xref: /openbmc/qemu/block/qcow.c (revision 91bfcdb0)
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->bs, 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->bs, 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->bs,
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->bs, 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->bs);
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->bs,
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->bs, l2_offset, l2_table,
409                 s->l2_size * sizeof(uint64_t)) < 0)
410             return 0;
411     } else {
412         if (bdrv_pread(bs->file->bs, l2_offset, l2_table,
413                        s->l2_size * sizeof(uint64_t)) !=
414             s->l2_size * sizeof(uint64_t))
415             return 0;
416     }
417     s->l2_cache_offsets[min_index] = l2_offset;
418     s->l2_cache_counts[min_index] = 1;
419  found:
420     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
421     cluster_offset = be64_to_cpu(l2_table[l2_index]);
422     if (!cluster_offset ||
423         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
424         if (!allocate)
425             return 0;
426         /* allocate a new cluster */
427         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
428             (n_end - n_start) < s->cluster_sectors) {
429             /* if the cluster is already compressed, we must
430                decompress it in the case it is not completely
431                overwritten */
432             if (decompress_cluster(bs, cluster_offset) < 0)
433                 return 0;
434             cluster_offset = bdrv_getlength(bs->file->bs);
435             cluster_offset = (cluster_offset + s->cluster_size - 1) &
436                 ~(s->cluster_size - 1);
437             /* write the cluster content */
438             if (bdrv_pwrite(bs->file->bs, cluster_offset, s->cluster_cache,
439                             s->cluster_size) !=
440                 s->cluster_size)
441                 return -1;
442         } else {
443             cluster_offset = bdrv_getlength(bs->file->bs);
444             if (allocate == 1) {
445                 /* round to cluster size */
446                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
447                     ~(s->cluster_size - 1);
448                 bdrv_truncate(bs->file->bs, cluster_offset + s->cluster_size);
449                 /* if encrypted, we must initialize the cluster
450                    content which won't be written */
451                 if (bs->encrypted &&
452                     (n_end - n_start) < s->cluster_sectors) {
453                     uint64_t start_sect;
454                     assert(s->cipher);
455                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
456                     memset(s->cluster_data + 512, 0x00, 512);
457                     for(i = 0; i < s->cluster_sectors; i++) {
458                         if (i < n_start || i >= n_end) {
459                             Error *err = NULL;
460                             if (encrypt_sectors(s, start_sect + i,
461                                                 s->cluster_data,
462                                                 s->cluster_data + 512, 1,
463                                                 true, &err) < 0) {
464                                 error_free(err);
465                                 errno = EIO;
466                                 return -1;
467                             }
468                             if (bdrv_pwrite(bs->file->bs,
469                                             cluster_offset + i * 512,
470                                             s->cluster_data, 512) != 512)
471                                 return -1;
472                         }
473                     }
474                 }
475             } else if (allocate == 2) {
476                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
477                     (uint64_t)compressed_size << (63 - s->cluster_bits);
478             }
479         }
480         /* update L2 table */
481         tmp = cpu_to_be64(cluster_offset);
482         l2_table[l2_index] = tmp;
483         if (bdrv_pwrite_sync(bs->file->bs, l2_offset + l2_index * sizeof(tmp),
484                 &tmp, sizeof(tmp)) < 0)
485             return 0;
486     }
487     return cluster_offset;
488 }
489 
490 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
491         int64_t sector_num, int nb_sectors, int *pnum)
492 {
493     BDRVQcowState *s = bs->opaque;
