xref: /openbmc/qemu/block/qcow.c (revision 1dcea719)
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 "qemu/memalign.h"
36 #include <zlib.h>
37 #include "qapi/qmp/qdict.h"
38 #include "qapi/qmp/qstring.h"
39 #include "qapi/qobject-input-visitor.h"
40 #include "qapi/qapi-visit-block-core.h"
41 #include "crypto/block.h"
42 #include "migration/blocker.h"
43 #include "crypto.h"
44 
45 /**************************************************************/
46 /* QEMU COW block driver with compression and encryption support */
47 
48 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
49 #define QCOW_VERSION 1
50 
51 #define QCOW_CRYPT_NONE 0
52 #define QCOW_CRYPT_AES  1
53 
54 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
55 
56 typedef struct QCowHeader {
57     uint32_t magic;
58     uint32_t version;
59     uint64_t backing_file_offset;
60     uint32_t backing_file_size;
61     uint32_t mtime;
62     uint64_t size; /* in bytes */
63     uint8_t cluster_bits;
64     uint8_t l2_bits;
65     uint16_t padding;
66     uint32_t crypt_method;
67     uint64_t l1_table_offset;
68 } QEMU_PACKED QCowHeader;
69 
70 #define L2_CACHE_SIZE 16
71 
72 typedef struct BDRVQcowState {
73     int cluster_bits;
74     int cluster_size;
75     int l2_bits;
76     int l2_size;
77     unsigned int l1_size;
78     uint64_t cluster_offset_mask;
79     uint64_t l1_table_offset;
80     uint64_t *l1_table;
81     uint64_t *l2_cache;
82     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
83     uint32_t l2_cache_counts[L2_CACHE_SIZE];
84     uint8_t *cluster_cache;
85     uint8_t *cluster_data;
86     uint64_t cluster_cache_offset;
87     QCryptoBlock *crypto; /* Disk encryption format driver */
88     uint32_t crypt_method_header;
89     CoMutex lock;
90     Error *migration_blocker;
91 } BDRVQcowState;
92 
93 static QemuOptsList qcow_create_opts;
94 
95 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
96 
97 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
98 {
99     const QCowHeader *cow_header = (const void *)buf;
100 
101     if (buf_size >= sizeof(QCowHeader) &&
102         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
103         be32_to_cpu(cow_header->version) == QCOW_VERSION)
104         return 100;
105     else
106         return 0;
107 }
108 
109 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
110                      Error **errp)
111 {
112     BDRVQcowState *s = bs->opaque;
113     unsigned int len, i, shift;
114     int ret;
115     QCowHeader header;
116     QCryptoBlockOpenOptions *crypto_opts = NULL;
117     unsigned int cflags = 0;
118     QDict *encryptopts = NULL;
119     const char *encryptfmt;
120 
121     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
122     encryptfmt = qdict_get_try_str(encryptopts, "format");
123 
124     ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
125     if (ret < 0) {
126         goto fail;
127     }
128 
129     ret = bdrv_pread(bs->file, 0, sizeof(header), &header, 0);
130     if (ret < 0) {
131         goto fail;
132     }
133     header.magic = be32_to_cpu(header.magic);
134     header.version = be32_to_cpu(header.version);
135     header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
136     header.backing_file_size = be32_to_cpu(header.backing_file_size);
137     header.mtime = be32_to_cpu(header.mtime);
138     header.size = be64_to_cpu(header.size);
139     header.crypt_method = be32_to_cpu(header.crypt_method);
140     header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
141 
142     if (header.magic != QCOW_MAGIC) {
143         error_setg(errp, "Image not in qcow format");
144         ret = -EINVAL;
145         goto fail;
146     }
147     if (header.version != QCOW_VERSION) {
148         error_setg(errp, "qcow (v%d) does not support qcow version %" PRIu32,
149                    QCOW_VERSION, header.version);
150         if (header.version == 2 || header.version == 3) {
151             error_append_hint(errp, "Try the 'qcow2' driver instead.\n");
152         }
153 
154         ret = -ENOTSUP;
155         goto fail;
156     }
157 
158     if (header.size <= 1) {
159         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
160         ret = -EINVAL;
161         goto fail;
162     }
163     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
164         error_setg(errp, "Cluster size must be between 512 and 64k");
165         ret = -EINVAL;
166         goto fail;
167     }
168 
169     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
170      * so bytes = num_entries << 3. */
171     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
172         error_setg(errp, "L2 table size must be between 512 and 64k");
173         ret = -EINVAL;
174         goto fail;
175     }
176 
177     s->crypt_method_header = header.crypt_method;
178     if (s->crypt_method_header) {
179         if (bdrv_uses_whitelist() &&
180             s->crypt_method_header == QCOW_CRYPT_AES) {
181             error_setg(errp,
182                        "Use of AES-CBC encrypted qcow images is no longer "
183                        "supported in system emulators");
184             error_append_hint(errp,
185                               "You can use 'qemu-img convert' to convert your "
186                               "image to an alternative supported format, such "
187                               "as unencrypted qcow, or raw with the LUKS "
188                               "format instead.\n");
