xref: /openbmc/qemu/block/qcow.c (revision 33577b47)
1 /*
2  * Block driver for the QCOW format
3  *
4  * Copyright (c) 2004-2006 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 #include "qemu/osdep.h"
25 #include "qemu-common.h"
26 #include "block/block_int.h"
27 #include "sysemu/block-backend.h"
28 #include "qemu/module.h"
29 #include <zlib.h>
30 #include "qapi/qmp/qerror.h"
31 #include "crypto/cipher.h"
32 #include "migration/migration.h"
33 
34 /**************************************************************/
35 /* QEMU COW block driver with compression and encryption support */
36 
37 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
38 #define QCOW_VERSION 1
39 
40 #define QCOW_CRYPT_NONE 0
41 #define QCOW_CRYPT_AES  1
42 
43 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
44 
45 typedef struct QCowHeader {
46     uint32_t magic;
47     uint32_t version;
48     uint64_t backing_file_offset;
49     uint32_t backing_file_size;
50     uint32_t mtime;
51     uint64_t size; /* in bytes */
52     uint8_t cluster_bits;
53     uint8_t l2_bits;
54     uint16_t padding;
55     uint32_t crypt_method;
56     uint64_t l1_table_offset;
57 } QEMU_PACKED QCowHeader;
58 
59 #define L2_CACHE_SIZE 16
60 
61 typedef struct BDRVQcowState {
62     int cluster_bits;
63     int cluster_size;
64     int cluster_sectors;
65     int l2_bits;
66     int l2_size;
67     unsigned int l1_size;
68     uint64_t cluster_offset_mask;
69     uint64_t l1_table_offset;
70     uint64_t *l1_table;
71     uint64_t *l2_cache;
72     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
73     uint32_t l2_cache_counts[L2_CACHE_SIZE];
74     uint8_t *cluster_cache;
75     uint8_t *cluster_data;
76     uint64_t cluster_cache_offset;
77     QCryptoCipher *cipher; /* NULL if no key yet */
78     uint32_t crypt_method_header;
79     CoMutex lock;
80     Error *migration_blocker;
81 } BDRVQcowState;
82 
83 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
84 
85 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
86 {
87     const QCowHeader *cow_header = (const void *)buf;
88 
89     if (buf_size >= sizeof(QCowHeader) &&
90         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
91         be32_to_cpu(cow_header->version) == QCOW_VERSION)
92         return 100;
93     else
94         return 0;
95 }
96 
97 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
98                      Error **errp)
99 {
100     BDRVQcowState *s = bs->opaque;
101     unsigned int len, i, shift;
102     int ret;
103     QCowHeader header;
104 
105     ret = bdrv_pread(bs->file->bs, 0, &header, sizeof(header));
106     if (ret < 0) {
107         goto fail;
108     }
109     be32_to_cpus(&header.magic);
110     be32_to_cpus(&header.version);
111     be64_to_cpus(&header.backing_file_offset);
112     be32_to_cpus(&header.backing_file_size);
113     be32_to_cpus(&header.mtime);
114     be64_to_cpus(&header.size);
115     be32_to_cpus(&header.crypt_method);
116     be64_to_cpus(&header.l1_table_offset);
117 
118     if (header.magic != QCOW_MAGIC) {
119         error_setg(errp, "Image not in qcow format");
120         ret = -EINVAL;
121         goto fail;
122     }
123     if (header.version != QCOW_VERSION) {
124         char version[64];
125         snprintf(version, sizeof(version), "QCOW version %" PRIu32,
126                  header.version);
127         error_setg(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
128                    bdrv_get_device_or_node_name(bs), "qcow", version);
129         ret = -ENOTSUP;
130         goto fail;
131     }
132 
133     if (header.size <= 1) {
134         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
135         ret = -EINVAL;
136         goto fail;
137     }
138     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
139         error_setg(errp, "Cluster size must be between 512 and 64k");
140         ret = -EINVAL;
141         goto fail;
142     }
143 
144     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
145      * so bytes = num_entries << 3. */
146     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
147         error_setg(errp, "L2 table size must be between 512 and 64k");
148         ret = -EINVAL;
149         goto fail;
150     }
151 
152     if (header.crypt_method > QCOW_CRYPT_AES) {
153         error_setg(errp, "invalid encryption method in qcow header");
154         ret = -EINVAL;
155         goto fail;
156     }