494     int index_in_cluster, n;
495     uint64_t cluster_offset;
496 
497     qemu_co_mutex_lock(&s->lock);
498     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
499     qemu_co_mutex_unlock(&s->lock);
500     index_in_cluster = sector_num & (s->cluster_sectors - 1);
501     n = s->cluster_sectors - index_in_cluster;
502     if (n > nb_sectors)
503         n = nb_sectors;
504     *pnum = n;
505     if (!cluster_offset) {
506         return 0;
507     }
508     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->cipher) {
509         return BDRV_BLOCK_DATA;
510     }
511     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
512     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
513 }
514 
515 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
516                              const uint8_t *buf, int buf_size)
517 {
518     z_stream strm1, *strm = &strm1;
519     int ret, out_len;
520 
521     memset(strm, 0, sizeof(*strm));
522 
523     strm->next_in = (uint8_t *)buf;
524     strm->avail_in = buf_size;
525     strm->next_out = out_buf;
526     strm->avail_out = out_buf_size;
527 
528     ret = inflateInit2(strm, -12);
529     if (ret != Z_OK)
530         return -1;
531     ret = inflate(strm, Z_FINISH);
532     out_len = strm->next_out - out_buf;
533     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
534         out_len != out_buf_size) {
535         inflateEnd(strm);
536         return -1;
537     }
538     inflateEnd(strm);
539     return 0;
540 }
541 
542 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
543 {
544     BDRVQcowState *s = bs->opaque;
545     int ret, csize;
546     uint64_t coffset;
547 
548     coffset = cluster_offset & s->cluster_offset_mask;
549     if (s->cluster_cache_offset != coffset) {
550         csize = cluster_offset >> (63 - s->cluster_bits);
551         csize &= (s->cluster_size - 1);
552         ret = bdrv_pread(bs->file->bs, coffset, s->cluster_data, csize);
553         if (ret != csize)
554             return -1;
555         if (decompress_buffer(s->cluster_cache, s->cluster_size,
556                               s->cluster_data, csize) < 0) {
557             return -1;
558         }
559         s->cluster_cache_offset = coffset;
560     }
561     return 0;
562 }
563 
564 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
565                          int nb_sectors, QEMUIOVector *qiov)
566 {
567     BDRVQcowState *s = bs->opaque;
568     int index_in_cluster;
569     int ret = 0, n;
570     uint64_t cluster_offset;
571     struct iovec hd_iov;
572     QEMUIOVector hd_qiov;
573     uint8_t *buf;
574     void *orig_buf;
575     Error *err = NULL;
576 
577     if (qiov->niov > 1) {
578         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
579         if (buf == NULL) {
580             return -ENOMEM;
581         }
582     } else {
583         orig_buf = NULL;
584         buf = (uint8_t *)qiov->iov->iov_base;
585     }
586 
587     qemu_co_mutex_lock(&s->lock);
588 
589     while (nb_sectors != 0) {
590         /* prepare next request */
591         cluster_offset = get_cluster_offset(bs, sector_num << 9,
592                                                  0, 0, 0, 0);
593         index_in_cluster = sector_num & (s->cluster_sectors - 1);
594         n = s->cluster_sectors - index_in_cluster;
595         if (n > nb_sectors) {
596             n = nb_sectors;
597         }
598 
599         if (!cluster_offset) {
600             if (bs->backing) {
601                 /* read from the base image */
602                 hd_iov.iov_base = (void *)buf;
603                 hd_iov.iov_len = n * 512;
604                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
605                 qemu_co_mutex_unlock(&s->lock);
606                 ret = bdrv_co_readv(bs->backing->bs, sector_num,
607                                     n, &hd_qiov);
608                 qemu_co_mutex_lock(&s->lock);
609                 if (ret < 0) {
610                     goto fail;
611                 }
612             } else {
613                 /* Note: in this case, no need to wait */
614                 memset(buf, 0, 512 * n);
615             }
616         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
617             /* add AIO support for compressed blocks ? */
618             if (decompress_cluster(bs, cluster_offset) < 0) {
619                 goto fail;
620             }
621             memcpy(buf,
622                    s->cluster_cache + index_in_cluster * 512, 512 * n);
623         } else {
