xref: /openbmc/qemu/block/qcow.c (revision 429d3ae2)
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 <zlib.h>
35 #include "qapi/qmp/qdict.h"
36 #include "qapi/qmp/qstring.h"
37 #include "qapi/qobject-input-visitor.h"
38 #include "qapi/qapi-visit-block-core.h"
39 #include "crypto/block.h"
40 #include "migration/blocker.h"
41 #include "crypto.h"
42 
43 /**************************************************************/
44 /* QEMU COW block driver with compression and encryption support */
45 
46 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
47 #define QCOW_VERSION 1
48 
49 #define QCOW_CRYPT_NONE 0
50 #define QCOW_CRYPT_AES  1
51 
52 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
53 
54 typedef struct QCowHeader {
55     uint32_t magic;
56     uint32_t version;
57     uint64_t backing_file_offset;
58     uint32_t backing_file_size;
59     uint32_t mtime;
60     uint64_t size; /* in bytes */
61     uint8_t cluster_bits;
62     uint8_t l2_bits;
63     uint16_t padding;
64     uint32_t crypt_method;
65     uint64_t l1_table_offset;
66 } QEMU_PACKED QCowHeader;
67 
68 #define L2_CACHE_SIZE 16
69 
70 typedef struct BDRVQcowState {
71     int cluster_bits;
72     int cluster_size;
73     int l2_bits;
74     int l2_size;
75     unsigned int l1_size;
76     uint64_t cluster_offset_mask;
77     uint64_t l1_table_offset;
78     uint64_t *l1_table;
79     uint64_t *l2_cache;
80     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
81     uint32_t l2_cache_counts[L2_CACHE_SIZE];
82     uint8_t *cluster_cache;
83     uint8_t *cluster_data;
84     uint64_t cluster_cache_offset;
85     QCryptoBlock *crypto; /* Disk encryption format driver */
86     uint32_t crypt_method_header;
87     CoMutex lock;
88     Error *migration_blocker;
89 } BDRVQcowState;
90 
91 static QemuOptsList qcow_create_opts;
92 
93 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
94 
95 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
96 {
97     const QCowHeader *cow_header = (const void *)buf;
98 
99     if (buf_size >= sizeof(QCowHeader) &&
100         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
101         be32_to_cpu(cow_header->version) == QCOW_VERSION)
102         return 100;
103     else
104         return 0;
105 }
106 
107 static QemuOptsList qcow_runtime_opts = {
108     .name = "qcow",
109     .head = QTAILQ_HEAD_INITIALIZER(qcow_runtime_opts.head),
110     .desc = {
111         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
112         { /* end of list */ }
113     },
114 };
115 
116 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
117                      Error **errp)
118 {
119     BDRVQcowState *s = bs->opaque;
120     unsigned int len, i, shift;
121     int ret;
122     QCowHeader header;
123     Error *local_err = NULL;
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_file,
133                                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     be32_to_cpus(&header.magic);
144     be32_to_cpus(&header.version);
145     be64_to_cpus(&header.backing_file_offset);
146     be32_to_cpus(&header.backing_file_size);
147     be32_to_cpus(&header.mtime);
148     be64_to_cpus(&header.size);
149     be32_to_cpus(&header.crypt_method);
150     be64_to_cpus(&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, "Unsupported qcow version %" PRIu32, header.version);
159         ret = -ENOTSUP;
160         goto fail;
161     }
162 
163     if (header.size <= 1) {
164         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
165         ret = -EINVAL;
166         goto fail;
167     }
168     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
169         error_setg(errp, "Cluster size must be between 512 and 64k");
170         ret = -EINVAL;
171         goto fail;
172     }
173 
174     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
175      * so bytes = num_entries << 3. */
176     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
177         error_setg(errp, "L2 table size must be between 512 and 64k");
178         ret = -EINVAL;
179         goto fail;
180     }
181 
182     s->crypt_method_header = header.crypt_method;
183     if (s->crypt_method_header) {
184         if (bdrv_uses_whitelist() &&
185             s->crypt_method_header == QCOW_CRYPT_AES) {
186             error_setg(errp,
187                        "Use of AES-CBC encrypted qcow images is no longer "
188                        "supported in system emulators");
189             error_append_hint(errp,
190                               "You can use 'qemu-img convert' to convert your "
191                               "image to an alternative supported format, such "
192                               "as unencrypted qcow, or raw with the LUKS "
193                               "format instead.\n");
194             ret = -ENOSYS;
195             goto fail;
196         }
197         if (s->crypt_method_header == QCOW_CRYPT_AES) {
198             if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
199                 error_setg(errp,
