE) {
861             g_free(tmp);
862             goto exit;
863         }
864     }
865 
866     g_free(tmp);
867     ret = 0;
868 exit:
869     bdrv_unref(qcow_bs);
870 cleanup:
871     g_free(backing_file);
872     return ret;
873 }
874 
875 static int qcow_make_empty(BlockDriverState *bs)
876 {
877     BDRVQcowState *s = bs->opaque;
878     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
879     int ret;
880 
881     memset(s->l1_table, 0, l1_length);
882     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
883             l1_length) < 0)
884         return -1;
885     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
886     if (ret < 0)
887         return ret;
888 
889     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
890     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
891     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
892 
893     return 0;
894 }
895 
896 /* XXX: put compressed sectors first, then all the cluster aligned
897    tables to avoid losing bytes in alignment */
898 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
899                                  const uint8_t *buf, int nb_sectors)
900 {
901     BDRVQcowState *s = bs->opaque;
902     z_stream strm;
903     int ret, out_len;
904     uint8_t *out_buf;
905     uint64_t cluster_offset;
906 
907     if (nb_sectors != s->cluster_sectors) {
908         ret = -EINVAL;
909 
910         /* Zero-pad last write if image size is not cluster aligned */
911         if (sector_num + nb_sectors == bs->total_sectors &&
912             nb_sectors < s->cluster_sectors) {
913             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
914             memset(pad_buf, 0, s->cluster_size);
915             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
916             ret = qcow_write_compressed(bs, sector_num,
917                                         pad_buf, s->cluster_sectors);
918             qemu_vfree(pad_buf);
919         }
920         return ret;
921     }
922 
923     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
924 
925     /* best compression, small window, no zlib header */
926     memset(&strm, 0, sizeof(strm));
927     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
928                        Z_DEFLATED, -12,
929                        9, Z_DEFAULT_STRATEGY);
930     if (ret != 0) {
931         ret = -EINVAL;
932         goto fail;
933     }
934 
935     strm.avail_in = s->cluster_size;
936     strm.next_in = (uint8_t *)buf;
937     strm.avail_out = s->cluster_size;
938     strm.next_out = out_buf;
939 
940     ret = deflate(&strm, Z_FINISH);
941     if (ret != Z_STREAM_END && ret != Z_OK) {
942         deflateEnd(&strm);
943         ret = -EINVAL;
944         goto fail;
945     }
946     out_len = strm.next_out - out_buf;
947 
948     deflateEnd(&strm);
949 
950     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
951         /* could not compress: write normal cluster */
952         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
953         if (ret < 0) {
954             goto fail;
955         }
956     } else {
957         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
958                                             out_len, 0, 0);
959         if (cluster_offset == 0) {
960             ret = -EIO;
961             goto fail;
962         }
963 
964         cluster_offset &= s->cluster_offset_mask;
965         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
966         if (ret < 0) {
967             goto fail;
968         }
969     }
970 
971     ret = 0;
972 fail:
973     g_free(out_buf);
974     return ret;
975 }
976 
977 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
978 {
979     BDRVQcowState *s = bs->opaque;
980     bdi->cluster_size = s->cluster_size;
981     return 0;
982 }
983 
984 static QemuOptsList qcow_create_opts = {
985     .name = "qcow-create-opts",
986     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
987     .desc = {
988         {
989             .name = BLOCK_OPT_SIZE,
990             .type = QEMU_OPT_SIZE,
991             .help = "Virtual disk size"
992         },
993         {
994             .name = BLOCK_OPT_BACKING_FILE,
995             .type = QEMU_OPT_STRING,
996             .help = "File name of a base image"
997         },
998         {
999             .name = BLOCK_OPT_ENCRYPT,
1000             .type = QEMU_OPT_BOOL,
1001             .help = "Encrypt the image",
1002             .def_value_str = "off"
1003         },
1004         { /* end of list */ }
1005     }
1006 };
1007 
1008 static BlockDriver bdrv_qcow = {
1009     .format_name	= "qcow",
1010     .instance_size	= sizeof(BDRVQcowState),
1011     .bdrv_probe		= qcow_probe,
1012     .bdrv_open		= qcow_open,
1013     .bdrv_close		= qcow_close,
1014     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1015     .bdrv_create            = qcow_create,
1016     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1017     .supports_backing       = true,
1018 
1019     .bdrv_co_readv          = qcow_co_readv,
1020     .bdrv_co_writev         = qcow_co_writev,
1021     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1022 
1023     .bdrv_set_key           = qcow_set_key,
1024     .bdrv_make_empty        = qcow_make_empty,
1025     .bdrv_write_compressed  = qcow_write_compressed,
1026     .bdrv_get_info          = qcow_get_info,
1027 
1028     .create_opts            = &qcow_create_opts,
1029 };
1030 
1031 static void bdrv_qcow_init(void)
1032 {
1033     bdrv_register(&bdrv_qcow);
1034 }
1035 
1036 block_init(bdrv_qcow_init);
1037