xref: /openbmc/qemu/block/qcow.c (revision e3a99063)
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     Error *local_err = NULL;
125     QCryptoBlockOpenOptions *crypto_opts = NULL;
126     unsigned int cflags = 0;
127     QDict *encryptopts = NULL;
128     const char *encryptfmt;
129 
130     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
131     encryptfmt = qdict_get_try_str(encryptopts, "format");
132 
133     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
134                                false, errp);
135     if (!bs->file) {
136         ret = -EINVAL;
137         goto fail;
138     }
139 
140     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
141     if (ret < 0) {
142         goto fail;
143     }
144     header.magic = be32_to_cpu(header.magic);
145     header.version = be32_to_cpu(header.version);
146     header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
147     header.backing_file_size = be32_to_cpu(header.backing_file_size);
148     header.mtime = be32_to_cpu(header.mtime);
149     header.size = be64_to_cpu(header.size);
150     header.crypt_method = be32_to_cpu(header.crypt_method);
151     header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
152 
153     if (header.magic != QCOW_MAGIC) {
154         error_setg(errp, "Image not in qcow format");
155         ret = -EINVAL;
156         goto fail;
157     }
158     if (header.version != QCOW_VERSION) {
159         error_setg(errp, "qcow (v%d) does not support qcow version %" PRIu32,
160                    QCOW_VERSION, header.version);
161         if (header.version == 2 || header.version == 3) {
162             error_append_hint(errp, "Try the 'qcow2' driver instead.\n");
163         }
164 
165         ret = -ENOTSUP;
166         goto fail;
167     }
168 
169     if (header.size <= 1) {
170         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
171         ret = -EINVAL;
172         goto fail;
173     }
174     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
175         error_setg(errp, "Cluster size must be between 512 and 64k");
176         ret = -EINVAL;
177         goto fail;
178     }
179 
180     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
181      * so bytes = num_entries << 3. */
182     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
183         error_setg(errp, "L2 table size must be between 512 and 64k");
184         ret = -EINVAL;
185         goto fail;
186     }
187 
188     s->crypt_method_header = header.crypt_method;
189     if (s->crypt_method_header) {
190         if (bdrv_uses_whitelist() &&
191             s->crypt_method_header == QCOW_CRYPT_AES) {
192             error_setg(errp,
193                        "Use of AES-CBC encrypted qcow images is no longer "
194                        "supported in system emulators");
195             error_append_hint(errp,
196                               "You can use 'qemu-img convert' to convert your "
197                               "image to an alternative supported format, such "
198                               "as unencrypted qcow, or raw with the LUKS "
199                               "format instead.\n");
200             ret = -ENOSYS;
201             goto fail;
202         }
203         if (s->crypt_method_header == QCOW_CRYPT_AES) {
204             if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
205                 error_setg(errp,
206                            "Header reported 'aes' encryption format but "
207                            "options specify '%s'", encryptfmt);
208                 ret = -EINVAL;
209                 goto fail;
210             }
211             qdict_put_str(encryptopts, "format", "qcow");
212             crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
213             if (!crypto_opts) {
214                 ret = -EINVAL;
215                 goto fail;
216             }
217 
218             if (flags & BDRV_O_NO_IO) {
219                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
220             }
221             s->crypto = qcrypto_block_open(crypto_opts, "encrypt.",
222                                            NULL, NULL, cflags, 1, errp);
223             if (!s->crypto) {
224                 ret = -EINVAL;
225                 goto fail;
226             }
227         } else {
228             error_setg(errp, "invalid encryption method in qcow header");
229             ret = -EINVAL;
230             goto fail;
231         }
232         bs->encrypted = true;
