xref: /openbmc/qemu/block/qcow.c (revision 10358b6a)
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 "qemu/aes.h"
29 #include "migration/migration.h"
30 
31 /**************************************************************/
32 /* QEMU COW block driver with compression and encryption support */
33 
34 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
35 #define QCOW_VERSION 1
36 
37 #define QCOW_CRYPT_NONE 0
38 #define QCOW_CRYPT_AES  1
39 
40 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
41 
42 typedef struct QCowHeader {
43     uint32_t magic;
44     uint32_t version;
45     uint64_t backing_file_offset;
46     uint32_t backing_file_size;
47     uint32_t mtime;
48     uint64_t size; /* in bytes */
49     uint8_t cluster_bits;
50     uint8_t l2_bits;
51     uint16_t padding;
52     uint32_t crypt_method;
53     uint64_t l1_table_offset;
54 } QEMU_PACKED QCowHeader;
55 
56 #define L2_CACHE_SIZE 16
57 
58 typedef struct BDRVQcowState {
59     int cluster_bits;
60     int cluster_size;
61     int cluster_sectors;
62     int l2_bits;
63     int l2_size;
64     unsigned int l1_size;
65     uint64_t cluster_offset_mask;
66     uint64_t l1_table_offset;
67     uint64_t *l1_table;
68     uint64_t *l2_cache;
69     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
70     uint32_t l2_cache_counts[L2_CACHE_SIZE];
71     uint8_t *cluster_cache;
72     uint8_t *cluster_data;
73     uint64_t cluster_cache_offset;
74     uint32_t crypt_method; /* current crypt method, 0 if no key yet */
75     uint32_t crypt_method_header;
76     AES_KEY aes_encrypt_key;
77     AES_KEY aes_decrypt_key;
78     CoMutex lock;
79     Error *migration_blocker;
80 } BDRVQcowState;
81 
82 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
83 
84 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
85 {
86     const QCowHeader *cow_header = (const void *)buf;
87 
88     if (buf_size >= sizeof(QCowHeader) &&
89         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
90         be32_to_cpu(cow_header->version) == QCOW_VERSION)
91         return 100;
92     else
93         return 0;
94 }
95 
96 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
97                      Error **errp)
98 {
99     BDRVQcowState *s = bs->opaque;
100     unsigned int len, i, shift;
101     int ret;
102     QCowHeader header;
103 
104     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
105     if (ret < 0) {
106         goto fail;
107     }
108     be32_to_cpus(&header.magic);
109     be32_to_cpus(&header.version);
110     be64_to_cpus(&header.backing_file_offset);
111     be32_to_cpus(&header.backing_file_size);
112     be32_to_cpus(&header.mtime);
113     be64_to_cpus(&header.size);
114     be32_to_cpus(&header.crypt_method);
115     be64_to_cpus(&header.l1_table_offset);
116 
117     if (header.magic != QCOW_MAGIC) {
118         error_setg(errp, "Image not in qcow format");
119         ret = -EINVAL;
120         goto fail;
121     }
122     if (header.version != QCOW_VERSION) {
123         char version[64];
124         snprintf(version, sizeof(version), "QCOW version %" PRIu32,
125                  header.version);
126         error_set(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
127                   bs->device_name, "qcow", 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     s->crypt_method_header = header.crypt_method;
157     if (s->crypt_method_header) {
158         bs->encrypted = 1;
159     }
160     s->cluster_bits = header.cluster_bits;
161     s->cluster_size = 1 << s->cluster_bits;
162     s->cluster_sectors = 1 << (s->cluster_bits - 9);
163     s->l2_bits = header.l2_bits;
164     s->l2_size = 1 << s->l2_bits;
165     bs->total_sectors = header.size / 512;
166     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
167 
168     /* read the level 1 table */
169     shift = s->cluster_bits + s->l2_bits;
170     if (header.size > UINT64_MAX - (1LL << shift)) {
171         error_setg(errp, "Image too large");
172         ret = -EINVAL;
173         goto fail;
174     } else {
