xref: /openbmc/qemu/block/qcow.c (revision fe44dc91)
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     Error *local_err = NULL;
108 
109     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
110     if (ret < 0) {
111         goto fail;
112     }
113     be32_to_cpus(&header.magic);
114     be32_to_cpus(&header.version);
115     be64_to_cpus(&header.backing_file_offset);
116     be32_to_cpus(&header.backing_file_size);
117     be32_to_cpus(&header.mtime);
118     be64_to_cpus(&header.size);
119     be32_to_cpus(&header.crypt_method);
120     be64_to_cpus(&header.l1_table_offset);
121 
122     if (header.magic != QCOW_MAGIC) {
123         error_setg(errp, "Image not in qcow format");
124         ret = -EINVAL;
125         goto fail;
126     }
127     if (header.version != QCOW_VERSION) {
128         error_setg(errp, "Unsupported qcow version %" PRIu32, header.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                                  QCRYPTO_CIPHER_MODE_CBC)) {
159         error_setg(errp, "AES cipher not available");
160         ret = -EINVAL;
161         goto fail;
162     }
163     s->crypt_method_header = header.crypt_method;
164     if (s->crypt_method_header) {
165         if (bdrv_uses_whitelist() &&
166             s->crypt_method_header == QCOW_CRYPT_AES) {
167             error_setg(errp,
168                        "Use of AES-CBC encrypted qcow images is no longer "
169                        "supported in system emulators");
170             error_append_hint(errp,
171                               "You can use 'qemu-img convert' to convert your "
172                               "image to an alternative supported format, such "
173                               "as unencrypted qcow, or raw with the LUKS "
174                               "format instead.\n");
175             ret = -ENOSYS;
176             goto fail;
177         }
178 
179         bs->encrypted = true;
180     }
181     s->cluster_bits = header.cluster_bits;
182     s->cluster_size = 1 << s->cluster_bits;
183     s->cluster_sectors = 1 << (s->cluster_bits - 9);
184     s->l2_bits = header.l2_bits;
185     s->l2_size = 1 << s->l2_bits;
186     bs->total_sectors = header.size / 512;
187     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
188 
189     /* read the level 1 table */
190     shift = s->cluster_bits + s->l2_bits;
191     if (header.size > UINT64_MAX - (1LL << shift)) {
192         error_setg(errp, "Image too large");
193         ret = -EINVAL;
194         goto fail;
195     } else {
196         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
197         if (l1_size > INT_MAX / sizeof(uint64_t)) {
198             error_setg(errp, "Image too large");
199             ret = -EINVAL;
200             goto fail;
201         }
202         s->l1_size = l1_size;
203     }
204 
205     s->l1_table_offset = header.l1_table_offset;
206     s->l1_table = g_try_new(uint64_t, s->l1_size);
207     if (s->l1_table == NULL) {
208         error_setg(errp, "Could not allocate memory for L1 table");
209         ret = -ENOMEM;
210         goto fail;
211     }
212 
213     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
214                s->l1_size * sizeof(uint64_t));
215     if (ret < 0) {
216         goto fail;
217     }
218 
219     for(i = 0;i < s->l1_size; i++) {
220         be64_to_cpus(&s->l1_table[i]);
221     }
222 
223     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
224     s->l2_cache =
225         qemu_try_blockalign(bs->file->bs,
226                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
227     if (s->l2_cache == NULL) {
228         error_setg(errp, "Could not allocate L2 table cache");
229         ret = -ENOMEM;
230         goto fail;
231     }
232     s->cluster_cache = g_malloc(s->cluster_size);
233     s->cluster_data = g_malloc(s->cluster_size);
234     s->cluster_cache_offset = -1;
235 
236     /* read the backing file name */
237     if (header.backing_file_offset != 0) {
238         len = header.backing_file_size;
239         if (len > 1023 || len >= sizeof(bs->backing_file)) {
240             error_setg(errp, "Backing file name too long");