233     } else {
234         if (encryptfmt) {
235             error_setg(errp, "No encryption in image header, but options "
236                        "specified format '%s'", encryptfmt);
237             ret = -EINVAL;
238             goto fail;
239         }
240     }
241     s->cluster_bits = header.cluster_bits;
242     s->cluster_size = 1 << s->cluster_bits;
243     s->l2_bits = header.l2_bits;
244     s->l2_size = 1 << s->l2_bits;
245     bs->total_sectors = header.size / 512;
246     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
247 
248     /* read the level 1 table */
249     shift = s->cluster_bits + s->l2_bits;
250     if (header.size > UINT64_MAX - (1LL << shift)) {
251         error_setg(errp, "Image too large");
252         ret = -EINVAL;
253         goto fail;
254     } else {
255         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
256         if (l1_size > INT_MAX / sizeof(uint64_t)) {
257             error_setg(errp, "Image too large");
258             ret = -EINVAL;
259             goto fail;
260         }
261         s->l1_size = l1_size;
262     }
263 
264     s->l1_table_offset = header.l1_table_offset;
265     s->l1_table = g_try_new(uint64_t, s->l1_size);
266     if (s->l1_table == NULL) {
267         error_setg(errp, "Could not allocate memory for L1 table");
268         ret = -ENOMEM;
269         goto fail;
270     }
271 
272     ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
273                s->l1_size * sizeof(uint64_t));
274     if (ret < 0) {
275         goto fail;
276     }
277 
278     for(i = 0;i < s->l1_size; i++) {
279         s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
280     }
281 
282     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
283     s->l2_cache =
284         qemu_try_blockalign(bs->file->bs,
285                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
286     if (s->l2_cache == NULL) {
287         error_setg(errp, "Could not allocate L2 table cache");
288         ret = -ENOMEM;
289         goto fail;
290     }
291     s->cluster_cache = g_malloc(s->cluster_size);
292     s->cluster_data = g_malloc(s->cluster_size);
293     s->cluster_cache_offset = -1;
294 
295     /* read the backing file name */
296     if (header.backing_file_offset != 0) {
297         len = header.backing_file_size;
298         if (len > 1023 || len >= sizeof(bs->backing_file)) {
299             error_setg(errp, "Backing file name too long");
300             ret = -EINVAL;
301             goto fail;
302         }
303         ret = bdrv_pread(bs->file, header.backing_file_offset,
304                    bs->auto_backing_file, len);
305         if (ret < 0) {
306             goto fail;
307         }
308         bs->auto_backing_file[len] = '\0';
309         pstrcpy(bs->backing_file, sizeof(bs->backing_file),
310                 bs->auto_backing_file);
311     }
312 
313     /* Disable migration when qcow images are used */
314     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
315                "does not support live migration",
316                bdrv_get_device_or_node_name(bs));
317     ret = migrate_add_blocker(s->migration_blocker, &local_err);
318     if (local_err) {
319         error_propagate(errp, local_err);
320         error_free(s->migration_blocker);
321         goto fail;
322     }
323 
324     qobject_unref(encryptopts);
325     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
326     qemu_co_mutex_init(&s->lock);
327     return 0;
328 
329  fail:
330     g_free(s->l1_table);
331     qemu_vfree(s->l2_cache);
332     g_free(s->cluster_cache);
333     g_free(s->cluster_data);
334     qcrypto_block_free(s->crypto);
335     qobject_unref(encryptopts);
336     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
337     return ret;
338 }
339 
340 
341 /* We have nothing to do for QCOW reopen, stubs just return
342  * success */
343 static int qcow_reopen_prepare(BDRVReopenState *state,
344                                BlockReopenQueue *queue, Error **errp)
345 {
346     return 0;
347 }
348 
349 
350 /* 'allocate' is:
351  *
352  * 0 to not allocate.
353  *
354  * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start'
355  * to 'n_end' within the cluster)
356  *
357  * 2 to allocate a compressed cluster of size
358  * 'compressed_size'. 'compressed_size' must be > 0 and <
359  * cluster_size
360  *
361  * return 0 if not allocated, 1 if *result is assigned, and negative
362  * errno on failure.