175         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
176         if (l1_size > INT_MAX / sizeof(uint64_t)) {
177             error_setg(errp, "Image too large");
178             ret = -EINVAL;
179             goto fail;
180         }
181         s->l1_size = l1_size;
182     }
183 
184     s->l1_table_offset = header.l1_table_offset;
185     s->l1_table = g_malloc(s->l1_size * sizeof(uint64_t));
186 
187     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
188                s->l1_size * sizeof(uint64_t));
189     if (ret < 0) {
190         goto fail;
191     }
192 
193     for(i = 0;i < s->l1_size; i++) {
194         be64_to_cpus(&s->l1_table[i]);
195     }
196     /* alloc L2 cache */
197     s->l2_cache = g_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
198     s->cluster_cache = g_malloc(s->cluster_size);
199     s->cluster_data = g_malloc(s->cluster_size);
200     s->cluster_cache_offset = -1;
201 
202     /* read the backing file name */
203     if (header.backing_file_offset != 0) {
204         len = header.backing_file_size;
205         if (len > 1023) {
206             error_setg(errp, "Backing file name too long");
207             ret = -EINVAL;
208             goto fail;
209         }
210         ret = bdrv_pread(bs->file, header.backing_file_offset,
211                    bs->backing_file, len);
212         if (ret < 0) {
213             goto fail;
214         }
215         bs->backing_file[len] = '\0';
216     }
217 
218     /* Disable migration when qcow images are used */
219     error_set(&s->migration_blocker,
220               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
221               "qcow", bs->device_name, "live migration");
222     migrate_add_blocker(s->migration_blocker);
223 
224     qemu_co_mutex_init(&s->lock);
225     return 0;
226 
227  fail:
228     g_free(s->l1_table);
229     g_free(s->l2_cache);
230     g_free(s->cluster_cache);
231     g_free(s->cluster_data);
232     return ret;
233 }
234 
235 
236 /* We have nothing to do for QCOW reopen, stubs just return
237  * success */
238 static int qcow_reopen_prepare(BDRVReopenState *state,
239                                BlockReopenQueue *queue, Error **errp)
240 {
241     return 0;
242 }
243 
244 static int qcow_set_key(BlockDriverState *bs, const char *key)
245 {
246     BDRVQcowState *s = bs->opaque;
247     uint8_t keybuf[16];
248     int len, i;
249 
250     memset(keybuf, 0, 16);
251     len = strlen(key);
252     if (len > 16)
253         len = 16;
254     /* XXX: we could compress the chars to 7 bits to increase
255        entropy */
256     for(i = 0;i < len;i++) {
257         keybuf[i] = key[i];
258     }
259     s->crypt_method = s->crypt_method_header;
260 
261     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
262         return -1;
263     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
264         return -1;
265     return 0;
266 }
267 
268 /* The crypt function is compatible with the linux cryptoloop
269    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
270    supported */
271 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
272                             uint8_t *out_buf, const uint8_t *in_buf,
273                             int nb_sectors, int enc,
274                             const AES_KEY *key)
275 {
276     union {
277         uint64_t ll[2];
278         uint8_t b[16];
279     } ivec;
280     int i;
281 
282     for(i = 0; i < nb_sectors; i++) {
283         ivec.ll[0] = cpu_to_le64(sector_num);
284         ivec.ll[1] = 0;
285         AES_cbc_encrypt(in_buf, out_buf, 512, key,
286                         ivec.b, enc);
287         sector_num++;
288         in_buf += 512;
289         out_buf += 512;
290     }
291 }
292 
293 /* 'allocate' is:
294  *
295  * 0 to not allocate.
296  *
297  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
298  * 'n_end')
299  *
300  * 2 to allocate a compressed cluster of size
301  * 'compressed_size'. 'compressed_size' must be > 0 and <
302  * cluster_size
303  *
304  * return 0 if not allocated.