241             ret = -EINVAL;
242             goto fail;
243         }
244         ret = bdrv_pread(bs->file, header.backing_file_offset,
245                    bs->backing_file, len);
246         if (ret < 0) {
247             goto fail;
248         }
249         bs->backing_file[len] = '\0';
250     }
251 
252     /* Disable migration when qcow images are used */
253     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
254                "does not support live migration",
255                bdrv_get_device_or_node_name(bs));
256     ret = migrate_add_blocker(s->migration_blocker, &local_err);
257     if (local_err) {
258         error_propagate(errp, local_err);
259         error_free(s->migration_blocker);
260         goto fail;
261     }
262 
263     qemu_co_mutex_init(&s->lock);
264     return 0;
265 
266  fail:
267     g_free(s->l1_table);
268     qemu_vfree(s->l2_cache);
269     g_free(s->cluster_cache);
270     g_free(s->cluster_data);
271     return ret;
272 }
273 
274 
275 /* We have nothing to do for QCOW reopen, stubs just return
276  * success */
277 static int qcow_reopen_prepare(BDRVReopenState *state,
278                                BlockReopenQueue *queue, Error **errp)
279 {
280     return 0;
281 }
282 
283 static int qcow_set_key(BlockDriverState *bs, const char *key)
284 {
285     BDRVQcowState *s = bs->opaque;
286     uint8_t keybuf[16];
287     int len, i;
288     Error *err;
289 
290     memset(keybuf, 0, 16);
291     len = strlen(key);
292     if (len > 16)
293         len = 16;
294     /* XXX: we could compress the chars to 7 bits to increase
295        entropy */
296     for(i = 0;i < len;i++) {
297         keybuf[i] = key[i];
298     }
299     assert(bs->encrypted);
300 
301     qcrypto_cipher_free(s->cipher);
302     s->cipher = qcrypto_cipher_new(
303         QCRYPTO_CIPHER_ALG_AES_128,
304         QCRYPTO_CIPHER_MODE_CBC,
305         keybuf, G_N_ELEMENTS(keybuf),
306         &err);
307 
308     if (!s->cipher) {
309         /* XXX would be nice if errors in this method could
310          * be properly propagate to the caller. Would need
311          * the bdrv_set_key() API signature to be fixed. */
312         error_free(err);
313         return -1;
314     }
315     return 0;
316 }
317 
318 /* The crypt function is compatible with the linux cryptoloop
319    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
320    supported */
321 static int encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
322                            uint8_t *out_buf, const uint8_t *in_buf,
323                            int nb_sectors, bool enc, Error **errp)
324 {
325     union {
326         uint64_t ll[2];
327         uint8_t b[16];
328     } ivec;
329     int i;
330     int ret;
331 
332     for(i = 0; i < nb_sectors; i++) {
333         ivec.ll[0] = cpu_to_le64(sector_num);
334         ivec.ll[1] = 0;
335         if (qcrypto_cipher_setiv(s->cipher,
336                                  ivec.b, G_N_ELEMENTS(ivec.b),
337                                  errp) < 0) {
338             return -1;
339         }
340         if (enc) {
341             ret = qcrypto_cipher_encrypt(s->cipher,
342                                          in_buf,
343                                          out_buf,
344                                          512,
345                                          errp);
346         } else {
347             ret = qcrypto_cipher_decrypt(s->cipher,
348                                          in_buf,
349                                          out_buf,
350                                          512,
351                                          errp);
352         }
353         if (ret < 0) {
354             return -1;
355         }
356         sector_num++;
357         in_buf += 512;
358         out_buf += 512;
359     }
360     return 0;
361 }
362 
363 /* 'allocate' is:
364  *
365  * 0 to not allocate.
366  *
367  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
368  * 'n_end')
369  *
370  * 2 to allocate a compressed cluster of size
371  * 'compressed_size'. 'compressed_size' must be > 0 and <
372  * cluster_size
373  *
374  * return 0 if not allocated.
