xref: /openbmc/qemu/block/qcow.c (revision 87417811)
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(s->migration_blocker, errp);
311     if (ret < 0) {
312         error_free(s->migration_blocker);
313         goto fail;
314     }
315 
316     qobject_unref(encryptopts);
317     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
318     qemu_co_mutex_init(&s->lock);
319     return 0;
320 
321  fail:
322     g_free(s->l1_table);
323     qemu_vfree(s->l2_cache);
324     g_free(s->cluster_cache);
325     g_free(s->cluster_data);
326     qcrypto_block_free(s->crypto);
327     qobject_unref(encryptopts);
328     qapi_free_QCryptoBlockOpenOptions(crypto_opts);
329     return ret;
330 }
331 
332 
333 /* We have nothing to do for QCOW reopen, stubs just return
334  * success */
335 static int qcow_reopen_prepare(BDRVReopenState *state,
336                                BlockReopenQueue *queue, Error **errp)
337 {
338     return 0;
339 }
340 
341 
342 /* 'allocate' is:
343  *
344  * 0 to not allocate.
345  *
346  * 1 to allocate a normal cluster (for sector-aligned byte offsets 'n_start'
347  * to 'n_end' within the cluster)
348  *
349  * 2 to allocate a compressed cluster of size
350  * 'compressed_size'. 'compressed_size' must be > 0 and <
351  * cluster_size
352  *
353  * return 0 if not allocated, 1 if *result is assigned, and negative
354  * errno on failure.
355  */
356 static int coroutine_fn GRAPH_RDLOCK
357 get_cluster_offset(BlockDriverState *bs, uint64_t offset, int allocate,
358                    int compressed_size, int n_start, int n_end,
359                    uint64_t *result)
360 {
361     BDRVQcowState *s = bs->opaque;
362     int min_index, i, j, l1_index, l2_index, ret;
363     int64_t l2_offset;
364     uint64_t *l2_table, cluster_offset, tmp;
365     uint32_t min_count;
366     int new_l2_table;
367 
368     *result = 0;
369     l1_index = offset >> (s->l2_bits + s->cluster_bits);
370     l2_offset = s->l1_table[l1_index];
371     new_l2_table = 0;
372     if (!l2_offset) {
373         if (!allocate)
374             return 0;
375         /* allocate a new l2 entry */
376         l2_offset = bdrv_co_getlength(bs->file->bs);
377         if (l2_offset < 0) {
378             return l2_offset;
379         }
380         /* round to cluster size */
381         l2_offset = QEMU_ALIGN_UP(l2_offset, s->cluster_size);
382         /* update the L1 entry */
383         s->l1_table[l1_index] = l2_offset;
384         tmp = cpu_to_be64(l2_offset);
385         BLKDBG_CO_EVENT(bs->file, BLKDBG_L1_UPDATE);
386         ret = bdrv_co_pwrite_sync(bs->file,
387                                   s->l1_table_offset + l1_index * sizeof(tmp),
388                                   sizeof(tmp), &tmp, 0);
389         if (ret < 0) {
390             return ret;
391         }
392         new_l2_table = 1;
393     }
394     for(i = 0; i < L2_CACHE_SIZE; i++) {
395         if (l2_offset == s->l2_cache_offsets[i]) {
396             /* increment the hit count */
397             if (++s->l2_cache_counts[i] == 0xffffffff) {
398                 for(j = 0; j < L2_CACHE_SIZE; j++) {
399                     s->l2_cache_counts[j] >>= 1;
400                 }
401             }
402             l2_table = s->l2_cache + (i << s->l2_bits);
403             goto found;
404         }
405     }
406     /* not found: load a new entry in the least used one */
407     min_index = 0;
408     min_count = 0xffffffff;
409     for(i = 0; i < L2_CACHE_SIZE; i++) {
410         if (s->l2_cache_counts[i] < min_count) {
411             min_count = s->l2_cache_counts[i];
412             min_index = i;
413         }
414     }
415     l2_table = s->l2_cache + (min_index << s->l2_bits);
416     BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_LOAD);
417     if (new_l2_table) {
418         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
419         ret = bdrv_co_pwrite_sync(bs->file, l2_offset,
420                                   s->l2_size * sizeof(uint64_t), l2_table, 0);
421         if (ret < 0) {
