xref: /openbmc/qemu/block/qcow.c (revision 6233759a)
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 coroutine_fn GRAPH_RDLOCK
96 decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
97 
98 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
99 {
100     const QCowHeader *cow_header = (const void *)buf;
101 
102     if (buf_size >= sizeof(QCowHeader) &&
103         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
104         be32_to_cpu(cow_header->version) == QCOW_VERSION)
105         return 100;
106     else
107         return 0;
108 }
109 
110 static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
111                      Error **errp)
112 {
113     BDRVQcowState *s = bs->opaque;
114     unsigned int len, i, shift;
115     int ret;
116     QCowHeader header;
117     QCryptoBlockOpenOptions *crypto_opts = NULL;
118     unsigned int cflags = 0;
119     QDict *encryptopts = NULL;
120     const char *encryptfmt;
121 
122     qdict_extract_subqdict(options, &encryptopts, "encrypt.");
123     encryptfmt = qdict_get_try_str(encryptopts, "format");
124 
125     ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
126     if (ret < 0) {
127         goto fail;
128     }
129 
130     ret = bdrv_pread(bs->file, 0, sizeof(header), &header, 0);
131     if (ret < 0) {
132         goto fail;
133     }
134     header.magic = be32_to_cpu(header.magic);
135     header.version = be32_to_cpu(header.version);
136     header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
137     header.backing_file_size = be32_to_cpu(header.backing_file_size);
138     header.mtime = be32_to_cpu(header.mtime);
139     header.size = be64_to_cpu(header.size);
140     header.crypt_method = be32_to_cpu(header.crypt_method);
141     header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
142 
143     if (header.magic != QCOW_MAGIC) {
144         error_setg(errp, "Image not in qcow format");
145         ret = -EINVAL;
146         goto fail;
147     }
148     if (header.version != QCOW_VERSION) {
149         error_setg(errp, "qcow (v%d) does not support qcow version %" PRIu32,
150                    QCOW_VERSION, header.version);
151         if (header.version == 2 || header.version == 3) {
152             error_append_hint(errp, "Try the 'qcow2' driver instead.\n");
153         }
154 
155         ret = -ENOTSUP;
156         goto fail;
157     }
158 
159     if (header.size <= 1) {
160         error_setg(errp, "Image size is too small (must be at least 2 bytes)");
161         ret = -EINVAL;
162         goto fail;
163     }
164     if (header.cluster_bits < 9 || header.cluster_bits > 16) {
165         error_setg(errp, "Cluster size must be between 512 and 64k");
166         ret = -EINVAL;
167         goto fail;
168     }
169 
170     /* l2_bits specifies number of entries; storing a uint64_t in each entry,
171      * so bytes = num_entries << 3. */
172     if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
173         error_setg(errp, "L2 table size must be between 512 and 64k");
174         ret = -EINVAL;
175         goto fail;
176     }
177 
178     s->crypt_method_header = header.crypt_method;
179     if (s->crypt_method_header) {
180         if (bdrv_uses_whitelist() &&
181             s->crypt_method_header == QCOW_CRYPT_AES) {
182             error_setg(errp,
183                        "Use of AES-CBC encrypted qcow images is no longer "
184                        "supported in system emulators");
185             error_append_hint(errp,
186                               "You can use 'qemu-img convert' to convert your "
187                               "image to an alternative supported format, such "
188                               "as unencrypted qcow, or raw with the LUKS "
189                               "format instead.\n");
190             ret = -ENOSYS;
191             goto fail;
192         }
193         if (s->crypt_method_header == QCOW_CRYPT_AES) {
194             if (encryptfmt && !g_str_equal(encryptfmt, "aes")) {
195                 error_setg(errp,
196                            "Header reported 'aes' encryption format but "
197                            "options specify '%s'", encryptfmt);
198                 ret = -EINVAL;
199                 goto fail;
200             }
201             qdict_put_str(encryptopts, "format", "qcow");
202             crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
203             if (!crypto_opts) {
204                 ret = -EINVAL;
205                 goto fail;
206             }
207 
208             if (flags & BDRV_O_NO_IO) {
209                 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
