xref: /openbmc/qemu/block/qcow.c (revision bfb27e60)
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_try_new(uint64_t, s->l1_size);
186     if (s->l1_table == NULL) {
187         error_setg(errp, "Could not allocate memory for L1 table");
188         ret = -ENOMEM;
189         goto fail;
190     }
191 
192     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
193                s->l1_size * sizeof(uint64_t));
194     if (ret < 0) {
195         goto fail;
196     }
197 
198     for(i = 0;i < s->l1_size; i++) {
199         be64_to_cpus(&s->l1_table[i]);
200     }
201 
202     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
203     s->l2_cache =
204         qemu_try_blockalign(bs->file,
205                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
206     if (s->l2_cache == NULL) {
207         error_setg(errp, "Could not allocate L2 table cache");
208         ret = -ENOMEM;
209         goto fail;
210     }
211     s->cluster_cache = g_malloc(s->cluster_size);
212     s->cluster_data = g_malloc(s->cluster_size);
213     s->cluster_cache_offset = -1;
214 
215     /* read the backing file name */
216     if (header.backing_file_offset != 0) {
217         len = header.backing_file_size;
218         if (len > 1023) {
219             error_setg(errp, "Backing file name too long");
220             ret = -EINVAL;
221             goto fail;
222         }
223         ret = bdrv_pread(bs->file, header.backing_file_offset,
224                    bs->backing_file, len);
225         if (ret < 0) {
226             goto fail;
227         }
228         bs->backing_file[len] = '\0';
229     }
230 
231     /* Disable migration when qcow images are used */
232     error_set(&s->migration_blocker,
233               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
234               "qcow", bs->device_name, "live migration");
235     migrate_add_blocker(s->migration_blocker);
236 
237     qemu_co_mutex_init(&s->lock);
238     return 0;
239 
240  fail:
241     g_free(s->l1_table);
242     qemu_vfree(s->l2_cache);
243     g_free(s->cluster_cache);
244     g_free(s->cluster_data);
245     return ret;
246 }
247 
248 
249 /* We have nothing to do for QCOW reopen, stubs just return
250  * success */
251 static int qcow_reopen_prepare(BDRVReopenState *state,
252                                BlockReopenQueue *queue, Error **errp)
253 {
254     return 0;
255 }
256 
257 static int qcow_set_key(BlockDriverState *bs, const char *key)
258 {
259     BDRVQcowState *s = bs->opaque;
260     uint8_t keybuf[16];
261     int len, i;
262 
263     memset(keybuf, 0, 16);
264     len = strlen(key);
265     if (len > 16)
266         len = 16;
267     /* XXX: we could compress the chars to 7 bits to increase
268        entropy */
269     for(i = 0;i < len;i++) {
270         keybuf[i] = key[i];
271     }
272     s->crypt_method = s->crypt_method_header;
273 
274     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
275         return -1;
276     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
277         return -1;
278     return 0;
279 }
280 
281 /* The crypt function is compatible with the linux cryptoloop
282    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
283    supported */
284 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
285                             uint8_t *out_buf, const uint8_t *in_buf,
286                             int nb_sectors, int enc,
287                             const AES_KEY *key)
288 {
289     union {
290         uint64_t ll[2];
291         uint8_t b[16];
292     } ivec;
293     int i;
294 
295     for(i = 0; i < nb_sectors; i++) {
296         ivec.ll[0] = cpu_to_le64(sector_num);
297         ivec.ll[1] = 0;
298         AES_cbc_encrypt(in_buf, out_buf, 512, key,
299                         ivec.b, enc);
300         sector_num++;
301         in_buf += 512;
302         out_buf += 512;
303     }
304 }
305 
306 /* 'allocate' is:
307  *
308  * 0 to not allocate.
309  *
310  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
311  * 'n_end')
312  *
313  * 2 to allocate a compressed cluster of size
314  * 'compressed_size'. 'compressed_size' must be > 0 and <
315  * cluster_size
316  *
317  * return 0 if not allocated.
