xref: /openbmc/qemu/block/qcow.c (revision bcad45de)
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 "qapi/error.h"
26 #include "qemu-common.h"
27 #include "qemu/error-report.h"
28 #include "block/block_int.h"
29 #include "sysemu/block-backend.h"
30 #include "qemu/module.h"
31 #include "qemu/bswap.h"
32 #include <zlib.h>
33 #include "qapi/qmp/qerror.h"
34 #include "crypto/cipher.h"
35 #include "migration/migration.h"
36 
37 /**************************************************************/
38 /* QEMU COW block driver with compression and encryption support */
39 
40 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
41 #define QCOW_VERSION 1
42 
43 #define QCOW_CRYPT_NONE 0
44 #define QCOW_CRYPT_AES  1
45 
46 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
47 
48 typedef struct QCowHeader {
49     uint32_t magic;
50     uint32_t version;
51     uint64_t backing_file_offset;
52     uint32_t backing_file_size;
53     uint32_t mtime;
54     uint64_t size; /* in bytes */
55     uint8_t cluster_bits;
56     uint8_t l2_bits;
57     uint16_t padding;
58     uint32_t crypt_method;
59     uint64_t l1_table_offset;
60 } QEMU_PACKED QCowHeader;
61 
62 #define L2_CACHE_SIZE 16
63 
64 typedef struct BDRVQcowState {
65     int cluster_bits;
66     int cluster_size;
67     int cluster_sectors;
68     int l2_bits;
69     int l2_size;
70     unsigned int l1_size;
71     uint64_t cluster_offset_mask;
72     uint64_t l1_table_offset;
73     uint64_t *l1_table;
74     uint64_t *l2_cache;
75     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
76     uint32_t l2_cache_counts[L2_CACHE_SIZE];
77     uint8_t *cluster_cache;
78     uint8_t *cluster_data;
79     uint64_t cluster_cache_offset;
80     QCryptoCipher *cipher; /* NULL if no key yet */
81     uint32_t crypt_method_header;
82     CoMutex lock;
83     Error *migration_blocker;
84 } BDRVQcowState;
85 
86 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
87 
88 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
89 {
90     const QCowHeader *cow_header = (const void *)buf;
91 
92     if (buf_size >= sizeof(QCowHeader) &&
93         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
94         be32_to_cpu(cow_header->version) == QCOW_VERSION)
95         return 100;
96     else
97         return 0;
98 }
99 
100 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
101                      Error **errp)
102 {
103     BDRVQcowState *s = bs->opaque;
104     unsigned int len, i, shift;
105     int ret;
106     QCowHeader header;
107 
108     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
109     if (ret < 0) {
110         goto fail;
111     }
112     be32_to_cpus(&header.magic);
113     be32_to_cpus(&header.version);
114     be64_to_cpus(&header.backing_file_offset);
115     be32_to_cpus(&header.backing_file_size);
116     be32_to_cpus(&header.mtime);
117     be64_to_cpus(&header.size);
118     be32_to_cpus(&header.crypt_method);
119     be64_to_cpus(&header.l1_table_offset);
120 
121     if (header.magic != QCOW_MAGIC) {
122         error_setg(errp, "Image not in qcow format");
123         ret = -EINVAL;
124         goto fail;
125     }
126     if (header.version != QCOW_VERSION) {
127         error_setg(errp, "Unsupported qcow version %" PRIu32, header.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     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128,
157                                  QCRYPTO_CIPHER_MODE_CBC)) {
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         if (bdrv_uses_whitelist() &&
165             s->crypt_method_header == QCOW_CRYPT_AES) {
166             error_setg(errp,
167                        "Use of AES-CBC encrypted qcow images is no longer "
168                        "supported in system emulators");
169             error_append_hint(errp,
170                               "You can use 'qemu-img convert' to convert your "
171                               "image to an alternative supported format, such "
172                               "as unencrypted qcow, or raw with the LUKS "
173                               "format instead.\n");
174             ret = -ENOSYS;
175             goto fail;
176         }
177 
178         bs->encrypted = true;
179     }
180     s->cluster_bits = header.cluster_bits;
181     s->cluster_size = 1 << s->cluster_bits;
182     s->cluster_sectors = 1 << (s->cluster_bits - 9);