200                            "Header reported 'aes' encryption format but "
201                            "options specify '%s'", encryptfmt);
202                 ret = -EINVAL;
203                 goto fail;
204             }
205             qdict_put_str(encryptopts, "format", "qcow");
206             crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
207             if (!crypto_opts) {
208                 ret = -EINVAL;
209                 goto fail;
210             }
211 
212             if (flags & BDRV_O_NO_IO) {
213                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
214             }
215             s->crypto = qcrypto_block_open(crypto_opts, "encrypt.",
216                                            NULL, NULL, cflags, errp);
217             if (!s->crypto) {
218                 ret = -EINVAL;
219                 goto fail;
220             }
221         } else {
222             error_setg(errp, "invalid encryption method in qcow header");
223             ret = -EINVAL;
224             goto fail;
225         }
226         bs->encrypted = true;
227     } else {
228         if (encryptfmt) {
229             error_setg(errp, "No encryption in image header, but options "
230                        "specified format '%s'", encryptfmt);
231             ret = -EINVAL;
232             goto fail;
233         }
234     }
235     s->cluster_bits = header.cluster_bits;
236     s->cluster_size = 1 << s->cluster_bits;
237     s->l2_bits = header.l2_bits;
238     s->l2_size = 1 << s->l2_bits;
239     bs->total_sectors = header.size / 512;
240     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
241 
242     /* read the level 1 table */
243     shift = s->cluster_bits + s->l2_bits;
244     if (header.size > UINT64_MAX - (1LL << shift)) {
245         error_setg(errp, "Image too large");
246         ret = -EINVAL;
247         goto fail;
248     } else {
249         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
250         if (l1_size > INT_MAX / sizeof(uint64_t)) {
251             error_setg(errp, "Image too large");
252             ret = -EINVAL;
253             goto fail;
254         }
255         s->l1_size = l1_size;
256     }
257 
258     s->l1_table_offset = header.l1_table_offset;
259     s->l1_table = g_try_new(uint64_t, s->l1_size);
260     if (s->l1_table == NULL) {
261         error_setg(errp, "Could not allocate memory for L1 table");
262         ret = -ENOMEM;
263         goto fail;
264     }
265 
266     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
267                s->l1_size * sizeof(uint64_t));
268     if (ret < 0) {
269         goto fail;
270     }
271 
272     for(i = 0;i < s->l1_size; i++) {
273         be64_to_cpus(&s->l1_table[i]);
274     }
275 
276     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
277     s->l2_cache =
278         qemu_try_blockalign(bs->file->bs,
279                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
280     if (s->l2_cache == NULL) {
281         error_setg(errp, "Could not allocate L2 table cache");
282         ret = -ENOMEM;
283         goto fail;
284     }
285     s->cluster_cache = g_malloc(s->cluster_size);
286     s->cluster_data = g_malloc(s->cluster_size);
287     s->cluster_cache_offset = -1;
288 
289     /* read the backing file name */
290     if (header.backing_file_offset != 0) {
291         len = header.backing_file_size;
292         if (len > 1023 || len >= sizeof(bs->backing_file)) {
293             error_setg(errp, "Backing file name too long");
294             ret = -EINVAL;
295             goto fail;
296         }
297         ret = bdrv_pread(bs->file, header.backing_file_offset,
298                    bs->backing_file, len);
299         if (ret < 0) {
300             goto fail;
301         }
302         bs->backing_file[len] = '\0';
303     }
304 
305     /* Disable migration when qcow images are used */
306     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
307                "does not support live migration",
308                bdrv_get_device_or_node_name(bs));
309     ret = migrate_add_blocker(s->migration_blocker, &local_err);
310     if (local_err) {
311         error_propagate(errp, local_err);
312         error_free(s->migration_blocker);
313         goto fail;
314     }
315 
316     qobject_unref(encryptopts);
317     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
318     qemu_co_mutex_init(&s->lock);
319     return 0;
320 
321  fail:
322     g_free(s->l1_table);
323     qemu_vfree(s->l2_cache);
324     g_free(s->cluster_cache);
325     g_free(s->cluster_data);
326     qcrypto_block_free(s->crypto);
327     qobject_unref(encryptopts);
328     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
329     return ret;
330 }
331 
332 
333 /* We have nothing to do for QCOW reopen, stubs just return
334  * success */
335 static int qcow_reopen_prepare(BDRVReopenState *state,
336                                BlockReopenQueue *queue, Error **errp)
337 {
338     return 0;
339 }
340 
341 
342 /* 'allocate' is:
343  *
344  * 0 to not allocate.