157     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) {
158         error_setg(errp, "AES cipher not available");
159         ret = -EINVAL;
160         goto fail;
161     }
162     s->crypt_method_header = header.crypt_method;
163     if (s->crypt_method_header) {
164         bs->encrypted = 1;
165     }
166     s->cluster_bits = header.cluster_bits;
167     s->cluster_size = 1 << s->cluster_bits;
168     s->cluster_sectors = 1 << (s->cluster_bits - 9);
169     s->l2_bits = header.l2_bits;
170     s->l2_size = 1 << s->l2_bits;
171     bs->total_sectors = header.size / 512;
172     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
173 
174     /* read the level 1 table */
175     shift = s->cluster_bits + s->l2_bits;
176     if (header.size > UINT64_MAX - (1LL << shift)) {
177         error_setg(errp, "Image too large");
178         ret = -EINVAL;
179         goto fail;
180     } else {
181         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
182         if (l1_size > INT_MAX / sizeof(uint64_t)) {
183             error_setg(errp, "Image too large");
184             ret = -EINVAL;
185             goto fail;
186         }
187         s->l1_size = l1_size;
188     }
189 
190     s->l1_table_offset = header.l1_table_offset;
191     s->l1_table = g_try_new(uint64_t, s->l1_size);
192     if (s->l1_table == NULL) {
193         error_setg(errp, "Could not allocate memory for L1 table");
194         ret = -ENOMEM;
195         goto fail;
196     }
197 
198     ret = bdrv_pread(bs->file->bs, s->l1_table_offset, s->l1_table,
199                s->l1_size * sizeof(uint64_t));
200     if (ret < 0) {
201         goto fail;
202     }
203 
204     for(i = 0;i < s->l1_size; i++) {
205         be64_to_cpus(&s->l1_table[i]);
206     }
207 
208     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
209     s->l2_cache =
210         qemu_try_blockalign(bs->file->bs,
211                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
212     if (s->l2_cache == NULL) {
213         error_setg(errp, "Could not allocate L2 table cache");
214         ret = -ENOMEM;
215         goto fail;
216     }
217     s->cluster_cache = g_malloc(s->cluster_size);
218     s->cluster_data = g_malloc(s->cluster_size);
219     s->cluster_cache_offset = -1;
220 
221     /* read the backing file name */
222     if (header.backing_file_offset != 0) {
223         len = header.backing_file_size;
224         if (len > 1023 || len >= sizeof(bs->backing_file)) {
225             error_setg(errp, "Backing file name too long");
226             ret = -EINVAL;
227             goto fail;
228         }
229         ret = bdrv_pread(bs->file->bs, header.backing_file_offset,
230                    bs->backing_file, len);
231         if (ret < 0) {
232             goto fail;
233         }
234         bs->backing_file[len] = '\0';
235     }
236 
237     /* Disable migration when qcow images are used */
238     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
239                "does not support live migration",
240                bdrv_get_device_or_node_name(bs));
241     migrate_add_blocker(s->migration_blocker);
242 
243     qemu_co_mutex_init(&s->lock);
244     return 0;
245 
246  fail:
247     g_free(s->l1_table);
248     qemu_vfree(s->l2_cache);
249     g_free(s->cluster_cache);
250     g_free(s->cluster_data);
251     return ret;
252 }
253 
254 
255 /* We have nothing to do for QCOW reopen, stubs just return
256  * success */
257 static int qcow_reopen_prepare(BDRVReopenState *state,
258                                BlockReopenQueue *queue, Error **errp)
259 {
260     return 0;
261 }
262 
263 static int qcow_set_key(BlockDriverState *bs, const char *key)
264 {
265     BDRVQcowState *s = bs->opaque;
266     uint8_t keybuf[16];
267     int len, i;
268     Error *err;
269 
270     memset(keybuf, 0, 16);
271     len = strlen(key);
272     if (len > 16)
273         len = 16;
274     /* XXX: we could compress the chars to 7 bits to increase
275        entropy */
276     for(i = 0;i < len;i++) {
277         keybuf[i] = key[i];
278     }
279     assert(bs->encrypted);
280 
281     qcrypto_cipher_free(s->cipher);
282     s->cipher = qcrypto_cipher_new(
283         QCRYPTO_CIPHER_ALG_AES_128,
284         QCRYPTO_CIPHER_MODE_CBC,
285         keybuf, G_N_ELEMENTS(keybuf),
286         &err);
287 
288     if (!s->cipher) {
289         /* XXX would be nice if errors in this method could
290          * be properly propagate to the caller. Would need
