]>
Commit | Line | Data |
---|---|---|
585f8587 FB |
1 | /* |
2 | * Block driver for the QCOW version 2 format | |
5fafdf24 | 3 | * |
585f8587 | 4 | * Copyright (c) 2004-2006 Fabrice Bellard |
5fafdf24 | 5 | * |
585f8587 FB |
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 | */ | |
e688df6b | 24 | |
80c71a24 | 25 | #include "qemu/osdep.h" |
2714f13d | 26 | |
609f45ea | 27 | #include "block/qdict.h" |
23588797 | 28 | #include "sysemu/block-backend.h" |
db725815 | 29 | #include "qemu/main-loop.h" |
1de7afc9 | 30 | #include "qemu/module.h" |
0d8c41da | 31 | #include "qcow2.h" |
1de7afc9 | 32 | #include "qemu/error-report.h" |
e688df6b | 33 | #include "qapi/error.h" |
9af23989 | 34 | #include "qapi/qapi-events-block-core.h" |
6b673957 MA |
35 | #include "qapi/qmp/qdict.h" |
36 | #include "qapi/qmp/qstring.h" | |
3cce16f4 | 37 | #include "trace.h" |
1bd0e2d1 | 38 | #include "qemu/option_int.h" |
f348b6d1 | 39 | #include "qemu/cutils.h" |
58369e22 | 40 | #include "qemu/bswap.h" |
b76b4f60 KW |
41 | #include "qapi/qobject-input-visitor.h" |
42 | #include "qapi/qapi-visit-block-core.h" | |
0d8c41da | 43 | #include "crypto.h" |
d710cf57 | 44 | #include "block/aio_task.h" |
585f8587 FB |
45 | |
46 | /* | |
47 | Differences with QCOW: | |
48 | ||
49 | - Support for multiple incremental snapshots. | |
50 | - Memory management by reference counts. | |
51 | - Clusters which have a reference count of one have the bit | |
52 | QCOW_OFLAG_COPIED to optimize write performance. | |
5fafdf24 | 53 | - Size of compressed clusters is stored in sectors to reduce bit usage |
585f8587 FB |
54 | in the cluster offsets. |
55 | - Support for storing additional data (such as the VM state) in the | |
3b46e624 | 56 | snapshots. |
585f8587 FB |
57 | - If a backing store is used, the cluster size is not constrained |
58 | (could be backported to QCOW). | |
59 | - L2 tables have always a size of one cluster. | |
60 | */ | |
61 | ||
9b80ddf3 AL |
62 | |
63 | typedef struct { | |
64 | uint32_t magic; | |
65 | uint32_t len; | |
c4217f64 | 66 | } QEMU_PACKED QCowExtension; |
21d82ac9 | 67 | |
7c80ab3f | 68 | #define QCOW2_EXT_MAGIC_END 0 |
8098969c | 69 | #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xe2792aca |
cfcc4c62 | 70 | #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857 |
4652b8f3 | 71 | #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77 |
88ddffae | 72 | #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875 |
93c24936 | 73 | #define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441 |
9b80ddf3 | 74 | |
c3c10f72 VSO |
75 | static int coroutine_fn |
76 | qcow2_co_preadv_compressed(BlockDriverState *bs, | |
9c4269d5 | 77 | uint64_t cluster_descriptor, |
c3c10f72 VSO |
78 | uint64_t offset, |
79 | uint64_t bytes, | |
df893d25 VSO |
80 | QEMUIOVector *qiov, |
81 | size_t qiov_offset); | |
c3c10f72 | 82 | |
7c80ab3f | 83 | static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename) |
585f8587 FB |
84 | { |
85 | const QCowHeader *cow_header = (const void *)buf; | |
3b46e624 | 86 | |
585f8587 FB |
87 | if (buf_size >= sizeof(QCowHeader) && |
88 | be32_to_cpu(cow_header->magic) == QCOW_MAGIC && | |
6744cbab | 89 | be32_to_cpu(cow_header->version) >= 2) |
585f8587 FB |
90 | return 100; |
91 | else | |
92 | return 0; | |
93 | } | |
94 | ||
9b80ddf3 | 95 | |
4652b8f3 DB |
96 | static ssize_t qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset, |
97 | uint8_t *buf, size_t buflen, | |
98 | void *opaque, Error **errp) | |
99 | { | |
100 | BlockDriverState *bs = opaque; | |
101 | BDRVQcow2State *s = bs->opaque; | |
102 | ssize_t ret; | |
103 | ||
104 | if ((offset + buflen) > s->crypto_header.length) { | |
105 | error_setg(errp, "Request for data outside of extension header"); | |
106 | return -1; | |
107 | } | |
108 | ||
109 | ret = bdrv_pread(bs->file, | |
110 | s->crypto_header.offset + offset, buf, buflen); | |
111 | if (ret < 0) { | |
112 | error_setg_errno(errp, -ret, "Could not read encryption header"); | |
113 | return -1; | |
114 | } | |
115 | return ret; | |
116 | } | |
117 | ||
118 | ||
119 | static ssize_t qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen, | |
120 | void *opaque, Error **errp) | |
121 | { | |
122 | BlockDriverState *bs = opaque; | |
123 | BDRVQcow2State *s = bs->opaque; | |
124 | int64_t ret; | |
125 | int64_t clusterlen; | |
126 | ||
127 | ret = qcow2_alloc_clusters(bs, headerlen); | |
128 | if (ret < 0) { | |
129 | error_setg_errno(errp, -ret, | |
130 | "Cannot allocate cluster for LUKS header size %zu", | |
131 | headerlen); | |
132 | return -1; | |
133 | } | |
134 | ||
135 | s->crypto_header.length = headerlen; | |
136 | s->crypto_header.offset = ret; | |
137 | ||
087ab8e7 DB |
138 | /* |
139 | * Zero fill all space in cluster so it has predictable | |
140 | * content, as we may not initialize some regions of the | |
141 | * header (eg only 1 out of 8 key slots will be initialized) | |
142 | */ | |
4652b8f3 | 143 | clusterlen = size_to_clusters(s, headerlen) * s->cluster_size; |
966b000f | 144 | assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen, false) == 0); |
4652b8f3 | 145 | ret = bdrv_pwrite_zeroes(bs->file, |
087ab8e7 DB |
146 | ret, |
147 | clusterlen, 0); | |
4652b8f3 DB |
148 | if (ret < 0) { |
149 | error_setg_errno(errp, -ret, "Could not zero fill encryption header"); | |
150 | return -1; | |
151 | } | |
152 | ||
153 | return ret; | |
154 | } | |
155 | ||
156 | ||
157 | static ssize_t qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset, | |
158 | const uint8_t *buf, size_t buflen, | |
159 | void *opaque, Error **errp) | |
160 | { | |
161 | BlockDriverState *bs = opaque; | |
162 | BDRVQcow2State *s = bs->opaque; | |
163 | ssize_t ret; | |
164 | ||
165 | if ((offset + buflen) > s->crypto_header.length) { | |
166 | error_setg(errp, "Request for data outside of extension header"); | |
167 | return -1; | |
168 | } | |
169 | ||
170 | ret = bdrv_pwrite(bs->file, | |
171 | s->crypto_header.offset + offset, buf, buflen); | |
172 | if (ret < 0) { | |
173 | error_setg_errno(errp, -ret, "Could not read encryption header"); | |
174 | return -1; | |
175 | } | |
176 | return ret; | |
177 | } | |
178 | ||
90766d9d ML |
179 | static QDict* |
180 | qcow2_extract_crypto_opts(QemuOpts *opts, const char *fmt, Error **errp) | |
181 | { | |
182 | QDict *cryptoopts_qdict; | |
183 | QDict *opts_qdict; | |
184 | ||
185 | /* Extract "encrypt." options into a qdict */ | |
186 | opts_qdict = qemu_opts_to_qdict(opts, NULL); | |
187 | qdict_extract_subqdict(opts_qdict, &cryptoopts_qdict, "encrypt."); | |
188 | qobject_unref(opts_qdict); | |
189 | qdict_put_str(cryptoopts_qdict, "format", fmt); | |
190 | return cryptoopts_qdict; | |
191 | } | |
4652b8f3 | 192 | |
a951a631 | 193 | /* |
9b80ddf3 AL |
194 | * read qcow2 extension and fill bs |
195 | * start reading from start_offset | |
196 | * finish reading upon magic of value 0 or when end_offset reached | |
197 | * unknown magic is skipped (future extension this version knows nothing about) | |
198 | * return 0 upon success, non-0 otherwise | |
199 | */ | |
7c80ab3f | 200 | static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset, |
3ef6c40a | 201 | uint64_t end_offset, void **p_feature_table, |
88ddffae VSO |
202 | int flags, bool *need_update_header, |
203 | Error **errp) | |
9b80ddf3 | 204 | { |
ff99129a | 205 | BDRVQcow2State *s = bs->opaque; |
9b80ddf3 AL |
206 | QCowExtension ext; |
207 | uint64_t offset; | |
75bab85c | 208 | int ret; |
88ddffae VSO |
209 | Qcow2BitmapHeaderExt bitmaps_ext; |
210 | ||
211 | if (need_update_header != NULL) { | |
212 | *need_update_header = false; | |
213 | } | |
9b80ddf3 AL |
214 | |
215 | #ifdef DEBUG_EXT | |
7c80ab3f | 216 | printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); |
9b80ddf3 AL |
217 | #endif |
218 | offset = start_offset; | |
219 | while (offset < end_offset) { | |
220 | ||
221 | #ifdef DEBUG_EXT | |
222 | /* Sanity check */ | |
223 | if (offset > s->cluster_size) | |
7c80ab3f | 224 | printf("qcow2_read_extension: suspicious offset %lu\n", offset); |
9b80ddf3 | 225 | |
9b2260cb | 226 | printf("attempting to read extended header in offset %lu\n", offset); |
9b80ddf3 AL |
227 | #endif |
228 | ||
cf2ab8fc | 229 | ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext)); |
3ef6c40a HR |
230 | if (ret < 0) { |
231 | error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: " | |
232 | "pread fail from offset %" PRIu64, offset); | |
9b80ddf3 AL |
233 | return 1; |
234 | } | |
3b698f52 PM |
235 | ext.magic = be32_to_cpu(ext.magic); |
236 | ext.len = be32_to_cpu(ext.len); | |
9b80ddf3 AL |
237 | offset += sizeof(ext); |
238 | #ifdef DEBUG_EXT | |
239 | printf("ext.magic = 0x%x\n", ext.magic); | |
240 | #endif | |
2ebafc85 | 241 | if (offset > end_offset || ext.len > end_offset - offset) { |
3ef6c40a | 242 | error_setg(errp, "Header extension too large"); |
64ca6aee KW |
243 | return -EINVAL; |
244 | } | |
245 | ||
9b80ddf3 | 246 | switch (ext.magic) { |
7c80ab3f | 247 | case QCOW2_EXT_MAGIC_END: |
9b80ddf3 | 248 | return 0; |
f965509c | 249 | |
7c80ab3f | 250 | case QCOW2_EXT_MAGIC_BACKING_FORMAT: |
f965509c | 251 | if (ext.len >= sizeof(bs->backing_format)) { |
521b2b5d HR |
252 | error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32 |
253 | " too large (>=%zu)", ext.len, | |
254 | sizeof(bs->backing_format)); | |
f965509c AL |
255 | return 2; |
256 | } | |
cf2ab8fc | 257 | ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len); |
3ef6c40a HR |
258 | if (ret < 0) { |
259 | error_setg_errno(errp, -ret, "ERROR: ext_backing_format: " | |
260 | "Could not read format name"); | |
f965509c | 261 | return 3; |
3ef6c40a | 262 | } |
f965509c | 263 | bs->backing_format[ext.len] = '\0'; |
e4603fe1 | 264 | s->image_backing_format = g_strdup(bs->backing_format); |
f965509c AL |
265 | #ifdef DEBUG_EXT |
266 | printf("Qcow2: Got format extension %s\n", bs->backing_format); | |
267 | #endif | |
f965509c AL |
268 | break; |
269 | ||
cfcc4c62 KW |
270 | case QCOW2_EXT_MAGIC_FEATURE_TABLE: |
271 | if (p_feature_table != NULL) { | |
5f14f31d | 272 | void *feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); |
cf2ab8fc | 273 | ret = bdrv_pread(bs->file, offset , feature_table, ext.len); |
cfcc4c62 | 274 | if (ret < 0) { |
3ef6c40a HR |
275 | error_setg_errno(errp, -ret, "ERROR: ext_feature_table: " |
276 | "Could not read table"); | |
cfcc4c62 KW |
277 | return ret; |
278 | } | |
279 | ||
280 | *p_feature_table = feature_table; | |
281 | } | |
282 | break; | |
283 | ||
4652b8f3 DB |
284 | case QCOW2_EXT_MAGIC_CRYPTO_HEADER: { |
285 | unsigned int cflags = 0; | |
286 | if (s->crypt_method_header != QCOW_CRYPT_LUKS) { | |
287 | error_setg(errp, "CRYPTO header extension only " | |
288 | "expected with LUKS encryption method"); | |
289 | return -EINVAL; | |
290 | } | |
291 | if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) { | |
292 | error_setg(errp, "CRYPTO header extension size %u, " | |
293 | "but expected size %zu", ext.len, | |
294 | sizeof(Qcow2CryptoHeaderExtension)); | |
295 | return -EINVAL; | |
296 | } | |
297 | ||
298 | ret = bdrv_pread(bs->file, offset, &s->crypto_header, ext.len); | |
299 | if (ret < 0) { | |
300 | error_setg_errno(errp, -ret, | |
301 | "Unable to read CRYPTO header extension"); | |
302 | return ret; | |
303 | } | |
3b698f52 PM |
304 | s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); |
305 | s->crypto_header.length = be64_to_cpu(s->crypto_header.length); | |
4652b8f3 DB |
306 | |
307 | if ((s->crypto_header.offset % s->cluster_size) != 0) { | |
308 | error_setg(errp, "Encryption header offset '%" PRIu64 "' is " | |
309 | "not a multiple of cluster size '%u'", | |
310 | s->crypto_header.offset, s->cluster_size); | |
311 | return -EINVAL; | |
312 | } | |
313 | ||
314 | if (flags & BDRV_O_NO_IO) { | |
315 | cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; | |
316 | } | |
1cd9a787 | 317 | s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", |
4652b8f3 | 318 | qcow2_crypto_hdr_read_func, |
8ac0f15f | 319 | bs, cflags, QCOW2_MAX_THREADS, errp); |
4652b8f3 DB |
320 | if (!s->crypto) { |
321 | return -EINVAL; | |
322 | } | |
323 | } break; | |
324 | ||
88ddffae VSO |
325 | case QCOW2_EXT_MAGIC_BITMAPS: |
326 | if (ext.len != sizeof(bitmaps_ext)) { | |
327 | error_setg_errno(errp, -ret, "bitmaps_ext: " | |
328 | "Invalid extension length"); | |
329 | return -EINVAL; | |
330 | } | |
331 | ||
332 | if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) { | |
c9ceb3ec HR |
333 | if (s->qcow_version < 3) { |
334 | /* Let's be a bit more specific */ | |
335 | warn_report("This qcow2 v2 image contains bitmaps, but " | |
336 | "they may have been modified by a program " | |
337 | "without persistent bitmap support; so now " | |
338 | "they must all be considered inconsistent"); | |
339 | } else { | |
340 | warn_report("a program lacking bitmap support " | |
341 | "modified this file, so all bitmaps are now " | |
342 | "considered inconsistent"); | |
343 | } | |
55d527a9 AF |
344 | error_printf("Some clusters may be leaked, " |
345 | "run 'qemu-img check -r' on the image " | |
88ddffae VSO |
346 | "file to fix."); |
347 | if (need_update_header != NULL) { | |
348 | /* Updating is needed to drop invalid bitmap extension. */ | |
349 | *need_update_header = true; | |
350 | } | |
351 | break; | |
352 | } | |
353 | ||
354 | ret = bdrv_pread(bs->file, offset, &bitmaps_ext, ext.len); | |
355 | if (ret < 0) { | |
356 | error_setg_errno(errp, -ret, "bitmaps_ext: " | |
357 | "Could not read ext header"); | |
358 | return ret; | |
359 | } | |
360 | ||
361 | if (bitmaps_ext.reserved32 != 0) { | |
362 | error_setg_errno(errp, -ret, "bitmaps_ext: " | |
363 | "Reserved field is not zero"); | |
364 | return -EINVAL; | |
365 | } | |
366 | ||
3b698f52 PM |
367 | bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps); |
368 | bitmaps_ext.bitmap_directory_size = | |
369 | be64_to_cpu(bitmaps_ext.bitmap_directory_size); | |
370 | bitmaps_ext.bitmap_directory_offset = | |
371 | be64_to_cpu(bitmaps_ext.bitmap_directory_offset); | |
88ddffae VSO |
372 | |
373 | if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) { | |
374 | error_setg(errp, | |
375 | "bitmaps_ext: Image has %" PRIu32 " bitmaps, " | |
376 | "exceeding the QEMU supported maximum of %d", | |
377 | bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS); | |
378 | return -EINVAL; | |
379 | } | |
380 | ||
381 | if (bitmaps_ext.nb_bitmaps == 0) { | |
382 | error_setg(errp, "found bitmaps extension with zero bitmaps"); | |
383 | return -EINVAL; | |
384 | } | |
385 | ||
74e60fb5 | 386 | if (offset_into_cluster(s, bitmaps_ext.bitmap_directory_offset)) { |
88ddffae VSO |
387 | error_setg(errp, "bitmaps_ext: " |
388 | "invalid bitmap directory offset"); | |
389 | return -EINVAL; | |
390 | } | |
391 | ||
392 | if (bitmaps_ext.bitmap_directory_size > | |
393 | QCOW2_MAX_BITMAP_DIRECTORY_SIZE) { | |
394 | error_setg(errp, "bitmaps_ext: " | |
395 | "bitmap directory size (%" PRIu64 ") exceeds " | |
396 | "the maximum supported size (%d)", | |
397 | bitmaps_ext.bitmap_directory_size, | |
398 | QCOW2_MAX_BITMAP_DIRECTORY_SIZE); | |
399 | return -EINVAL; | |
400 | } | |
401 | ||
402 | s->nb_bitmaps = bitmaps_ext.nb_bitmaps; | |
403 | s->bitmap_directory_offset = | |
404 | bitmaps_ext.bitmap_directory_offset; | |
405 | s->bitmap_directory_size = | |
406 | bitmaps_ext.bitmap_directory_size; | |
407 | ||
408 | #ifdef DEBUG_EXT | |
409 | printf("Qcow2: Got bitmaps extension: " | |
410 | "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n", | |
411 | s->bitmap_directory_offset, s->nb_bitmaps); | |
412 | #endif | |
413 | break; | |
414 | ||
9b890bdc KW |
415 | case QCOW2_EXT_MAGIC_DATA_FILE: |
416 | { | |
417 | s->image_data_file = g_malloc0(ext.len + 1); | |
418 | ret = bdrv_pread(bs->file, offset, s->image_data_file, ext.len); | |
419 | if (ret < 0) { | |
420 | error_setg_errno(errp, -ret, | |
421 | "ERROR: Could not read data file name"); | |
422 | return ret; | |
423 | } | |
424 | #ifdef DEBUG_EXT | |
425 | printf("Qcow2: Got external data file %s\n", s->image_data_file); | |
426 | #endif | |
427 | break; | |
428 | } | |
429 | ||
9b80ddf3 | 430 | default: |
75bab85c | 431 | /* unknown magic - save it in case we need to rewrite the header */ |
4096974e EB |
432 | /* If you add a new feature, make sure to also update the fast |
433 | * path of qcow2_make_empty() to deal with it. */ | |
75bab85c KW |
434 | { |
435 | Qcow2UnknownHeaderExtension *uext; | |
436 | ||
437 | uext = g_malloc0(sizeof(*uext) + ext.len); | |
438 | uext->magic = ext.magic; | |
439 | uext->len = ext.len; | |
440 | QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); | |
441 | ||
cf2ab8fc | 442 | ret = bdrv_pread(bs->file, offset , uext->data, uext->len); |
75bab85c | 443 | if (ret < 0) { |
3ef6c40a HR |
444 | error_setg_errno(errp, -ret, "ERROR: unknown extension: " |
445 | "Could not read data"); | |
75bab85c KW |
446 | return ret; |
447 | } | |
75bab85c | 448 | } |
9b80ddf3 AL |
449 | break; |
450 | } | |
fd29b4bb KW |
451 | |
452 | offset += ((ext.len + 7) & ~7); | |
9b80ddf3 AL |
453 | } |
454 | ||
455 | return 0; | |
456 | } | |
457 | ||
75bab85c KW |
458 | static void cleanup_unknown_header_ext(BlockDriverState *bs) |
459 | { | |
ff99129a | 460 | BDRVQcow2State *s = bs->opaque; |
75bab85c KW |
461 | Qcow2UnknownHeaderExtension *uext, *next; |
462 | ||
463 | QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) { | |
464 | QLIST_REMOVE(uext, next); | |
465 | g_free(uext); | |
466 | } | |
467 | } | |
9b80ddf3 | 468 | |
a55448b3 HR |
469 | static void report_unsupported_feature(Error **errp, Qcow2Feature *table, |
470 | uint64_t mask) | |
cfcc4c62 | 471 | { |
7cdca2e2 | 472 | g_autoptr(GString) features = g_string_sized_new(60); |
12ac6d3d | 473 | |
cfcc4c62 KW |
474 | while (table && table->name[0] != '\0') { |
475 | if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) { | |
12ac6d3d | 476 | if (mask & (1ULL << table->bit)) { |
7cdca2e2 AG |
477 | if (features->len > 0) { |
478 | g_string_append(features, ", "); | |
479 | } | |
480 | g_string_append_printf(features, "%.46s", table->name); | |
12ac6d3d | 481 | mask &= ~(1ULL << table->bit); |
cfcc4c62 KW |
482 | } |
483 | } | |
484 | table++; | |
485 | } | |
486 | ||
487 | if (mask) { | |
7cdca2e2 AG |
488 | if (features->len > 0) { |
489 | g_string_append(features, ", "); | |
490 | } | |
491 | g_string_append_printf(features, | |
492 | "Unknown incompatible feature: %" PRIx64, mask); | |
cfcc4c62 | 493 | } |
12ac6d3d | 494 | |
7cdca2e2 | 495 | error_setg(errp, "Unsupported qcow2 feature(s): %s", features->str); |
cfcc4c62 KW |
496 | } |
497 | ||
bfe8043e SH |
498 | /* |
499 | * Sets the dirty bit and flushes afterwards if necessary. | |
500 | * | |
501 | * The incompatible_features bit is only set if the image file header was | |
502 | * updated successfully. Therefore it is not required to check the return | |
503 | * value of this function. | |
504 | */ | |
280d3735 | 505 | int qcow2_mark_dirty(BlockDriverState *bs) |
bfe8043e | 506 | { |
ff99129a | 507 | BDRVQcow2State *s = bs->opaque; |
bfe8043e SH |
508 | uint64_t val; |
509 | int ret; | |
510 | ||
511 | assert(s->qcow_version >= 3); | |
512 | ||
513 | if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { | |
514 | return 0; /* already dirty */ | |
515 | } | |
516 | ||
517 | val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); | |
d9ca2ea2 | 518 | ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features), |
bfe8043e SH |
519 | &val, sizeof(val)); |
520 | if (ret < 0) { | |
521 | return ret; | |
522 | } | |
9a4f4c31 | 523 | ret = bdrv_flush(bs->file->bs); |
bfe8043e SH |
524 | if (ret < 0) { |
525 | return ret; | |
526 | } | |
527 | ||
528 | /* Only treat image as dirty if the header was updated successfully */ | |
529 | s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; | |
530 | return 0; | |
531 | } | |
532 | ||
c61d0004 SH |
533 | /* |
534 | * Clears the dirty bit and flushes before if necessary. Only call this | |
535 | * function when there are no pending requests, it does not guard against | |
536 | * concurrent requests dirtying the image. | |
537 | */ | |
538 | static int qcow2_mark_clean(BlockDriverState *bs) | |
539 | { | |
ff99129a | 540 | BDRVQcow2State *s = bs->opaque; |
c61d0004 SH |
541 | |
542 | if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { | |
4c2e5f8f KW |
543 | int ret; |
544 | ||
545 | s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; | |
546 | ||
8b220eb7 | 547 | ret = qcow2_flush_caches(bs); |
c61d0004 SH |
548 | if (ret < 0) { |
549 | return ret; | |
550 | } | |
551 | ||
c61d0004 SH |
552 | return qcow2_update_header(bs); |
553 | } | |
554 | return 0; | |
555 | } | |
556 | ||
69c98726 HR |
557 | /* |
558 | * Marks the image as corrupt. | |
559 | */ | |
560 | int qcow2_mark_corrupt(BlockDriverState *bs) | |
561 | { | |
ff99129a | 562 | BDRVQcow2State *s = bs->opaque; |
69c98726 HR |
563 | |
564 | s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT; | |
565 | return qcow2_update_header(bs); | |
566 | } | |
567 | ||
568 | /* | |
569 | * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes | |
570 | * before if necessary. | |
571 | */ | |
572 | int qcow2_mark_consistent(BlockDriverState *bs) | |
573 | { | |
ff99129a | 574 | BDRVQcow2State *s = bs->opaque; |
69c98726 HR |
575 | |
576 | if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { | |
8b220eb7 | 577 | int ret = qcow2_flush_caches(bs); |
69c98726 HR |
578 | if (ret < 0) { |
579 | return ret; | |
580 | } | |
581 | ||
582 | s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT; | |
583 | return qcow2_update_header(bs); | |
584 | } | |
585 | return 0; | |
586 | } | |
587 | ||
8bc584fe HR |
588 | static void qcow2_add_check_result(BdrvCheckResult *out, |
589 | const BdrvCheckResult *src, | |
590 | bool set_allocation_info) | |
591 | { | |
592 | out->corruptions += src->corruptions; | |
593 | out->leaks += src->leaks; | |
594 | out->check_errors += src->check_errors; | |
595 | out->corruptions_fixed += src->corruptions_fixed; | |
596 | out->leaks_fixed += src->leaks_fixed; | |
597 | ||
598 | if (set_allocation_info) { | |
599 | out->image_end_offset = src->image_end_offset; | |
600 | out->bfi = src->bfi; | |
601 | } | |
602 | } | |
603 | ||
2fd61638 PB |
604 | static int coroutine_fn qcow2_co_check_locked(BlockDriverState *bs, |
605 | BdrvCheckResult *result, | |
606 | BdrvCheckMode fix) | |
acbe5982 | 607 | { |
8bc584fe HR |
608 | BdrvCheckResult snapshot_res = {}; |
609 | BdrvCheckResult refcount_res = {}; | |
610 | int ret; | |
611 | ||
612 | memset(result, 0, sizeof(*result)); | |
613 | ||
614 | ret = qcow2_check_read_snapshot_table(bs, &snapshot_res, fix); | |
8bc584fe | 615 | if (ret < 0) { |
fe446b5d | 616 | qcow2_add_check_result(result, &snapshot_res, false); |
8bc584fe HR |
617 | return ret; |
618 | } | |
619 | ||
620 | ret = qcow2_check_refcounts(bs, &refcount_res, fix); | |
621 | qcow2_add_check_result(result, &refcount_res, true); | |
fe446b5d HR |
622 | if (ret < 0) { |
623 | qcow2_add_check_result(result, &snapshot_res, false); | |
624 | return ret; | |
625 | } | |
626 | ||
627 | ret = qcow2_check_fix_snapshot_table(bs, &snapshot_res, fix); | |
628 | qcow2_add_check_result(result, &snapshot_res, false); | |
acbe5982 SH |
629 | if (ret < 0) { |
630 | return ret; | |
631 | } | |
632 | ||
633 | if (fix && result->check_errors == 0 && result->corruptions == 0) { | |
24530f3e HR |
634 | ret = qcow2_mark_clean(bs); |
635 | if (ret < 0) { | |
636 | return ret; | |
637 | } | |
638 | return qcow2_mark_consistent(bs); | |
acbe5982 SH |
639 | } |
640 | return ret; | |
641 | } | |
642 | ||
2fd61638 PB |
643 | static int coroutine_fn qcow2_co_check(BlockDriverState *bs, |
644 | BdrvCheckResult *result, | |
645 | BdrvCheckMode fix) | |
646 | { | |
647 | BDRVQcow2State *s = bs->opaque; | |
648 | int ret; | |
649 | ||
650 | qemu_co_mutex_lock(&s->lock); | |
651 | ret = qcow2_co_check_locked(bs, result, fix); | |
652 | qemu_co_mutex_unlock(&s->lock); | |
653 | return ret; | |
654 | } | |
655 | ||
0cf0e598 AG |
656 | int qcow2_validate_table(BlockDriverState *bs, uint64_t offset, |
657 | uint64_t entries, size_t entry_len, | |
658 | int64_t max_size_bytes, const char *table_name, | |
659 | Error **errp) | |
8c7de283 | 660 | { |
ff99129a | 661 | BDRVQcow2State *s = bs->opaque; |
8c7de283 | 662 | |
0cf0e598 AG |
663 | if (entries > max_size_bytes / entry_len) { |
664 | error_setg(errp, "%s too large", table_name); | |
665 | return -EFBIG; | |
8c7de283 KW |
666 | } |
667 | ||
0cf0e598 AG |
668 | /* Use signed INT64_MAX as the maximum even for uint64_t header fields, |
669 | * because values will be passed to qemu functions taking int64_t. */ | |
670 | if ((INT64_MAX - entries * entry_len < offset) || | |
671 | (offset_into_cluster(s, offset) != 0)) { | |
672 | error_setg(errp, "%s offset invalid", table_name); | |
8c7de283 KW |
673 | return -EINVAL; |
674 | } | |
675 | ||
676 | return 0; | |
677 | } | |
678 | ||
8a2ce0bc AG |
679 | static const char *const mutable_opts[] = { |
680 | QCOW2_OPT_LAZY_REFCOUNTS, | |
681 | QCOW2_OPT_DISCARD_REQUEST, | |
682 | QCOW2_OPT_DISCARD_SNAPSHOT, | |
683 | QCOW2_OPT_DISCARD_OTHER, | |
684 | QCOW2_OPT_OVERLAP, | |
685 | QCOW2_OPT_OVERLAP_TEMPLATE, | |
686 | QCOW2_OPT_OVERLAP_MAIN_HEADER, | |
