xref: /openbmc/qemu/block/qcow.c (revision db1015e9)
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 
25 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu/error-report.h"
28 #include "block/block_int.h"
29 #include "block/qdict.h"
30 #include "sysemu/block-backend.h"
31 #include "qemu/module.h"
32 #include "qemu/option.h"
33 #include "qemu/bswap.h"
34 #include "qemu/cutils.h"
35 #include <zlib.h>
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qstring.h"
38 #include "qapi/qobject-input-visitor.h"
39 #include "qapi/qapi-visit-block-core.h"
40 #include "crypto/block.h"
41 #include "migration/blocker.h"
42 #include "crypto.h"
43 
44 /**************************************************************/
45 /* QEMU COW block driver with compression and encryption support */
46 
47 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
48 #define QCOW_VERSION 1
49 
50 #define QCOW_CRYPT_NONE 0
51 #define QCOW_CRYPT_AES  1
52 
53 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
54 
55 typedef struct QCowHeader {
56     uint32_t magic;
57     uint32_t version;
58     uint64_t backing_file_offset;
59     uint32_t backing_file_size;
60     uint32_t mtime;
61     uint64_t size; /* in bytes */
62     uint8_t cluster_bits;
63     uint8_t l2_bits;
64     uint16_t padding;
65     uint32_t crypt_method;
66     uint64_t l1_table_offset;
67 } QEMU_PACKED QCowHeader;
68 
69 #define L2_CACHE_SIZE 16
70 
71 typedef struct BDRVQcowState {
72     int cluster_bits;
73     int cluster_size;
74     int l2_bits;
75     int l2_size;
76     unsigned int l1_size;
77     uint64_t cluster_offset_mask;
78     uint64_t l1_table_offset;
79     uint64_t *l1_table;
80     uint64_t *l2_cache;
81     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
82     uint32_t l2_cache_counts[L2_CACHE_SIZE];
83     uint8_t *cluster_cache;
84     uint8_t *cluster_data;
85     uint64_t cluster_cache_offset;
86     QCryptoBlock *crypto; /* Disk encryption format driver */
87     uint32_t crypt_method_header;
88     CoMutex lock;
89     Error *migration_blocker;
90 } BDRVQcowState;
91 
92 static QemuOptsList qcow_create_opts;
93 
94 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
95 
96 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
97 {
98     const QCowHeader *cow_header = (const void *)buf;
99 
100     if (buf_size >= sizeof(QCowHeader) &&
101         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
102         be32_to_cpu(cow_header->version) == QCOW_VERSION)
103         return 100;
104     else
105         return 0;
106 }
107 
108 static QemuOptsList qcow_runtime_opts = {
109     .name = "qcow",
110     .head = QTAILQ_HEAD_INITIALIZER(qcow_runtime_opts.head),
111     .desc = {
112         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
113         { /* end of list */ }
114     },
115 };
116 
117 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
118                      Error **errp)
119 {
120     BDRVQcowState *s = bs->opaque;
121     unsigned int len, i, shift;
122     int ret;
123     QCowHeader header;
124     QCryptoBlockOpenOptions *crypto_opts = NULL;
125     unsigned int cflags = 0;
126     QDict *encryptopts = NULL;
127     const char *encryptfmt;
128 
129     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
130     encryptfmt = qdict_get_try_str(encryptopts, "format");
131 
132     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
133                                BDRV_CHILD_IMAGE, false, errp);
134     if (!bs->file) {
135         ret = -EINVAL;
136         goto fail;
137     }
138 
139     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
140     if (ret < 0) {
141         goto fail;
142     }
143     header.magic = be32_to_cpu(header.magic);
144     header.version = be32_to_cpu(header.version);
145     header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
146     header.backing_file_size = be32_to_cpu(header.backing_file_size);
147     header.mtime = be32_to_cpu(header.mtime);
148     header.size = be64_to_cpu(header.size);
149     header.crypt_method = be32_to_cpu(header.crypt_method);
150     header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
151 
152     if (header.magic != QCOW_MAGIC) {
153         error_setg(errp, "Image not in qcow format");
154         ret = -EINVAL;
155         goto fail;
156     }
157     if (header.version != QCOW_VERSION) {
158         error_setg(errp, "qcow (v%d) does not support qcow version %" PRIu32,
159                    QCOW_VERSION, header.version);
160         if (header.version == 2 || header.version == 3) {
161             error_append_hint(errp, "Try the 'qcow2' driver instead.\n");
162         }
163 
164         ret = -ENOTSUP;
165         goto fail;
166     }
167 
168     if (header.size <= 1) {
169         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
170         ret = -EINVAL;
171         goto fail;
172     }
173     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
174         error_setg(errp, "Cluster size must be between 512 and 64k");
175         ret = -EINVAL;
176         goto fail;
177     }
178 
179     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
180      * so bytes = num_entries << 3. */
181     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
182         error_setg(errp, "L2 table size must be between 512 and 64k");
183         ret = -EINVAL;
184         goto fail;
185     }
186 
187     s->crypt_method_header = header.crypt_method;
188     if (s->crypt_method_header) {
189         if (bdrv_uses_whitelist() &&
190             s->crypt_method_header == QCOW_CRYPT_AES) {
191             error_setg(errp,
192                        "Use of AES-CBC encrypted qcow images is no longer "
193                        "supported in system emulators");
