xref: /openbmc/qemu/block/qcow.c (revision 1f7685fa)
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 <zlib.h>
32 #include "qapi/qmp/qerror.h"
33 #include "crypto/cipher.h"
34 #include "migration/migration.h"
35 
36 /**************************************************************/
37 /* QEMU COW block driver with compression and encryption support */
38 
39 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
40 #define QCOW_VERSION 1
41 
42 #define QCOW_CRYPT_NONE 0
43 #define QCOW_CRYPT_AES  1
44 
45 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
46 
47 typedef struct QCowHeader {
48     uint32_t magic;
49     uint32_t version;
50     uint64_t backing_file_offset;
51     uint32_t backing_file_size;
52     uint32_t mtime;
53     uint64_t size; /* in bytes */
54     uint8_t cluster_bits;
55     uint8_t l2_bits;
56     uint16_t padding;
57     uint32_t crypt_method;
58     uint64_t l1_table_offset;
59 } QEMU_PACKED QCowHeader;
60 
61 #define L2_CACHE_SIZE 16
62 
63 typedef struct BDRVQcowState {
64     int cluster_bits;
65     int cluster_size;
66     int cluster_sectors;
67     int l2_bits;
68     int l2_size;
69     unsigned int l1_size;
70     uint64_t cluster_offset_mask;
71     uint64_t l1_table_offset;
72     uint64_t *l1_table;
73     uint64_t *l2_cache;
74     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
75     uint32_t l2_cache_counts[L2_CACHE_SIZE];
76     uint8_t *cluster_cache;
77     uint8_t *cluster_data;
78     uint64_t cluster_cache_offset;
79     QCryptoCipher *cipher; /* NULL if no key yet */
80     uint32_t crypt_method_header;
81     CoMutex lock;
82     Error *migration_blocker;
83 } BDRVQcowState;
84 
85 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
86 
87 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
88 {
89     const QCowHeader *cow_header = (const void *)buf;
90 
91     if (buf_size >= sizeof(QCowHeader) &&
92         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
93         be32_to_cpu(cow_header->version) == QCOW_VERSION)
94         return 100;
95     else
96         return 0;
97 }
98 
99 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
100                      Error **errp)
101 {
102     BDRVQcowState *s = bs->opaque;
103     unsigned int len, i, shift;
104     int ret;
105     QCowHeader header;
106 
107     ret = bdrv_pread(bs->file->bs, 0, &header, sizeof(header));
108     if (ret < 0) {
109         goto fail;
110     }
111     be32_to_cpus(&header.magic);
112     be32_to_cpus(&header.version);
113     be64_to_cpus(&header.backing_file_offset);
114     be32_to_cpus(&header.backing_file_size);
115     be32_to_cpus(&header.mtime);
116     be64_to_cpus(&header.size);
117     be32_to_cpus(&header.crypt_method);
118     be64_to_cpus(&header.l1_table_offset);
119 
120     if (header.magic != QCOW_MAGIC) {
121         error_setg(errp, "Image not in qcow format");
122         ret = -EINVAL;
123         goto fail;
124     }
125     if (header.version != QCOW_VERSION) {
126         error_setg(errp, "Unsupported qcow version %" PRIu32, header.version);
127         ret = -ENOTSUP;
128         goto fail;
129     }
130 
131     if (header.size <= 1) {
132         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
133         ret = -EINVAL;
134         goto fail;
135     }
136     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
137         error_setg(errp, "Cluster size must be between 512 and 64k");
138         ret = -EINVAL;
139         goto fail;
140     }
141 
142     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
143      * so bytes = num_entries << 3. */
144     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
145         error_setg(errp, "L2 table size must be between 512 and 64k");
146         ret = -EINVAL;
147         goto fail;
148     }
149 
150     if (header.crypt_method > QCOW_CRYPT_AES) {
151         error_setg(errp, "invalid encryption method in qcow header");
152         ret = -EINVAL;
153         goto fail;
154     }
155     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) {
156         error_setg(errp, "AES cipher not available");
157         ret = -EINVAL;
158         goto fail;
159     }
160     s->crypt_method_header = header.crypt_method;
161     if (s->crypt_method_header) {
162         if (bdrv_uses_whitelist() &&
163             s->crypt_method_header == QCOW_CRYPT_AES) {
164             error_report("qcow built-in AES encryption is deprecated");