687 | QCOW2_OPT_OVERLAP_ACTIVE_L1, | |
688 | QCOW2_OPT_OVERLAP_ACTIVE_L2, | |
689 | QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, | |
690 | QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, | |
691 | QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, | |
692 | QCOW2_OPT_OVERLAP_INACTIVE_L1, | |
693 | QCOW2_OPT_OVERLAP_INACTIVE_L2, | |
694 | QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, | |
695 | QCOW2_OPT_CACHE_SIZE, | |
696 | QCOW2_OPT_L2_CACHE_SIZE, | |
697 | QCOW2_OPT_L2_CACHE_ENTRY_SIZE, | |
698 | QCOW2_OPT_REFCOUNT_CACHE_SIZE, | |
699 | QCOW2_OPT_CACHE_CLEAN_INTERVAL, | |
700 | NULL | |
701 | }; | |
702 | ||
74c4510a KW |
703 | static QemuOptsList qcow2_runtime_opts = { |
704 | .name = "qcow2", | |
705 | .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), | |
706 | .desc = { | |
707 | { | |
64aa99d3 | 708 | .name = QCOW2_OPT_LAZY_REFCOUNTS, |
74c4510a KW |
709 | .type = QEMU_OPT_BOOL, |
710 | .help = "Postpone refcount updates", | |
711 | }, | |
67af674e KW |
712 | { |
713 | .name = QCOW2_OPT_DISCARD_REQUEST, | |
714 | .type = QEMU_OPT_BOOL, | |
715 | .help = "Pass guest discard requests to the layer below", | |
716 | }, | |
717 | { | |
718 | .name = QCOW2_OPT_DISCARD_SNAPSHOT, | |
719 | .type = QEMU_OPT_BOOL, | |
720 | .help = "Generate discard requests when snapshot related space " | |
721 | "is freed", | |
722 | }, | |
723 | { | |
724 | .name = QCOW2_OPT_DISCARD_OTHER, | |
725 | .type = QEMU_OPT_BOOL, | |
726 | .help = "Generate discard requests when other clusters are freed", | |
727 | }, | |
05de7e86 HR |
728 | { |
729 | .name = QCOW2_OPT_OVERLAP, | |
730 | .type = QEMU_OPT_STRING, | |
731 | .help = "Selects which overlap checks to perform from a range of " | |
732 | "templates (none, constant, cached, all)", | |
733 | }, | |
ee42b5ce HR |
734 | { |
735 | .name = QCOW2_OPT_OVERLAP_TEMPLATE, | |
736 | .type = QEMU_OPT_STRING, | |
737 | .help = "Selects which overlap checks to perform from a range of " | |
738 | "templates (none, constant, cached, all)", | |
739 | }, | |
05de7e86 HR |
740 | { |
741 | .name = QCOW2_OPT_OVERLAP_MAIN_HEADER, | |
742 | .type = QEMU_OPT_BOOL, | |
743 | .help = "Check for unintended writes into the main qcow2 header", | |
744 | }, | |
745 | { | |
746 | .name = QCOW2_OPT_OVERLAP_ACTIVE_L1, | |
747 | .type = QEMU_OPT_BOOL, | |
748 | .help = "Check for unintended writes into the active L1 table", | |
749 | }, | |
750 | { | |
751 | .name = QCOW2_OPT_OVERLAP_ACTIVE_L2, | |
752 | .type = QEMU_OPT_BOOL, | |
753 | .help = "Check for unintended writes into an active L2 table", | |
754 | }, | |
755 | { | |
756 | .name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, | |
757 | .type = QEMU_OPT_BOOL, | |
758 | .help = "Check for unintended writes into the refcount table", | |
759 | }, | |
760 | { | |
761 | .name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, | |
762 | .type = QEMU_OPT_BOOL, | |
763 | .help = "Check for unintended writes into a refcount block", | |
764 | }, | |
765 | { | |
766 | .name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, | |
767 | .type = QEMU_OPT_BOOL, | |
768 | .help = "Check for unintended writes into the snapshot table", | |
769 | }, | |
770 | { | |
771 | .name = QCOW2_OPT_OVERLAP_INACTIVE_L1, | |
772 | .type = QEMU_OPT_BOOL, | |
773 | .help = "Check for unintended writes into an inactive L1 table", | |
774 | }, | |
775 | { | |
776 | .name = QCOW2_OPT_OVERLAP_INACTIVE_L2, | |
777 | .type = QEMU_OPT_BOOL, | |
778 | .help = "Check for unintended writes into an inactive L2 table", | |
779 | }, | |
0e4e4318 VSO |
780 | { |
781 | .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, | |
782 | .type = QEMU_OPT_BOOL, | |
783 | .help = "Check for unintended writes into the bitmap directory", | |
784 | }, | |
6c1c8d5d HR |
785 | { |
786 | .name = QCOW2_OPT_CACHE_SIZE, | |
787 | .type = QEMU_OPT_SIZE, | |
788 | .help = "Maximum combined metadata (L2 tables and refcount blocks) " | |
789 | "cache size", | |
790 | }, | |
791 | { | |
792 | .name = QCOW2_OPT_L2_CACHE_SIZE, | |
793 | .type = QEMU_OPT_SIZE, | |
794 | .help = "Maximum L2 table cache size", | |
795 | }, | |
1221fe6f AG |
796 | { |
797 | .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE, | |
798 | .type = QEMU_OPT_SIZE, | |
799 | .help = "Size of each entry in the L2 cache", | |
800 | }, | |
6c1c8d5d HR |
801 | { |
802 | .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE, | |
803 | .type = QEMU_OPT_SIZE, | |
804 | .help = "Maximum refcount block cache size", | |
805 | }, | |
279621c0 AG |
806 | { |
807 | .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL, | |
808 | .type = QEMU_OPT_NUMBER, | |
809 | .help = "Clean unused cache entries after this time (in seconds)", | |
810 | }, | |
4652b8f3 DB |
811 | BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", |
812 | "ID of secret providing qcow2 AES key or LUKS passphrase"), | |
74c4510a KW |
813 | { /* end of list */ } |
814 | }, | |
815 | }; | |
816 | ||
4092e99d | 817 | static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { |
0e4e4318 VSO |
818 | [QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, |
819 | [QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, | |
820 | [QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, | |
821 | [QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, | |
822 | [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, | |
823 | [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, | |
824 | [QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, | |
825 | [QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, | |
826 | [QCOW2_OL_BITMAP_DIRECTORY_BITNR] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY, | |
4092e99d HR |
827 | }; |
828 | ||
279621c0 AG |
829 | static void cache_clean_timer_cb(void *opaque) |
830 | { | |
831 | BlockDriverState *bs = opaque; | |
ff99129a | 832 | BDRVQcow2State *s = bs->opaque; |
b2f68bff AG |
833 | qcow2_cache_clean_unused(s->l2_table_cache); |
834 | qcow2_cache_clean_unused(s->refcount_block_cache); | |
279621c0 AG |
835 | timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + |
836 | (int64_t) s->cache_clean_interval * 1000); | |
837 | } | |
838 | ||
839 | static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context) | |
840 | { | |
ff99129a | 841 | BDRVQcow2State *s = bs->opaque; |
279621c0 AG |
842 | if (s->cache_clean_interval > 0) { |
843 | s->cache_clean_timer = aio_timer_new(context, QEMU_CLOCK_VIRTUAL, | |
844 | SCALE_MS, cache_clean_timer_cb, | |
845 | bs); | |
846 | timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + | |
847 | (int64_t) s->cache_clean_interval * 1000); | |
848 | } | |
849 | } | |
850 | ||
851 | static void cache_clean_timer_del(BlockDriverState *bs) | |
852 | { | |
ff99129a | 853 | BDRVQcow2State *s = bs->opaque; |
279621c0 | 854 | if (s->cache_clean_timer) { |
279621c0 AG |
855 | timer_free(s->cache_clean_timer); |
856 | s->cache_clean_timer = NULL; | |
857 | } | |
858 | } | |
859 | ||
860 | static void qcow2_detach_aio_context(BlockDriverState *bs) | |
861 | { | |
862 | cache_clean_timer_del(bs); | |
863 | } | |
864 | ||
865 | static void qcow2_attach_aio_context(BlockDriverState *bs, | |
866 | AioContext *new_context) | |
867 | { | |
868 | cache_clean_timer_init(bs, new_context); | |
869 | } | |
870 | ||
bc85ef26 HR |
871 | static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts, |
872 | uint64_t *l2_cache_size, | |
1221fe6f | 873 | uint64_t *l2_cache_entry_size, |
6c1c8d5d HR |
874 | uint64_t *refcount_cache_size, Error **errp) |
875 | { | |
ff99129a | 876 | BDRVQcow2State *s = bs->opaque; |
b749562d | 877 | uint64_t combined_cache_size, l2_cache_max_setting; |
6c1c8d5d | 878 | bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set; |
af39bd0d | 879 | bool l2_cache_entry_size_set; |
7af5eea9 | 880 | int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size; |
b749562d | 881 | uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE; |
b70d0820 AG |
882 | uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size); |
883 | /* An L2 table is always one cluster in size so the max cache size | |
884 | * should be a multiple of the cluster size. */ | |
c8fd8554 | 885 | uint64_t max_l2_cache = ROUND_UP(max_l2_entries * l2_entry_size(s), |
b70d0820 | 886 | s->cluster_size); |
6c1c8d5d HR |
887 | |
888 | combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE); | |
889 | l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE); | |
890 | refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE); | |
af39bd0d | 891 | l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE); |
6c1c8d5d HR |
892 | |
893 | combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0); | |
b749562d LB |
894 | l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE, |
895 | DEFAULT_L2_CACHE_MAX_SIZE); | |
6c1c8d5d HR |
896 | *refcount_cache_size = qemu_opt_get_size(opts, |
897 | QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0); | |
898 | ||
1221fe6f AG |
899 | *l2_cache_entry_size = qemu_opt_get_size( |
900 | opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size); | |
901 | ||
b749562d LB |
902 | *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting); |
903 | ||
6c1c8d5d HR |
904 | if (combined_cache_size_set) { |
905 | if (l2_cache_size_set && refcount_cache_size_set) { | |
906 | error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE | |
907 | " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set " | |
308999e9 | 908 | "at the same time"); |
6c1c8d5d | 909 | return; |
b749562d LB |
910 | } else if (l2_cache_size_set && |
911 | (l2_cache_max_setting > combined_cache_size)) { | |
6c1c8d5d HR |
912 | error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed " |
913 | QCOW2_OPT_CACHE_SIZE); | |
914 | return; | |
915 | } else if (*refcount_cache_size > combined_cache_size) { | |
916 | error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed " | |
917 | QCOW2_OPT_CACHE_SIZE); | |
918 | return; | |
919 | } | |
920 | ||
921 | if (l2_cache_size_set) { | |
922 | *refcount_cache_size = combined_cache_size - *l2_cache_size; | |
923 | } else if (refcount_cache_size_set) { | |
924 | *l2_cache_size = combined_cache_size - *refcount_cache_size; | |
925 | } else { | |
52253998 AG |
926 | /* Assign as much memory as possible to the L2 cache, and |
927 | * use the remainder for the refcount cache */ | |
928 | if (combined_cache_size >= max_l2_cache + min_refcount_cache) { | |
929 | *l2_cache_size = max_l2_cache; | |
930 | *refcount_cache_size = combined_cache_size - *l2_cache_size; | |
931 | } else { | |
932 | *refcount_cache_size = | |
933 | MIN(combined_cache_size, min_refcount_cache); | |
934 | *l2_cache_size = combined_cache_size - *refcount_cache_size; | |
935 | } | |
6c1c8d5d | 936 | } |
6c1c8d5d | 937 | } |
af39bd0d AG |
938 | |
939 | /* | |
940 | * If the L2 cache is not enough to cover the whole disk then | |
941 | * default to 4KB entries. Smaller entries reduce the cost of | |
942 | * loads and evictions and increase I/O performance. | |
943 | */ | |
944 | if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) { | |
945 | *l2_cache_entry_size = MIN(s->cluster_size, 4096); | |
946 | } | |
947 | ||
657ada52 LB |
948 | /* l2_cache_size and refcount_cache_size are ensured to have at least |
949 | * their minimum values in qcow2_update_options_prepare() */ | |
1221fe6f AG |
950 | |
951 | if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) || | |
952 | *l2_cache_entry_size > s->cluster_size || | |
953 | !is_power_of_2(*l2_cache_entry_size)) { | |
954 | error_setg(errp, "L2 cache entry size must be a power of two " | |
955 | "between %d and the cluster size (%d)", | |
956 | 1 << MIN_CLUSTER_BITS, s->cluster_size); | |
957 | return; | |
958 | } | |
6c1c8d5d HR |
959 | } |
960 | ||
ee55b173 KW |
961 | typedef struct Qcow2ReopenState { |
962 | Qcow2Cache *l2_table_cache; | |
963 | Qcow2Cache *refcount_block_cache; | |
3c2e511a | 964 | int l2_slice_size; /* Number of entries in a slice of the L2 table */ |
ee55b173 KW |
965 | bool use_lazy_refcounts; |
966 | int overlap_check; | |
967 | bool discard_passthrough[QCOW2_DISCARD_MAX]; | |
968 | uint64_t cache_clean_interval; | |
b25b387f | 969 | QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */ |
ee55b173 KW |
970 | } Qcow2ReopenState; |
971 | ||
972 | static int qcow2_update_options_prepare(BlockDriverState *bs, | |
973 | Qcow2ReopenState *r, | |
974 | QDict *options, int flags, | |
975 | Error **errp) | |
4c75d1a1 KW |
976 | { |
977 | BDRVQcow2State *s = bs->opaque; | |
94edf3fb | 978 | QemuOpts *opts = NULL; |
4c75d1a1 KW |
979 | const char *opt_overlap_check, *opt_overlap_check_template; |
980 | int overlap_check_template = 0; | |
1221fe6f | 981 | uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size; |
4c75d1a1 | 982 | int i; |
b25b387f DB |
983 | const char *encryptfmt; |
984 | QDict *encryptopts = NULL; | |
94edf3fb | 985 | Error *local_err = NULL; |
4c75d1a1 KW |
986 | int ret; |
987 | ||
b25b387f DB |
988 | qdict_extract_subqdict(options, &encryptopts, "encrypt."); |
989 | encryptfmt = qdict_get_try_str(encryptopts, "format"); | |
990 | ||
94edf3fb | 991 | opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); |
af175e85 | 992 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { |
94edf3fb KW |
993 | ret = -EINVAL; |
994 | goto fail; | |
995 | } | |
996 | ||
997 | /* get L2 table/refcount block cache size from command line options */ | |
1221fe6f AG |
998 | read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size, |
999 | &refcount_cache_size, &local_err); | |
94edf3fb KW |
1000 | if (local_err) { |
1001 | error_propagate(errp, local_err); | |
1002 | ret = -EINVAL; | |
1003 | goto fail; | |
1004 | } | |
1005 | ||
1221fe6f | 1006 | l2_cache_size /= l2_cache_entry_size; |
94edf3fb KW |
1007 | if (l2_cache_size < MIN_L2_CACHE_SIZE) { |
1008 | l2_cache_size = MIN_L2_CACHE_SIZE; | |
1009 | } | |
1010 | if (l2_cache_size > INT_MAX) { | |
1011 | error_setg(errp, "L2 cache size too big"); | |
1012 | ret = -EINVAL; | |
1013 | goto fail; | |
1014 | } | |
1015 | ||
1016 | refcount_cache_size /= s->cluster_size; | |
1017 | if (refcount_cache_size < MIN_REFCOUNT_CACHE_SIZE) { | |
1018 | refcount_cache_size = MIN_REFCOUNT_CACHE_SIZE; | |
1019 | } | |
1020 | if (refcount_cache_size > INT_MAX) { | |
1021 | error_setg(errp, "Refcount cache size too big"); | |
1022 | ret = -EINVAL; | |
1023 | goto fail; | |
1024 | } | |
1025 | ||
5b0959a7 KW |
1026 | /* alloc new L2 table/refcount block cache, flush old one */ |
1027 | if (s->l2_table_cache) { | |
1028 | ret = qcow2_cache_flush(bs, s->l2_table_cache); | |
1029 | if (ret) { | |
1030 | error_setg_errno(errp, -ret, "Failed to flush the L2 table cache"); | |
1031 | goto fail; | |
1032 | } | |
1033 | } | |
1034 | ||
1035 | if (s->refcount_block_cache) { | |
1036 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
1037 | if (ret) { | |
1038 | error_setg_errno(errp, -ret, | |
1039 | "Failed to flush the refcount block cache"); | |
1040 | goto fail; | |
1041 | } | |
1042 | } | |
1043 | ||
c8fd8554 | 1044 | r->l2_slice_size = l2_cache_entry_size / l2_entry_size(s); |
1221fe6f AG |
1045 | r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size, |
1046 | l2_cache_entry_size); | |
1047 | r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size, | |
1048 | s->cluster_size); | |
ee55b173 | 1049 | if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) { |
94edf3fb KW |
1050 | error_setg(errp, "Could not allocate metadata caches"); |
1051 | ret = -ENOMEM; | |
1052 | goto fail; | |
1053 | } | |
1054 | ||
1055 | /* New interval for cache cleanup timer */ | |
ee55b173 | 1056 | r->cache_clean_interval = |
5b0959a7 | 1057 | qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL, |
e957b50b | 1058 | DEFAULT_CACHE_CLEAN_INTERVAL); |
91203f08 AG |
1059 | #ifndef CONFIG_LINUX |
1060 | if (r->cache_clean_interval != 0) { | |
1061 | error_setg(errp, QCOW2_OPT_CACHE_CLEAN_INTERVAL | |
1062 | " not supported on this host"); | |
1063 | ret = -EINVAL; | |
1064 | goto fail; | |
1065 | } | |
1066 | #endif | |
ee55b173 | 1067 | if (r->cache_clean_interval > UINT_MAX) { |
94edf3fb KW |
1068 | error_setg(errp, "Cache clean interval too big"); |
1069 | ret = -EINVAL; | |
1070 | goto fail; | |
1071 | } | |
94edf3fb | 1072 | |
5b0959a7 | 1073 | /* lazy-refcounts; flush if going from enabled to disabled */ |
ee55b173 | 1074 | r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, |
4c75d1a1 | 1075 | (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); |
ee55b173 | 1076 | if (r->use_lazy_refcounts && s->qcow_version < 3) { |
007dbc39 KW |
1077 | error_setg(errp, "Lazy refcounts require a qcow2 image with at least " |
1078 | "qemu 1.1 compatibility level"); | |
1079 | ret = -EINVAL; | |
1080 | goto fail; | |
1081 | } | |
4c75d1a1 | 1082 | |
5b0959a7 KW |
1083 | if (s->use_lazy_refcounts && !r->use_lazy_refcounts) { |
1084 | ret = qcow2_mark_clean(bs); | |
1085 | if (ret < 0) { | |
1086 | error_setg_errno(errp, -ret, "Failed to disable lazy refcounts"); | |
1087 | goto fail; | |
1088 | } | |
1089 | } | |
1090 | ||
007dbc39 | 1091 | /* Overlap check options */ |
4c75d1a1 KW |
1092 | opt_overlap_check = qemu_opt_get(opts, QCOW2_OPT_OVERLAP); |
1093 | opt_overlap_check_template = qemu_opt_get(opts, QCOW2_OPT_OVERLAP_TEMPLATE); | |
1094 | if (opt_overlap_check_template && opt_overlap_check && | |
1095 | strcmp(opt_overlap_check_template, opt_overlap_check)) | |
1096 | { | |
1097 | error_setg(errp, "Conflicting values for qcow2 options '" | |
1098 | QCOW2_OPT_OVERLAP "' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE | |
1099 | "' ('%s')", opt_overlap_check, opt_overlap_check_template); | |
1100 | ret = -EINVAL; | |
1101 | goto fail; | |
1102 | } | |
1103 | if (!opt_overlap_check) { | |
1104 | opt_overlap_check = opt_overlap_check_template ?: "cached"; | |
1105 | } | |
1106 | ||
1107 | if (!strcmp(opt_overlap_check, "none")) { | |
1108 | overlap_check_template = 0; | |
1109 | } else if (!strcmp(opt_overlap_check, "constant")) { | |
1110 | overlap_check_template = QCOW2_OL_CONSTANT; | |
1111 | } else if (!strcmp(opt_overlap_check, "cached")) { | |
1112 | overlap_check_template = QCOW2_OL_CACHED; | |
1113 | } else if (!strcmp(opt_overlap_check, "all")) { | |
1114 | overlap_check_template = QCOW2_OL_ALL; | |
1115 | } else { | |
1116 | error_setg(errp, "Unsupported value '%s' for qcow2 option " | |
1117 | "'overlap-check'. Allowed are any of the following: " | |
1118 | "none, constant, cached, all", opt_overlap_check); | |
1119 | ret = -EINVAL; | |
1120 | goto fail; | |
1121 | } | |
1122 | ||
ee55b173 | 1123 | r->overlap_check = 0; |
4c75d1a1 KW |
1124 | for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { |
1125 | /* overlap-check defines a template bitmask, but every flag may be | |
1126 | * overwritten through the associated boolean option */ | |
ee55b173 | 1127 | r->overlap_check |= |
4c75d1a1 KW |
1128 | qemu_opt_get_bool(opts, overlap_bool_option_names[i], |
1129 | overlap_check_template & (1 << i)) << i; | |
1130 | } | |
1131 | ||
ee55b173 KW |
1132 | r->discard_passthrough[QCOW2_DISCARD_NEVER] = false; |
1133 | r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true; | |
1134 | r->discard_passthrough[QCOW2_DISCARD_REQUEST] = | |
007dbc39 KW |
1135 | qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, |
1136 | flags & BDRV_O_UNMAP); | |
ee55b173 | 1137 | r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = |
007dbc39 | 1138 | qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true); |
ee55b173 | 1139 | r->discard_passthrough[QCOW2_DISCARD_OTHER] = |
007dbc39 KW |
1140 | qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false); |
1141 | ||
b25b387f DB |
1142 | switch (s->crypt_method_header) { |
1143 | case QCOW_CRYPT_NONE: | |
1144 | if (encryptfmt) { | |
1145 | error_setg(errp, "No encryption in image header, but options " | |
1146 | "specified format '%s'", encryptfmt); | |
1147 | ret = -EINVAL; | |
1148 | goto fail; | |
1149 | } | |
1150 | break; | |
1151 | ||
1152 | case QCOW_CRYPT_AES: | |
1153 | if (encryptfmt && !g_str_equal(encryptfmt, "aes")) { | |
1154 | error_setg(errp, | |
1155 | "Header reported 'aes' encryption format but " | |
1156 | "options specify '%s'", encryptfmt); | |
1157 | ret = -EINVAL; | |
1158 | goto fail; | |
1159 | } | |
796d3239 MA |
1160 | qdict_put_str(encryptopts, "format", "qcow"); |
1161 | r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp); | |
b25b387f DB |
1162 | break; |
1163 | ||
4652b8f3 DB |
1164 | case QCOW_CRYPT_LUKS: |
1165 | if (encryptfmt && !g_str_equal(encryptfmt, "luks")) { | |
1166 | error_setg(errp, | |
1167 | "Header reported 'luks' encryption format but " | |
1168 | "options specify '%s'", encryptfmt); | |
1169 | ret = -EINVAL; | |
1170 | goto fail; | |
1171 | } | |
796d3239 MA |
1172 | qdict_put_str(encryptopts, "format", "luks"); |
1173 | r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp); | |
4652b8f3 DB |
1174 | break; |
1175 | ||
b25b387f DB |
1176 | default: |
1177 | error_setg(errp, "Unsupported encryption method %d", | |
1178 | s->crypt_method_header); | |
1179 | break; | |
1180 | } | |
1181 | if (s->crypt_method_header != QCOW_CRYPT_NONE && !r->crypto_opts) { | |
1182 | ret = -EINVAL; | |
1183 | goto fail; | |
1184 | } | |
1185 | ||
4c75d1a1 KW |
1186 | ret = 0; |
1187 | fail: | |
cb3e7f08 | 1188 | qobject_unref(encryptopts); |
94edf3fb KW |
1189 | qemu_opts_del(opts); |
1190 | opts = NULL; | |
ee55b173 KW |
1191 | return ret; |
1192 | } | |
1193 | ||
1194 | static void qcow2_update_options_commit(BlockDriverState *bs, | |
1195 | Qcow2ReopenState *r) | |
1196 | { | |
1197 | BDRVQcow2State *s = bs->opaque; | |
1198 | int i; | |
1199 | ||
5b0959a7 | 1200 | if (s->l2_table_cache) { |
e64d4072 | 1201 | qcow2_cache_destroy(s->l2_table_cache); |
5b0959a7 KW |
1202 | } |
1203 | if (s->refcount_block_cache) { | |
e64d4072 | 1204 | qcow2_cache_destroy(s->refcount_block_cache); |
5b0959a7 | 1205 | } |
ee55b173 KW |
1206 | s->l2_table_cache = r->l2_table_cache; |
1207 | s->refcount_block_cache = r->refcount_block_cache; | |
3c2e511a | 1208 | s->l2_slice_size = r->l2_slice_size; |
ee55b173 KW |
1209 | |
1210 | s->overlap_check = r->overlap_check; | |
1211 | s->use_lazy_refcounts = r->use_lazy_refcounts; | |
1212 | ||
1213 | for (i = 0; i < QCOW2_DISCARD_MAX; i++) { | |
1214 | s->discard_passthrough[i] = r->discard_passthrough[i]; | |
1215 | } | |
1216 | ||
5b0959a7 KW |
1217 | if (s->cache_clean_interval != r->cache_clean_interval) { |
1218 | cache_clean_timer_del(bs); | |
1219 | s->cache_clean_interval = r->cache_clean_interval; | |
1220 | cache_clean_timer_init(bs, bdrv_get_aio_context(bs)); | |
1221 | } | |
b25b387f DB |
1222 | |
1223 | qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); | |
1224 | s->crypto_opts = r->crypto_opts; | |
ee55b173 KW |
1225 | } |
1226 | ||
1227 | static void qcow2_update_options_abort(BlockDriverState *bs, | |
1228 | Qcow2ReopenState *r) | |
1229 | { | |
1230 | if (r->l2_table_cache) { | |
e64d4072 | 1231 | qcow2_cache_destroy(r->l2_table_cache); |
ee55b173 KW |
1232 | } |
1233 | if (r->refcount_block_cache) { | |
e64d4072 | 1234 | qcow2_cache_destroy(r->refcount_block_cache); |
ee55b173 | 1235 | } |
b25b387f | 1236 | qapi_free_QCryptoBlockOpenOptions(r->crypto_opts); |
ee55b173 KW |
1237 | } |
1238 | ||
1239 | static int qcow2_update_options(BlockDriverState *bs, QDict *options, | |
1240 | int flags, Error **errp) | |
1241 | { | |
1242 | Qcow2ReopenState r = {}; | |
1243 | int ret; | |
1244 | ||
1245 | ret = qcow2_update_options_prepare(bs, &r, options, flags, errp); | |
1246 | if (ret >= 0) { | |
1247 | qcow2_update_options_commit(bs, &r); | |
1248 | } else { | |
1249 | qcow2_update_options_abort(bs, &r); | |
1250 | } | |
94edf3fb | 1251 | |
4c75d1a1 KW |
1252 | return ret; |
1253 | } | |
1254 | ||
572ad978 DP |
1255 | static int validate_compression_type(BDRVQcow2State *s, Error **errp) |
1256 | { | |
1257 | switch (s->compression_type) { | |
1258 | case QCOW2_COMPRESSION_TYPE_ZLIB: | |
d298ac10 DP |
1259 | #ifdef CONFIG_ZSTD |
1260 | case QCOW2_COMPRESSION_TYPE_ZSTD: | |
1261 | #endif | |
572ad978 DP |
1262 | break; |
1263 | ||
1264 | default: | |
1265 | error_setg(errp, "qcow2: unknown compression type: %u", | |
1266 | s->compression_type); | |
1267 | return -ENOTSUP; | |
1268 | } | |
1269 | ||
1270 | /* | |
1271 | * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB | |
1272 | * the incompatible feature flag must be set | |
1273 | */ | |
1274 | if (s->compression_type == QCOW2_COMPRESSION_TYPE_ZLIB) { | |
1275 | if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) { | |
1276 | error_setg(errp, "qcow2: Compression type incompatible feature " | |
1277 | "bit must not be set"); | |
1278 | return -EINVAL; | |
1279 | } | |
1280 | } else { | |
1281 | if (!(s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION)) { | |
1282 | error_setg(errp, "qcow2: Compression type incompatible feature " | |
1283 | "bit must be set"); | |
1284 | return -EINVAL; | |
1285 | } | |
1286 | } | |
1287 | ||
1288 | return 0; | |
1289 | } | |
1290 | ||
1fafcd93 PB |
1291 | /* Called with s->lock held. */ |
1292 | static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, | |
1293 | int flags, Error **errp) | |
585f8587 | 1294 | { |
bc520249 | 1295 | ERRP_GUARD(); |
ff99129a | 1296 | BDRVQcow2State *s = bs->opaque; |