189             ret = -ENOSYS;
190             goto fail;
191         }
192         if (s->crypt_method_header == QCOW_CRYPT_AES) {
193             if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
194                 error_setg(errp,
195                            "Header reported 'aes' encryption format but "
196                            "options specify '%s'", encryptfmt);
197                 ret = -EINVAL;
198                 goto fail;
199             }
200             qdict_put_str(encryptopts, "format", "qcow");
201             crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
202             if (!crypto_opts) {
203                 ret = -EINVAL;
204                 goto fail;
205             }
206 
207             if (flags & BDRV_O_NO_IO) {
208                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
209             }
210             s->crypto = qcrypto_block_open(crypto_opts, "encrypt.",
211                                            NULL, NULL, cflags, 1, errp);
212             if (!s->crypto) {
213                 ret = -EINVAL;
214                 goto fail;
215             }
216         } else {
217             error_setg(errp, "invalid encryption method in qcow header");
218             ret = -EINVAL;
219             goto fail;
220         }
221         bs->encrypted = true;
222     } else {
223         if (encryptfmt) {
224             error_setg(errp, "No encryption in image header, but options "
225                        "specified format '%s'", encryptfmt);
226             ret = -EINVAL;
227             goto fail;
228         }
229     }
230     s->cluster_bits = header.cluster_bits;
231     s->cluster_size = 1 << s->cluster_bits;
232     s->l2_bits = header.l2_bits;
233     s->l2_size = 1 << s->l2_bits;
234     bs->total_sectors = header.size / 512;
235     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
236 
237     /* read the level 1 table */
238     shift = s->cluster_bits + s->l2_bits;
239     if (header.size > UINT64_MAX - (1LL << shift)) {
240         error_setg(errp, "Image too large");
241         ret = -EINVAL;
242         goto fail;
243     } else {
244         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
245         if (l1_size > INT_MAX / sizeof(uint64_t)) {
246             error_setg(errp, "Image too large");
247             ret = -EINVAL;
248             goto fail;
249         }
250         s->l1_size = l1_size;
251     }
252 
253     s->l1_table_offset = header.l1_table_offset;
254     s->l1_table = g_try_new(uint64_t, s->l1_size);
255     if (s->l1_table == NULL) {
256         error_setg(errp, "Could not allocate memory for L1 table");
257         ret = -ENOMEM;
258         goto fail;
259     }
260 
261     ret = bdrv_pread(bs->file, s->l1_table_offset,
262                      s->l1_size * sizeof(uint64_t), s->l1_table, 0);
263     if (ret < 0) {
264         goto fail;
265     }
266 
267     for(i = 0;i < s->l1_size; i++) {
268         s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
269     }
270 
271     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
272     s->l2_cache =
273         qemu_try_blockalign(bs->file->bs,
274                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
275     if (s->l2_cache == NULL) {
276         error_setg(errp, "Could not allocate L2 table cache");
277         ret = -ENOMEM;
278         goto fail;
279     }
280     s->cluster_cache = g_malloc(s->cluster_size);
281     s->cluster_data = g_malloc(s->cluster_size);
282     s->cluster_cache_offset = -1;
283 
284     /* read the backing file name */
285     if (header.backing_file_offset != 0) {
286         len = header.backing_file_size;
287         if (len > 1023 || len >= sizeof(bs->backing_file)) {
288             error_setg(errp, "Backing file name too long");
289             ret = -EINVAL;
290             goto fail;
291         }
292         ret = bdrv_pread(bs->file, header.backing_file_offset, len,
293                          bs->auto_backing_file, 0);
294         if (ret < 0) {
295             goto fail;
296         }
297         bs->auto_backing_file[len] = '\0';
298         pstrcpy(bs->backing_file, sizeof(bs->backing_file),
299                 bs->auto_backing_file);
300     }
301 
302     /* Disable migration when qcow images are used */
303     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
304                "does not support live migration",
305                bdrv_get_device_or_node_name(bs));
306     ret = migrate_add_blocker(s->migration_blocker, errp);
307     if (ret < 0) {
308         error_free(s->migration_blocker);
309         goto fail;
310     }
311 
312     qobject_unref(encryptopts);
313     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
314     qemu_co_mutex_init(&s->lock);
315     return 0;
316 
317  fail:
318     g_free(s->l1_table);
319     qemu_vfree(s->l2_cache);
320     g_free(s->cluster_cache);
321     g_free(s->cluster_data);
322     qcrypto_block_free(s->crypto);
323     qobject_unref(encryptopts);
324     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
325     return ret;
326 }
327 
328 
329 /* We have nothing to do for QCOW reopen, stubs just return
330  * success */
331 static int qcow_reopen_prepare(BDRVReopenState *state,
332                                BlockReopenQueue *queue, Error **errp)
333 {
334     return 0;
335 }
336 
337 
338 /* 'allocate' is:
339  *
340  * 0 to not allocate.