345  *
346  * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start'
347  * to 'n_end' within the cluster)
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, 1 if *result is assigned, and negative
354  * errno on failure.
355  */
356 static int get_cluster_offset(BlockDriverState *bs,
357                               uint64_t offset, int allocate,
358                               int compressed_size,
359                               int n_start, int n_end, uint64_t *result)
360 {
361     BDRVQcowState *s = bs->opaque;
362     int min_index, i, j, l1_index, l2_index, ret;
363     int64_t l2_offset;
364     uint64_t *l2_table, cluster_offset, tmp;
365     uint32_t min_count;
366     int new_l2_table;
367 
368     *result = 0;
369     l1_index = offset >> (s->l2_bits + s->cluster_bits);
370     l2_offset = s->l1_table[l1_index];
371     new_l2_table = 0;
372     if (!l2_offset) {
373         if (!allocate)
374             return 0;
375         /* allocate a new l2 entry */
376         l2_offset = bdrv_getlength(bs->file->bs);
377         if (l2_offset < 0) {
378             return l2_offset;
379         }
380         /* round to cluster size */
381         l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size);
382         /* update the L1 entry */
383         s->l1_table[l1_index] = l2_offset;
384         tmp = cpu_to_be64(l2_offset);
385         BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
386         ret = bdrv_pwrite_sync(bs->file,
387                                s->l1_table_offset + l1_index * sizeof(tmp),
388                                &tmp, sizeof(tmp));
389         if (ret < 0) {
390             return ret;
391         }
392         new_l2_table = 1;
393     }
394     for(i = 0; i < L2_CACHE_SIZE; i++) {
395         if (l2_offset == s->l2_cache_offsets[i]) {
396             /* increment the hit count */
397             if (++s->l2_cache_counts[i] == 0xffffffff) {
398                 for(j = 0; j < L2_CACHE_SIZE; j++) {
399                     s->l2_cache_counts[j] >>= 1;
400                 }
401             }
402             l2_table = s->l2_cache + (i << s->l2_bits);
403             goto found;
404         }
405     }
406     /* not found: load a new entry in the least used one */
407     min_index = 0;
408     min_count = 0xffffffff;
409     for(i = 0; i < L2_CACHE_SIZE; i++) {
410         if (s->l2_cache_counts[i] < min_count) {
411             min_count = s->l2_cache_counts[i];
412             min_index = i;
413         }
414     }
415     l2_table = s->l2_cache + (min_index << s->l2_bits);
416     BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
417     if (new_l2_table) {
418         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
419         ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
420                                s->l2_size * sizeof(uint64_t));
421         if (ret < 0) {
422             return ret;
423         }
424     } else {
425         ret = bdrv_pread(bs->file, l2_offset, l2_table,
426                          s->l2_size * sizeof(uint64_t));
427         if (ret < 0) {
428             return ret;
429         }
430     }
431     s->l2_cache_offsets[min_index] = l2_offset;
432     s->l2_cache_counts[min_index] = 1;
433  found:
434     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
435     cluster_offset = be64_to_cpu(l2_table[l2_index]);
436     if (!cluster_offset ||
437         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
438         if (!allocate)
439             return 0;
440         BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
441         assert(QEMU_IS_ALIGNED(n_start | n_end, BDRV_SECTOR_SIZE));
442         /* allocate a new cluster */
443         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
444             (n_end - n_start) < s->cluster_size) {
445             /* if the cluster is already compressed, we must
446                decompress it in the case it is not completely
447                overwritten */
448             if (decompress_cluster(bs, cluster_offset) < 0) {
449                 return -EIO;
450             }
451             cluster_offset = bdrv_getlength(bs->file->bs);
452             if ((int64_t) cluster_offset < 0) {
453                 return cluster_offset;
454             }
455             cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
456             /* write the cluster content */
457             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
458             ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
459                               s->cluster_size);
460             if (ret < 0) {
461                 return ret;
462             }
463         } else {
464             cluster_offset = bdrv_getlength(bs->file->bs);
465             if ((int64_t) cluster_offset < 0) {
466                 return cluster_offset;
467             }
468             if (allocate == 1) {
469                 /* round to cluster size */
470                 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