291          * the bdrv_set_key() API signature to be fixed. */
292         error_free(err);
293         return -1;
294     }
295     return 0;
296 }
297 
298 /* The crypt function is compatible with the linux cryptoloop
299    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
300    supported */
301 static int encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
302                            uint8_t *out_buf, const uint8_t *in_buf,
303                            int nb_sectors, bool enc, Error **errp)
304 {
305     union {
306         uint64_t ll[2];
307         uint8_t b[16];
308     } ivec;
309     int i;
310     int ret;
311 
312     for(i = 0; i < nb_sectors; i++) {
313         ivec.ll[0] = cpu_to_le64(sector_num);
314         ivec.ll[1] = 0;
315         if (qcrypto_cipher_setiv(s->cipher,
316                                  ivec.b, G_N_ELEMENTS(ivec.b),
317                                  errp) < 0) {
318             return -1;
319         }
320         if (enc) {
321             ret = qcrypto_cipher_encrypt(s->cipher,
322                                          in_buf,
323                                          out_buf,
324                                          512,
325                                          errp);
326         } else {
327             ret = qcrypto_cipher_decrypt(s->cipher,
328                                          in_buf,
329                                          out_buf,
330                                          512,
331                                          errp);
332         }
333         if (ret < 0) {
334             return -1;
335         }
336         sector_num++;
337         in_buf += 512;
338         out_buf += 512;
339     }
340     return 0;
341 }
342 
343 /* 'allocate' is:
344  *
345  * 0 to not allocate.
346  *
347  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
348  * 'n_end')
349  *
350  * 2 to allocate a compressed cluster of size
351  * 'compressed_size'. 'compressed_size' must be > 0 and <
352  * cluster_size
353  *
354  * return 0 if not allocated.
355  */
356 static uint64_t get_cluster_offset(BlockDriverState *bs,
357                                    uint64_t offset, int allocate,
358                                    int compressed_size,
359                                    int n_start, int n_end)
360 {
361     BDRVQcowState *s = bs->opaque;
362     int min_index, i, j, l1_index, l2_index;
363     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
364     uint32_t min_count;
365     int new_l2_table;
366 
367     l1_index = offset >> (s->l2_bits + s->cluster_bits);
368     l2_offset = s->l1_table[l1_index];
369     new_l2_table = 0;
370     if (!l2_offset) {
371         if (!allocate)
372             return 0;
373         /* allocate a new l2 entry */
374         l2_offset = bdrv_getlength(bs->file->bs);
375         /* round to cluster size */
376         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
377         /* update the L1 entry */
378         s->l1_table[l1_index] = l2_offset;
379         tmp = cpu_to_be64(l2_offset);
380         if (bdrv_pwrite_sync(bs->file->bs,
381                 s->l1_table_offset + l1_index * sizeof(tmp),
382                 &tmp, sizeof(tmp)) < 0)
383             return 0;
384         new_l2_table = 1;
385     }
386     for(i = 0; i < L2_CACHE_SIZE; i++) {
387         if (l2_offset == s->l2_cache_offsets[i]) {
388             /* increment the hit count */
389             if (++s->l2_cache_counts[i] == 0xffffffff) {
390                 for(j = 0; j < L2_CACHE_SIZE; j++) {
391                     s->l2_cache_counts[j] >>= 1;
392                 }
393             }
394             l2_table = s->l2_cache + (i << s->l2_bits);
395             goto found;
396         }
397     }
398     /* not found: load a new entry in the least used one */
399     min_index = 0;
400     min_count = 0xffffffff;
401     for(i = 0; i < L2_CACHE_SIZE; i++) {
402         if (s->l2_cache_counts[i] < min_count) {
403             min_count = s->l2_cache_counts[i];
404             min_index = i;
405         }
406     }
407     l2_table = s->l2_cache + (min_index << s->l2_bits);
408     if (new_l2_table) {
409         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
410         if (bdrv_pwrite_sync(bs->file->bs, l2_offset, l2_table,
411                 s->l2_size * sizeof(uint64_t)) < 0)
412             return 0;
413     } else {