305  */
306 static uint64_t get_cluster_offset(BlockDriverState *bs,
307                                    uint64_t offset, int allocate,
308                                    int compressed_size,
309                                    int n_start, int n_end)
310 {
311     BDRVQcowState *s = bs->opaque;
312     int min_index, i, j, l1_index, l2_index;
313     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
314     uint32_t min_count;
315     int new_l2_table;
316 
317     l1_index = offset >> (s->l2_bits + s->cluster_bits);
318     l2_offset = s->l1_table[l1_index];
319     new_l2_table = 0;
320     if (!l2_offset) {
321         if (!allocate)
322             return 0;
323         /* allocate a new l2 entry */
324         l2_offset = bdrv_getlength(bs->file);
325         /* round to cluster size */
326         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
327         /* update the L1 entry */
328         s->l1_table[l1_index] = l2_offset;
329         tmp = cpu_to_be64(l2_offset);
330         if (bdrv_pwrite_sync(bs->file,
331                 s->l1_table_offset + l1_index * sizeof(tmp),
332                 &tmp, sizeof(tmp)) < 0)
333             return 0;
334         new_l2_table = 1;
335     }
336     for(i = 0; i < L2_CACHE_SIZE; i++) {
337         if (l2_offset == s->l2_cache_offsets[i]) {
338             /* increment the hit count */
339             if (++s->l2_cache_counts[i] == 0xffffffff) {
340                 for(j = 0; j < L2_CACHE_SIZE; j++) {
341                     s->l2_cache_counts[j] >>= 1;
342                 }
343             }
344             l2_table = s->l2_cache + (i << s->l2_bits);
345             goto found;
346         }
347     }
348     /* not found: load a new entry in the least used one */
349     min_index = 0;
350     min_count = 0xffffffff;
351     for(i = 0; i < L2_CACHE_SIZE; i++) {
352         if (s->l2_cache_counts[i] < min_count) {
353             min_count = s->l2_cache_counts[i];
354             min_index = i;
355         }
356     }
357     l2_table = s->l2_cache + (min_index << s->l2_bits);
358     if (new_l2_table) {
359         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
360         if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
361                 s->l2_size * sizeof(uint64_t)) < 0)
362             return 0;
363     } else {
364         if (bdrv_pread(bs->file, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
365             s->l2_size * sizeof(uint64_t))
366             return 0;
367     }
368     s->l2_cache_offsets[min_index] = l2_offset;
369     s->l2_cache_counts[min_index] = 1;
370  found:
371     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
372     cluster_offset = be64_to_cpu(l2_table[l2_index]);
373     if (!cluster_offset ||
374         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
375         if (!allocate)
376             return 0;
377         /* allocate a new cluster */
378         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
379             (n_end - n_start) < s->cluster_sectors) {
380             /* if the cluster is already compressed, we must
381                decompress it in the case it is not completely
382                overwritten */
383             if (decompress_cluster(bs, cluster_offset) < 0)
384                 return 0;
385             cluster_offset = bdrv_getlength(bs->file);
386             cluster_offset = (cluster_offset + s->cluster_size - 1) &
387                 ~(s->cluster_size - 1);
388             /* write the cluster content */
389             if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache, s->cluster_size) !=
390                 s->cluster_size)
391                 return -1;
392         } else {
393             cluster_offset = bdrv_getlength(bs->file);
394             if (allocate == 1) {
395                 /* round to cluster size */
396                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
397                     ~(s->cluster_size - 1);
398                 bdrv_truncate(bs->file, cluster_offset + s->cluster_size);
399                 /* if encrypted, we must initialize the cluster
400                    content which won't be written */
401                 if (s->crypt_method &&
402                     (n_end - n_start) < s->cluster_sectors) {
403                     uint64_t start_sect;
404                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
405                     memset(s->cluster_data + 512, 0x00, 512);
406                     for(i = 0; i < s->cluster_sectors; i++) {
407                         if (i < n_start || i >= n_end) {
408                             encrypt_sectors(s, start_sect + i,