165             error_printf("Support for it will be removed in a future release.\n"
166                          "You can use 'qemu-img convert' to switch to an\n"
167                          "unencrypted qcow image, or a LUKS raw image.\n");
168         }
169 
170         bs->encrypted = 1;
171     }
172     s->cluster_bits = header.cluster_bits;
173     s->cluster_size = 1 << s->cluster_bits;
174     s->cluster_sectors = 1 << (s->cluster_bits - 9);
175     s->l2_bits = header.l2_bits;
176     s->l2_size = 1 << s->l2_bits;
177     bs->total_sectors = header.size / 512;
178     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
179 
180     /* read the level 1 table */
181     shift = s->cluster_bits + s->l2_bits;
182     if (header.size > UINT64_MAX - (1LL << shift)) {
183         error_setg(errp, "Image too large");
184         ret = -EINVAL;
185         goto fail;
186     } else {
187         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
188         if (l1_size > INT_MAX / sizeof(uint64_t)) {
189             error_setg(errp, "Image too large");
190             ret = -EINVAL;
191             goto fail;
192         }
193         s->l1_size = l1_size;
194     }
195 
196     s->l1_table_offset = header.l1_table_offset;
197     s->l1_table = g_try_new(uint64_t, s->l1_size);
198     if (s->l1_table == NULL) {
199         error_setg(errp, "Could not allocate memory for L1 table");
200         ret = -ENOMEM;
201         goto fail;
202     }
203 
204     ret = bdrv_pread(bs->file->bs, s->l1_table_offset, s->l1_table,
205                s->l1_size * sizeof(uint64_t));
206     if (ret < 0) {
207         goto fail;
208     }
209 
210     for(i = 0;i < s->l1_size; i++) {
211         be64_to_cpus(&s->l1_table[i]);
212     }
213 
214     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
215     s->l2_cache =
216         qemu_try_blockalign(bs->file->bs,
217                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
218     if (s->l2_cache == NULL) {
219         error_setg(errp, "Could not allocate L2 table cache");
220         ret = -ENOMEM;
221         goto fail;
222     }
223     s->cluster_cache = g_malloc(s->cluster_size);
224     s->cluster_data = g_malloc(s->cluster_size);
225     s->cluster_cache_offset = -1;
226 
227     /* read the backing file name */
228     if (header.backing_file_offset != 0) {
229         len = header.backing_file_size;
230         if (len > 1023 || len >= sizeof(bs->backing_file)) {
231             error_setg(errp, "Backing file name too long");
232             ret = -EINVAL;
233             goto fail;
234         }
235         ret = bdrv_pread(bs->file->bs, header.backing_file_offset,
236                    bs->backing_file, len);
237         if (ret < 0) {
238             goto fail;
239         }
240         bs->backing_file[len] = '\0';
241     }
242 
243     /* Disable migration when qcow images are used */
244     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
245                "does not support live migration",
246                bdrv_get_device_or_node_name(bs));
247     migrate_add_blocker(s->migration_blocker);
248 
249     qemu_co_mutex_init(&s->lock);
250     return 0;
251 
252  fail:
253     g_free(s->l1_table);
254     qemu_vfree(s->l2_cache);
255     g_free(s->cluster_cache);
256     g_free(s->cluster_data);
257     return ret;
258 }
259 
260 
261 /* We have nothing to do for QCOW reopen, stubs just return
262  * success */
263 static int qcow_reopen_prepare(BDRVReopenState *state,
264                                BlockReopenQueue *queue, Error **errp)
265 {
266     return 0;
267 }
268 
269 static int qcow_set_key(BlockDriverState *bs, const char *key)
270 {
271     BDRVQcowState *s = bs->opaque;
272     uint8_t keybuf[16];
273     int len, i;
274     Error *err;
275 
276     memset(keybuf, 0, 16);
277     len = strlen(key);
278     if (len > 16)
279         len = 16;
280     /* XXX: we could compress the chars to 7 bits to increase
281        entropy */
282     for(i = 0;i < len;i++) {
283         keybuf[i] = key[i];
284     }
285     assert(bs->encrypted);
286 
287     qcrypto_cipher_free(s->cipher);
288     s->cipher = qcrypto_cipher_new(
289         QCRYPTO_CIPHER_ALG_AES_128,
290         QCRYPTO_CIPHER_MODE_CBC,
291         keybuf, G_N_ELEMENTS(keybuf),
292         &err);
293 
294     if (!s->cipher) {
295         /* XXX would be nice if errors in this method could
296          * be properly propagate to the caller. Would need