210             }
211             s->crypto = qcrypto_block_open(crypto_opts, "encrypt.",
212                                            NULL, NULL, cflags, 1, errp);
213             if (!s->crypto) {
214                 ret = -EINVAL;
215                 goto fail;
216             }
217         } else {
218             error_setg(errp, "invalid encryption method in qcow header");
219             ret = -EINVAL;
220             goto fail;
221         }
222         bs->encrypted = true;
223     } else {
224         if (encryptfmt) {
225             error_setg(errp, "No encryption in image header, but options "
226                        "specified format '%s'", encryptfmt);
227             ret = -EINVAL;
228             goto fail;
229         }
230     }
231     s->cluster_bits = header.cluster_bits;
232     s->cluster_size = 1 << s->cluster_bits;
233     s->l2_bits = header.l2_bits;
234     s->l2_size = 1 << s->l2_bits;
235     bs->total_sectors = header.size / 512;
236     s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
237 
238     /* read the level 1 table */
239     shift = s->cluster_bits + s->l2_bits;
240     if (header.size > UINT64_MAX - (1LL << shift)) {
241         error_setg(errp, "Image too large");
242         ret = -EINVAL;
243         goto fail;
244     } else {
245         uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
246         if (l1_size > INT_MAX / sizeof(uint64_t)) {
247             error_setg(errp, "Image too large");
248             ret = -EINVAL;
249             goto fail;
250         }
251         s->l1_size = l1_size;
252     }
253 
254     s->l1_table_offset = header.l1_table_offset;
255     s->l1_table = g_try_new(uint64_t, s->l1_size);
256     if (s->l1_table == NULL) {
257         error_setg(errp, "Could not allocate memory for L1 table");
258         ret = -ENOMEM;
259         goto fail;
260     }
261 
262     ret = bdrv_pread(bs->file, s->l1_table_offset,
263                      s->l1_size * sizeof(uint64_t), s->l1_table, 0);
264     if (ret < 0) {
265         goto fail;
266     }
267 
268     for(i = 0;i < s->l1_size; i++) {
269         s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
270     }
271 
272     /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
273     s->l2_cache =
274         qemu_try_blockalign(bs->file->bs,
275                             s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
276     if (s->l2_cache == NULL) {
277         error_setg(errp, "Could not allocate L2 table cache");
278         ret = -ENOMEM;
279         goto fail;
280     }
281     s->cluster_cache = g_malloc(s->cluster_size);
282     s->cluster_data = g_malloc(s->cluster_size);
283     s->cluster_cache_offset = -1;
284 
285     /* read the backing file name */
286     if (header.backing_file_offset != 0) {
287         len = header.backing_file_size;
288         if (len > 1023 || len >= sizeof(bs->backing_file)) {
289             error_setg(errp, "Backing file name too long");
290             ret = -EINVAL;
291             goto fail;
292         }
293         ret = bdrv_pread(bs->file, header.backing_file_offset, len,
294                          bs->auto_backing_file, 0);
295         if (ret < 0) {
296             goto fail;
297         }
298         bs->auto_backing_file[len] = '\0';
299         pstrcpy(bs->backing_file, sizeof(bs->backing_file),
300                 bs->auto_backing_file);
301     }
302 
303     /* Disable migration when qcow images are used */
304     bdrv_graph_rdlock_main_loop();
305     error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
306                "does not support live migration",
307                bdrv_get_device_or_node_name(bs));
308     bdrv_graph_rdunlock_main_loop();
309 
310     ret = migrate_add_blocker_normal(&s->migration_blocker, errp);
311     if (ret < 0) {
312         goto fail;
313     }
314 
315     qobject_unref(encryptopts);
316     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
317     qemu_co_mutex_init(&s->lock);
318     return 0;
319 
320  fail:
321     g_free(s->l1_table);
322     qemu_vfree(s->l2_cache);
323     g_free(s->cluster_cache);
324     g_free(s->cluster_data);
325     qcrypto_block_free(s->crypto);
326     qobject_unref(encryptopts);
327     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
328     return ret;
329 }
330 
331 
332 /* We have nothing to do for QCOW reopen, stubs just return
333  * success */
334 static int qcow_reopen_prepare(BDRVReopenState *state,
335                                BlockReopenQueue *queue, Error **errp)
336 {
337     return 0;
338 }
339 
340 
341 /* 'allocate' is:
342  *
343  * 0 to not allocate.