409                                             s->cluster_data,
410                                             s->cluster_data + 512, 1, 1,
411                                             &s->aes_encrypt_key);
412                             if (bdrv_pwrite(bs->file, cluster_offset + i * 512,
413                                             s->cluster_data, 512) != 512)
414                                 return -1;
415                         }
416                     }
417                 }
418             } else if (allocate == 2) {
419                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
420                     (uint64_t)compressed_size << (63 - s->cluster_bits);
421             }
422         }
423         /* update L2 table */
424         tmp = cpu_to_be64(cluster_offset);
425         l2_table[l2_index] = tmp;
426         if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
427                 &tmp, sizeof(tmp)) < 0)
428             return 0;
429     }
430     return cluster_offset;
431 }
432 
433 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
434         int64_t sector_num, int nb_sectors, int *pnum)
435 {
436     BDRVQcowState *s = bs->opaque;
437     int index_in_cluster, n;
438     uint64_t cluster_offset;
439 
440     qemu_co_mutex_lock(&s->lock);
441     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
442     qemu_co_mutex_unlock(&s->lock);
443     index_in_cluster = sector_num & (s->cluster_sectors - 1);
444     n = s->cluster_sectors - index_in_cluster;
445     if (n > nb_sectors)
446         n = nb_sectors;
447     *pnum = n;
448     if (!cluster_offset) {
449         return 0;
450     }
451     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypt_method) {
452         return BDRV_BLOCK_DATA;
453     }
454     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
455     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
456 }
457 
458 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
459                              const uint8_t *buf, int buf_size)
460 {
461     z_stream strm1, *strm = &strm1;
462     int ret, out_len;
463 
464     memset(strm, 0, sizeof(*strm));
465 
466     strm->next_in = (uint8_t *)buf;
467     strm->avail_in = buf_size;
468     strm->next_out = out_buf;
469     strm->avail_out = out_buf_size;
470 
471     ret = inflateInit2(strm, -12);
472     if (ret != Z_OK)
473         return -1;
474     ret = inflate(strm, Z_FINISH);
475     out_len = strm->next_out - out_buf;
476     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
477         out_len != out_buf_size) {
478         inflateEnd(strm);
479         return -1;
480     }
481     inflateEnd(strm);
482     return 0;
483 }
484 
485 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
486 {
487     BDRVQcowState *s = bs->opaque;
488     int ret, csize;
489     uint64_t coffset;
490 
491     coffset = cluster_offset & s->cluster_offset_mask;
492     if (s->cluster_cache_offset != coffset) {
493         csize = cluster_offset >> (63 - s->cluster_bits);
494         csize &= (s->cluster_size - 1);
495         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
496         if (ret != csize)
497             return -1;
498         if (decompress_buffer(s->cluster_cache, s->cluster_size,
499                               s->cluster_data, csize) < 0) {
500             return -1;
501         }
502         s->cluster_cache_offset = coffset;
503     }
504     return 0;
505 }
506 
507 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
508                          int nb_sectors, QEMUIOVector *qiov)
509 {
510     BDRVQcowState *s = bs->opaque;
511     int index_in_cluster;
512     int ret = 0, n;
513     uint64_t cluster_offset;
514     struct iovec hd_iov;
515     QEMUIOVector hd_qiov;
516     uint8_t *buf;
517     void *orig_buf;
518 
519     if (qiov->niov > 1) {
520         buf = orig_buf = qemu_blockalign(bs, qiov->size);
521     } else {
522         orig_buf = NULL;
523         buf = (uint8_t *)qiov->iov->iov_base;
524     }
525 
526     qemu_co_mutex_lock(&s->lock);
527 
528     while (nb_sectors != 0) {
529         /* prepare next request */
530         cluster_offset = get_cluster_offset(bs, sector_num << 9,
531                                                  0, 0, 0, 0);
532         index_in_cluster = sector_num & (s->cluster_sectors - 1);
533         n = s->cluster_sectors - index_in_cluster;
534         if (n > nb_sectors) {
535             n = nb_sectors;
536         }
537 
538         if (!cluster_offset) {
539             if (bs->backing_hd) {
540                 /* read from the base image */
541                 hd_iov.iov_base = (void *)buf;