624             if ((cluster_offset & 511) != 0) {
625                 goto fail;
626             }
627             hd_iov.iov_base = (void *)buf;
628             hd_iov.iov_len = n * 512;
629             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
630             qemu_co_mutex_unlock(&s->lock);
631             ret = bdrv_co_readv(bs->file->bs,
632                                 (cluster_offset >> 9) + index_in_cluster,
633                                 n, &hd_qiov);
634             qemu_co_mutex_lock(&s->lock);
635             if (ret < 0) {
636                 break;
637             }
638             if (bs->encrypted) {
639                 assert(s->cipher);
640                 if (encrypt_sectors(s, sector_num, buf, buf,
641                                     n, false, &err) < 0) {
642                     goto fail;
643                 }
644             }
645         }
646         ret = 0;
647 
648         nb_sectors -= n;
649         sector_num += n;
650         buf += n * 512;
651     }
652 
653 done:
654     qemu_co_mutex_unlock(&s->lock);
655 
656     if (qiov->niov > 1) {
657         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
658         qemu_vfree(orig_buf);
659     }
660 
661     return ret;
662 
663 fail:
664     error_free(err);
665     ret = -EIO;
666     goto done;
667 }
668 
669 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
670                           int nb_sectors, QEMUIOVector *qiov)
671 {
672     BDRVQcowState *s = bs->opaque;
673     int index_in_cluster;
674     uint64_t cluster_offset;
675     const uint8_t *src_buf;
676     int ret = 0, n;
677     uint8_t *cluster_data = NULL;
678     struct iovec hd_iov;
679     QEMUIOVector hd_qiov;
680     uint8_t *buf;
681     void *orig_buf;
682 
683     s->cluster_cache_offset = -1; /* disable compressed cache */
684 
685     if (qiov->niov > 1) {
686         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
687         if (buf == NULL) {
688             return -ENOMEM;
689         }
690         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
691     } else {
692         orig_buf = NULL;
693         buf = (uint8_t *)qiov->iov->iov_base;
694     }
695 
696     qemu_co_mutex_lock(&s->lock);
697 
698     while (nb_sectors != 0) {
699 
700         index_in_cluster = sector_num & (s->cluster_sectors - 1);
701         n = s->cluster_sectors - index_in_cluster;
702         if (n > nb_sectors) {
703             n = nb_sectors;
704         }
705         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
706                                             index_in_cluster,
707                                             index_in_cluster + n);
708         if (!cluster_offset || (cluster_offset & 511) != 0) {
709             ret = -EIO;
710             break;
711         }
712         if (bs->encrypted) {
713             Error *err = NULL;
714             assert(s->cipher);
715             if (!cluster_data) {
716                 cluster_data = g_malloc0(s->cluster_size);
717             }
718             if (encrypt_sectors(s, sector_num, cluster_data, buf,
719                                 n, true, &err) < 0) {
720                 error_free(err);
721                 ret = -EIO;
722                 break;
723             }
724             src_buf = cluster_data;
725         } else {
726             src_buf = buf;
727         }
728 
729         hd_iov.iov_base = (void *)src_buf;
730         hd_iov.iov_len = n * 512;
731         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
732         qemu_co_mutex_unlock(&s->lock);
733         ret = bdrv_co_writev(bs->file->bs,
734                              (cluster_offset >> 9) + index_in_cluster,
735                              n, &hd_qiov);
736         qemu_co_mutex_lock(&s->lock);
737         if (ret < 0) {
738             break;
739         }
740         ret = 0;
741 
742         nb_sectors -= n;
743         sector_num += n;
744         buf += n * 512;
745     }
746     qemu_co_mutex_unlock(&s->lock);
747 
748     if (qiov->niov > 1) {
749         qemu_vfree(orig_buf);
750     }
751     g_free(cluster_data);
752 
753     return ret;
754 }
755 
756 static void qcow_close(BlockDriverState *bs)
757 {
758     BDRVQcowState *s = bs->opaque;
759 
760     qcrypto_cipher_free(s->cipher);
761     s->cipher = NULL;
762     g_free(s->l1_table);
763     qemu_vfree(s->l2_cache);
764     g_free(s->cluster_cache);
765     g_free(s->cluster_data);
766 
767     migrate_del_blocker(s->migration_blocker);
768     error_free(s->migration_blocker);