375  */
376 static uint64_t get_cluster_offset(BlockDriverState *bs,
377                                    uint64_t offset, int allocate,
378                                    int compressed_size,
379                                    int n_start, int n_end)
380 {
381     BDRVQcowState *s = bs->opaque;
382     int min_index, i, j, l1_index, l2_index;
383     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
384     uint32_t min_count;
385     int new_l2_table;
386 
387     l1_index = offset >> (s->l2_bits + s->cluster_bits);
388     l2_offset = s->l1_table[l1_index];
389     new_l2_table = 0;
390     if (!l2_offset) {
391         if (!allocate)
392             return 0;
393         /* allocate a new l2 entry */
394         l2_offset = bdrv_getlength(bs->file->bs);
395         /* round to cluster size */
396         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
397         /* update the L1 entry */
398         s->l1_table[l1_index] = l2_offset;
399         tmp = cpu_to_be64(l2_offset);
400         if (bdrv_pwrite_sync(bs->file,
401                 s->l1_table_offset + l1_index * sizeof(tmp),
402                 &tmp, sizeof(tmp)) < 0)
403             return 0;
404         new_l2_table = 1;
405     }
406     for(i = 0; i < L2_CACHE_SIZE; i++) {
407         if (l2_offset == s->l2_cache_offsets[i]) {
408             /* increment the hit count */
409             if (++s->l2_cache_counts[i] == 0xffffffff) {
410                 for(j = 0; j < L2_CACHE_SIZE; j++) {
411                     s->l2_cache_counts[j] >>= 1;
412                 }
413             }
414             l2_table = s->l2_cache + (i << s->l2_bits);
415             goto found;
416         }
417     }
418     /* not found: load a new entry in the least used one */
419     min_index = 0;
420     min_count = 0xffffffff;
421     for(i = 0; i < L2_CACHE_SIZE; i++) {
422         if (s->l2_cache_counts[i] < min_count) {
423             min_count = s->l2_cache_counts[i];
424             min_index = i;
425         }
426     }
427     l2_table = s->l2_cache + (min_index << s->l2_bits);
428     if (new_l2_table) {
429         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
430         if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
431                 s->l2_size * sizeof(uint64_t)) < 0)
432             return 0;
433     } else {
434         if (bdrv_pread(bs->file, l2_offset, l2_table,
435                        s->l2_size * sizeof(uint64_t)) !=
436             s->l2_size * sizeof(uint64_t))
437             return 0;
438     }
439     s->l2_cache_offsets[min_index] = l2_offset;
440     s->l2_cache_counts[min_index] = 1;
441  found:
442     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
443     cluster_offset = be64_to_cpu(l2_table[l2_index]);
444     if (!cluster_offset ||
445         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
446         if (!allocate)
447             return 0;
448         /* allocate a new cluster */
449         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
450             (n_end - n_start) < s->cluster_sectors) {
451             /* if the cluster is already compressed, we must
452                decompress it in the case it is not completely
453                overwritten */
454             if (decompress_cluster(bs, cluster_offset) < 0)
455                 return 0;
456             cluster_offset = bdrv_getlength(bs->file->bs);
457             cluster_offset = (cluster_offset + s->cluster_size - 1) &
458                 ~(s->cluster_size - 1);
459             /* write the cluster content */
460             if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
461                             s->cluster_size) !=
462                 s->cluster_size)
463                 return -1;
464         } else {
465             cluster_offset = bdrv_getlength(bs->file->bs);
466             if (allocate == 1) {
467                 /* round to cluster size */
468                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
469                     ~(s->cluster_size - 1);
470                 bdrv_truncate(bs->file->bs, cluster_offset + s->cluster_size);
471                 /* if encrypted, we must initialize the cluster
472                    content which won't be written */
473                 if (bs->encrypted &&
474                     (n_end - n_start) < s->cluster_sectors) {
475                     uint64_t start_sect;
476                     assert(s->cipher);
477                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
478                     memset(s->cluster_data + 512, 0x00, 512);
479                     for(i = 0; i < s->cluster_sectors; i++) {
480                         if (i < n_start || i >= n_end) {
481                             Error *err = NULL;
482                             if (encrypt_sectors(s, start_sect + i,
483                                                 s->cluster_data,