471                 if (cluster_offset + s->cluster_size > INT64_MAX) {
472                     return -E2BIG;
473                 }
474                 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size,
475                                     PREALLOC_MODE_OFF, NULL);
476                 if (ret < 0) {
477                     return ret;
478                 }
479                 /* if encrypted, we must initialize the cluster
480                    content which won't be written */
481                 if (bs->encrypted &&
482                     (n_end - n_start) < s->cluster_size) {
483                     uint64_t start_offset;
484                     assert(s->crypto);
485                     start_offset = offset & ~(s->cluster_size - 1);
486                     for (i = 0; i < s->cluster_size; i += BDRV_SECTOR_SIZE) {
487                         if (i < n_start || i >= n_end) {
488                             memset(s->cluster_data, 0x00, BDRV_SECTOR_SIZE);
489                             if (qcrypto_block_encrypt(s->crypto,
490                                                       start_offset + i,
491                                                       s->cluster_data,
492                                                       BDRV_SECTOR_SIZE,
493                                                       NULL) < 0) {
494                                 return -EIO;
495                             }
496                             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
497                             ret = bdrv_pwrite(bs->file,
498                                               cluster_offset + i,
499                                               s->cluster_data,
500                                               BDRV_SECTOR_SIZE);
501                             if (ret < 0) {
502                                 return ret;
503                             }
504                         }
505                     }
506                 }
507             } else if (allocate == 2) {
508                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
509                     (uint64_t)compressed_size << (63 - s->cluster_bits);
510             }
511         }
512         /* update L2 table */
513         tmp = cpu_to_be64(cluster_offset);
514         l2_table[l2_index] = tmp;
515         if (allocate == 2) {
516             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
517         } else {
518             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
519         }
520         ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
521                                &tmp, sizeof(tmp));
522         if (ret < 0) {
523             return ret;
524         }
525     }
526     *result = cluster_offset;
527     return 1;
528 }
529 
530 static int coroutine_fn qcow_co_block_status(BlockDriverState *bs,
531                                              bool want_zero,
532                                              int64_t offset, int64_t bytes,
533                                              int64_t *pnum, int64_t *map,
534                                              BlockDriverState **file)
535 {
536     BDRVQcowState *s = bs->opaque;
537     int index_in_cluster, ret;
538     int64_t n;
539     uint64_t cluster_offset;
540 
541     qemu_co_mutex_lock(&s->lock);
542     ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
543     qemu_co_mutex_unlock(&s->lock);
544     if (ret < 0) {
545         return ret;
546     }
547     index_in_cluster = offset & (s->cluster_size - 1);
548     n = s->cluster_size - index_in_cluster;
549     if (n > bytes) {
550         n = bytes;
551     }
552     *pnum = n;
553     if (!cluster_offset) {
554         return 0;
555     }
556     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) {
557         return BDRV_BLOCK_DATA;
558     }
559     *map = cluster_offset | index_in_cluster;
560     *file = bs->file->bs;
561     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
562 }
563 
564 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
565                              const uint8_t *buf, int buf_size)
566 {
567     z_stream strm1, *strm = &strm1;
568     int ret, out_len;
569 
570     memset(strm, 0, sizeof(*strm));
571 
572     strm->next_in = (uint8_t *)buf;
573     strm->avail_in = buf_size;
574     strm->next_out = out_buf;
575     strm->avail_out = out_buf_size;
576 
577     ret = inflateInit2(strm, -12);
578     if (ret != Z_OK)
579         return -1;
580     ret = inflate(strm, Z_FINISH);
581     out_len = strm->next_out - out_buf;
582     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
583         out_len != out_buf_size) {
584         inflateEnd(strm);
585         return -1;
586     }
587     inflateEnd(strm);
588     return 0;
589 }
590 
591 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
592 {
593     BDRVQcowState *s = bs->opaque;
594     int ret, csize;
595     uint64_t coffset;
596 