414         if (bdrv_pread(bs->file->bs, l2_offset, l2_table,
415                        s->l2_size * sizeof(uint64_t)) !=
416             s->l2_size * sizeof(uint64_t))
417             return 0;
418     }
419     s->l2_cache_offsets[min_index] = l2_offset;
420     s->l2_cache_counts[min_index] = 1;
421  found:
422     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
423     cluster_offset = be64_to_cpu(l2_table[l2_index]);
424     if (!cluster_offset ||
425         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
426         if (!allocate)
427             return 0;
428         /* allocate a new cluster */
429         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
430             (n_end - n_start) < s->cluster_sectors) {
431             /* if the cluster is already compressed, we must
432                decompress it in the case it is not completely
433                overwritten */
434             if (decompress_cluster(bs, cluster_offset) < 0)
435                 return 0;
436             cluster_offset = bdrv_getlength(bs->file->bs);
437             cluster_offset = (cluster_offset + s->cluster_size - 1) &
438                 ~(s->cluster_size - 1);
439             /* write the cluster content */
440             if (bdrv_pwrite(bs->file->bs, cluster_offset, s->cluster_cache,
441                             s->cluster_size) !=
442                 s->cluster_size)
443                 return -1;
444         } else {
445             cluster_offset = bdrv_getlength(bs->file->bs);
446             if (allocate == 1) {
447                 /* round to cluster size */
448                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
449                     ~(s->cluster_size - 1);
450                 bdrv_truncate(bs->file->bs, cluster_offset + s->cluster_size);
451                 /* if encrypted, we must initialize the cluster
452                    content which won't be written */
453                 if (bs->encrypted &&
454                     (n_end - n_start) < s->cluster_sectors) {
455                     uint64_t start_sect;
456                     assert(s->cipher);
457                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
458                     memset(s->cluster_data + 512, 0x00, 512);
459                     for(i = 0; i < s->cluster_sectors; i++) {
460                         if (i < n_start || i >= n_end) {
461                             Error *err = NULL;
462                             if (encrypt_sectors(s, start_sect + i,
463                                                 s->cluster_data,
464                                                 s->cluster_data + 512, 1,
465                                                 true, &err) < 0) {
466                                 error_free(err);
467                                 errno = EIO;
468                                 return -1;
469                             }
470                             if (bdrv_pwrite(bs->file->bs,
471                                             cluster_offset + i * 512,
472                                             s->cluster_data, 512) != 512)
473                                 return -1;
474                         }
475                     }
476                 }
477             } else if (allocate == 2) {
478                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
479                     (uint64_t)compressed_size << (63 - s->cluster_bits);
480             }
481         }
482         /* update L2 table */
483         tmp = cpu_to_be64(cluster_offset);
484         l2_table[l2_index] = tmp;
485         if (bdrv_pwrite_sync(bs->file->bs, l2_offset + l2_index * sizeof(tmp),
486                 &tmp, sizeof(tmp)) < 0)
487             return 0;
488     }
489     return cluster_offset;
490 }
491 
492 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
493         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
494 {
495     BDRVQcowState *s = bs->opaque;
496     int index_in_cluster, n;
497     uint64_t cluster_offset;
498 
499     qemu_co_mutex_lock(&s->lock);
500     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
501     qemu_co_mutex_unlock(&s->lock);
502     index_in_cluster = sector_num & (s->cluster_sectors - 1);
503     n = s->cluster_sectors - index_in_cluster;
504     if (n > nb_sectors)
505         n = nb_sectors;
506     *pnum = n;
507     if (!cluster_offset) {
508         return 0;
509     }
510     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->cipher) {