484                                                 s->cluster_data + 512, 1,
485                                                 true, &err) < 0) {
486                                 error_free(err);
487                                 errno = EIO;
488                                 return -1;
489                             }
490                             if (bdrv_pwrite(bs->file,
491                                             cluster_offset + i * 512,
492                                             s->cluster_data, 512) != 512)
493                                 return -1;
494                         }
495                     }
496                 }
497             } else if (allocate == 2) {
498                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
499                     (uint64_t)compressed_size << (63 - s->cluster_bits);
500             }
501         }
502         /* update L2 table */
503         tmp = cpu_to_be64(cluster_offset);
504         l2_table[l2_index] = tmp;
505         if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
506                 &tmp, sizeof(tmp)) < 0)
507             return 0;
508     }
509     return cluster_offset;
510 }
511 
512 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
513         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
514 {
515     BDRVQcowState *s = bs->opaque;
516     int index_in_cluster, n;
517     uint64_t cluster_offset;
518 
519     qemu_co_mutex_lock(&s->lock);
520     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
521     qemu_co_mutex_unlock(&s->lock);
522     index_in_cluster = sector_num & (s->cluster_sectors - 1);
523     n = s->cluster_sectors - index_in_cluster;
524     if (n > nb_sectors)
525         n = nb_sectors;
526     *pnum = n;
527     if (!cluster_offset) {
528         return 0;
529     }
530     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->cipher) {
531         return BDRV_BLOCK_DATA;
532     }
533     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
534     *file = bs->file->bs;
535     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
536 }
537 
538 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
539                              const uint8_t *buf, int buf_size)
540 {
541     z_stream strm1, *strm = &strm1;
542     int ret, out_len;
543 
544     memset(strm, 0, sizeof(*strm));
545 
546     strm->next_in = (uint8_t *)buf;
547     strm->avail_in = buf_size;
548     strm->next_out = out_buf;
549     strm->avail_out = out_buf_size;
550 
551     ret = inflateInit2(strm, -12);
552     if (ret != Z_OK)
553         return -1;
554     ret = inflate(strm, Z_FINISH);
555     out_len = strm->next_out - out_buf;
556     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
557         out_len != out_buf_size) {
558         inflateEnd(strm);
559         return -1;
560     }
561     inflateEnd(strm);
562     return 0;
563 }
564 
565 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
566 {
567     BDRVQcowState *s = bs->opaque;
568     int ret, csize;
569     uint64_t coffset;
570 
571     coffset = cluster_offset & s->cluster_offset_mask;
572     if (s->cluster_cache_offset != coffset) {
573         csize = cluster_offset >> (63 - s->cluster_bits);
574         csize &= (s->cluster_size - 1);
575         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
576         if (ret != csize)
577             return -1;
578         if (decompress_buffer(s->cluster_cache, s->cluster_size,
579                               s->cluster_data, csize) < 0) {
580             return -1;
581         }
582         s->cluster_cache_offset = coffset;
583     }
584     return 0;
585 }
586 
587 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
588                          int nb_sectors, QEMUIOVector *qiov)
589 {
590     BDRVQcowState *s = bs->opaque;
591     int index_in_cluster;
592     int ret = 0, n;
593     uint64_t cluster_offset;
594     struct iovec hd_iov;
595     QEMUIOVector hd_qiov;
596     uint8_t *buf;
597     void *orig_buf;
598     Error *err = NULL;
599 
600     if (qiov->niov > 1) {
601         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
602         if (buf == NULL) {
603             return -ENOMEM;
604         }
605     } else {
606         orig_buf = NULL;
607         buf = (uint8_t *)qiov->iov->iov_base;
608     }
609 
610     qemu_co_mutex_lock(&s->lock);
611 
612     while (nb_sectors != 0) {
613         /* prepare next request */
614         cluster_offset = get_cluster_offset(bs, sector_num << 9,
615                                                  0, 0, 0, 0);
616         index_in_cluster = sector_num & (s->cluster_sectors - 1);
617         n = s->cluster_sectors - index_in_cluster;
618         if (n > nb_sectors) {
619             n = nb_sectors;
620         }
621 
622         if (!cluster_offset) {
623             if (bs->backing) {
624                 /* read from the base image */