318  */
319 static uint64_t get_cluster_offset(BlockDriverState *bs,
320                                    uint64_t offset, int allocate,
321                                    int compressed_size,
322                                    int n_start, int n_end)
323 {
324     BDRVQcowState *s = bs->opaque;
325     int min_index, i, j, l1_index, l2_index;
326     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
327     uint32_t min_count;
328     int new_l2_table;
329 
330     l1_index = offset >> (s->l2_bits + s->cluster_bits);
331     l2_offset = s->l1_table[l1_index];
332     new_l2_table = 0;
333     if (!l2_offset) {
334         if (!allocate)
335             return 0;
336         /* allocate a new l2 entry */
337         l2_offset = bdrv_getlength(bs->file);
338         /* round to cluster size */
339         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
340         /* update the L1 entry */
341         s->l1_table[l1_index] = l2_offset;
342         tmp = cpu_to_be64(l2_offset);
343         if (bdrv_pwrite_sync(bs->file,
344                 s->l1_table_offset + l1_index * sizeof(tmp),
345                 &tmp, sizeof(tmp)) < 0)
346             return 0;
347         new_l2_table = 1;
348     }
349     for(i = 0; i < L2_CACHE_SIZE; i++) {
350         if (l2_offset == s->l2_cache_offsets[i]) {
351             /* increment the hit count */
352             if (++s->l2_cache_counts[i] == 0xffffffff) {
353                 for(j = 0; j < L2_CACHE_SIZE; j++) {
354                     s->l2_cache_counts[j] >>= 1;
355                 }
356             }
357             l2_table = s->l2_cache + (i << s->l2_bits);
358             goto found;
359         }
360     }
361     /* not found: load a new entry in the least used one */
362     min_index = 0;
363     min_count = 0xffffffff;
364     for(i = 0; i < L2_CACHE_SIZE; i++) {
365         if (s->l2_cache_counts[i] < min_count) {
366             min_count = s->l2_cache_counts[i];
367             min_index = i;
368         }
369     }
370     l2_table = s->l2_cache + (min_index << s->l2_bits);
371     if (new_l2_table) {
372         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
373         if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
374                 s->l2_size * sizeof(uint64_t)) < 0)
375             return 0;
376     } else {
377         if (bdrv_pread(bs->file, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
378             s->l2_size * sizeof(uint64_t))
379             return 0;
380     }
381     s->l2_cache_offsets[min_index] = l2_offset;
382     s->l2_cache_counts[min_index] = 1;
383  found:
384     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
385     cluster_offset = be64_to_cpu(l2_table[l2_index]);
386     if (!cluster_offset ||
387         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
388         if (!allocate)
389             return 0;
390         /* allocate a new cluster */
391         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
392             (n_end - n_start) < s->cluster_sectors) {
393             /* if the cluster is already compressed, we must
394                decompress it in the case it is not completely
395                overwritten */
396             if (decompress_cluster(bs, cluster_offset) < 0)
397                 return 0;
398             cluster_offset = bdrv_getlength(bs->file);
399             cluster_offset = (cluster_offset + s->cluster_size - 1) &
400                 ~(s->cluster_size - 1);
401             /* write the cluster content */
402             if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache, s->cluster_size) !=
403                 s->cluster_size)
404                 return -1;
405         } else {
406             cluster_offset = bdrv_getlength(bs->file);
407             if (allocate == 1) {
408                 /* round to cluster size */
409                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
410                     ~(s->cluster_size - 1);
411                 bdrv_truncate(bs->file, cluster_offset + s->cluster_size);
412                 /* if encrypted, we must initialize the cluster
413                    content which won't be written */
414                 if (s->crypt_method &&
415                     (n_end - n_start) < s->cluster_sectors) {
416                     uint64_t start_sect;
417                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
418                     memset(s->cluster_data + 512, 0x00, 512);
419                     for(i = 0; i < s->cluster_sectors; i++) {
420                         if (i < n_start || i >= n_end) {
421                             encrypt_sectors(s, start_sect + i,