511         return BDRV_BLOCK_DATA;
512     }
513     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
514     *file = bs->file->bs;
515     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
516 }
517 
518 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
519                              const uint8_t *buf, int buf_size)
520 {
521     z_stream strm1, *strm = &strm1;
522     int ret, out_len;
523 
524     memset(strm, 0, sizeof(*strm));
525 
526     strm->next_in = (uint8_t *)buf;
527     strm->avail_in = buf_size;
528     strm->next_out = out_buf;
529     strm->avail_out = out_buf_size;
530 
531     ret = inflateInit2(strm, -12);
532     if (ret != Z_OK)
533         return -1;
534     ret = inflate(strm, Z_FINISH);
535     out_len = strm->next_out - out_buf;
536     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
537         out_len != out_buf_size) {
538         inflateEnd(strm);
539         return -1;
540     }
541     inflateEnd(strm);
542     return 0;
543 }
544 
545 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
546 {
547     BDRVQcowState *s = bs->opaque;
548     int ret, csize;
549     uint64_t coffset;
550 
551     coffset = cluster_offset & s->cluster_offset_mask;
552     if (s->cluster_cache_offset != coffset) {
553         csize = cluster_offset >> (63 - s->cluster_bits);
554         csize &= (s->cluster_size - 1);
555         ret = bdrv_pread(bs->file->bs, coffset, s->cluster_data, csize);
556         if (ret != csize)
557             return -1;
558         if (decompress_buffer(s->cluster_cache, s->cluster_size,
559                               s->cluster_data, csize) < 0) {
560             return -1;
561         }
562         s->cluster_cache_offset = coffset;
563     }
564     return 0;
565 }
566 
567 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
568                          int nb_sectors, QEMUIOVector *qiov)
569 {
570     BDRVQcowState *s = bs->opaque;
571     int index_in_cluster;
572     int ret = 0, n;
573     uint64_t cluster_offset;
574     struct iovec hd_iov;
575     QEMUIOVector hd_qiov;
576     uint8_t *buf;
577     void *orig_buf;
578     Error *err = NULL;
579 
580     if (qiov->niov > 1) {
581         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
582         if (buf == NULL) {
583             return -ENOMEM;
584         }
585     } else {
586         orig_buf = NULL;
587         buf = (uint8_t *)qiov->iov->iov_base;
588     }
589 
590     qemu_co_mutex_lock(&s->lock);
591 
592     while (nb_sectors != 0) {
593         /* prepare next request */
594         cluster_offset = get_cluster_offset(bs, sector_num << 9,
595                                                  0, 0, 0, 0);
596         index_in_cluster = sector_num & (s->cluster_sectors - 1);
597         n = s->cluster_sectors - index_in_cluster;
598         if (n > nb_sectors) {
599             n = nb_sectors;
600         }
601 
602         if (!cluster_offset) {
603             if (bs->backing) {
604                 /* read from the base image */
605                 hd_iov.iov_base = (void *)buf;
606                 hd_iov.iov_len = n * 512;
607                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
608                 qemu_co_mutex_unlock(&s->lock);
609                 ret = bdrv_co_readv(bs->backing->bs, sector_num,
610                                     n, &hd_qiov);
611                 qemu_co_mutex_lock(&s->lock);
612                 if (ret < 0) {
613                     goto fail;
614                 }
615             } else {
616                 /* Note: in this case, no need to wait */
617                 memset(buf, 0, 512 * n);
618             }
619         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
620             /* add AIO support for compressed blocks ? */
621             if (decompress_cluster(bs, cluster_offset) < 0) {
622                 goto fail;
623             }
624             memcpy(buf,
625                    s->cluster_cache + index_in_cluster * 512, 512 * n);
626         } else {
627             if ((cluster_offset & 511) != 0) {
628                 goto fail;
629             }
630             hd_iov.iov_base = (void *)buf;
631             hd_iov.iov_len = n * 512;
632             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
633             qemu_co_mutex_unlock(&s->lock);
634             ret = bdrv_co_readv(bs->file->bs,
635                                 (cluster_offset >> 9) + index_in_cluster,