363  */
364 static int get_cluster_offset(BlockDriverState *bs,
365                               uint64_t offset, int allocate,
366                               int compressed_size,
367                               int n_start, int n_end, uint64_t *result)
368 {
369     BDRVQcowState *s = bs->opaque;
370     int min_index, i, j, l1_index, l2_index, ret;
371     int64_t l2_offset;
372     uint64_t *l2_table, cluster_offset, tmp;
373     uint32_t min_count;
374     int new_l2_table;
375 
376     *result = 0;
377     l1_index = offset >> (s->l2_bits + s->cluster_bits);
378     l2_offset = s->l1_table[l1_index];
379     new_l2_table = 0;
380     if (!l2_offset) {
381         if (!allocate)
382             return 0;
383         /* allocate a new l2 entry */
384         l2_offset = bdrv_getlength(bs->file->bs);
385         if (l2_offset < 0) {
386             return l2_offset;
387         }
388         /* round to cluster size */
389         l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size);
390         /* update the L1 entry */
391         s->l1_table[l1_index] = l2_offset;
392         tmp = cpu_to_be64(l2_offset);
393         BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
394         ret = bdrv_pwrite_sync(bs->file,
395                                s->l1_table_offset + l1_index * sizeof(tmp),
396                                &tmp, sizeof(tmp));
397         if (ret < 0) {
398             return ret;
399         }
400         new_l2_table = 1;
401     }
402     for(i = 0; i < L2_CACHE_SIZE; i++) {
403         if (l2_offset == s->l2_cache_offsets[i]) {
404             /* increment the hit count */
405             if (++s->l2_cache_counts[i] == 0xffffffff) {
406                 for(j = 0; j < L2_CACHE_SIZE; j++) {
407                     s->l2_cache_counts[j] >>= 1;
408                 }
409             }
410             l2_table = s->l2_cache + (i << s->l2_bits);
411             goto found;
412         }
413     }
414     /* not found: load a new entry in the least used one */
415     min_index = 0;
416     min_count = 0xffffffff;
417     for(i = 0; i < L2_CACHE_SIZE; i++) {
418         if (s->l2_cache_counts[i] < min_count) {
419             min_count = s->l2_cache_counts[i];
420             min_index = i;
421         }
422     }
423     l2_table = s->l2_cache + (min_index << s->l2_bits);
424     BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
425     if (new_l2_table) {
426         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
427         ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
428                                s->l2_size * sizeof(uint64_t));
429         if (ret < 0) {
430             return ret;
431         }
432     } else {
433         ret = bdrv_pread(bs->file, l2_offset, l2_table,
434                          s->l2_size * sizeof(uint64_t));
435         if (ret < 0) {
436             return ret;
437         }
438     }
439     s->l2_cache_offsets[min_index] = l2_offset;
440     s->l2_cache_counts[min_index] = 1;
441  found:
442     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
443     cluster_offset = be64_to_cpu(l2_table[l2_index]);
444     if (!cluster_offset ||
445         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
446         if (!allocate)
447             return 0;
448         BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
449         assert(QEMU_IS_ALIGNED(n_start | n_end, BDRV_SECTOR_SIZE));
450         /* allocate a new cluster */
451         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
452             (n_end - n_start) < s->cluster_size) {
453             /* if the cluster is already compressed, we must
454                decompress it in the case it is not completely
455                overwritten */
456             if (decompress_cluster(bs, cluster_offset) < 0) {
457                 return -EIO;
458             }
459             cluster_offset = bdrv_getlength(bs->file->bs);
460             if ((int64_t) cluster_offset < 0) {
461                 return cluster_offset;
462             }
463             cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
464             /* write the cluster content */
465             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
466             ret = bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
467                               s->cluster_size);
468             if (ret < 0) {
469                 return ret;
470             }
471         } else {
472             cluster_offset = bdrv_getlength(bs->file->bs);
473             if ((int64_t) cluster_offset < 0) {
474                 return cluster_offset;
475             }
476             if (allocate == 1) {
477                 /* round to cluster size */
478                 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
479                 if (cluster_offset + s->cluster_size > INT64_MAX) {