625                 hd_iov.iov_base = (void *)buf;
626                 hd_iov.iov_len = n * 512;
627                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
628                 qemu_co_mutex_unlock(&s->lock);
629                 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov);
630                 qemu_co_mutex_lock(&s->lock);
631                 if (ret < 0) {
632                     goto fail;
633                 }
634             } else {
635                 /* Note: in this case, no need to wait */
636                 memset(buf, 0, 512 * n);
637             }
638         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
639             /* add AIO support for compressed blocks ? */
640             if (decompress_cluster(bs, cluster_offset) < 0) {
641                 goto fail;
642             }
643             memcpy(buf,
644                    s->cluster_cache + index_in_cluster * 512, 512 * n);
645         } else {
646             if ((cluster_offset & 511) != 0) {
647                 goto fail;
648             }
649             hd_iov.iov_base = (void *)buf;
650             hd_iov.iov_len = n * 512;
651             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
652             qemu_co_mutex_unlock(&s->lock);
653             ret = bdrv_co_readv(bs->file,
654                                 (cluster_offset >> 9) + index_in_cluster,
655                                 n, &hd_qiov);
656             qemu_co_mutex_lock(&s->lock);
657             if (ret < 0) {
658                 break;
659             }
660             if (bs->encrypted) {
661                 assert(s->cipher);
662                 if (encrypt_sectors(s, sector_num, buf, buf,
663                                     n, false, &err) < 0) {
664                     goto fail;
665                 }
666             }
667         }
668         ret = 0;
669 
670         nb_sectors -= n;
671         sector_num += n;
672         buf += n * 512;
673     }
674 
675 done:
676     qemu_co_mutex_unlock(&s->lock);
677 
678     if (qiov->niov > 1) {
679         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
680         qemu_vfree(orig_buf);
681     }
682 
683     return ret;
684 
685 fail:
686     error_free(err);
687     ret = -EIO;
688     goto done;
689 }
690 
691 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
692                           int nb_sectors, QEMUIOVector *qiov)
693 {
694     BDRVQcowState *s = bs->opaque;
695     int index_in_cluster;
696     uint64_t cluster_offset;
697     const uint8_t *src_buf;
698     int ret = 0, n;
699     uint8_t *cluster_data = NULL;
700     struct iovec hd_iov;
701     QEMUIOVector hd_qiov;
702     uint8_t *buf;
703     void *orig_buf;
704 
705     s->cluster_cache_offset = -1; /* disable compressed cache */
706 
707     if (qiov->niov > 1) {
708         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
709         if (buf == NULL) {
710             return -ENOMEM;
711         }
712         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
713     } else {
714         orig_buf = NULL;
715         buf = (uint8_t *)qiov->iov->iov_base;
716     }
717 
718     qemu_co_mutex_lock(&s->lock);
719 
720     while (nb_sectors != 0) {
721 
722         index_in_cluster = sector_num & (s->cluster_sectors - 1);
723         n = s->cluster_sectors - index_in_cluster;
724         if (n > nb_sectors) {
725             n = nb_sectors;
726         }
727         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
728                                             index_in_cluster,
729                                             index_in_cluster + n);
730         if (!cluster_offset || (cluster_offset & 511) != 0) {
731             ret = -EIO;
732             break;
733         }
734         if (bs->encrypted) {
735             Error *err = NULL;
736             assert(s->cipher);
737             if (!cluster_data) {
738                 cluster_data = g_malloc0(s->cluster_size);
739             }
740             if (encrypt_sectors(s, sector_num, cluster_data, buf,
741                                 n, true, &err) < 0) {
742                 error_free(err);
743                 ret = -EIO;
744                 break;
745             }
746             src_buf = cluster_data;
747         } else {
748             src_buf = buf;
749         }
750 
751         hd_iov.iov_base = (void *)src_buf;
752         hd_iov.iov_len = n * 512;
753         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
754         qemu_co_mutex_unlock(&s->lock);
755         ret = bdrv_co_writev(bs->file,
756                              (cluster_offset >> 9) + index_in_cluster,
757                              n, &hd_qiov);
758         qemu_co_mutex_lock(&s->lock);
759         if (ret < 0) {
760             break;
761         }
762         ret = 0;
763 
764         nb_sectors -= n;
765         sector_num += n;
766         buf += n * 512;
767     }