6d33e8e7 KW |
1297 | unsigned int len, i; |
1298 | int ret = 0; | |
585f8587 | 1299 | QCowHeader header; |
74c4510a | 1300 | Error *local_err = NULL; |
9b80ddf3 | 1301 | uint64_t ext_end; |
2cf7cfa1 | 1302 | uint64_t l1_vm_state_index; |
88ddffae | 1303 | bool update_header = false; |
585f8587 | 1304 | |
cf2ab8fc | 1305 | ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); |
6d85a57e | 1306 | if (ret < 0) { |
3ef6c40a | 1307 | error_setg_errno(errp, -ret, "Could not read qcow2 header"); |
585f8587 | 1308 | goto fail; |
6d85a57e | 1309 | } |
3b698f52 PM |
1310 | header.magic = be32_to_cpu(header.magic); |
1311 | header.version = be32_to_cpu(header.version); | |
1312 | header.backing_file_offset = be64_to_cpu(header.backing_file_offset); | |
1313 | header.backing_file_size = be32_to_cpu(header.backing_file_size); | |
1314 | header.size = be64_to_cpu(header.size); | |
1315 | header.cluster_bits = be32_to_cpu(header.cluster_bits); | |
1316 | header.crypt_method = be32_to_cpu(header.crypt_method); | |
1317 | header.l1_table_offset = be64_to_cpu(header.l1_table_offset); | |
1318 | header.l1_size = be32_to_cpu(header.l1_size); | |
1319 | header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset); | |
1320 | header.refcount_table_clusters = | |
1321 | be32_to_cpu(header.refcount_table_clusters); | |
1322 | header.snapshots_offset = be64_to_cpu(header.snapshots_offset); | |
1323 | header.nb_snapshots = be32_to_cpu(header.nb_snapshots); | |
3b46e624 | 1324 | |
e8cdcec1 | 1325 | if (header.magic != QCOW_MAGIC) { |
3ef6c40a | 1326 | error_setg(errp, "Image is not in qcow2 format"); |
76abe407 | 1327 | ret = -EINVAL; |
585f8587 | 1328 | goto fail; |
6d85a57e | 1329 | } |
6744cbab | 1330 | if (header.version < 2 || header.version > 3) { |
a55448b3 | 1331 | error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version); |
6744cbab KW |
1332 | ret = -ENOTSUP; |
1333 | goto fail; | |
1334 | } | |
1335 | ||
1336 | s->qcow_version = header.version; | |
1337 | ||
24342f2c KW |
1338 | /* Initialise cluster size */ |
1339 | if (header.cluster_bits < MIN_CLUSTER_BITS || | |
1340 | header.cluster_bits > MAX_CLUSTER_BITS) { | |
521b2b5d HR |
1341 | error_setg(errp, "Unsupported cluster size: 2^%" PRIu32, |
1342 | header.cluster_bits); | |
24342f2c KW |
1343 | ret = -EINVAL; |
1344 | goto fail; | |
1345 | } | |
1346 | ||
1347 | s->cluster_bits = header.cluster_bits; | |
1348 | s->cluster_size = 1 << s->cluster_bits; | |
24342f2c | 1349 | |
6744cbab KW |
1350 | /* Initialise version 3 header fields */ |
1351 | if (header.version == 2) { | |
1352 | header.incompatible_features = 0; | |
1353 | header.compatible_features = 0; | |
1354 | header.autoclear_features = 0; | |
1355 | header.refcount_order = 4; | |
1356 | header.header_length = 72; | |
1357 | } else { | |
3b698f52 PM |
1358 | header.incompatible_features = |
1359 | be64_to_cpu(header.incompatible_features); | |
1360 | header.compatible_features = be64_to_cpu(header.compatible_features); | |
1361 | header.autoclear_features = be64_to_cpu(header.autoclear_features); | |
1362 | header.refcount_order = be32_to_cpu(header.refcount_order); | |
1363 | header.header_length = be32_to_cpu(header.header_length); | |
24342f2c KW |
1364 | |
1365 | if (header.header_length < 104) { | |
1366 | error_setg(errp, "qcow2 header too short"); | |
1367 | ret = -EINVAL; | |
1368 | goto fail; | |
1369 | } | |
1370 | } | |
1371 | ||
1372 | if (header.header_length > s->cluster_size) { | |
1373 | error_setg(errp, "qcow2 header exceeds cluster size"); | |
1374 | ret = -EINVAL; | |
1375 | goto fail; | |
6744cbab KW |
1376 | } |
1377 | ||
1378 | if (header.header_length > sizeof(header)) { | |
1379 | s->unknown_header_fields_size = header.header_length - sizeof(header); | |
1380 | s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); | |
cf2ab8fc | 1381 | ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields, |
6744cbab KW |
1382 | s->unknown_header_fields_size); |
1383 | if (ret < 0) { | |
3ef6c40a HR |
1384 | error_setg_errno(errp, -ret, "Could not read unknown qcow2 header " |
1385 | "fields"); | |
6744cbab KW |
1386 | goto fail; |
1387 | } | |
1388 | } | |
1389 | ||
a1b3955c KW |
1390 | if (header.backing_file_offset > s->cluster_size) { |
1391 | error_setg(errp, "Invalid backing file offset"); | |
1392 | ret = -EINVAL; | |
1393 | goto fail; | |
1394 | } | |
1395 | ||
cfcc4c62 KW |
1396 | if (header.backing_file_offset) { |
1397 | ext_end = header.backing_file_offset; | |
1398 | } else { | |
1399 | ext_end = 1 << header.cluster_bits; | |
1400 | } | |
1401 | ||
6744cbab KW |
1402 | /* Handle feature bits */ |
1403 | s->incompatible_features = header.incompatible_features; | |
1404 | s->compatible_features = header.compatible_features; | |
1405 | s->autoclear_features = header.autoclear_features; | |
1406 | ||
572ad978 DP |
1407 | /* |
1408 | * Handle compression type | |
1409 | * Older qcow2 images don't contain the compression type header. | |
1410 | * Distinguish them by the header length and use | |
1411 | * the only valid (default) compression type in that case | |
1412 | */ | |
1413 | if (header.header_length > offsetof(QCowHeader, compression_type)) { | |
1414 | s->compression_type = header.compression_type; | |
1415 | } else { | |
1416 | s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB; | |
1417 | } | |
1418 | ||
1419 | ret = validate_compression_type(s, errp); | |
1420 | if (ret) { | |
1421 | goto fail; | |
1422 | } | |
1423 | ||
c61d0004 | 1424 | if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) { |
cfcc4c62 KW |
1425 | void *feature_table = NULL; |
1426 | qcow2_read_extensions(bs, header.header_length, ext_end, | |
88ddffae | 1427 | &feature_table, flags, NULL, NULL); |
a55448b3 | 1428 | report_unsupported_feature(errp, feature_table, |
c61d0004 SH |
1429 | s->incompatible_features & |
1430 | ~QCOW2_INCOMPAT_MASK); | |
6744cbab | 1431 | ret = -ENOTSUP; |
c5a33ee9 | 1432 | g_free(feature_table); |
6744cbab KW |
1433 | goto fail; |
1434 | } | |
1435 | ||
69c98726 HR |
1436 | if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) { |
1437 | /* Corrupt images may not be written to unless they are being repaired | |
1438 | */ | |
1439 | if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) { | |
3ef6c40a HR |
1440 | error_setg(errp, "qcow2: Image is corrupt; cannot be opened " |
1441 | "read/write"); | |
69c98726 HR |
1442 | ret = -EACCES; |
1443 | goto fail; | |
1444 | } | |
1445 | } | |
1446 | ||
d0346b55 AG |
1447 | s->subclusters_per_cluster = |
1448 | has_subclusters(s) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER : 1; | |
1449 | s->subcluster_size = s->cluster_size / s->subclusters_per_cluster; | |
1450 | s->subcluster_bits = ctz32(s->subcluster_size); | |
1451 | ||
7be20252 AG |
1452 | if (s->subcluster_size < (1 << MIN_CLUSTER_BITS)) { |
1453 | error_setg(errp, "Unsupported subcluster size: %d", s->subcluster_size); | |
1454 | ret = -EINVAL; | |
1455 | goto fail; | |
1456 | } | |
1457 | ||
6744cbab | 1458 | /* Check support for various header values */ |
b72faf9f HR |
1459 | if (header.refcount_order > 6) { |
1460 | error_setg(errp, "Reference count entry width too large; may not " | |
1461 | "exceed 64 bits"); | |
1462 | ret = -EINVAL; | |
e8cdcec1 KW |
1463 | goto fail; |
1464 | } | |
b6481f37 | 1465 | s->refcount_order = header.refcount_order; |
346a53df HR |
1466 | s->refcount_bits = 1 << s->refcount_order; |
1467 | s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1); | |
1468 | s->refcount_max += s->refcount_max - 1; | |
6744cbab | 1469 | |
585f8587 | 1470 | s->crypt_method_header = header.crypt_method; |
6d85a57e | 1471 | if (s->crypt_method_header) { |
e6ff69bf DB |
1472 | if (bdrv_uses_whitelist() && |
1473 | s->crypt_method_header == QCOW_CRYPT_AES) { | |
8c0dcbc4 DB |
1474 | error_setg(errp, |
1475 | "Use of AES-CBC encrypted qcow2 images is no longer " | |
1476 | "supported in system emulators"); | |
1477 | error_append_hint(errp, | |
1478 | "You can use 'qemu-img convert' to convert your " | |
1479 | "image to an alternative supported format, such " | |
1480 | "as unencrypted qcow2, or raw with the LUKS " | |
1481 | "format instead.\n"); | |
1482 | ret = -ENOSYS; | |
1483 | goto fail; | |
e6ff69bf DB |
1484 | } |
1485 | ||
4652b8f3 DB |
1486 | if (s->crypt_method_header == QCOW_CRYPT_AES) { |
1487 | s->crypt_physical_offset = false; | |
1488 | } else { | |
1489 | /* Assuming LUKS and any future crypt methods we | |
1490 | * add will all use physical offsets, due to the | |
1491 | * fact that the alternative is insecure... */ | |
1492 | s->crypt_physical_offset = true; | |
1493 | } | |
1494 | ||
54115412 | 1495 | bs->encrypted = true; |
6d85a57e | 1496 | } |
24342f2c | 1497 | |
c8fd8554 | 1498 | s->l2_bits = s->cluster_bits - ctz32(l2_entry_size(s)); |
585f8587 | 1499 | s->l2_size = 1 << s->l2_bits; |
1d13d654 HR |
1500 | /* 2^(s->refcount_order - 3) is the refcount width in bytes */ |
1501 | s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3); | |
1502 | s->refcount_block_size = 1 << s->refcount_block_bits; | |
bd016b91 | 1503 | bs->total_sectors = header.size / BDRV_SECTOR_SIZE; |
585f8587 FB |
1504 | s->csize_shift = (62 - (s->cluster_bits - 8)); |
1505 | s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; | |
1506 | s->cluster_offset_mask = (1LL << s->csize_shift) - 1; | |
5dab2fad | 1507 | |
585f8587 | 1508 | s->refcount_table_offset = header.refcount_table_offset; |
5fafdf24 | 1509 | s->refcount_table_size = |
585f8587 FB |
1510 | header.refcount_table_clusters << (s->cluster_bits - 3); |
1511 | ||
951053a9 AG |
1512 | if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) { |
1513 | error_setg(errp, "Image does not contain a reference count table"); | |
1514 | ret = -EINVAL; | |
1515 | goto fail; | |
1516 | } | |
1517 | ||
0cf0e598 AG |
1518 | ret = qcow2_validate_table(bs, s->refcount_table_offset, |
1519 | header.refcount_table_clusters, | |
1520 | s->cluster_size, QCOW_MAX_REFTABLE_SIZE, | |
1521 | "Reference count table", errp); | |
8c7de283 | 1522 | if (ret < 0) { |
8c7de283 KW |
1523 | goto fail; |
1524 | } | |
1525 | ||
8bc584fe HR |
1526 | if (!(flags & BDRV_O_CHECK)) { |
1527 | /* | |
1528 | * The total size in bytes of the snapshot table is checked in | |
1529 | * qcow2_read_snapshots() because the size of each snapshot is | |
1530 | * variable and we don't know it yet. | |
1531 | * Here we only check the offset and number of snapshots. | |
1532 | */ | |
1533 | ret = qcow2_validate_table(bs, header.snapshots_offset, | |
1534 | header.nb_snapshots, | |
1535 | sizeof(QCowSnapshotHeader), | |
1536 | sizeof(QCowSnapshotHeader) * | |
1537 | QCOW_MAX_SNAPSHOTS, | |
1538 | "Snapshot table", errp); | |
1539 | if (ret < 0) { | |
1540 | goto fail; | |
1541 | } | |
ce48f2f4 KW |
1542 | } |
1543 | ||
585f8587 | 1544 | /* read the level 1 table */ |
0cf0e598 | 1545 | ret = qcow2_validate_table(bs, header.l1_table_offset, |
02b1ecfa | 1546 | header.l1_size, L1E_SIZE, |
0cf0e598 AG |
1547 | QCOW_MAX_L1_SIZE, "Active L1 table", errp); |
1548 | if (ret < 0) { | |
2d51c32c KW |
1549 | goto fail; |
1550 | } | |
585f8587 | 1551 | s->l1_size = header.l1_size; |
0cf0e598 | 1552 | s->l1_table_offset = header.l1_table_offset; |
2cf7cfa1 KW |
1553 | |
1554 | l1_vm_state_index = size_to_l1(s, header.size); | |
1555 | if (l1_vm_state_index > INT_MAX) { | |
3ef6c40a | 1556 | error_setg(errp, "Image is too big"); |
2cf7cfa1 KW |
1557 | ret = -EFBIG; |
1558 | goto fail; | |
1559 | } | |
1560 | s->l1_vm_state_index = l1_vm_state_index; | |
1561 | ||
585f8587 FB |
1562 | /* the L1 table must contain at least enough entries to put |
1563 | header.size bytes */ | |
6d85a57e | 1564 | if (s->l1_size < s->l1_vm_state_index) { |
3ef6c40a | 1565 | error_setg(errp, "L1 table is too small"); |
6d85a57e | 1566 | ret = -EINVAL; |
585f8587 | 1567 | goto fail; |
6d85a57e | 1568 | } |
2d51c32c | 1569 | |
d191d12d | 1570 | if (s->l1_size > 0) { |
02b1ecfa | 1571 | s->l1_table = qemu_try_blockalign(bs->file->bs, s->l1_size * L1E_SIZE); |
de82815d KW |
1572 | if (s->l1_table == NULL) { |
1573 | error_setg(errp, "Could not allocate L1 table"); | |
1574 | ret = -ENOMEM; | |
1575 | goto fail; | |
1576 | } | |
cf2ab8fc | 1577 | ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, |
02b1ecfa | 1578 | s->l1_size * L1E_SIZE); |
6d85a57e | 1579 | if (ret < 0) { |
3ef6c40a | 1580 | error_setg_errno(errp, -ret, "Could not read L1 table"); |
d191d12d | 1581 | goto fail; |
6d85a57e | 1582 | } |
d191d12d | 1583 | for(i = 0;i < s->l1_size; i++) { |
3b698f52 | 1584 | s->l1_table[i] = be64_to_cpu(s->l1_table[i]); |
d191d12d | 1585 | } |
585f8587 | 1586 | } |
29c1a730 | 1587 | |
94edf3fb KW |
1588 | /* Parse driver-specific options */ |
1589 | ret = qcow2_update_options(bs, options, flags, errp); | |
90efa0ea KW |
1590 | if (ret < 0) { |
1591 | goto fail; | |
1592 | } | |
1593 | ||
06d9260f | 1594 | s->flags = flags; |
3b46e624 | 1595 | |
6d85a57e JS |
1596 | ret = qcow2_refcount_init(bs); |
1597 | if (ret != 0) { | |
3ef6c40a | 1598 | error_setg_errno(errp, -ret, "Could not initialize refcount handling"); |
585f8587 | 1599 | goto fail; |
6d85a57e | 1600 | } |
585f8587 | 1601 | |
72cf2d4f | 1602 | QLIST_INIT(&s->cluster_allocs); |
0b919fae | 1603 | QTAILQ_INIT(&s->discards); |
f214978a | 1604 | |
9b80ddf3 | 1605 | /* read qcow2 extensions */ |
3ef6c40a | 1606 | if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, |
af175e85 | 1607 | flags, &update_header, errp)) { |
6d85a57e | 1608 | ret = -EINVAL; |
9b80ddf3 | 1609 | goto fail; |
6d85a57e | 1610 | } |
9b80ddf3 | 1611 | |
0e8c08be | 1612 | /* Open external data file */ |
8b1869da HR |
1613 | s->data_file = bdrv_open_child(NULL, options, "data-file", bs, |
1614 | &child_of_bds, BDRV_CHILD_DATA, | |
bc520249 VSO |
1615 | true, errp); |
1616 | if (*errp) { | |
0e8c08be KW |
1617 | ret = -EINVAL; |
1618 | goto fail; | |
1619 | } | |
1620 | ||
1621 | if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) { | |
9b890bdc KW |
1622 | if (!s->data_file && s->image_data_file) { |
1623 | s->data_file = bdrv_open_child(s->image_data_file, options, | |
8b1869da HR |
1624 | "data-file", bs, &child_of_bds, |
1625 | BDRV_CHILD_DATA, false, errp); | |
9b890bdc KW |
1626 | if (!s->data_file) { |
1627 | ret = -EINVAL; | |
1628 | goto fail; | |
1629 | } | |
1630 | } | |
0e8c08be KW |
1631 | if (!s->data_file) { |
1632 | error_setg(errp, "'data-file' is required for this image"); | |
1633 | ret = -EINVAL; | |
1634 | goto fail; | |
1635 | } | |
8b1869da HR |
1636 | |
1637 | /* No data here */ | |
1638 | bs->file->role &= ~BDRV_CHILD_DATA; | |
1639 | ||
1640 | /* Must succeed because we have given up permissions if anything */ | |
1641 | bdrv_child_refresh_perms(bs, bs->file, &error_abort); | |
0e8c08be KW |
1642 | } else { |
1643 | if (s->data_file) { | |
1644 | error_setg(errp, "'data-file' can only be set for images with an " | |
1645 | "external data file"); | |
1646 | ret = -EINVAL; | |
1647 | goto fail; | |
6c3944dc KW |
1648 | } |
1649 | ||
1650 | s->data_file = bs->file; | |
1651 | ||
1652 | if (data_file_is_raw(bs)) { | |
1653 | error_setg(errp, "data-file-raw requires a data file"); | |
1654 | ret = -EINVAL; | |
1655 | goto fail; | |
0e8c08be KW |
1656 | } |
1657 | } | |
93c24936 | 1658 | |
4652b8f3 DB |
1659 | /* qcow2_read_extension may have set up the crypto context |
1660 | * if the crypt method needs a header region, some methods | |
1661 | * don't need header extensions, so must check here | |
1662 | */ | |
1663 | if (s->crypt_method_header && !s->crypto) { | |
1664 | if (s->crypt_method_header == QCOW_CRYPT_AES) { | |
1665 | unsigned int cflags = 0; | |
1666 | if (flags & BDRV_O_NO_IO) { | |
1667 | cflags |= QCRYPTO_BLOCK_OPEN_NO_IO; | |
1668 | } | |
1cd9a787 | 1669 | s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.", |
8ac0f15f VSO |
1670 | NULL, NULL, cflags, |
1671 | QCOW2_MAX_THREADS, errp); | |
4652b8f3 DB |
1672 | if (!s->crypto) { |
1673 | ret = -EINVAL; | |
1674 | goto fail; | |
1675 | } | |
1676 | } else if (!(flags & BDRV_O_NO_IO)) { | |
1677 | error_setg(errp, "Missing CRYPTO header for crypt method %d", | |
1678 | s->crypt_method_header); | |
b25b387f DB |
1679 | ret = -EINVAL; |
1680 | goto fail; | |
1681 | } | |
1682 | } | |
1683 | ||
585f8587 FB |
1684 | /* read the backing file name */ |
1685 | if (header.backing_file_offset != 0) { | |
1686 | len = header.backing_file_size; | |
9a29e18f | 1687 | if (len > MIN(1023, s->cluster_size - header.backing_file_offset) || |
e729fa6a | 1688 | len >= sizeof(bs->backing_file)) { |
6d33e8e7 KW |
1689 | error_setg(errp, "Backing file name too long"); |
1690 | ret = -EINVAL; | |
1691 | goto fail; | |
6d85a57e | 1692 | } |
cf2ab8fc | 1693 | ret = bdrv_pread(bs->file, header.backing_file_offset, |
998c2019 | 1694 | bs->auto_backing_file, len); |
6d85a57e | 1695 | if (ret < 0) { |
3ef6c40a | 1696 | error_setg_errno(errp, -ret, "Could not read backing file name"); |
585f8587 | 1697 | goto fail; |
6d85a57e | 1698 | } |
998c2019 HR |
1699 | bs->auto_backing_file[len] = '\0'; |
1700 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), | |
1701 | bs->auto_backing_file); | |
1702 | s->image_backing_file = g_strdup(bs->auto_backing_file); | |
585f8587 | 1703 | } |
42deb29f | 1704 | |
8bc584fe HR |
1705 | /* |
1706 | * Internal snapshots; skip reading them in check mode, because | |
1707 | * we do not need them then, and we do not want to abort because | |
1708 | * of a broken table. | |
1709 | */ | |
1710 | if (!(flags & BDRV_O_CHECK)) { | |
1711 | s->snapshots_offset = header.snapshots_offset; | |
1712 | s->nb_snapshots = header.nb_snapshots; | |
11b128f4 | 1713 | |
8bc584fe HR |
1714 | ret = qcow2_read_snapshots(bs, errp); |
1715 | if (ret < 0) { | |
1716 | goto fail; | |
1717 | } | |
6d85a57e | 1718 | } |
585f8587 | 1719 | |
af7b708d | 1720 | /* Clear unknown autoclear feature bits */ |
88ddffae | 1721 | update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK; |
d1258dd0 VSO |
1722 | update_header = |
1723 | update_header && !bs->read_only && !(flags & BDRV_O_INACTIVE); | |
1724 | if (update_header) { | |
88ddffae | 1725 | s->autoclear_features &= QCOW2_AUTOCLEAR_MASK; |
d1258dd0 VSO |
1726 | } |
1727 | ||
9c98f145 VSO |
1728 | /* == Handle persistent dirty bitmaps == |
1729 | * | |
1730 | * We want load dirty bitmaps in three cases: | |
1731 | * | |
1732 | * 1. Normal open of the disk in active mode, not related to invalidation | |
1733 | * after migration. | |
1734 | * | |
1735 | * 2. Invalidation of the target vm after pre-copy phase of migration, if | |
1736 | * bitmaps are _not_ migrating through migration channel, i.e. | |
1737 | * 'dirty-bitmaps' capability is disabled. | |
1738 | * | |
1739 | * 3. Invalidation of source vm after failed or canceled migration. | |
1740 | * This is a very interesting case. There are two possible types of | |
1741 | * bitmaps: | |
1742 | * | |
1743 | * A. Stored on inactivation and removed. They should be loaded from the | |
1744 | * image. | |
1745 | * | |
1746 | * B. Not stored: not-persistent bitmaps and bitmaps, migrated through | |
1747 | * the migration channel (with dirty-bitmaps capability). | |
1748 | * | |
1749 | * On the other hand, there are two possible sub-cases: | |
1750 | * | |
1751 | * 3.1 disk was changed by somebody else while were inactive. In this | |
1752 | * case all in-RAM dirty bitmaps (both persistent and not) are | |
1753 | * definitely invalid. And we don't have any method to determine | |
1754 | * this. | |
1755 | * | |
1756 | * Simple and safe thing is to just drop all the bitmaps of type B on | |
1757 | * inactivation. But in this case we lose bitmaps in valid 4.2 case. | |
1758 | * | |
1759 | * On the other hand, resuming source vm, if disk was already changed | |
1760 | * is a bad thing anyway: not only bitmaps, the whole vm state is | |
1761 | * out of sync with disk. | |
1762 | * | |
1763 | * This means, that user or management tool, who for some reason | |
1764 | * decided to resume source vm, after disk was already changed by | |
1765 | * target vm, should at least drop all dirty bitmaps by hand. | |
1766 | * | |
1767 | * So, we can ignore this case for now, but TODO: "generation" | |
1768 | * extension for qcow2, to determine, that image was changed after | |
1769 | * last inactivation. And if it is changed, we will drop (or at least | |
1770 | * mark as 'invalid' all the bitmaps of type B, both persistent | |
1771 | * and not). | |
1772 | * | |
1773 | * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved | |
1774 | * to disk ('dirty-bitmaps' capability disabled), or not saved | |
1775 | * ('dirty-bitmaps' capability enabled), but we don't need to care | |
1776 | * of: let's load bitmaps as always: stored bitmaps will be loaded, | |
1777 | * and not stored has flag IN_USE=1 in the image and will be skipped | |
1778 | * on loading. | |
1779 | * | |
1780 | * One remaining possible case when we don't want load bitmaps: | |
1781 | * | |
1782 | * 4. Open disk in inactive mode in target vm (bitmaps are migrating or | |
1783 | * will be loaded on invalidation, no needs try loading them before) | |
1784 | */ | |
1785 | ||
1786 | if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) { | |
1787 | /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */ | |
1788 | bool header_updated = qcow2_load_dirty_bitmaps(bs, &local_err); | |
66be5c3e T |
1789 | if (local_err != NULL) { |
1790 | error_propagate(errp, local_err); | |
1791 | ret = -EINVAL; | |
1792 | goto fail; | |
1793 | } | |
9c98f145 VSO |
1794 | |
1795 | update_header = update_header && !header_updated; | |
d1258dd0 | 1796 | } |
d1258dd0 VSO |
1797 | |
1798 | if (update_header) { | |
af7b708d SH |
1799 | ret = qcow2_update_header(bs); |
1800 | if (ret < 0) { | |
3ef6c40a | 1801 | error_setg_errno(errp, -ret, "Could not update qcow2 header"); |
af7b708d SH |
1802 | goto fail; |
1803 | } | |
1804 | } | |
1805 | ||
3b650816 KW |
1806 | bs->supported_zero_flags = header.version >= 3 ? |
1807 | BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0; | |
f01643fb | 1808 | bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; |
68d100e9 | 1809 | |
c61d0004 | 1810 | /* Repair image if dirty */ |
04c01a5c | 1811 | if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only && |
058f8f16 | 1812 | (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) { |
c61d0004 SH |
1813 | BdrvCheckResult result = {0}; |
1814 | ||
2fd61638 PB |
1815 | ret = qcow2_co_check_locked(bs, &result, |
1816 | BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); | |
791fff50 HR |
1817 | if (ret < 0 || result.check_errors) { |
1818 | if (ret >= 0) { | |
1819 | ret = -EIO; | |
1820 | } | |
3ef6c40a | 1821 | error_setg_errno(errp, -ret, "Could not repair dirty image"); |
c61d0004 SH |
1822 | goto fail; |
1823 | } | |
1824 | } | |
1825 | ||
585f8587 | 1826 | #ifdef DEBUG_ALLOC |
6cbc3031 PH |
1827 | { |
1828 | BdrvCheckResult result = {0}; | |
b35278f7 | 1829 | qcow2_check_refcounts(bs, &result, 0); |
6cbc3031 | 1830 | } |
585f8587 | 1831 | #endif |
ceb029cd | 1832 | |
6f13a316 | 1833 | qemu_co_queue_init(&s->thread_task_queue); |
ceb029cd | 1834 | |
6d85a57e | 1835 | return ret; |
585f8587 FB |
1836 | |
1837 | fail: | |
9b890bdc | 1838 | g_free(s->image_data_file); |
0e8c08be KW |
1839 | if (has_data_file(bs)) { |
1840 | bdrv_unref_child(bs, s->data_file); | |
808cf3cb | 1841 | s->data_file = NULL; |
0e8c08be | 1842 | } |
6744cbab | 1843 | g_free(s->unknown_header_fields); |
75bab85c | 1844 | cleanup_unknown_header_ext(bs); |
ed6ccf0f KW |
1845 | qcow2_free_snapshots(bs); |
1846 | qcow2_refcount_close(bs); | |
de82815d | 1847 | qemu_vfree(s->l1_table); |
cf93980e HR |
1848 | /* else pre-write overlap checks in cache_destroy may crash */ |
1849 | s->l1_table = NULL; | |
279621c0 | 1850 | cache_clean_timer_del(bs); |
29c1a730 | 1851 | if (s->l2_table_cache) { |
e64d4072 | 1852 | qcow2_cache_destroy(s->l2_table_cache); |
29c1a730 | 1853 | } |
c5a33ee9 | 1854 | if (s->refcount_block_cache) { |
e64d4072 | 1855 | qcow2_cache_destroy(s->refcount_block_cache); |
c5a33ee9 | 1856 | } |
b25b387f DB |
1857 | qcrypto_block_free(s->crypto); |
1858 | qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); | |
6d85a57e | 1859 | return ret; |
585f8587 FB |
1860 | } |
1861 | ||
1fafcd93 PB |
1862 | typedef struct QCow2OpenCo { |
1863 | BlockDriverState *bs; | |
1864 | QDict *options; | |
1865 | int flags; | |
1866 | Error **errp; | |
1867 | int ret; | |
1868 | } QCow2OpenCo; | |
1869 | ||
1870 | static void coroutine_fn qcow2_open_entry(void *opaque) | |
1871 | { | |
1872 | QCow2OpenCo *qoc = opaque; | |
1873 | BDRVQcow2State *s = qoc->bs->opaque; | |
1874 | ||
1875 | qemu_co_mutex_lock(&s->lock); | |
1876 | qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp); | |
1877 | qemu_co_mutex_unlock(&s->lock); | |
1878 | } | |
1879 | ||
4e4bf5c4 KW |
1880 | static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, |
1881 | Error **errp) | |
1882 | { | |
1fafcd93 PB |
1883 | BDRVQcow2State *s = bs->opaque; |
1884 | QCow2OpenCo qoc = { | |
1885 | .bs = bs, | |
1886 | .options = options, | |
1887 | .flags = flags, | |
1888 | .errp = errp, | |
1889 | .ret = -EINPROGRESS | |
1890 | }; | |
1891 | ||
8b1869da HR |
1892 | bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds, |
1893 | BDRV_CHILD_IMAGE, false, errp); | |