422             return ret;
423         }
424     } else {
425         ret = bdrv_co_pread(bs->file, l2_offset,
426                             s->l2_size * sizeof(uint64_t), l2_table, 0);
427         if (ret < 0) {
428             return ret;
429         }
430     }
431     s->l2_cache_offsets[min_index] = l2_offset;
432     s->l2_cache_counts[min_index] = 1;
433  found:
434     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
435     cluster_offset = be64_to_cpu(l2_table[l2_index]);
436     if (!cluster_offset ||
437         ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
438         if (!allocate)
439             return 0;
440         BLKDBG_CO_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC);
441         assert(QEMU_IS_ALIGNED(n_start | n_end, BDRV_SECTOR_SIZE));
442         /* allocate a new cluster */
443         if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
444             (n_end - n_start) < s->cluster_size) {
445             /* if the cluster is already compressed, we must
446                decompress it in the case it is not completely
447                overwritten */
448             if (decompress_cluster(bs, cluster_offset) < 0) {
449                 return -EIO;
450             }
451             cluster_offset = bdrv_co_getlength(bs->file->bs);
452             if ((int64_t) cluster_offset < 0) {
453                 return cluster_offset;
454             }
455             cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
456             /* write the cluster content */
457             BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_AIO);
458             ret = bdrv_co_pwrite(bs->file, cluster_offset, s->cluster_size,
459                                  s->cluster_cache, 0);
460             if (ret < 0) {
461                 return ret;
462             }
463         } else {
464             cluster_offset = bdrv_co_getlength(bs->file->bs);
465             if ((int64_t) cluster_offset < 0) {
466                 return cluster_offset;
467             }
468             if (allocate == 1) {
469                 /* round to cluster size */
470                 cluster_offset = QEMU_ALIGN_UP(cluster_offset, s->cluster_size);
471                 if (cluster_offset + s->cluster_size > INT64_MAX) {
472                     return -E2BIG;
473                 }
474                 ret = bdrv_co_truncate(bs->file,
475                                        cluster_offset + s->cluster_size,
476                                        false, PREALLOC_MODE_OFF, 0, NULL);
477                 if (ret < 0) {
478                     return ret;
479                 }
480                 /* if encrypted, we must initialize the cluster
481                    content which won't be written */
482                 if (bs->encrypted &&
483                     (n_end - n_start) < s->cluster_size) {
484                     uint64_t start_offset;
485                     assert(s->crypto);
486                     start_offset = offset & ~(s->cluster_size - 1);
487                     for (i = 0; i < s->cluster_size; i += BDRV_SECTOR_SIZE) {
488                         if (i < n_start || i >= n_end) {
489                             memset(s->cluster_data, 0x00, BDRV_SECTOR_SIZE);
490                             if (qcrypto_block_encrypt(s->crypto,
491                                                       start_offset + i,
492                                                       s->cluster_data,
493                                                       BDRV_SECTOR_SIZE,
494                                                       NULL) < 0) {
495                                 return -EIO;
496                             }
497                             BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_AIO);
498                             ret = bdrv_co_pwrite(bs->file, cluster_offset + i,
499                                                  BDRV_SECTOR_SIZE,
500                                                  s->cluster_data, 0);