341  *
342  * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start'
343  * to 'n_end' within the cluster)
344  *
345  * 2 to allocate a compressed cluster of size
346  * 'compressed_size'. 'compressed_size' must be > 0 and <
347  * cluster_size
348  *
349  * return 0 if not allocated, 1 if *result is assigned, and negative
350  * errno on failure.
351  */
352 static int get_cluster_offset(BlockDriverState *bs,
353                               uint64_t offset, int allocate,
354                               int compressed_size,
355                               int n_start, int n_end, uint64_t *result)
356 {
357     BDRVQcowState *s = bs->opaque;
358     int min_index, i, j, l1_index, l2_index, ret;
359     int64_t l2_offset;
360     uint64_t *l2_table, cluster_offset, tmp;
361     uint32_t min_count;
362     int new_l2_table;
363 
364     *result = 0;
365     l1_index = offset >> (s->l2_bits + s->cluster_bits);
366     l2_offset = s->l1_table[l1_index];
367     new_l2_table = 0;
368     if (!l2_offset) {
369         if (!allocate)
370             return 0;
371         /* allocate a new l2 entry */
372         l2_offset = bdrv_getlength(bs->file->bs);
373         if (l2_offset < 0) {
374             return l2_offset;
375         }
376         /* round to cluster size */
377         l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size);
378         /* update the L1 entry */
379         s->l1_table[l1_index] = l2_offset;
380         tmp = cpu_to_be64(l2_offset);
381         BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
382         ret = bdrv_pwrite_sync(bs->file,
383                                s->l1_table_offset + l1_index * sizeof(tmp),
384                                sizeof(tmp), &tmp, 0);
385         if (ret < 0) {
386             return ret;
387         }
388         new_l2_table = 1;
389     }
390     for(i = 0; i < L2_CACHE_SIZE; i++) {
391         if (l2_offset == s->l2_cache_offsets[i]) {
392             /* increment the hit count */
393             if (++s->l2_cache_counts[i] == 0xffffffff) {
394                 for(j = 0; j < L2_CACHE_SIZE; j++) {
395                     s->l2_cache_counts[j] >>= 1;
396                 }
397             }
398             l2_table = s->l2_cache + (i << s->l2_bits);
399             goto found;
400         }
401     }
402     /* not found: load a new entry in the least used one */
403     min_index = 0;
404     min_count = 0xffffffff;
405     for(i = 0; i < L2_CACHE_SIZE; i++) {
406         if (s->l2_cache_counts[i] < min_count) {
407             min_count = s->l2_cache_counts[i];
408             min_index = i;
409         }
410     }
411     l2_table = s->l2_cache + (min_index << s->l2_bits);
412     BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
413     if (new_l2_table) {
414         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
415         ret = bdrv_pwrite_sync(bs->file, l2_offset,
416                                s->l2_size * sizeof(uint64_t), l2_table, 0);
417         if (ret < 0) {
418             return ret;
419         }
420     } else {
421         ret = bdrv_pread(bs->file, l2_offset, s->l2_size * sizeof(uint64_t),
422                          l2_table, 0);
423         if (ret < 0) {
424             return ret;
425         }
426     }
427     s->l2_cache_offsets[min_index] = l2_offset;
428     s->l2_cache_counts[min_index] = 1;
429  found:
430     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
431     cluster_offset = be64_to_cpu(l2_table[l2_index]);
432     if (!cluster_offset ||
433         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
434         if (!allocate)
435             return 0;
436         BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
437         assert(QEMU_IS_ALIGNED(n_start | n_end, BDRV_SECTOR_SIZE));
438         /* allocate a new cluster */
439         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
440             (n_end - n_start) < s->cluster_size) {
441             /* if the cluster is already compressed, we must
442                decompress it in the case it is not completely
443                overwritten */
444             if (decompress_cluster(bs, cluster_offset) < 0) {
445                 return -EIO;
446             }
447             cluster_offset = bdrv_getlength(bs->file->bs);
448             if ((int64_t) cluster_offset < 0) {
449                 return cluster_offset;
450             }
451             cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
452             /* write the cluster content */
453             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
454             ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_size,
455                               s->cluster_cache, 0);
456             if (ret < 0) {
457                 return ret;
458             }
459         } else {
460             cluster_offset = bdrv_getlength(bs->file->bs);
461             if ((int64_t) cluster_offset < 0) {
462                 return cluster_offset;
463             }
464             if (allocate == 1) {
465                 /* round to cluster size */
466                 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
467                 if (cluster_offset + s->cluster_size > INT64_MAX) {
468                     return -E2BIG;
469                 }
470                 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size,
471                                     false, PREALLOC_MODE_OFF, 0, NULL);
472                 if (ret < 0) {
473                     return ret;
474                 }