636                                 n, &hd_qiov);
637             qemu_co_mutex_lock(&s->lock);
638             if (ret < 0) {
639                 break;
640             }
641             if (bs->encrypted) {
642                 assert(s->cipher);
643                 if (encrypt_sectors(s, sector_num, buf, buf,
644                                     n, false, &err) < 0) {
645                     goto fail;
646                 }
647             }
648         }
649         ret = 0;
650 
651         nb_sectors -= n;
652         sector_num += n;
653         buf += n * 512;
654     }
655 
656 done:
657     qemu_co_mutex_unlock(&s->lock);
658 
659     if (qiov->niov > 1) {
660         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
661         qemu_vfree(orig_buf);
662     }
663 
664     return ret;
665 
666 fail:
667     error_free(err);
668     ret = -EIO;
669     goto done;
670 }
671 
672 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
673                           int nb_sectors, QEMUIOVector *qiov)
674 {
675     BDRVQcowState *s = bs->opaque;
676     int index_in_cluster;
677     uint64_t cluster_offset;
678     const uint8_t *src_buf;
679     int ret = 0, n;
680     uint8_t *cluster_data = NULL;
681     struct iovec hd_iov;
682     QEMUIOVector hd_qiov;
683     uint8_t *buf;
684     void *orig_buf;
685 
686     s->cluster_cache_offset = -1; /* disable compressed cache */
687 
688     if (qiov->niov > 1) {
689         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
690         if (buf == NULL) {
691             return -ENOMEM;
692         }
693         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
694     } else {
695         orig_buf = NULL;
696         buf = (uint8_t *)qiov->iov->iov_base;
697     }
698 
699     qemu_co_mutex_lock(&s->lock);
700 
701     while (nb_sectors != 0) {
702 
703         index_in_cluster = sector_num & (s->cluster_sectors - 1);
704         n = s->cluster_sectors - index_in_cluster;
705         if (n > nb_sectors) {
706             n = nb_sectors;
707         }
708         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
709                                             index_in_cluster,
710                                             index_in_cluster + n);
711         if (!cluster_offset || (cluster_offset & 511) != 0) {
712             ret = -EIO;
713             break;
714         }
715         if (bs->encrypted) {
716             Error *err = NULL;
717             assert(s->cipher);
718             if (!cluster_data) {
719                 cluster_data = g_malloc0(s->cluster_size);
720             }
721             if (encrypt_sectors(s, sector_num, cluster_data, buf,
722                                 n, true, &err) < 0) {
723                 error_free(err);
724                 ret = -EIO;
725                 break;
726             }
727             src_buf = cluster_data;
728         } else {
729             src_buf = buf;
730         }
731 
732         hd_iov.iov_base = (void *)src_buf;
733         hd_iov.iov_len = n * 512;
734         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
735         qemu_co_mutex_unlock(&s->lock);
736         ret = bdrv_co_writev(bs->file->bs,
737                              (cluster_offset >> 9) + index_in_cluster,
738                              n, &hd_qiov);
739         qemu_co_mutex_lock(&s->lock);
740         if (ret < 0) {
741             break;
742         }
743         ret = 0;
744 
745         nb_sectors -= n;
746         sector_num += n;
747         buf += n * 512;
748     }
749     qemu_co_mutex_unlock(&s->lock);
750 
751     if (qiov->niov > 1) {
752         qemu_vfree(orig_buf);
753     }
754     g_free(cluster_data);
755 
756     return ret;
757 }
758 
759 static void qcow_close(BlockDriverState *bs)
760 {
761     BDRVQcowState *s = bs->opaque;
762 
763     qcrypto_cipher_free(s->cipher);
764     s->cipher = NULL;
765     g_free(s->l1_table);
766     qemu_vfree(s->l2_cache);
767     g_free(s->cluster_cache);
768     g_free(s->cluster_data);
769 
770     migrate_del_blocker(s->migration_blocker);
771     error_free(s->migration_blocker);
772 }
773 
774 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
775 {
776     int header_size, backing_filename_len, l1_size, shift, i;
777     QCowHeader header;
778     uint8_t *tmp;
779     int64_t total_size = 0;
780     char *backing_file = NULL;
781     int flags = 0;
782     Error *local_err = NULL;
