xref: /openbmc/qemu/block/qcow.c (revision d341d9f3)
1 /*
2  * Block driver for the QCOW format
3  *
4  * Copyright (c) 2004-2006 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 #include "qemu/osdep.h"
25 #include "qemu-common.h"
26 #include "block/block_int.h"
27 #include "qemu/module.h"
28 #include <zlib.h>
29 #include "qapi/qmp/qerror.h"
30 #include "crypto/cipher.h"
31 #include "migration/migration.h"
32 
33 /**************************************************************/
34 /* QEMU COW block driver with compression and encryption support */
35 
36 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
37 #define QCOW_VERSION 1
38 
39 #define QCOW_CRYPT_NONE 0
40 #define QCOW_CRYPT_AES  1
41 
42 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
43 
44 typedef struct QCowHeader {
45     uint32_t magic;
46     uint32_t version;
47     uint64_t backing_file_offset;
48     uint32_t backing_file_size;
49     uint32_t mtime;
50     uint64_t size; /* in bytes */
51     uint8_t cluster_bits;
52     uint8_t l2_bits;
53     uint16_t padding;
54     uint32_t crypt_method;
55     uint64_t l1_table_offset;
56 } QEMU_PACKED QCowHeader;
57 
58 #define L2_CACHE_SIZE 16
59 
60 typedef struct BDRVQcowState {
61     int cluster_bits;
62     int cluster_size;
63     int cluster_sectors;
64     int l2_bits;
65     int l2_size;
66     unsigned int l1_size;
67     uint64_t cluster_offset_mask;
68     uint64_t l1_table_offset;
69     uint64_t *l1_table;
70     uint64_t *l2_cache;
71     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
72     uint32_t l2_cache_counts[L2_CACHE_SIZE];
73     uint8_t *cluster_cache;
74     uint8_t *cluster_data;
75     uint64_t cluster_cache_offset;
76     QCryptoCipher *cipher; /* NULL if no key yet */
77     uint32_t crypt_method_header;
78     CoMutex lock;
79     Error *migration_blocker;
80 } BDRVQcowState;
81 
82 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
83 
84 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
85 {
86     const QCowHeader *cow_header = (const void *)buf;
87 
88     if (buf_size >= sizeof(QCowHeader) &&
89         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
90         be32_to_cpu(cow_header->version) == QCOW_VERSION)
91         return 100;
92     else
93         return 0;
94 }
95 
96 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
97                      Error **errp)
98 {
99     BDRVQcowState *s = bs->opaque;
100     unsigned int len, i, shift;
101     int ret;
102     QCowHeader header;
103 
104     ret = bdrv_pread(bs->file->bs, 0, &header, sizeof(header));
105     if (ret < 0) {
106         goto fail;
107     }
108     be32_to_cpus(&header.magic);
109     be32_to_cpus(&header.version);
110     be64_to_cpus(&header.backing_file_offset);
111     be32_to_cpus(&header.backing_file_size);
112     be32_to_cpus(&header.mtime);
113     be64_to_cpus(&header.size);
114     be32_to_cpus(&header.crypt_method);
115     be64_to_cpus(&header.l1_table_offset);
116 
117     if (header.magic != QCOW_MAGIC) {
118         error_setg(errp, "Image not in qcow format");
119         ret = -EINVAL;
120         goto fail;
121     }
122     if (header.version != QCOW_VERSION) {
123         char version[64];
124         snprintf(version, sizeof(version), "QCOW version %" PRIu32,
125                  header.version);
126         error_setg(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
127                    bdrv_get_device_or_node_name(bs), "qcow", version);
128         ret = -ENOTSUP;
129         goto fail;
130     }
131 
132     if (header.size <= 1) {
133         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
134         ret = -EINVAL;
135         goto fail;
136     }
137     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
138         error_setg(errp, "Cluster size must be between 512 and 64k");
139         ret = -EINVAL;
140         goto fail;
141     }
142 
143     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
144      * so bytes = num_entries << 3. */
145     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
146         error_setg(errp, "L2 table size must be between 512 and 64k");
147         ret = -EINVAL;
148         goto fail;
149     }
150 
151     if (header.crypt_method > QCOW_CRYPT_AES) {
152         error_setg(errp, "invalid encryption method in qcow header");
153         ret = -EINVAL;
154         goto fail;
155     }
156     if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) {