183     s->l2_bits = header.l2_bits;
184     s->l2_size = 1 << s->l2_bits;
185     bs->total_sectors = header.size / 512;
186     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
187 
188     /* read the level 1 table */
189     shift = s->cluster_bits + s->l2_bits;
190     if (header.size > UINT64_MAX - (1LL << shift)) {
191         error_setg(errp, "Image too large");
192         ret = -EINVAL;
193         goto fail;
194     } else {
195         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
196         if (l1_size > INT_MAX / sizeof(uint64_t)) {
197             error_setg(errp, "Image too large");
198             ret = -EINVAL;
199             goto fail;
200         }
201         s->l1_size = l1_size;
202     }
203 
204     s->l1_table_offset = header.l1_table_offset;
205     s->l1_table = g_try_new(uint64_t, s->l1_size);
206     if (s->l1_table == NULL) {
207         error_setg(errp, "Could not allocate memory for L1 table");
208         ret = -ENOMEM;
209         goto fail;
210     }
211 
212     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
213                s->l1_size * sizeof(uint64_t));
214     if (ret < 0) {
215         goto fail;
216     }
217 
218     for(i = 0;i < s->l1_size; i++) {
219         be64_to_cpus(&s->l1_table[i]);
220     }
221 
222     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
223     s->l2_cache =
224         qemu_try_blockalign(bs->file->bs,
225                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
226     if (s->l2_cache == NULL) {
227         error_setg(errp, "Could not allocate L2 table cache");
228         ret = -ENOMEM;
229         goto fail;
230     }
231     s->cluster_cache = g_malloc(s->cluster_size);
232     s->cluster_data = g_malloc(s->cluster_size);
233     s->cluster_cache_offset = -1;
234 
235     /* read the backing file name */
236     if (header.backing_file_offset != 0) {
237         len = header.backing_file_size;
238         if (len > 1023 || len >= sizeof(bs->backing_file)) {
239             error_setg(errp, "Backing file name too long");
240             ret = -EINVAL;
241             goto fail;
242         }
243         ret = bdrv_pread(bs->file, header.backing_file_offset,
244                    bs->backing_file, len);
245         if (ret < 0) {
246             goto fail;
247         }
248         bs->backing_file[len] = '\0';
249     }
250 
251     /* Disable migration when qcow images are used */
252     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
253                "does not support live migration",
254                bdrv_get_device_or_node_name(bs));
255     migrate_add_blocker(s->migration_blocker);
256 
257     qemu_co_mutex_init(&s->lock);
258     return 0;
259 
260  fail:
261     g_free(s->l1_table);
262     qemu_vfree(s->l2_cache);
263     g_free(s->cluster_cache);
264     g_free(s->cluster_data);
265     return ret;
266 }
267 
268 
269 /* We have nothing to do for QCOW reopen, stubs just return
270  * success */
271 static int qcow_reopen_prepare(BDRVReopenState *state,
272                                BlockReopenQueue *queue, Error **errp)
273 {
274     return 0;
275 }
276 
277 static int qcow_set_key(BlockDriverState *bs, const char *key)
278 {
279     BDRVQcowState *s = bs->opaque;
280     uint8_t keybuf[16];
281     int len, i;
282     Error *err;
283 
284     memset(keybuf, 0, 16);
285     len = strlen(key);
286     if (len > 16)
287         len = 16;
288     /* XXX: we could compress the chars to 7 bits to increase
289        entropy */
290     for(i = 0;i < len;i++) {
291         keybuf[i] = key[i];
292     }
293     assert(bs->encrypted);
294 
295     qcrypto_cipher_free(s->cipher);
296     s->cipher = qcrypto_cipher_new(
297         QCRYPTO_CIPHER_ALG_AES_128,
298         QCRYPTO_CIPHER_MODE_CBC,
299         keybuf, G_N_ELEMENTS(keybuf),
300         &err);
301 
302     if (!s->cipher) {
303         /* XXX would be nice if errors in this method could
304          * be properly propagate to the caller. Would need
305          * the bdrv_set_key() API signature to be fixed. */
306         error_free(err);
307         return -1;
308     }
309     return 0;
310 }
311 
312 /* The crypt function is compatible with the linux cryptoloop
313    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
314    supported */
315 static int encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