4e4bf5c4 KW |
1894 | if (!bs->file) { |
1895 | return -EINVAL; | |
1896 | } | |
1897 | ||
1fafcd93 PB |
1898 | /* Initialise locks */ |
1899 | qemu_co_mutex_init(&s->lock); | |
1900 | ||
1901 | if (qemu_in_coroutine()) { | |
1902 | /* From bdrv_co_create. */ | |
1903 | qcow2_open_entry(&qoc); | |
1904 | } else { | |
4720cbee | 1905 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
1fafcd93 PB |
1906 | qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc)); |
1907 | BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS); | |
1908 | } | |
1909 | return qoc.ret; | |
4e4bf5c4 KW |
1910 | } |
1911 | ||
3baca891 | 1912 | static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp) |
d34682cd | 1913 | { |
ff99129a | 1914 | BDRVQcow2State *s = bs->opaque; |
d34682cd | 1915 | |
a84178cc EB |
1916 | if (bs->encrypted) { |
1917 | /* Encryption works on a sector granularity */ | |
6f8f015c | 1918 | bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto); |
a84178cc | 1919 | } |
a6841a2d | 1920 | bs->bl.pwrite_zeroes_alignment = s->subcluster_size; |
ecdbead6 | 1921 | bs->bl.pdiscard_alignment = s->cluster_size; |
d34682cd KW |
1922 | } |
1923 | ||
21d82ac9 JC |
1924 | static int qcow2_reopen_prepare(BDRVReopenState *state, |
1925 | BlockReopenQueue *queue, Error **errp) | |
1926 | { | |
5b0959a7 | 1927 | Qcow2ReopenState *r; |
4c2e5f8f KW |
1928 | int ret; |
1929 | ||
5b0959a7 KW |
1930 | r = g_new0(Qcow2ReopenState, 1); |
1931 | state->opaque = r; | |
1932 | ||
1933 | ret = qcow2_update_options_prepare(state->bs, r, state->options, | |
1934 | state->flags, errp); | |
1935 | if (ret < 0) { | |
1936 | goto fail; | |
1937 | } | |
1938 | ||
1939 | /* We need to write out any unwritten data if we reopen read-only. */ | |
4c2e5f8f | 1940 | if ((state->flags & BDRV_O_RDWR) == 0) { |
169b8793 VSO |
1941 | ret = qcow2_reopen_bitmaps_ro(state->bs, errp); |
1942 | if (ret < 0) { | |
1943 | goto fail; | |
1944 | } | |
1945 | ||
4c2e5f8f KW |
1946 | ret = bdrv_flush(state->bs); |
1947 | if (ret < 0) { | |
5b0959a7 | 1948 | goto fail; |
4c2e5f8f KW |
1949 | } |
1950 | ||
1951 | ret = qcow2_mark_clean(state->bs); | |
1952 | if (ret < 0) { | |
5b0959a7 | 1953 | goto fail; |
4c2e5f8f KW |
1954 | } |
1955 | } | |
1956 | ||
21d82ac9 | 1957 | return 0; |
5b0959a7 KW |
1958 | |
1959 | fail: | |
1960 | qcow2_update_options_abort(state->bs, r); | |
1961 | g_free(r); | |
1962 | return ret; | |
1963 | } | |
1964 | ||
1965 | static void qcow2_reopen_commit(BDRVReopenState *state) | |
1966 | { | |
1967 | qcow2_update_options_commit(state->bs, state->opaque); | |
65eb7c85 PK |
1968 | g_free(state->opaque); |
1969 | } | |
1970 | ||
1971 | static void qcow2_reopen_commit_post(BDRVReopenState *state) | |
1972 | { | |
4dd09f62 VSO |
1973 | if (state->flags & BDRV_O_RDWR) { |
1974 | Error *local_err = NULL; | |
1975 | ||
1976 | if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) { | |
1977 | /* | |
1978 | * This is not fatal, bitmaps just left read-only, so all following | |
1979 | * writes will fail. User can remove read-only bitmaps to unblock | |
1980 | * writes or retry reopen. | |
1981 | */ | |
1982 | error_reportf_err(local_err, | |
1983 | "%s: Failed to make dirty bitmaps writable: ", | |
1984 | bdrv_get_node_name(state->bs)); | |
1985 | } | |
1986 | } | |
5b0959a7 KW |
1987 | } |
1988 | ||
1989 | static void qcow2_reopen_abort(BDRVReopenState *state) | |
1990 | { | |
1991 | qcow2_update_options_abort(state->bs, state->opaque); | |
1992 | g_free(state->opaque); | |
21d82ac9 JC |
1993 | } |
1994 | ||
5365f44d KW |
1995 | static void qcow2_join_options(QDict *options, QDict *old_options) |
1996 | { | |
1997 | bool has_new_overlap_template = | |
1998 | qdict_haskey(options, QCOW2_OPT_OVERLAP) || | |
1999 | qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE); | |
2000 | bool has_new_total_cache_size = | |
2001 | qdict_haskey(options, QCOW2_OPT_CACHE_SIZE); | |
2002 | bool has_all_cache_options; | |
2003 | ||
2004 | /* New overlap template overrides all old overlap options */ | |
2005 | if (has_new_overlap_template) { | |
2006 | qdict_del(old_options, QCOW2_OPT_OVERLAP); | |
2007 | qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE); | |
2008 | qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER); | |
2009 | qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1); | |
2010 | qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2); | |
2011 | qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE); | |
2012 | qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK); | |
2013 | qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE); | |
2014 | qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1); | |
2015 | qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2); | |
2016 | } | |
2017 | ||
2018 | /* New total cache size overrides all old options */ | |
2019 | if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) { | |
2020 | qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE); | |
2021 | qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); | |
2022 | } | |
2023 | ||
2024 | qdict_join(options, old_options, false); | |
2025 | ||
2026 | /* | |
2027 | * If after merging all cache size options are set, an old total size is | |
2028 | * overwritten. Do keep all options, however, if all three are new. The | |
2029 | * resulting error message is what we want to happen. | |
2030 | */ | |
2031 | has_all_cache_options = | |
2032 | qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) || | |
2033 | qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) || | |
2034 | qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE); | |
2035 | ||
2036 | if (has_all_cache_options && !has_new_total_cache_size) { | |
2037 | qdict_del(options, QCOW2_OPT_CACHE_SIZE); | |
2038 | } | |
2039 | } | |
2040 | ||
a320fb04 EB |
2041 | static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs, |
2042 | bool want_zero, | |
2043 | int64_t offset, int64_t count, | |
2044 | int64_t *pnum, int64_t *map, | |
2045 | BlockDriverState **file) | |
585f8587 | 2046 | { |
ff99129a | 2047 | BDRVQcow2State *s = bs->opaque; |
388e5816 | 2048 | uint64_t host_offset; |
ecfe1863 | 2049 | unsigned int bytes; |
10dabdc5 | 2050 | QCow2SubclusterType type; |
74e60fb5 | 2051 | int ret, status = 0; |
585f8587 | 2052 | |
5e978550 KW |
2053 | qemu_co_mutex_lock(&s->lock); |
2054 | ||
69f47505 VSO |
2055 | if (!s->metadata_preallocation_checked) { |
2056 | ret = qcow2_detect_metadata_preallocation(bs); | |
2057 | s->metadata_preallocation = (ret == 1); | |
2058 | s->metadata_preallocation_checked = true; | |
2059 | } | |
2060 | ||
a320fb04 | 2061 | bytes = MIN(INT_MAX, count); |
ca4a0bb8 | 2062 | ret = qcow2_get_host_offset(bs, offset, &bytes, &host_offset, &type); |
f8a2e5e3 | 2063 | qemu_co_mutex_unlock(&s->lock); |
1c46efaa | 2064 | if (ret < 0) { |
d663640c | 2065 | return ret; |
1c46efaa | 2066 | } |
095a9c58 | 2067 | |
a320fb04 | 2068 | *pnum = bytes; |
ecfe1863 | 2069 | |
10dabdc5 | 2070 | if ((type == QCOW2_SUBCLUSTER_NORMAL || |
97490a14 AG |
2071 | type == QCOW2_SUBCLUSTER_ZERO_ALLOC || |
2072 | type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) && !s->crypto) { | |
388e5816 | 2073 | *map = host_offset; |
37be1403 | 2074 | *file = s->data_file->bs; |
a320fb04 | 2075 | status |= BDRV_BLOCK_OFFSET_VALID; |
4bc74be9 | 2076 | } |
10dabdc5 AG |
2077 | if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN || |
2078 | type == QCOW2_SUBCLUSTER_ZERO_ALLOC) { | |
4bc74be9 | 2079 | status |= BDRV_BLOCK_ZERO; |
97490a14 AG |
2080 | } else if (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && |
2081 | type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) { | |
4bc74be9 PB |
2082 | status |= BDRV_BLOCK_DATA; |
2083 | } | |
69f47505 VSO |
2084 | if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) && |
2085 | (status & BDRV_BLOCK_OFFSET_VALID)) | |
2086 | { | |
2087 | status |= BDRV_BLOCK_RECURSE; | |
2088 | } | |
4bc74be9 | 2089 | return status; |
585f8587 FB |
2090 | } |
2091 | ||
fd9fcd37 FZ |
2092 | static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs, |
2093 | QCowL2Meta **pl2meta, | |
2094 | bool link_l2) | |
2095 | { | |
2096 | int ret = 0; | |
2097 | QCowL2Meta *l2meta = *pl2meta; | |
2098 | ||
2099 | while (l2meta != NULL) { | |
2100 | QCowL2Meta *next; | |
2101 | ||
354d930d | 2102 | if (link_l2) { |
fd9fcd37 FZ |
2103 | ret = qcow2_alloc_cluster_link_l2(bs, l2meta); |
2104 | if (ret) { | |
2105 | goto out; | |
2106 | } | |
8b24cd14 KW |
2107 | } else { |
2108 | qcow2_alloc_cluster_abort(bs, l2meta); | |
fd9fcd37 FZ |
2109 | } |
2110 | ||
2111 | /* Take the request off the list of running requests */ | |
f7bd5bba | 2112 | QLIST_REMOVE(l2meta, next_in_flight); |
fd9fcd37 FZ |
2113 | |
2114 | qemu_co_queue_restart_all(&l2meta->dependent_requests); | |
2115 | ||
2116 | next = l2meta->next; | |
2117 | g_free(l2meta); | |
2118 | l2meta = next; | |
2119 | } | |
2120 | out: | |
2121 | *pl2meta = l2meta; | |
2122 | return ret; | |
2123 | } | |
2124 | ||
88f468e5 VSO |
2125 | static coroutine_fn int |
2126 | qcow2_co_preadv_encrypted(BlockDriverState *bs, | |
9c4269d5 | 2127 | uint64_t host_offset, |
88f468e5 VSO |
2128 | uint64_t offset, |
2129 | uint64_t bytes, | |
2130 | QEMUIOVector *qiov, | |
2131 | uint64_t qiov_offset) | |
2132 | { | |
2133 | int ret; | |
2134 | BDRVQcow2State *s = bs->opaque; | |
2135 | uint8_t *buf; | |
2136 | ||
2137 | assert(bs->encrypted && s->crypto); | |
2138 | assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); | |
2139 | ||
2140 | /* | |
2141 | * For encrypted images, read everything into a temporary | |
2142 | * contiguous buffer on which the AES functions can work. | |
2143 | * Also, decryption in a separate buffer is better as it | |
2144 | * prevents the guest from learning information about the | |
2145 | * encrypted nature of the virtual disk. | |
2146 | */ | |
2147 | ||
2148 | buf = qemu_try_blockalign(s->data_file->bs, bytes); | |
2149 | if (buf == NULL) { | |
2150 | return -ENOMEM; | |
2151 | } | |
2152 | ||
2153 | BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); | |
9c4269d5 | 2154 | ret = bdrv_co_pread(s->data_file, host_offset, bytes, buf, 0); |
88f468e5 VSO |
2155 | if (ret < 0) { |
2156 | goto fail; | |
2157 | } | |
2158 | ||
9c4269d5 | 2159 | if (qcow2_co_decrypt(bs, host_offset, offset, buf, bytes) < 0) |
88f468e5 VSO |
2160 | { |
2161 | ret = -EIO; | |
2162 | goto fail; | |
2163 | } | |
2164 | qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes); | |
2165 | ||
2166 | fail: | |
2167 | qemu_vfree(buf); | |
2168 | ||
2169 | return ret; | |
2170 | } | |
2171 | ||
d710cf57 VSO |
2172 | typedef struct Qcow2AioTask { |
2173 | AioTask task; | |
2174 | ||
2175 | BlockDriverState *bs; | |
10dabdc5 | 2176 | QCow2SubclusterType subcluster_type; /* only for read */ |
9c4269d5 | 2177 | uint64_t host_offset; /* or full descriptor in compressed clusters */ |
d710cf57 VSO |
2178 | uint64_t offset; |
2179 | uint64_t bytes; | |
2180 | QEMUIOVector *qiov; | |
2181 | uint64_t qiov_offset; | |
2182 | QCowL2Meta *l2meta; /* only for write */ | |
2183 | } Qcow2AioTask; | |
2184 | ||
2185 | static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task); | |
2186 | static coroutine_fn int qcow2_add_task(BlockDriverState *bs, | |
2187 | AioTaskPool *pool, | |
2188 | AioTaskFunc func, | |
10dabdc5 | 2189 | QCow2SubclusterType subcluster_type, |
9c4269d5 | 2190 | uint64_t host_offset, |
d710cf57 VSO |
2191 | uint64_t offset, |
2192 | uint64_t bytes, | |
2193 | QEMUIOVector *qiov, | |
2194 | size_t qiov_offset, | |
2195 | QCowL2Meta *l2meta) | |
2196 | { | |
2197 | Qcow2AioTask local_task; | |
2198 | Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task; | |
2199 | ||
2200 | *task = (Qcow2AioTask) { | |
2201 | .task.func = func, | |
2202 | .bs = bs, | |
10dabdc5 | 2203 | .subcluster_type = subcluster_type, |
d710cf57 | 2204 | .qiov = qiov, |
9c4269d5 | 2205 | .host_offset = host_offset, |
d710cf57 VSO |
2206 | .offset = offset, |
2207 | .bytes = bytes, | |
2208 | .qiov_offset = qiov_offset, | |
2209 | .l2meta = l2meta, | |
2210 | }; | |
2211 | ||
2212 | trace_qcow2_add_task(qemu_coroutine_self(), bs, pool, | |
2213 | func == qcow2_co_preadv_task_entry ? "read" : "write", | |
10dabdc5 | 2214 | subcluster_type, host_offset, offset, bytes, |
d710cf57 VSO |
2215 | qiov, qiov_offset); |
2216 | ||
2217 | if (!pool) { | |
2218 | return func(&task->task); | |
2219 | } | |
2220 | ||
2221 | aio_task_pool_start_task(pool, &task->task); | |
2222 | ||
2223 | return 0; | |
2224 | } | |
2225 | ||
88f468e5 | 2226 | static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs, |
10dabdc5 | 2227 | QCow2SubclusterType subc_type, |
9c4269d5 | 2228 | uint64_t host_offset, |
88f468e5 VSO |
2229 | uint64_t offset, uint64_t bytes, |
2230 | QEMUIOVector *qiov, | |
2231 | size_t qiov_offset) | |
2232 | { | |
2233 | BDRVQcow2State *s = bs->opaque; | |
88f468e5 | 2234 | |
10dabdc5 AG |
2235 | switch (subc_type) { |
2236 | case QCOW2_SUBCLUSTER_ZERO_PLAIN: | |
2237 | case QCOW2_SUBCLUSTER_ZERO_ALLOC: | |
88f468e5 VSO |
2238 | /* Both zero types are handled in qcow2_co_preadv_part */ |
2239 | g_assert_not_reached(); | |
2240 | ||
10dabdc5 | 2241 | case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: |
97490a14 | 2242 | case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: |
88f468e5 VSO |
2243 | assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */ |
2244 | ||
2245 | BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); | |
2246 | return bdrv_co_preadv_part(bs->backing, offset, bytes, | |
2247 | qiov, qiov_offset, 0); | |
2248 | ||
10dabdc5 | 2249 | case QCOW2_SUBCLUSTER_COMPRESSED: |
9c4269d5 | 2250 | return qcow2_co_preadv_compressed(bs, host_offset, |
88f468e5 VSO |
2251 | offset, bytes, qiov, qiov_offset); |
2252 | ||
10dabdc5 | 2253 | case QCOW2_SUBCLUSTER_NORMAL: |
88f468e5 | 2254 | if (bs->encrypted) { |
9c4269d5 | 2255 | return qcow2_co_preadv_encrypted(bs, host_offset, |
88f468e5 VSO |
2256 | offset, bytes, qiov, qiov_offset); |
2257 | } | |
2258 | ||
2259 | BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); | |
9c4269d5 | 2260 | return bdrv_co_preadv_part(s->data_file, host_offset, |
88f468e5 VSO |
2261 | bytes, qiov, qiov_offset, 0); |
2262 | ||
2263 | default: | |
2264 | g_assert_not_reached(); | |
2265 | } | |
2266 | ||
2267 | g_assert_not_reached(); | |
2268 | } | |
2269 | ||
d710cf57 VSO |
2270 | static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task) |
2271 | { | |
2272 | Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); | |
2273 | ||
2274 | assert(!t->l2meta); | |
2275 | ||
10dabdc5 AG |
2276 | return qcow2_co_preadv_task(t->bs, t->subcluster_type, |
2277 | t->host_offset, t->offset, t->bytes, | |
2278 | t->qiov, t->qiov_offset); | |
d710cf57 VSO |
2279 | } |
2280 | ||
df893d25 VSO |
2281 | static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs, |
2282 | uint64_t offset, uint64_t bytes, | |
2283 | QEMUIOVector *qiov, | |
2284 | size_t qiov_offset, int flags) | |
585f8587 | 2285 | { |
ff99129a | 2286 | BDRVQcow2State *s = bs->opaque; |
d710cf57 | 2287 | int ret = 0; |
ecfe1863 | 2288 | unsigned int cur_bytes; /* number of bytes in current iteration */ |
388e5816 | 2289 | uint64_t host_offset = 0; |
10dabdc5 | 2290 | QCow2SubclusterType type; |
d710cf57 | 2291 | AioTaskPool *aio = NULL; |
585f8587 | 2292 | |
d710cf57 | 2293 | while (bytes != 0 && aio_task_pool_status(aio) == 0) { |
5ebaa27e | 2294 | /* prepare next request */ |
ecfe1863 | 2295 | cur_bytes = MIN(bytes, INT_MAX); |
b25b387f | 2296 | if (s->crypto) { |
ecfe1863 KW |
2297 | cur_bytes = MIN(cur_bytes, |
2298 | QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); | |
585f8587 | 2299 | } |
5ebaa27e | 2300 | |
f24196d3 | 2301 | qemu_co_mutex_lock(&s->lock); |
ca4a0bb8 AG |
2302 | ret = qcow2_get_host_offset(bs, offset, &cur_bytes, |
2303 | &host_offset, &type); | |
f24196d3 | 2304 | qemu_co_mutex_unlock(&s->lock); |
8af36488 | 2305 | if (ret < 0) { |
d710cf57 | 2306 | goto out; |
8af36488 | 2307 | } |
bd28f835 | 2308 | |
10dabdc5 AG |
2309 | if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN || |
2310 | type == QCOW2_SUBCLUSTER_ZERO_ALLOC || | |
97490a14 AG |
2311 | (type == QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && !bs->backing) || |
2312 | (type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && !bs->backing)) | |
88f468e5 | 2313 | { |
df893d25 | 2314 | qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes); |
88f468e5 | 2315 | } else { |
d710cf57 VSO |
2316 | if (!aio && cur_bytes != bytes) { |
2317 | aio = aio_task_pool_new(QCOW2_MAX_WORKERS); | |
2318 | } | |
ca4a0bb8 | 2319 | ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, type, |
9c4269d5 | 2320 | host_offset, offset, cur_bytes, |
d710cf57 | 2321 | qiov, qiov_offset, NULL); |
5ebaa27e | 2322 | if (ret < 0) { |
d710cf57 | 2323 | goto out; |
5ebaa27e | 2324 | } |
faf575c1 | 2325 | } |
f141eafe | 2326 | |
ecfe1863 KW |
2327 | bytes -= cur_bytes; |
2328 | offset += cur_bytes; | |
df893d25 | 2329 | qiov_offset += cur_bytes; |
5ebaa27e | 2330 | } |
68d100e9 | 2331 | |
d710cf57 VSO |
2332 | out: |
2333 | if (aio) { | |
2334 | aio_task_pool_wait_all(aio); | |
2335 | if (ret == 0) { | |
2336 | ret = aio_task_pool_status(aio); | |
2337 | } | |
2338 | g_free(aio); | |
2339 | } | |
2340 | ||
2341 | return ret; | |
585f8587 FB |
2342 | } |
2343 | ||
ee22a9d8 AG |
2344 | /* Check if it's possible to merge a write request with the writing of |
2345 | * the data from the COW regions */ | |
2346 | static bool merge_cow(uint64_t offset, unsigned bytes, | |
5396234b VSO |
2347 | QEMUIOVector *qiov, size_t qiov_offset, |
2348 | QCowL2Meta *l2meta) | |
ee22a9d8 AG |
2349 | { |
2350 | QCowL2Meta *m; | |
2351 | ||
2352 | for (m = l2meta; m != NULL; m = m->next) { | |
2353 | /* If both COW regions are empty then there's nothing to merge */ | |
2354 | if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) { | |
2355 | continue; | |
2356 | } | |
2357 | ||
c8bb23cb AN |
2358 | /* If COW regions are handled already, skip this too */ |
2359 | if (m->skip_cow) { | |
2360 | continue; | |
2361 | } | |
2362 | ||
3441ad4b AG |
2363 | /* |
2364 | * The write request should start immediately after the first | |
2365 | * COW region. This does not always happen because the area | |
2366 | * touched by the request can be larger than the one defined | |
2367 | * by @m (a single request can span an area consisting of a | |
2368 | * mix of previously unallocated and allocated clusters, that | |
2369 | * is why @l2meta is a list). | |
2370 | */ | |
ee22a9d8 | 2371 | if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) { |
3441ad4b AG |
2372 | /* In this case the request starts before this region */ |
2373 | assert(offset < l2meta_cow_start(m)); | |
2374 | assert(m->cow_start.nb_bytes == 0); | |
ee22a9d8 AG |
2375 | continue; |
2376 | } | |
2377 | ||
3441ad4b AG |
2378 | /* The write request should end immediately before the second |
2379 | * COW region (see above for why it does not always happen) */ | |
ee22a9d8 | 2380 | if (m->offset + m->cow_end.offset != offset + bytes) { |
3441ad4b AG |
2381 | assert(offset + bytes > m->offset + m->cow_end.offset); |
2382 | assert(m->cow_end.nb_bytes == 0); | |
ee22a9d8 AG |
2383 | continue; |
2384 | } | |
2385 | ||
2386 | /* Make sure that adding both COW regions to the QEMUIOVector | |
2387 | * does not exceed IOV_MAX */ | |
5396234b | 2388 | if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) { |
ee22a9d8 AG |
2389 | continue; |
2390 | } | |
2391 | ||
5396234b VSO |
2392 | m->data_qiov = qiov; |
2393 | m->data_qiov_offset = qiov_offset; | |
ee22a9d8 AG |
2394 | return true; |
2395 | } | |
2396 | ||
2397 | return false; | |
2398 | } | |
2399 | ||
46cd1e8a AG |
2400 | /* |
2401 | * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error. | |
2402 | * Note that returning 0 does not guarantee non-zero data. | |
2403 | */ | |
2404 | static int is_zero_cow(BlockDriverState *bs, QCowL2Meta *m) | |
c8bb23cb AN |
2405 | { |
2406 | /* | |
2407 | * This check is designed for optimization shortcut so it must be | |
2408 | * efficient. | |
46cd1e8a AG |
2409 | * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is |
2410 | * faster (but not as accurate and can result in false negatives). | |
c8bb23cb | 2411 | */ |
46cd1e8a AG |
2412 | int ret = bdrv_co_is_zero_fast(bs, m->offset + m->cow_start.offset, |
2413 | m->cow_start.nb_bytes); | |
2414 | if (ret <= 0) { | |
2415 | return ret; | |
2416 | } | |
2417 | ||
2418 | return bdrv_co_is_zero_fast(bs, m->offset + m->cow_end.offset, | |
2419 | m->cow_end.nb_bytes); | |
c8bb23cb AN |
2420 | } |
2421 | ||
2422 | static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta) | |
2423 | { | |
2424 | BDRVQcow2State *s = bs->opaque; | |
2425 | QCowL2Meta *m; | |
2426 | ||
2427 | if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) { | |
2428 | return 0; | |
2429 | } | |
2430 | ||
2431 | if (bs->encrypted) { | |
2432 | return 0; | |
2433 | } | |
2434 | ||
2435 | for (m = l2meta; m != NULL; m = m->next) { | |
2436 | int ret; | |
bf4a66ee AG |
2437 | uint64_t start_offset = m->alloc_offset + m->cow_start.offset; |
2438 | unsigned nb_bytes = m->cow_end.offset + m->cow_end.nb_bytes - | |
2439 | m->cow_start.offset; | |
c8bb23cb AN |
2440 | |
2441 | if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) { | |
2442 | continue; | |
2443 | } | |
2444 | ||
46cd1e8a AG |
2445 | ret = is_zero_cow(bs, m); |
2446 | if (ret < 0) { | |
2447 | return ret; | |
2448 | } else if (ret == 0) { | |
c8bb23cb AN |
2449 | continue; |
2450 | } | |
2451 | ||
2452 | /* | |
2453 | * instead of writing zero COW buffers, | |
2454 | * efficiently zero out the whole clusters | |
2455 | */ | |
2456 | ||
bf4a66ee | 2457 | ret = qcow2_pre_write_overlap_check(bs, 0, start_offset, nb_bytes, |
c8bb23cb AN |
2458 | true); |
2459 | if (ret < 0) { | |
2460 | return ret; | |
2461 | } | |
2462 | ||
2463 | BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE); | |
bf4a66ee | 2464 | ret = bdrv_co_pwrite_zeroes(s->data_file, start_offset, nb_bytes, |
c8bb23cb AN |
2465 | BDRV_REQ_NO_FALLBACK); |
2466 | if (ret < 0) { | |
2467 | if (ret != -ENOTSUP && ret != -EAGAIN) { | |
2468 | return ret; | |
2469 | } | |
2470 | continue; | |
2471 | } | |
2472 | ||
2473 | trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters); | |
2474 | m->skip_cow = true; | |
2475 | } | |
2476 | return 0; | |
2477 | } | |
2478 | ||
6aa7a263 VSO |
2479 | /* |
2480 | * qcow2_co_pwritev_task | |
2481 | * Called with s->lock unlocked | |
2482 | * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must | |
2483 | * not use it somehow after qcow2_co_pwritev_task() call | |
2484 | */ | |
2485 | static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs, | |
9c4269d5 | 2486 | uint64_t host_offset, |
6aa7a263 VSO |
2487 | uint64_t offset, uint64_t bytes, |
2488 | QEMUIOVector *qiov, | |
2489 | uint64_t qiov_offset, | |
2490 | QCowL2Meta *l2meta) | |
2491 | { | |
2492 | int ret; | |
2493 | BDRVQcow2State *s = bs->opaque; | |
2494 | void *crypt_buf = NULL; | |
6aa7a263 VSO |
2495 | QEMUIOVector encrypted_qiov; |
2496 | ||
2497 | if (bs->encrypted) { | |
2498 | assert(s->crypto); | |
2499 | assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); | |
2500 | crypt_buf = qemu_try_blockalign(bs->file->bs, bytes); | |
2501 | if (crypt_buf == NULL) { | |
2502 | ret = -ENOMEM; | |
2503 | goto out_unlocked; | |
2504 | } | |
2505 | qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes); | |
2506 | ||
9c4269d5 | 2507 | if (qcow2_co_encrypt(bs, host_offset, offset, crypt_buf, bytes) < 0) { |
6aa7a263 VSO |
2508 | ret = -EIO; |
2509 | goto out_unlocked; | |
2510 | } | |
2511 | ||
2512 | qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes); | |
2513 | qiov = &encrypted_qiov; | |
2514 | qiov_offset = 0; | |
2515 | } | |
2516 | ||
2517 | /* Try to efficiently initialize the physical space with zeroes */ | |
2518 | ret = handle_alloc_space(bs, l2meta); | |
2519 | if (ret < 0) { | |
2520 | goto out_unlocked; | |
2521 | } | |
2522 | ||
2523 | /* | |
2524 | * If we need to do COW, check if it's possible to merge the | |
2525 | * writing of the guest data together with that of the COW regions. | |
2526 | * If it's not possible (or not necessary) then write the | |
2527 | * guest data now. | |
2528 | */ | |
2529 | if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) { | |
2530 | BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); | |
9c4269d5 AG |
2531 | trace_qcow2_writev_data(qemu_coroutine_self(), host_offset); |
2532 | ret = bdrv_co_pwritev_part(s->data_file, host_offset, | |
6aa7a263 VSO |
2533 | bytes, qiov, qiov_offset, 0); |
2534 | if (ret < 0) { | |
2535 | goto out_unlocked; | |
2536 | } | |
2537 | } | |
2538 | ||
2539 | qemu_co_mutex_lock(&s->lock); | |
2540 | ||
2541 | ret = qcow2_handle_l2meta(bs, &l2meta, true); | |
2542 | goto out_locked; | |
2543 | ||
2544 | out_unlocked: | |
2545 | qemu_co_mutex_lock(&s->lock); | |
2546 | ||
2547 | out_locked: | |
2548 | qcow2_handle_l2meta(bs, &l2meta, false); | |
2549 | qemu_co_mutex_unlock(&s->lock); | |
2550 | ||
2551 | qemu_vfree(crypt_buf); | |
2552 | ||
2553 | return ret; | |
2554 | } | |
2555 | ||
d710cf57 VSO |
2556 | static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task) |
2557 | { | |