542                 hd_iov.iov_len = n * 512;
543                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
544                 qemu_co_mutex_unlock(&s->lock);
545                 ret = bdrv_co_readv(bs->backing_hd, sector_num,
546                                     n, &hd_qiov);
547                 qemu_co_mutex_lock(&s->lock);
548                 if (ret < 0) {
549                     goto fail;
550                 }
551             } else {
552                 /* Note: in this case, no need to wait */
553                 memset(buf, 0, 512 * n);
554             }
555         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
556             /* add AIO support for compressed blocks ? */
557             if (decompress_cluster(bs, cluster_offset) < 0) {
558                 goto fail;
559             }
560             memcpy(buf,
561                    s->cluster_cache + index_in_cluster * 512, 512 * n);
562         } else {
563             if ((cluster_offset & 511) != 0) {
564                 goto fail;
565             }
566             hd_iov.iov_base = (void *)buf;
567             hd_iov.iov_len = n * 512;
568             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
569             qemu_co_mutex_unlock(&s->lock);
570             ret = bdrv_co_readv(bs->file,
571                                 (cluster_offset >> 9) + index_in_cluster,
572                                 n, &hd_qiov);
573             qemu_co_mutex_lock(&s->lock);
574             if (ret < 0) {
575                 break;
576             }
577             if (s->crypt_method) {
578                 encrypt_sectors(s, sector_num, buf, buf,
579                                 n, 0,
580                                 &s->aes_decrypt_key);
581             }
582         }
583         ret = 0;
584 
585         nb_sectors -= n;
586         sector_num += n;
587         buf += n * 512;
588     }
589 
590 done:
591     qemu_co_mutex_unlock(&s->lock);
592 
593     if (qiov->niov > 1) {
594         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
595         qemu_vfree(orig_buf);
596     }
597 
598     return ret;
599 
600 fail:
601     ret = -EIO;
602     goto done;
603 }
604 
605 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
606                           int nb_sectors, QEMUIOVector *qiov)
607 {
608     BDRVQcowState *s = bs->opaque;
609     int index_in_cluster;
610     uint64_t cluster_offset;
611     const uint8_t *src_buf;
612     int ret = 0, n;
613     uint8_t *cluster_data = NULL;
614     struct iovec hd_iov;
615     QEMUIOVector hd_qiov;
616     uint8_t *buf;
617     void *orig_buf;
618 
619     s->cluster_cache_offset = -1; /* disable compressed cache */
620 
621     if (qiov->niov > 1) {
622         buf = orig_buf = qemu_blockalign(bs, qiov->size);
623         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
624     } else {
625         orig_buf = NULL;
626         buf = (uint8_t *)qiov->iov->iov_base;
627     }
628 
629     qemu_co_mutex_lock(&s->lock);
630 
631     while (nb_sectors != 0) {
632 
633         index_in_cluster = sector_num & (s->cluster_sectors - 1);
634         n = s->cluster_sectors - index_in_cluster;
635         if (n > nb_sectors) {
636             n = nb_sectors;
637         }
638         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
639                                             index_in_cluster,
640                                             index_in_cluster + n);
641         if (!cluster_offset || (cluster_offset & 511) != 0) {
642             ret = -EIO;
643             break;
644         }
645         if (s->crypt_method) {
646             if (!cluster_data) {
647                 cluster_data = g_malloc0(s->cluster_size);
648             }
649             encrypt_sectors(s, sector_num, cluster_data, buf,
650                             n, 1, &s->aes_encrypt_key);
651             src_buf = cluster_data;
652         } else {
653             src_buf = buf;
654         }
655 
656         hd_iov.iov_base = (void *)src_buf;
657         hd_iov.iov_len = n * 512;
658         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
659         qemu_co_mutex_unlock(&s->lock);
660         ret = bdrv_co_writev(bs->file,
661                              (cluster_offset >> 9) + index_in_cluster,
662                              n, &hd_qiov);
663         qemu_co_mutex_lock(&s->lock);
664         if (ret < 0) {
665             break;
666         }
667         ret = 0;
668 
669         nb_sectors -= n;
670         sector_num += n;
671         buf += n * 512;
672     }
673     qemu_co_mutex_unlock(&s->lock);
674 
675     if (qiov->niov > 1) {