297          * the bdrv_set_key() API signature to be fixed. */
298         error_free(err);
299         return -1;
300     }
301     return 0;
302 }
303 
304 /* The crypt function is compatible with the linux cryptoloop
305    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
306    supported */
307 static int encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
308                            uint8_t *out_buf, const uint8_t *in_buf,
309                            int nb_sectors, bool enc, Error **errp)
310 {
311     union {
312         uint64_t ll[2];
313         uint8_t b[16];
314     } ivec;
315     int i;
316     int ret;
317 
318     for(i = 0; i < nb_sectors; i++) {
319         ivec.ll[0] = cpu_to_le64(sector_num);
320         ivec.ll[1] = 0;
321         if (qcrypto_cipher_setiv(s->cipher,
322                                  ivec.b, G_N_ELEMENTS(ivec.b),
323                                  errp) < 0) {
324             return -1;
325         }
326         if (enc) {
327             ret = qcrypto_cipher_encrypt(s->cipher,
328                                          in_buf,
329                                          out_buf,
330                                          512,
331                                          errp);
332         } else {
333             ret = qcrypto_cipher_decrypt(s->cipher,
334                                          in_buf,
335                                          out_buf,
336                                          512,
337                                          errp);
338         }
339         if (ret < 0) {
340             return -1;
341         }
342         sector_num++;
343         in_buf += 512;
344         out_buf += 512;
345     }
346     return 0;
347 }
348 
349 /* 'allocate' is:
350  *
351  * 0 to not allocate.
352  *
353  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
354  * 'n_end')
355  *
356  * 2 to allocate a compressed cluster of size
357  * 'compressed_size'. 'compressed_size' must be > 0 and <
358  * cluster_size
359  *
360  * return 0 if not allocated.
361  */
362 static uint64_t get_cluster_offset(BlockDriverState *bs,
363                                    uint64_t offset, int allocate,
364                                    int compressed_size,
365                                    int n_start, int n_end)
366 {
367     BDRVQcowState *s = bs->opaque;
368     int min_index, i, j, l1_index, l2_index;
369     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
370     uint32_t min_count;
371     int new_l2_table;
372 
373     l1_index = offset >> (s->l2_bits + s->cluster_bits);
374     l2_offset = s->l1_table[l1_index];
375     new_l2_table = 0;
376     if (!l2_offset) {
377         if (!allocate)
378             return 0;
379         /* allocate a new l2 entry */
380         l2_offset = bdrv_getlength(bs->file->bs);
381         /* round to cluster size */
382         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
383         /* update the L1 entry */
384         s->l1_table[l1_index] = l2_offset;
385         tmp = cpu_to_be64(l2_offset);
386         if (bdrv_pwrite_sync(bs->file->bs,
387                 s->l1_table_offset + l1_index * sizeof(tmp),
388                 &tmp, sizeof(tmp)) < 0)
389             return 0;
390         new_l2_table = 1;
391     }
392     for(i = 0; i < L2_CACHE_SIZE; i++) {
393         if (l2_offset == s->l2_cache_offsets[i]) {
394             /* increment the hit count */
395             if (++s->l2_cache_counts[i] == 0xffffffff) {
396                 for(j = 0; j < L2_CACHE_SIZE; j++) {
397                     s->l2_cache_counts[j] >>= 1;
398                 }
399             }
400             l2_table = s->l2_cache + (i << s->l2_bits);
401             goto found;
402         }
403     }
404     /* not found: load a new entry in the least used one */
405     min_index = 0;
406     min_count = 0xffffffff;
407     for(i = 0; i < L2_CACHE_SIZE; i++) {
408         if (s->l2_cache_counts[i] < min_count) {
409             min_count = s->l2_cache_counts[i];
410             min_index = i;
411         }
412     }
413     l2_table = s->l2_cache + (min_index << s->l2_bits);
414     if (new_l2_table) {
415         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
416         if (bdrv_pwrite_sync(bs->file->bs, l2_offset, l2_table,
417                 s->l2_size * sizeof(uint64_t)) < 0)
418             return 0;
419     } else {