475                 /* if encrypted, we must initialize the cluster
476                    content which won't be written */
477                 if (bs->encrypted &&
478                     (n_end - n_start) < s->cluster_size) {
479                     uint64_t start_offset;
480                     assert(s->crypto);
481                     start_offset = offset & ~(s->cluster_size - 1);
482                     for (i = 0; i < s->cluster_size; i += BDRV_SECTOR_SIZE) {
483                         if (i < n_start || i >= n_end) {
484                             memset(s->cluster_data, 0x00, BDRV_SECTOR_SIZE);
485                             if (qcrypto_block_encrypt(s->crypto,
486                                                       start_offset + i,
487                                                       s->cluster_data,
488                                                       BDRV_SECTOR_SIZE,
489                                                       NULL) < 0) {
490                                 return -EIO;
491                             }
492                             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
493                             ret = bdrv_pwrite(bs->file, cluster_offset + i,
494                                               BDRV_SECTOR_SIZE,
495                                               s->cluster_data, 0);
496                             if (ret < 0) {
497                                 return ret;
498                             }
499                         }
500                     }
501                 }
502             } else if (allocate == 2) {
503                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
504                     (uint64_t)compressed_size << (63 - s->cluster_bits);
505             }
506         }
507         /* update L2 table */
508         tmp = cpu_to_be64(cluster_offset);
509         l2_table[l2_index] = tmp;
510         if (allocate == 2) {
511             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
512         } else {
513             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
514         }
515         ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
516                                sizeof(tmp), &tmp, 0);
517         if (ret < 0) {
518             return ret;
519         }
520     }
521     *result = cluster_offset;
522     return 1;
523 }
524 
525 static int coroutine_fn qcow_co_block_status(BlockDriverState *bs,
526                                              bool want_zero,
527                                              int64_t offset, int64_t bytes,
528                                              int64_t *pnum, int64_t *map,
529                                              BlockDriverState **file)
530 {
531     BDRVQcowState *s = bs->opaque;
532     int index_in_cluster, ret;
533     int64_t n;
534     uint64_t cluster_offset;
535 
536     qemu_co_mutex_lock(&s->lock);
537     ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
538     qemu_co_mutex_unlock(&s->lock);
539     if (ret < 0) {
540         return ret;
541     }
542     index_in_cluster = offset & (s->cluster_size - 1);
543     n = s->cluster_size - index_in_cluster;
544     if (n > bytes) {
545         n = bytes;
546     }
547     *pnum = n;
548     if (!cluster_offset) {
549         return 0;
550     }
551     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) {
552         return BDRV_BLOCK_DATA;
553     }
554     *map = cluster_offset | index_in_cluster;
555     *file = bs->file->bs;
556     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
557 }
558 
559 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
560                              const uint8_t *buf, int buf_size)
561 {
562     z_stream strm1, *strm = &strm1;
563     int ret, out_len;
564 
565     memset(strm, 0, sizeof(*strm));
566 
567     strm->next_in = (uint8_t *)buf;
568     strm->avail_in = buf_size;
569     strm->next_out = out_buf;
570     strm->avail_out = out_buf_size;
571 
572     ret = inflateInit2(strm, -12);
573     if (ret != Z_OK)
574         return -1;
575     ret = inflate(strm, Z_FINISH);
576     out_len = strm->next_out - out_buf;
577     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
578         out_len != out_buf_size) {
579         inflateEnd(strm);
580         return -1;
581     }
582     inflateEnd(strm);
583     return 0;
584 }
585 
586 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
587 {
588     BDRVQcowState *s = bs->opaque;
589     int ret, csize;
590     uint64_t coffset;
591 
592     coffset = cluster_offset & s->cluster_offset_mask;
593     if (s->cluster_cache_offset != coffset) {
594         csize = cluster_offset >> (63 - s->cluster_bits);
595         csize &= (s->cluster_size - 1);
596         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
597         ret = bdrv_pread(bs->file, coffset, csize, s->cluster_data, 0);
598         if (ret < 0)
599             return -1;
600         if (decompress_buffer(s->cluster_cache, s->cluster_size,
601                               s->cluster_data, csize) < 0) {
602             return -1;
603         }
604         s->cluster_cache_offset = coffset;
605     }
606     return 0;
607 }
608 
609 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp)
610 {
611     /* At least encrypted images require 512-byte alignment. Apply the
612      * limit universally, rather than just on encrypted images, as
613      * it's easier to let the block layer handle rounding than to
614      * audit this code further. */
615     bs->bl.request_alignment = BDRV_SECTOR_SIZE;