344  *
345  * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start'
346  * to 'n_end' within the cluster)
347  *
348  * 2 to allocate a compressed cluster of size
349  * 'compressed_size'. 'compressed_size' must be > 0 and <
350  * cluster_size
351  *
352  * return 0 if not allocated, 1 if *result is assigned, and negative
353  * errno on failure.
354  */
355 static int coroutine_fn GRAPH_RDLOCK
356 get_cluster_offset(BlockDriverState *bs, uint64_t offset, int allocate,
357                    int compressed_size, int n_start, int n_end,
358                    uint64_t *result)
359 {
360     BDRVQcowState *s = bs->opaque;
361     int min_index, i, j, l1_index, l2_index, ret;
362     int64_t l2_offset;
363     uint64_t *l2_table, cluster_offset, tmp;
364     uint32_t min_count;
365     int new_l2_table;
366 
367     *result = 0;
368     l1_index = offset >> (s->l2_bits + s->cluster_bits);
369     l2_offset = s->l1_table[l1_index];
370     new_l2_table = 0;
371     if (!l2_offset) {
372         if (!allocate)
373             return 0;
374         /* allocate a new l2 entry */
375         l2_offset = bdrv_co_getlength(bs->file->bs);
376         if (l2_offset < 0) {
377             return l2_offset;
378         }
379         /* round to cluster size */
380         l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size);
381         /* update the L1 entry */
382         s->l1_table[l1_index] = l2_offset;
383         tmp = cpu_to_be64(l2_offset);
384         BLKDBG_CO_EVENT(bs->file, BLKDBG_L1_UPDATE);
385         ret = bdrv_co_pwrite_sync(bs->file,
386                                   s->l1_table_offset + l1_index * sizeof(tmp),
387                                   sizeof(tmp), &tmp, 0);
388         if (ret < 0) {
389             return ret;
390         }
391         new_l2_table = 1;
392     }
393     for(i = 0; i < L2_CACHE_SIZE; i++) {
394         if (l2_offset == s->l2_cache_offsets[i]) {
395             /* increment the hit count */
396             if (++s->l2_cache_counts[i] == 0xffffffff) {
397                 for(j = 0; j < L2_CACHE_SIZE; j++) {
398                     s->l2_cache_counts[j] >>= 1;
399                 }
400             }
401             l2_table = s->l2_cache + (i << s->l2_bits);
402             goto found;
403         }
404     }
405     /* not found: load a new entry in the least used one */
406     min_index = 0;
407     min_count = 0xffffffff;
408     for(i = 0; i < L2_CACHE_SIZE; i++) {
409         if (s->l2_cache_counts[i] < min_count) {
410             min_count = s->l2_cache_counts[i];
411             min_index = i;
412         }
413     }
414     l2_table = s->l2_cache + (min_index << s->l2_bits);
415     BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_LOAD);
416     if (new_l2_table) {
417         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
418         ret = bdrv_co_pwrite_sync(bs->file, l2_offset,
419                                   s->l2_size * sizeof(uint64_t), l2_table, 0);
420         if (ret < 0) {
421             return ret;
422         }
423     } else {
424         ret = bdrv_co_pread(bs->file, l2_offset,
425                             s->l2_size * sizeof(uint64_t), l2_table, 0);
426         if (ret < 0) {
427             return ret;
428         }
429     }
430     s->l2_cache_offsets[min_index] = l2_offset;
431     s->l2_cache_counts[min_index] = 1;
432  found:
433     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
434     cluster_offset = be64_to_cpu(l2_table[l2_index]);
435     if (!cluster_offset ||
436         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
437         if (!allocate)
438             return 0;
439         BLKDBG_CO_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
440         assert(QEMU_IS_ALIGNED(n_start | n_end, BDRV_SECTOR_SIZE));
441         /* allocate a new cluster */
442         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
443             (n_end - n_start) < s->cluster_size) {
444             /* if the cluster is already compressed, we must
445                decompress it in the case it is not completely
446                overwritten */
447             if (decompress_cluster(bs, cluster_offset) < 0) {