420         if (bdrv_pread(bs->file->bs, l2_offset, l2_table,
421                        s->l2_size * sizeof(uint64_t)) !=
422             s->l2_size * sizeof(uint64_t))
423             return 0;
424     }
425     s->l2_cache_offsets[min_index] = l2_offset;
426     s->l2_cache_counts[min_index] = 1;
427  found:
428     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
429     cluster_offset = be64_to_cpu(l2_table[l2_index]);
430     if (!cluster_offset ||
431         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
432         if (!allocate)
433             return 0;
434         /* allocate a new cluster */
435         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
436             (n_end - n_start) < s->cluster_sectors) {
437             /* if the cluster is already compressed, we must
438                decompress it in the case it is not completely
439                overwritten */
440             if (decompress_cluster(bs, cluster_offset) < 0)
441                 return 0;
442             cluster_offset = bdrv_getlength(bs->file->bs);
443             cluster_offset = (cluster_offset + s->cluster_size - 1) &
444                 ~(s->cluster_size - 1);
445             /* write the cluster content */
446             if (bdrv_pwrite(bs->file->bs, cluster_offset, s->cluster_cache,
447                             s->cluster_size) !=
448                 s->cluster_size)
449                 return -1;
450         } else {
451             cluster_offset = bdrv_getlength(bs->file->bs);
452             if (allocate == 1) {
453                 /* round to cluster size */
454                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
455                     ~(s->cluster_size - 1);
456                 bdrv_truncate(bs->file->bs, cluster_offset + s->cluster_size);
457                 /* if encrypted, we must initialize the cluster
458                    content which won't be written */
459                 if (bs->encrypted &&
460                     (n_end - n_start) < s->cluster_sectors) {
461                     uint64_t start_sect;
462                     assert(s->cipher);
463                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
464                     memset(s->cluster_data + 512, 0x00, 512);
465                     for(i = 0; i < s->cluster_sectors; i++) {
466                         if (i < n_start || i >= n_end) {
467                             Error *err = NULL;
468                             if (encrypt_sectors(s, start_sect + i,
469                                                 s->cluster_data,
470                                                 s->cluster_data + 512, 1,
471                                                 true, &err) < 0) {
472                                 error_free(err);
473                                 errno = EIO;
474                                 return -1;
475                             }
476                             if (bdrv_pwrite(bs->file->bs,
477                                             cluster_offset + i * 512,
478                                             s->cluster_data, 512) != 512)
479                                 return -1;
480                         }
481                     }
482                 }
483             } else if (allocate == 2) {
484                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
485                     (uint64_t)compressed_size << (63 - s->cluster_bits);
486             }
487         }
488         /* update L2 table */
489         tmp = cpu_to_be64(cluster_offset);
490         l2_table[l2_index] = tmp;
491         if (bdrv_pwrite_sync(bs->file->bs, l2_offset + l2_index * sizeof(tmp),
492                 &tmp, sizeof(tmp)) < 0)
493             return 0;
494     }
495     return cluster_offset;
496 }
497 
498 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
499         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
500 {
501     BDRVQcowState *s = bs->opaque;
502     int index_in_cluster, n;
503     uint64_t cluster_offset;
504 
505     qemu_co_mutex_lock(&s->lock);
506     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
507     qemu_co_mutex_unlock(&s->lock);
508     index_in_cluster = sector_num & (s->cluster_sectors - 1);
509     n = s->cluster_sectors - index_in_cluster;
510     if (n > nb_sectors)
511         n = nb_sectors;
512     *pnum = n;
513     if (!cluster_offset) {
514         return 0;
515     }