676         qemu_vfree(orig_buf);
677     }
678     g_free(cluster_data);
679 
680     return ret;
681 }
682 
683 static void qcow_close(BlockDriverState *bs)
684 {
685     BDRVQcowState *s = bs->opaque;
686 
687     g_free(s->l1_table);
688     g_free(s->l2_cache);
689     g_free(s->cluster_cache);
690     g_free(s->cluster_data);
691 
692     migrate_del_blocker(s->migration_blocker);
693     error_free(s->migration_blocker);
694 }
695 
696 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
697 {
698     int header_size, backing_filename_len, l1_size, shift, i;
699     QCowHeader header;
700     uint8_t *tmp;
701     int64_t total_size = 0;
702     char *backing_file = NULL;
703     int flags = 0;
704     Error *local_err = NULL;
705     int ret;
706     BlockDriverState *qcow_bs;
707 
708     /* Read out options */
709     total_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0) / 512;
710     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
711     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
712         flags |= BLOCK_FLAG_ENCRYPT;
713     }
714 
715     ret = bdrv_create_file(filename, opts, &local_err);
716     if (ret < 0) {
717         error_propagate(errp, local_err);
718         goto cleanup;
719     }
720 
721     qcow_bs = NULL;
722     ret = bdrv_open(&qcow_bs, filename, NULL, NULL,
723                     BDRV_O_RDWR | BDRV_O_PROTOCOL, NULL, &local_err);
724     if (ret < 0) {
725         error_propagate(errp, local_err);
726         goto cleanup;
727     }
728 
729     ret = bdrv_truncate(qcow_bs, 0);
730     if (ret < 0) {
731         goto exit;
732     }
733 
734     memset(&header, 0, sizeof(header));
735     header.magic = cpu_to_be32(QCOW_MAGIC);
736     header.version = cpu_to_be32(QCOW_VERSION);
737     header.size = cpu_to_be64(total_size * 512);
738     header_size = sizeof(header);
739     backing_filename_len = 0;
740     if (backing_file) {
741         if (strcmp(backing_file, "fat:")) {
742             header.backing_file_offset = cpu_to_be64(header_size);
743             backing_filename_len = strlen(backing_file);
744             header.backing_file_size = cpu_to_be32(backing_filename_len);
745             header_size += backing_filename_len;
746         } else {
747             /* special backing file for vvfat */
748             backing_file = NULL;
749         }
750         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
751                                     unmodified sectors */
752         header.l2_bits = 12; /* 32 KB L2 tables */
753     } else {
754         header.cluster_bits = 12; /* 4 KB clusters */
755         header.l2_bits = 9; /* 4 KB L2 tables */
756     }
757     header_size = (header_size + 7) & ~7;
758     shift = header.cluster_bits + header.l2_bits;
759     l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
760 
761     header.l1_table_offset = cpu_to_be64(header_size);
762     if (flags & BLOCK_FLAG_ENCRYPT) {
763         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
764     } else {
765         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
766     }
767 
768     /* write all the data */
769     ret = bdrv_pwrite(qcow_bs, 0, &header, sizeof(header));
770     if (ret != sizeof(header)) {
771         goto exit;
772     }
773 
774     if (backing_file) {
775         ret = bdrv_pwrite(qcow_bs, sizeof(header),
776             backing_file, backing_filename_len);
777         if (ret != backing_filename_len) {
778             goto exit;
779         }
780     }
781 
782     tmp = g_malloc0(BDRV_SECTOR_SIZE);
783     for (i = 0; i < ((sizeof(uint64_t)*l1_size + BDRV_SECTOR_SIZE - 1)/
784         BDRV_SECTOR_SIZE); i++) {
785         ret = bdrv_pwrite(qcow_bs, header_size +
786             BDRV_SECTOR_SIZE*i, tmp, BDRV_SECTOR_SIZE);
787         if (ret != BDRV_SECTOR_SIZE) {
788             g_free(tmp);
789             goto exit;
790         }
791     }
792 
793     g_free(tmp);
794     ret = 0;
795 exit:
796     bdrv_unref(qcow_bs);
797 cleanup:
798     g_free(backing_file);
799     return ret;
800 }
801 
802 static int qcow_make_empty(BlockDriverState *bs)
803 {
804     BDRVQcowState *s = bs->opaque;
805     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
806     int ret;
807 
808     memset(s->l1_table, 0, l1_length);
809     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
810             l1_length) < 0)
811         return -1;
812     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
813     if (ret < 0)
814         return ret;
815 
816     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
817     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