616 }
617 
618 static coroutine_fn int qcow_co_preadv(BlockDriverState *bs, int64_t offset,
619                                        int64_t bytes, QEMUIOVector *qiov,
620                                        BdrvRequestFlags flags)
621 {
622     BDRVQcowState *s = bs->opaque;
623     int offset_in_cluster;
624     int ret = 0, n;
625     uint64_t cluster_offset;
626     uint8_t *buf;
627     void *orig_buf;
628 
629     assert(!flags);
630     if (qiov->niov > 1) {
631         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
632         if (buf == NULL) {
633             return -ENOMEM;
634         }
635     } else {
636         orig_buf = NULL;
637         buf = (uint8_t *)qiov->iov->iov_base;
638     }
639 
640     qemu_co_mutex_lock(&s->lock);
641 
642     while (bytes != 0) {
643         /* prepare next request */
644         ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
645         if (ret < 0) {
646             break;
647         }
648         offset_in_cluster = offset & (s->cluster_size - 1);
649         n = s->cluster_size - offset_in_cluster;
650         if (n > bytes) {
651             n = bytes;
652         }
653 
654         if (!cluster_offset) {
655             if (bs->backing) {
656                 /* read from the base image */
657                 qemu_co_mutex_unlock(&s->lock);
658                 /* qcow2 emits this on bs->file instead of bs->backing */
659                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
660                 ret = bdrv_co_pread(bs->backing, offset, n, buf, 0);
661                 qemu_co_mutex_lock(&s->lock);
662                 if (ret < 0) {
663                     break;
664                 }
665             } else {
666                 /* Note: in this case, no need to wait */
667                 memset(buf, 0, n);
668             }
669         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
670             /* add AIO support for compressed blocks ? */
671             if (decompress_cluster(bs, cluster_offset) < 0) {
672                 ret = -EIO;
673                 break;
674             }
675             memcpy(buf, s->cluster_cache + offset_in_cluster, n);
676         } else {
677             if ((cluster_offset & 511) != 0) {
678                 ret = -EIO;
679                 break;
680             }
681             qemu_co_mutex_unlock(&s->lock);
682             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
683             ret = bdrv_co_pread(bs->file, cluster_offset + offset_in_cluster,
684                                 n, buf, 0);
685             qemu_co_mutex_lock(&s->lock);
686             if (ret < 0) {
687                 break;
688             }
689             if (bs->encrypted) {
690                 assert(s->crypto);
691                 if (qcrypto_block_decrypt(s->crypto,
692                                           offset, buf, n, NULL) < 0) {
693                     ret = -EIO;
694                     break;
695                 }
696             }
697         }
698         ret = 0;
699 
700         bytes -= n;
701         offset += n;
702         buf += n;
703     }
704 
705     qemu_co_mutex_unlock(&s->lock);
706 
707     if (qiov->niov > 1) {
708         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
709         qemu_vfree(orig_buf);
710     }
711 
712     return ret;
713 }
714 
715 static coroutine_fn int qcow_co_pwritev(BlockDriverState *bs, int64_t offset,
716                                         int64_t bytes, QEMUIOVector *qiov,
717                                         BdrvRequestFlags flags)
718 {
719     BDRVQcowState *s = bs->opaque;
720     int offset_in_cluster;
721     uint64_t cluster_offset;
722     int ret = 0, n;
723     uint8_t *buf;
724     void *orig_buf;
725 
726     assert(!flags);
727     s->cluster_cache_offset = -1; /* disable compressed cache */
728 
729     /* We must always copy the iov when encrypting, so we
730      * don't modify the original data buffer during encryption */
731     if (bs->encrypted || qiov->niov > 1) {
732         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
733         if (buf == NULL) {
734             return -ENOMEM;
735         }
736         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
737     } else {
738         orig_buf = NULL;
739         buf = (uint8_t *)qiov->iov->iov_base;
740     }
741 
742     qemu_co_mutex_lock(&s->lock);
743 
744     while (bytes != 0) {
745         offset_in_cluster = offset & (s->cluster_size - 1);
746         n = s->cluster_size - offset_in_cluster;
747         if (n > bytes) {
748             n = bytes;
749         }
750         ret = get_cluster_offset(bs, offset, 1, 0, offset_in_cluster,
751                                  offset_in_cluster + n, &cluster_offset);
752         if (ret < 0) {
753             break;
754         }
755         if (!cluster_offset || (cluster_offset & 511) != 0) {
756             ret = -EIO;
757             break;
758         }
759         if (bs->encrypted) {
760             assert(s->crypto);
761             if (qcrypto_block_encrypt(s->crypto, offset, buf, n, NULL) < 0) {
762                 ret = -EIO;
763                 break;
764             }
765         }
766 
767         qemu_co_mutex_unlock(&s->lock);
768         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
769         ret = bdrv_co_pwrite(bs->file, cluster_offset + offset_in_cluster,
770                              n, buf, 0);
771         qemu_co_mutex_lock(&s->lock);
772         if (ret < 0) {
773             break;
774         }
775         ret = 0;