516     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->cipher) {
517         return BDRV_BLOCK_DATA;
518     }
519     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
520     *file = bs->file->bs;
521     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
522 }
523 
524 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
525                              const uint8_t *buf, int buf_size)
526 {
527     z_stream strm1, *strm = &strm1;
528     int ret, out_len;
529 
530     memset(strm, 0, sizeof(*strm));
531 
532     strm->next_in = (uint8_t *)buf;
533     strm->avail_in = buf_size;
534     strm->next_out = out_buf;
535     strm->avail_out = out_buf_size;
536 
537     ret = inflateInit2(strm, -12);
538     if (ret != Z_OK)
539         return -1;
540     ret = inflate(strm, Z_FINISH);
541     out_len = strm->next_out - out_buf;
542     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
543         out_len != out_buf_size) {
544         inflateEnd(strm);
545         return -1;
546     }
547     inflateEnd(strm);
548     return 0;
549 }
550 
551 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
552 {
553     BDRVQcowState *s = bs->opaque;
554     int ret, csize;
555     uint64_t coffset;
556 
557     coffset = cluster_offset & s->cluster_offset_mask;
558     if (s->cluster_cache_offset != coffset) {
559         csize = cluster_offset >> (63 - s->cluster_bits);
560         csize &= (s->cluster_size - 1);
561         ret = bdrv_pread(bs->file->bs, coffset, s->cluster_data, csize);
562         if (ret != csize)
563             return -1;
564         if (decompress_buffer(s->cluster_cache, s->cluster_size,
565                               s->cluster_data, csize) < 0) {
566             return -1;
567         }
568         s->cluster_cache_offset = coffset;
569     }
570     return 0;
571 }
572 
573 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
574                          int nb_sectors, QEMUIOVector *qiov)
575 {
576     BDRVQcowState *s = bs->opaque;
577     int index_in_cluster;
578     int ret = 0, n;
579     uint64_t cluster_offset;
580     struct iovec hd_iov;
581     QEMUIOVector hd_qiov;
582     uint8_t *buf;
583     void *orig_buf;
584     Error *err = NULL;
585 
586     if (qiov->niov > 1) {
587         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
588         if (buf == NULL) {
589             return -ENOMEM;
590         }
591     } else {
592         orig_buf = NULL;
593         buf = (uint8_t *)qiov->iov->iov_base;
594     }
595 
596     qemu_co_mutex_lock(&s->lock);
597 
598     while (nb_sectors != 0) {
599         /* prepare next request */
600         cluster_offset = get_cluster_offset(bs, sector_num << 9,
601                                                  0, 0, 0, 0);
602         index_in_cluster = sector_num & (s->cluster_sectors - 1);
603         n = s->cluster_sectors - index_in_cluster;
604         if (n > nb_sectors) {
605             n = nb_sectors;
606         }
607 
608         if (!cluster_offset) {
609             if (bs->backing) {
610                 /* read from the base image */
611                 hd_iov.iov_base = (void *)buf;
612                 hd_iov.iov_len = n * 512;
613                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
614                 qemu_co_mutex_unlock(&s->lock);
615                 ret = bdrv_co_readv(bs->backing->bs, sector_num,
616                                     n, &hd_qiov);
617                 qemu_co_mutex_lock(&s->lock);
618                 if (ret < 0) {
619                     goto fail;
620                 }
621             } else {
622                 /* Note: in this case, no need to wait */
623                 memset(buf, 0, 512 * n);
624             }
625         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
626             /* add AIO support for compressed blocks ? */
627             if (decompress_cluster(bs, cluster_offset) < 0) {
628                 goto fail;
629             }
630             memcpy(buf,
631                    s->cluster_cache + index_in_cluster * 512, 512 * n);
632         } else {
633             if ((cluster_offset & 511) != 0) {
634                 goto fail;
635             }
636             hd_iov.iov_base = (void *)buf;
637             hd_iov.iov_len = n * 512;
638             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
639             qemu_co_mutex_unlock(&s->lock);
640             ret = bdrv_co_readv(bs->file->bs,