157         error_setg(errp, "AES cipher not available");
158         ret = -EINVAL;
159         goto fail;
160     }
161     s->crypt_method_header = header.crypt_method;
162     if (s->crypt_method_header) {
163         bs->encrypted = 1;
164     }
165     s->cluster_bits = header.cluster_bits;
166     s->cluster_size = 1 << s->cluster_bits;
167     s->cluster_sectors = 1 << (s->cluster_bits - 9);
168     s->l2_bits = header.l2_bits;
169     s->l2_size = 1 << s->l2_bits;
170     bs->total_sectors = header.size / 512;
171     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
172 
173     /* read the level 1 table */
174     shift = s->cluster_bits + s->l2_bits;
175     if (header.size > UINT64_MAX - (1LL << shift)) {
176         error_setg(errp, "Image too large");
177         ret = -EINVAL;
178         goto fail;
179     } else {
180         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
181         if (l1_size > INT_MAX / sizeof(uint64_t)) {
182             error_setg(errp, "Image too large");
183             ret = -EINVAL;
184             goto fail;
185         }
186         s->l1_size = l1_size;
187     }
188 
189     s->l1_table_offset = header.l1_table_offset;
190     s->l1_table = g_try_new(uint64_t, s->l1_size);
191     if (s->l1_table == NULL) {
192         error_setg(errp, "Could not allocate memory for L1 table");
193         ret = -ENOMEM;
194         goto fail;
195     }
196 
197     ret = bdrv_pread(bs->file->bs, s->l1_table_offset, s->l1_table,
198                s->l1_size * sizeof(uint64_t));
199     if (ret < 0) {
200         goto fail;
201     }
202 
203     for(i = 0;i < s->l1_size; i++) {
204         be64_to_cpus(&s->l1_table[i]);
205     }
206 
207     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
208     s->l2_cache =
209         qemu_try_blockalign(bs->file->bs,
210                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
211     if (s->l2_cache == NULL) {
212         error_setg(errp, "Could not allocate L2 table cache");
213         ret = -ENOMEM;
214         goto fail;
215     }
216     s->cluster_cache = g_malloc(s->cluster_size);
217     s->cluster_data = g_malloc(s->cluster_size);
218     s->cluster_cache_offset = -1;
219 
220     /* read the backing file name */
221     if (header.backing_file_offset != 0) {
222         len = header.backing_file_size;
223         if (len > 1023 || len >= sizeof(bs->backing_file)) {
224             error_setg(errp, "Backing file name too long");
225             ret = -EINVAL;
226             goto fail;
227         }
228         ret = bdrv_pread(bs->file->bs, header.backing_file_offset,
229                    bs->backing_file, len);
230         if (ret < 0) {
231             goto fail;
232         }
233         bs->backing_file[len] = '\0';
234     }
235 
236     /* Disable migration when qcow images are used */
237     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
238                "does not support live migration",
239                bdrv_get_device_or_node_name(bs));
240     migrate_add_blocker(s->migration_blocker);
241 
242     qemu_co_mutex_init(&s->lock);
243     return 0;
244 
245  fail:
246     g_free(s->l1_table);
247     qemu_vfree(s->l2_cache);
248     g_free(s->cluster_cache);
249     g_free(s->cluster_data);
250     return ret;
251 }
252 
253 
254 /* We have nothing to do for QCOW reopen, stubs just return
255  * success */
256 static int qcow_reopen_prepare(BDRVReopenState *state,
257                                BlockReopenQueue *queue, Error **errp)
258 {
259     return 0;
260 }
261 
262 static int qcow_set_key(BlockDriverState *bs, const char *key)
263 {
264     BDRVQcowState *s = bs->opaque;
265     uint8_t keybuf[16];
266     int len, i;
267     Error *err;
268 
269     memset(keybuf, 0, 16);
270     len = strlen(key);
271     if (len > 16)
272         len = 16;
273     /* XXX: we could compress the chars to 7 bits to increase
274        entropy */
275     for(i = 0;i < len;i++) {
276         keybuf[i] = key[i];
277     }
278     assert(bs->encrypted);
279 
280     qcrypto_cipher_free(s->cipher);
281     s->cipher = qcrypto_cipher_new(
282         QCRYPTO_CIPHER_ALG_AES_128,
283         QCRYPTO_CIPHER_MODE_CBC,
284         keybuf, G_N_ELEMENTS(keybuf),
285         &err);
286 
287     if (!s->cipher) {
288         /* XXX would be nice if errors in this method could
289          * be properly propagate to the caller. Would need