597     coffset = cluster_offset & s->cluster_offset_mask;
598     if (s->cluster_cache_offset != coffset) {
599         csize = cluster_offset >> (63 - s->cluster_bits);
600         csize &= (s->cluster_size - 1);
601         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
602         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
603         if (ret != csize)
604             return -1;
605         if (decompress_buffer(s->cluster_cache, s->cluster_size,
606                               s->cluster_data, csize) < 0) {
607             return -1;
608         }
609         s->cluster_cache_offset = coffset;
610     }
611     return 0;
612 }
613 
614 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp)
615 {
616     /* At least encrypted images require 512-byte alignment. Apply the
617      * limit universally, rather than just on encrypted images, as
618      * it's easier to let the block layer handle rounding than to
619      * audit this code further. */
620     bs->bl.request_alignment = BDRV_SECTOR_SIZE;
621 }
622 
623 static coroutine_fn int qcow_co_preadv(BlockDriverState *bs, uint64_t offset,
624                                        uint64_t bytes, QEMUIOVector *qiov,
625                                        int flags)
626 {
627     BDRVQcowState *s = bs->opaque;
628     int offset_in_cluster;
629     int ret = 0, n;
630     uint64_t cluster_offset;
631     struct iovec hd_iov;
632     QEMUIOVector hd_qiov;
633     uint8_t *buf;
634     void *orig_buf;
635 
636     assert(!flags);
637     if (qiov->niov > 1) {
638         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
639         if (buf == NULL) {
640             return -ENOMEM;
641         }
642     } else {
643         orig_buf = NULL;
644         buf = (uint8_t *)qiov->iov->iov_base;
645     }
646 
647     qemu_co_mutex_lock(&s->lock);
648 
649     while (bytes != 0) {
650         /* prepare next request */
651         ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
652         if (ret < 0) {
653             break;
654         }
655         offset_in_cluster = offset & (s->cluster_size - 1);
656         n = s->cluster_size - offset_in_cluster;
657         if (n > bytes) {
658             n = bytes;
659         }
660 
661         if (!cluster_offset) {
662             if (bs->backing) {
663                 /* read from the base image */
664                 hd_iov.iov_base = (void *)buf;
665                 hd_iov.iov_len = n;
666                 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
667                 qemu_co_mutex_unlock(&s->lock);
668                 /* qcow2 emits this on bs->file instead of bs->backing */
669                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
670                 ret = bdrv_co_preadv(bs->backing, offset, n, &hd_qiov, 0);
671                 qemu_co_mutex_lock(&s->lock);
672                 if (ret < 0) {
673                     break;
674                 }
675             } else {
676                 /* Note: in this case, no need to wait */
677                 memset(buf, 0, n);
678             }
679         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
680             /* add AIO support for compressed blocks ? */
681             if (decompress_cluster(bs, cluster_offset) < 0) {
682                 ret = -EIO;
683                 break;
684             }
685             memcpy(buf, s->cluster_cache + offset_in_cluster, n);
686         } else {
687             if ((cluster_offset & 511) != 0) {
688                 ret = -EIO;
689                 break;
690             }
691             hd_iov.iov_base = (void *)buf;
692             hd_iov.iov_len = n;
693             qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
694             qemu_co_mutex_unlock(&s->lock);
695             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
696             ret = bdrv_co_preadv(bs->file, cluster_offset + offset_in_cluster,
697                                  n, &hd_qiov, 0);
698             qemu_co_mutex_lock(&s->lock);
699             if (ret < 0) {
700                 break;
701             }
702             if (bs->encrypted) {
703                 assert(s->crypto);
704                 if (qcrypto_block_decrypt(s->crypto,
705                                           offset, buf, n, NULL) < 0) {
706                     ret = -EIO;
707                     break;
708                 }
709             }
710         }
711         ret = 0;
712 
713         bytes -= n;
714         offset += n;
715         buf += n;
716     }
717 
718     qemu_co_mutex_unlock(&s->lock);
719 
720     if (qiov->niov > 1) {
721         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
722         qemu_vfree(orig_buf);
723     }
724 
725     return ret;
726 }
727 
728 static coroutine_fn int qcow_co_pwritev(BlockDriverState *bs, uint64_t offset,
729                                         uint64_t bytes, QEMUIOVector *qiov,
730                                         int flags)
731 {
732     BDRVQcowState *s = bs->opaque;