501                             if (ret < 0) {
502                                 return ret;
503                             }
504                         }
505                     }
506                 }
507             } else if (allocate == 2) {
508                 cluster_offset |= QCOW_OFLAG_COMPRESSED |
509                     (uint64_t)compressed_size << (63 - s->cluster_bits);
510             }
511         }
512         /* update L2 table */
513         tmp = cpu_to_be64(cluster_offset);
514         l2_table[l2_index] = tmp;
515         if (allocate == 2) {
516             BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
517         } else {
518             BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_UPDATE);
519         }
520         ret = bdrv_co_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
521                                   sizeof(tmp), &tmp, 0);
522         if (ret < 0) {
523             return ret;
524         }
525     }
526     *result = cluster_offset;
527     return 1;
528 }
529 
530 static int coroutine_fn GRAPH_RDLOCK
531 qcow_co_block_status(BlockDriverState *bs, bool want_zero,
532                      int64_t offset, int64_t bytes, int64_t *pnum,
533                      int64_t *map, BlockDriverState **file)
534 {
535     BDRVQcowState *s = bs->opaque;
536     int index_in_cluster, ret;
537     int64_t n;
538     uint64_t cluster_offset;
539 
540     qemu_co_mutex_lock(&s->lock);
541     ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
542     qemu_co_mutex_unlock(&s->lock);
543     if (ret < 0) {
544         return ret;
545     }
546     index_in_cluster = offset & (s->cluster_size - 1);
547     n = s->cluster_size - index_in_cluster;
548     if (n > bytes) {
549         n = bytes;
550     }
551     *pnum = n;
552     if (!cluster_offset) {
553         return 0;
554     }
555     if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
556         return BDRV_BLOCK_DATA | BDRV_BLOCK_COMPRESSED;
557     }
558     if (s->crypto) {
559         return BDRV_BLOCK_DATA;
560     }
561     *map = cluster_offset | index_in_cluster;
562     *file = bs->file->bs;
563     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
564 }
565 
566 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
567                              const uint8_t *buf, int buf_size)
568 {
569     z_stream strm1, *strm = &strm1;
570     int ret, out_len;
571 
572     memset(strm, 0, sizeof(*strm));
573 
574     strm->next_in = (uint8_t *)buf;
575     strm->avail_in = buf_size;
576     strm->next_out = out_buf;
577     strm->avail_out = out_buf_size;
578 
579     ret = inflateInit2(strm, -12);
580     if (ret != Z_OK)
581         return -1;
582     ret = inflate(strm, Z_FINISH);
583     out_len = strm->next_out - out_buf;
584     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
585         out_len != out_buf_size) {
586         inflateEnd(strm);
587         return -1;
588     }
589     inflateEnd(strm);
590     return 0;
591 }
592 
593 static int coroutine_fn GRAPH_RDLOCK
594 decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
595 {
596     BDRVQcowState *s = bs->opaque;
597     int ret, csize;
598     uint64_t coffset;
599 
600     coffset = cluster_offset & s->cluster_offset_mask;
601     if (s->cluster_cache_offset != coffset) {
602         csize = cluster_offset >> (63 - s->cluster_bits);
603         csize &= (s->cluster_size - 1);
604         BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
605         ret = bdrv_co_pread(bs->file, coffset, csize, s->cluster_data, 0);
606         if (ret < 0)
607             return -1;
608         if (decompress_buffer(s->cluster_cache, s->cluster_size,
609                               s->cluster_data, csize) < 0) {
610             return -1;
611         }
612         s->cluster_cache_offset = coffset;
613     }
614     return 0;
615 }
616 
617 static void qcow_refresh_limits(BlockDriverState *bs, Error **errp)
618 {
619     /* At least encrypted images require 512-byte alignment. Apply the