422                                             s->cluster_data,
423                                             s->cluster_data + 512, 1, 1,
424                                             &s->aes_encrypt_key);
425                             if (bdrv_pwrite(bs->file, cluster_offset + i * 512,
426                                             s->cluster_data, 512) != 512)
427                                 return -1;
428                         }
429                     }
430                 }
431             } else if (allocate == 2) {
432                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
433                     (uint64_t)compressed_size << (63 - s->cluster_bits);
434             }
435         }
436         /* update L2 table */
437         tmp = cpu_to_be64(cluster_offset);
438         l2_table[l2_index] = tmp;
439         if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
440                 &tmp, sizeof(tmp)) < 0)
441             return 0;
442     }
443     return cluster_offset;
444 }
445 
446 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
447         int64_t sector_num, int nb_sectors, int *pnum)
448 {
449     BDRVQcowState *s = bs->opaque;
450     int index_in_cluster, n;
451     uint64_t cluster_offset;
452 
453     qemu_co_mutex_lock(&s->lock);
454     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
455     qemu_co_mutex_unlock(&s->lock);
456     index_in_cluster = sector_num & (s->cluster_sectors - 1);
457     n = s->cluster_sectors - index_in_cluster;
458     if (n > nb_sectors)
459         n = nb_sectors;
460     *pnum = n;
461     if (!cluster_offset) {
462         return 0;
463     }
464     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypt_method) {
465         return BDRV_BLOCK_DATA;
466     }
467     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
468     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
469 }
470 
471 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
472                              const uint8_t *buf, int buf_size)
473 {
474     z_stream strm1, *strm = &strm1;
475     int ret, out_len;
476 
477     memset(strm, 0, sizeof(*strm));
478 
479     strm->next_in = (uint8_t *)buf;
480     strm->avail_in = buf_size;
481     strm->next_out = out_buf;
482     strm->avail_out = out_buf_size;
483 
484     ret = inflateInit2(strm, -12);
485     if (ret != Z_OK)
486         return -1;
487     ret = inflate(strm, Z_FINISH);
488     out_len = strm->next_out - out_buf;
489     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
490         out_len != out_buf_size) {
491         inflateEnd(strm);
492         return -1;
493     }
494     inflateEnd(strm);
495     return 0;
496 }
497 
498 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
499 {
500     BDRVQcowState *s = bs->opaque;
501     int ret, csize;
502     uint64_t coffset;
503 
504     coffset = cluster_offset & s->cluster_offset_mask;
505     if (s->cluster_cache_offset != coffset) {
506         csize = cluster_offset >> (63 - s->cluster_bits);
507         csize &= (s->cluster_size - 1);
508         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
509         if (ret != csize)
510             return -1;
511         if (decompress_buffer(s->cluster_cache, s->cluster_size,
512                               s->cluster_data, csize) < 0) {
513             return -1;
514         }
515         s->cluster_cache_offset = coffset;
516     }
517     return 0;
518 }
519 
520 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
521                          int nb_sectors, QEMUIOVector *qiov)
522 {
523     BDRVQcowState *s = bs->opaque;
524     int index_in_cluster;
525     int ret = 0, n;
526     uint64_t cluster_offset;
527     struct iovec hd_iov;
528     QEMUIOVector hd_qiov;
529     uint8_t *buf;
530     void *orig_buf;
531 
532     if (qiov->niov > 1) {
533         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
534         if (buf == NULL) {
535             return -ENOMEM;
536         }
537     } else {
538         orig_buf = NULL;
539         buf = (uint8_t *)qiov->iov->iov_base;
540     }
541 
542     qemu_co_mutex_lock(&s->lock);
543 
544     while (nb_sectors != 0) {
545         /* prepare next request */
546         cluster_offset = get_cluster_offset(bs, sector_num << 9,
547                                                  0, 0, 0, 0);
548         index_in_cluster = sector_num & (s->cluster_sectors - 1);
549         n = s->cluster_sectors - index_in_cluster;
550         if (n > nb_sectors) {
551             n = nb_sectors;
552         }
553 
554         if (!cluster_offset) {
555             if (bs->backing_hd) {
556                 /* read from the base image */
557                 hd_iov.iov_base = (void *)buf;