769 }
770 
771 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
772 {
773     int header_size, backing_filename_len, l1_size, shift, i;
774     QCowHeader header;
775     uint8_t *tmp;
776     int64_t total_size = 0;
777     char *backing_file = NULL;
778     int flags = 0;
779     Error *local_err = NULL;
780     int ret;
781     BlockDriverState *qcow_bs;
782 
783     /* Read out options */
784     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
785                           BDRV_SECTOR_SIZE);
786     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
787     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
788         flags |= BLOCK_FLAG_ENCRYPT;
789     }
790 
791     ret = bdrv_create_file(filename, opts, &local_err);
792     if (ret < 0) {
793         error_propagate(errp, local_err);
794         goto cleanup;
795     }
796 
797     qcow_bs = NULL;
798     ret = bdrv_open(&qcow_bs, filename, NULL, NULL,
799                     BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
800     if (ret < 0) {
801         error_propagate(errp, local_err);
802         goto cleanup;
803     }
804 
805     ret = bdrv_truncate(qcow_bs, 0);
806     if (ret < 0) {
807         goto exit;
808     }
809 
810     memset(&header, 0, sizeof(header));
811     header.magic = cpu_to_be32(QCOW_MAGIC);
812     header.version = cpu_to_be32(QCOW_VERSION);
813     header.size = cpu_to_be64(total_size);
814     header_size = sizeof(header);
815     backing_filename_len = 0;
816     if (backing_file) {
817         if (strcmp(backing_file, "fat:")) {
818             header.backing_file_offset = cpu_to_be64(header_size);
819             backing_filename_len = strlen(backing_file);
820             header.backing_file_size = cpu_to_be32(backing_filename_len);
821             header_size += backing_filename_len;
822         } else {
823             /* special backing file for vvfat */
824             backing_file = NULL;
825         }
826         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
827                                     unmodified sectors */
828         header.l2_bits = 12; /* 32 KB L2 tables */
829     } else {
830         header.cluster_bits = 12; /* 4 KB clusters */
831         header.l2_bits = 9; /* 4 KB L2 tables */
832     }
833     header_size = (header_size + 7) & ~7;
834     shift = header.cluster_bits + header.l2_bits;
835     l1_size = (total_size + (1LL << shift) - 1) >> shift;
836 
837     header.l1_table_offset = cpu_to_be64(header_size);
838     if (flags & BLOCK_FLAG_ENCRYPT) {
839         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
840     } else {
841         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
842     }
843 
844     /* write all the data */
845     ret = bdrv_pwrite(qcow_bs, 0, &header, sizeof(header));
846     if (ret != sizeof(header)) {
847         goto exit;
848     }
849 
850     if (backing_file) {
851         ret = bdrv_pwrite(qcow_bs, sizeof(header),
852             backing_file, backing_filename_len);
853         if (ret != backing_filename_len) {
854             goto exit;
855         }
856     }
857 
858     tmp = g_malloc0(BDRV_SECTOR_SIZE);
859     for (i = 0; i < ((sizeof(uint64_t)*l1_size + BDRV_SECTOR_SIZE - 1)/
860         BDRV_SECTOR_SIZE); i++) {
861         ret = bdrv_pwrite(qcow_bs, header_size +
862             BDRV_SECTOR_SIZE*i, tmp, BDRV_SECTOR_SIZE);
863         if (ret != BDRV_SECTOR_SIZE) {
864             g_free(tmp);
865             goto exit;
866         }
867     }
868 
869     g_free(tmp);
870     ret = 0;
871 exit:
872     bdrv_unref(qcow_bs);
873 cleanup:
874     g_free(backing_file);
875     return ret;
876 }
877 
878 static int qcow_make_empty(BlockDriverState *bs)
879 {
880     BDRVQcowState *s = bs->opaque;
881     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
882     int ret;
883 
884     memset(s->l1_table, 0, l1_length);
885     if (bdrv_pwrite_sync(bs->file->bs, s->l1_table_offset, s->l1_table,
886             l1_length) < 0)
887         return -1;
888     ret = bdrv_truncate(bs->file->bs, s->l1_table_offset + l1_length);
889     if (ret < 0)
890         return ret;
891 
892     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
893     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
894     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
895 
896     return 0;
897 }
898 
899 /* XXX: put compressed sectors first, then all the cluster aligned
900    tables to avoid losing bytes in alignment */