480                     return -E2BIG;
481                 }
482                 ret = bdrv_truncate(bs->file, cluster_offset + s->cluster_size,
483                                     false, PREALLOC_MODE_OFF, 0, NULL);
484                 if (ret < 0) {
485                     return ret;
486                 }
487                 /* if encrypted, we must initialize the cluster
488                    content which won't be written */
489                 if (bs->encrypted &&
490                     (n_end - n_start) < s->cluster_size) {
491                     uint64_t start_offset;
492                     assert(s->crypto);
493                     start_offset = offset & ~(s->cluster_size - 1);
494                     for (i = 0; i < s->cluster_size; i += BDRV_SECTOR_SIZE) {
495                         if (i < n_start || i >= n_end) {
496                             memset(s->cluster_data, 0x00, BDRV_SECTOR_SIZE);
497                             if (qcrypto_block_encrypt(s->crypto,
498                                                       start_offset + i,
499                                                       s->cluster_data,
500                                                       BDRV_SECTOR_SIZE,
501                                                       NULL) < 0) {
502                                 return -EIO;
503                             }
504                             BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
505                             ret = bdrv_pwrite(bs->file,
506                                               cluster_offset + i,
507                                               s->cluster_data,
508                                               BDRV_SECTOR_SIZE);
509                             if (ret < 0) {
510                                 return ret;
511                             }
512                         }
513                     }
514                 }
515             } else if (allocate == 2) {
516                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
517                     (uint64_t)compressed_size << (63 - s->cluster_bits);
518             }
519         }
520         /* update L2 table */
521         tmp = cpu_to_be64(cluster_offset);
522         l2_table[l2_index] = tmp;
523         if (allocate == 2) {
524             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
525         } else {
526             BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
527         }
528         ret = bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
529                                &tmp, sizeof(tmp));
530         if (ret < 0) {
531             return ret;
532         }
533     }
534     *result = cluster_offset;
535     return 1;
536 }
537 
538 static int coroutine_fn qcow_co_block_status(BlockDriverState *bs,
539                                              bool want_zero,
540                                              int64_t offset, int64_t bytes,
541                                              int64_t *pnum, int64_t *map,
542                                              BlockDriverState **file)
543 {
544     BDRVQcowState *s = bs->opaque;
545     int index_in_cluster, ret;
546     int64_t n;
547     uint64_t cluster_offset;
548 
549     qemu_co_mutex_lock(&s->lock);
550     ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
551     qemu_co_mutex_unlock(&s->lock);
552     if (ret < 0) {
553         return ret;
554     }
555     index_in_cluster = offset & (s->cluster_size - 1);
556     n = s->cluster_size - index_in_cluster;
557     if (n > bytes) {
558         n = bytes;
559     }
560     *pnum = n;
561     if (!cluster_offset) {
562         return 0;
563     }
564     if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->crypto) {
565         return BDRV_BLOCK_DATA;
566     }
567     *map = cluster_offset | index_in_cluster;
568     *file = bs->file->bs;
569     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
570 }
571 
572 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
573                              const uint8_t *buf, int buf_size)
574 {
575     z_stream strm1, *strm = &strm1;
576     int ret, out_len;
577 
578     memset(strm, 0, sizeof(*strm));
579 
580     strm->next_in = (uint8_t *)buf;
581     strm->avail_in = buf_size;
582     strm->next_out = out_buf;
583     strm->avail_out = out_buf_size;
584 
585     ret = inflateInit2(strm, -12);
586     if (ret != Z_OK)
587         return -1;
588     ret = inflate(strm, Z_FINISH);
589     out_len = strm->next_out - out_buf;
590     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
591         out_len != out_buf_size) {
592         inflateEnd(strm);
593         return -1;
594     }
595     inflateEnd(strm);
596     return 0;
597 }
598 
599 static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
600 {
601     BDRVQcowState *s = bs->opaque;
602     int ret, csize;