558                 hd_iov.iov_len = n * 512;
559                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
560                 qemu_co_mutex_unlock(&s->lock);
561                 ret = bdrv_co_readv(bs->backing_hd, sector_num,
562                                     n, &hd_qiov);
563                 qemu_co_mutex_lock(&s->lock);
564                 if (ret < 0) {
565                     goto fail;
566                 }
567             } else {
568                 /* Note: in this case, no need to wait */
569                 memset(buf, 0, 512 * n);
570             }
571         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
572             /* add AIO support for compressed blocks ? */
573             if (decompress_cluster(bs, cluster_offset) < 0) {
574                 goto fail;
575             }
576             memcpy(buf,
577                    s->cluster_cache + index_in_cluster * 512, 512 * n);
578         } else {
579             if ((cluster_offset & 511) != 0) {
580                 goto fail;
581             }
582             hd_iov.iov_base = (void *)buf;
583             hd_iov.iov_len = n * 512;
584             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
585             qemu_co_mutex_unlock(&s->lock);
586             ret = bdrv_co_readv(bs->file,
587                                 (cluster_offset >> 9) + index_in_cluster,
588                                 n, &hd_qiov);
589             qemu_co_mutex_lock(&s->lock);
590             if (ret < 0) {
591                 break;
592             }
593             if (s->crypt_method) {
594                 encrypt_sectors(s, sector_num, buf, buf,
595                                 n, 0,
596                                 &s->aes_decrypt_key);
597             }
598         }
599         ret = 0;
600 
601         nb_sectors -= n;
602         sector_num += n;
603         buf += n * 512;
604     }
605 
606 done:
607     qemu_co_mutex_unlock(&s->lock);
608 
609     if (qiov->niov > 1) {
610         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
611         qemu_vfree(orig_buf);
612     }
613 
614     return ret;
615 
616 fail:
617     ret = -EIO;
618     goto done;
619 }
620 
621 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
622                           int nb_sectors, QEMUIOVector *qiov)
623 {
624     BDRVQcowState *s = bs->opaque;
625     int index_in_cluster;
626     uint64_t cluster_offset;
627     const uint8_t *src_buf;
628     int ret = 0, n;
629     uint8_t *cluster_data = NULL;
630     struct iovec hd_iov;
631     QEMUIOVector hd_qiov;
632     uint8_t *buf;
633     void *orig_buf;
634 
635     s->cluster_cache_offset = -1; /* disable compressed cache */
636 
637     if (qiov->niov > 1) {
638         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
639         if (buf == NULL) {
640             return -ENOMEM;
641         }
642         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
643     } else {
644         orig_buf = NULL;
645         buf = (uint8_t *)qiov->iov->iov_base;
646     }
647 
648     qemu_co_mutex_lock(&s->lock);
649 
650     while (nb_sectors != 0) {
651 
652         index_in_cluster = sector_num & (s->cluster_sectors - 1);
653         n = s->cluster_sectors - index_in_cluster;
654         if (n > nb_sectors) {
655             n = nb_sectors;
656         }
657         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
658                                             index_in_cluster,
659                                             index_in_cluster + n);
660         if (!cluster_offset || (cluster_offset & 511) != 0) {
661             ret = -EIO;
662             break;
663         }
664         if (s->crypt_method) {
665             if (!cluster_data) {
666                 cluster_data = g_malloc0(s->cluster_size);
667             }
668             encrypt_sectors(s, sector_num, cluster_data, buf,
669                             n, 1, &s->aes_encrypt_key);
670             src_buf = cluster_data;
671         } else {
672             src_buf = buf;
673         }
674 
675         hd_iov.iov_base = (void *)src_buf;
676         hd_iov.iov_len = n * 512;
677         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
678         qemu_co_mutex_unlock(&s->lock);
679         ret = bdrv_co_writev(bs->file,
680                              (cluster_offset >> 9) + index_in_cluster,
681                              n, &hd_qiov);
682         qemu_co_mutex_lock(&s->lock);
683         if (ret < 0) {
684             break;
685         }
686         ret = 0;
687 
688         nb_sectors -= n;
689         sector_num += n;
690         buf += n * 512;
691     }
692     qemu_co_mutex_unlock(&s->lock);
693 