776 
777         bytes -= n;
778         offset += n;
779         buf += n;
780     }
781     qemu_co_mutex_unlock(&s->lock);
782 
783     qemu_vfree(orig_buf);
784 
785     return ret;
786 }
787 
788 static void qcow_close(BlockDriverState *bs)
789 {
790     BDRVQcowState *s = bs->opaque;
791 
792     qcrypto_block_free(s->crypto);
793     s->crypto = NULL;
794     g_free(s->l1_table);
795     qemu_vfree(s->l2_cache);
796     g_free(s->cluster_cache);
797     g_free(s->cluster_data);
798 
799     migrate_del_blocker(s->migration_blocker);
800     error_free(s->migration_blocker);
801 }
802 
803 static int coroutine_fn qcow_co_create(BlockdevCreateOptions *opts,
804                                        Error **errp)
805 {
806     BlockdevCreateOptionsQcow *qcow_opts;
807     int header_size, backing_filename_len, l1_size, shift, i;
808     QCowHeader header;
809     uint8_t *tmp;
810     int64_t total_size = 0;
811     int ret;
812     BlockDriverState *bs;
813     BlockBackend *qcow_blk;
814     QCryptoBlock *crypto = NULL;
815 
816     assert(opts->driver == BLOCKDEV_DRIVER_QCOW);
817     qcow_opts = &opts->u.qcow;
818 
819     /* Sanity checks */
820     total_size = qcow_opts->size;
821     if (total_size == 0) {
822         error_setg(errp, "Image size is too small, cannot be zero length");
823         return -EINVAL;
824     }
825 
826     if (qcow_opts->has_encrypt &&
827         qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW)
828     {
829         error_setg(errp, "Unsupported encryption format");
830         return -EINVAL;
831     }
832 
833     /* Create BlockBackend to write to the image */
834     bs = bdrv_open_blockdev_ref(qcow_opts->file, errp);
835     if (bs == NULL) {
836         return -EIO;
837     }
838 
839     qcow_blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE,
840                                BLK_PERM_ALL, errp);
841     if (!qcow_blk) {
842         ret = -EPERM;
843         goto exit;
844     }
845     blk_set_allow_write_beyond_eof(qcow_blk, true);
846 
847     /* Create image format */
848     memset(&header, 0, sizeof(header));
849     header.magic = cpu_to_be32(QCOW_MAGIC);
850     header.version = cpu_to_be32(QCOW_VERSION);
851     header.size = cpu_to_be64(total_size);
852     header_size = sizeof(header);
853     backing_filename_len = 0;
854     if (qcow_opts->has_backing_file) {
855         if (strcmp(qcow_opts->backing_file, "fat:")) {
856             header.backing_file_offset = cpu_to_be64(header_size);
857             backing_filename_len = strlen(qcow_opts->backing_file);
858             header.backing_file_size = cpu_to_be32(backing_filename_len);
859             header_size += backing_filename_len;
860         } else {
861             /* special backing file for vvfat */
862             qcow_opts->has_backing_file = false;
863         }
864         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
865                                     unmodified sectors */
866         header.l2_bits = 12; /* 32 KB L2 tables */
867     } else {
868         header.cluster_bits = 12; /* 4 KB clusters */
869         header.l2_bits = 9; /* 4 KB L2 tables */
870     }
871     header_size = (header_size + 7) & ~7;
872     shift = header.cluster_bits + header.l2_bits;
873     l1_size = (total_size + (1LL << shift) - 1) >> shift;
874 
875     header.l1_table_offset = cpu_to_be64(header_size);
876 
877     if (qcow_opts->has_encrypt) {
878         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
879 
880         crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.",
881                                       NULL, NULL, NULL, errp);
882         if (!crypto) {
883             ret = -EINVAL;
884             goto exit;
885         }
886     } else {
887         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
888     }
889 
890     /* write all the data */
891     ret = blk_pwrite(qcow_blk, 0, sizeof(header), &header, 0);
892     if (ret < 0) {
893         goto exit;
894     }
895 
896     if (qcow_opts->has_backing_file) {
897         ret = blk_pwrite(qcow_blk, sizeof(header), backing_filename_len,
898                          qcow_opts->backing_file, 0);
899         if (ret < 0) {
900             goto exit;
901         }
902     }
903 
904     tmp = g_malloc0(BDRV_SECTOR_SIZE);
905     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
906          i++) {
907         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
908                          BDRV_SECTOR_SIZE, tmp, 0);
909         if (ret < 0) {
910             g_free(tmp);
911             goto exit;
912         }
913     }
914 
915     g_free(tmp);
916     ret = 0;
917 exit:
918     blk_unref(qcow_blk);
919     bdrv_unref(bs);
920     qcrypto_block_free(crypto);
921     return ret;
922 }
923 
924 static int coroutine_fn qcow_co_create_opts(BlockDriver *drv,
925                                             const char *filename,
926                                             QemuOpts *opts, Error **errp)
927 {
928     BlockdevCreateOptions *create_options = NULL;
929     BlockDriverState *bs = NULL;
930     QDict *qdict = NULL;
931     Visitor *v;
932     const char *val;
933     int ret;
934     char *backing_fmt;
935 
936     static const QDictRenames opt_renames[] = {
937         { BLOCK_OPT_BACKING_FILE,       "backing-file" },
938         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
939         { NULL, NULL },
940     };
941 
942     /*
943      * We can't actually store a backing format, but can check that
944      * the user's request made sense.