194             error_append_hint(errp,
195                               "You can use 'qemu-img convert' to convert your "
196                               "image to an alternative supported format, such "
197                               "as unencrypted qcow, or raw with the LUKS "
198                               "format instead.\n");
199             ret = -ENOSYS;
200             goto fail;
201         }
202         if (s->crypt_method_header == QCOW_CRYPT_AES) {
203             if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
204                 error_setg(errp,
205                            "Header reported 'aes' encryption format but "
206                            "options specify '%s'", encryptfmt);
207                 ret = -EINVAL;
208                 goto fail;
209             }
210             qdict_put_str(encryptopts, "format", "qcow");
211             crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
212             if (!crypto_opts) {
213                 ret = -EINVAL;
214                 goto fail;
215             }
216 
217             if (flags & BDRV_O_NO_IO) {
218                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
219             }
220             s->crypto = qcrypto_block_open(crypto_opts, "encrypt.",
221                                            NULL, NULL, cflags, 1, errp);
222             if (!s->crypto) {
223                 ret = -EINVAL;
224                 goto fail;
225             }
226         } else {
227             error_setg(errp, "invalid encryption method in qcow header");
228             ret = -EINVAL;
229             goto fail;
230         }
231         bs->encrypted = true;
232     } else {
233         if (encryptfmt) {
234             error_setg(errp, "No encryption in image header, but options "
235                        "specified format '%s'", encryptfmt);
236             ret = -EINVAL;
237             goto fail;
238         }
239     }
240     s->cluster_bits = header.cluster_bits;
241     s->cluster_size = 1 << s->cluster_bits;
242     s->l2_bits = header.l2_bits;
243     s->l2_size = 1 << s->l2_bits;
244     bs->total_sectors = header.size / 512;
245     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
246 
247     /* read the level 1 table */
248     shift = s->cluster_bits + s->l2_bits;
249     if (header.size > UINT64_MAX - (1LL << shift)) {
250         error_setg(errp, "Image too large");
251         ret = -EINVAL;
252         goto fail;
253     } else {
254         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
255         if (l1_size > INT_MAX / sizeof(uint64_t)) {
256             error_setg(errp, "Image too large");
257             ret = -EINVAL;
258             goto fail;
259         }
260         s->l1_size = l1_size;
261     }
262 
263     s->l1_table_offset = header.l1_table_offset;
264     s->l1_table = g_try_new(uint64_t, s->l1_size);
265     if (s->l1_table == NULL) {
266         error_setg(errp, "Could not allocate memory for L1 table");
267         ret = -ENOMEM;
268         goto fail;
269     }
270 
271     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
272                s->l1_size * sizeof(uint64_t));
273     if (ret < 0) {
274         goto fail;
275     }
276 
277     for(i = 0;i < s->l1_size; i++) {
278         s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
279     }
280 
281     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
282     s->l2_cache =
283         qemu_try_blockalign(bs->file->bs,
284                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
285     if (s->l2_cache == NULL) {
286         error_setg(errp, "Could not allocate L2 table cache");
287         ret = -ENOMEM;
288         goto fail;
289     }
290     s->cluster_cache = g_malloc(s->cluster_size);
291     s->cluster_data = g_malloc(s->cluster_size);
292     s->cluster_cache_offset = -1;
293 
294     /* read the backing file name */
295     if (header.backing_file_offset != 0) {
296         len = header.backing_file_size;
297         if (len > 1023 || len >= sizeof(bs->backing_file)) {
298             error_setg(errp, "Backing file name too long");
299             ret = -EINVAL;
300             goto fail;
301         }
302         ret = bdrv_pread(bs->file, header.backing_file_offset,
303                    bs->auto_backing_file, len);
304         if (ret < 0) {
305             goto fail;
306         }
307         bs->auto_backing_file[len] = '\0';
308         pstrcpy(bs->backing_file, sizeof(bs->backing_file),
309                 bs->auto_backing_file);
310     }
311 
312     /* Disable migration when qcow images are used */
313     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
314                "does not support live migration",
315                bdrv_get_device_or_node_name(bs));
316     ret = migrate_add_blocker(s->migration_blocker, errp);
317     if (ret < 0) {
318         error_free(s->migration_blocker);
319         goto fail;
320     }
321 
322     qobject_unref(encryptopts);
323     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
324     qemu_co_mutex_init(&s->lock);
325     return 0;
326 
327  fail:
328     g_free(s->l1_table);
329     qemu_vfree(s->l2_cache);
330     g_free(s->cluster_cache);
331     g_free(s->cluster_data);
332     qcrypto_block_free(s->crypto);
333     qobject_unref(encryptopts);
334     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
335     return ret;
336 }
337 
338 
339 /* We have nothing to do for QCOW reopen, stubs just return
340  * success */
341 static int qcow_reopen_prepare(BDRVReopenState *state,
342                                BlockReopenQueue *queue, Error **errp)
343 {
344     return 0;
345 }
346 
347 
348 /* 'allocate' is:
349  *
350  * 0 to not allocate.