290          * the bdrv_set_key() API signature to be fixed. */
291         error_free(err);
292         return -1;
293     }
294     return 0;
295 }
296 
297 /* The crypt function is compatible with the linux cryptoloop
298    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
299    supported */
300 static int encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
301                            uint8_t *out_buf, const uint8_t *in_buf,
302                            int nb_sectors, bool enc, Error **errp)
303 {
304     union {
305         uint64_t ll[2];
306         uint8_t b[16];
307     } ivec;
308     int i;
309     int ret;
310 
311     for(i = 0; i < nb_sectors; i++) {
312         ivec.ll[0] = cpu_to_le64(sector_num);
313         ivec.ll[1] = 0;
314         if (qcrypto_cipher_setiv(s->cipher,
315                                  ivec.b, G_N_ELEMENTS(ivec.b),
316                                  errp) < 0) {
317             return -1;
318         }
319         if (enc) {
320             ret = qcrypto_cipher_encrypt(s->cipher,
321                                          in_buf,
322                                          out_buf,
323                                          512,
324                                          errp);
325         } else {
326             ret = qcrypto_cipher_decrypt(s->cipher,
327                                          in_buf,
328                                          out_buf,
329                                          512,
330                                          errp);
331         }
332         if (ret < 0) {
333             return -1;
334         }
335         sector_num++;
336         in_buf += 512;
337         out_buf += 512;
338     }
339     return 0;
340 }
341 
342 /* 'allocate' is:
343  *
344  * 0 to not allocate.
345  *
346  * 1 to allocate a normal cluster (for sector indexes 'n_start' to
347  * 'n_end')
348  *
349  * 2 to allocate a compressed cluster of size
350  * 'compressed_size'. 'compressed_size' must be > 0 and <
351  * cluster_size
352  *
353  * return 0 if not allocated.
354  */
355 static uint64_t get_cluster_offset(BlockDriverState *bs,
356                                    uint64_t offset, int allocate,
357                                    int compressed_size,
358                                    int n_start, int n_end)
359 {
360     BDRVQcowState *s = bs->opaque;
361     int min_index, i, j, l1_index, l2_index;
362     uint64_t l2_offset, *l2_table, cluster_offset, tmp;
363     uint32_t min_count;
364     int new_l2_table;
365 
366     l1_index = offset >> (s->l2_bits + s->cluster_bits);
367     l2_offset = s->l1_table[l1_index];
368     new_l2_table = 0;
369     if (!l2_offset) {
370         if (!allocate)
371             return 0;
372         /* allocate a new l2 entry */
373         l2_offset = bdrv_getlength(bs->file->bs);
374         /* round to cluster size */
375         l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
376         /* update the L1 entry */
377         s->l1_table[l1_index] = l2_offset;
378         tmp = cpu_to_be64(l2_offset);
379         if (bdrv_pwrite_sync(bs->file->bs,
380                 s->l1_table_offset + l1_index * sizeof(tmp),
381                 &tmp, sizeof(tmp)) < 0)
382             return 0;
383         new_l2_table = 1;
384     }
385     for(i = 0; i < L2_CACHE_SIZE; i++) {
386         if (l2_offset == s->l2_cache_offsets[i]) {
387             /* increment the hit count */
388             if (++s->l2_cache_counts[i] == 0xffffffff) {
389                 for(j = 0; j < L2_CACHE_SIZE; j++) {
390                     s->l2_cache_counts[j] >>= 1;
391                 }
392             }
393             l2_table = s->l2_cache + (i << s->l2_bits);
394             goto found;
395         }
396     }
397     /* not found: load a new entry in the least used one */
398     min_index = 0;
399     min_count = 0xffffffff;
400     for(i = 0; i < L2_CACHE_SIZE; i++) {
401         if (s->l2_cache_counts[i] < min_count) {
402             min_count = s->l2_cache_counts[i];
403             min_index = i;
404         }
405     }
406     l2_table = s->l2_cache + (min_index << s->l2_bits);
407     if (new_l2_table) {
408         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
409         if (bdrv_pwrite_sync(bs->file->bs, l2_offset, l2_table,
410                 s->l2_size * sizeof(uint64_t)) < 0)
411             return 0;
412     } else {