2558 | Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); | |
2559 | ||
10dabdc5 | 2560 | assert(!t->subcluster_type); |
d710cf57 | 2561 | |
9c4269d5 | 2562 | return qcow2_co_pwritev_task(t->bs, t->host_offset, |
d710cf57 VSO |
2563 | t->offset, t->bytes, t->qiov, t->qiov_offset, |
2564 | t->l2meta); | |
2565 | } | |
2566 | ||
5396234b VSO |
2567 | static coroutine_fn int qcow2_co_pwritev_part( |
2568 | BlockDriverState *bs, uint64_t offset, uint64_t bytes, | |
2569 | QEMUIOVector *qiov, size_t qiov_offset, int flags) | |
585f8587 | 2570 | { |
ff99129a | 2571 | BDRVQcow2State *s = bs->opaque; |
d46a0bb2 | 2572 | int offset_in_cluster; |
68d100e9 | 2573 | int ret; |
d46a0bb2 | 2574 | unsigned int cur_bytes; /* number of sectors in current iteration */ |
bfd0989a | 2575 | uint64_t host_offset; |
8d2497c3 | 2576 | QCowL2Meta *l2meta = NULL; |
d710cf57 | 2577 | AioTaskPool *aio = NULL; |
c2271403 | 2578 | |
d46a0bb2 | 2579 | trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes); |
3cce16f4 | 2580 | |
d710cf57 | 2581 | while (bytes != 0 && aio_task_pool_status(aio) == 0) { |
3fc48d09 | 2582 | |
f50f88b9 | 2583 | l2meta = NULL; |
cf5c1a23 | 2584 | |
3cce16f4 | 2585 | trace_qcow2_writev_start_part(qemu_coroutine_self()); |
d46a0bb2 KW |
2586 | offset_in_cluster = offset_into_cluster(s, offset); |
2587 | cur_bytes = MIN(bytes, INT_MAX); | |
2588 | if (bs->encrypted) { | |
2589 | cur_bytes = MIN(cur_bytes, | |
2590 | QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size | |
2591 | - offset_in_cluster); | |
5ebaa27e | 2592 | } |
095a9c58 | 2593 | |
6aa7a263 VSO |
2594 | qemu_co_mutex_lock(&s->lock); |
2595 | ||
bfd0989a AG |
2596 | ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes, |
2597 | &host_offset, &l2meta); | |
5ebaa27e | 2598 | if (ret < 0) { |
5447c3a0 | 2599 | goto out_locked; |
5ebaa27e | 2600 | } |
148da7ea | 2601 | |
bfd0989a | 2602 | ret = qcow2_pre_write_overlap_check(bs, 0, host_offset, |
5447c3a0 VSO |
2603 | cur_bytes, true); |
2604 | if (ret < 0) { | |
2605 | goto out_locked; | |
2606 | } | |
2607 | ||
2608 | qemu_co_mutex_unlock(&s->lock); | |
2609 | ||
d710cf57 VSO |
2610 | if (!aio && cur_bytes != bytes) { |
2611 | aio = aio_task_pool_new(QCOW2_MAX_WORKERS); | |
2612 | } | |
2613 | ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0, | |
bfd0989a | 2614 | host_offset, offset, |
9c4269d5 | 2615 | cur_bytes, qiov, qiov_offset, l2meta); |
6aa7a263 | 2616 | l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */ |
c8bb23cb | 2617 | if (ret < 0) { |
6aa7a263 | 2618 | goto fail_nometa; |
f50f88b9 | 2619 | } |
0fa9131a | 2620 | |
d46a0bb2 KW |
2621 | bytes -= cur_bytes; |
2622 | offset += cur_bytes; | |
6aa7a263 | 2623 | qiov_offset += cur_bytes; |
d46a0bb2 | 2624 | trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes); |
5ebaa27e | 2625 | } |
3fc48d09 | 2626 | ret = 0; |
faf575c1 | 2627 | |
5447c3a0 VSO |
2628 | qemu_co_mutex_lock(&s->lock); |
2629 | ||
2630 | out_locked: | |
fd9fcd37 | 2631 | qcow2_handle_l2meta(bs, &l2meta, false); |
0fa9131a | 2632 | |
a8c57408 PB |
2633 | qemu_co_mutex_unlock(&s->lock); |
2634 | ||
6aa7a263 | 2635 | fail_nometa: |
d710cf57 VSO |
2636 | if (aio) { |
2637 | aio_task_pool_wait_all(aio); | |
2638 | if (ret == 0) { | |
2639 | ret = aio_task_pool_status(aio); | |
2640 | } | |
2641 | g_free(aio); | |
2642 | } | |
2643 | ||
3cce16f4 | 2644 | trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); |
42496d62 | 2645 | |
68d100e9 | 2646 | return ret; |
585f8587 FB |
2647 | } |
2648 | ||
ec6d8912 KW |
2649 | static int qcow2_inactivate(BlockDriverState *bs) |
2650 | { | |
2651 | BDRVQcow2State *s = bs->opaque; | |
2652 | int ret, result = 0; | |
5f72826e | 2653 | Error *local_err = NULL; |
ec6d8912 | 2654 | |
644ddbb7 | 2655 | qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err); |
83a8c775 PB |
2656 | if (local_err != NULL) { |
2657 | result = -EINVAL; | |
132adb68 VSO |
2658 | error_reportf_err(local_err, "Lost persistent bitmaps during " |
2659 | "inactivation of node '%s': ", | |
2660 | bdrv_get_device_or_node_name(bs)); | |
83a8c775 PB |
2661 | } |
2662 | ||
ec6d8912 KW |
2663 | ret = qcow2_cache_flush(bs, s->l2_table_cache); |
2664 | if (ret) { | |
2665 | result = ret; | |
2666 | error_report("Failed to flush the L2 table cache: %s", | |
2667 | strerror(-ret)); | |
2668 | } | |
2669 | ||
2670 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
2671 | if (ret) { | |
2672 | result = ret; | |
2673 | error_report("Failed to flush the refcount block cache: %s", | |
2674 | strerror(-ret)); | |
2675 | } | |
2676 | ||
2677 | if (result == 0) { | |
2678 | qcow2_mark_clean(bs); | |
2679 | } | |
2680 | ||
2681 | return result; | |
2682 | } | |
2683 | ||
7c80ab3f | 2684 | static void qcow2_close(BlockDriverState *bs) |
585f8587 | 2685 | { |
ff99129a | 2686 | BDRVQcow2State *s = bs->opaque; |
de82815d | 2687 | qemu_vfree(s->l1_table); |
cf93980e HR |
2688 | /* else pre-write overlap checks in cache_destroy may crash */ |
2689 | s->l1_table = NULL; | |
29c1a730 | 2690 | |
140fd5a6 | 2691 | if (!(s->flags & BDRV_O_INACTIVE)) { |
ec6d8912 | 2692 | qcow2_inactivate(bs); |
27eb6c09 | 2693 | } |
c61d0004 | 2694 | |
279621c0 | 2695 | cache_clean_timer_del(bs); |
e64d4072 AG |
2696 | qcow2_cache_destroy(s->l2_table_cache); |
2697 | qcow2_cache_destroy(s->refcount_block_cache); | |
29c1a730 | 2698 | |
b25b387f DB |
2699 | qcrypto_block_free(s->crypto); |
2700 | s->crypto = NULL; | |
4aebf0f0 | 2701 | qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); |
f6fa64f6 | 2702 | |
6744cbab | 2703 | g_free(s->unknown_header_fields); |
75bab85c | 2704 | cleanup_unknown_header_ext(bs); |
6744cbab | 2705 | |
9b890bdc | 2706 | g_free(s->image_data_file); |
e4603fe1 KW |
2707 | g_free(s->image_backing_file); |
2708 | g_free(s->image_backing_format); | |
2709 | ||
0e8c08be KW |
2710 | if (has_data_file(bs)) { |
2711 | bdrv_unref_child(bs, s->data_file); | |
808cf3cb | 2712 | s->data_file = NULL; |
0e8c08be KW |
2713 | } |
2714 | ||
ed6ccf0f | 2715 | qcow2_refcount_close(bs); |
28c1202b | 2716 | qcow2_free_snapshots(bs); |
585f8587 FB |
2717 | } |
2718 | ||
2b148f39 PB |
2719 | static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs, |
2720 | Error **errp) | |
06d9260f | 2721 | { |
ff99129a | 2722 | BDRVQcow2State *s = bs->opaque; |
06d9260f | 2723 | int flags = s->flags; |
b25b387f | 2724 | QCryptoBlock *crypto = NULL; |
acdfb480 | 2725 | QDict *options; |
5a8a30db KW |
2726 | Error *local_err = NULL; |
2727 | int ret; | |
06d9260f AL |
2728 | |
2729 | /* | |
2730 | * Backing files are read-only which makes all of their metadata immutable, | |
2731 | * that means we don't have to worry about reopening them here. | |
2732 | */ | |
2733 | ||
b25b387f DB |
2734 | crypto = s->crypto; |
2735 | s->crypto = NULL; | |
06d9260f AL |
2736 | |
2737 | qcow2_close(bs); | |
2738 | ||
ff99129a | 2739 | memset(s, 0, sizeof(BDRVQcow2State)); |
d475e5ac | 2740 | options = qdict_clone_shallow(bs->options); |
5a8a30db | 2741 | |
140fd5a6 | 2742 | flags &= ~BDRV_O_INACTIVE; |
2b148f39 | 2743 | qemu_co_mutex_lock(&s->lock); |
4e4bf5c4 | 2744 | ret = qcow2_do_open(bs, options, flags, &local_err); |
2b148f39 | 2745 | qemu_co_mutex_unlock(&s->lock); |
cb3e7f08 | 2746 | qobject_unref(options); |
5a8a30db | 2747 | if (local_err) { |
4b576648 MA |
2748 | error_propagate_prepend(errp, local_err, |
2749 | "Could not reopen qcow2 layer: "); | |
191fb11b | 2750 | bs->drv = NULL; |
5a8a30db KW |
2751 | return; |
2752 | } else if (ret < 0) { | |
2753 | error_setg_errno(errp, -ret, "Could not reopen qcow2 layer"); | |
191fb11b | 2754 | bs->drv = NULL; |
5a8a30db KW |
2755 | return; |
2756 | } | |
acdfb480 | 2757 | |
b25b387f | 2758 | s->crypto = crypto; |
06d9260f AL |
2759 | } |
2760 | ||
e24e49e6 KW |
2761 | static size_t header_ext_add(char *buf, uint32_t magic, const void *s, |
2762 | size_t len, size_t buflen) | |
2763 | { | |
2764 | QCowExtension *ext_backing_fmt = (QCowExtension*) buf; | |
2765 | size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); | |
2766 | ||
2767 | if (buflen < ext_len) { | |
2768 | return -ENOSPC; | |
2769 | } | |
2770 | ||
2771 | *ext_backing_fmt = (QCowExtension) { | |
2772 | .magic = cpu_to_be32(magic), | |
2773 | .len = cpu_to_be32(len), | |
2774 | }; | |
0647d47c SH |
2775 | |
2776 | if (len) { | |
2777 | memcpy(buf + sizeof(QCowExtension), s, len); | |
2778 | } | |
e24e49e6 KW |
2779 | |
2780 | return ext_len; | |
2781 | } | |
2782 | ||
756e6736 | 2783 | /* |
e24e49e6 KW |
2784 | * Updates the qcow2 header, including the variable length parts of it, i.e. |
2785 | * the backing file name and all extensions. qcow2 was not designed to allow | |
2786 | * such changes, so if we run out of space (we can only use the first cluster) | |
2787 | * this function may fail. | |
756e6736 KW |
2788 | * |
2789 | * Returns 0 on success, -errno in error cases. | |
2790 | */ | |
e24e49e6 | 2791 | int qcow2_update_header(BlockDriverState *bs) |
756e6736 | 2792 | { |
ff99129a | 2793 | BDRVQcow2State *s = bs->opaque; |
e24e49e6 KW |
2794 | QCowHeader *header; |
2795 | char *buf; | |
2796 | size_t buflen = s->cluster_size; | |
756e6736 | 2797 | int ret; |
e24e49e6 KW |
2798 | uint64_t total_size; |
2799 | uint32_t refcount_table_clusters; | |
6744cbab | 2800 | size_t header_length; |
75bab85c | 2801 | Qcow2UnknownHeaderExtension *uext; |
756e6736 | 2802 | |
e24e49e6 | 2803 | buf = qemu_blockalign(bs, buflen); |
756e6736 | 2804 | |
e24e49e6 KW |
2805 | /* Header structure */ |
2806 | header = (QCowHeader*) buf; | |
756e6736 | 2807 | |
e24e49e6 KW |
2808 | if (buflen < sizeof(*header)) { |
2809 | ret = -ENOSPC; | |
2810 | goto fail; | |
756e6736 KW |
2811 | } |
2812 | ||
6744cbab | 2813 | header_length = sizeof(*header) + s->unknown_header_fields_size; |
e24e49e6 KW |
2814 | total_size = bs->total_sectors * BDRV_SECTOR_SIZE; |
2815 | refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3); | |
2816 | ||
572ad978 DP |
2817 | ret = validate_compression_type(s, NULL); |
2818 | if (ret) { | |
2819 | goto fail; | |
2820 | } | |
2821 | ||
e24e49e6 | 2822 | *header = (QCowHeader) { |
6744cbab | 2823 | /* Version 2 fields */ |
e24e49e6 | 2824 | .magic = cpu_to_be32(QCOW_MAGIC), |
6744cbab | 2825 | .version = cpu_to_be32(s->qcow_version), |
e24e49e6 KW |
2826 | .backing_file_offset = 0, |
2827 | .backing_file_size = 0, | |
2828 | .cluster_bits = cpu_to_be32(s->cluster_bits), | |
2829 | .size = cpu_to_be64(total_size), | |
2830 | .crypt_method = cpu_to_be32(s->crypt_method_header), | |
2831 | .l1_size = cpu_to_be32(s->l1_size), | |
2832 | .l1_table_offset = cpu_to_be64(s->l1_table_offset), | |
2833 | .refcount_table_offset = cpu_to_be64(s->refcount_table_offset), | |
2834 | .refcount_table_clusters = cpu_to_be32(refcount_table_clusters), | |
2835 | .nb_snapshots = cpu_to_be32(s->nb_snapshots), | |
2836 | .snapshots_offset = cpu_to_be64(s->snapshots_offset), | |
6744cbab KW |
2837 | |
2838 | /* Version 3 fields */ | |
2839 | .incompatible_features = cpu_to_be64(s->incompatible_features), | |
2840 | .compatible_features = cpu_to_be64(s->compatible_features), | |
2841 | .autoclear_features = cpu_to_be64(s->autoclear_features), | |
b6481f37 | 2842 | .refcount_order = cpu_to_be32(s->refcount_order), |
6744cbab | 2843 | .header_length = cpu_to_be32(header_length), |
572ad978 | 2844 | .compression_type = s->compression_type, |
e24e49e6 | 2845 | }; |
756e6736 | 2846 | |
6744cbab KW |
2847 | /* For older versions, write a shorter header */ |
2848 | switch (s->qcow_version) { | |
2849 | case 2: | |
2850 | ret = offsetof(QCowHeader, incompatible_features); | |
2851 | break; | |
2852 | case 3: | |
2853 | ret = sizeof(*header); | |
2854 | break; | |
2855 | default: | |
b6c14762 JM |
2856 | ret = -EINVAL; |
2857 | goto fail; | |
6744cbab KW |
2858 | } |
2859 | ||
2860 | buf += ret; | |
2861 | buflen -= ret; | |
2862 | memset(buf, 0, buflen); | |
2863 | ||
2864 | /* Preserve any unknown field in the header */ | |
2865 | if (s->unknown_header_fields_size) { | |
2866 | if (buflen < s->unknown_header_fields_size) { | |
2867 | ret = -ENOSPC; | |
2868 | goto fail; | |
2869 | } | |
2870 | ||
2871 | memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); | |
2872 | buf += s->unknown_header_fields_size; | |
2873 | buflen -= s->unknown_header_fields_size; | |
2874 | } | |
756e6736 | 2875 | |
e24e49e6 | 2876 | /* Backing file format header extension */ |
e4603fe1 | 2877 | if (s->image_backing_format) { |
e24e49e6 | 2878 | ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, |
e4603fe1 KW |
2879 | s->image_backing_format, |
2880 | strlen(s->image_backing_format), | |
e24e49e6 KW |
2881 | buflen); |
2882 | if (ret < 0) { | |
2883 | goto fail; | |
756e6736 KW |
2884 | } |
2885 | ||
e24e49e6 KW |
2886 | buf += ret; |
2887 | buflen -= ret; | |
756e6736 KW |
2888 | } |
2889 | ||
9b890bdc KW |
2890 | /* External data file header extension */ |
2891 | if (has_data_file(bs) && s->image_data_file) { | |
2892 | ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE, | |
2893 | s->image_data_file, strlen(s->image_data_file), | |
2894 | buflen); | |
2895 | if (ret < 0) { | |
2896 | goto fail; | |
2897 | } | |
2898 | ||
2899 | buf += ret; | |
2900 | buflen -= ret; | |
2901 | } | |
2902 | ||
4652b8f3 DB |
2903 | /* Full disk encryption header pointer extension */ |
2904 | if (s->crypto_header.offset != 0) { | |
3b698f52 PM |
2905 | s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset); |
2906 | s->crypto_header.length = cpu_to_be64(s->crypto_header.length); | |
4652b8f3 DB |
2907 | ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER, |
2908 | &s->crypto_header, sizeof(s->crypto_header), | |
2909 | buflen); | |
3b698f52 PM |
2910 | s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset); |
2911 | s->crypto_header.length = be64_to_cpu(s->crypto_header.length); | |
4652b8f3 DB |
2912 | if (ret < 0) { |
2913 | goto fail; | |
2914 | } | |
2915 | buf += ret; | |
2916 | buflen -= ret; | |
2917 | } | |
2918 | ||
e7be13ad EB |
2919 | /* |
2920 | * Feature table. A mere 8 feature names occupies 392 bytes, and | |
2921 | * when coupled with the v3 minimum header of 104 bytes plus the | |
2922 | * 8-byte end-of-extension marker, that would leave only 8 bytes | |
2923 | * for a backing file name in an image with 512-byte clusters. | |
2924 | * Thus, we choose to omit this header for cluster sizes 4k and | |
2925 | * smaller. | |
2926 | */ | |
2927 | if (s->qcow_version >= 3 && s->cluster_size > 4096) { | |
bb40ebce | 2928 | static const Qcow2Feature features[] = { |
1a4828c7 KW |
2929 | { |
2930 | .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, | |
2931 | .bit = QCOW2_INCOMPAT_DIRTY_BITNR, | |
2932 | .name = "dirty bit", | |
2933 | }, | |
2934 | { | |
2935 | .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, | |
2936 | .bit = QCOW2_INCOMPAT_CORRUPT_BITNR, | |
2937 | .name = "corrupt bit", | |
2938 | }, | |
93c24936 KW |
2939 | { |
2940 | .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, | |
2941 | .bit = QCOW2_INCOMPAT_DATA_FILE_BITNR, | |
2942 | .name = "external data file", | |
2943 | }, | |
572ad978 DP |
2944 | { |
2945 | .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, | |
2946 | .bit = QCOW2_INCOMPAT_COMPRESSION_BITNR, | |
2947 | .name = "compression type", | |
2948 | }, | |
7be20252 AG |
2949 | { |
2950 | .type = QCOW2_FEAT_TYPE_INCOMPATIBLE, | |
2951 | .bit = QCOW2_INCOMPAT_EXTL2_BITNR, | |
2952 | .name = "extended L2 entries", | |
2953 | }, | |
1a4828c7 KW |
2954 | { |
2955 | .type = QCOW2_FEAT_TYPE_COMPATIBLE, | |
2956 | .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, | |
2957 | .name = "lazy refcounts", | |
2958 | }, | |
bb40ebce EB |
2959 | { |
2960 | .type = QCOW2_FEAT_TYPE_AUTOCLEAR, | |
2961 | .bit = QCOW2_AUTOCLEAR_BITMAPS_BITNR, | |
2962 | .name = "bitmaps", | |
2963 | }, | |
2964 | { | |
2965 | .type = QCOW2_FEAT_TYPE_AUTOCLEAR, | |
2966 | .bit = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR, | |
2967 | .name = "raw external data", | |
2968 | }, | |
1a4828c7 KW |
2969 | }; |
2970 | ||
2971 | ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, | |
2972 | features, sizeof(features), buflen); | |
2973 | if (ret < 0) { | |
2974 | goto fail; | |
2975 | } | |
2976 | buf += ret; | |
2977 | buflen -= ret; | |
cfcc4c62 | 2978 | } |
cfcc4c62 | 2979 | |
88ddffae VSO |
2980 | /* Bitmap extension */ |
2981 | if (s->nb_bitmaps > 0) { | |
2982 | Qcow2BitmapHeaderExt bitmaps_header = { | |
2983 | .nb_bitmaps = cpu_to_be32(s->nb_bitmaps), | |
2984 | .bitmap_directory_size = | |
2985 | cpu_to_be64(s->bitmap_directory_size), | |
2986 | .bitmap_directory_offset = | |
2987 | cpu_to_be64(s->bitmap_directory_offset) | |
2988 | }; | |
2989 | ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS, | |
2990 | &bitmaps_header, sizeof(bitmaps_header), | |
2991 | buflen); | |
2992 | if (ret < 0) { | |
2993 | goto fail; | |
2994 | } | |
2995 | buf += ret; | |
2996 | buflen -= ret; | |
2997 | } | |
2998 | ||
75bab85c KW |
2999 | /* Keep unknown header extensions */ |
3000 | QLIST_FOREACH(uext, &s->unknown_header_ext, next) { | |
3001 | ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); | |
3002 | if (ret < 0) { | |
3003 | goto fail; | |
3004 | } | |
3005 | ||
3006 | buf += ret; | |
3007 | buflen -= ret; | |
3008 | } | |
3009 | ||
e24e49e6 KW |
3010 | /* End of header extensions */ |
3011 | ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); | |
756e6736 KW |
3012 | if (ret < 0) { |
3013 | goto fail; | |
3014 | } | |
3015 | ||
e24e49e6 KW |
3016 | buf += ret; |
3017 | buflen -= ret; | |
756e6736 | 3018 | |
e24e49e6 | 3019 | /* Backing file name */ |
e4603fe1 KW |
3020 | if (s->image_backing_file) { |
3021 | size_t backing_file_len = strlen(s->image_backing_file); | |
e24e49e6 KW |
3022 | |
3023 | if (buflen < backing_file_len) { | |
3024 | ret = -ENOSPC; | |
3025 | goto fail; | |
3026 | } | |
3027 | ||
00ea1881 | 3028 | /* Using strncpy is ok here, since buf is not NUL-terminated. */ |
e4603fe1 | 3029 | strncpy(buf, s->image_backing_file, buflen); |
e24e49e6 KW |
3030 | |
3031 | header->backing_file_offset = cpu_to_be64(buf - ((char*) header)); | |
3032 | header->backing_file_size = cpu_to_be32(backing_file_len); | |
756e6736 KW |
3033 | } |
3034 | ||
e24e49e6 | 3035 | /* Write the new header */ |
d9ca2ea2 | 3036 | ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size); |
756e6736 KW |
3037 | if (ret < 0) { |
3038 | goto fail; | |
3039 | } | |
3040 | ||
3041 | ret = 0; | |
3042 | fail: | |
e24e49e6 | 3043 | qemu_vfree(header); |
756e6736 KW |
3044 | return ret; |
3045 | } | |
3046 | ||
3047 | static int qcow2_change_backing_file(BlockDriverState *bs, | |
3048 | const char *backing_file, const char *backing_fmt) | |
3049 | { | |
ff99129a | 3050 | BDRVQcow2State *s = bs->opaque; |
e4603fe1 | 3051 | |
6c3944dc KW |
3052 | /* Adding a backing file means that the external data file alone won't be |
3053 | * enough to make sense of the content */ | |
3054 | if (backing_file && data_file_is_raw(bs)) { | |
3055 | return -EINVAL; | |
3056 | } | |
3057 | ||
4e876bcf HR |
3058 | if (backing_file && strlen(backing_file) > 1023) { |
3059 | return -EINVAL; | |
3060 | } | |
3061 | ||
998c2019 HR |
3062 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), |
3063 | backing_file ?: ""); | |
e24e49e6 KW |
3064 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); |
3065 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); | |
3066 | ||
e4603fe1 KW |
3067 | g_free(s->image_backing_file); |
3068 | g_free(s->image_backing_format); | |
3069 | ||
3070 | s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL; | |
3071 | s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL; | |
3072 | ||
e24e49e6 | 3073 | return qcow2_update_header(bs); |
756e6736 KW |
3074 | } |
3075 | ||
60900b7b KW |
3076 | static int qcow2_set_up_encryption(BlockDriverState *bs, |
3077 | QCryptoBlockCreateOptions *cryptoopts, | |
3078 | Error **errp) | |
3079 | { | |
3080 | BDRVQcow2State *s = bs->opaque; | |
3081 | QCryptoBlock *crypto = NULL; | |
3082 | int fmt, ret; | |
3083 | ||
3084 | switch (cryptoopts->format) { | |
3085 | case Q_CRYPTO_BLOCK_FORMAT_LUKS: | |
3086 | fmt = QCOW_CRYPT_LUKS; | |
3087 | break; | |
3088 | case Q_CRYPTO_BLOCK_FORMAT_QCOW: | |
3089 | fmt = QCOW_CRYPT_AES; | |
3090 | break; | |
3091 | default: | |
3092 | error_setg(errp, "Crypto format not supported in qcow2"); | |
3093 | return -EINVAL; | |
b25b387f | 3094 | } |
60900b7b | 3095 | |
4652b8f3 | 3096 | s->crypt_method_header = fmt; |
b25b387f | 3097 | |
1cd9a787 | 3098 | crypto = qcrypto_block_create(cryptoopts, "encrypt.", |
4652b8f3 DB |
3099 | qcow2_crypto_hdr_init_func, |
3100 | qcow2_crypto_hdr_write_func, | |
b25b387f DB |
3101 | bs, errp); |
3102 | if (!crypto) { | |
60900b7b | 3103 | return -EINVAL; |
b25b387f DB |
3104 | } |
3105 | ||
3106 | ret = qcow2_update_header(bs); | |
3107 | if (ret < 0) { | |
3108 | error_setg_errno(errp, -ret, "Could not write encryption header"); | |
3109 | goto out; | |
3110 | } | |
3111 | ||
60900b7b | 3112 | ret = 0; |
b25b387f | 3113 | out: |
b25b387f | 3114 | qcrypto_block_free(crypto); |
b25b387f DB |
3115 | return ret; |
3116 | } | |
3117 | ||
7bc45dc1 HR |
3118 | /** |
3119 | * Preallocates metadata structures for data clusters between @offset (in the | |
3120 | * guest disk) and @new_length (which is thus generally the new guest disk | |
3121 | * size). | |
3122 | * | |
3123 | * Returns: 0 on success, -errno on failure. | |
3124 | */ | |
47e86b86 | 3125 | static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset, |
718c0fce KW |
3126 | uint64_t new_length, PreallocMode mode, |
3127 | Error **errp) | |
a35e1c17 | 3128 | { |
93e32b3e | 3129 | BDRVQcow2State *s = bs->opaque; |
d46a0bb2 | 3130 | uint64_t bytes; |
060bee89 | 3131 | uint64_t host_offset = 0; |
718c0fce | 3132 | int64_t file_length; |
d46a0bb2 | 3133 | unsigned int cur_bytes; |
148da7ea | 3134 | int ret; |
1a52b73d | 3135 | QCowL2Meta *meta = NULL, *m; |
a35e1c17 | 3136 | |
7bc45dc1 HR |
3137 | assert(offset <= new_length); |
3138 | bytes = new_length - offset; | |
a35e1c17 | 3139 | |
d46a0bb2 | 3140 | while (bytes) { |
f29fbf7c | 3141 | cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size)); |
bfd0989a AG |
3142 | ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes, |
3143 | &host_offset, &meta); | |
148da7ea | 3144 | if (ret < 0) { |
360bd074 | 3145 | error_setg_errno(errp, -ret, "Allocating clusters failed"); |
1a52b73d | 3146 | goto out; |
a35e1c17 KW |
3147 | } |
3148 | ||
1a52b73d AG |
3149 | for (m = meta; m != NULL; m = m->next) { |
3150 | m->prealloc = true; | |
3151 | } | |
c792707f | 3152 | |
1a52b73d AG |
3153 | ret = qcow2_handle_l2meta(bs, &meta, true); |
3154 | if (ret < 0) { | |
3155 | error_setg_errno(errp, -ret, "Mapping clusters failed"); | |
3156 | goto out; | |
f50f88b9 | 3157 | } |
f214978a | 3158 | |
a35e1c17 KW |
3159 | /* TODO Preallocate data if requested */ |
3160 | ||
d46a0bb2 KW |
3161 | bytes -= cur_bytes; |
3162 | offset += cur_bytes; | |
a35e1c17 KW |
3163 | } |
3164 | ||
3165 | /* | |
3166 | * It is expected that the image file is large enough to actually contain | |
3167 | * all of the allocated clusters (otherwise we get failing reads after | |
3168 | * EOF). Extend the image to the last allocated sector. | |
3169 | */ | |
718c0fce KW |
3170 | file_length = bdrv_getlength(s->data_file->bs); |
3171 | if (file_length < 0) { | |
3172 | error_setg_errno(errp, -file_length, "Could not get file size"); | |
1a52b73d AG |
3173 | ret = file_length; |
3174 | goto out; | |
718c0fce KW |
3175 | } |
3176 | ||
3177 | if (host_offset + cur_bytes > file_length) { | |
3178 | if (mode == PREALLOC_MODE_METADATA) { | |
3179 | mode = PREALLOC_MODE_OFF; | |
3180 | } | |
c80d8b06 | 3181 | ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false, |
7b8e4857 | 3182 | mode, 0, errp); |
19dbcbf7 | 3183 | if (ret < 0) { |
1a52b73d | 3184 | goto out; |
19dbcbf7 | 3185 | } |
a35e1c17 KW |
3186 | } |
3187 | ||
1a52b73d AG |
3188 | ret = 0; |
3189 | ||
3190 | out: | |
3191 | qcow2_handle_l2meta(bs, &meta, false); | |
3192 | return ret; | |
a35e1c17 KW |
3193 | } |
3194 | ||
7c5bcc42 SH |
3195 | /* qcow2_refcount_metadata_size: |
3196 | * @clusters: number of clusters to refcount (including data and L1/L2 tables) | |
3197 | * @cluster_size: size of a cluster, in bytes | |
3198 | * @refcount_order: refcount bits power-of-2 exponent | |
12cc30a8 HR |
3199 | * @generous_increase: allow for the refcount table to be 1.5x as large as it |
3200 | * needs to be | |
7c5bcc42 SH |
3201 | * |
3202 | * Returns: Number of bytes required for refcount blocks and table metadata. | |
3203 | */ | |
12cc30a8 HR |
3204 | int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size, |
3205 | int refcount_order, bool generous_increase, | |
3206 | uint64_t *refblock_count) | |
7c5bcc42 SH |
3207 | { |
3208 | /* | |
3209 | * Every host cluster is reference-counted, including metadata (even | |
3210 | * refcount metadata is recursively included). | |
3211 | * | |
3212 | * An accurate formula for the size of refcount metadata size is difficult | |
3213 | * to derive. An easier method of calculation is finding the fixed point | |
3214 | * where no further refcount blocks or table clusters are required to | |
3215 | * reference count every cluster. | |
3216 | */ | |
02b1ecfa | 3217 | int64_t blocks_per_table_cluster = cluster_size / REFTABLE_ENTRY_SIZE; |