316                            uint8_t *out_buf, const uint8_t *in_buf,
317                            int nb_sectors, bool enc, Error **errp)
318 {
319     union {
320         uint64_t ll[2];
321         uint8_t b[16];
322     } ivec;
323     int i;
324     int ret;
325 
326     for(i = 0; i < nb_sectors; i++) {
327         ivec.ll[0] = cpu_to_le64(sector_num);
328         ivec.ll[1] = 0;
329         if (qcrypto_cipher_setiv(s->cipher,
330                                  ivec.b, G_N_ELEMENTS(ivec.b),
331                                  errp) < 0) {
332             return -1;
333         }
334         if (enc) {
335             ret = qcrypto_cipher_encrypt(s->cipher,
336                                          in_buf,
337                                          out_buf,
338                                          512,
339                                          errp);
340         } else {
341             ret = qcrypto_cipher_decrypt(s->cipher,
342                                          in_buf,
343                                          out_buf,
344                                          512,
345                                          errp);
346         }
347         if (ret < 0) {
348             return -1;
349         }
350         sector_num++;
351         in_buf += 512;
352         out_buf += 512;
353     }
354     return 0;
355 }
356 
357 /* 'allocate' is:
358  *
359  * 0 to not allocate.
360  *
361  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
362  * 'n_end')
363  *
364  * 2 to allocate a compressed cluster of size
365  * 'compressed_size'. 'compressed_size' must be > 0 and <
366  * cluster_size
367  *
368  * return 0 if not allocated.
369  */
370 static uint64_t get_cluster_offset(BlockDriverState *bs,
371                                    uint64_t offset, int allocate,
372                                    int compressed_size,
373                                    int n_start, int n_end)
374 {
375     BDRVQcowState *s = bs->opaque;
376     int min_index, i, j, l1_index, l2_index;
377     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
378     uint32_t min_count;
379     int new_l2_table;
380 
381     l1_index = offset >> (s->l2_bits + s->cluster_bits);
382     l2_offset = s->l1_table[l1_index];
383     new_l2_table = 0;
384     if (!l2_offset) {
385         if (!allocate)
386             return 0;
387         /* allocate a new l2 entry */
388         l2_offset = bdrv_getlength(bs->file->bs);
389         /* round to cluster size */
390         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
391         /* update the L1 entry */
392         s->l1_table[l1_index] = l2_offset;
393         tmp = cpu_to_be64(l2_offset);
394         if (bdrv_pwrite_sync(bs->file,
395                 s->l1_table_offset + l1_index * sizeof(tmp),
396                 &tmp, sizeof(tmp)) < 0)
397             return 0;
398         new_l2_table = 1;
399     }
400     for(i = 0; i < L2_CACHE_SIZE; i++) {
401         if (l2_offset == s->l2_cache_offsets[i]) {
402             /* increment the hit count */
403             if (++s->l2_cache_counts[i] == 0xffffffff) {
404                 for(j = 0; j < L2_CACHE_SIZE; j++) {
405                     s->l2_cache_counts[j] >>= 1;
406                 }
407             }
408             l2_table = s->l2_cache + (i << s->l2_bits);
409             goto found;
410         }
411     }
412     /* not found: load a new entry in the least used one */
413     min_index = 0;
414     min_count = 0xffffffff;
415     for(i = 0; i < L2_CACHE_SIZE; i++) {
416         if (s->l2_cache_counts[i] < min_count) {
417             min_count = s->l2_cache_counts[i];
418             min_index = i;
419         }
420     }
421     l2_table = s->l2_cache + (min_index << s->l2_bits);
422     if (new_l2_table) {
423         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
424         if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
425                 s->l2_size * sizeof(uint64_t)) < 0)
426             return 0;
427     } else {
428         if (bdrv_pread(bs->file, l2_offset, l2_table,
429                        s->l2_size * sizeof(uint64_t)) !=
430             s->l2_size * sizeof(uint64_t))
431             return 0;
432     }
433     s->l2_cache_offsets[min_index] = l2_offset;
434     s->l2_cache_counts[min_index] = 1;
435  found:
436     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