603     uint64_t coffset;
604 
605     coffset = cluster_offset & s->cluster_offset_mask;
606     if (s->cluster_cache_offset != coffset) {
607         csize = cluster_offset >> (63 - s->cluster_bits);
608         csize &= (s->cluster_size - 1);
609         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
610         ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
611         if (ret != csize)
612             return -1;
613         if (decompress_buffer(s->cluster_cache, s->cluster_size,
614                               s->cluster_data, csize) < 0) {
615             return -1;
616         }
617         s->cluster_cache_offset = coffset;
618     }
619     return 0;
620 }
621 
622 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp)
623 {
624     /* At least encrypted images require 512-byte alignment. Apply the
625      * limit universally, rather than just on encrypted images, as
626      * it's easier to let the block layer handle rounding than to
627      * audit this code further. */
628     bs->bl.request_alignment = BDRV_SECTOR_SIZE;
629 }
630 
631 static coroutine_fn int qcow_co_preadv(BlockDriverState *bs, uint64_t offset,
632                                        uint64_t bytes, QEMUIOVector *qiov,
633                                        int flags)
634 {
635     BDRVQcowState *s = bs->opaque;
636     int offset_in_cluster;
637     int ret = 0, n;
638     uint64_t cluster_offset;
639     uint8_t *buf;
640     void *orig_buf;
641 
642     assert(!flags);
643     if (qiov->niov > 1) {
644         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
645         if (buf == NULL) {
646             return -ENOMEM;
647         }
648     } else {
649         orig_buf = NULL;
650         buf = (uint8_t *)qiov->iov->iov_base;
651     }
652 
653     qemu_co_mutex_lock(&s->lock);
654 
655     while (bytes != 0) {
656         /* prepare next request */
657         ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
658         if (ret < 0) {
659             break;
660         }
661         offset_in_cluster = offset & (s->cluster_size - 1);
662         n = s->cluster_size - offset_in_cluster;
663         if (n > bytes) {
664             n = bytes;
665         }
666 
667         if (!cluster_offset) {
668             if (bs->backing) {
669                 /* read from the base image */
670                 qemu_co_mutex_unlock(&s->lock);
671                 /* qcow2 emits this on bs->file instead of bs->backing */
672                 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
673                 ret = bdrv_co_pread(bs->backing, offset, n, buf, 0);
674                 qemu_co_mutex_lock(&s->lock);
675                 if (ret < 0) {
676                     break;
677                 }
678             } else {
679                 /* Note: in this case, no need to wait */
680                 memset(buf, 0, n);
681             }
682         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
683             /* add AIO support for compressed blocks ? */
684             if (decompress_cluster(bs, cluster_offset) < 0) {
685                 ret = -EIO;
686                 break;
687             }
688             memcpy(buf, s->cluster_cache + offset_in_cluster, n);
689         } else {
690             if ((cluster_offset & 511) != 0) {
691                 ret = -EIO;
692                 break;
693             }
694             qemu_co_mutex_unlock(&s->lock);
695             BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
696             ret = bdrv_co_pread(bs->file, cluster_offset + offset_in_cluster,
697                                 n, buf, 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     uint8_t *buf;
737     void *orig_buf;
738 
739     assert(!flags);
740     s->cluster_cache_offset = -1; /* disable compressed cache */
741 
742     /* We must always copy the iov when encrypting, so we
743      * don't modify the original data buffer during encryption */
744     if (bs->encrypted || qiov->niov > 1) {
745         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
746         if (buf == NULL) {
747             return -ENOMEM;
748         }
749         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
750     } else {
751         orig_buf = NULL;
752         buf = (uint8_t *)qiov->iov->iov_base;
753     }
754 
755     qemu_co_mutex_lock(&s->lock);
756 
757     while (bytes != 0) {
758         offset_in_cluster = offset & (s->cluster_size - 1);
759         n = s->cluster_size - offset_in_cluster;
760         if (n > bytes) {
761             n = bytes;
762         }
763         ret = get_cluster_offset(bs, offset, 1, 0, offset_in_cluster,
764                                  offset_in_cluster + n, &cluster_offset);
765         if (ret < 0) {
766             break;
767         }
768         if (!cluster_offset || (cluster_offset & 511) != 0) {