945      */
946     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
947     if (backing_fmt && !bdrv_find_format(backing_fmt)) {
948         error_setg(errp, "unrecognized backing format '%s'", backing_fmt);
949         ret = -EINVAL;
950         goto fail;
951     }
952 
953     /* Parse options and convert legacy syntax */
954     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true);
955 
956     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
957     if (val && !strcmp(val, "on")) {
958         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
959     } else if (val && !strcmp(val, "off")) {
960         qdict_del(qdict, BLOCK_OPT_ENCRYPT);
961     }
962 
963     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
964     if (val && !strcmp(val, "aes")) {
965         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
966     }
967 
968     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
969         ret = -EINVAL;
970         goto fail;
971     }
972 
973     /* Create and open the file (protocol layer) */
974     ret = bdrv_create_file(filename, opts, errp);
975     if (ret < 0) {
976         goto fail;
977     }
978 
979     bs = bdrv_open(filename, NULL, NULL,
980                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
981     if (bs == NULL) {
982         ret = -EIO;
983         goto fail;
984     }
985 
986     /* Now get the QAPI type BlockdevCreateOptions */
987     qdict_put_str(qdict, "driver", "qcow");
988     qdict_put_str(qdict, "file", bs->node_name);
989 
990     v = qobject_input_visitor_new_flat_confused(qdict, errp);
991     if (!v) {
992         ret = -EINVAL;
993         goto fail;
994     }
995 
996     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
997     visit_free(v);
998     if (!create_options) {
999         ret = -EINVAL;
1000         goto fail;
1001     }
1002 
1003     /* Silently round up size */
1004     assert(create_options->driver == BLOCKDEV_DRIVER_QCOW);
1005     create_options->u.qcow.size =
1006         ROUND_UP(create_options->u.qcow.size, BDRV_SECTOR_SIZE);
1007 
1008     /* Create the qcow image (format layer) */
1009     ret = qcow_co_create(create_options, errp);
1010     if (ret < 0) {
1011         goto fail;
1012     }
1013 
1014     ret = 0;
1015 fail:
1016     g_free(backing_fmt);
1017     qobject_unref(qdict);
1018     bdrv_unref(bs);
1019     qapi_free_BlockdevCreateOptions(create_options);
1020     return ret;
1021 }
1022 
1023 static int qcow_make_empty(BlockDriverState *bs)
1024 {
1025     BDRVQcowState *s = bs->opaque;
1026     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1027     int ret;
1028 
1029     memset(s->l1_table, 0, l1_length);
1030     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, l1_length, s->l1_table,
1031                          0) < 0)
1032         return -1;
1033     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length, false,
1034                         PREALLOC_MODE_OFF, 0, NULL);
1035     if (ret < 0)
1036         return ret;
1037 
1038     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
1039     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
1040     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
1041 
1042     return 0;
1043 }
1044 
1045 /* XXX: put compressed sectors first, then all the cluster aligned
1046    tables to avoid losing bytes in alignment */
1047 static coroutine_fn int
1048 qcow_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes,
1049                            QEMUIOVector *qiov)
1050 {
1051     BDRVQcowState *s = bs->opaque;
1052     z_stream strm;
1053     int ret, out_len;
1054     uint8_t *buf, *out_buf;
1055     uint64_t cluster_offset;
1056 
1057     buf = qemu_blockalign(bs, s->cluster_size);
1058     if (bytes != s->cluster_size) {
1059         if (bytes > s->cluster_size ||
1060             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
1061         {
1062             qemu_vfree(buf);
1063             return -EINVAL;
1064         }
1065         /* Zero-pad last write if image size is not cluster aligned */
1066         memset(buf + bytes, 0, s->cluster_size - bytes);
1067     }
1068     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
1069 
1070     out_buf = g_malloc(s->cluster_size);
1071 
1072     /* best compression, small window, no zlib header */
1073     memset(&strm, 0, sizeof(strm));
1074     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1075                        Z_DEFLATED, -12,