351  *
352  * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start'
353  * to 'n_end' within the cluster)
354  *
355  * 2 to allocate a compressed cluster of size
356  * 'compressed_size'. 'compressed_size' must be > 0 and <
357  * cluster_size
358  *
359  * return 0 if not allocated, 1 if *result is assigned, and negative
360  * errno on failure.
361  */
362 static int get_cluster_offset(BlockDriverState *bs,
363                               uint64_t offset, int allocate,
364                               int compressed_size,
365                               int n_start, int n_end, uint64_t *result)
366 {
367     BDRVQcowState *s = bs->opaque;
368     int min_index, i, j, l1_index, l2_index, ret;
369     int64_t l2_offset;
370     uint64_t *l2_table, cluster_offset, tmp;
371     uint32_t min_count;
372     int new_l2_table;
373 
374     *result = 0;
375     l1_index = offset >> (s->l2_bits + s->cluster_bits);
376     l2_offset = s->l1_table[l1_index];
377     new_l2_table = 0;
378     if (!l2_offset) {
379         if (!allocate)
380             return 0;
381         /* allocate a new l2 entry */
382         l2_offset = bdrv_getlength(bs->file->bs);
383         if (l2_offset < 0) {
384             return l2_offset;
385         }
386         /* round to cluster size */
387         l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size);
388         /* update the L1 entry */
389         s->l1_table[l1_index] = l2_offset;
390         tmp = cpu_to_be64(l2_offset);
391         BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
392         ret = bdrv_pwrite_sync(bs->file,
393                                s->l1_table_offset + l1_index * sizeof(tmp),
394                                &tmp, sizeof(tmp));
395         if (ret < 0) {
396             return ret;
397         }
398         new_l2_table = 1;
399     }
400     for(i = 0; i < L2_CACHE_SIZE; i++) {
401         if (l2_offset == s->l2_cache_offsets[i]) {
402             /* increment the hit count */
403             if (++s->l2_cache_counts[i] == 0xffffffff) {
404                 for(j = 0; j < L2_CACHE_SIZE; j++) {
405                     s->l2_cache_counts[j] >>= 1;
406                 }
407             }
408             l2_table = s->l2_cache + (i << s->l2_bits);
409             goto found;
410         }
411     }
412     /* not found: load a new entry in the least used one */
413     min_index = 0;
414     min_count = 0xffffffff;
415     for(i = 0; i < L2_CACHE_SIZE; i++) {
416         if (s->l2_cache_counts[i] < min_count) {
417             min_count = s->l2_cache_counts[i];
418             min_index = i;
419         }
420     }
421     l2_table = s->l2_cache + (min_index << s->l2_bits);
422     BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
423     if (new_l2_table) {
424         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
425         ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
426                                s->l2_size * sizeof(uint64_t));
427         if (ret < 0) {
428             return ret;
429         }
430     } else {
431         ret = bdrv_pread(bs->file, l2_offset, l2_table,
432                          s->l2_size * sizeof(uint64_t));
433         if (ret < 0) {
434             return ret;
435         }
436     }
437     s->l2_cache_offsets[min_index] = l2_offset;
438     s->l2_cache_counts[min_index] = 1;
439  found:
440     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
441     cluster_offset = be64_to_cpu(l2_table[l2_index]);
442     if (!cluster_offset ||
443         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
444         if (!allocate)
445             return 0;
446         BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
447         assert(QEMU_IS_ALIGNED(n_start | n_end, BDRV_SECTOR_SIZE));
448         /* allocate a new cluster */
449         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
450             (n_end - n_start) < s->cluster_size) {
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 -EIO;
456             }
457             cluster_offset = bdrv_getlength(bs->file->bs);
458             if ((int64_t) cluster_offset < 0) {
459                 return cluster_offset;
460             }
461             cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
462             /* write the cluster content */
463             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
464             ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
465                               s->cluster_size);
466             if (ret < 0) {
467                 return ret;
468             }
469         } else {
470             cluster_offset = bdrv_getlength(bs->file->bs);
471             if ((int64_t) cluster_offset < 0) {
472                 return cluster_offset;
473             }
474             if (allocate == 1) {
475                 /* round to cluster size */
476                 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
477                 if (cluster_offset + s->cluster_size > INT64_MAX) {
478                     return -E2BIG;
479                 }
480                 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size,
481                                     false, PREALLOC_MODE_OFF, 0, NULL);
482                 if (ret < 0) {
483                     return ret;
484                 }
485                 /* if encrypted, we must initialize the cluster
486                    content which won't be written */
487                 if (bs->encrypted &&
488                     (n_end - n_start) < s->cluster_size) {