437     cluster_offset = be64_to_cpu(l2_table[l2_index]);
438     if (!cluster_offset ||
439         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
440         if (!allocate)
441             return 0;
442         /* allocate a new cluster */
443         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
444             (n_end - n_start) < s->cluster_sectors) {
445             /* if the cluster is already compressed, we must
446                decompress it in the case it is not completely
447                overwritten */
448             if (decompress_cluster(bs, cluster_offset) < 0)
449                 return 0;
450             cluster_offset = bdrv_getlength(bs->file->bs);
451             cluster_offset = (cluster_offset + s->cluster_size - 1) &
452                 ~(s->cluster_size - 1);
453             /* write the cluster content */
454             if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
455                             s->cluster_size) !=
456                 s->cluster_size)
457                 return -1;
458         } else {
459             cluster_offset = bdrv_getlength(bs->file->bs);
460             if (allocate == 1) {
461                 /* round to cluster size */
462                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
463                     ~(s->cluster_size - 1);
464                 bdrv_truncate(bs->file->bs, cluster_offset + s->cluster_size);
465                 /* if encrypted, we must initialize the cluster
466                    content which won't be written */
467                 if (bs->encrypted &&
468                     (n_end - n_start) < s->cluster_sectors) {
469                     uint64_t start_sect;
470                     assert(s->cipher);
471                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
472                     memset(s->cluster_data + 512, 0x00, 512);
473                     for(i = 0; i < s->cluster_sectors; i++) {
474                         if (i < n_start || i >= n_end) {
475                             Error *err = NULL;
476                             if (encrypt_sectors(s, start_sect + i,
477                                                 s->cluster_data,
478                                                 s->cluster_data + 512, 1,
479                                                 true, &err) < 0) {
480                                 error_free(err);
481                                 errno = EIO;
482                                 return -1;
483                             }
484                             if (bdrv_pwrite(bs->file,
485                                             cluster_offset + i * 512,
486                                             s->cluster_data, 512) != 512)
487                                 return -1;
488                         }
489                     }
490                 }
491             } else if (allocate == 2) {
492                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
493                     (uint64_t)compressed_size << (63 - s->cluster_bits);
494             }
495         }
496         /* update L2 table */
497         tmp = cpu_to_be64(cluster_offset);
498         l2_table[l2_index] = tmp;
499         if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
500                 &tmp, sizeof(tmp)) < 0)
501             return 0;
502     }
503     return cluster_offset;
504 }
505 
506 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
507         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
508 {
509     BDRVQcowState *s = bs->opaque;
510     int index_in_cluster, n;
511     uint64_t cluster_offset;
512 
513     qemu_co_mutex_lock(&s->lock);
514     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
515     qemu_co_mutex_unlock(&s->lock);
516     index_in_cluster = sector_num & (s->cluster_sectors - 1);
517     n = s->cluster_sectors - index_in_cluster;
518     if (n > nb_sectors)
519         n = nb_sectors;
520     *pnum = n;
521     if (!cluster_offset) {
522         return 0;
523     }
524     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->cipher) {
525         return BDRV_BLOCK_DATA;
526     }
527     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
528     *file = bs->file->bs;
529     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
530 }
531 
532 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