768     qemu_co_mutex_unlock(&s->lock);
769 
770     if (qiov->niov > 1) {
771         qemu_vfree(orig_buf);
772     }
773     g_free(cluster_data);
774 
775     return ret;
776 }
777 
778 static void qcow_close(BlockDriverState *bs)
779 {
780     BDRVQcowState *s = bs->opaque;
781 
782     qcrypto_cipher_free(s->cipher);
783     s->cipher = NULL;
784     g_free(s->l1_table);
785     qemu_vfree(s->l2_cache);
786     g_free(s->cluster_cache);
787     g_free(s->cluster_data);
788 
789     migrate_del_blocker(s->migration_blocker);
790     error_free(s->migration_blocker);
791 }
792 
793 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
794 {
795     int header_size, backing_filename_len, l1_size, shift, i;
796     QCowHeader header;
797     uint8_t *tmp;
798     int64_t total_size = 0;
799     char *backing_file = NULL;
800     int flags = 0;
801     Error *local_err = NULL;
802     int ret;
803     BlockBackend *qcow_blk;
804 
805     /* Read out options */
806     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
807                           BDRV_SECTOR_SIZE);
808     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
809     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
810         flags |= BLOCK_FLAG_ENCRYPT;
811     }
812 
813     ret = bdrv_create_file(filename, opts, &local_err);
814     if (ret < 0) {
815         error_propagate(errp, local_err);
816         goto cleanup;
817     }
818 
819     qcow_blk = blk_new_open(filename, NULL, NULL,
820                             BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
821     if (qcow_blk == NULL) {
822         error_propagate(errp, local_err);
823         ret = -EIO;
824         goto cleanup;
825     }
826 
827     blk_set_allow_write_beyond_eof(qcow_blk, true);
828 
829     ret = blk_truncate(qcow_blk, 0);
830     if (ret < 0) {
831         goto exit;
832     }
833 
834     memset(&header, 0, sizeof(header));
835     header.magic = cpu_to_be32(QCOW_MAGIC);
836     header.version = cpu_to_be32(QCOW_VERSION);
837     header.size = cpu_to_be64(total_size);
838     header_size = sizeof(header);
839     backing_filename_len = 0;
840     if (backing_file) {
841         if (strcmp(backing_file, "fat:")) {
842             header.backing_file_offset = cpu_to_be64(header_size);
843             backing_filename_len = strlen(backing_file);
844             header.backing_file_size = cpu_to_be32(backing_filename_len);
845             header_size += backing_filename_len;
846         } else {
847             /* special backing file for vvfat */
848             backing_file = NULL;
849         }
850         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
851                                     unmodified sectors */
852         header.l2_bits = 12; /* 32 KB L2 tables */
853     } else {
854         header.cluster_bits = 12; /* 4 KB clusters */
855         header.l2_bits = 9; /* 4 KB L2 tables */
856     }
857     header_size = (header_size + 7) & ~7;
858     shift = header.cluster_bits + header.l2_bits;
859     l1_size = (total_size + (1LL << shift) - 1) >> shift;
860 
861     header.l1_table_offset = cpu_to_be64(header_size);
862     if (flags & BLOCK_FLAG_ENCRYPT) {
863         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
864     } else {
865         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
866     }
867 
868     /* write all the data */
869     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
870     if (ret != sizeof(header)) {
871         goto exit;
872     }
873 
874     if (backing_file) {
875         ret = blk_pwrite(qcow_blk, sizeof(header),
876                          backing_file, backing_filename_len, 0);
877         if (ret != backing_filename_len) {
878             goto exit;
879         }
880     }
881 
882     tmp = g_malloc0(BDRV_SECTOR_SIZE);
883     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
884          i++) {
885         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
886                          tmp, BDRV_SECTOR_SIZE, 0);
887         if (ret != BDRV_SECTOR_SIZE) {
888             g_free(tmp);
889             goto exit;
890         }
891     }
892 
893     g_free(tmp);
894     ret = 0;
895 exit:
896     blk_unref(qcow_blk);
897 cleanup:
898     g_free(backing_file);
899     return ret;
900 }
901 
902 static int qcow_make_empty(BlockDriverState *bs)
903 {
904     BDRVQcowState *s = bs->opaque;
905     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
906     int ret;
907 
908     memset(s->l1_table, 0, l1_length);
909     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
910             l1_length) < 0)
911         return -1;
912     ret = bdrv_truncate(bs->file->bs, s->l1_table_offset + l1_length);
913     if (ret < 0)
914         return ret;
915 
916     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
917     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
918     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