413         if (bdrv_pread(bs->file->bs, l2_offset, l2_table,
414                        s->l2_size * sizeof(uint64_t)) !=
415             s->l2_size * sizeof(uint64_t))
416             return 0;
417     }
418     s->l2_cache_offsets[min_index] = l2_offset;
419     s->l2_cache_counts[min_index] = 1;
420  found:
421     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
422     cluster_offset = be64_to_cpu(l2_table[l2_index]);
423     if (!cluster_offset ||
424         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
425         if (!allocate)
426             return 0;
427         /* allocate a new cluster */
428         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
429             (n_end - n_start) < s->cluster_sectors) {
430             /* if the cluster is already compressed, we must
431                decompress it in the case it is not completely
432                overwritten */
433             if (decompress_cluster(bs, cluster_offset) < 0)
434                 return 0;
435             cluster_offset = bdrv_getlength(bs->file->bs);
436             cluster_offset = (cluster_offset + s->cluster_size - 1) &
437                 ~(s->cluster_size - 1);
438             /* write the cluster content */
439             if (bdrv_pwrite(bs->file->bs, cluster_offset, s->cluster_cache,
440                             s->cluster_size) !=
441                 s->cluster_size)
442                 return -1;
443         } else {
444             cluster_offset = bdrv_getlength(bs->file->bs);
445             if (allocate == 1) {
446                 /* round to cluster size */
447                 cluster_offset = (cluster_offset + s->cluster_size - 1) &
448                     ~(s->cluster_size - 1);
449                 bdrv_truncate(bs->file->bs, cluster_offset + s->cluster_size);
450                 /* if encrypted, we must initialize the cluster
451                    content which won't be written */
452                 if (bs->encrypted &&
453                     (n_end - n_start) < s->cluster_sectors) {
454                     uint64_t start_sect;
455                     assert(s->cipher);
456                     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
457                     memset(s->cluster_data + 512, 0x00, 512);
458                     for(i = 0; i < s->cluster_sectors; i++) {
459                         if (i < n_start || i >= n_end) {
460                             Error *err = NULL;
461                             if (encrypt_sectors(s, start_sect + i,
462                                                 s->cluster_data,
463                                                 s->cluster_data + 512, 1,
464                                                 true, &err) < 0) {
465                                 error_free(err);
466                                 errno = EIO;
467                                 return -1;
468                             }
469                             if (bdrv_pwrite(bs->file->bs,
470                                             cluster_offset + i * 512,
471                                             s->cluster_data, 512) != 512)
472                                 return -1;
473                         }
474                     }
475                 }
476             } else if (allocate == 2) {
477                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
478                     (uint64_t)compressed_size << (63 - s->cluster_bits);
479             }
480         }
481         /* update L2 table */
482         tmp = cpu_to_be64(cluster_offset);
483         l2_table[l2_index] = tmp;
484         if (bdrv_pwrite_sync(bs->file->bs, l2_offset + l2_index * sizeof(tmp),
485                 &tmp, sizeof(tmp)) < 0)
486             return 0;
487     }
488     return cluster_offset;
489 }
490 
491 static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
492         int64_t sector_num, int nb_sectors, int *pnum)
493 {
494     BDRVQcowState *s = bs->opaque;
495     int index_in_cluster, n;
496     uint64_t cluster_offset;
497 
498     qemu_co_mutex_lock(&s->lock);
499     cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
500     qemu_co_mutex_unlock(&s->lock);
501     index_in_cluster = sector_num & (s->cluster_sectors - 1);
502     n = s->cluster_sectors - index_in_cluster;
503     if (n > nb_sectors)
504         n = nb_sectors;
505     *pnum = n;
506     if (!cluster_offset) {
507         return 0;