533                              const uint8_t *buf, int buf_size)
534 {
535     z_stream strm1, *strm = &strm1;
536     int ret, out_len;
537 
538     memset(strm, 0, sizeof(*strm));
539 
540     strm->next_in = (uint8_t *)buf;
541     strm->avail_in = buf_size;
542     strm->next_out = out_buf;
543     strm->avail_out = out_buf_size;
544 
545     ret = inflateInit2(strm, -12);
546     if (ret != Z_OK)
547         return -1;
548     ret = inflate(strm, Z_FINISH);
549     out_len = strm->next_out - out_buf;
550     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
551         out_len != out_buf_size) {
552         inflateEnd(strm);
553         return -1;
554     }
555     inflateEnd(strm);
556     return 0;
557 }
558 
559 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
560 {
561     BDRVQcowState *s = bs->opaque;
562     int ret, csize;
563     uint64_t coffset;
564 
565     coffset = cluster_offset & s->cluster_offset_mask;
566     if (s->cluster_cache_offset != coffset) {
567         csize = cluster_offset >> (63 - s->cluster_bits);
568         csize &= (s->cluster_size - 1);
569         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
570         if (ret != csize)
571             return -1;
572         if (decompress_buffer(s->cluster_cache, s->cluster_size,
573                               s->cluster_data, csize) < 0) {
574             return -1;
575         }
576         s->cluster_cache_offset = coffset;
577     }
578     return 0;
579 }
580 
581 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
582                          int nb_sectors, QEMUIOVector *qiov)
583 {
584     BDRVQcowState *s = bs->opaque;
585     int index_in_cluster;
586     int ret = 0, n;
587     uint64_t cluster_offset;
588     struct iovec hd_iov;
589     QEMUIOVector hd_qiov;
590     uint8_t *buf;
591     void *orig_buf;
592     Error *err = NULL;
593 
594     if (qiov->niov > 1) {
595         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
596         if (buf == NULL) {
597             return -ENOMEM;
598         }
599     } else {
600         orig_buf = NULL;
601         buf = (uint8_t *)qiov->iov->iov_base;
602     }
603 
604     qemu_co_mutex_lock(&s->lock);
605 
606     while (nb_sectors != 0) {
607         /* prepare next request */
608         cluster_offset = get_cluster_offset(bs, sector_num << 9,
609                                                  0, 0, 0, 0);
610         index_in_cluster = sector_num & (s->cluster_sectors - 1);
611         n = s->cluster_sectors - index_in_cluster;
612         if (n > nb_sectors) {
613             n = nb_sectors;
614         }
615 
616         if (!cluster_offset) {
617             if (bs->backing) {
618                 /* read from the base image */
619                 hd_iov.iov_base = (void *)buf;
620                 hd_iov.iov_len = n * 512;
621                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
622                 qemu_co_mutex_unlock(&s->lock);
623                 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov);
624                 qemu_co_mutex_lock(&s->lock);
625                 if (ret < 0) {
626                     goto fail;
627                 }
628             } else {
629                 /* Note: in this case, no need to wait */
630                 memset(buf, 0, 512 * n);
631             }
632         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
633             /* add AIO support for compressed blocks ? */
634             if (decompress_cluster(bs, cluster_offset) < 0) {
635                 goto fail;
636             }
637             memcpy(buf,
638                    s->cluster_cache + index_in_cluster * 512, 512 * n);
639         } else {
640             if ((cluster_offset & 511) != 0) {
641                 goto fail;
642             }
643             hd_iov.iov_base = (void *)buf;
644             hd_iov.iov_len = n * 512;
645             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
646             qemu_co_mutex_unlock(&s->lock);
647             ret = bdrv_co_readv(bs->file,
648                                 (cluster_offset >> 9) + index_in_cluster,
649                                 n, &hd_qiov);
650             qemu_co_mutex_lock(&s->lock);
651             if (ret < 0) {
652                 break;
653             }
654             if (bs->encrypted) {
655                 assert(s->cipher);