7c5bcc42 SH |
3218 | int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order); |
3219 | int64_t table = 0; /* number of refcount table clusters */ | |
3220 | int64_t blocks = 0; /* number of refcount block clusters */ | |
3221 | int64_t last; | |
3222 | int64_t n = 0; | |
3223 | ||
3224 | do { | |
3225 | last = n; | |
3226 | blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block); | |
3227 | table = DIV_ROUND_UP(blocks, blocks_per_table_cluster); | |
3228 | n = clusters + blocks + table; | |
12cc30a8 HR |
3229 | |
3230 | if (n == last && generous_increase) { | |
3231 | clusters += DIV_ROUND_UP(table, 2); | |
3232 | n = 0; /* force another loop */ | |
3233 | generous_increase = false; | |
3234 | } | |
7c5bcc42 SH |
3235 | } while (n != last); |
3236 | ||
12cc30a8 HR |
3237 | if (refblock_count) { |
3238 | *refblock_count = blocks; | |
3239 | } | |
3240 | ||
7c5bcc42 SH |
3241 | return (blocks + table) * cluster_size; |
3242 | } | |
3243 | ||
95c67e3b SH |
3244 | /** |
3245 | * qcow2_calc_prealloc_size: | |
3246 | * @total_size: virtual disk size in bytes | |
3247 | * @cluster_size: cluster size in bytes | |
3248 | * @refcount_order: refcount bits power-of-2 exponent | |
0dd07b29 | 3249 | * @extended_l2: true if the image has extended L2 entries |
95c67e3b SH |
3250 | * |
3251 | * Returns: Total number of bytes required for the fully allocated image | |
3252 | * (including metadata). | |
3253 | */ | |
3254 | static int64_t qcow2_calc_prealloc_size(int64_t total_size, | |
3255 | size_t cluster_size, | |
0dd07b29 AG |
3256 | int refcount_order, |
3257 | bool extended_l2) | |
95c67e3b | 3258 | { |
95c67e3b | 3259 | int64_t meta_size = 0; |
7c5bcc42 | 3260 | uint64_t nl1e, nl2e; |
9e029689 | 3261 | int64_t aligned_total_size = ROUND_UP(total_size, cluster_size); |
0dd07b29 | 3262 | size_t l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL; |
95c67e3b SH |
3263 | |
3264 | /* header: 1 cluster */ | |
3265 | meta_size += cluster_size; | |
3266 | ||
3267 | /* total size of L2 tables */ | |
3268 | nl2e = aligned_total_size / cluster_size; | |
0dd07b29 AG |
3269 | nl2e = ROUND_UP(nl2e, cluster_size / l2e_size); |
3270 | meta_size += nl2e * l2e_size; | |
95c67e3b SH |
3271 | |
3272 | /* total size of L1 tables */ | |
0dd07b29 | 3273 | nl1e = nl2e * l2e_size / cluster_size; |
02b1ecfa AG |
3274 | nl1e = ROUND_UP(nl1e, cluster_size / L1E_SIZE); |
3275 | meta_size += nl1e * L1E_SIZE; | |
95c67e3b | 3276 | |
7c5bcc42 SH |
3277 | /* total size of refcount table and blocks */ |
3278 | meta_size += qcow2_refcount_metadata_size( | |
3279 | (meta_size + aligned_total_size) / cluster_size, | |
12cc30a8 | 3280 | cluster_size, refcount_order, false, NULL); |
95c67e3b SH |
3281 | |
3282 | return meta_size + aligned_total_size; | |
3283 | } | |
3284 | ||
7be20252 AG |
3285 | static bool validate_cluster_size(size_t cluster_size, bool extended_l2, |
3286 | Error **errp) | |
a9420734 | 3287 | { |
29ca9e45 | 3288 | int cluster_bits = ctz32(cluster_size); |
a9420734 KW |
3289 | if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || |
3290 | (1 << cluster_bits) != cluster_size) | |
3291 | { | |
3ef6c40a HR |
3292 | error_setg(errp, "Cluster size must be a power of two between %d and " |
3293 | "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); | |
29ca9e45 KW |
3294 | return false; |
3295 | } | |
7be20252 AG |
3296 | |
3297 | if (extended_l2) { | |
3298 | unsigned min_cluster_size = | |
3299 | (1 << MIN_CLUSTER_BITS) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER; | |
3300 | if (cluster_size < min_cluster_size) { | |
3301 | error_setg(errp, "Extended L2 entries are only supported with " | |
3302 | "cluster sizes of at least %u bytes", min_cluster_size); | |
3303 | return false; | |
3304 | } | |
3305 | } | |
3306 | ||
29ca9e45 KW |
3307 | return true; |
3308 | } | |
3309 | ||
7be20252 AG |
3310 | static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, bool extended_l2, |
3311 | Error **errp) | |
29ca9e45 KW |
3312 | { |
3313 | size_t cluster_size; | |
3314 | ||
3315 | cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, | |
3316 | DEFAULT_CLUSTER_SIZE); | |
7be20252 | 3317 | if (!validate_cluster_size(cluster_size, extended_l2, errp)) { |
0eb4a8c1 SH |
3318 | return 0; |
3319 | } | |
3320 | return cluster_size; | |
3321 | } | |
3322 | ||
3323 | static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp) | |
3324 | { | |
3325 | char *buf; | |
3326 | int ret; | |
3327 | ||
3328 | buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL); | |
3329 | if (!buf) { | |
3330 | ret = 3; /* default */ | |
3331 | } else if (!strcmp(buf, "0.10")) { | |
3332 | ret = 2; | |
3333 | } else if (!strcmp(buf, "1.1")) { | |
3334 | ret = 3; | |
3335 | } else { | |
3336 | error_setg(errp, "Invalid compatibility level: '%s'", buf); | |
3337 | ret = -EINVAL; | |
3338 | } | |
3339 | g_free(buf); | |
3340 | return ret; | |
3341 | } | |
3342 | ||
3343 | static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version, | |
3344 | Error **errp) | |
3345 | { | |
3346 | uint64_t refcount_bits; | |
3347 | ||
3348 | refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16); | |
3349 | if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) { | |
3350 | error_setg(errp, "Refcount width must be a power of two and may not " | |
3351 | "exceed 64 bits"); | |
3352 | return 0; | |
3353 | } | |
3354 | ||
3355 | if (version < 3 && refcount_bits != 16) { | |
3356 | error_setg(errp, "Different refcount widths than 16 bits require " | |
3357 | "compatibility level 1.1 or above (use compat=1.1 or " | |
3358 | "greater)"); | |
3359 | return 0; | |
a9420734 KW |
3360 | } |
3361 | ||
0eb4a8c1 SH |
3362 | return refcount_bits; |
3363 | } | |
3364 | ||
c274393a | 3365 | static int coroutine_fn |
60900b7b | 3366 | qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp) |
0eb4a8c1 | 3367 | { |
29ca9e45 | 3368 | BlockdevCreateOptionsQcow2 *qcow2_opts; |
0eb4a8c1 SH |
3369 | QDict *options; |
3370 | ||
a9420734 KW |
3371 | /* |
3372 | * Open the image file and write a minimal qcow2 header. | |
3373 | * | |
3374 | * We keep things simple and start with a zero-sized image. We also | |
3375 | * do without refcount blocks or a L1 table for now. We'll fix the | |
3376 | * inconsistency later. | |
3377 | * | |
3378 | * We do need a refcount table because growing the refcount table means | |
a951a631 | 3379 | * allocating two new refcount blocks - the second of which would be at |
a9420734 KW |
3380 | * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file |
3381 | * size for any qcow2 image. | |
3382 | */ | |
e1d74bc6 KW |
3383 | BlockBackend *blk = NULL; |
3384 | BlockDriverState *bs = NULL; | |
dcc98687 | 3385 | BlockDriverState *data_bs = NULL; |
f8413b3c | 3386 | QCowHeader *header; |
29ca9e45 KW |
3387 | size_t cluster_size; |
3388 | int version; | |
3389 | int refcount_order; | |
5f14f31d | 3390 | uint64_t *refcount_table; |
a9420734 | 3391 | int ret; |
572ad978 | 3392 | uint8_t compression_type = QCOW2_COMPRESSION_TYPE_ZLIB; |
a9420734 | 3393 | |
29ca9e45 KW |
3394 | assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2); |
3395 | qcow2_opts = &create_options->u.qcow2; | |
3396 | ||
e1d74bc6 KW |
3397 | bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp); |
3398 | if (bs == NULL) { | |
3399 | return -EIO; | |
3400 | } | |
3401 | ||
3402 | /* Validate options and set default values */ | |
29ca9e45 | 3403 | if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) { |
3afea402 AG |
3404 | error_setg(errp, "Image size must be a multiple of %u bytes", |
3405 | (unsigned) BDRV_SECTOR_SIZE); | |
29ca9e45 KW |
3406 | ret = -EINVAL; |
3407 | goto out; | |
3408 | } | |
3409 | ||
3410 | if (qcow2_opts->has_version) { | |
3411 | switch (qcow2_opts->version) { | |
3412 | case BLOCKDEV_QCOW2_VERSION_V2: | |
3413 | version = 2; | |
3414 | break; | |
3415 | case BLOCKDEV_QCOW2_VERSION_V3: | |
3416 | version = 3; | |
3417 | break; | |
3418 | default: | |
3419 | g_assert_not_reached(); | |
3420 | } | |
3421 | } else { | |
3422 | version = 3; | |
3423 | } | |
3424 | ||
3425 | if (qcow2_opts->has_cluster_size) { | |
3426 | cluster_size = qcow2_opts->cluster_size; | |
3427 | } else { | |
3428 | cluster_size = DEFAULT_CLUSTER_SIZE; | |
3429 | } | |
3430 | ||
7be20252 AG |
3431 | if (!qcow2_opts->has_extended_l2) { |
3432 | qcow2_opts->extended_l2 = false; | |
3433 | } | |
3434 | if (qcow2_opts->extended_l2) { | |
3435 | if (version < 3) { | |
3436 | error_setg(errp, "Extended L2 entries are only supported with " | |
3437 | "compatibility level 1.1 and above (use version=v3 or " | |
3438 | "greater)"); | |
3439 | ret = -EINVAL; | |
3440 | goto out; | |
3441 | } | |
3442 | } | |
3443 | ||
3444 | if (!validate_cluster_size(cluster_size, qcow2_opts->extended_l2, errp)) { | |
e1d74bc6 KW |
3445 | ret = -EINVAL; |
3446 | goto out; | |
29ca9e45 KW |
3447 | } |
3448 | ||
3449 | if (!qcow2_opts->has_preallocation) { | |
3450 | qcow2_opts->preallocation = PREALLOC_MODE_OFF; | |
3451 | } | |
3452 | if (qcow2_opts->has_backing_file && | |
2118771d AG |
3453 | qcow2_opts->preallocation != PREALLOC_MODE_OFF && |
3454 | !qcow2_opts->extended_l2) | |
29ca9e45 | 3455 | { |
2118771d AG |
3456 | error_setg(errp, "Backing file and preallocation can only be used at " |
3457 | "the same time if extended_l2 is on"); | |
e1d74bc6 KW |
3458 | ret = -EINVAL; |
3459 | goto out; | |
29ca9e45 KW |
3460 | } |
3461 | if (qcow2_opts->has_backing_fmt && !qcow2_opts->has_backing_file) { | |
3462 | error_setg(errp, "Backing format cannot be used without backing file"); | |
e1d74bc6 KW |
3463 | ret = -EINVAL; |
3464 | goto out; | |
29ca9e45 KW |
3465 | } |
3466 | ||
3467 | if (!qcow2_opts->has_lazy_refcounts) { | |
3468 | qcow2_opts->lazy_refcounts = false; | |
3469 | } | |
3470 | if (version < 3 && qcow2_opts->lazy_refcounts) { | |
3471 | error_setg(errp, "Lazy refcounts only supported with compatibility " | |
b76b4f60 | 3472 | "level 1.1 and above (use version=v3 or greater)"); |
e1d74bc6 KW |
3473 | ret = -EINVAL; |
3474 | goto out; | |
29ca9e45 KW |
3475 | } |
3476 | ||
3477 | if (!qcow2_opts->has_refcount_bits) { | |
3478 | qcow2_opts->refcount_bits = 16; | |
3479 | } | |
3480 | if (qcow2_opts->refcount_bits > 64 || | |
3481 | !is_power_of_2(qcow2_opts->refcount_bits)) | |
3482 | { | |
3483 | error_setg(errp, "Refcount width must be a power of two and may not " | |
3484 | "exceed 64 bits"); | |
e1d74bc6 KW |
3485 | ret = -EINVAL; |
3486 | goto out; | |
29ca9e45 KW |
3487 | } |
3488 | if (version < 3 && qcow2_opts->refcount_bits != 16) { | |
3489 | error_setg(errp, "Different refcount widths than 16 bits require " | |
b76b4f60 | 3490 | "compatibility level 1.1 or above (use version=v3 or " |
29ca9e45 | 3491 | "greater)"); |
e1d74bc6 KW |
3492 | ret = -EINVAL; |
3493 | goto out; | |
29ca9e45 KW |
3494 | } |
3495 | refcount_order = ctz32(qcow2_opts->refcount_bits); | |
3496 | ||
6c3944dc KW |
3497 | if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) { |
3498 | error_setg(errp, "data-file-raw requires data-file"); | |
3499 | ret = -EINVAL; | |
3500 | goto out; | |
3501 | } | |
3502 | if (qcow2_opts->data_file_raw && qcow2_opts->has_backing_file) { | |
3503 | error_setg(errp, "Backing file and data-file-raw cannot be used at " | |
3504 | "the same time"); | |
3505 | ret = -EINVAL; | |
3506 | goto out; | |
3507 | } | |
3508 | ||
dcc98687 KW |
3509 | if (qcow2_opts->data_file) { |
3510 | if (version < 3) { | |
3511 | error_setg(errp, "External data files are only supported with " | |
3512 | "compatibility level 1.1 and above (use version=v3 or " | |
3513 | "greater)"); | |
3514 | ret = -EINVAL; | |
3515 | goto out; | |
3516 | } | |
3517 | data_bs = bdrv_open_blockdev_ref(qcow2_opts->data_file, errp); | |
a0cf8363 | 3518 | if (data_bs == NULL) { |
dcc98687 KW |
3519 | ret = -EIO; |
3520 | goto out; | |
3521 | } | |
3522 | } | |
29ca9e45 | 3523 | |
572ad978 DP |
3524 | if (qcow2_opts->has_compression_type && |
3525 | qcow2_opts->compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) { | |
3526 | ||
3527 | ret = -EINVAL; | |
3528 | ||
3529 | if (version < 3) { | |
3530 | error_setg(errp, "Non-zlib compression type is only supported with " | |
3531 | "compatibility level 1.1 and above (use version=v3 or " | |
3532 | "greater)"); | |
3533 | goto out; | |
3534 | } | |
3535 | ||
3536 | switch (qcow2_opts->compression_type) { | |
d298ac10 DP |
3537 | #ifdef CONFIG_ZSTD |
3538 | case QCOW2_COMPRESSION_TYPE_ZSTD: | |
3539 | break; | |
3540 | #endif | |
572ad978 DP |
3541 | default: |
3542 | error_setg(errp, "Unknown compression type"); | |
3543 | goto out; | |
3544 | } | |
3545 | ||
3546 | compression_type = qcow2_opts->compression_type; | |
3547 | } | |
3548 | ||
29ca9e45 | 3549 | /* Create BlockBackend to write to the image */ |
a3aeeab5 EB |
3550 | blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL, |
3551 | errp); | |
3552 | if (!blk) { | |
3553 | ret = -EPERM; | |
cbf2b7c4 | 3554 | goto out; |
a9420734 | 3555 | } |
23588797 KW |
3556 | blk_set_allow_write_beyond_eof(blk, true); |
3557 | ||
a9420734 | 3558 | /* Write the header */ |
f8413b3c KW |
3559 | QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); |
3560 | header = g_malloc0(cluster_size); | |
3561 | *header = (QCowHeader) { | |
3562 | .magic = cpu_to_be32(QCOW_MAGIC), | |
3563 | .version = cpu_to_be32(version), | |
0eb4a8c1 | 3564 | .cluster_bits = cpu_to_be32(ctz32(cluster_size)), |
f8413b3c KW |
3565 | .size = cpu_to_be64(0), |
3566 | .l1_table_offset = cpu_to_be64(0), | |
3567 | .l1_size = cpu_to_be32(0), | |
3568 | .refcount_table_offset = cpu_to_be64(cluster_size), | |
3569 | .refcount_table_clusters = cpu_to_be32(1), | |
bd4b167f | 3570 | .refcount_order = cpu_to_be32(refcount_order), |
572ad978 DP |
3571 | /* don't deal with endianness since compression_type is 1 byte long */ |
3572 | .compression_type = compression_type, | |
f8413b3c KW |
3573 | .header_length = cpu_to_be32(sizeof(*header)), |
3574 | }; | |
a9420734 | 3575 | |
b25b387f DB |
3576 | /* We'll update this to correct value later */ |
3577 | header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); | |
a9420734 | 3578 | |
29ca9e45 | 3579 | if (qcow2_opts->lazy_refcounts) { |
f8413b3c | 3580 | header->compatible_features |= |
bfe8043e SH |
3581 | cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); |
3582 | } | |
dcc98687 KW |
3583 | if (data_bs) { |
3584 | header->incompatible_features |= | |
3585 | cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE); | |
3586 | } | |
6c3944dc KW |
3587 | if (qcow2_opts->data_file_raw) { |
3588 | header->autoclear_features |= | |
3589 | cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW); | |
3590 | } | |
572ad978 DP |
3591 | if (compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) { |
3592 | header->incompatible_features |= | |
3593 | cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION); | |
3594 | } | |
bfe8043e | 3595 | |
7be20252 AG |
3596 | if (qcow2_opts->extended_l2) { |
3597 | header->incompatible_features |= | |
3598 | cpu_to_be64(QCOW2_INCOMPAT_EXTL2); | |
3599 | } | |
3600 | ||
8341f00d | 3601 | ret = blk_pwrite(blk, 0, header, cluster_size, 0); |
f8413b3c | 3602 | g_free(header); |
a9420734 | 3603 | if (ret < 0) { |
3ef6c40a | 3604 | error_setg_errno(errp, -ret, "Could not write qcow2 header"); |
a9420734 KW |
3605 | goto out; |
3606 | } | |
3607 | ||
b106ad91 KW |
3608 | /* Write a refcount table with one refcount block */ |
3609 | refcount_table = g_malloc0(2 * cluster_size); | |
3610 | refcount_table[0] = cpu_to_be64(2 * cluster_size); | |
8341f00d | 3611 | ret = blk_pwrite(blk, cluster_size, refcount_table, 2 * cluster_size, 0); |
7267c094 | 3612 | g_free(refcount_table); |
a9420734 KW |
3613 | |
3614 | if (ret < 0) { | |
3ef6c40a | 3615 | error_setg_errno(errp, -ret, "Could not write refcount table"); |
a9420734 KW |
3616 | goto out; |
3617 | } | |
3618 | ||
23588797 KW |
3619 | blk_unref(blk); |
3620 | blk = NULL; | |
a9420734 KW |
3621 | |
3622 | /* | |
3623 | * And now open the image and make it consistent first (i.e. increase the | |
3624 | * refcount of the cluster that is occupied by the header and the refcount | |
3625 | * table) | |
3626 | */ | |
e6641719 | 3627 | options = qdict_new(); |
46f5ac20 | 3628 | qdict_put_str(options, "driver", "qcow2"); |
cbf2b7c4 | 3629 | qdict_put_str(options, "file", bs->node_name); |
dcc98687 KW |
3630 | if (data_bs) { |
3631 | qdict_put_str(options, "data-file", data_bs->node_name); | |
3632 | } | |
cbf2b7c4 | 3633 | blk = blk_new_open(NULL, NULL, options, |
55880601 | 3634 | BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH, |
af175e85 | 3635 | errp); |
23588797 | 3636 | if (blk == NULL) { |
23588797 | 3637 | ret = -EIO; |
a9420734 KW |
3638 | goto out; |
3639 | } | |
3640 | ||
23588797 | 3641 | ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size); |
a9420734 | 3642 | if (ret < 0) { |
3ef6c40a HR |
3643 | error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 " |
3644 | "header and refcount table"); | |
a9420734 KW |
3645 | goto out; |
3646 | ||
3647 | } else if (ret != 0) { | |
3648 | error_report("Huh, first cluster in empty image is already in use?"); | |
3649 | abort(); | |
3650 | } | |
3651 | ||
9b890bdc KW |
3652 | /* Set the external data file if necessary */ |
3653 | if (data_bs) { | |
3654 | BDRVQcow2State *s = blk_bs(blk)->opaque; | |
3655 | s->image_data_file = g_strdup(data_bs->filename); | |
3656 | } | |
3657 | ||
b527c9b3 | 3658 | /* Create a full header (including things like feature table) */ |
23588797 | 3659 | ret = qcow2_update_header(blk_bs(blk)); |
b527c9b3 KW |
3660 | if (ret < 0) { |
3661 | error_setg_errno(errp, -ret, "Could not update qcow2 header"); | |
3662 | goto out; | |
3663 | } | |
3664 | ||
a9420734 | 3665 | /* Okay, now that we have a valid image, let's give it the right size */ |
c80d8b06 | 3666 | ret = blk_truncate(blk, qcow2_opts->size, false, qcow2_opts->preallocation, |
8c6242b6 | 3667 | 0, errp); |
a9420734 | 3668 | if (ret < 0) { |
ed3d2ec9 | 3669 | error_prepend(errp, "Could not resize image: "); |
a9420734 KW |
3670 | goto out; |
3671 | } | |
3672 | ||
a951a631 | 3673 | /* Want a backing file? There you go. */ |
29ca9e45 KW |
3674 | if (qcow2_opts->has_backing_file) { |
3675 | const char *backing_format = NULL; | |
3676 | ||
3677 | if (qcow2_opts->has_backing_fmt) { | |
3678 | backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt); | |
3679 | } | |
3680 | ||
3681 | ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file, | |
e54ee1b3 | 3682 | backing_format, false); |
a9420734 | 3683 | if (ret < 0) { |
3ef6c40a | 3684 | error_setg_errno(errp, -ret, "Could not assign backing file '%s' " |
29ca9e45 KW |
3685 | "with format '%s'", qcow2_opts->backing_file, |
3686 | backing_format); | |
a9420734 KW |
3687 | goto out; |
3688 | } | |
3689 | } | |
3690 | ||
b25b387f | 3691 | /* Want encryption? There you go. */ |
60900b7b KW |
3692 | if (qcow2_opts->has_encrypt) { |
3693 | ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp); | |
b25b387f DB |
3694 | if (ret < 0) { |
3695 | goto out; | |
3696 | } | |
3697 | } | |
3698 | ||
23588797 KW |
3699 | blk_unref(blk); |
3700 | blk = NULL; | |
ba2ab2f2 | 3701 | |
b25b387f DB |
3702 | /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning. |
3703 | * Using BDRV_O_NO_IO, since encryption is now setup we don't want to | |
3704 | * have to setup decryption context. We're not doing any I/O on the top | |
3705 | * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does | |
3706 | * not have effect. | |
3707 | */ | |
e6641719 | 3708 | options = qdict_new(); |
46f5ac20 | 3709 | qdict_put_str(options, "driver", "qcow2"); |
cbf2b7c4 | 3710 | qdict_put_str(options, "file", bs->node_name); |
dcc98687 KW |
3711 | if (data_bs) { |
3712 | qdict_put_str(options, "data-file", data_bs->node_name); | |
3713 | } | |
cbf2b7c4 | 3714 | blk = blk_new_open(NULL, NULL, options, |
b25b387f | 3715 | BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO, |
af175e85 | 3716 | errp); |
23588797 | 3717 | if (blk == NULL) { |
23588797 | 3718 | ret = -EIO; |
ba2ab2f2 HR |
3719 | goto out; |
3720 | } | |
3721 | ||
a9420734 KW |
3722 | ret = 0; |
3723 | out: | |
e1d74bc6 KW |
3724 | blk_unref(blk); |
3725 | bdrv_unref(bs); | |
dcc98687 | 3726 | bdrv_unref(data_bs); |
a9420734 KW |
3727 | return ret; |
3728 | } | |
de5f3f40 | 3729 | |
b92902df ML |
3730 | static int coroutine_fn qcow2_co_create_opts(BlockDriver *drv, |
3731 | const char *filename, | |
3732 | QemuOpts *opts, | |
efc75e2a | 3733 | Error **errp) |
de5f3f40 | 3734 | { |
b76b4f60 | 3735 | BlockdevCreateOptions *create_options = NULL; |
92adf9db | 3736 | QDict *qdict; |
b76b4f60 | 3737 | Visitor *v; |
cbf2b7c4 | 3738 | BlockDriverState *bs = NULL; |
9b890bdc | 3739 | BlockDriverState *data_bs = NULL; |
b76b4f60 | 3740 | const char *val; |
3ef6c40a | 3741 | int ret; |
de5f3f40 | 3742 | |
b76b4f60 KW |
3743 | /* Only the keyval visitor supports the dotted syntax needed for |
3744 | * encryption, so go through a QDict before getting a QAPI type. Ignore | |
3745 | * options meant for the protocol layer so that the visitor doesn't | |
3746 | * complain. */ | |
3747 | qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts, | |
3748 | true); | |
3749 | ||
3750 | /* Handle encryption options */ | |
3751 | val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT); | |
3752 | if (val && !strcmp(val, "on")) { | |
3753 | qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow"); | |
3754 | } else if (val && !strcmp(val, "off")) { | |
3755 | qdict_del(qdict, BLOCK_OPT_ENCRYPT); | |
3756 | } | |
3757 | ||
3758 | val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT); | |
3759 | if (val && !strcmp(val, "aes")) { | |
3760 | qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow"); | |
3761 | } | |
3762 | ||
3763 | /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into | |
3764 | * version=v2/v3 below. */ | |
3765 | val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL); | |
3766 | if (val && !strcmp(val, "0.10")) { | |
3767 | qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2"); | |
3768 | } else if (val && !strcmp(val, "1.1")) { | |
3769 | qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3"); | |
3770 | } | |
3771 | ||
3772 | /* Change legacy command line options into QMP ones */ | |
3773 | static const QDictRenames opt_renames[] = { | |
3774 | { BLOCK_OPT_BACKING_FILE, "backing-file" }, | |
3775 | { BLOCK_OPT_BACKING_FMT, "backing-fmt" }, | |
3776 | { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" }, | |
3777 | { BLOCK_OPT_LAZY_REFCOUNTS, "lazy-refcounts" }, | |
7be20252 | 3778 | { BLOCK_OPT_EXTL2, "extended-l2" }, |
b76b4f60 KW |
3779 | { BLOCK_OPT_REFCOUNT_BITS, "refcount-bits" }, |
3780 | { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT }, | |
3781 | { BLOCK_OPT_COMPAT_LEVEL, "version" }, | |
6c3944dc | 3782 | { BLOCK_OPT_DATA_FILE_RAW, "data-file-raw" }, |
572ad978 | 3783 | { BLOCK_OPT_COMPRESSION_TYPE, "compression-type" }, |
b76b4f60 KW |
3784 | { NULL, NULL }, |
3785 | }; | |
3786 | ||
3787 | if (!qdict_rename_keys(qdict, opt_renames, errp)) { | |
29ca9e45 KW |
3788 | ret = -EINVAL; |
3789 | goto finish; | |
3790 | } | |
60900b7b | 3791 | |
b76b4f60 KW |
3792 | /* Create and open the file (protocol layer) */ |
3793 | ret = bdrv_create_file(filename, opts, errp); | |
3794 | if (ret < 0) { | |
0eb4a8c1 SH |
3795 | goto finish; |
3796 | } | |
b76b4f60 KW |
3797 | |
3798 | bs = bdrv_open(filename, NULL, NULL, | |
3799 | BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); | |
3800 | if (bs == NULL) { | |
3801 | ret = -EIO; | |
1bd0e2d1 CL |
3802 | goto finish; |
3803 | } | |
0eb4a8c1 | 3804 | |
9b890bdc KW |
3805 | /* Create and open an external data file (protocol layer) */ |
3806 | val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE); | |
3807 | if (val) { | |
3808 | ret = bdrv_create_file(val, opts, errp); | |
3809 | if (ret < 0) { | |
3810 | goto finish; | |
3811 | } | |
3812 | ||
3813 | data_bs = bdrv_open(val, NULL, NULL, | |
3814 | BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, | |
3815 | errp); | |
3816 | if (data_bs == NULL) { | |
3817 | ret = -EIO; | |
3818 | goto finish; | |
3819 | } | |
3820 | ||
3821 | qdict_del(qdict, BLOCK_OPT_DATA_FILE); | |
3822 | qdict_put_str(qdict, "data-file", data_bs->node_name); | |
3823 | } | |
3824 | ||
b76b4f60 KW |
3825 | /* Set 'driver' and 'node' options */ |
3826 | qdict_put_str(qdict, "driver", "qcow2"); | |
3827 | qdict_put_str(qdict, "file", bs->node_name); | |
3828 | ||
3829 | /* Now get the QAPI type BlockdevCreateOptions */ | |
af91062e MA |
3830 | v = qobject_input_visitor_new_flat_confused(qdict, errp); |
3831 | if (!v) { | |
1bd0e2d1 CL |
3832 | ret = -EINVAL; |
3833 | goto finish; | |
3834 | } | |
3835 | ||
b11a093c | 3836 | visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp); |
b76b4f60 | 3837 | visit_free(v); |
b11a093c | 3838 | if (!create_options) { |
bd4b167f HR |
3839 | ret = -EINVAL; |
3840 | goto finish; | |
3841 | } | |
3842 | ||
b76b4f60 KW |
3843 | /* Silently round up size */ |
3844 | create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size, | |
3845 | BDRV_SECTOR_SIZE); | |
cbf2b7c4 KW |
3846 | |
3847 | /* Create the qcow2 image (format layer) */ | |