448                 return -EIO;
449             }
450             cluster_offset = bdrv_co_getlength(bs->file->bs);
451             if ((int64_t) cluster_offset < 0) {
452                 return cluster_offset;
453             }
454             cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
455             /* write the cluster content */
456             BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_AIO);
457             ret = bdrv_co_pwrite(bs->file, cluster_offset, s->cluster_size,
458                                  s->cluster_cache, 0);
459             if (ret < 0) {
460                 return ret;
461             }
462         } else {
463             cluster_offset = bdrv_co_getlength(bs->file->bs);
464             if ((int64_t) cluster_offset < 0) {
465                 return cluster_offset;
466             }
467             if (allocate == 1) {
468                 /* round to cluster size */
469                 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
470                 if (cluster_offset + s->cluster_size > INT64_MAX) {
471                     return -E2BIG;
472                 }
473                 ret = bdrv_co_truncate(bs->file,
474                                        cluster_offset + s->cluster_size,
475                                        false, PREALLOC_MODE_OFF, 0, 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_CO_EVENT(bs->file, BLKDBG_WRITE_AIO);
497                             ret = bdrv_co_pwrite(bs->file, cluster_offset + i,
498                                                  BDRV_SECTOR_SIZE,
499                                                  s->cluster_data, 0);
500                             if (ret < 0) {
501                                 return ret;
502                             }
503                         }
504                     }
505                 }
506             } else if (allocate == 2) {
507                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
508                     (uint64_t)compressed_size << (63 - s->cluster_bits);
509             }
510         }
511         /* update L2 table */
512         tmp = cpu_to_be64(cluster_offset);
513         l2_table[l2_index] = tmp;
514         if (allocate == 2) {
515             BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
516         } else {
517             BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_UPDATE);
518         }
519         ret = bdrv_co_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
520                                   sizeof(tmp), &tmp, 0);
521         if (ret < 0) {
522             return ret;
523         }
524     }
525     *result = cluster_offset;
526     return 1;
527 }
528 
529 static int coroutine_fn GRAPH_RDLOCK
530 qcow_co_block_status(BlockDriverState *bs, bool want_zero,
531                      int64_t offset, int64_t bytes, int64_t *pnum,
532                      int64_t *map, BlockDriverState **file)
533 {
534     BDRVQcowState *s = bs->opaque;
535     int index_in_cluster, ret;
536     int64_t n;
537     uint64_t cluster_offset;
538 
539     qemu_co_mutex_lock(&s->lock);
540     ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
541     qemu_co_mutex_unlock(&s->lock);
542     if (ret < 0) {
543         return ret;
544     }
545     index_in_cluster = offset & (s->cluster_size - 1);
546     n = s->cluster_size - index_in_cluster;
547     if (n > bytes) {
548         n = bytes;
549     }
550     *pnum = n;
551     if (!cluster_offset) {
552         return 0;
553     }
554     if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
555         return BDRV_BLOCK_DATA | BDRV_BLOCK_COMPRESSED;
556     }
557     if (s->crypto) {
558         return BDRV_BLOCK_DATA;
559     }
560     *map = cluster_offset | index_in_cluster;
561     *file = bs->file->bs;
562     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
563 }
564 
565 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
566                              const uint8_t *buf, int buf_size)
567 {
568     z_stream strm1, *strm = &strm1;
569     int ret, out_len;
570 
571     memset(strm, 0, sizeof(*strm));
572 
573     strm->next_in = (uint8_t *)buf;
574     strm->avail_in = buf_size;