919 
920     return 0;
921 }
922 
923 /* XXX: put compressed sectors first, then all the cluster aligned
924    tables to avoid losing bytes in alignment */
925 static coroutine_fn int
926 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
927                            uint64_t bytes, QEMUIOVector *qiov)
928 {
929     BDRVQcowState *s = bs->opaque;
930     QEMUIOVector hd_qiov;
931     struct iovec iov;
932     z_stream strm;
933     int ret, out_len;
934     uint8_t *buf, *out_buf;
935     uint64_t cluster_offset;
936 
937     buf = qemu_blockalign(bs, s->cluster_size);
938     if (bytes != s->cluster_size) {
939         if (bytes > s->cluster_size ||
940             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
941         {
942             qemu_vfree(buf);
943             return -EINVAL;
944         }
945         /* Zero-pad last write if image size is not cluster aligned */
946         memset(buf + bytes, 0, s->cluster_size - bytes);
947     }
948     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
949 
950     out_buf = g_malloc(s->cluster_size);
951 
952     /* best compression, small window, no zlib header */
953     memset(&strm, 0, sizeof(strm));
954     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
955                        Z_DEFLATED, -12,
956                        9, Z_DEFAULT_STRATEGY);
957     if (ret != 0) {
958         ret = -EINVAL;
959         goto fail;
960     }
961 
962     strm.avail_in = s->cluster_size;
963     strm.next_in = (uint8_t *)buf;
964     strm.avail_out = s->cluster_size;
965     strm.next_out = out_buf;
966 
967     ret = deflate(&strm, Z_FINISH);
968     if (ret != Z_STREAM_END && ret != Z_OK) {
969         deflateEnd(&strm);
970         ret = -EINVAL;
971         goto fail;
972     }
973     out_len = strm.next_out - out_buf;
974 
975     deflateEnd(&strm);
976 
977     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
978         /* could not compress: write normal cluster */
979         ret = qcow_co_writev(bs, offset >> BDRV_SECTOR_BITS,
980                              bytes >> BDRV_SECTOR_BITS, qiov);
981         if (ret < 0) {
982             goto fail;
983         }
984         goto success;
985     }
986     qemu_co_mutex_lock(&s->lock);
987     cluster_offset = get_cluster_offset(bs, offset, 2, out_len, 0, 0);
988     qemu_co_mutex_unlock(&s->lock);
989     if (cluster_offset == 0) {
990         ret = -EIO;
991         goto fail;
992     }
993     cluster_offset &= s->cluster_offset_mask;
994 
995     iov = (struct iovec) {
996         .iov_base   = out_buf,
997         .iov_len    = out_len,
998     };
999     qemu_iovec_init_external(&hd_qiov, &iov, 1);
1000     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
1001     if (ret < 0) {
1002         goto fail;
1003     }
1004 success:
1005     ret = 0;
1006 fail:
1007     qemu_vfree(buf);
1008     g_free(out_buf);
1009     return ret;
1010 }
1011 
1012 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1013 {
1014     BDRVQcowState *s = bs->opaque;
1015     bdi->cluster_size = s->cluster_size;
1016     return 0;
1017 }
1018 
1019 static QemuOptsList qcow_create_opts = {
1020     .name = "qcow-create-opts",
1021     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1022     .desc = {
1023         {
1024             .name = BLOCK_OPT_SIZE,
1025             .type = QEMU_OPT_SIZE,
1026             .help = "Virtual disk size"
1027         },
1028         {
1029             .name = BLOCK_OPT_BACKING_FILE,
1030             .type = QEMU_OPT_STRING,
1031             .help = "File name of a base image"
1032         },
1033         {
1034             .name = BLOCK_OPT_ENCRYPT,
1035             .type = QEMU_OPT_BOOL,
1036             .help = "Encrypt the image",
1037             .def_value_str = "off"
1038         },
1039         { /* end of list */ }
1040     }
1041 };
1042 
1043 static BlockDriver bdrv_qcow = {
1044     .format_name	= "qcow",
1045     .instance_size	= sizeof(BDRVQcowState),
1046     .bdrv_probe		= qcow_probe,
1047     .bdrv_open		= qcow_open,
1048     .bdrv_close		= qcow_close,
1049     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1050     .bdrv_create            = qcow_create,
1051     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1052     .supports_backing       = true,
1053 
1054     .bdrv_co_readv          = qcow_co_readv,
1055     .bdrv_co_writev         = qcow_co_writev,
1056     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1057 
1058     .bdrv_set_key           = qcow_set_key,
1059     .bdrv_make_empty        = qcow_make_empty,
1060     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1061     .bdrv_get_info          = qcow_get_info,
1062 
1063     .create_opts            = &qcow_create_opts,
1064 };
1065 
1066 static void bdrv_qcow_init(void)
1067 {
1068     bdrv_register(&bdrv_qcow);
1069 }
1070 
1071 block_init(bdrv_qcow_init);
1072