694     if (qiov->niov > 1) {
695         qemu_vfree(orig_buf);
696     }
697     g_free(cluster_data);
698 
699     return ret;
700 }
701 
702 static void qcow_close(BlockDriverState *bs)
703 {
704     BDRVQcowState *s = bs->opaque;
705 
706     g_free(s->l1_table);
707     qemu_vfree(s->l2_cache);
708     g_free(s->cluster_cache);
709     g_free(s->cluster_data);
710 
711     migrate_del_blocker(s->migration_blocker);
712     error_free(s->migration_blocker);
713 }
714 
715 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
716 {
717     int header_size, backing_filename_len, l1_size, shift, i;
718     QCowHeader header;
719     uint8_t *tmp;
720     int64_t total_size = 0;
721     char *backing_file = NULL;
722     int flags = 0;
723     Error *local_err = NULL;
724     int ret;
725     BlockDriverState *qcow_bs;
726 
727     /* Read out options */
728     total_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0) / 512;
729     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
730     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
731         flags |= BLOCK_FLAG_ENCRYPT;
732     }
733 
734     ret = bdrv_create_file(filename, opts, &local_err);
735     if (ret < 0) {
736         error_propagate(errp, local_err);
737         goto cleanup;
738     }
739 
740     qcow_bs = NULL;
741     ret = bdrv_open(&qcow_bs, filename, NULL, NULL,
742                     BDRV_O_RDWR | BDRV_O_PROTOCOL, NULL, &local_err);
743     if (ret < 0) {
744         error_propagate(errp, local_err);
745         goto cleanup;
746     }
747 
748     ret = bdrv_truncate(qcow_bs, 0);
749     if (ret < 0) {
750         goto exit;
751     }
752 
753     memset(&header, 0, sizeof(header));
754     header.magic = cpu_to_be32(QCOW_MAGIC);
755     header.version = cpu_to_be32(QCOW_VERSION);
756     header.size = cpu_to_be64(total_size * 512);
757     header_size = sizeof(header);
758     backing_filename_len = 0;
759     if (backing_file) {
760         if (strcmp(backing_file, "fat:")) {
761             header.backing_file_offset = cpu_to_be64(header_size);
762             backing_filename_len = strlen(backing_file);
763             header.backing_file_size = cpu_to_be32(backing_filename_len);
764             header_size += backing_filename_len;
765         } else {
766             /* special backing file for vvfat */
767             backing_file = NULL;
768         }
769         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
770                                     unmodified sectors */
771         header.l2_bits = 12; /* 32 KB L2 tables */
772     } else {
773         header.cluster_bits = 12; /* 4 KB clusters */
774         header.l2_bits = 9; /* 4 KB L2 tables */
775     }
776     header_size = (header_size + 7) & ~7;
777     shift = header.cluster_bits + header.l2_bits;
778     l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
779 
780     header.l1_table_offset = cpu_to_be64(header_size);
781     if (flags & BLOCK_FLAG_ENCRYPT) {
782         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
783     } else {
784         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
785     }
786 
787     /* write all the data */
788     ret = bdrv_pwrite(qcow_bs, 0, &header, sizeof(header));
789     if (ret != sizeof(header)) {
790         goto exit;
791     }
792 
793     if (backing_file) {
794         ret = bdrv_pwrite(qcow_bs, sizeof(header),
795             backing_file, backing_filename_len);
796         if (ret != backing_filename_len) {
797             goto exit;
798         }
799     }
800 
801     tmp = g_malloc0(BDRV_SECTOR_SIZE);
802     for (i = 0; i < ((sizeof(uint64_t)*l1_size + BDRV_SECTOR_SIZE - 1)/
803         BDRV_SECTOR_SIZE); i++) {
804         ret = bdrv_pwrite(qcow_bs, header_size +
805             BDRV_SECTOR_SIZE*i, tmp, BDRV_SECTOR_SIZE);
806         if (ret != BDRV_SECTOR_SIZE) {
807             g_free(tmp);
808             goto exit;
809         }
810     }
811 
812     g_free(tmp);
813     ret = 0;
814 exit:
815     bdrv_unref(qcow_bs);
816 cleanup:
817     g_free(backing_file);
818     return ret;
819 }
820 
821 static int qcow_make_empty(BlockDriverState *bs)
822 {
823     BDRVQcowState *s = bs->opaque;
824     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
825     int ret;
826 
827     memset(s->l1_table, 0, l1_length);
828     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
829             l1_length) < 0)
830         return -1;
831     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
832     if (ret < 0)
833         return ret;
834 
835     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
836     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