769             ret = -EIO;
770             break;
771         }
772         if (bs->encrypted) {
773             assert(s->crypto);
774             if (qcrypto_block_encrypt(s->crypto, offset, buf, n, NULL) < 0) {
775                 ret = -EIO;
776                 break;
777             }
778         }
779 
780         qemu_co_mutex_unlock(&s->lock);
781         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
782         ret = bdrv_co_pwrite(bs->file, cluster_offset + offset_in_cluster,
783                              n, buf, 0);
784         qemu_co_mutex_lock(&s->lock);
785         if (ret < 0) {
786             break;
787         }
788         ret = 0;
789 
790         bytes -= n;
791         offset += n;
792         buf += n;
793     }
794     qemu_co_mutex_unlock(&s->lock);
795 
796     qemu_vfree(orig_buf);
797 
798     return ret;
799 }
800 
801 static void qcow_close(BlockDriverState *bs)
802 {
803     BDRVQcowState *s = bs->opaque;
804 
805     qcrypto_block_free(s->crypto);
806     s->crypto = NULL;
807     g_free(s->l1_table);
808     qemu_vfree(s->l2_cache);
809     g_free(s->cluster_cache);
810     g_free(s->cluster_data);
811 
812     migrate_del_blocker(s->migration_blocker);
813     error_free(s->migration_blocker);
814 }
815 
816 static int coroutine_fn qcow_co_create(BlockdevCreateOptions *opts,
817                                        Error **errp)
818 {
819     BlockdevCreateOptionsQcow *qcow_opts;
820     int header_size, backing_filename_len, l1_size, shift, i;
821     QCowHeader header;
822     uint8_t *tmp;
823     int64_t total_size = 0;
824     int ret;
825     BlockDriverState *bs;
826     BlockBackend *qcow_blk;
827     QCryptoBlock *crypto = NULL;
828 
829     assert(opts->driver == BLOCKDEV_DRIVER_QCOW);
830     qcow_opts = &opts->u.qcow;
831 
832     /* Sanity checks */
833     total_size = qcow_opts->size;
834     if (total_size == 0) {
835         error_setg(errp, "Image size is too small, cannot be zero length");
836         return -EINVAL;
837     }
838 
839     if (qcow_opts->has_encrypt &&
840         qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW)
841     {
842         error_setg(errp, "Unsupported encryption format");
843         return -EINVAL;
844     }
845 
846     /* Create BlockBackend to write to the image */
847     bs = bdrv_open_blockdev_ref(qcow_opts->file, errp);
848     if (bs == NULL) {
849         return -EIO;
850     }
851 
852     qcow_blk = blk_new(bdrv_get_aio_context(bs),
853                        BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
854     ret = blk_insert_bs(qcow_blk, bs, errp);
855     if (ret < 0) {
856         goto exit;
857     }
858     blk_set_allow_write_beyond_eof(qcow_blk, true);
859 
860     /* Create image format */
861     memset(&header, 0, sizeof(header));
862     header.magic = cpu_to_be32(QCOW_MAGIC);
863     header.version = cpu_to_be32(QCOW_VERSION);
864     header.size = cpu_to_be64(total_size);
865     header_size = sizeof(header);
866     backing_filename_len = 0;
867     if (qcow_opts->has_backing_file) {
868         if (strcmp(qcow_opts->backing_file, "fat:")) {
869             header.backing_file_offset = cpu_to_be64(header_size);
870             backing_filename_len = strlen(qcow_opts->backing_file);
871             header.backing_file_size = cpu_to_be32(backing_filename_len);
872             header_size += backing_filename_len;
873         } else {
874             /* special backing file for vvfat */
875             qcow_opts->has_backing_file = false;
876         }
877         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
878                                     unmodified sectors */
879         header.l2_bits = 12; /* 32 KB L2 tables */
880     } else {
881         header.cluster_bits = 12; /* 4 KB clusters */
882         header.l2_bits = 9; /* 4 KB L2 tables */
883     }
884     header_size = (header_size + 7) & ~7;
885     shift = header.cluster_bits + header.l2_bits;
886     l1_size = (total_size + (1LL << shift) - 1) >> shift;
887 
888     header.l1_table_offset = cpu_to_be64(header_size);
889 
890     if (qcow_opts->has_encrypt) {
891         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
892 
893         crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.",
894                                       NULL, NULL, NULL, errp);
895         if (!crypto) {
896             ret = -EINVAL;
897             goto exit;
898         }
899     } else {
900         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
901     }
902 
903     /* write all the data */
904     ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
905     if (ret != sizeof(header)) {
906         goto exit;
907     }
908 
909     if (qcow_opts->has_backing_file) {
910         ret = blk_pwrite(qcow_blk, sizeof(header),
911                          qcow_opts->backing_file, backing_filename_len, 0);