508     }
509     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->cipher) {
510         return BDRV_BLOCK_DATA;
511     }
512     cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
513     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
514 }
515 
516 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
517                              const uint8_t *buf, int buf_size)
518 {
519     z_stream strm1, *strm = &strm1;
520     int ret, out_len;
521 
522     memset(strm, 0, sizeof(*strm));
523 
524     strm->next_in = (uint8_t *)buf;
525     strm->avail_in = buf_size;
526     strm->next_out = out_buf;
527     strm->avail_out = out_buf_size;
528 
529     ret = inflateInit2(strm, -12);
530     if (ret != Z_OK)
531         return -1;
532     ret = inflate(strm, Z_FINISH);
533     out_len = strm->next_out - out_buf;
534     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
535         out_len != out_buf_size) {
536         inflateEnd(strm);
537         return -1;
538     }
539     inflateEnd(strm);
540     return 0;
541 }
542 
543 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
544 {
545     BDRVQcowState *s = bs->opaque;
546     int ret, csize;
547     uint64_t coffset;
548 
549     coffset = cluster_offset & s->cluster_offset_mask;
550     if (s->cluster_cache_offset != coffset) {
551         csize = cluster_offset >> (63 - s->cluster_bits);
552         csize &= (s->cluster_size - 1);
553         ret = bdrv_pread(bs->file->bs, coffset, s->cluster_data, csize);
554         if (ret != csize)
555             return -1;
556         if (decompress_buffer(s->cluster_cache, s->cluster_size,
557                               s->cluster_data, csize) < 0) {
558             return -1;
559         }
560         s->cluster_cache_offset = coffset;
561     }
562     return 0;
563 }
564 
565 static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
566                          int nb_sectors, QEMUIOVector *qiov)
567 {
568     BDRVQcowState *s = bs->opaque;
569     int index_in_cluster;
570     int ret = 0, n;
571     uint64_t cluster_offset;
572     struct iovec hd_iov;
573     QEMUIOVector hd_qiov;
574     uint8_t *buf;
575     void *orig_buf;
576     Error *err = NULL;
577 
578     if (qiov->niov > 1) {
579         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
580         if (buf == NULL) {
581             return -ENOMEM;
582         }
583     } else {
584         orig_buf = NULL;
585         buf = (uint8_t *)qiov->iov->iov_base;
586     }
587 
588     qemu_co_mutex_lock(&s->lock);
589 
590     while (nb_sectors != 0) {
591         /* prepare next request */
592         cluster_offset = get_cluster_offset(bs, sector_num << 9,
593                                                  0, 0, 0, 0);
594         index_in_cluster = sector_num & (s->cluster_sectors - 1);
595         n = s->cluster_sectors - index_in_cluster;
596         if (n > nb_sectors) {
597             n = nb_sectors;
598         }
599 
600         if (!cluster_offset) {
601             if (bs->backing) {
602                 /* read from the base image */
603                 hd_iov.iov_base = (void *)buf;
604                 hd_iov.iov_len = n * 512;
605                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
606                 qemu_co_mutex_unlock(&s->lock);
607                 ret = bdrv_co_readv(bs->backing->bs, sector_num,
608                                     n, &hd_qiov);
609                 qemu_co_mutex_lock(&s->lock);
610                 if (ret < 0) {
611                     goto fail;
612                 }
613             } else {
614                 /* Note: in this case, no need to wait */
615                 memset(buf, 0, 512 * n);
616             }
617         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
618             /* add AIO support for compressed blocks ? */
619             if (decompress_cluster(bs, cluster_offset) < 0) {
620                 goto fail;
621             }
622             memcpy(buf,
623                    s->cluster_cache + index_in_cluster * 512, 512 * n);
624         } else {
625             if ((cluster_offset & 511) != 0) {
626                 goto fail;
627             }
628             hd_iov.iov_base = (void *)buf;
629             hd_iov.iov_len = n * 512;
630             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
631             qemu_co_mutex_unlock(&s->lock);
632             ret = bdrv_co_readv(bs->file->bs,