1076                        9, Z_DEFAULT_STRATEGY);
1077     if (ret != 0) {
1078         ret = -EINVAL;
1079         goto fail;
1080     }
1081 
1082     strm.avail_in = s->cluster_size;
1083     strm.next_in = (uint8_t *)buf;
1084     strm.avail_out = s->cluster_size;
1085     strm.next_out = out_buf;
1086 
1087     ret = deflate(&strm, Z_FINISH);
1088     if (ret != Z_STREAM_END && ret != Z_OK) {
1089         deflateEnd(&strm);
1090         ret = -EINVAL;
1091         goto fail;
1092     }
1093     out_len = strm.next_out - out_buf;
1094 
1095     deflateEnd(&strm);
1096 
1097     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1098         /* could not compress: write normal cluster */
1099         ret = qcow_co_pwritev(bs, offset, bytes, qiov, 0);
1100         if (ret < 0) {
1101             goto fail;
1102         }
1103         goto success;
1104     }
1105     qemu_co_mutex_lock(&s->lock);
1106     ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1107     qemu_co_mutex_unlock(&s->lock);
1108     if (ret < 0) {
1109         goto fail;
1110     }
1111     if (cluster_offset == 0) {
1112         ret = -EIO;
1113         goto fail;
1114     }
1115     cluster_offset &= s->cluster_offset_mask;
1116 
1117     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1118     ret = bdrv_co_pwrite(bs->file, cluster_offset, out_len, out_buf, 0);
1119     if (ret < 0) {
1120         goto fail;
1121     }
1122 success:
1123     ret = 0;
1124 fail:
1125     qemu_vfree(buf);
1126     g_free(out_buf);
1127     return ret;
1128 }
1129 
1130 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1131 {
1132     BDRVQcowState *s = bs->opaque;
1133     bdi->cluster_size = s->cluster_size;
1134     return 0;
1135 }
1136 
1137 static QemuOptsList qcow_create_opts = {
1138     .name = "qcow-create-opts",
1139     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1140     .desc = {
1141         {
1142             .name = BLOCK_OPT_SIZE,
1143             .type = QEMU_OPT_SIZE,
1144             .help = "Virtual disk size"
1145         },
1146         {
1147             .name = BLOCK_OPT_BACKING_FILE,
1148             .type = QEMU_OPT_STRING,
1149             .help = "File name of a base image"
1150         },
1151         {
1152             .name = BLOCK_OPT_BACKING_FMT,
1153             .type = QEMU_OPT_STRING,
1154             .help = "Format of the backing image",
1155         },
1156         {
1157             .name = BLOCK_OPT_ENCRYPT,
1158             .type = QEMU_OPT_BOOL,
1159             .help = "Encrypt the image with format 'aes'. (Deprecated "
1160                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1161         },
1162         {
1163             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1164             .type = QEMU_OPT_STRING,
1165             .help = "Encrypt the image, format choices: 'aes'",
1166         },
1167         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1168         { /* end of list */ }
1169     }
1170 };
1171 
1172 static const char *const qcow_strong_runtime_opts[] = {
1173     "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
1174 
1175     NULL
1176 };
1177 
1178 static BlockDriver bdrv_qcow = {
1179     .format_name	= "qcow",
1180     .instance_size	= sizeof(BDRVQcowState),
1181     .bdrv_probe		= qcow_probe,
1182     .bdrv_open		= qcow_open,
1183     .bdrv_close		= qcow_close,
1184     .bdrv_child_perm        = bdrv_default_perms,
1185     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1186     .bdrv_co_create         = qcow_co_create,
1187     .bdrv_co_create_opts    = qcow_co_create_opts,
1188     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1189     .is_format              = true,
1190     .supports_backing       = true,
1191     .bdrv_refresh_limits    = qcow_refresh_limits,
1192 
1193     .bdrv_co_preadv         = qcow_co_preadv,
1194     .bdrv_co_pwritev        = qcow_co_pwritev,
1195     .bdrv_co_block_status   = qcow_co_block_status,
1196 
1197     .bdrv_make_empty        = qcow_make_empty,
1198     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1199     .bdrv_get_info          = qcow_get_info,
1200 
1201     .create_opts            = &qcow_create_opts,
1202     .strong_runtime_opts    = qcow_strong_runtime_opts,
1203 };
1204 
1205 static void bdrv_qcow_init(void)
1206 {
1207     bdrv_register(&bdrv_qcow);
1208 }
1209 
1210 block_init(bdrv_qcow_init);
1211