783     int ret;
784     BlockBackend *qcow_blk;
785 
786     /* Read out options */
787     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
788                           BDRV_SECTOR_SIZE);
789     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
790     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
791         flags |= BLOCK_FLAG_ENCRYPT;
792     }
793 
794     ret = bdrv_create_file(filename, opts, &local_err);
795     if (ret < 0) {
796         error_propagate(errp, local_err);
797         goto cleanup;
798     }
799 
800     qcow_blk = blk_new_open("image", filename, NULL, NULL,
801                             BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_PROTOCOL,
802                             &local_err);
803     if (qcow_blk == NULL) {
804         error_propagate(errp, local_err);
805         ret = -EIO;
806         goto cleanup;
807     }
808 
809     blk_set_allow_write_beyond_eof(qcow_blk, true);
810 
811     ret = blk_truncate(qcow_blk, 0);
812     if (ret < 0) {
813         goto exit;
814     }
815 
816     memset(&header, 0, sizeof(header));
817     header.magic = cpu_to_be32(QCOW_MAGIC);
818     header.version = cpu_to_be32(QCOW_VERSION);
819     header.size = cpu_to_be64(total_size);
820     header_size = sizeof(header);
821     backing_filename_len = 0;
822     if (backing_file) {
823         if (strcmp(backing_file, "fat:")) {
824             header.backing_file_offset = cpu_to_be64(header_size);
825             backing_filename_len = strlen(backing_file);
826             header.backing_file_size = cpu_to_be32(backing_filename_len);
827             header_size += backing_filename_len;
828         } else {
829             /* special backing file for vvfat */
830             backing_file = NULL;
831         }
832         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
833                                     unmodified sectors */
834         header.l2_bits = 12; /* 32 KB L2 tables */
835     } else {
836         header.cluster_bits = 12; /* 4 KB clusters */
837         header.l2_bits = 9; /* 4 KB L2 tables */
838     }
839     header_size = (header_size + 7) & ~7;
840     shift = header.cluster_bits + header.l2_bits;
841     l1_size = (total_size + (1LL << shift) - 1) >> shift;
842 
843     header.l1_table_offset = cpu_to_be64(header_size);
844     if (flags & BLOCK_FLAG_ENCRYPT) {
845         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
846     } else {
847         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
848     }
849 
850     /* write all the data */
851     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header));
852     if (ret != sizeof(header)) {
853         goto exit;
854     }
855 
856     if (backing_file) {
857         ret = blk_pwrite(qcow_blk, sizeof(header),
858             backing_file, backing_filename_len);
859         if (ret != backing_filename_len) {
860             goto exit;
861         }
862     }
863 
864     tmp = g_malloc0(BDRV_SECTOR_SIZE);
865     for (i = 0; i < ((sizeof(uint64_t)*l1_size + BDRV_SECTOR_SIZE - 1)/
866         BDRV_SECTOR_SIZE); i++) {
867         ret = blk_pwrite(qcow_blk, header_size +
868             BDRV_SECTOR_SIZE*i, tmp, BDRV_SECTOR_SIZE);
869         if (ret != BDRV_SECTOR_SIZE) {
870             g_free(tmp);
871             goto exit;
872         }
873     }
874 
875     g_free(tmp);
876     ret = 0;
877 exit:
878     blk_unref(qcow_blk);
879 cleanup:
880     g_free(backing_file);
881     return ret;
882 }
883 
884 static int qcow_make_empty(BlockDriverState *bs)
885 {
886     BDRVQcowState *s = bs->opaque;
887     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
888     int ret;
889 
890     memset(s->l1_table, 0, l1_length);
891     if (bdrv_pwrite_sync(bs->file->bs, s->l1_table_offset, s->l1_table,
892             l1_length) < 0)
893         return -1;
894     ret = bdrv_truncate(bs->file->bs, s->l1_table_offset + l1_length);
895     if (ret < 0)
896         return ret;
897 
898     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
899     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
900     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
901 
902     return 0;
903 }
904 
905 /* XXX: put compressed sectors first, then all the cluster aligned
906    tables to avoid losing bytes in alignment */
907 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
908                                  const uint8_t *buf, int nb_sectors)