489                     uint64_t start_offset;
490                     assert(s->crypto);
491                     start_offset = offset & ~(s->cluster_size - 1);
492                     for (i = 0; i < s->cluster_size; i += BDRV_SECTOR_SIZE) {
493                         if (i < n_start || i >= n_end) {
494                             memset(s->cluster_data, 0x00, BDRV_SECTOR_SIZE);
495                             if (qcrypto_block_encrypt(s->crypto,
496                                                       start_offset + i,
497                                                       s->cluster_data,
498                                                       BDRV_SECTOR_SIZE,
499                                                       NULL) < 0) {
500                                 return -EIO;
501                             }
502                             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
503                             ret = bdrv_pwrite(bs->file,
504                                               cluster_offset + i,
505                                               s->cluster_data,
506                                               BDRV_SECTOR_SIZE);
507                             if (ret < 0) {
508                                 return ret;
509                             }
510                         }
511                     }
512                 }
513             } else if (allocate == 2) {
514                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
515                     (uint64_t)compressed_size << (63 - s->cluster_bits);
516             }
517         }
518         /* update L2 table */
519         tmp = cpu_to_be64(cluster_offset);
520         l2_table[l2_index] = tmp;
521         if (allocate == 2) {
522             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
523         } else {
524             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
525         }
526         ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
527                                &tmp, sizeof(tmp));
528         if (ret < 0) {
529             return ret;
530         }
531     }
532     *result = cluster_offset;
533     return 1;
534 }
535 
536 static int coroutine_fn qcow_co_block_status(BlockDriverState *bs,
537                                              bool want_zero,
538                                              int64_t offset, int64_t bytes,
539                                              int64_t *pnum, int64_t *map,
540                                              BlockDriverState **file)
541 {
542     BDRVQcowState *s = bs->opaque;
543     int index_in_cluster, ret;
544     int64_t n;
545     uint64_t cluster_offset;
546 
547     qemu_co_mutex_lock(&s->lock);
548     ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
549     qemu_co_mutex_unlock(&s->lock);
550     if (ret < 0) {
551         return ret;
552     }
553     index_in_cluster = offset & (s->cluster_size - 1);
554     n = s->cluster_size - index_in_cluster;
555     if (n > bytes) {
556         n = bytes;
557     }
558     *pnum = n;
559     if (!cluster_offset) {
560         return 0;
561     }
562     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) {
563         return BDRV_BLOCK_DATA;
564     }
565     *map = cluster_offset | index_in_cluster;
566     *file = bs->file->bs;
567     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
568 }
569 
570 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
571                              const uint8_t *buf, int buf_size)
572 {
573     z_stream strm1, *strm = &strm1;
574     int ret, out_len;
575 
576     memset(strm, 0, sizeof(*strm));
577 
578     strm->next_in = (uint8_t *)buf;
579     strm->avail_in = buf_size;
580     strm->next_out = out_buf;
581     strm->avail_out = out_buf_size;
582 
583     ret = inflateInit2(strm, -12);
584     if (ret != Z_OK)
585         return -1;
586     ret = inflate(strm, Z_FINISH);
587     out_len = strm->next_out - out_buf;
588     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
589         out_len != out_buf_size) {
590         inflateEnd(strm);
591         return -1;
592     }
593     inflateEnd(strm);
594     return 0;
595 }
596 
597 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
598 {
599     BDRVQcowState *s = bs->opaque;
600     int ret, csize;
601     uint64_t coffset;
602 
603     coffset = cluster_offset & s->cluster_offset_mask;
604     if (s->cluster_cache_offset != coffset) {
605         csize = cluster_offset >> (63 - s->cluster_bits);
606         csize &= (s->cluster_size - 1);
607         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
608         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
609         if (ret != csize)
610             return -1;
611         if (decompress_buffer(s->cluster_cache, s->cluster_size,
612                               s->cluster_data, csize) < 0) {
613             return -1;
614         }
615         s->cluster_cache_offset = coffset;
616     }
617     return 0;
618 }
619 
620 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp)
621 {
622     /* At least encrypted images require 512-byte alignment. Apply the
623      * limit universally, rather than just on encrypted images, as
624      * it's easier to let the block layer handle rounding than to
625      * audit this code further. */
626     bs->bl.request_alignment = BDRV_SECTOR_SIZE;
627 }
628 
629 static coroutine_fn int qcow_co_preadv(BlockDriverState *bs, uint64_t offset,