912         if (ret != backing_filename_len) {
913             goto exit;
914         }
915     }
916 
917     tmp = g_malloc0(BDRV_SECTOR_SIZE);
918     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
919          i++) {
920         ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
921                          tmp, BDRV_SECTOR_SIZE, 0);
922         if (ret != BDRV_SECTOR_SIZE) {
923             g_free(tmp);
924             goto exit;
925         }
926     }
927 
928     g_free(tmp);
929     ret = 0;
930 exit:
931     blk_unref(qcow_blk);
932     bdrv_unref(bs);
933     qcrypto_block_free(crypto);
934     return ret;
935 }
936 
937 static int coroutine_fn qcow_co_create_opts(BlockDriver *drv,
938                                             const char *filename,
939                                             QemuOpts *opts, Error **errp)
940 {
941     BlockdevCreateOptions *create_options = NULL;
942     BlockDriverState *bs = NULL;
943     QDict *qdict;
944     Visitor *v;
945     const char *val;
946     Error *local_err = NULL;
947     int ret;
948 
949     static const QDictRenames opt_renames[] = {
950         { BLOCK_OPT_BACKING_FILE,       "backing-file" },
951         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
952         { NULL, NULL },
953     };
954 
955     /* Parse options and convert legacy syntax */
956     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true);
957 
958     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
959     if (val && !strcmp(val, "on")) {
960         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
961     } else if (val && !strcmp(val, "off")) {
962         qdict_del(qdict, BLOCK_OPT_ENCRYPT);
963     }
964 
965     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
966     if (val && !strcmp(val, "aes")) {
967         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
968     }
969 
970     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
971         ret = -EINVAL;
972         goto fail;
973     }
974 
975     /* Create and open the file (protocol layer) */
976     ret = bdrv_create_file(filename, opts, &local_err);
977     if (ret < 0) {
978         error_propagate(errp, local_err);
979         goto fail;
980     }
981 
982     bs = bdrv_open(filename, NULL, NULL,
983                    BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
984     if (bs == NULL) {
985         ret = -EIO;
986         goto fail;
987     }
988 
989     /* Now get the QAPI type BlockdevCreateOptions */
990     qdict_put_str(qdict, "driver", "qcow");
991     qdict_put_str(qdict, "file", bs->node_name);
992 
993     v = qobject_input_visitor_new_flat_confused(qdict, errp);
994     if (!v) {
995         ret = -EINVAL;
996         goto fail;
997     }
998 
999     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
1000     visit_free(v);
1001 
1002     if (local_err) {
1003         error_propagate(errp, local_err);
1004         ret = -EINVAL;
1005         goto fail;
1006     }
1007 
1008     /* Silently round up size */
1009     assert(create_options->driver == BLOCKDEV_DRIVER_QCOW);
1010     create_options->u.qcow.size =
1011         ROUND_UP(create_options->u.qcow.size, BDRV_SECTOR_SIZE);
1012 
1013     /* Create the qcow image (format layer) */
1014     ret = qcow_co_create(create_options, errp);
1015     if (ret < 0) {
1016         goto fail;
1017     }
1018 
1019     ret = 0;
1020 fail:
1021     qobject_unref(qdict);
1022     bdrv_unref(bs);
1023     qapi_free_BlockdevCreateOptions(create_options);
1024     return ret;
1025 }
1026 
1027 static int qcow_make_empty(BlockDriverState *bs)
1028 {
1029     BDRVQcowState *s = bs->opaque;
1030     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1031     int ret;
1032 
1033     memset(s->l1_table, 0, l1_length);
1034     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
1035             l1_length) < 0)
1036         return -1;
1037     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length, false,
1038                         PREALLOC_MODE_OFF, 0, NULL);
1039     if (ret < 0)
1040         return ret;
1041 
1042     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
1043     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
1044     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
1045 
1046     return 0;
1047 }
1048 
1049 /* XXX: put compressed sectors first, then all the cluster aligned
1050    tables to avoid losing bytes in alignment */
1051 static coroutine_fn int
1052 qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1053                            uint64_t bytes, QEMUIOVector *qiov)
1054 {
1055     BDRVQcowState *s = bs->opaque;
1056     z_stream strm;
1057     int ret, out_len;
1058     uint8_t *buf, *out_buf;
1059     uint64_t cluster_offset;
1060 
1061     buf = qemu_blockalign(bs, s->cluster_size);
1062     if (bytes != s->cluster_size) {