620      * limit universally, rather than just on encrypted images, as
621      * it's easier to let the block layer handle rounding than to
622      * audit this code further. */
623     bs->bl.request_alignment = BDRV_SECTOR_SIZE;
624 }
625 
626 static int coroutine_fn GRAPH_RDLOCK
627 qcow_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
628                QEMUIOVector *qiov, BdrvRequestFlags flags)
629 {
630     BDRVQcowState *s = bs->opaque;
631     int offset_in_cluster;
632     int ret = 0, n;
633     uint64_t cluster_offset;
634     uint8_t *buf;
635     void *orig_buf;
636 
637     if (qiov->niov > 1) {
638         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
639         if (buf == NULL) {
640             return -ENOMEM;
641         }
642     } else {
643         orig_buf = NULL;
644         buf = (uint8_t *)qiov->iov->iov_base;
645     }
646 
647     qemu_co_mutex_lock(&s->lock);
648 
649     while (bytes != 0) {
650         /* prepare next request */
651         ret = get_cluster_offset(bs, offset, 0, 0, 0, 0, &cluster_offset);
652         if (ret < 0) {
653             break;
654         }
655         offset_in_cluster = offset & (s->cluster_size - 1);
656         n = s->cluster_size - offset_in_cluster;
657         if (n > bytes) {
658             n = bytes;
659         }
660 
661         if (!cluster_offset) {
662             if (bs->backing) {
663                 /* read from the base image */
664                 qemu_co_mutex_unlock(&s->lock);
665                 /* qcow2 emits this on bs->file instead of bs->backing */
666                 BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
667                 ret = bdrv_co_pread(bs->backing, offset, n, buf, 0);
668                 qemu_co_mutex_lock(&s->lock);
669                 if (ret < 0) {
670                     break;
671                 }
672             } else {
673                 /* Note: in this case, no need to wait */
674                 memset(buf, 0, n);
675             }
676         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
677             /* add AIO support for compressed blocks ? */
678             if (decompress_cluster(bs, cluster_offset) < 0) {
679                 ret = -EIO;
680                 break;
681             }
682             memcpy(buf, s->cluster_cache + offset_in_cluster, n);
683         } else {
684             if ((cluster_offset & 511) != 0) {
685                 ret = -EIO;
686                 break;
687             }
688             qemu_co_mutex_unlock(&s->lock);
689             BLKDBG_CO_EVENT(bs->file, BLKDBG_READ_AIO);
690             ret = bdrv_co_pread(bs->file, cluster_offset + offset_in_cluster,
691                                 n, buf, 0);
692             qemu_co_mutex_lock(&s->lock);
693             if (ret < 0) {
694                 break;
695             }
696             if (bs->encrypted) {
697                 assert(s->crypto);
698                 if (qcrypto_block_decrypt(s->crypto,
699                                           offset, buf, n, NULL) < 0) {
700                     ret = -EIO;
701                     break;
702                 }
703             }
704         }
705         ret = 0;
706 
707         bytes -= n;
708         offset += n;
709         buf += n;
710     }
711 
712     qemu_co_mutex_unlock(&s->lock);
713 
714     if (qiov->niov > 1) {
715         qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
716         qemu_vfree(orig_buf);
717     }
718 
719     return ret;
720 }
721 
722 static int coroutine_fn GRAPH_RDLOCK
723 qcow_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
724                 QEMUIOVector *qiov, BdrvRequestFlags flags)
725 {
726     BDRVQcowState *s = bs->opaque;
727     int offset_in_cluster;
728     uint64_t cluster_offset;
729     int ret = 0, n;
730     uint8_t *buf;
731     void *orig_buf;
732 
733     s->cluster_cache_offset = -1; /* disable compressed cache */
734 
735     /* We must always copy the iov when encrypting, so we
736      * don't modify the original data buffer during encryption */