641                                 (cluster_offset >> 9) + index_in_cluster,
642                                 n, &hd_qiov);
643             qemu_co_mutex_lock(&s->lock);
644             if (ret < 0) {
645                 break;
646             }
647             if (bs->encrypted) {
648                 assert(s->cipher);
649                 if (encrypt_sectors(s, sector_num, buf, buf,
650                                     n, false, &err) < 0) {
651                     goto fail;
652                 }
653             }
654         }
655         ret = 0;
656 
657         nb_sectors -= n;
658         sector_num += n;
659         buf += n * 512;
660     }
661 
662 done:
663     qemu_co_mutex_unlock(&s->lock);
664 
665     if (qiov->niov > 1) {
666         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
667         qemu_vfree(orig_buf);
668     }
669 
670     return ret;
671 
672 fail:
673     error_free(err);
674     ret = -EIO;
675     goto done;
676 }
677 
678 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
679                           int nb_sectors, QEMUIOVector *qiov)
680 {
681     BDRVQcowState *s = bs->opaque;
682     int index_in_cluster;
683     uint64_t cluster_offset;
684     const uint8_t *src_buf;
685     int ret = 0, n;
686     uint8_t *cluster_data = NULL;
687     struct iovec hd_iov;
688     QEMUIOVector hd_qiov;
689     uint8_t *buf;
690     void *orig_buf;
691 
692     s->cluster_cache_offset = -1; /* disable compressed cache */
693 
694     if (qiov->niov > 1) {
695         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
696         if (buf == NULL) {
697             return -ENOMEM;
698         }
699         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
700     } else {
701         orig_buf = NULL;
702         buf = (uint8_t *)qiov->iov->iov_base;
703     }
704 
705     qemu_co_mutex_lock(&s->lock);
706 
707     while (nb_sectors != 0) {
708 
709         index_in_cluster = sector_num & (s->cluster_sectors - 1);
710         n = s->cluster_sectors - index_in_cluster;
711         if (n > nb_sectors) {
712             n = nb_sectors;
713         }
714         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
715                                             index_in_cluster,
716                                             index_in_cluster + n);
717         if (!cluster_offset || (cluster_offset & 511) != 0) {
718             ret = -EIO;
719             break;
720         }
721         if (bs->encrypted) {
722             Error *err = NULL;
723             assert(s->cipher);
724             if (!cluster_data) {
725                 cluster_data = g_malloc0(s->cluster_size);
726             }
727             if (encrypt_sectors(s, sector_num, cluster_data, buf,
728                                 n, true, &err) < 0) {
729                 error_free(err);
730                 ret = -EIO;
731                 break;
732             }
733             src_buf = cluster_data;
734         } else {
735             src_buf = buf;
736         }
737 
738         hd_iov.iov_base = (void *)src_buf;
739         hd_iov.iov_len = n * 512;
740         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
741         qemu_co_mutex_unlock(&s->lock);
742         ret = bdrv_co_writev(bs->file->bs,
743                              (cluster_offset >> 9) + index_in_cluster,
744                              n, &hd_qiov);
745         qemu_co_mutex_lock(&s->lock);
746         if (ret < 0) {
747             break;
748         }
749         ret = 0;
750 
751         nb_sectors -= n;
752         sector_num += n;
753         buf += n * 512;
754     }
755     qemu_co_mutex_unlock(&s->lock);
756 
757     if (qiov->niov > 1) {
758         qemu_vfree(orig_buf);
759     }
760     g_free(cluster_data);
761 
762     return ret;
763 }
764 
765 static void qcow_close(BlockDriverState *bs)
766 {
767     BDRVQcowState *s = bs->opaque;
768 
769     qcrypto_cipher_free(s->cipher);
770     s->cipher = NULL;
771     g_free(s->l1_table);
772     qemu_vfree(s->l2_cache);
773     g_free(s->cluster_cache);
774     g_free(s->cluster_data);
775 
776     migrate_del_blocker(s->migration_blocker);
777     error_free(s->migration_blocker);
778 }
779 
780 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
781 {
782     int header_size, backing_filename_len, l1_size, shift, i;
783     QCowHeader header;
784     uint8_t *tmp;
785     int64_t total_size = 0;
786     char *backing_file = NULL;