656                 if (encrypt_sectors(s, sector_num, buf, buf,
657                                     n, false, &err) < 0) {
658                     goto fail;
659                 }
660             }
661         }
662         ret = 0;
663 
664         nb_sectors -= n;
665         sector_num += n;
666         buf += n * 512;
667     }
668 
669 done:
670     qemu_co_mutex_unlock(&s->lock);
671 
672     if (qiov->niov > 1) {
673         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
674         qemu_vfree(orig_buf);
675     }
676 
677     return ret;
678 
679 fail:
680     error_free(err);
681     ret = -EIO;
682     goto done;
683 }
684 
685 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
686                           int nb_sectors, QEMUIOVector *qiov)
687 {
688     BDRVQcowState *s = bs->opaque;
689     int index_in_cluster;
690     uint64_t cluster_offset;
691     const uint8_t *src_buf;
692     int ret = 0, n;
693     uint8_t *cluster_data = NULL;
694     struct iovec hd_iov;
695     QEMUIOVector hd_qiov;
696     uint8_t *buf;
697     void *orig_buf;
698 
699     s->cluster_cache_offset = -1; /* disable compressed cache */
700 
701     if (qiov->niov > 1) {
702         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
703         if (buf == NULL) {
704             return -ENOMEM;
705         }
706         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
707     } else {
708         orig_buf = NULL;
709         buf = (uint8_t *)qiov->iov->iov_base;
710     }
711 
712     qemu_co_mutex_lock(&s->lock);
713 
714     while (nb_sectors != 0) {
715 
716         index_in_cluster = sector_num & (s->cluster_sectors - 1);
717         n = s->cluster_sectors - index_in_cluster;
718         if (n > nb_sectors) {
719             n = nb_sectors;
720         }
721         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
722                                             index_in_cluster,
723                                             index_in_cluster + n);
724         if (!cluster_offset || (cluster_offset & 511) != 0) {
725             ret = -EIO;
726             break;
727         }
728         if (bs->encrypted) {
729             Error *err = NULL;
730             assert(s->cipher);
731             if (!cluster_data) {
732                 cluster_data = g_malloc0(s->cluster_size);
733             }
734             if (encrypt_sectors(s, sector_num, cluster_data, buf,
735                                 n, true, &err) < 0) {
736                 error_free(err);
737                 ret = -EIO;
738                 break;
739             }
740             src_buf = cluster_data;
741         } else {
742             src_buf = buf;
743         }
744 
745         hd_iov.iov_base = (void *)src_buf;
746         hd_iov.iov_len = n * 512;
747         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
748         qemu_co_mutex_unlock(&s->lock);
749         ret = bdrv_co_writev(bs->file,
750                              (cluster_offset >> 9) + index_in_cluster,
751                              n, &hd_qiov);
752         qemu_co_mutex_lock(&s->lock);
753         if (ret < 0) {
754             break;
755         }
756         ret = 0;
757 
758         nb_sectors -= n;
759         sector_num += n;
760         buf += n * 512;
761     }
762     qemu_co_mutex_unlock(&s->lock);
763 
764     if (qiov->niov > 1) {
765         qemu_vfree(orig_buf);
766     }
767     g_free(cluster_data);
768 
769     return ret;
770 }
771 
772 static void qcow_close(BlockDriverState *bs)
773 {
774     BDRVQcowState *s = bs->opaque;
775 
776     qcrypto_cipher_free(s->cipher);
777     s->cipher = NULL;
778     g_free(s->l1_table);
779     qemu_vfree(s->l2_cache);
780     g_free(s->cluster_cache);
781     g_free(s->cluster_data);
782 
783     migrate_del_blocker(s->migration_blocker);
784     error_free(s->migration_blocker);
785 }
786 
787 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
788 {
789     int header_size, backing_filename_len, l1_size, shift, i;
790     QCowHeader header;
791     uint8_t *tmp;
792     int64_t total_size = 0;
793     char *backing_file = NULL;
794     int flags = 0;
795     Error *local_err = NULL;
796     int ret;
797     BlockBackend *qcow_blk;
798 
799     /* Read out options */
800     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
801                           BDRV_SECTOR_SIZE);