818     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
819 
820     return 0;
821 }
822 
823 /* XXX: put compressed sectors first, then all the cluster aligned
824    tables to avoid losing bytes in alignment */
825 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
826                                  const uint8_t *buf, int nb_sectors)
827 {
828     BDRVQcowState *s = bs->opaque;
829     z_stream strm;
830     int ret, out_len;
831     uint8_t *out_buf;
832     uint64_t cluster_offset;
833 
834     if (nb_sectors != s->cluster_sectors) {
835         ret = -EINVAL;
836 
837         /* Zero-pad last write if image size is not cluster aligned */
838         if (sector_num + nb_sectors == bs->total_sectors &&
839             nb_sectors < s->cluster_sectors) {
840             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
841             memset(pad_buf, 0, s->cluster_size);
842             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
843             ret = qcow_write_compressed(bs, sector_num,
844                                         pad_buf, s->cluster_sectors);
845             qemu_vfree(pad_buf);
846         }
847         return ret;
848     }
849 
850     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
851 
852     /* best compression, small window, no zlib header */
853     memset(&strm, 0, sizeof(strm));
854     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
855                        Z_DEFLATED, -12,
856                        9, Z_DEFAULT_STRATEGY);
857     if (ret != 0) {
858         ret = -EINVAL;
859         goto fail;
860     }
861 
862     strm.avail_in = s->cluster_size;
863     strm.next_in = (uint8_t *)buf;
864     strm.avail_out = s->cluster_size;
865     strm.next_out = out_buf;
866 
867     ret = deflate(&strm, Z_FINISH);
868     if (ret != Z_STREAM_END && ret != Z_OK) {
869         deflateEnd(&strm);
870         ret = -EINVAL;
871         goto fail;
872     }
873     out_len = strm.next_out - out_buf;
874 
875     deflateEnd(&strm);
876 
877     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
878         /* could not compress: write normal cluster */
879         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
880         if (ret < 0) {
881             goto fail;
882         }
883     } else {
884         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
885                                             out_len, 0, 0);
886         if (cluster_offset == 0) {
887             ret = -EIO;
888             goto fail;
889         }
890 
891         cluster_offset &= s->cluster_offset_mask;
892         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
893         if (ret < 0) {
894             goto fail;
895         }
896     }
897 
898     ret = 0;
899 fail:
900     g_free(out_buf);
901     return ret;
902 }
903 
904 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
905 {
906     BDRVQcowState *s = bs->opaque;
907     bdi->cluster_size = s->cluster_size;
908     return 0;
909 }
910 
911 static QemuOptsList qcow_create_opts = {
912     .name = "qcow-create-opts",
913     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
914     .desc = {
915         {
916             .name = BLOCK_OPT_SIZE,
917             .type = QEMU_OPT_SIZE,
918             .help = "Virtual disk size"
919         },
920         {
921             .name = BLOCK_OPT_BACKING_FILE,
922             .type = QEMU_OPT_STRING,
923             .help = "File name of a base image"
924         },
925         {
926             .name = BLOCK_OPT_ENCRYPT,
927             .type = QEMU_OPT_BOOL,
928             .help = "Encrypt the image",
929             .def_value_str = "off"
930         },
931         { /* end of list */ }
932     }
933 };
934 
935 static BlockDriver bdrv_qcow = {
936     .format_name	= "qcow",
937     .instance_size	= sizeof(BDRVQcowState),
938     .bdrv_probe		= qcow_probe,
939     .bdrv_open		= qcow_open,
940     .bdrv_close		= qcow_close,
941     .bdrv_reopen_prepare    = qcow_reopen_prepare,
942     .bdrv_create            = qcow_create,
943     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
944 
945     .bdrv_co_readv          = qcow_co_readv,
946     .bdrv_co_writev         = qcow_co_writev,
947     .bdrv_co_get_block_status   = qcow_co_get_block_status,
948 
949     .bdrv_set_key           = qcow_set_key,
950     .bdrv_make_empty        = qcow_make_empty,
951     .bdrv_write_compressed  = qcow_write_compressed,
952     .bdrv_get_info          = qcow_get_info,
953 
954     .create_opts            = &qcow_create_opts,
955 };
956 
957 static void bdrv_qcow_init(void)
958 {
959     bdrv_register(&bdrv_qcow);
960 }
961 
962 block_init(bdrv_qcow_init);
963