b76b4f60 | 3848 | ret = qcow2_co_create(create_options, errp); |
6094cbeb | 3849 | finish: |
cbf2b7c4 | 3850 | if (ret < 0) { |
6094cbeb ML |
3851 | bdrv_co_delete_file_noerr(bs); |
3852 | bdrv_co_delete_file_noerr(data_bs); | |
3853 | } else { | |
3854 | ret = 0; | |
cbf2b7c4 | 3855 | } |
1bd0e2d1 | 3856 | |
cb3e7f08 | 3857 | qobject_unref(qdict); |
cbf2b7c4 | 3858 | bdrv_unref(bs); |
9b890bdc | 3859 | bdrv_unref(data_bs); |
b76b4f60 | 3860 | qapi_free_BlockdevCreateOptions(create_options); |
3ef6c40a | 3861 | return ret; |
de5f3f40 KW |
3862 | } |
3863 | ||
2928abce | 3864 | |
f06f6b66 | 3865 | static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes) |
2928abce | 3866 | { |
31826642 EB |
3867 | int64_t nr; |
3868 | int res; | |
f06f6b66 EB |
3869 | |
3870 | /* Clamp to image length, before checking status of underlying sectors */ | |
8cbf74b2 EB |
3871 | if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { |
3872 | bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset; | |
fbaa6bb3 EB |
3873 | } |
3874 | ||
f06f6b66 | 3875 | if (!bytes) { |
ebb718a5 EB |
3876 | return true; |
3877 | } | |
67c095c8 VSO |
3878 | |
3879 | /* | |
3880 | * bdrv_block_status_above doesn't merge different types of zeros, for | |
3881 | * example, zeros which come from the region which is unallocated in | |
3882 | * the whole backing chain, and zeros which come because of a short | |
3883 | * backing file. So, we need a loop. | |
3884 | */ | |
3885 | do { | |
3886 | res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL); | |
3887 | offset += nr; | |
3888 | bytes -= nr; | |
3889 | } while (res >= 0 && (res & BDRV_BLOCK_ZERO) && nr && bytes); | |
3890 | ||
3891 | return res >= 0 && (res & BDRV_BLOCK_ZERO) && bytes == 0; | |
2928abce DL |
3892 | } |
3893 | ||
5544b59f | 3894 | static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 3895 | int64_t offset, int bytes, BdrvRequestFlags flags) |
621f0589 KW |
3896 | { |
3897 | int ret; | |
ff99129a | 3898 | BDRVQcow2State *s = bs->opaque; |
621f0589 | 3899 | |
a6841a2d AG |
3900 | uint32_t head = offset_into_subcluster(s, offset); |
3901 | uint32_t tail = ROUND_UP(offset + bytes, s->subcluster_size) - | |
3902 | (offset + bytes); | |
2928abce | 3903 | |
f5a5ca79 MP |
3904 | trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes); |
3905 | if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) { | |
fbaa6bb3 EB |
3906 | tail = 0; |
3907 | } | |
5a64e942 | 3908 | |
ebb718a5 | 3909 | if (head || tail) { |
ebb718a5 | 3910 | uint64_t off; |
ecfe1863 | 3911 | unsigned int nr; |
10dabdc5 | 3912 | QCow2SubclusterType type; |
2928abce | 3913 | |
a6841a2d | 3914 | assert(head + bytes + tail <= s->subcluster_size); |
2928abce | 3915 | |
ebb718a5 | 3916 | /* check whether remainder of cluster already reads as zero */ |
f06f6b66 | 3917 | if (!(is_zero(bs, offset - head, head) && |
a6841a2d | 3918 | is_zero(bs, offset + bytes, tail))) { |
2928abce DL |
3919 | return -ENOTSUP; |
3920 | } | |
3921 | ||
2928abce DL |
3922 | qemu_co_mutex_lock(&s->lock); |
3923 | /* We can have new write after previous check */ | |
a6841a2d AG |
3924 | offset -= head; |
3925 | bytes = s->subcluster_size; | |
3926 | nr = s->subcluster_size; | |
ca4a0bb8 AG |
3927 | ret = qcow2_get_host_offset(bs, offset, &nr, &off, &type); |
3928 | if (ret < 0 || | |
10dabdc5 | 3929 | (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && |
97490a14 | 3930 | type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && |
10dabdc5 AG |
3931 | type != QCOW2_SUBCLUSTER_ZERO_PLAIN && |
3932 | type != QCOW2_SUBCLUSTER_ZERO_ALLOC)) { | |
2928abce | 3933 | qemu_co_mutex_unlock(&s->lock); |
580384d6 | 3934 | return ret < 0 ? ret : -ENOTSUP; |
2928abce DL |
3935 | } |
3936 | } else { | |
3937 | qemu_co_mutex_lock(&s->lock); | |
621f0589 KW |
3938 | } |
3939 | ||
f5a5ca79 | 3940 | trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes); |
5a64e942 | 3941 | |
a6841a2d AG |
3942 | /* Whatever is left can use real zero subclusters */ |
3943 | ret = qcow2_subcluster_zeroize(bs, offset, bytes, flags); | |
621f0589 KW |
3944 | qemu_co_mutex_unlock(&s->lock); |
3945 | ||
3946 | return ret; | |
3947 | } | |
3948 | ||
82e8a788 | 3949 | static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs, |
f5a5ca79 | 3950 | int64_t offset, int bytes) |
5ea929e3 | 3951 | { |
6db39ae2 | 3952 | int ret; |
ff99129a | 3953 | BDRVQcow2State *s = bs->opaque; |
6db39ae2 | 3954 | |
80f5c011 AG |
3955 | /* If the image does not support QCOW_OFLAG_ZERO then discarding |
3956 | * clusters could expose stale data from the backing file. */ | |
3957 | if (s->qcow_version < 3 && bs->backing) { | |
3958 | return -ENOTSUP; | |
3959 | } | |
3960 | ||
f5a5ca79 MP |
3961 | if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) { |
3962 | assert(bytes < s->cluster_size); | |
048c5fd1 EB |
3963 | /* Ignore partial clusters, except for the special case of the |
3964 | * complete partial cluster at the end of an unaligned file */ | |
3965 | if (!QEMU_IS_ALIGNED(offset, s->cluster_size) || | |
f5a5ca79 | 3966 | offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) { |
048c5fd1 EB |
3967 | return -ENOTSUP; |
3968 | } | |
49228d1e EB |
3969 | } |
3970 | ||
6db39ae2 | 3971 | qemu_co_mutex_lock(&s->lock); |
f5a5ca79 | 3972 | ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST, |
d2cb36af | 3973 | false); |
6db39ae2 PB |
3974 | qemu_co_mutex_unlock(&s->lock); |
3975 | return ret; | |
5ea929e3 KW |
3976 | } |
3977 | ||
fd9fcd37 FZ |
3978 | static int coroutine_fn |
3979 | qcow2_co_copy_range_from(BlockDriverState *bs, | |
3980 | BdrvChild *src, uint64_t src_offset, | |
3981 | BdrvChild *dst, uint64_t dst_offset, | |
67b51fb9 VSO |
3982 | uint64_t bytes, BdrvRequestFlags read_flags, |
3983 | BdrvRequestFlags write_flags) | |
fd9fcd37 FZ |
3984 | { |
3985 | BDRVQcow2State *s = bs->opaque; | |
3986 | int ret; | |
3987 | unsigned int cur_bytes; /* number of bytes in current iteration */ | |
3988 | BdrvChild *child = NULL; | |
67b51fb9 | 3989 | BdrvRequestFlags cur_write_flags; |
fd9fcd37 FZ |
3990 | |
3991 | assert(!bs->encrypted); | |
3992 | qemu_co_mutex_lock(&s->lock); | |
3993 | ||
3994 | while (bytes != 0) { | |
3995 | uint64_t copy_offset = 0; | |
10dabdc5 | 3996 | QCow2SubclusterType type; |
fd9fcd37 FZ |
3997 | /* prepare next request */ |
3998 | cur_bytes = MIN(bytes, INT_MAX); | |
67b51fb9 | 3999 | cur_write_flags = write_flags; |
fd9fcd37 | 4000 | |
ca4a0bb8 AG |
4001 | ret = qcow2_get_host_offset(bs, src_offset, &cur_bytes, |
4002 | ©_offset, &type); | |
fd9fcd37 FZ |
4003 | if (ret < 0) { |
4004 | goto out; | |
4005 | } | |
4006 | ||
ca4a0bb8 | 4007 | switch (type) { |
10dabdc5 | 4008 | case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: |
97490a14 | 4009 | case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: |
fd9fcd37 FZ |
4010 | if (bs->backing && bs->backing->bs) { |
4011 | int64_t backing_length = bdrv_getlength(bs->backing->bs); | |
4012 | if (src_offset >= backing_length) { | |
67b51fb9 | 4013 | cur_write_flags |= BDRV_REQ_ZERO_WRITE; |
fd9fcd37 FZ |
4014 | } else { |
4015 | child = bs->backing; | |
4016 | cur_bytes = MIN(cur_bytes, backing_length - src_offset); | |
4017 | copy_offset = src_offset; | |
4018 | } | |
4019 | } else { | |
67b51fb9 | 4020 | cur_write_flags |= BDRV_REQ_ZERO_WRITE; |
fd9fcd37 FZ |
4021 | } |
4022 | break; | |
4023 | ||
10dabdc5 AG |
4024 | case QCOW2_SUBCLUSTER_ZERO_PLAIN: |
4025 | case QCOW2_SUBCLUSTER_ZERO_ALLOC: | |
67b51fb9 | 4026 | cur_write_flags |= BDRV_REQ_ZERO_WRITE; |
fd9fcd37 FZ |
4027 | break; |
4028 | ||
10dabdc5 | 4029 | case QCOW2_SUBCLUSTER_COMPRESSED: |
fd9fcd37 FZ |
4030 | ret = -ENOTSUP; |
4031 | goto out; | |
fd9fcd37 | 4032 | |
10dabdc5 | 4033 | case QCOW2_SUBCLUSTER_NORMAL: |
966b000f | 4034 | child = s->data_file; |
fd9fcd37 FZ |
4035 | break; |
4036 | ||
4037 | default: | |
4038 | abort(); | |
4039 | } | |
4040 | qemu_co_mutex_unlock(&s->lock); | |
4041 | ret = bdrv_co_copy_range_from(child, | |
4042 | copy_offset, | |
4043 | dst, dst_offset, | |
67b51fb9 | 4044 | cur_bytes, read_flags, cur_write_flags); |
fd9fcd37 FZ |
4045 | qemu_co_mutex_lock(&s->lock); |
4046 | if (ret < 0) { | |
4047 | goto out; | |
4048 | } | |
4049 | ||
4050 | bytes -= cur_bytes; | |
4051 | src_offset += cur_bytes; | |
4052 | dst_offset += cur_bytes; | |
4053 | } | |
4054 | ret = 0; | |
4055 | ||
4056 | out: | |
4057 | qemu_co_mutex_unlock(&s->lock); | |
4058 | return ret; | |
4059 | } | |
4060 | ||
4061 | static int coroutine_fn | |
4062 | qcow2_co_copy_range_to(BlockDriverState *bs, | |
4063 | BdrvChild *src, uint64_t src_offset, | |
4064 | BdrvChild *dst, uint64_t dst_offset, | |
67b51fb9 VSO |
4065 | uint64_t bytes, BdrvRequestFlags read_flags, |
4066 | BdrvRequestFlags write_flags) | |
fd9fcd37 FZ |
4067 | { |
4068 | BDRVQcow2State *s = bs->opaque; | |
fd9fcd37 FZ |
4069 | int ret; |
4070 | unsigned int cur_bytes; /* number of sectors in current iteration */ | |
bfd0989a | 4071 | uint64_t host_offset; |
fd9fcd37 FZ |
4072 | QCowL2Meta *l2meta = NULL; |
4073 | ||
4074 | assert(!bs->encrypted); | |
fd9fcd37 FZ |
4075 | |
4076 | qemu_co_mutex_lock(&s->lock); | |
4077 | ||
4078 | while (bytes != 0) { | |
4079 | ||
4080 | l2meta = NULL; | |
4081 | ||
fd9fcd37 FZ |
4082 | cur_bytes = MIN(bytes, INT_MAX); |
4083 | ||
4084 | /* TODO: | |
4085 | * If src->bs == dst->bs, we could simply copy by incrementing | |
4086 | * the refcnt, without copying user data. | |
4087 | * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */ | |
bfd0989a AG |
4088 | ret = qcow2_alloc_host_offset(bs, dst_offset, &cur_bytes, |
4089 | &host_offset, &l2meta); | |
fd9fcd37 FZ |
4090 | if (ret < 0) { |
4091 | goto fail; | |
4092 | } | |
4093 | ||
bfd0989a AG |
4094 | ret = qcow2_pre_write_overlap_check(bs, 0, host_offset, cur_bytes, |
4095 | true); | |
fd9fcd37 FZ |
4096 | if (ret < 0) { |
4097 | goto fail; | |
4098 | } | |
4099 | ||
4100 | qemu_co_mutex_unlock(&s->lock); | |
bfd0989a | 4101 | ret = bdrv_co_copy_range_to(src, src_offset, s->data_file, host_offset, |
67b51fb9 | 4102 | cur_bytes, read_flags, write_flags); |
fd9fcd37 FZ |
4103 | qemu_co_mutex_lock(&s->lock); |
4104 | if (ret < 0) { | |
4105 | goto fail; | |
4106 | } | |
4107 | ||
4108 | ret = qcow2_handle_l2meta(bs, &l2meta, true); | |
4109 | if (ret) { | |
4110 | goto fail; | |
4111 | } | |
4112 | ||
4113 | bytes -= cur_bytes; | |
e06f4639 | 4114 | src_offset += cur_bytes; |
fd9fcd37 FZ |
4115 | dst_offset += cur_bytes; |
4116 | } | |
4117 | ret = 0; | |
4118 | ||
4119 | fail: | |
4120 | qcow2_handle_l2meta(bs, &l2meta, false); | |
4121 | ||
4122 | qemu_co_mutex_unlock(&s->lock); | |
4123 | ||
fd9fcd37 FZ |
4124 | trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); |
4125 | ||
4126 | return ret; | |
4127 | } | |
4128 | ||
061ca8a3 | 4129 | static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, |
c80d8b06 | 4130 | bool exact, PreallocMode prealloc, |
92b92799 | 4131 | BdrvRequestFlags flags, Error **errp) |
419b19d9 | 4132 | { |
ff99129a | 4133 | BDRVQcow2State *s = bs->opaque; |
95b98f34 | 4134 | uint64_t old_length; |
2cf7cfa1 KW |
4135 | int64_t new_l1_size; |
4136 | int ret; | |
45b4949c | 4137 | QDict *options; |
419b19d9 | 4138 | |
772d1f97 HR |
4139 | if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA && |
4140 | prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL) | |
4141 | { | |
8243ccb7 | 4142 | error_setg(errp, "Unsupported preallocation mode '%s'", |
977c736f | 4143 | PreallocMode_str(prealloc)); |
8243ccb7 HR |
4144 | return -ENOTSUP; |
4145 | } | |
4146 | ||
3afea402 AG |
4147 | if (!QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)) { |
4148 | error_setg(errp, "The new size must be a multiple of %u", | |
4149 | (unsigned) BDRV_SECTOR_SIZE); | |
419b19d9 SH |
4150 | return -EINVAL; |
4151 | } | |
4152 | ||
061ca8a3 KW |
4153 | qemu_co_mutex_lock(&s->lock); |
4154 | ||
7fa140ab EB |
4155 | /* |
4156 | * Even though we store snapshot size for all images, it was not | |
4157 | * required until v3, so it is not safe to proceed for v2. | |
4158 | */ | |
4159 | if (s->nb_snapshots && s->qcow_version < 3) { | |
4160 | error_setg(errp, "Can't resize a v2 image which has snapshots"); | |
061ca8a3 KW |
4161 | ret = -ENOTSUP; |
4162 | goto fail; | |
419b19d9 SH |
4163 | } |
4164 | ||
ee1244a2 | 4165 | /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */ |
d19c6b36 | 4166 | if (qcow2_truncate_bitmaps_check(bs, errp)) { |
061ca8a3 KW |
4167 | ret = -ENOTSUP; |
4168 | goto fail; | |
88ddffae VSO |
4169 | } |
4170 | ||
bd016b91 | 4171 | old_length = bs->total_sectors * BDRV_SECTOR_SIZE; |
46b732cd | 4172 | new_l1_size = size_to_l1(s, offset); |
95b98f34 | 4173 | |
95b98f34 | 4174 | if (offset < old_length) { |
163bc39d | 4175 | int64_t last_cluster, old_file_size; |
46b732cd PB |
4176 | if (prealloc != PREALLOC_MODE_OFF) { |
4177 | error_setg(errp, | |
4178 | "Preallocation can't be used for shrinking an image"); | |
061ca8a3 KW |
4179 | ret = -EINVAL; |
4180 | goto fail; | |
46b732cd | 4181 | } |
419b19d9 | 4182 | |
46b732cd PB |
4183 | ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size), |
4184 | old_length - ROUND_UP(offset, | |
4185 | s->cluster_size), | |
4186 | QCOW2_DISCARD_ALWAYS, true); | |
4187 | if (ret < 0) { | |
4188 | error_setg_errno(errp, -ret, "Failed to discard cropped clusters"); | |
061ca8a3 | 4189 | goto fail; |
46b732cd PB |
4190 | } |
4191 | ||
4192 | ret = qcow2_shrink_l1_table(bs, new_l1_size); | |
4193 | if (ret < 0) { | |
4194 | error_setg_errno(errp, -ret, | |
4195 | "Failed to reduce the number of L2 tables"); | |
061ca8a3 | 4196 | goto fail; |
46b732cd PB |
4197 | } |
4198 | ||
4199 | ret = qcow2_shrink_reftable(bs); | |
4200 | if (ret < 0) { | |
4201 | error_setg_errno(errp, -ret, | |
4202 | "Failed to discard unused refblocks"); | |
061ca8a3 | 4203 | goto fail; |
46b732cd | 4204 | } |
163bc39d PB |
4205 | |
4206 | old_file_size = bdrv_getlength(bs->file->bs); | |
4207 | if (old_file_size < 0) { | |
4208 | error_setg_errno(errp, -old_file_size, | |
4209 | "Failed to inquire current file length"); | |
061ca8a3 KW |
4210 | ret = old_file_size; |
4211 | goto fail; | |
163bc39d PB |
4212 | } |
4213 | last_cluster = qcow2_get_last_cluster(bs, old_file_size); | |
4214 | if (last_cluster < 0) { | |
4215 | error_setg_errno(errp, -last_cluster, | |
4216 | "Failed to find the last cluster"); | |
061ca8a3 KW |
4217 | ret = last_cluster; |
4218 | goto fail; | |
163bc39d PB |
4219 | } |
4220 | if ((last_cluster + 1) * s->cluster_size < old_file_size) { | |
233521b1 HR |
4221 | Error *local_err = NULL; |
4222 | ||
e61a28a9 HR |
4223 | /* |
4224 | * Do not pass @exact here: It will not help the user if | |
4225 | * we get an error here just because they wanted to shrink | |
4226 | * their qcow2 image (on a block device) with qemu-img. | |
4227 | * (And on the qcow2 layer, the @exact requirement is | |
4228 | * always fulfilled, so there is no need to pass it on.) | |
4229 | */ | |
061ca8a3 | 4230 | bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size, |
7b8e4857 | 4231 | false, PREALLOC_MODE_OFF, 0, &local_err); |
233521b1 HR |
4232 | if (local_err) { |
4233 | warn_reportf_err(local_err, | |
4234 | "Failed to truncate the tail of the image: "); | |
163bc39d PB |
4235 | } |
4236 | } | |
46b732cd PB |
4237 | } else { |
4238 | ret = qcow2_grow_l1_table(bs, new_l1_size, true); | |
4239 | if (ret < 0) { | |
4240 | error_setg_errno(errp, -ret, "Failed to grow the L1 table"); | |
061ca8a3 | 4241 | goto fail; |
46b732cd | 4242 | } |
419b19d9 SH |
4243 | } |
4244 | ||
95b98f34 HR |
4245 | switch (prealloc) { |
4246 | case PREALLOC_MODE_OFF: | |
718c0fce | 4247 | if (has_data_file(bs)) { |
e61a28a9 HR |
4248 | /* |
4249 | * If the caller wants an exact resize, the external data | |
4250 | * file should be resized to the exact target size, too, | |
4251 | * so we pass @exact here. | |
4252 | */ | |
7b8e4857 KW |
4253 | ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, 0, |
4254 | errp); | |
718c0fce KW |
4255 | if (ret < 0) { |
4256 | goto fail; | |
4257 | } | |
4258 | } | |
95b98f34 HR |
4259 | break; |
4260 | ||
4261 | case PREALLOC_MODE_METADATA: | |
718c0fce | 4262 | ret = preallocate_co(bs, old_length, offset, prealloc, errp); |
95b98f34 | 4263 | if (ret < 0) { |
061ca8a3 | 4264 | goto fail; |
95b98f34 HR |
4265 | } |
4266 | break; | |
4267 | ||
772d1f97 HR |
4268 | case PREALLOC_MODE_FALLOC: |
4269 | case PREALLOC_MODE_FULL: | |
4270 | { | |
4271 | int64_t allocation_start, host_offset, guest_offset; | |
4272 | int64_t clusters_allocated; | |
4b96fa38 | 4273 | int64_t old_file_size, last_cluster, new_file_size; |
772d1f97 | 4274 | uint64_t nb_new_data_clusters, nb_new_l2_tables; |
40dee943 | 4275 | bool subclusters_need_allocation = false; |
772d1f97 | 4276 | |
966b000f KW |
4277 | /* With a data file, preallocation means just allocating the metadata |
4278 | * and forwarding the truncate request to the data file */ | |
4279 | if (has_data_file(bs)) { | |
718c0fce | 4280 | ret = preallocate_co(bs, old_length, offset, prealloc, errp); |
966b000f | 4281 | if (ret < 0) { |
966b000f KW |
4282 | goto fail; |
4283 | } | |
4284 | break; | |
4285 | } | |
4286 | ||
772d1f97 HR |
4287 | old_file_size = bdrv_getlength(bs->file->bs); |
4288 | if (old_file_size < 0) { | |
4289 | error_setg_errno(errp, -old_file_size, | |
4290 | "Failed to inquire current file length"); | |
061ca8a3 KW |
4291 | ret = old_file_size; |
4292 | goto fail; | |
772d1f97 | 4293 | } |
4b96fa38 HR |
4294 | |
4295 | last_cluster = qcow2_get_last_cluster(bs, old_file_size); | |
4296 | if (last_cluster >= 0) { | |
4297 | old_file_size = (last_cluster + 1) * s->cluster_size; | |
4298 | } else { | |
4299 | old_file_size = ROUND_UP(old_file_size, s->cluster_size); | |
4300 | } | |
772d1f97 | 4301 | |
a5675f39 AG |
4302 | nb_new_data_clusters = (ROUND_UP(offset, s->cluster_size) - |
4303 | start_of_cluster(s, old_length)) >> s->cluster_bits; | |
772d1f97 HR |
4304 | |
4305 | /* This is an overestimation; we will not actually allocate space for | |
4306 | * these in the file but just make sure the new refcount structures are | |
4307 | * able to cover them so we will not have to allocate new refblocks | |
4308 | * while entering the data blocks in the potentially new L2 tables. | |
4309 | * (We do not actually care where the L2 tables are placed. Maybe they | |
4310 | * are already allocated or they can be placed somewhere before | |
4311 | * @old_file_size. It does not matter because they will be fully | |
4312 | * allocated automatically, so they do not need to be covered by the | |
4313 | * preallocation. All that matters is that we will not have to allocate | |
4314 | * new refcount structures for them.) */ | |
4315 | nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters, | |
c8fd8554 | 4316 | s->cluster_size / l2_entry_size(s)); |
772d1f97 HR |
4317 | /* The cluster range may not be aligned to L2 boundaries, so add one L2 |
4318 | * table for a potential head/tail */ | |
4319 | nb_new_l2_tables++; | |
4320 | ||
4321 | allocation_start = qcow2_refcount_area(bs, old_file_size, | |
4322 | nb_new_data_clusters + | |
4323 | nb_new_l2_tables, | |
4324 | true, 0, 0); | |
4325 | if (allocation_start < 0) { | |
4326 | error_setg_errno(errp, -allocation_start, | |
4327 | "Failed to resize refcount structures"); | |
061ca8a3 KW |
4328 | ret = allocation_start; |
4329 | goto fail; | |
772d1f97 HR |
4330 | } |
4331 | ||
4332 | clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start, | |
4333 | nb_new_data_clusters); | |
4334 | if (clusters_allocated < 0) { | |
4335 | error_setg_errno(errp, -clusters_allocated, | |
4336 | "Failed to allocate data clusters"); | |
061ca8a3 KW |
4337 | ret = clusters_allocated; |
4338 | goto fail; | |
772d1f97 HR |
4339 | } |
4340 | ||
4341 | assert(clusters_allocated == nb_new_data_clusters); | |
4342 | ||
4343 | /* Allocate the data area */ | |
4344 | new_file_size = allocation_start + | |
4345 | nb_new_data_clusters * s->cluster_size; | |
eb8a0cf3 KW |
4346 | /* |
4347 | * Image file grows, so @exact does not matter. | |
4348 | * | |
4349 | * If we need to zero out the new area, try first whether the protocol | |
4350 | * driver can already take care of this. | |
4351 | */ | |
4352 | if (flags & BDRV_REQ_ZERO_WRITE) { | |
4353 | ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, | |
4354 | BDRV_REQ_ZERO_WRITE, NULL); | |
4355 | if (ret >= 0) { | |
4356 | flags &= ~BDRV_REQ_ZERO_WRITE; | |
40dee943 AG |
4357 | /* Ensure that we read zeroes and not backing file data */ |
4358 | subclusters_need_allocation = true; | |
eb8a0cf3 KW |
4359 | } |
4360 | } else { | |
4361 | ret = -1; | |
4362 | } | |
4363 | if (ret < 0) { | |
4364 | ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0, | |
4365 | errp); | |
4366 | } | |
772d1f97 HR |
4367 | if (ret < 0) { |
4368 | error_prepend(errp, "Failed to resize underlying file: "); | |
4369 | qcow2_free_clusters(bs, allocation_start, | |
4370 | nb_new_data_clusters * s->cluster_size, | |
4371 | QCOW2_DISCARD_OTHER); | |
061ca8a3 | 4372 | goto fail; |
772d1f97 HR |
4373 | } |
4374 | ||
4375 | /* Create the necessary L2 entries */ | |
4376 | host_offset = allocation_start; | |
4377 | guest_offset = old_length; | |
4378 | while (nb_new_data_clusters) { | |
13bec229 AG |
4379 | int64_t nb_clusters = MIN( |
4380 | nb_new_data_clusters, | |
4381 | s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset)); | |
a5675f39 AG |
4382 | unsigned cow_start_length = offset_into_cluster(s, guest_offset); |
4383 | QCowL2Meta allocation; | |
4384 | guest_offset = start_of_cluster(s, guest_offset); | |
4385 | allocation = (QCowL2Meta) { | |
772d1f97 HR |
4386 | .offset = guest_offset, |
4387 | .alloc_offset = host_offset, | |
4388 | .nb_clusters = nb_clusters, | |
a5675f39 AG |
4389 | .cow_start = { |
4390 | .offset = 0, | |
4391 | .nb_bytes = cow_start_length, | |
4392 | }, | |
4393 | .cow_end = { | |
4394 | .offset = nb_clusters << s->cluster_bits, | |
4395 | .nb_bytes = 0, | |
4396 | }, | |
40dee943 | 4397 | .prealloc = !subclusters_need_allocation, |
772d1f97 HR |
4398 | }; |
4399 | qemu_co_queue_init(&allocation.dependent_requests); | |
4400 | ||
4401 | ret = qcow2_alloc_cluster_link_l2(bs, &allocation); | |
4402 | if (ret < 0) { | |
4403 | error_setg_errno(errp, -ret, "Failed to update L2 tables"); | |
4404 | qcow2_free_clusters(bs, host_offset, | |
4405 | nb_new_data_clusters * s->cluster_size, | |
4406 | QCOW2_DISCARD_OTHER); | |
061ca8a3 | 4407 | goto fail; |
772d1f97 HR |
4408 | } |
4409 | ||
4410 | guest_offset += nb_clusters * s->cluster_size; | |
4411 | host_offset += nb_clusters * s->cluster_size; | |
4412 | nb_new_data_clusters -= nb_clusters; | |
4413 | } | |
4414 | break; | |
4415 | } | |
4416 | ||
95b98f34 HR |
4417 | default: |
4418 | g_assert_not_reached(); | |
4419 | } | |
4420 | ||
f01643fb | 4421 | if ((flags & BDRV_REQ_ZERO_WRITE) && offset > old_length) { |
a6841a2d | 4422 | uint64_t zero_start = QEMU_ALIGN_UP(old_length, s->subcluster_size); |
f01643fb KW |
4423 | |
4424 | /* | |
a6841a2d AG |
4425 | * Use zero clusters as much as we can. qcow2_subcluster_zeroize() |
4426 | * requires a subcluster-aligned start. The end may be unaligned if | |
4427 | * it is at the end of the image (which it is here). | |
f01643fb | 4428 | */ |
e4d7019e | 4429 | if (offset > zero_start) { |
a6841a2d AG |
4430 | ret = qcow2_subcluster_zeroize(bs, zero_start, offset - zero_start, |
4431 | 0); | |
e4d7019e AG |
4432 | if (ret < 0) { |
4433 | error_setg_errno(errp, -ret, "Failed to zero out new clusters"); | |
4434 | goto fail; | |
4435 | } | |
f01643fb KW |
4436 | } |
4437 | ||
4438 | /* Write explicit zeros for the unaligned head */ | |
4439 | if (zero_start > old_length) { | |
e4d7019e | 4440 | uint64_t len = MIN(zero_start, offset) - old_length; |
f01643fb KW |
4441 | uint8_t *buf = qemu_blockalign0(bs, len); |
4442 | QEMUIOVector qiov; | |
4443 | qemu_iovec_init_buf(&qiov, buf, len); | |
4444 | ||
4445 | qemu_co_mutex_unlock(&s->lock); | |
4446 | ret = qcow2_co_pwritev_part(bs, old_length, len, &qiov, 0, 0); | |
4447 | qemu_co_mutex_lock(&s->lock); | |
4448 | ||
4449 | qemu_vfree(buf); | |
4450 | if (ret < 0) { | |
4451 | error_setg_errno(errp, -ret, "Failed to zero out the new area"); | |
4452 | goto fail; | |
4453 | } | |
4454 | } | |
4455 | } | |
4456 | ||
95b98f34 HR |
4457 | if (prealloc != PREALLOC_MODE_OFF) { |
4458 | /* Flush metadata before actually changing the image size */ | |
061ca8a3 | 4459 | ret = qcow2_write_caches(bs); |
95b98f34 HR |
4460 | if (ret < 0) { |
4461 | error_setg_errno(errp, -ret, | |
4462 | "Failed to flush the preallocated area to disk"); | |
061ca8a3 | 4463 | goto fail; |
95b98f34 HR |
4464 | } |
4465 | } | |
4466 | ||
45b4949c LB |
4467 | bs->total_sectors = offset / BDRV_SECTOR_SIZE; |
4468 | ||
419b19d9 SH |
4469 | /* write updated header.size */ |
4470 | offset = cpu_to_be64(offset); | |
d9ca2ea2 | 4471 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), |
02b1ecfa | 4472 | &offset, sizeof(offset)); |
419b19d9 | 4473 | if (ret < 0) { |
f59adb32 | 4474 | error_setg_errno(errp, -ret, "Failed to update the image size"); |
061ca8a3 | 4475 | goto fail; |
419b19d9 SH |
4476 | } |
4477 | ||
4478 | s->l1_vm_state_index = new_l1_size; | |
45b4949c LB |
4479 | |
4480 | /* Update cache sizes */ | |
4481 | options = qdict_clone_shallow(bs->options); | |