630                                        uint64_t bytes, QEMUIOVector *qiov,
631                                        int flags)
632 {
633     BDRVQcowState *s = bs->opaque;
634     int offset_in_cluster;
635     int ret = 0, n;
636     uint64_t cluster_offset;
637     uint8_t *buf;
638     void *orig_buf;
639 
640     assert(!flags);
641     if (qiov->niov > 1) {
642         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
643         if (buf == NULL) {
644             return -ENOMEM;
645         }
646     } else {
647         orig_buf = NULL;
648         buf = (uint8_t *)qiov->iov->iov_base;
649     }
650 
651     qemu_co_mutex_lock(&s->lock);
652 
653     while (bytes != 0) {
654         /* prepare next request */
655         ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
656         if (ret < 0) {
657             break;
658         }
659         offset_in_cluster = offset & (s->cluster_size - 1);
660         n = s->cluster_size - offset_in_cluster;
661         if (n > bytes) {
662             n = bytes;
663         }
664 
665         if (!cluster_offset) {
666             if (bs->backing) {
667                 /* read from the base image */
668                 qemu_co_mutex_unlock(&s->lock);
669                 /* qcow2 emits this on bs->file instead of bs->backing */
670                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
671                 ret = bdrv_co_pread(bs->backing, offset, n, buf, 0);
672                 qemu_co_mutex_lock(&s->lock);
673                 if (ret < 0) {
674                     break;
675                 }
676             } else {
677                 /* Note: in this case, no need to wait */
678                 memset(buf, 0, n);
679             }
680         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
681             /* add AIO support for compressed blocks ? */
682             if (decompress_cluster(bs, cluster_offset) < 0) {
683                 ret = -EIO;
684                 break;
685             }
686             memcpy(buf, s->cluster_cache + offset_in_cluster, n);
687         } else {
688             if ((cluster_offset & 511) != 0) {
689                 ret = -EIO;
690                 break;
691             }
692             qemu_co_mutex_unlock(&s->lock);
693             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
694             ret = bdrv_co_pread(bs->file, cluster_offset + offset_in_cluster,
695                                 n, buf, 0);
696             qemu_co_mutex_lock(&s->lock);
697             if (ret < 0) {
698                 break;
699             }
700             if (bs->encrypted) {
701                 assert(s->crypto);
702                 if (qcrypto_block_decrypt(s->crypto,
703                                           offset, buf, n, NULL) < 0) {
704                     ret = -EIO;
705                     break;
706                 }
707             }
708         }
709         ret = 0;
710 
711         bytes -= n;
712         offset += n;
713         buf += n;
714     }
715 
716     qemu_co_mutex_unlock(&s->lock);
717 
718     if (qiov->niov > 1) {
719         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
720         qemu_vfree(orig_buf);
721     }
722 
723     return ret;
724 }
725 
726 static coroutine_fn int qcow_co_pwritev(BlockDriverState *bs, uint64_t offset,
727                                         uint64_t bytes, QEMUIOVector *qiov,
728                                         int flags)
729 {
730     BDRVQcowState *s = bs->opaque;
731     int offset_in_cluster;
732     uint64_t cluster_offset;
733     int ret = 0, n;
734     uint8_t *buf;
735     void *orig_buf;
736 
737     assert(!flags);
738     s->cluster_cache_offset = -1; /* disable compressed cache */
739 
740     /* We must always copy the iov when encrypting, so we
741      * don't modify the original data buffer during encryption */
742     if (bs->encrypted || qiov->niov > 1) {
743         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
744         if (buf == NULL) {
745             return -ENOMEM;
746         }
747         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
748     } else {
749         orig_buf = NULL;
750         buf = (uint8_t *)qiov->iov->iov_base;
751     }
752 
753     qemu_co_mutex_lock(&s->lock);
754 
755     while (bytes != 0) {
756         offset_in_cluster = offset & (s->cluster_size - 1);
757         n = s->cluster_size - offset_in_cluster;
758         if (n > bytes) {
759             n = bytes;
760         }
761         ret = get_cluster_offset(bs, offset, 1, 0, offset_in_cluster,
762                                  offset_in_cluster + n, &cluster_offset);
763         if (ret < 0) {
764             break;
765         }
766         if (!cluster_offset || (cluster_offset & 511) != 0) {
767             ret = -EIO;
768             break;
769         }
770         if (bs->encrypted) {
771             assert(s->crypto);
772             if (qcrypto_block_encrypt(s->crypto, offset, buf, n, NULL) < 0) {
773                 ret = -EIO;
774                 break;
775             }
776         }
777 
778         qemu_co_mutex_unlock(&s->lock);
779         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
780         ret = bdrv_co_pwrite(bs->file, cluster_offset + offset_in_cluster,
781                              n, buf, 0);
782         qemu_co_mutex_lock(&s->lock);
783         if (ret < 0) {
784             break;
785         }
786         ret = 0;
787 
788         bytes -= n;
789         offset += n;
790         buf += n;
791     }