737     if (bs->encrypted || qiov->niov > 1) {
738         buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
739         if (buf == NULL) {
740             return -ENOMEM;
741         }
742         qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
743     } else {
744         orig_buf = NULL;
745         buf = (uint8_t *)qiov->iov->iov_base;
746     }
747 
748     qemu_co_mutex_lock(&s->lock);
749 
750     while (bytes != 0) {
751         offset_in_cluster = offset & (s->cluster_size - 1);
752         n = s->cluster_size - offset_in_cluster;
753         if (n > bytes) {
754             n = bytes;
755         }
756         ret = get_cluster_offset(bs, offset, 1, 0, offset_in_cluster,
757                                  offset_in_cluster + n, &cluster_offset);
758         if (ret < 0) {
759             break;
760         }
761         if (!cluster_offset || (cluster_offset & 511) != 0) {
762             ret = -EIO;
763             break;
764         }
765         if (bs->encrypted) {
766             assert(s->crypto);
767             if (qcrypto_block_encrypt(s->crypto, offset, buf, n, NULL) < 0) {
768                 ret = -EIO;
769                 break;
770             }
771         }
772 
773         qemu_co_mutex_unlock(&s->lock);
774         BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_AIO);
775         ret = bdrv_co_pwrite(bs->file, cluster_offset + offset_in_cluster,
776                              n, buf, 0);
777         qemu_co_mutex_lock(&s->lock);
778         if (ret < 0) {
779             break;
780         }
781         ret = 0;
782 
783         bytes -= n;
784         offset += n;
785         buf += n;
786     }
787     qemu_co_mutex_unlock(&s->lock);
788 
789     qemu_vfree(orig_buf);
790 
791     return ret;
792 }
793 
794 static void qcow_close(BlockDriverState *bs)
795 {
796     BDRVQcowState *s = bs->opaque;
797 
798     qcrypto_block_free(s->crypto);
799     s->crypto = NULL;
800     g_free(s->l1_table);
801     qemu_vfree(s->l2_cache);
802     g_free(s->cluster_cache);
803     g_free(s->cluster_data);
804 
805     migrate_del_blocker(s->migration_blocker);
806     error_free(s->migration_blocker);
807 }
808 
809 static int coroutine_fn GRAPH_UNLOCKED
810 qcow_co_create(BlockdevCreateOptions *opts, Error **errp)
811 {
812     BlockdevCreateOptionsQcow *qcow_opts;
813     int header_size, backing_filename_len, l1_size, shift, i;
814     QCowHeader header;
815     uint8_t *tmp;
816     int64_t total_size = 0;
817     int ret;
818     BlockDriverState *bs;
819     BlockBackend *qcow_blk;
820     QCryptoBlock *crypto = NULL;
821 
822     assert(opts->driver == BLOCKDEV_DRIVER_QCOW);
823     qcow_opts = &opts->u.qcow;
824 
825     /* Sanity checks */
826     total_size = qcow_opts->size;
827     if (total_size == 0) {
828         error_setg(errp, "Image size is too small, cannot be zero length");
829         return -EINVAL;
830     }
831 
832     if (qcow_opts->encrypt &&
833         qcow_opts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_QCOW)
834     {
835         error_setg(errp, "Unsupported encryption format");
836         return -EINVAL;
837     }
838 
839     /* Create BlockBackend to write to the image */
840     bs = bdrv_co_open_blockdev_ref(qcow_opts->file, errp);
841     if (bs == NULL) {
842         return -EIO;
843     }
844 
845     qcow_blk = blk_co_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE,
846                                   BLK_PERM_ALL, errp);
847     if (!qcow_blk) {
848         ret = -EPERM;
849         goto exit;
850     }
851     blk_set_allow_write_beyond_eof(qcow_blk, true);
852 
853     /* Create image format */
854     memset(&header, 0, sizeof(header));
855     header.magic = cpu_to_be32(QCOW_MAGIC);
856     header.version = cpu_to_be32(QCOW_VERSION);
857     header.size = cpu_to_be64(total_size);
858     header_size = sizeof(header);
859     backing_filename_len = 0;
860     if (qcow_opts->backing_file) {
861         if (strcmp(qcow_opts->backing_file, "fat:")) {