787     int flags = 0;
788     Error *local_err = NULL;
789     int ret;
790     BlockBackend *qcow_blk;
791 
792     /* Read out options */
793     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
794                           BDRV_SECTOR_SIZE);
795     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
796     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
797         flags |= BLOCK_FLAG_ENCRYPT;
798     }
799 
800     ret = bdrv_create_file(filename, opts, &local_err);
801     if (ret < 0) {
802         error_propagate(errp, local_err);
803         goto cleanup;
804     }
805 
806     qcow_blk = blk_new_open(filename, NULL, NULL,
807                             BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
808     if (qcow_blk == NULL) {
809         error_propagate(errp, local_err);
810         ret = -EIO;
811         goto cleanup;
812     }
813 
814     blk_set_allow_write_beyond_eof(qcow_blk, true);
815 
816     ret = blk_truncate(qcow_blk, 0);
817     if (ret < 0) {
818         goto exit;
819     }
820 
821     memset(&header, 0, sizeof(header));
822     header.magic = cpu_to_be32(QCOW_MAGIC);
823     header.version = cpu_to_be32(QCOW_VERSION);
824     header.size = cpu_to_be64(total_size);
825     header_size = sizeof(header);
826     backing_filename_len = 0;
827     if (backing_file) {
828         if (strcmp(backing_file, "fat:")) {
829             header.backing_file_offset = cpu_to_be64(header_size);
830             backing_filename_len = strlen(backing_file);
831             header.backing_file_size = cpu_to_be32(backing_filename_len);
832             header_size += backing_filename_len;
833         } else {
834             /* special backing file for vvfat */
835             backing_file = NULL;
836         }
837         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
838                                     unmodified sectors */
839         header.l2_bits = 12; /* 32 KB L2 tables */
840     } else {
841         header.cluster_bits = 12; /* 4 KB clusters */
842         header.l2_bits = 9; /* 4 KB L2 tables */
843     }
844     header_size = (header_size + 7) & ~7;
845     shift = header.cluster_bits + header.l2_bits;
846     l1_size = (total_size + (1LL << shift) - 1) >> shift;
847 
848     header.l1_table_offset = cpu_to_be64(header_size);
849     if (flags & BLOCK_FLAG_ENCRYPT) {
850         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
851     } else {
852         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
853     }
854 
855     /* write all the data */
856     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header));
857     if (ret != sizeof(header)) {
858         goto exit;
859     }
860 
861     if (backing_file) {
862         ret = blk_pwrite(qcow_blk, sizeof(header),
863             backing_file, backing_filename_len);
864         if (ret != backing_filename_len) {
865             goto exit;
866         }
867     }
868 
869     tmp = g_malloc0(BDRV_SECTOR_SIZE);
870     for (i = 0; i < ((sizeof(uint64_t)*l1_size + BDRV_SECTOR_SIZE - 1)/
871         BDRV_SECTOR_SIZE); i++) {
872         ret = blk_pwrite(qcow_blk, header_size +
873             BDRV_SECTOR_SIZE*i, tmp, BDRV_SECTOR_SIZE);
874         if (ret != BDRV_SECTOR_SIZE) {
875             g_free(tmp);
876             goto exit;
877         }
878     }
879 
880     g_free(tmp);
881     ret = 0;
882 exit:
883     blk_unref(qcow_blk);
884 cleanup:
885     g_free(backing_file);
886     return ret;
887 }
888 
889 static int qcow_make_empty(BlockDriverState *bs)
890 {
891     BDRVQcowState *s = bs->opaque;
892     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
893     int ret;
894 
895     memset(s->l1_table, 0, l1_length);
896     if (bdrv_pwrite_sync(bs->file->bs, s->l1_table_offset, s->l1_table,
897             l1_length) < 0)
898         return -1;
899     ret = bdrv_truncate(bs->file->bs, s->l1_table_offset + l1_length);
900     if (ret < 0)
901         return ret;
902 
903     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
904     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
905     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
906 
907     return 0;
908 }
909 
910 /* XXX: put compressed sectors first, then all the cluster aligned
911    tables to avoid losing bytes in alignment */
912 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
913                                  const uint8_t *buf, int nb_sectors)