802     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
803     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
804         flags |= BLOCK_FLAG_ENCRYPT;
805     }
806 
807     ret = bdrv_create_file(filename, opts, &local_err);
808     if (ret < 0) {
809         error_propagate(errp, local_err);
810         goto cleanup;
811     }
812 
813     qcow_blk = blk_new_open(filename, NULL, NULL,
814                             BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
815     if (qcow_blk == NULL) {
816         error_propagate(errp, local_err);
817         ret = -EIO;
818         goto cleanup;
819     }
820 
821     blk_set_allow_write_beyond_eof(qcow_blk, true);
822 
823     ret = blk_truncate(qcow_blk, 0);
824     if (ret < 0) {
825         goto exit;
826     }
827 
828     memset(&header, 0, sizeof(header));
829     header.magic = cpu_to_be32(QCOW_MAGIC);
830     header.version = cpu_to_be32(QCOW_VERSION);
831     header.size = cpu_to_be64(total_size);
832     header_size = sizeof(header);
833     backing_filename_len = 0;
834     if (backing_file) {
835         if (strcmp(backing_file, "fat:")) {
836             header.backing_file_offset = cpu_to_be64(header_size);
837             backing_filename_len = strlen(backing_file);
838             header.backing_file_size = cpu_to_be32(backing_filename_len);
839             header_size += backing_filename_len;
840         } else {
841             /* special backing file for vvfat */
842             backing_file = NULL;
843         }
844         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
845                                     unmodified sectors */
846         header.l2_bits = 12; /* 32 KB L2 tables */
847     } else {
848         header.cluster_bits = 12; /* 4 KB clusters */
849         header.l2_bits = 9; /* 4 KB L2 tables */
850     }
851     header_size = (header_size + 7) & ~7;
852     shift = header.cluster_bits + header.l2_bits;
853     l1_size = (total_size + (1LL << shift) - 1) >> shift;
854 
855     header.l1_table_offset = cpu_to_be64(header_size);
856     if (flags & BLOCK_FLAG_ENCRYPT) {
857         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
858     } else {
859         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
860     }
861 
862     /* write all the data */
863     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
864     if (ret != sizeof(header)) {
865         goto exit;
866     }
867 
868     if (backing_file) {
869         ret = blk_pwrite(qcow_blk, sizeof(header),
870                          backing_file, backing_filename_len, 0);
871         if (ret != backing_filename_len) {
872             goto exit;
873         }
874     }
875 
876     tmp = g_malloc0(BDRV_SECTOR_SIZE);
877     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
878          i++) {
879         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
880                          tmp, BDRV_SECTOR_SIZE, 0);
881         if (ret != BDRV_SECTOR_SIZE) {
882             g_free(tmp);
883             goto exit;
884         }
885     }
886 
887     g_free(tmp);
888     ret = 0;
889 exit:
890     blk_unref(qcow_blk);
891 cleanup:
892     g_free(backing_file);
893     return ret;
894 }
895 
896 static int qcow_make_empty(BlockDriverState *bs)
897 {
898     BDRVQcowState *s = bs->opaque;
899     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
900     int ret;
901 
902     memset(s->l1_table, 0, l1_length);
903     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
904             l1_length) < 0)
905         return -1;
906     ret = bdrv_truncate(bs->file->bs, s->l1_table_offset + l1_length);
907     if (ret < 0)
908         return ret;
909 
910     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
911     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
912     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
913 
914     return 0;
915 }
916 
917 /* XXX: put compressed sectors first, then all the cluster aligned
918    tables to avoid losing bytes in alignment */
919 static coroutine_fn int
920 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
921                            uint64_t bytes, QEMUIOVector *qiov)
922 {
923     BDRVQcowState *s = bs->opaque;
924     QEMUIOVector hd_qiov;
925     struct iovec iov;
926     z_stream strm;
927     int ret, out_len;
928     uint8_t *buf, *out_buf;