901 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
902                                  const uint8_t *buf, int nb_sectors)
903 {
904     BDRVQcowState *s = bs->opaque;
905     z_stream strm;
906     int ret, out_len;
907     uint8_t *out_buf;
908     uint64_t cluster_offset;
909 
910     if (nb_sectors != s->cluster_sectors) {
911         ret = -EINVAL;
912 
913         /* Zero-pad last write if image size is not cluster aligned */
914         if (sector_num + nb_sectors == bs->total_sectors &&
915             nb_sectors < s->cluster_sectors) {
916             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
917             memset(pad_buf, 0, s->cluster_size);
918             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
919             ret = qcow_write_compressed(bs, sector_num,
920                                         pad_buf, s->cluster_sectors);
921             qemu_vfree(pad_buf);
922         }
923         return ret;
924     }
925 
926     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
927 
928     /* best compression, small window, no zlib header */
929     memset(&strm, 0, sizeof(strm));
930     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
931                        Z_DEFLATED, -12,
932                        9, Z_DEFAULT_STRATEGY);
933     if (ret != 0) {
934         ret = -EINVAL;
935         goto fail;
936     }
937 
938     strm.avail_in = s->cluster_size;
939     strm.next_in = (uint8_t *)buf;
940     strm.avail_out = s->cluster_size;
941     strm.next_out = out_buf;
942 
943     ret = deflate(&strm, Z_FINISH);
944     if (ret != Z_STREAM_END && ret != Z_OK) {
945         deflateEnd(&strm);
946         ret = -EINVAL;
947         goto fail;
948     }
949     out_len = strm.next_out - out_buf;
950 
951     deflateEnd(&strm);
952 
953     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
954         /* could not compress: write normal cluster */
955         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
956         if (ret < 0) {
957             goto fail;
958         }
959     } else {
960         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
961                                             out_len, 0, 0);
962         if (cluster_offset == 0) {
963             ret = -EIO;
964             goto fail;
965         }
966 
967         cluster_offset &= s->cluster_offset_mask;
968         ret = bdrv_pwrite(bs->file->bs, cluster_offset, out_buf, out_len);
969         if (ret < 0) {
970             goto fail;
971         }
972     }
973 
974     ret = 0;
975 fail:
976     g_free(out_buf);
977     return ret;
978 }
979 
980 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
981 {
982     BDRVQcowState *s = bs->opaque;
983     bdi->cluster_size = s->cluster_size;
984     return 0;
985 }
986 
987 static QemuOptsList qcow_create_opts = {
988     .name = "qcow-create-opts",
989     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
990     .desc = {
991         {
992             .name = BLOCK_OPT_SIZE,
993             .type = QEMU_OPT_SIZE,
994             .help = "Virtual disk size"
995         },
996         {
997             .name = BLOCK_OPT_BACKING_FILE,
998             .type = QEMU_OPT_STRING,
999             .help = "File name of a base image"
1000         },
1001         {
1002             .name = BLOCK_OPT_ENCRYPT,
1003             .type = QEMU_OPT_BOOL,
1004             .help = "Encrypt the image",
1005             .def_value_str = "off"
1006         },
1007         { /* end of list */ }
1008     }
1009 };
1010 
1011 static BlockDriver bdrv_qcow = {
1012     .format_name	= "qcow",
1013     .instance_size	= sizeof(BDRVQcowState),
1014     .bdrv_probe		= qcow_probe,
1015     .bdrv_open		= qcow_open,
1016     .bdrv_close		= qcow_close,
1017     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1018     .bdrv_create            = qcow_create,
1019     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1020     .supports_backing       = true,
1021 
1022     .bdrv_co_readv          = qcow_co_readv,
1023     .bdrv_co_writev         = qcow_co_writev,
1024     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1025 
1026     .bdrv_set_key           = qcow_set_key,
1027     .bdrv_make_empty        = qcow_make_empty,
1028     .bdrv_write_compressed  = qcow_write_compressed,
1029     .bdrv_get_info          = qcow_get_info,
1030 
1031     .create_opts            = &qcow_create_opts,
1032 };
1033 
1034 static void bdrv_qcow_init(void)
1035 {
1036     bdrv_register(&bdrv_qcow);
1037 }
1038 
1039 block_init(bdrv_qcow_init);
1040