575     strm->next_out = out_buf;
576     strm->avail_out = out_buf_size;
577 
578     ret = inflateInit2(strm, -12);
579     if (ret != Z_OK)
580         return -1;
581     ret = inflate(strm, Z_FINISH);
582     out_len = strm->next_out - out_buf;
583     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
584         out_len != out_buf_size) {
585         inflateEnd(strm);
586         return -1;
587     }
588     inflateEnd(strm);
589     return 0;
590 }
591 
592 static int coroutine_fn GRAPH_RDLOCK
593 decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
594 {
595     BDRVQcowState *s = bs->opaque;
596     int ret, csize;
597     uint64_t coffset;
598 
599     coffset = cluster_offset & s->cluster_offset_mask;
600     if (s->cluster_cache_offset != coffset) {
601         csize = cluster_offset >> (63 - s->cluster_bits);
602         csize &= (s->cluster_size - 1);
603         BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
604         ret = bdrv_co_pread(bs->file, coffset, csize, s->cluster_data, 0);
605         if (ret < 0)
606             return -1;
607         if (decompress_buffer(s->cluster_cache, s->cluster_size,
608                               s->cluster_data, csize) < 0) {
609             return -1;
610         }
611         s->cluster_cache_offset = coffset;
612     }
613     return 0;
614 }
615 
616 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp)
617 {
618     /* At least encrypted images require 512-byte alignment. Apply the
619      * limit universally, rather than just on encrypted images, as
620      * it's easier to let the block layer handle rounding than to
621      * audit this code further. */
622     bs->bl.request_alignment = BDRV_SECTOR_SIZE;
623 }
624 
625 static int coroutine_fn GRAPH_RDLOCK
626 qcow_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
627                QEMUIOVector *qiov, BdrvRequestFlags flags)
628 {
629     BDRVQcowState *s = bs->opaque;
630     int offset_in_cluster;
631     int ret = 0, n;
632     uint64_t cluster_offset;
633     uint8_t *buf;
634     void *orig_buf;
635 
636     if (qiov->niov > 1) {
637         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
638         if (buf == NULL) {
639             return -ENOMEM;
640         }
641     } else {
642         orig_buf = NULL;
643         buf = (uint8_t *)qiov->iov->iov_base;
644     }
645 
646     qemu_co_mutex_lock(&s->lock);
647 
648     while (bytes != 0) {
649         /* prepare next request */
650         ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
651         if (ret < 0) {
652             break;
653         }
654         offset_in_cluster = offset & (s->cluster_size - 1);
655         n = s->cluster_size - offset_in_cluster;
656         if (n > bytes) {
657             n = bytes;
658         }
659 
660         if (!cluster_offset) {
661             if (bs->backing) {
662                 /* read from the base image */
663                 qemu_co_mutex_unlock(&s->lock);
664                 /* qcow2 emits this on bs->file instead of bs->backing */
665                 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
666                 ret = bdrv_co_pread(bs->backing, offset, n, buf, 0);
667                 qemu_co_mutex_lock(&s->lock);
668                 if (ret < 0) {
669                     break;
670                 }
671             } else {
672                 /* Note: in this case, no need to wait */
673                 memset(buf, 0, n);
674             }
675         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
676             /* add AIO support for compressed blocks ? */
677             if (decompress_cluster(bs, cluster_offset) < 0) {
678                 ret = -EIO;
679                 break;
680             }
681             memcpy(buf, s->cluster_cache + offset_in_cluster, n);
682         } else {
683             if ((cluster_offset & 511) != 0) {
684                 ret = -EIO;
685                 break;
686             }
687             qemu_co_mutex_unlock(&s->lock);
688             BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_AIO);
689             ret = bdrv_co_pread(bs->file, cluster_offset + offset_in_cluster,
690                                 n, buf, 0);