1063         if (bytes > s->cluster_size ||
1064             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
1065         {
1066             qemu_vfree(buf);
1067             return -EINVAL;
1068         }
1069         /* Zero-pad last write if image size is not cluster aligned */
1070         memset(buf + bytes, 0, s->cluster_size - bytes);
1071     }
1072     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
1073 
1074     out_buf = g_malloc(s->cluster_size);
1075 
1076     /* best compression, small window, no zlib header */
1077     memset(&strm, 0, sizeof(strm));
1078     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1079                        Z_DEFLATED, -12,
1080                        9, Z_DEFAULT_STRATEGY);
1081     if (ret != 0) {
1082         ret = -EINVAL;
1083         goto fail;
1084     }
1085 
1086     strm.avail_in = s->cluster_size;
1087     strm.next_in = (uint8_t *)buf;
1088     strm.avail_out = s->cluster_size;
1089     strm.next_out = out_buf;
1090 
1091     ret = deflate(&strm, Z_FINISH);
1092     if (ret != Z_STREAM_END && ret != Z_OK) {
1093         deflateEnd(&strm);
1094         ret = -EINVAL;
1095         goto fail;
1096     }
1097     out_len = strm.next_out - out_buf;
1098 
1099     deflateEnd(&strm);
1100 
1101     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1102         /* could not compress: write normal cluster */
1103         ret = qcow_co_pwritev(bs, offset, bytes, qiov, 0);
1104         if (ret < 0) {
1105             goto fail;
1106         }
1107         goto success;
1108     }
1109     qemu_co_mutex_lock(&s->lock);
1110     ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1111     qemu_co_mutex_unlock(&s->lock);
1112     if (ret < 0) {
1113         goto fail;
1114     }
1115     if (cluster_offset == 0) {
1116         ret = -EIO;
1117         goto fail;
1118     }
1119     cluster_offset &= s->cluster_offset_mask;
1120 
1121     BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1122     ret = bdrv_co_pwrite(bs->file, cluster_offset, out_len, out_buf, 0);
1123     if (ret < 0) {
1124         goto fail;
1125     }
1126 success:
1127     ret = 0;
1128 fail:
1129     qemu_vfree(buf);
1130     g_free(out_buf);
1131     return ret;
1132 }
1133 
1134 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1135 {
1136     BDRVQcowState *s = bs->opaque;
1137     bdi->cluster_size = s->cluster_size;
1138     return 0;
1139 }
1140 
1141 static QemuOptsList qcow_create_opts = {
1142     .name = "qcow-create-opts",
1143     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1144     .desc = {
1145         {
1146             .name = BLOCK_OPT_SIZE,
1147             .type = QEMU_OPT_SIZE,
1148             .help = "Virtual disk size"
1149         },
1150         {
1151             .name = BLOCK_OPT_BACKING_FILE,
1152             .type = QEMU_OPT_STRING,
1153             .help = "File name of a base image"
1154         },
1155         {
1156             .name = BLOCK_OPT_ENCRYPT,
1157             .type = QEMU_OPT_BOOL,
1158             .help = "Encrypt the image with format 'aes'. (Deprecated "
1159                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1160         },
1161         {
1162             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1163             .type = QEMU_OPT_STRING,
1164             .help = "Encrypt the image, format choices: 'aes'",
1165         },
1166         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1167         { /* end of list */ }
1168     }
1169 };
1170 
1171 static const char *const qcow_strong_runtime_opts[] = {
1172     "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
1173 
1174     NULL
1175 };
1176 
1177 static BlockDriver bdrv_qcow = {
1178     .format_name	= "qcow",
1179     .instance_size	= sizeof(BDRVQcowState),
1180     .bdrv_probe		= qcow_probe,
1181     .bdrv_open		= qcow_open,
1182     .bdrv_close		= qcow_close,
1183     .bdrv_child_perm        = bdrv_format_default_perms,
1184     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1185     .bdrv_co_create         = qcow_co_create,
1186     .bdrv_co_create_opts    = qcow_co_create_opts,
1187     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1188     .supports_backing       = true,
1189     .bdrv_refresh_limits    = qcow_refresh_limits,
1190 
1191     .bdrv_co_preadv         = qcow_co_preadv,
1192     .bdrv_co_pwritev        = qcow_co_pwritev,
1193     .bdrv_co_block_status   = qcow_co_block_status,
1194 
1195     .bdrv_make_empty        = qcow_make_empty,
1196     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1197     .bdrv_get_info          = qcow_get_info,
1198 
1199     .create_opts            = &qcow_create_opts,
1200     .strong_runtime_opts    = qcow_strong_runtime_opts,
1201 };
1202 
1203 static void bdrv_qcow_init(void)
1204 {
1205     bdrv_register(&bdrv_qcow);
1206 }
1207 
1208 block_init(bdrv_qcow_init);
1209