792     qemu_co_mutex_unlock(&s->lock);
793 
794     qemu_vfree(orig_buf);
795 
796     return ret;
797 }
798 
799 static void qcow_close(BlockDriverState *bs)
800 {
801     BDRVQcowState *s = bs->opaque;
802 
803     qcrypto_block_free(s->crypto);
804     s->crypto = NULL;
805     g_free(s->l1_table);
806     qemu_vfree(s->l2_cache);
807     g_free(s->cluster_cache);
808     g_free(s->cluster_data);
809 
810     migrate_del_blocker(s->migration_blocker);
811     error_free(s->migration_blocker);
812 }
813 
814 static int coroutine_fn qcow_co_create(BlockdevCreateOptions *opts,
815                                        Error **errp)
816 {
817     BlockdevCreateOptionsQcow *qcow_opts;
818     int header_size, backing_filename_len, l1_size, shift, i;
819     QCowHeader header;
820     uint8_t *tmp;
821     int64_t total_size = 0;
822     int ret;
823     BlockDriverState *bs;
824     BlockBackend *qcow_blk;
825     QCryptoBlock *crypto = NULL;
826 
827     assert(opts->driver == BLOCKDEV_DRIVER_QCOW);
828     qcow_opts = &opts->u.qcow;
829 
830     /* Sanity checks */
831     total_size = qcow_opts->size;
832     if (total_size == 0) {
833         error_setg(errp, "Image size is too small, cannot be zero length");
834         return -EINVAL;
835     }
836 
837     if (qcow_opts->has_encrypt &&
838         qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW)
839     {
840         error_setg(errp, "Unsupported encryption format");
841         return -EINVAL;
842     }
843 
844     /* Create BlockBackend to write to the image */
845     bs = bdrv_open_blockdev_ref(qcow_opts->file, errp);
846     if (bs == NULL) {
847         return -EIO;
848     }
849 
850     qcow_blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE,
851                                BLK_PERM_ALL, errp);
852     if (!qcow_blk) {
853         ret = -EPERM;
854         goto exit;
855     }
856     blk_set_allow_write_beyond_eof(qcow_blk, true);
857 
858     /* Create image format */
859     memset(&header, 0, sizeof(header));
860     header.magic = cpu_to_be32(QCOW_MAGIC);
861     header.version = cpu_to_be32(QCOW_VERSION);
862     header.size = cpu_to_be64(total_size);
863     header_size = sizeof(header);
864     backing_filename_len = 0;
865     if (qcow_opts->has_backing_file) {
866         if (strcmp(qcow_opts->backing_file, "fat:")) {
867             header.backing_file_offset = cpu_to_be64(header_size);
868             backing_filename_len = strlen(qcow_opts->backing_file);
869             header.backing_file_size = cpu_to_be32(backing_filename_len);
870             header_size += backing_filename_len;
871         } else {
872             /* special backing file for vvfat */
873             qcow_opts->has_backing_file = false;
874         }
875         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
876                                     unmodified sectors */
877         header.l2_bits = 12; /* 32 KB L2 tables */
878     } else {
879         header.cluster_bits = 12; /* 4 KB clusters */
880         header.l2_bits = 9; /* 4 KB L2 tables */
881     }
882     header_size = (header_size + 7) & ~7;
883     shift = header.cluster_bits + header.l2_bits;
884     l1_size = (total_size + (1LL << shift) - 1) >> shift;
885 
886     header.l1_table_offset = cpu_to_be64(header_size);
887 
888     if (qcow_opts->has_encrypt) {
889         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
890 
891         crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.",
892                                       NULL, NULL, NULL, errp);
893         if (!crypto) {
894             ret = -EINVAL;
895             goto exit;
896         }
897     } else {
898         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
899     }
900 
901     /* write all the data */
902     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
903     if (ret != sizeof(header)) {
904         goto exit;
905     }
906 
907     if (qcow_opts->has_backing_file) {
908         ret = blk_pwrite(qcow_blk, sizeof(header),
909                          qcow_opts->backing_file, backing_filename_len, 0);
910         if (ret != backing_filename_len) {
911             goto exit;
912         }
913     }
914 
915     tmp = g_malloc0(BDRV_SECTOR_SIZE);
916     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
917          i++) {
918         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
919                          tmp, BDRV_SECTOR_SIZE, 0);
920         if (ret != BDRV_SECTOR_SIZE) {
921             g_free(tmp);
922             goto exit;
923         }
924     }
925 
926     g_free(tmp);
927     ret = 0;
928 exit:
929     blk_unref(qcow_blk);
930     bdrv_unref(bs);
931     qcrypto_block_free(crypto);
932     return ret;
933 }
934 
935 static int coroutine_fn qcow_co_create_opts(BlockDriver *drv,
936                                             const char *filename,
937                                             QemuOpts *opts, Error **errp)
938 {
939     BlockdevCreateOptions *create_options = NULL;
940     BlockDriverState *bs = NULL;
941     QDict *qdict = NULL;
942     Visitor *v;
943     const char *val;
944     int ret;
945     char *backing_fmt;
946 
947     static const QDictRenames opt_renames[] = {
948         { BLOCK_OPT_BACKING_FILE,       "backing-file" },
949         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
950         { NULL, NULL },
951     };
952 
953     /*
954      * We can't actually store a backing format, but can check that
955      * the user's request made sense.