862             header.backing_file_offset = cpu_to_be64(header_size);
863             backing_filename_len = strlen(qcow_opts->backing_file);
864             header.backing_file_size = cpu_to_be32(backing_filename_len);
865             header_size += backing_filename_len;
866         } else {
867             /* special backing file for vvfat */
868             qcow_opts->backing_file = NULL;
869         }
870         header.cluster_bits = 9; /* 512 byte cluster to avoid copying
871                                     unmodified sectors */
872         header.l2_bits = 12; /* 32 KB L2 tables */
873     } else {
874         header.cluster_bits = 12; /* 4 KB clusters */
875         header.l2_bits = 9; /* 4 KB L2 tables */
876     }
877     header_size = (header_size + 7) & ~7;
878     shift = header.cluster_bits + header.l2_bits;
879     l1_size = (total_size + (1LL << shift) - 1) >> shift;
880 
881     header.l1_table_offset = cpu_to_be64(header_size);
882 
883     if (qcow_opts->encrypt) {
884         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
885 
886         crypto = qcrypto_block_create(qcow_opts->encrypt, "encrypt.",
887                                       NULL, NULL, NULL, errp);
888         if (!crypto) {
889             ret = -EINVAL;
890             goto exit;
891         }
892     } else {
893         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
894     }
895 
896     /* write all the data */
897     ret = blk_co_pwrite(qcow_blk, 0, sizeof(header), &header, 0);
898     if (ret < 0) {
899         goto exit;
900     }
901 
902     if (qcow_opts->backing_file) {
903         ret = blk_co_pwrite(qcow_blk, sizeof(header), backing_filename_len,
904                             qcow_opts->backing_file, 0);
905         if (ret < 0) {
906             goto exit;
907         }
908     }
909 
910     tmp = g_malloc0(BDRV_SECTOR_SIZE);
911     for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
912          i++) {
913         ret = blk_co_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
914                             BDRV_SECTOR_SIZE, tmp, 0);
915         if (ret < 0) {
916             g_free(tmp);
917             goto exit;
918         }
919     }
920 
921     g_free(tmp);
922     ret = 0;
923 exit:
924     blk_co_unref(qcow_blk);
925     bdrv_co_unref(bs);
926     qcrypto_block_free(crypto);
927     return ret;
928 }
929 
930 static int coroutine_fn GRAPH_UNLOCKED
931 qcow_co_create_opts(BlockDriver *drv, const char *filename,
932                     QemuOpts *opts, Error **errp)
933 {
934     BlockdevCreateOptions *create_options = NULL;
935     BlockDriverState *bs = NULL;
936     QDict *qdict = NULL;
937     Visitor *v;
938     const char *val;
939     int ret;
940     char *backing_fmt;
941 
942     static const QDictRenames opt_renames[] = {
943         { BLOCK_OPT_BACKING_FILE,       "backing-file" },
944         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
945         { NULL, NULL },
946     };
947 
948     /*
949      * We can't actually store a backing format, but can check that
950      * the user's request made sense.
951      */
952     backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT);
953     if (backing_fmt && !bdrv_find_format(backing_fmt)) {
954         error_setg(errp, "unrecognized backing format '%s'", backing_fmt);
955         ret = -EINVAL;
956         goto fail;
957     }
958 
959     /* Parse options and convert legacy syntax */
960     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qcow_create_opts, true);
961 
962     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
963     if (val && !strcmp(val, "on")) {
964         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
965     } else if (val && !strcmp(val, "off")) {
966         qdict_del(qdict, BLOCK_OPT_ENCRYPT);
967     }
968 
969     val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
970     if (val && !strcmp(val, "aes")) {
971         qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
972     }
973 
974     if (!qdict_rename_keys(qdict, opt_renames, errp)) {