837     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
838 
839     return 0;
840 }
841 
842 /* XXX: put compressed sectors first, then all the cluster aligned
843    tables to avoid losing bytes in alignment */
844 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
845                                  const uint8_t *buf, int nb_sectors)
846 {
847     BDRVQcowState *s = bs->opaque;
848     z_stream strm;
849     int ret, out_len;
850     uint8_t *out_buf;
851     uint64_t cluster_offset;
852 
853     if (nb_sectors != s->cluster_sectors) {
854         ret = -EINVAL;
855 
856         /* Zero-pad last write if image size is not cluster aligned */
857         if (sector_num + nb_sectors == bs->total_sectors &&
858             nb_sectors < s->cluster_sectors) {
859             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
860             memset(pad_buf, 0, s->cluster_size);
861             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
862             ret = qcow_write_compressed(bs, sector_num,
863                                         pad_buf, s->cluster_sectors);
864             qemu_vfree(pad_buf);
865         }
866         return ret;
867     }
868 
869     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
870 
871     /* best compression, small window, no zlib header */
872     memset(&strm, 0, sizeof(strm));
873     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
874                        Z_DEFLATED, -12,
875                        9, Z_DEFAULT_STRATEGY);
876     if (ret != 0) {
877         ret = -EINVAL;
878         goto fail;
879     }
880 
881     strm.avail_in = s->cluster_size;
882     strm.next_in = (uint8_t *)buf;
883     strm.avail_out = s->cluster_size;
884     strm.next_out = out_buf;
885 
886     ret = deflate(&strm, Z_FINISH);
887     if (ret != Z_STREAM_END && ret != Z_OK) {
888         deflateEnd(&strm);
889         ret = -EINVAL;
890         goto fail;
891     }
892     out_len = strm.next_out - out_buf;
893 
894     deflateEnd(&strm);
895 
896     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
897         /* could not compress: write normal cluster */
898         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
899         if (ret < 0) {
900             goto fail;
901         }
902     } else {
903         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
904                                             out_len, 0, 0);
905         if (cluster_offset == 0) {
906             ret = -EIO;
907             goto fail;
908         }
909 
910         cluster_offset &= s->cluster_offset_mask;
911         ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
912         if (ret < 0) {
913             goto fail;
914         }
915     }
916 
917     ret = 0;
918 fail:
919     g_free(out_buf);
920     return ret;
921 }
922 
923 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
924 {
925     BDRVQcowState *s = bs->opaque;
926     bdi->cluster_size = s->cluster_size;
927     return 0;
928 }
929 
930 static QemuOptsList qcow_create_opts = {
931     .name = "qcow-create-opts",
932     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
933     .desc = {
934         {
935             .name = BLOCK_OPT_SIZE,
936             .type = QEMU_OPT_SIZE,
937             .help = "Virtual disk size"
938         },
939         {
940             .name = BLOCK_OPT_BACKING_FILE,
941             .type = QEMU_OPT_STRING,
942             .help = "File name of a base image"
943         },
944         {
945             .name = BLOCK_OPT_ENCRYPT,
946             .type = QEMU_OPT_BOOL,
947             .help = "Encrypt the image",
948             .def_value_str = "off"
949         },
950         { /* end of list */ }
951     }
952 };
953 
954 static BlockDriver bdrv_qcow = {
955     .format_name	= "qcow",
956     .instance_size	= sizeof(BDRVQcowState),
957     .bdrv_probe		= qcow_probe,
958     .bdrv_open		= qcow_open,
959     .bdrv_close		= qcow_close,
960     .bdrv_reopen_prepare    = qcow_reopen_prepare,
961     .bdrv_create            = qcow_create,
962     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
963     .supports_backing       = true,
964 
965     .bdrv_co_readv          = qcow_co_readv,
966     .bdrv_co_writev         = qcow_co_writev,
967     .bdrv_co_get_block_status   = qcow_co_get_block_status,
968 
969     .bdrv_set_key           = qcow_set_key,
970     .bdrv_make_empty        = qcow_make_empty,
971     .bdrv_write_compressed  = qcow_write_compressed,
972     .bdrv_get_info          = qcow_get_info,
973 
974     .create_opts            = &qcow_create_opts,
975 };
976 
977 static void bdrv_qcow_init(void)
978 {
979     bdrv_register(&bdrv_qcow);
980 }
981 
982 block_init(bdrv_qcow_init);
983