909 {
910     BDRVQcowState *s = bs->opaque;
911     z_stream strm;
912     int ret, out_len;
913     uint8_t *out_buf;
914     uint64_t cluster_offset;
915 
916     if (nb_sectors != s->cluster_sectors) {
917         ret = -EINVAL;
918 
919         /* Zero-pad last write if image size is not cluster aligned */
920         if (sector_num + nb_sectors == bs->total_sectors &&
921             nb_sectors < s->cluster_sectors) {
922             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
923             memset(pad_buf, 0, s->cluster_size);
924             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
925             ret = qcow_write_compressed(bs, sector_num,
926                                         pad_buf, s->cluster_sectors);
927             qemu_vfree(pad_buf);
928         }
929         return ret;
930     }
931 
932     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
933 
934     /* best compression, small window, no zlib header */
935     memset(&strm, 0, sizeof(strm));
936     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
937                        Z_DEFLATED, -12,
938                        9, Z_DEFAULT_STRATEGY);
939     if (ret != 0) {
940         ret = -EINVAL;
941         goto fail;
942     }
943 
944     strm.avail_in = s->cluster_size;
945     strm.next_in = (uint8_t *)buf;
946     strm.avail_out = s->cluster_size;
947     strm.next_out = out_buf;
948 
949     ret = deflate(&strm, Z_FINISH);
950     if (ret != Z_STREAM_END && ret != Z_OK) {
951         deflateEnd(&strm);
952         ret = -EINVAL;
953         goto fail;
954     }
955     out_len = strm.next_out - out_buf;
956 
957     deflateEnd(&strm);
958 
959     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
960         /* could not compress: write normal cluster */
961         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
962         if (ret < 0) {
963             goto fail;
964         }
965     } else {
966         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
967                                             out_len, 0, 0);
968         if (cluster_offset == 0) {
969             ret = -EIO;
970             goto fail;
971         }
972 
973         cluster_offset &= s->cluster_offset_mask;
974         ret = bdrv_pwrite(bs->file->bs, cluster_offset, out_buf, out_len);
975         if (ret < 0) {
976             goto fail;
977         }
978     }
979 
980     ret = 0;
981 fail:
982     g_free(out_buf);
983     return ret;
984 }
985 
986 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
987 {
988     BDRVQcowState *s = bs->opaque;
989     bdi->cluster_size = s->cluster_size;
990     return 0;
991 }
992 
993 static QemuOptsList qcow_create_opts = {
994     .name = "qcow-create-opts",
995     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
996     .desc = {
997         {
998             .name = BLOCK_OPT_SIZE,
999             .type = QEMU_OPT_SIZE,
1000             .help = "Virtual disk size"
1001         },
1002         {
1003             .name = BLOCK_OPT_BACKING_FILE,
1004             .type = QEMU_OPT_STRING,
1005             .help = "File name of a base image"
1006         },
1007         {
1008             .name = BLOCK_OPT_ENCRYPT,
1009             .type = QEMU_OPT_BOOL,
1010             .help = "Encrypt the image",
1011             .def_value_str = "off"
1012         },
1013         { /* end of list */ }
1014     }
1015 };
1016 
1017 static BlockDriver bdrv_qcow = {
1018     .format_name	= "qcow",
1019     .instance_size	= sizeof(BDRVQcowState),
1020     .bdrv_probe		= qcow_probe,
1021     .bdrv_open		= qcow_open,
1022     .bdrv_close		= qcow_close,
1023     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1024     .bdrv_create            = qcow_create,
1025     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1026     .supports_backing       = true,
1027 
1028     .bdrv_co_readv          = qcow_co_readv,
1029     .bdrv_co_writev         = qcow_co_writev,
1030     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1031 
1032     .bdrv_set_key           = qcow_set_key,
1033     .bdrv_make_empty        = qcow_make_empty,
1034     .bdrv_write_compressed  = qcow_write_compressed,
1035     .bdrv_get_info          = qcow_get_info,
1036 
1037     .create_opts            = &qcow_create_opts,
1038 };
1039 
1040 static void bdrv_qcow_init(void)
1041 {
1042     bdrv_register(&bdrv_qcow);
1043 }
1044 
1045 block_init(bdrv_qcow_init);
1046