633                                 (cluster_offset >> 9) + index_in_cluster,
634                                 n, &hd_qiov);
635             qemu_co_mutex_lock(&s->lock);
636             if (ret < 0) {
637                 break;
638             }
639             if (bs->encrypted) {
640                 assert(s->cipher);
641                 if (encrypt_sectors(s, sector_num, buf, buf,
642                                     n, false, &err) < 0) {
643                     goto fail;
644                 }
645             }
646         }
647         ret = 0;
648 
649         nb_sectors -= n;
650         sector_num += n;
651         buf += n * 512;
652     }
653 
654 done:
655     qemu_co_mutex_unlock(&s->lock);
656 
657     if (qiov->niov > 1) {
658         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
659         qemu_vfree(orig_buf);
660     }
661 
662     return ret;
663 
664 fail:
665     error_free(err);
666     ret = -EIO;
667     goto done;
668 }
669 
670 static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
671                           int nb_sectors, QEMUIOVector *qiov)
672 {
673     BDRVQcowState *s = bs->opaque;
674     int index_in_cluster;
675     uint64_t cluster_offset;
676     const uint8_t *src_buf;
677     int ret = 0, n;
678     uint8_t *cluster_data = NULL;
679     struct iovec hd_iov;
680     QEMUIOVector hd_qiov;
681     uint8_t *buf;
682     void *orig_buf;
683 
684     s->cluster_cache_offset = -1; /* disable compressed cache */
685 
686     if (qiov->niov > 1) {
687         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
688         if (buf == NULL) {
689             return -ENOMEM;
690         }
691         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
692     } else {
693         orig_buf = NULL;
694         buf = (uint8_t *)qiov->iov->iov_base;
695     }
696 
697     qemu_co_mutex_lock(&s->lock);
698 
699     while (nb_sectors != 0) {
700 
701         index_in_cluster = sector_num & (s->cluster_sectors - 1);
702         n = s->cluster_sectors - index_in_cluster;
703         if (n > nb_sectors) {
704             n = nb_sectors;
705         }
706         cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
707                                             index_in_cluster,
708                                             index_in_cluster + n);
709         if (!cluster_offset || (cluster_offset & 511) != 0) {
710             ret = -EIO;
711             break;
712         }
713         if (bs->encrypted) {
714             Error *err = NULL;
715             assert(s->cipher);
716             if (!cluster_data) {
717                 cluster_data = g_malloc0(s->cluster_size);
718             }
719             if (encrypt_sectors(s, sector_num, cluster_data, buf,
720                                 n, true, &err) < 0) {
721                 error_free(err);
722                 ret = -EIO;
723                 break;
724             }
725             src_buf = cluster_data;
726         } else {
727             src_buf = buf;
728         }
729 
730         hd_iov.iov_base = (void *)src_buf;
731         hd_iov.iov_len = n * 512;
732         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
733         qemu_co_mutex_unlock(&s->lock);
734         ret = bdrv_co_writev(bs->file->bs,
735                              (cluster_offset >> 9) + index_in_cluster,
736                              n, &hd_qiov);
737         qemu_co_mutex_lock(&s->lock);
738         if (ret < 0) {
739             break;
740         }
741         ret = 0;
742 
743         nb_sectors -= n;
744         sector_num += n;
745         buf += n * 512;
746     }
747     qemu_co_mutex_unlock(&s->lock);
748 
749     if (qiov->niov > 1) {
750         qemu_vfree(orig_buf);
751     }
752     g_free(cluster_data);
753 
754     return ret;
755 }
756 
757 static void qcow_close(BlockDriverState *bs)
758 {
759     BDRVQcowState *s = bs->opaque;
760 
761     qcrypto_cipher_free(s->cipher);
762     s->cipher = NULL;
763     g_free(s->l1_table);
764     qemu_vfree(s->l2_cache);
765     g_free(s->cluster_cache);
766     g_free(s->cluster_data);
767 
768     migrate_del_blocker(s->migration_blocker);
769     error_free(s->migration_blocker);
770 }
771 
772 static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
773 {
774     int header_size, backing_filename_len, l1_size, shift, i;
775     QCowHeader header;
776     uint8_t *tmp;
777     int64_t total_size = 0;