975         ret = -EINVAL;
976         goto fail;
977     }
978 
979     /* Create and open the file (protocol layer) */
980     ret = bdrv_co_create_file(filename, opts, errp);
981     if (ret < 0) {
982         goto fail;
983     }
984 
985     bs = bdrv_co_open(filename, NULL, NULL,
986                       BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
987     if (bs == NULL) {
988         ret = -EIO;
989         goto fail;
990     }
991 
992     /* Now get the QAPI type BlockdevCreateOptions */
993     qdict_put_str(qdict, "driver", "qcow");
994     qdict_put_str(qdict, "file", bs->node_name);
995 
996     v = qobject_input_visitor_new_flat_confused(qdict, errp);
997     if (!v) {
998         ret = -EINVAL;
999         goto fail;
1000     }
1001 
1002     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
1003     visit_free(v);
1004     if (!create_options) {
1005         ret = -EINVAL;
1006         goto fail;
1007     }
1008 
1009     /* Silently round up size */
1010     assert(create_options->driver == BLOCKDEV_DRIVER_QCOW);
1011     create_options->u.qcow.size =
1012         ROUND_UP(create_options->u.qcow.size, BDRV_SECTOR_SIZE);
1013 
1014     /* Create the qcow image (format layer) */
1015     ret = qcow_co_create(create_options, errp);
1016     if (ret < 0) {
1017         goto fail;
1018     }
1019 
1020     ret = 0;
1021 fail:
1022     g_free(backing_fmt);
1023     qobject_unref(qdict);
1024     bdrv_co_unref(bs);
1025     qapi_free_BlockdevCreateOptions(create_options);
1026     return ret;
1027 }
1028 
1029 static int qcow_make_empty(BlockDriverState *bs)
1030 {
1031     BDRVQcowState *s = bs->opaque;
1032     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1033     int ret;
1034 
1035     memset(s->l1_table, 0, l1_length);
1036     if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, l1_length, s->l1_table,
1037                          0) < 0)
1038         return -1;
1039     ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length, false,
1040                         PREALLOC_MODE_OFF, 0, NULL);
1041     if (ret < 0)
1042         return ret;
1043 
1044     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
1045     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
1046     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
1047 
1048     return 0;
1049 }
1050 
1051 /* XXX: put compressed sectors first, then all the cluster aligned
1052    tables to avoid losing bytes in alignment */
1053 static int coroutine_fn GRAPH_RDLOCK
1054 qcow_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes,
1055                            QEMUIOVector *qiov)
1056 {
1057     BDRVQcowState *s = bs->opaque;
1058     z_stream strm;
1059     int ret, out_len;
1060     uint8_t *buf, *out_buf;
1061     uint64_t cluster_offset;
1062 
1063     buf = qemu_blockalign(bs, s->cluster_size);
1064     if (bytes != s->cluster_size) {
1065         if (bytes > s->cluster_size ||
1066             offset + bytes != bs->total_sectors << BDRV_SECTOR_BITS)
1067         {
1068             qemu_vfree(buf);
1069             return -EINVAL;
1070         }
1071         /* Zero-pad last write if image size is not cluster aligned */
1072         memset(buf + bytes, 0, s->cluster_size - bytes);
1073     }
1074     qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
1075 
1076     out_buf = g_malloc(s->cluster_size);
1077 
1078     /* best compression, small window, no zlib header */
1079     memset(&strm, 0, sizeof(strm));
1080     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1081                        Z_DEFLATED, -12,
1082                        9, Z_DEFAULT_STRATEGY);
1083     if (ret != 0) {
1084         ret = -EINVAL;
1085         goto fail;
1086     }
1087 
1088     strm.avail_in = s->cluster_size;
1089     strm.next_in = (uint8_t *)buf;
1090     strm.avail_out = s->cluster_size;
1091     strm.next_out = out_buf;
1092 
1093     ret = deflate(&strm, Z_FINISH);
1094     if (ret != Z_STREAM_END && ret != Z_OK) {
1095         deflateEnd(&strm);