733     int offset_in_cluster;
734     uint64_t cluster_offset;
735     int ret = 0, n;
736     struct iovec hd_iov;
737     QEMUIOVector hd_qiov;
738     uint8_t *buf;
739     void *orig_buf;
740 
741     assert(!flags);
742     s->cluster_cache_offset = -1; /* disable compressed cache */
743 
744     /* We must always copy the iov when encrypting, so we
745      * don't modify the original data buffer during encryption */
746     if (bs->encrypted || qiov->niov > 1) {
747         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
748         if (buf == NULL) {
749             return -ENOMEM;
750         }
751         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
752     } else {
753         orig_buf = NULL;
754         buf = (uint8_t *)qiov->iov->iov_base;
755     }
756 
757     qemu_co_mutex_lock(&s->lock);
758 
759     while (bytes != 0) {
760         offset_in_cluster = offset & (s->cluster_size - 1);
761         n = s->cluster_size - offset_in_cluster;
762         if (n > bytes) {
763             n = bytes;
764         }
765         ret = get_cluster_offset(bs, offset, 1, 0, offset_in_cluster,
766                                  offset_in_cluster + n, &cluster_offset);
767         if (ret < 0) {
768             break;
769         }
770         if (!cluster_offset || (cluster_offset & 511) != 0) {
771             ret = -EIO;
772             break;
773         }
774         if (bs->encrypted) {
775             assert(s->crypto);
776             if (qcrypto_block_encrypt(s->crypto, offset, buf, n, NULL) < 0) {
777                 ret = -EIO;
778                 break;
779             }
780         }
781 
782         hd_iov.iov_base = (void *)buf;
783         hd_iov.iov_len = n;
784         qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
785         qemu_co_mutex_unlock(&s->lock);
786         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
787         ret = bdrv_co_pwritev(bs->file, cluster_offset + offset_in_cluster,
788                               n, &hd_qiov, 0);
789         qemu_co_mutex_lock(&s->lock);
790         if (ret < 0) {
791             break;
792         }
793         ret = 0;
794 
795         bytes -= n;
796         offset += n;
797         buf += n;
798     }
799     qemu_co_mutex_unlock(&s->lock);
800 
801     qemu_vfree(orig_buf);
802 
803     return ret;
804 }
805 
806 static void qcow_close(BlockDriverState *bs)
807 {
808     BDRVQcowState *s = bs->opaque;
809 
810     qcrypto_block_free(s->crypto);
811     s->crypto = NULL;
812     g_free(s->l1_table);
813     qemu_vfree(s->l2_cache);
814     g_free(s->cluster_cache);
815     g_free(s->cluster_data);
816 
817     migrate_del_blocker(s->migration_blocker);
818     error_free(s->migration_blocker);
819 }
820 
821 static int coroutine_fn qcow_co_create(BlockdevCreateOptions *opts,
822                                        Error **errp)
823 {
824     BlockdevCreateOptionsQcow *qcow_opts;
825     int header_size, backing_filename_len, l1_size, shift, i;
826     QCowHeader header;
827     uint8_t *tmp;
828     int64_t total_size = 0;
829     int ret;
830     BlockDriverState *bs;
831     BlockBackend *qcow_blk;
832     QCryptoBlock *crypto = NULL;
833 
834     assert(opts->driver == BLOCKDEV_DRIVER_QCOW);
835     qcow_opts = &opts->u.qcow;
836 
837     /* Sanity checks */
838     total_size = qcow_opts->size;
839     if (total_size == 0) {
840         error_setg(errp, "Image size is too small, cannot be zero length");
841         return -EINVAL;
842     }
843 
844     if (qcow_opts->has_encrypt &&
845         qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW)
846     {
847         error_setg(errp, "Unsupported encryption format");
848         return -EINVAL;
849     }
850 
851     /* Create BlockBackend to write to the image */
852     bs = bdrv_open_blockdev_ref(qcow_opts->file, errp);
853     if (bs == NULL) {
854         return -EIO;
855     }
856 
857     qcow_blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
858     ret = blk_insert_bs(qcow_blk, bs, errp);
859     if (ret < 0) {
860         goto exit;
861     }
862     blk_set_allow_write_beyond_eof(qcow_blk, true);
863 
864     /* Create image format */
865     ret = blk_truncate(qcow_blk, 0, PREALLOC_MODE_OFF, errp);
866     if (ret < 0) {
867         goto exit;
868     }
869 
870     memset(&header, 0, sizeof(header));
871     header.magic = cpu_to_be32(QCOW_MAGIC);
872     header.version = cpu_to_be32(QCOW_VERSION);
873     header.size = cpu_to_be64(total_size);
874     header_size = sizeof(header);
875     backing_filename_len = 0;
876     if (qcow_opts->has_backing_file) {
877         if (strcmp(qcow_opts->backing_file, "fat:")) {
878             header.backing_file_offset = cpu_to_be64(header_size);
879             backing_filename_len = strlen(qcow_opts->backing_file);
880             header.backing_file_size = cpu_to_be32(backing_filename_len);
881             header_size += backing_filename_len;
882         } else {
883             /* special backing file for vvfat */