778     char *backing_file = NULL;
779     int flags = 0;
780     Error *local_err = NULL;
781     int ret;
782     BlockDriverState *qcow_bs;
783 
784     /* Read out options */
785     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
786                           BDRV_SECTOR_SIZE);
787     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
788     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
789         flags |= BLOCK_FLAG_ENCRYPT;
790     }
791 
792     ret = bdrv_create_file(filename, opts, &local_err);
793     if (ret < 0) {
794         error_propagate(errp, local_err);
795         goto cleanup;
796     }
797 
798     qcow_bs = NULL;
799     ret = bdrv_open(&qcow_bs, filename, NULL, NULL,
800                     BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
801     if (ret < 0) {
802         error_propagate(errp, local_err);
803         goto cleanup;
804     }
805 
806     ret = bdrv_truncate(qcow_bs, 0);
807     if (ret < 0) {
808         goto exit;
809     }
810 
811     memset(&header, 0, sizeof(header));
812     header.magic = cpu_to_be32(QCOW_MAGIC);
813     header.version = cpu_to_be32(QCOW_VERSION);
814     header.size = cpu_to_be64(total_size);
815     header_size = sizeof(header);
816     backing_filename_len = 0;
817     if (backing_file) {
818         if (strcmp(backing_file, "fat:")) {
819             header.backing_file_offset = cpu_to_be64(header_size);
820             backing_filename_len = strlen(backing_file);
821             header.backing_file_size = cpu_to_be32(backing_filename_len);
822             header_size += backing_filename_len;
823         } else {
824             /* special backing file for vvfat */
825             backing_file = NULL;
826         }
827         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
828                                     unmodified sectors */
829         header.l2_bits = 12; /* 32 KB L2 tables */
830     } else {
831         header.cluster_bits = 12; /* 4 KB clusters */
832         header.l2_bits = 9; /* 4 KB L2 tables */
833     }
834     header_size = (header_size + 7) & ~7;
835     shift = header.cluster_bits + header.l2_bits;
836     l1_size = (total_size + (1LL << shift) - 1) >> shift;
837 
838     header.l1_table_offset = cpu_to_be64(header_size);
839     if (flags & BLOCK_FLAG_ENCRYPT) {
840         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
841     } else {
842         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
843     }
844 
845     /* write all the data */
846     ret = bdrv_pwrite(qcow_bs, 0, &header, sizeof(header));
847     if (ret != sizeof(header)) {
848         goto exit;
849     }
850 
851     if (backing_file) {
852         ret = bdrv_pwrite(qcow_bs, sizeof(header),
853             backing_file, backing_filename_len);
854         if (ret != backing_filename_len) {
855             goto exit;
856         }
857     }
858 
859     tmp = g_malloc0(BDRV_SECTOR_SIZE);
860     for (i = 0; i < ((sizeof(uint64_t)*l1_size + BDRV_SECTOR_SIZE - 1)/
861         BDRV_SECTOR_SIZE); i++) {
862         ret = bdrv_pwrite(qcow_bs, header_size +
863             BDRV_SECTOR_SIZE*i, tmp, BDRV_SECTOR_SIZE);
864         if (ret != BDRV_SECTOR_SIZE) {
865             g_free(tmp);
866             goto exit;
867         }
868     }
869 
870     g_free(tmp);
871     ret = 0;
872 exit:
873     bdrv_unref(qcow_bs);
874 cleanup:
875     g_free(backing_file);
876     return ret;
877 }
878 
879 static int qcow_make_empty(BlockDriverState *bs)
880 {
881     BDRVQcowState *s = bs->opaque;
882     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
883     int ret;
884 
885     memset(s->l1_table, 0, l1_length);
886     if (bdrv_pwrite_sync(bs->file->bs, s->l1_table_offset, s->l1_table,
887             l1_length) < 0)
888         return -1;
889     ret = bdrv_truncate(bs->file->bs, s->l1_table_offset + l1_length);
890     if (ret < 0)
891         return ret;
892 
893     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
894     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
895     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
896 
897     return 0;
898 }
899 
900 /* XXX: put compressed sectors first, then all the cluster aligned
901    tables to avoid losing bytes in alignment */
902 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
903                                  const uint8_t *buf, int nb_sectors)
904 {