929     uint64_t cluster_offset;
930 
931     buf = qemu_blockalign(bs, s->cluster_size);
932     if (bytes != s->cluster_size) {
933         if (bytes > s->cluster_size ||
934             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
935         {
936             qemu_vfree(buf);
937             return -EINVAL;
938         }
939         /* Zero-pad last write if image size is not cluster aligned */
940         memset(buf + bytes, 0, s->cluster_size - bytes);
941     }
942     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
943 
944     out_buf = g_malloc(s->cluster_size);
945 
946     /* best compression, small window, no zlib header */
947     memset(&strm, 0, sizeof(strm));
948     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
949                        Z_DEFLATED, -12,
950                        9, Z_DEFAULT_STRATEGY);
951     if (ret != 0) {
952         ret = -EINVAL;
953         goto fail;
954     }
955 
956     strm.avail_in = s->cluster_size;
957     strm.next_in = (uint8_t *)buf;
958     strm.avail_out = s->cluster_size;
959     strm.next_out = out_buf;
960 
961     ret = deflate(&strm, Z_FINISH);
962     if (ret != Z_STREAM_END && ret != Z_OK) {
963         deflateEnd(&strm);
964         ret = -EINVAL;
965         goto fail;
966     }
967     out_len = strm.next_out - out_buf;
968 
969     deflateEnd(&strm);
970 
971     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
972         /* could not compress: write normal cluster */
973         ret = qcow_co_writev(bs, offset >> BDRV_SECTOR_BITS,
974                              bytes >> BDRV_SECTOR_BITS, qiov);
975         if (ret < 0) {
976             goto fail;
977         }
978         goto success;
979     }
980     qemu_co_mutex_lock(&s->lock);
981     cluster_offset = get_cluster_offset(bs, offset, 2, out_len, 0, 0);
982     qemu_co_mutex_unlock(&s->lock);
983     if (cluster_offset == 0) {
984         ret = -EIO;
985         goto fail;
986     }
987     cluster_offset &= s->cluster_offset_mask;
988 
989     iov = (struct iovec) {
990         .iov_base   = out_buf,
991         .iov_len    = out_len,
992     };
993     qemu_iovec_init_external(&hd_qiov, &iov, 1);
994     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
995     if (ret < 0) {
996         goto fail;
997     }
998 success:
999     ret = 0;
1000 fail:
1001     qemu_vfree(buf);
1002     g_free(out_buf);
1003     return ret;
1004 }
1005 
1006 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1007 {
1008     BDRVQcowState *s = bs->opaque;
1009     bdi->cluster_size = s->cluster_size;
1010     return 0;
1011 }
1012 
1013 static QemuOptsList qcow_create_opts = {
1014     .name = "qcow-create-opts",
1015     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1016     .desc = {
1017         {
1018             .name = BLOCK_OPT_SIZE,
1019             .type = QEMU_OPT_SIZE,
1020             .help = "Virtual disk size"
1021         },
1022         {
1023             .name = BLOCK_OPT_BACKING_FILE,
1024             .type = QEMU_OPT_STRING,
1025             .help = "File name of a base image"
1026         },
1027         {
1028             .name = BLOCK_OPT_ENCRYPT,
1029             .type = QEMU_OPT_BOOL,
1030             .help = "Encrypt the image",
1031             .def_value_str = "off"
1032         },
1033         { /* end of list */ }
1034     }
1035 };
1036 
1037 static BlockDriver bdrv_qcow = {
1038     .format_name	= "qcow",
1039     .instance_size	= sizeof(BDRVQcowState),
1040     .bdrv_probe		= qcow_probe,
1041     .bdrv_open		= qcow_open,
1042     .bdrv_close		= qcow_close,
1043     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1044     .bdrv_create            = qcow_create,
1045     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1046     .supports_backing       = true,
1047 
1048     .bdrv_co_readv          = qcow_co_readv,
1049     .bdrv_co_writev         = qcow_co_writev,
1050     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1051 
1052     .bdrv_set_key           = qcow_set_key,
1053     .bdrv_make_empty        = qcow_make_empty,
1054     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1055     .bdrv_get_info          = qcow_get_info,
1056 
1057     .create_opts            = &qcow_create_opts,
1058 };
1059 
1060 static void bdrv_qcow_init(void)
1061 {
1062     bdrv_register(&bdrv_qcow);
1063 }
1064 
1065 block_init(bdrv_qcow_init);
1066