956      */
957     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
958     if (backing_fmt && !bdrv_find_format(backing_fmt)) {
959         error_setg(errp, "unrecognized backing format '%s'", backing_fmt);
960         ret = -EINVAL;
961         goto fail;
962     }
963 
964     /* Parse options and convert legacy syntax */
965     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true);
966 
967     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
968     if (val && !strcmp(val, "on")) {
969         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
970     } else if (val && !strcmp(val, "off")) {
971         qdict_del(qdict, BLOCK_OPT_ENCRYPT);
972     }
973 
974     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
975     if (val && !strcmp(val, "aes")) {
976         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
977     }
978 
979     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
980         ret = -EINVAL;
981         goto fail;
982     }
983 
984     /* Create and open the file (protocol layer) */
985     ret = bdrv_create_file(filename, opts, errp);
986     if (ret < 0) {
987         goto fail;
988     }
989 
990     bs = bdrv_open(filename, NULL, NULL,
991                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
992     if (bs == NULL) {
993         ret = -EIO;
994         goto fail;
995     }
996 
997     /* Now get the QAPI type BlockdevCreateOptions */
998     qdict_put_str(qdict, "driver", "qcow");
999     qdict_put_str(qdict, "file", bs->node_name);
1000 
1001     v = qobject_input_visitor_new_flat_confused(qdict, errp);
1002     if (!v) {
1003         ret = -EINVAL;
1004         goto fail;
1005     }
1006 
1007     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
1008     visit_free(v);
1009     if (!create_options) {
1010         ret = -EINVAL;
1011         goto fail;
1012     }
1013 
1014     /* Silently round up size */
1015     assert(create_options->driver == BLOCKDEV_DRIVER_QCOW);
1016     create_options->u.qcow.size =
1017         ROUND_UP(create_options->u.qcow.size, BDRV_SECTOR_SIZE);
1018 
1019     /* Create the qcow image (format layer) */
1020     ret = qcow_co_create(create_options, errp);
1021     if (ret < 0) {
1022         goto fail;
1023     }
1024 
1025     ret = 0;
1026 fail:
1027     g_free(backing_fmt);
1028     qobject_unref(qdict);
1029     bdrv_unref(bs);
1030     qapi_free_BlockdevCreateOptions(create_options);
1031     return ret;
1032 }
1033 
1034 static int qcow_make_empty(BlockDriverState *bs)
1035 {
1036     BDRVQcowState *s = bs->opaque;
1037     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1038     int ret;
1039 
1040     memset(s->l1_table, 0, l1_length);
1041     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
1042             l1_length) < 0)
1043         return -1;
1044     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length, false,
1045                         PREALLOC_MODE_OFF, 0, NULL);
1046     if (ret < 0)
1047         return ret;
1048 
1049     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
1050     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
1051     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
1052 
1053     return 0;
1054 }
1055 
1056 /* XXX: put compressed sectors first, then all the cluster aligned
1057    tables to avoid losing bytes in alignment */
1058 static coroutine_fn int
1059 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1060                            uint64_t bytes, QEMUIOVector *qiov)
1061 {
1062     BDRVQcowState *s = bs->opaque;
1063     z_stream strm;
1064     int ret, out_len;
1065     uint8_t *buf, *out_buf;
1066     uint64_t cluster_offset;
1067 
1068     buf = qemu_blockalign(bs, s->cluster_size);
1069     if (bytes != s->cluster_size) {
1070         if (bytes > s->cluster_size ||
1071             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
1072         {
1073             qemu_vfree(buf);
1074             return -EINVAL;
1075         }
1076         /* Zero-pad last write if image size is not cluster aligned */
1077         memset(buf + bytes, 0, s->cluster_size - bytes);
1078     }
1079     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
1080 
1081     out_buf = g_malloc(s->cluster_size);
1082 
1083     /* best compression, small window, no zlib header */
1084     memset(&strm, 0, sizeof(strm));
1085     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1086                        Z_DEFLATED, -12,