691             qemu_co_mutex_lock(&s->lock);
692             if (ret < 0) {
693                 break;
694             }
695             if (bs->encrypted) {
696                 assert(s->crypto);
697                 if (qcrypto_block_decrypt(s->crypto,
698                                           offset, buf, n, NULL) < 0) {
699                     ret = -EIO;
700                     break;
701                 }
702             }
703         }
704         ret = 0;
705 
706         bytes -= n;
707         offset += n;
708         buf += n;
709     }
710 
711     qemu_co_mutex_unlock(&s->lock);
712 
713     if (qiov->niov > 1) {
714         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
715         qemu_vfree(orig_buf);
716     }
717 
718     return ret;
719 }
720 
721 static int coroutine_fn GRAPH_RDLOCK
722 qcow_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
723                 QEMUIOVector *qiov, BdrvRequestFlags flags)
724 {
725     BDRVQcowState *s = bs->opaque;
726     int offset_in_cluster;
727     uint64_t cluster_offset;
728     int ret = 0, n;
729     uint8_t *buf;
730     void *orig_buf;
731 
732     s->cluster_cache_offset = -1; /* disable compressed cache */
733 
734     /* We must always copy the iov when encrypting, so we
735      * don't modify the original data buffer during encryption */
736     if (bs->encrypted || qiov->niov > 1) {
737         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
738         if (buf == NULL) {
739             return -ENOMEM;
740         }
741         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
742     } else {
743         orig_buf = NULL;
744         buf = (uint8_t *)qiov->iov->iov_base;
745     }
746 
747     qemu_co_mutex_lock(&s->lock);
748 
749     while (bytes != 0) {
750         offset_in_cluster = offset & (s->cluster_size - 1);
751         n = s->cluster_size - offset_in_cluster;
752         if (n > bytes) {
753             n = bytes;
754         }
755         ret = get_cluster_offset(bs, offset, 1, 0, offset_in_cluster,
756                                  offset_in_cluster + n, &cluster_offset);
757         if (ret < 0) {
758             break;
759         }
760         if (!cluster_offset || (cluster_offset & 511) != 0) {
761             ret = -EIO;
762             break;
763         }
764         if (bs->encrypted) {
765             assert(s->crypto);
766             if (qcrypto_block_encrypt(s->crypto, offset, buf, n, NULL) < 0) {
767                 ret = -EIO;
768                 break;
769             }
770         }
771 
772         qemu_co_mutex_unlock(&s->lock);
773         BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_AIO);
774         ret = bdrv_co_pwrite(bs->file, cluster_offset + offset_in_cluster,
775                              n, buf, 0);
776         qemu_co_mutex_lock(&s->lock);
777         if (ret < 0) {
778             break;
779         }
780         ret = 0;
781 
782         bytes -= n;
783         offset += n;
784         buf += n;
785     }
786     qemu_co_mutex_unlock(&s->lock);
787 
788     qemu_vfree(orig_buf);
789 
790     return ret;
791 }
792 
793 static void qcow_close(BlockDriverState *bs)
794 {
795     BDRVQcowState *s = bs->opaque;
796 
797     qcrypto_block_free(s->crypto);
798     s->crypto = NULL;
799     g_free(s->l1_table);
800     qemu_vfree(s->l2_cache);
801     g_free(s->cluster_cache);
802     g_free(s->cluster_data);
803 
804     migrate_del_blocker(&s->migration_blocker);
805 }
806 
807 static int coroutine_fn GRAPH_UNLOCKED
808 qcow_co_create(BlockdevCreateOptions *opts, Error **errp)
809 {
810     BlockdevCreateOptionsQcow *qcow_opts;
811     int header_size, backing_filename_len, l1_size, shift, i;
812     QCowHeader header;
813     uint8_t *tmp;
814     int64_t total_size = 0;
815     int ret;
816     BlockDriverState *bs;
817     BlockBackend *qcow_blk;
818     QCryptoBlock *crypto = NULL;
819 
820     assert(opts->driver == BLOCKDEV_DRIVER_QCOW);
821     qcow_opts = &opts->u.qcow;
822 
823     /* Sanity checks */
824     total_size = qcow_opts->size;
825     if (total_size == 0) {
826         error_setg(errp, "Image size is too small, cannot be zero length");
827         return -EINVAL;