1096         ret = -EINVAL;
1097         goto fail;
1098     }
1099     out_len = strm.next_out - out_buf;
1100 
1101     deflateEnd(&strm);
1102 
1103     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1104         /* could not compress: write normal cluster */
1105         ret = qcow_co_pwritev(bs, offset, bytes, qiov, 0);
1106         if (ret < 0) {
1107             goto fail;
1108         }
1109         goto success;
1110     }
1111     qemu_co_mutex_lock(&s->lock);
1112     ret = get_cluster_offset(bs, offset, 2, out_len, 0, 0, &cluster_offset);
1113     qemu_co_mutex_unlock(&s->lock);
1114     if (ret < 0) {
1115         goto fail;
1116     }
1117     if (cluster_offset == 0) {
1118         ret = -EIO;
1119         goto fail;
1120     }
1121     cluster_offset &= s->cluster_offset_mask;
1122 
1123     BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1124     ret = bdrv_co_pwrite(bs->file, cluster_offset, out_len, out_buf, 0);
1125     if (ret < 0) {
1126         goto fail;
1127     }
1128 success:
1129     ret = 0;
1130 fail:
1131     qemu_vfree(buf);
1132     g_free(out_buf);
1133     return ret;
1134 }
1135 
1136 static int coroutine_fn
1137 qcow_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1138 {
1139     BDRVQcowState *s = bs->opaque;
1140     bdi->cluster_size = s->cluster_size;
1141     return 0;
1142 }
1143 
1144 static QemuOptsList qcow_create_opts = {
1145     .name = "qcow-create-opts",
1146     .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1147     .desc = {
1148         {
1149             .name = BLOCK_OPT_SIZE,
1150             .type = QEMU_OPT_SIZE,
1151             .help = "Virtual disk size"
1152         },
1153         {
1154             .name = BLOCK_OPT_BACKING_FILE,
1155             .type = QEMU_OPT_STRING,
1156             .help = "File name of a base image"
1157         },
1158         {
1159             .name = BLOCK_OPT_BACKING_FMT,
1160             .type = QEMU_OPT_STRING,
1161             .help = "Format of the backing image",
1162         },
1163         {
1164             .name = BLOCK_OPT_ENCRYPT,
1165             .type = QEMU_OPT_BOOL,
1166             .help = "Encrypt the image with format 'aes'. (Deprecated "
1167                     "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)",
1168         },
1169         {
1170             .name = BLOCK_OPT_ENCRYPT_FORMAT,
1171             .type = QEMU_OPT_STRING,
1172             .help = "Encrypt the image, format choices: 'aes'",
1173         },
1174         BLOCK_CRYPTO_OPT_DEF_QCOW_KEY_SECRET("encrypt."),
1175         { /* end of list */ }
1176     }
1177 };
1178 
1179 static const char *const qcow_strong_runtime_opts[] = {
1180     "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
1181 
1182     NULL
1183 };
1184 
1185 static BlockDriver bdrv_qcow = {
1186     .format_name	= "qcow",
1187     .instance_size	= sizeof(BDRVQcowState),
1188     .bdrv_probe		= qcow_probe,
1189     .bdrv_open		= qcow_open,
1190     .bdrv_close		= qcow_close,
1191     .bdrv_child_perm        = bdrv_default_perms,
1192     .bdrv_reopen_prepare    = qcow_reopen_prepare,
1193     .bdrv_co_create         = qcow_co_create,
1194     .bdrv_co_create_opts    = qcow_co_create_opts,
1195     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1196     .is_format              = true,
1197     .supports_backing       = true,
1198     .bdrv_refresh_limits    = qcow_refresh_limits,
1199 
1200     .bdrv_co_preadv         = qcow_co_preadv,
1201     .bdrv_co_pwritev        = qcow_co_pwritev,
1202     .bdrv_co_block_status   = qcow_co_block_status,
1203 
1204     .bdrv_make_empty        = qcow_make_empty,
1205     .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
1206     .bdrv_co_get_info       = qcow_co_get_info,
1207 
1208     .create_opts            = &qcow_create_opts,
1209     .strong_runtime_opts    = qcow_strong_runtime_opts,
1210 };
1211 
1212 static void bdrv_qcow_init(void)
1213 {
1214     bdrv_register(&bdrv_qcow);
1215 }
1216 
1217 block_init(bdrv_qcow_init);
1218