1087                        9, Z_DEFAULT_STRATEGY);
1088     if (ret != 0) {
1089         ret = -EINVAL;
1090         goto fail;
1091     }
1092 
1093     strm.avail_in = s->cluster_size;
1094     strm.next_in = (uint8_t *)buf;
1095     strm.avail_out = s->cluster_size;
1096     strm.next_out = out_buf;
1097 
1098     ret = deflate(&strm, Z_FINISH);
1099     if (ret != Z_STREAM_END && ret != Z_OK) {
1100         deflateEnd(&strm);
1101         ret = -EINVAL;
1102         goto fail;
1103     }
1104     out_len = strm.next_out - out_buf;
1105 
1106     deflateEnd(&strm);
1107 
1108     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1109         /* could not compress: write normal cluster */
1110         ret = qcow_co_pwritev(bs, offset, bytes, qiov, 0);
1111         if (ret < 0) {
1112             goto fail;
1113         }
1114         goto success;
1115     }
1116     qemu_co_mutex_lock(&s->lock);
1117     ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1118     qemu_co_mutex_unlock(&s->lock);
1119     if (ret < 0) {
1120         goto fail;
1121     }
1122     if (cluster_offset == 0) {
1123         ret = -EIO;
1124         goto fail;
1125     }
1126     cluster_offset &= s->cluster_offset_mask;
1127 
1128     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1129     ret = bdrv_co_pwrite(bs->file, cluster_offset, out_len, out_buf, 0);
1130     if (ret < 0) {
1131         goto fail;
1132     }
1133 success:
1134     ret = 0;
1135 fail:
1136     qemu_vfree(buf);
1137     g_free(out_buf);
1138     return ret;
1139 }
1140 
1141 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1142 {
1143     BDRVQcowState *s = bs->opaque;
1144     bdi->cluster_size = s->cluster_size;
1145     return 0;
1146 }
1147 
1148 static QemuOptsList qcow_create_opts = {
1149     .name = "qcow-create-opts",
1150     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1151     .desc = {
1152         {
1153             .name = BLOCK_OPT_SIZE,
1154             .type = QEMU_OPT_SIZE,
1155             .help = "Virtual disk size"
1156         },
1157         {
1158             .name = BLOCK_OPT_BACKING_FILE,
1159             .type = QEMU_OPT_STRING,
1160             .help = "File name of a base image"
1161         },
1162         {
1163             .name = BLOCK_OPT_BACKING_FMT,
1164             .type = QEMU_OPT_STRING,
1165             .help = "Format of the backing image",
1166         },
1167         {
1168             .name = BLOCK_OPT_ENCRYPT,
1169             .type = QEMU_OPT_BOOL,
1170             .help = "Encrypt the image with format 'aes'. (Deprecated "
1171                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1172         },
1173         {
1174             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1175             .type = QEMU_OPT_STRING,
1176             .help = "Encrypt the image, format choices: 'aes'",
1177         },
1178         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1179         { /* end of list */ }
1180     }
1181 };
1182 
1183 static const char *const qcow_strong_runtime_opts[] = {
1184     "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
1185 
1186     NULL
1187 };
1188 
1189 static BlockDriver bdrv_qcow = {
1190     .format_name	= "qcow",
1191     .instance_size	= sizeof(BDRVQcowState),
1192     .bdrv_probe		= qcow_probe,
1193     .bdrv_open		= qcow_open,
1194     .bdrv_close		= qcow_close,
1195     .bdrv_child_perm        = bdrv_default_perms,
1196     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1197     .bdrv_co_create         = qcow_co_create,
1198     .bdrv_co_create_opts    = qcow_co_create_opts,
1199     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1200     .is_format              = true,
1201     .supports_backing       = true,
1202     .bdrv_refresh_limits    = qcow_refresh_limits,
1203 
1204     .bdrv_co_preadv         = qcow_co_preadv,
1205     .bdrv_co_pwritev        = qcow_co_pwritev,
1206     .bdrv_co_block_status   = qcow_co_block_status,
1207 
1208     .bdrv_make_empty        = qcow_make_empty,
1209     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1210     .bdrv_get_info          = qcow_get_info,
1211 
1212     .create_opts            = &qcow_create_opts,
1213     .strong_runtime_opts    = qcow_strong_runtime_opts,
1214 };
1215 
1216 static void bdrv_qcow_init(void)
1217 {
1218     bdrv_register(&bdrv_qcow);
1219 }
1220 
1221 block_init(bdrv_qcow_init);
1222