884             qcow_opts->has_backing_file = false;
885         }
886         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
887                                     unmodified sectors */
888         header.l2_bits = 12; /* 32 KB L2 tables */
889     } else {
890         header.cluster_bits = 12; /* 4 KB clusters */
891         header.l2_bits = 9; /* 4 KB L2 tables */
892     }
893     header_size = (header_size + 7) & ~7;
894     shift = header.cluster_bits + header.l2_bits;
895     l1_size = (total_size + (1LL << shift) - 1) >> shift;
896 
897     header.l1_table_offset = cpu_to_be64(header_size);
898 
899     if (qcow_opts->has_encrypt) {
900         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
901 
902         crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.",
903                                       NULL, NULL, NULL, errp);
904         if (!crypto) {
905             ret = -EINVAL;
906             goto exit;
907         }
908     } else {
909         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
910     }
911 
912     /* write all the data */
913     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
914     if (ret != sizeof(header)) {
915         goto exit;
916     }
917 
918     if (qcow_opts->has_backing_file) {
919         ret = blk_pwrite(qcow_blk, sizeof(header),
920                          qcow_opts->backing_file, backing_filename_len, 0);
921         if (ret != backing_filename_len) {
922             goto exit;
923         }
924     }
925 
926     tmp = g_malloc0(BDRV_SECTOR_SIZE);
927     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
928          i++) {
929         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
930                          tmp, BDRV_SECTOR_SIZE, 0);
931         if (ret != BDRV_SECTOR_SIZE) {
932             g_free(tmp);
933             goto exit;
934         }
935     }
936 
937     g_free(tmp);
938     ret = 0;
939 exit:
940     blk_unref(qcow_blk);
941     qcrypto_block_free(crypto);
942     return ret;
943 }
944 
945 static int coroutine_fn qcow_co_create_opts(const char *filename,
946                                             QemuOpts *opts, Error **errp)
947 {
948     BlockdevCreateOptions *create_options = NULL;
949     BlockDriverState *bs = NULL;
950     QDict *qdict;
951     Visitor *v;
952     const char *val;
953     Error *local_err = NULL;
954     int ret;
955 
956     static const QDictRenames opt_renames[] = {
957         { BLOCK_OPT_BACKING_FILE,       "backing-file" },
958         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
959         { NULL, NULL },
960     };
961 
962     /* Parse options and convert legacy syntax */
963     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true);
964 
965     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
966     if (val && !strcmp(val, "on")) {
967         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
968     } else if (val && !strcmp(val, "off")) {
969         qdict_del(qdict, BLOCK_OPT_ENCRYPT);
970     }
971 
972     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
973     if (val && !strcmp(val, "aes")) {
974         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
975     }
976 
977     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
978         ret = -EINVAL;
979         goto fail;
980     }
981 
982     /* Create and open the file (protocol layer) */
983     ret = bdrv_create_file(filename, opts, &local_err);
984     if (ret < 0) {
985         error_propagate(errp, local_err);
986         goto fail;
987     }
988 
989     bs = bdrv_open(filename, NULL, NULL,
990                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
991     if (bs == NULL) {
992         ret = -EIO;
993         goto fail;
994     }
995 
996     /* Now get the QAPI type BlockdevCreateOptions */
997     qdict_put_str(qdict, "driver", "qcow");
998     qdict_put_str(qdict, "file", bs->node_name);
999 
1000     v = qobject_input_visitor_new_flat_confused(qdict, errp);
1001     if (!v) {
1002         ret = -EINVAL;
1003         goto fail;
1004     }
1005 
1006     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
1007     visit_free(v);
1008 
1009     if (local_err) {
1010         error_propagate(errp, local_err);
1011         ret = -EINVAL;
1012         goto fail;
1013     }
1014 
1015     /* Silently round up size */
1016     assert(create_options->driver == BLOCKDEV_DRIVER_QCOW);
1017     create_options->u.qcow.size =
1018         ROUND_UP(create_options->u.qcow.size, BDRV_SECTOR_SIZE);
1019 
1020     /* Create the qcow image (format layer) */
1021     ret = qcow_co_create(create_options, errp);
1022     if (ret < 0) {
1023         goto fail;
1024     }
1025 
1026     ret = 0;
1027 fail:
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,
1045                         PREALLOC_MODE_OFF, 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     QEMUIOVector hd_qiov;
1064     struct iovec iov;
1065     z_stream strm;
1066     int ret, out_len;