914 {
915     BDRVQcowState *s = bs->opaque;
916     z_stream strm;
917     int ret, out_len;
918     uint8_t *out_buf;
919     uint64_t cluster_offset;
920 
921     if (nb_sectors != s->cluster_sectors) {
922         ret = -EINVAL;
923 
924         /* Zero-pad last write if image size is not cluster aligned */
925         if (sector_num + nb_sectors == bs->total_sectors &&
926             nb_sectors < s->cluster_sectors) {
927             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
928             memset(pad_buf, 0, s->cluster_size);
929             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
930             ret = qcow_write_compressed(bs, sector_num,
931                                         pad_buf, s->cluster_sectors);
932             qemu_vfree(pad_buf);
933         }
934         return ret;
935     }
936 
937     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
938 
939     /* best compression, small window, no zlib header */
940     memset(&strm, 0, sizeof(strm));
941     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
942                        Z_DEFLATED, -12,
943                        9, Z_DEFAULT_STRATEGY);
944     if (ret != 0) {
945         ret = -EINVAL;
946         goto fail;
947     }
948 
949     strm.avail_in = s->cluster_size;
950     strm.next_in = (uint8_t *)buf;
951     strm.avail_out = s->cluster_size;
952     strm.next_out = out_buf;
953 
954     ret = deflate(&strm, Z_FINISH);
955     if (ret != Z_STREAM_END && ret != Z_OK) {
956         deflateEnd(&strm);
957         ret = -EINVAL;
958         goto fail;
959     }
960     out_len = strm.next_out - out_buf;
961 
962     deflateEnd(&strm);
963 
964     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
965         /* could not compress: write normal cluster */
966         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
967         if (ret < 0) {
968             goto fail;
969         }
970     } else {
971         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
972                                             out_len, 0, 0);
973         if (cluster_offset == 0) {
974             ret = -EIO;
975             goto fail;
976         }
977 
978         cluster_offset &= s->cluster_offset_mask;
979         ret = bdrv_pwrite(bs->file->bs, cluster_offset, out_buf, out_len);
980         if (ret < 0) {
981             goto fail;
982         }
983     }
984 
985     ret = 0;
986 fail:
987     g_free(out_buf);
988     return ret;
989 }
990 
991 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
992 {
993     BDRVQcowState *s = bs->opaque;
994     bdi->cluster_size = s->cluster_size;
995     return 0;
996 }
997 
998 static QemuOptsList qcow_create_opts = {
999     .name = "qcow-create-opts",
1000     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1001     .desc = {
1002         {
1003             .name = BLOCK_OPT_SIZE,
1004             .type = QEMU_OPT_SIZE,
1005             .help = "Virtual disk size"
1006         },
1007         {
1008             .name = BLOCK_OPT_BACKING_FILE,
1009             .type = QEMU_OPT_STRING,
1010             .help = "File name of a base image"
1011         },
1012         {
1013             .name = BLOCK_OPT_ENCRYPT,
1014             .type = QEMU_OPT_BOOL,
1015             .help = "Encrypt the image",
1016             .def_value_str = "off"
1017         },
1018         { /* end of list */ }
1019     }
1020 };
1021 
1022 static BlockDriver bdrv_qcow = {
1023     .format_name	= "qcow",
1024     .instance_size	= sizeof(BDRVQcowState),
1025     .bdrv_probe		= qcow_probe,
1026     .bdrv_open		= qcow_open,
1027     .bdrv_close		= qcow_close,
1028     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1029     .bdrv_create            = qcow_create,
1030     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1031     .supports_backing       = true,
1032 
1033     .bdrv_co_readv          = qcow_co_readv,
1034     .bdrv_co_writev         = qcow_co_writev,
1035     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1036 
1037     .bdrv_set_key           = qcow_set_key,
1038     .bdrv_make_empty        = qcow_make_empty,
1039     .bdrv_write_compressed  = qcow_write_compressed,
1040     .bdrv_get_info          = qcow_get_info,
1041 
1042     .create_opts            = &qcow_create_opts,
1043 };
1044 
1045 static void bdrv_qcow_init(void)
1046 {
1047     bdrv_register(&bdrv_qcow);
1048 }
1049 
1050 block_init(bdrv_qcow_init);
1051