905     BDRVQcowState *s = bs->opaque;
906     z_stream strm;
907     int ret, out_len;
908     uint8_t *out_buf;
909     uint64_t cluster_offset;
910 
911     if (nb_sectors != s->cluster_sectors) {
912         ret = -EINVAL;
913 
914         /* Zero-pad last write if image size is not cluster aligned */
915         if (sector_num + nb_sectors == bs->total_sectors &&
916             nb_sectors < s->cluster_sectors) {
917             uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
918             memset(pad_buf, 0, s->cluster_size);
919             memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE);
920             ret = qcow_write_compressed(bs, sector_num,
921                                         pad_buf, s->cluster_sectors);
922             qemu_vfree(pad_buf);
923         }
924         return ret;
925     }
926 
927     out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
928 
929     /* best compression, small window, no zlib header */
930     memset(&strm, 0, sizeof(strm));
931     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
932                        Z_DEFLATED, -12,
933                        9, Z_DEFAULT_STRATEGY);
934     if (ret != 0) {
935         ret = -EINVAL;
936         goto fail;
937     }
938 
939     strm.avail_in = s->cluster_size;
940     strm.next_in = (uint8_t *)buf;
941     strm.avail_out = s->cluster_size;
942     strm.next_out = out_buf;
943 
944     ret = deflate(&strm, Z_FINISH);
945     if (ret != Z_STREAM_END && ret != Z_OK) {
946         deflateEnd(&strm);
947         ret = -EINVAL;
948         goto fail;
949     }
950     out_len = strm.next_out - out_buf;
951 
952     deflateEnd(&strm);
953 
954     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
955         /* could not compress: write normal cluster */
956         ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
957         if (ret < 0) {
958             goto fail;
959         }
960     } else {
961         cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
962                                             out_len, 0, 0);
963         if (cluster_offset == 0) {
964             ret = -EIO;
965             goto fail;
966         }
967 
968         cluster_offset &= s->cluster_offset_mask;
969         ret = bdrv_pwrite(bs->file->bs, cluster_offset, out_buf, out_len);
970         if (ret < 0) {
971             goto fail;
972         }
973     }
974 
975     ret = 0;
976 fail:
977     g_free(out_buf);
978     return ret;
979 }
980 
981 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
982 {
983     BDRVQcowState *s = bs->opaque;
984     bdi->cluster_size = s->cluster_size;
985     return 0;
986 }
987 
988 static QemuOptsList qcow_create_opts = {
989     .name = "qcow-create-opts",
990     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
991     .desc = {
992         {
993             .name = BLOCK_OPT_SIZE,
994             .type = QEMU_OPT_SIZE,
995             .help = "Virtual disk size"
996         },
997         {
998             .name = BLOCK_OPT_BACKING_FILE,
999             .type = QEMU_OPT_STRING,
1000             .help = "File name of a base image"
1001         },
1002         {
1003             .name = BLOCK_OPT_ENCRYPT,
1004             .type = QEMU_OPT_BOOL,
1005             .help = "Encrypt the image",
1006             .def_value_str = "off"
1007         },
1008         { /* end of list */ }
1009     }
1010 };
1011 
1012 static BlockDriver bdrv_qcow = {
1013     .format_name	= "qcow",
1014     .instance_size	= sizeof(BDRVQcowState),
1015     .bdrv_probe		= qcow_probe,
1016     .bdrv_open		= qcow_open,
1017     .bdrv_close		= qcow_close,
1018     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1019     .bdrv_create            = qcow_create,
1020     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1021     .supports_backing       = true,
1022 
1023     .bdrv_co_readv          = qcow_co_readv,
1024     .bdrv_co_writev         = qcow_co_writev,
1025     .bdrv_co_get_block_status   = qcow_co_get_block_status,
1026 
1027     .bdrv_set_key           = qcow_set_key,
1028     .bdrv_make_empty        = qcow_make_empty,
1029     .bdrv_write_compressed  = qcow_write_compressed,
1030     .bdrv_get_info          = qcow_get_info,
1031 
1032     .create_opts            = &qcow_create_opts,
1033 };
1034 
1035 static void bdrv_qcow_init(void)
1036 {
1037     bdrv_register(&bdrv_qcow);
1038 }
1039 
1040 block_init(bdrv_qcow_init);
1041