1067     uint8_t *buf, *out_buf;
1068     uint64_t cluster_offset;
1069 
1070     buf = qemu_blockalign(bs, s->cluster_size);
1071     if (bytes != s->cluster_size) {
1072         if (bytes > s->cluster_size ||
1073             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
1074         {
1075             qemu_vfree(buf);
1076             return -EINVAL;
1077         }
1078         /* Zero-pad last write if image size is not cluster aligned */
1079         memset(buf + bytes, 0, s->cluster_size - bytes);
1080     }
1081     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
1082 
1083     out_buf = g_malloc(s->cluster_size);
1084 
1085     /* best compression, small window, no zlib header */
1086     memset(&strm, 0, sizeof(strm));
1087     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1088                        Z_DEFLATED, -12,
1089                        9, Z_DEFAULT_STRATEGY);
1090     if (ret != 0) {
1091         ret = -EINVAL;
1092         goto fail;
1093     }
1094 
1095     strm.avail_in = s->cluster_size;
1096     strm.next_in = (uint8_t *)buf;
1097     strm.avail_out = s->cluster_size;
1098     strm.next_out = out_buf;
1099 
1100     ret = deflate(&strm, Z_FINISH);
1101     if (ret != Z_STREAM_END && ret != Z_OK) {
1102         deflateEnd(&strm);
1103         ret = -EINVAL;
1104         goto fail;
1105     }
1106     out_len = strm.next_out - out_buf;
1107 
1108     deflateEnd(&strm);
1109 
1110     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1111         /* could not compress: write normal cluster */
1112         ret = qcow_co_pwritev(bs, offset, bytes, qiov, 0);
1113         if (ret < 0) {
1114             goto fail;
1115         }
1116         goto success;
1117     }
1118     qemu_co_mutex_lock(&s->lock);
1119     ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1120     qemu_co_mutex_unlock(&s->lock);
1121     if (ret < 0) {
1122         goto fail;
1123     }
1124     if (cluster_offset == 0) {
1125         ret = -EIO;
1126         goto fail;
1127     }
1128     cluster_offset &= s->cluster_offset_mask;
1129 
1130     iov = (struct iovec) {
1131         .iov_base   = out_buf,
1132         .iov_len    = out_len,
1133     };
1134     qemu_iovec_init_external(&hd_qiov, &iov, 1);
1135     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1136     ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
1137     if (ret < 0) {
1138         goto fail;
1139     }
1140 success:
1141     ret = 0;
1142 fail:
1143     qemu_vfree(buf);
1144     g_free(out_buf);
1145     return ret;
1146 }
1147 
1148 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1149 {
1150     BDRVQcowState *s = bs->opaque;
1151     bdi->cluster_size = s->cluster_size;
1152     return 0;
1153 }
1154 
1155 static QemuOptsList qcow_create_opts = {
1156     .name = "qcow-create-opts",
1157     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1158     .desc = {
1159         {
1160             .name = BLOCK_OPT_SIZE,
1161             .type = QEMU_OPT_SIZE,
1162             .help = "Virtual disk size"
1163         },
1164         {
1165             .name = BLOCK_OPT_BACKING_FILE,
1166             .type = QEMU_OPT_STRING,
1167             .help = "File name of a base image"
1168         },
1169         {
1170             .name = BLOCK_OPT_ENCRYPT,
1171             .type = QEMU_OPT_BOOL,
1172             .help = "Encrypt the image with format 'aes'. (Deprecated "
1173                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1174         },
1175         {
1176             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1177             .type = QEMU_OPT_STRING,
1178             .help = "Encrypt the image, format choices: 'aes'",
1179         },
1180         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1181         { /* end of list */ }
1182     }
1183 };
1184 
1185 static BlockDriver bdrv_qcow = {
1186     .format_name	= "qcow",
1187     .instance_size	= sizeof(BDRVQcowState),
1188     .bdrv_probe		= qcow_probe,
1189     .bdrv_open		= qcow_open,
1190     .bdrv_close		= qcow_close,
1191     .bdrv_child_perm        = bdrv_format_default_perms,
1192     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1193     .bdrv_co_create         = qcow_co_create,
1194     .bdrv_co_create_opts    = qcow_co_create_opts,
1195     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1196     .supports_backing       = true,
1197     .bdrv_refresh_limits    = qcow_refresh_limits,
1198 
1199     .bdrv_co_preadv         = qcow_co_preadv,
1200     .bdrv_co_pwritev        = qcow_co_pwritev,
1201     .bdrv_co_block_status   = qcow_co_block_status,
1202 
1203     .bdrv_make_empty        = qcow_make_empty,
1204     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1205     .bdrv_get_info          = qcow_get_info,
1206 
1207     .create_opts            = &qcow_create_opts,
1208 };
1209 
1210 static void bdrv_qcow_init(void)
1211 {
1212     bdrv_register(&bdrv_qcow);
1213 }
1214 
1215 block_init(bdrv_qcow_init);
1216