828     }
829 
830     if (qcow_opts->encrypt &&
831         qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW)
832     {
833         error_setg(errp, "Unsupported encryption format");
834         return -EINVAL;
835     }
836 
837     /* Create BlockBackend to write to the image */
838     bs = bdrv_co_open_blockdev_ref(qcow_opts->file, errp);
839     if (bs == NULL) {
840         return -EIO;
841     }
842 
843     qcow_blk = blk_co_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE,
844                                   BLK_PERM_ALL, errp);
845     if (!qcow_blk) {
846         ret = -EPERM;
847         goto exit;
848     }
849     blk_set_allow_write_beyond_eof(qcow_blk, true);
850 
851     /* Create image format */
852     memset(&header, 0, sizeof(header));
853     header.magic = cpu_to_be32(QCOW_MAGIC);
854     header.version = cpu_to_be32(QCOW_VERSION);
855     header.size = cpu_to_be64(total_size);
856     header_size = sizeof(header);
857     backing_filename_len = 0;
858     if (qcow_opts->backing_file) {
859         if (strcmp(qcow_opts->backing_file, "fat:")) {
860             header.backing_file_offset = cpu_to_be64(header_size);
861             backing_filename_len = strlen(qcow_opts->backing_file);
862             header.backing_file_size = cpu_to_be32(backing_filename_len);
863             header_size += backing_filename_len;
864         } else {
865             /* special backing file for vvfat */
866             qcow_opts->backing_file = NULL;
867         }
868         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
869                                     unmodified sectors */
870         header.l2_bits = 12; /* 32 KB L2 tables */
871     } else {
872         header.cluster_bits = 12; /* 4 KB clusters */
873         header.l2_bits = 9; /* 4 KB L2 tables */
874     }
875     header_size = (header_size + 7) & ~7;
876     shift = header.cluster_bits + header.l2_bits;
877     l1_size = (total_size + (1LL << shift) - 1) >> shift;
878 
879     header.l1_table_offset = cpu_to_be64(header_size);
880 
881     if (qcow_opts->encrypt) {
882         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
883 
884         crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.",
885                                       NULL, NULL, NULL, errp);
886         if (!crypto) {
887             ret = -EINVAL;
888             goto exit;
889         }
890     } else {
891         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
892     }
893 
894     /* write all the data */
895     ret = blk_co_pwrite(qcow_blk, 0, sizeof(header), &header, 0);
896     if (ret < 0) {
897         goto exit;
898     }
899 
900     if (qcow_opts->backing_file) {
901         ret = blk_co_pwrite(qcow_blk, sizeof(header), backing_filename_len,
902                             qcow_opts->backing_file, 0);
903         if (ret < 0) {
904             goto exit;
905         }
906     }
907 
908     tmp = g_malloc0(BDRV_SECTOR_SIZE);
909     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
910          i++) {
911         ret = blk_co_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
912                             BDRV_SECTOR_SIZE, tmp, 0);
913         if (ret < 0) {
914             g_free(tmp);
915             goto exit;
916         }
917     }
918 
919     g_free(tmp);
920     ret = 0;
921 exit:
922     blk_co_unref(qcow_blk);
923     bdrv_co_unref(bs);
924     qcrypto_block_free(crypto);
925     return ret;
926 }
927 
928 static int coroutine_fn GRAPH_UNLOCKED
929 qcow_co_create_opts(BlockDriver *drv, const char *filename,
930                     QemuOpts *opts, Error **errp)
931 {
932     BlockdevCreateOptions *create_options = NULL;
933     BlockDriverState *bs = NULL;
934     QDict *qdict = NULL;
935     Visitor *v;
936     const char *val;
937     int ret;
938     char *backing_fmt;
939 
940     static const QDictRenames opt_renames[] = {
941         { BLOCK_OPT_BACKING_FILE,       "backing-file" },
942         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
943         { NULL, NULL },
944     };
945 
946     /*
947      * We can't actually store a backing format, but can check that
948      * the user's request made sense.
949      */
950     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
951     if (backing_fmt && !bdrv_find_format(backing_fmt)) {
952         error_setg(errp, "unrecognized backing format '%s'", backing_fmt);
953         ret = -EINVAL;
954         goto fail;
955     }
956 
957     /* Parse options and convert legacy syntax */
958     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true);
959 
960     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
961     if (val && !strcmp(val, "on")) {
962         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
963     } else if (val && !strcmp(val, "off")) {
964         qdict_del(qdict, BLOCK_OPT_ENCRYPT);
965     }
966 
967     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
968     if (val && !strcmp(val, "aes")) {
969         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
970     }
971 
972     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
973         ret = -EINVAL;
974         goto fail;
975     }
976 
977     /* Create and open the file (protocol layer) */
978     ret = bdrv_co_create_file(filename, opts, errp);
979     if (ret < 0) {
980         goto fail;
981     }
982 
983     bs = bdrv_co_open(filename, NULL, NULL,
984                       BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
985     if (bs == NULL) {
986         ret = -EIO;
987         goto fail;
988     }
989 
990     /* Now get the QAPI type BlockdevCreateOptions */
991     qdict_put_str(qdict, "driver", "qcow");
992     qdict_put_str(qdict, "file", bs->node_name);
993 
994     v = qobject_input_visitor_new_flat_confused(qdict, errp);
995     if (!v) {
996         ret = -EINVAL;
997         goto fail;
998     }
999 
1000     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
1001     visit_free(v);
1002     if (!create_options) {
1003         ret = -EINVAL;
1004         goto fail;
1005     }
1006 
1007     /* Silently round up size */
1008     assert(create_options->driver == BLOCKDEV_DRIVER_QCOW);
1009     create_options->u.qcow.size =
1010         ROUND_UP(create_options->u.qcow.size, BDRV_SECTOR_SIZE);
1011 
1012     /* Create the qcow image (format layer) */
1013     ret = qcow_co_create(create_options, errp);
1014     if (ret < 0) {
1015         goto fail;
1016     }
1017 
1018     ret = 0;
1019 fail:
1020     g_free(backing_fmt);
1021     qobject_unref(qdict);
1022     bdrv_co_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, l1_length, s->l1_table,
1035                          0) < 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 int coroutine_fn GRAPH_RDLOCK
1052 qcow_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes,
1053                            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_CO_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 coroutine_fn
1135 qcow_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1136 {
1137     BDRVQcowState *s = bs->opaque;
1138     bdi->cluster_size = s->cluster_size;
1139     return 0;
1140 }
1141 
1142 static QemuOptsList qcow_create_opts = {
1143     .name = "qcow-create-opts",
1144     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1145     .desc = {
1146         {
1147             .name = BLOCK_OPT_SIZE,
1148             .type = QEMU_OPT_SIZE,
1149             .help = "Virtual disk size"
1150         },
1151         {
1152             .name = BLOCK_OPT_BACKING_FILE,
1153             .type = QEMU_OPT_STRING,
1154             .help = "File name of a base image"
1155         },
1156         {
1157             .name = BLOCK_OPT_BACKING_FMT,
1158             .type = QEMU_OPT_STRING,
1159             .help = "Format of the backing image",
1160         },
1161         {
1162             .name = BLOCK_OPT_ENCRYPT,
1163             .type = QEMU_OPT_BOOL,
1164             .help = "Encrypt the image with format 'aes'. (Deprecated "
1165                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1166         },
1167         {
1168             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1169             .type = QEMU_OPT_STRING,
1170             .help = "Encrypt the image, format choices: 'aes'",
1171         },
1172         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1173         { /* end of list */ }
1174     }
1175 };
1176 
1177 static const char *const qcow_strong_runtime_opts[] = {
1178     "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
1179 
1180     NULL
1181 };
1182 
1183 static BlockDriver bdrv_qcow = {
1184     .format_name	= "qcow",
1185     .instance_size	= sizeof(BDRVQcowState),
1186     .bdrv_probe		= qcow_probe,
1187     .bdrv_open		= qcow_open,
1188     .bdrv_close		= qcow_close,
1189     .bdrv_child_perm        = bdrv_default_perms,
1190     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1191     .bdrv_co_create         = qcow_co_create,
1192     .bdrv_co_create_opts    = qcow_co_create_opts,
1193     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1194     .is_format              = true,
1195     .supports_backing       = true,
1196     .bdrv_refresh_limits    = qcow_refresh_limits,
1197 
1198     .bdrv_co_preadv         = qcow_co_preadv,
1199     .bdrv_co_pwritev        = qcow_co_pwritev,
1200     .bdrv_co_block_status   = qcow_co_block_status,
1201 
1202     .bdrv_make_empty        = qcow_make_empty,
1203     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1204     .bdrv_co_get_info       = qcow_co_get_info,
1205 
1206     .create_opts            = &qcow_create_opts,
1207     .strong_runtime_opts    = qcow_strong_runtime_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