4482 | ret = qcow2_update_options(bs, options, s->flags, errp); | |
4483 | qobject_unref(options); | |
4484 | if (ret < 0) { | |
4485 | goto fail; | |
4486 | } | |
061ca8a3 KW |
4487 | ret = 0; |
4488 | fail: | |
4489 | qemu_co_mutex_unlock(&s->lock); | |
4490 | return ret; | |
419b19d9 SH |
4491 | } |
4492 | ||
fcccefc5 | 4493 | static coroutine_fn int |
0d483dce | 4494 | qcow2_co_pwritev_compressed_task(BlockDriverState *bs, |
5396234b VSO |
4495 | uint64_t offset, uint64_t bytes, |
4496 | QEMUIOVector *qiov, size_t qiov_offset) | |
20d97356 | 4497 | { |
ff99129a | 4498 | BDRVQcow2State *s = bs->opaque; |
2714f13d | 4499 | int ret; |
e1f4a37a | 4500 | ssize_t out_len; |
fcccefc5 | 4501 | uint8_t *buf, *out_buf; |
77e023ff | 4502 | uint64_t cluster_offset; |
20d97356 | 4503 | |
0d483dce AS |
4504 | assert(bytes == s->cluster_size || (bytes < s->cluster_size && |
4505 | (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS))); | |
3e3b838f | 4506 | |
a2c0ca6f | 4507 | buf = qemu_blockalign(bs, s->cluster_size); |
0d483dce | 4508 | if (bytes < s->cluster_size) { |
a2c0ca6f PB |
4509 | /* Zero-pad last write if image size is not cluster aligned */ |
4510 | memset(buf + bytes, 0, s->cluster_size - bytes); | |
f4d38bef | 4511 | } |
5396234b | 4512 | qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes); |
20d97356 | 4513 | |
ebf7bba0 | 4514 | out_buf = g_malloc(s->cluster_size); |
20d97356 | 4515 | |
6994fd78 VSO |
4516 | out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1, |
4517 | buf, s->cluster_size); | |
e1f4a37a | 4518 | if (out_len == -ENOMEM) { |
20d97356 | 4519 | /* could not compress: write normal cluster */ |
5396234b | 4520 | ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0); |
8f1efd00 KW |
4521 | if (ret < 0) { |
4522 | goto fail; | |
4523 | } | |
fcccefc5 | 4524 | goto success; |
e1f4a37a AG |
4525 | } else if (out_len < 0) { |
4526 | ret = -EINVAL; | |
4527 | goto fail; | |
fcccefc5 | 4528 | } |
cf93980e | 4529 | |
fcccefc5 | 4530 | qemu_co_mutex_lock(&s->lock); |
77e023ff KW |
4531 | ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len, |
4532 | &cluster_offset); | |
4533 | if (ret < 0) { | |
fcccefc5 | 4534 | qemu_co_mutex_unlock(&s->lock); |
fcccefc5 PB |
4535 | goto fail; |
4536 | } | |
cf93980e | 4537 | |
966b000f | 4538 | ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true); |
fcccefc5 PB |
4539 | qemu_co_mutex_unlock(&s->lock); |
4540 | if (ret < 0) { | |
4541 | goto fail; | |
20d97356 BS |
4542 | } |
4543 | ||
966b000f | 4544 | BLKDBG_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED); |
b00cb15b | 4545 | ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0); |
fcccefc5 PB |
4546 | if (ret < 0) { |
4547 | goto fail; | |
4548 | } | |
4549 | success: | |
8f1efd00 KW |
4550 | ret = 0; |
4551 | fail: | |
fcccefc5 | 4552 | qemu_vfree(buf); |
7267c094 | 4553 | g_free(out_buf); |
8f1efd00 | 4554 | return ret; |
20d97356 BS |
4555 | } |
4556 | ||
0d483dce AS |
4557 | static coroutine_fn int qcow2_co_pwritev_compressed_task_entry(AioTask *task) |
4558 | { | |
4559 | Qcow2AioTask *t = container_of(task, Qcow2AioTask, task); | |
4560 | ||
10dabdc5 | 4561 | assert(!t->subcluster_type && !t->l2meta); |
0d483dce AS |
4562 | |
4563 | return qcow2_co_pwritev_compressed_task(t->bs, t->offset, t->bytes, t->qiov, | |
4564 | t->qiov_offset); | |
4565 | } | |
4566 | ||
4567 | /* | |
4568 | * XXX: put compressed sectors first, then all the cluster aligned | |
4569 | * tables to avoid losing bytes in alignment | |
4570 | */ | |
4571 | static coroutine_fn int | |
4572 | qcow2_co_pwritev_compressed_part(BlockDriverState *bs, | |
4573 | uint64_t offset, uint64_t bytes, | |
4574 | QEMUIOVector *qiov, size_t qiov_offset) | |
4575 | { | |
4576 | BDRVQcow2State *s = bs->opaque; | |
4577 | AioTaskPool *aio = NULL; | |
4578 | int ret = 0; | |
4579 | ||
4580 | if (has_data_file(bs)) { | |
4581 | return -ENOTSUP; | |
4582 | } | |
4583 | ||
4584 | if (bytes == 0) { | |
4585 | /* | |
4586 | * align end of file to a sector boundary to ease reading with | |
4587 | * sector based I/Os | |
4588 | */ | |
4589 | int64_t len = bdrv_getlength(bs->file->bs); | |
4590 | if (len < 0) { | |
4591 | return len; | |
4592 | } | |
7b8e4857 KW |
4593 | return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, 0, |
4594 | NULL); | |
0d483dce AS |
4595 | } |
4596 | ||
4597 | if (offset_into_cluster(s, offset)) { | |
4598 | return -EINVAL; | |
4599 | } | |
4600 | ||
fb43d2d4 AG |
4601 | if (offset_into_cluster(s, bytes) && |
4602 | (offset + bytes) != (bs->total_sectors << BDRV_SECTOR_BITS)) { | |
4603 | return -EINVAL; | |
4604 | } | |
4605 | ||
0d483dce AS |
4606 | while (bytes && aio_task_pool_status(aio) == 0) { |
4607 | uint64_t chunk_size = MIN(bytes, s->cluster_size); | |
4608 | ||
4609 | if (!aio && chunk_size != bytes) { | |
4610 | aio = aio_task_pool_new(QCOW2_MAX_WORKERS); | |
4611 | } | |
4612 | ||
4613 | ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_compressed_task_entry, | |
4614 | 0, 0, offset, chunk_size, qiov, qiov_offset, NULL); | |
4615 | if (ret < 0) { | |
4616 | break; | |
4617 | } | |
4618 | qiov_offset += chunk_size; | |
4619 | offset += chunk_size; | |
4620 | bytes -= chunk_size; | |
4621 | } | |
4622 | ||
4623 | if (aio) { | |
4624 | aio_task_pool_wait_all(aio); | |
4625 | if (ret == 0) { | |
4626 | ret = aio_task_pool_status(aio); | |
4627 | } | |
4628 | g_free(aio); | |
4629 | } | |
4630 | ||
4631 | return ret; | |
4632 | } | |
4633 | ||
c3c10f72 VSO |
4634 | static int coroutine_fn |
4635 | qcow2_co_preadv_compressed(BlockDriverState *bs, | |
9c4269d5 | 4636 | uint64_t cluster_descriptor, |
c3c10f72 VSO |
4637 | uint64_t offset, |
4638 | uint64_t bytes, | |
df893d25 VSO |
4639 | QEMUIOVector *qiov, |
4640 | size_t qiov_offset) | |
f4b3e2a9 VSO |
4641 | { |
4642 | BDRVQcow2State *s = bs->opaque; | |
c3c10f72 | 4643 | int ret = 0, csize, nb_csectors; |
f4b3e2a9 | 4644 | uint64_t coffset; |
c3c10f72 | 4645 | uint8_t *buf, *out_buf; |
c3c10f72 | 4646 | int offset_in_cluster = offset_into_cluster(s, offset); |
f4b3e2a9 | 4647 | |
9c4269d5 AG |
4648 | coffset = cluster_descriptor & s->cluster_offset_mask; |
4649 | nb_csectors = ((cluster_descriptor >> s->csize_shift) & s->csize_mask) + 1; | |
b6c24694 AG |
4650 | csize = nb_csectors * QCOW2_COMPRESSED_SECTOR_SIZE - |
4651 | (coffset & ~QCOW2_COMPRESSED_SECTOR_MASK); | |
f4b3e2a9 | 4652 | |
c3c10f72 VSO |
4653 | buf = g_try_malloc(csize); |
4654 | if (!buf) { | |
4655 | return -ENOMEM; | |
4656 | } | |
f4b3e2a9 | 4657 | |
c3c10f72 | 4658 | out_buf = qemu_blockalign(bs, s->cluster_size); |
c068a1cd | 4659 | |
c3c10f72 | 4660 | BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
b00cb15b | 4661 | ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0); |
c3c10f72 VSO |
4662 | if (ret < 0) { |
4663 | goto fail; | |
f4b3e2a9 | 4664 | } |
c3c10f72 | 4665 | |
e23c9d7a | 4666 | if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) { |
c3c10f72 VSO |
4667 | ret = -EIO; |
4668 | goto fail; | |
4669 | } | |
4670 | ||
df893d25 | 4671 | qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes); |
c3c10f72 VSO |
4672 | |
4673 | fail: | |
4674 | qemu_vfree(out_buf); | |
4675 | g_free(buf); | |
4676 | ||
4677 | return ret; | |
f4b3e2a9 VSO |
4678 | } |
4679 | ||
94054183 HR |
4680 | static int make_completely_empty(BlockDriverState *bs) |
4681 | { | |
ff99129a | 4682 | BDRVQcow2State *s = bs->opaque; |
ed3d2ec9 | 4683 | Error *local_err = NULL; |
94054183 HR |
4684 | int ret, l1_clusters; |
4685 | int64_t offset; | |
4686 | uint64_t *new_reftable = NULL; | |
4687 | uint64_t rt_entry, l1_size2; | |
4688 | struct { | |
4689 | uint64_t l1_offset; | |
4690 | uint64_t reftable_offset; | |
4691 | uint32_t reftable_clusters; | |
4692 | } QEMU_PACKED l1_ofs_rt_ofs_cls; | |
4693 | ||
4694 | ret = qcow2_cache_empty(bs, s->l2_table_cache); | |
4695 | if (ret < 0) { | |
4696 | goto fail; | |
4697 | } | |
4698 | ||
4699 | ret = qcow2_cache_empty(bs, s->refcount_block_cache); | |
4700 | if (ret < 0) { | |
4701 | goto fail; | |
4702 | } | |
4703 | ||
4704 | /* Refcounts will be broken utterly */ | |
4705 | ret = qcow2_mark_dirty(bs); | |
4706 | if (ret < 0) { | |
4707 | goto fail; | |
4708 | } | |
4709 | ||
4710 | BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); | |
4711 | ||
02b1ecfa AG |
4712 | l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE); |
4713 | l1_size2 = (uint64_t)s->l1_size * L1E_SIZE; | |
94054183 HR |
4714 | |
4715 | /* After this call, neither the in-memory nor the on-disk refcount | |
4716 | * information accurately describe the actual references */ | |
4717 | ||
720ff280 | 4718 | ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset, |
74021bc4 | 4719 | l1_clusters * s->cluster_size, 0); |
94054183 HR |
4720 | if (ret < 0) { |
4721 | goto fail_broken_refcounts; | |
4722 | } | |
4723 | memset(s->l1_table, 0, l1_size2); | |
4724 | ||
4725 | BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE); | |
4726 | ||
4727 | /* Overwrite enough clusters at the beginning of the sectors to place | |
4728 | * the refcount table, a refcount block and the L1 table in; this may | |
4729 | * overwrite parts of the existing refcount and L1 table, which is not | |
4730 | * an issue because the dirty flag is set, complete data loss is in fact | |
4731 | * desired and partial data loss is consequently fine as well */ | |
720ff280 | 4732 | ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size, |
74021bc4 | 4733 | (2 + l1_clusters) * s->cluster_size, 0); |
94054183 HR |
4734 | /* This call (even if it failed overall) may have overwritten on-disk |
4735 | * refcount structures; in that case, the in-memory refcount information | |
4736 | * will probably differ from the on-disk information which makes the BDS | |
4737 | * unusable */ | |
4738 | if (ret < 0) { | |
4739 | goto fail_broken_refcounts; | |
4740 | } | |
4741 | ||
4742 | BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); | |
4743 | BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE); | |
4744 | ||
4745 | /* "Create" an empty reftable (one cluster) directly after the image | |
4746 | * header and an empty L1 table three clusters after the image header; | |
4747 | * the cluster between those two will be used as the first refblock */ | |
f1f7a1dd PM |
4748 | l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size); |
4749 | l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size); | |
4750 | l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1); | |
d9ca2ea2 | 4751 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset), |
94054183 HR |
4752 | &l1_ofs_rt_ofs_cls, sizeof(l1_ofs_rt_ofs_cls)); |
4753 | if (ret < 0) { | |
4754 | goto fail_broken_refcounts; | |
4755 | } | |
4756 | ||
4757 | s->l1_table_offset = 3 * s->cluster_size; | |
4758 | ||
02b1ecfa | 4759 | new_reftable = g_try_new0(uint64_t, s->cluster_size / REFTABLE_ENTRY_SIZE); |
94054183 HR |
4760 | if (!new_reftable) { |
4761 | ret = -ENOMEM; | |
4762 | goto fail_broken_refcounts; | |
4763 | } | |
4764 | ||
4765 | s->refcount_table_offset = s->cluster_size; | |
02b1ecfa | 4766 | s->refcount_table_size = s->cluster_size / REFTABLE_ENTRY_SIZE; |
7061a078 | 4767 | s->max_refcount_table_index = 0; |
94054183 HR |
4768 | |
4769 | g_free(s->refcount_table); | |
4770 | s->refcount_table = new_reftable; | |
4771 | new_reftable = NULL; | |
4772 | ||
4773 | /* Now the in-memory refcount information again corresponds to the on-disk | |
4774 | * information (reftable is empty and no refblocks (the refblock cache is | |
4775 | * empty)); however, this means some clusters (e.g. the image header) are | |
4776 | * referenced, but not refcounted, but the normal qcow2 code assumes that | |
4777 | * the in-memory information is always correct */ | |
4778 | ||
4779 | BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC); | |
4780 | ||
4781 | /* Enter the first refblock into the reftable */ | |
4782 | rt_entry = cpu_to_be64(2 * s->cluster_size); | |
d9ca2ea2 | 4783 | ret = bdrv_pwrite_sync(bs->file, s->cluster_size, |
94054183 HR |
4784 | &rt_entry, sizeof(rt_entry)); |
4785 | if (ret < 0) { | |
4786 | goto fail_broken_refcounts; | |
4787 | } | |
4788 | s->refcount_table[0] = 2 * s->cluster_size; | |
4789 | ||
4790 | s->free_cluster_index = 0; | |
4791 | assert(3 + l1_clusters <= s->refcount_block_size); | |
4792 | offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2); | |
4793 | if (offset < 0) { | |
4794 | ret = offset; | |
4795 | goto fail_broken_refcounts; | |
4796 | } else if (offset > 0) { | |
4797 | error_report("First cluster in emptied image is in use"); | |
4798 | abort(); | |
4799 | } | |
4800 | ||
4801 | /* Now finally the in-memory information corresponds to the on-disk | |
4802 | * structures and is correct */ | |
4803 | ret = qcow2_mark_clean(bs); | |
4804 | if (ret < 0) { | |
4805 | goto fail; | |
4806 | } | |
4807 | ||
c80d8b06 | 4808 | ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false, |
7b8e4857 | 4809 | PREALLOC_MODE_OFF, 0, &local_err); |
94054183 | 4810 | if (ret < 0) { |
ed3d2ec9 | 4811 | error_report_err(local_err); |
94054183 HR |
4812 | goto fail; |
4813 | } | |
4814 | ||
4815 | return 0; | |
4816 | ||
4817 | fail_broken_refcounts: | |
4818 | /* The BDS is unusable at this point. If we wanted to make it usable, we | |
4819 | * would have to call qcow2_refcount_close(), qcow2_refcount_init(), | |
4820 | * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init() | |
4821 | * again. However, because the functions which could have caused this error | |
4822 | * path to be taken are used by those functions as well, it's very likely | |
4823 | * that that sequence will fail as well. Therefore, just eject the BDS. */ | |
4824 | bs->drv = NULL; | |
4825 | ||
4826 | fail: | |
4827 | g_free(new_reftable); | |
4828 | return ret; | |
4829 | } | |
4830 | ||
491d27e2 HR |
4831 | static int qcow2_make_empty(BlockDriverState *bs) |
4832 | { | |
ff99129a | 4833 | BDRVQcow2State *s = bs->opaque; |
d2cb36af EB |
4834 | uint64_t offset, end_offset; |
4835 | int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size); | |
94054183 HR |
4836 | int l1_clusters, ret = 0; |
4837 | ||
02b1ecfa | 4838 | l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE); |
94054183 | 4839 | |
4096974e | 4840 | if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps && |
f0603329 | 4841 | 3 + l1_clusters <= s->refcount_block_size && |
db04524f KW |
4842 | s->crypt_method_header != QCOW_CRYPT_LUKS && |
4843 | !has_data_file(bs)) { | |
4096974e EB |
4844 | /* The following function only works for qcow2 v3 images (it |
4845 | * requires the dirty flag) and only as long as there are no | |
4846 | * features that reserve extra clusters (such as snapshots, | |
4847 | * LUKS header, or persistent bitmaps), because it completely | |
4848 | * empties the image. Furthermore, the L1 table and three | |
4849 | * additional clusters (image header, refcount table, one | |
db04524f KW |
4850 | * refcount block) have to fit inside one refcount block. It |
4851 | * only resets the image file, i.e. does not work with an | |
4852 | * external data file. */ | |
94054183 HR |
4853 | return make_completely_empty(bs); |
4854 | } | |
491d27e2 | 4855 | |
94054183 HR |
4856 | /* This fallback code simply discards every active cluster; this is slow, |
4857 | * but works in all cases */ | |
d2cb36af EB |
4858 | end_offset = bs->total_sectors * BDRV_SECTOR_SIZE; |
4859 | for (offset = 0; offset < end_offset; offset += step) { | |
491d27e2 HR |
4860 | /* As this function is generally used after committing an external |
4861 | * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the | |
4862 | * default action for this kind of discard is to pass the discard, | |
4863 | * which will ideally result in an actually smaller image file, as | |
4864 | * is probably desired. */ | |
d2cb36af EB |
4865 | ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset), |
4866 | QCOW2_DISCARD_SNAPSHOT, true); | |
491d27e2 HR |
4867 | if (ret < 0) { |
4868 | break; | |
4869 | } | |
4870 | } | |
4871 | ||
4872 | return ret; | |
4873 | } | |
4874 | ||
a968168c | 4875 | static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) |
20d97356 | 4876 | { |
ff99129a | 4877 | BDRVQcow2State *s = bs->opaque; |
29c1a730 KW |
4878 | int ret; |
4879 | ||
8b94ff85 | 4880 | qemu_co_mutex_lock(&s->lock); |
8b220eb7 | 4881 | ret = qcow2_write_caches(bs); |
8b94ff85 | 4882 | qemu_co_mutex_unlock(&s->lock); |
29c1a730 | 4883 | |
8b220eb7 | 4884 | return ret; |
eb489bb1 KW |
4885 | } |
4886 | ||
c501c352 SH |
4887 | static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs, |
4888 | Error **errp) | |
4889 | { | |
4890 | Error *local_err = NULL; | |
4891 | BlockMeasureInfo *info; | |
4892 | uint64_t required = 0; /* bytes that contribute to required size */ | |
4893 | uint64_t virtual_size; /* disk size as seen by guest */ | |
4894 | uint64_t refcount_bits; | |
4895 | uint64_t l2_tables; | |
61914f89 | 4896 | uint64_t luks_payload_size = 0; |
c501c352 SH |
4897 | size_t cluster_size; |
4898 | int version; | |
4899 | char *optstr; | |
4900 | PreallocMode prealloc; | |
4901 | bool has_backing_file; | |
61914f89 | 4902 | bool has_luks; |
7be20252 | 4903 | bool extended_l2; |
0dd07b29 | 4904 | size_t l2e_size; |
c501c352 SH |
4905 | |
4906 | /* Parse image creation options */ | |
7be20252 AG |
4907 | extended_l2 = qemu_opt_get_bool_del(opts, BLOCK_OPT_EXTL2, false); |
4908 | ||
4909 | cluster_size = qcow2_opt_get_cluster_size_del(opts, extended_l2, | |
4910 | &local_err); | |
c501c352 SH |
4911 | if (local_err) { |
4912 | goto err; | |
4913 | } | |
4914 | ||
4915 | version = qcow2_opt_get_version_del(opts, &local_err); | |
4916 | if (local_err) { | |
4917 | goto err; | |
4918 | } | |
4919 | ||
4920 | refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err); | |
4921 | if (local_err) { | |
4922 | goto err; | |
4923 | } | |
4924 | ||
4925 | optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
f7abe0ec | 4926 | prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr, |
06c60b6c | 4927 | PREALLOC_MODE_OFF, &local_err); |
c501c352 SH |
4928 | g_free(optstr); |
4929 | if (local_err) { | |
4930 | goto err; | |
4931 | } | |
4932 | ||
4933 | optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); | |
4934 | has_backing_file = !!optstr; | |
4935 | g_free(optstr); | |
4936 | ||
61914f89 SH |
4937 | optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT); |
4938 | has_luks = optstr && strcmp(optstr, "luks") == 0; | |
4939 | g_free(optstr); | |
4940 | ||
4941 | if (has_luks) { | |
6d49d3a8 | 4942 | g_autoptr(QCryptoBlockCreateOptions) create_opts = NULL; |
90766d9d | 4943 | QDict *cryptoopts = qcow2_extract_crypto_opts(opts, "luks", errp); |
61914f89 SH |
4944 | size_t headerlen; |
4945 | ||
6d49d3a8 SH |
4946 | create_opts = block_crypto_create_opts_init(cryptoopts, errp); |
4947 | qobject_unref(cryptoopts); | |
4948 | if (!create_opts) { | |
4949 | goto err; | |
4950 | } | |
4951 | ||
4952 | if (!qcrypto_block_calculate_payload_offset(create_opts, | |
4953 | "encrypt.", | |
4954 | &headerlen, | |
4955 | &local_err)) { | |
61914f89 SH |
4956 | goto err; |
4957 | } | |
4958 | ||
4959 | luks_payload_size = ROUND_UP(headerlen, cluster_size); | |
4960 | } | |
4961 | ||
9e029689 AG |
4962 | virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); |
4963 | virtual_size = ROUND_UP(virtual_size, cluster_size); | |
c501c352 SH |
4964 | |
4965 | /* Check that virtual disk size is valid */ | |
0dd07b29 | 4966 | l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL; |
c501c352 | 4967 | l2_tables = DIV_ROUND_UP(virtual_size / cluster_size, |
0dd07b29 | 4968 | cluster_size / l2e_size); |
02b1ecfa | 4969 | if (l2_tables * L1E_SIZE > QCOW_MAX_L1_SIZE) { |
c501c352 SH |
4970 | error_setg(&local_err, "The image size is too large " |
4971 | "(try using a larger cluster size)"); | |
4972 | goto err; | |
4973 | } | |
4974 | ||
4975 | /* Account for input image */ | |
4976 | if (in_bs) { | |
4977 | int64_t ssize = bdrv_getlength(in_bs); | |
4978 | if (ssize < 0) { | |
4979 | error_setg_errno(&local_err, -ssize, | |
4980 | "Unable to get image virtual_size"); | |
4981 | goto err; | |
4982 | } | |
4983 | ||
9e029689 | 4984 | virtual_size = ROUND_UP(ssize, cluster_size); |
c501c352 SH |
4985 | |
4986 | if (has_backing_file) { | |
4987 | /* We don't how much of the backing chain is shared by the input | |
4988 | * image and the new image file. In the worst case the new image's | |
4989 | * backing file has nothing in common with the input image. Be | |
4990 | * conservative and assume all clusters need to be written. | |
4991 | */ | |
4992 | required = virtual_size; | |
4993 | } else { | |
b85ee453 | 4994 | int64_t offset; |
31826642 | 4995 | int64_t pnum = 0; |
c501c352 | 4996 | |
31826642 EB |
4997 | for (offset = 0; offset < ssize; offset += pnum) { |
4998 | int ret; | |
c501c352 | 4999 | |
31826642 EB |
5000 | ret = bdrv_block_status_above(in_bs, NULL, offset, |
5001 | ssize - offset, &pnum, NULL, | |
5002 | NULL); | |
c501c352 SH |
5003 | if (ret < 0) { |
5004 | error_setg_errno(&local_err, -ret, | |
5005 | "Unable to get block status"); | |
5006 | goto err; | |
5007 | } | |
5008 | ||
5009 | if (ret & BDRV_BLOCK_ZERO) { | |
5010 | /* Skip zero regions (safe with no backing file) */ | |
5011 | } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) == | |
5012 | (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) { | |
5013 | /* Extend pnum to end of cluster for next iteration */ | |
31826642 | 5014 | pnum = ROUND_UP(offset + pnum, cluster_size) - offset; |
c501c352 SH |
5015 | |
5016 | /* Count clusters we've seen */ | |
31826642 | 5017 | required += offset % cluster_size + pnum; |
c501c352 SH |
5018 | } |
5019 | } | |
5020 | } | |
5021 | } | |
5022 | ||
5023 | /* Take into account preallocation. Nothing special is needed for | |
5024 | * PREALLOC_MODE_METADATA since metadata is always counted. | |
5025 | */ | |
5026 | if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) { | |
5027 | required = virtual_size; | |
5028 | } | |
5029 | ||
5d72c68b | 5030 | info = g_new0(BlockMeasureInfo, 1); |
0dd07b29 | 5031 | info->fully_allocated = luks_payload_size + |
c501c352 | 5032 | qcow2_calc_prealloc_size(virtual_size, cluster_size, |
0dd07b29 | 5033 | ctz32(refcount_bits), extended_l2); |
c501c352 | 5034 | |
5d72c68b EB |
5035 | /* |
5036 | * Remove data clusters that are not required. This overestimates the | |
c501c352 | 5037 | * required size because metadata needed for the fully allocated file is |
5d72c68b EB |
5038 | * still counted. Show bitmaps only if both source and destination |
5039 | * would support them. | |
c501c352 SH |
5040 | */ |
5041 | info->required = info->fully_allocated - virtual_size + required; | |
5d72c68b EB |
5042 | info->has_bitmaps = version >= 3 && in_bs && |
5043 | bdrv_supports_persistent_dirty_bitmap(in_bs); | |
5044 | if (info->has_bitmaps) { | |
5045 | info->bitmaps = qcow2_get_persistent_dirty_bitmap_size(in_bs, | |
5046 | cluster_size); | |
5047 | } | |
c501c352 SH |
5048 | return info; |
5049 | ||
5050 | err: | |
5051 | error_propagate(errp, local_err); | |
5052 | return NULL; | |
5053 | } | |
5054 | ||
7c80ab3f | 5055 | static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
20d97356 | 5056 | { |
ff99129a | 5057 | BDRVQcow2State *s = bs->opaque; |
20d97356 | 5058 | bdi->cluster_size = s->cluster_size; |
7c80ab3f | 5059 | bdi->vm_state_offset = qcow2_vm_state_offset(s); |
20d97356 BS |
5060 | return 0; |
5061 | } | |
5062 | ||
1bf6e9ca AS |
5063 | static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs, |
5064 | Error **errp) | |
37764dfb | 5065 | { |
ff99129a | 5066 | BDRVQcow2State *s = bs->opaque; |
0a12f6f8 DB |
5067 | ImageInfoSpecific *spec_info; |
5068 | QCryptoBlockInfo *encrypt_info = NULL; | |
37764dfb | 5069 | |
0a12f6f8 | 5070 | if (s->crypto != NULL) { |
83bad8cb VSO |
5071 | encrypt_info = qcrypto_block_get_info(s->crypto, errp); |
5072 | if (!encrypt_info) { | |
1bf6e9ca AS |
5073 | return NULL; |
5074 | } | |
0a12f6f8 DB |
5075 | } |
5076 | ||
5077 | spec_info = g_new(ImageInfoSpecific, 1); | |
37764dfb | 5078 | *spec_info = (ImageInfoSpecific){ |
6a8f9661 | 5079 | .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2, |
b8968c87 | 5080 | .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1), |
37764dfb HR |
5081 | }; |
5082 | if (s->qcow_version == 2) { | |
32bafa8f | 5083 | *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ |
0709c5a1 HR |
5084 | .compat = g_strdup("0.10"), |
5085 | .refcount_bits = s->refcount_bits, | |
37764dfb HR |
5086 | }; |
5087 | } else if (s->qcow_version == 3) { | |
b8968c87 | 5088 | Qcow2BitmapInfoList *bitmaps; |
83bad8cb | 5089 | if (!qcow2_get_bitmap_info_list(bs, &bitmaps, errp)) { |
b8968c87 | 5090 | qapi_free_ImageInfoSpecific(spec_info); |
71eaec2e | 5091 | qapi_free_QCryptoBlockInfo(encrypt_info); |
b8968c87 AS |
5092 | return NULL; |
5093 | } | |
32bafa8f | 5094 | *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){ |
37764dfb HR |
5095 | .compat = g_strdup("1.1"), |
5096 | .lazy_refcounts = s->compatible_features & | |
5097 | QCOW2_COMPAT_LAZY_REFCOUNTS, | |
5098 | .has_lazy_refcounts = true, | |
9009b196 HR |
5099 | .corrupt = s->incompatible_features & |
5100 | QCOW2_INCOMPAT_CORRUPT, | |
5101 | .has_corrupt = true, | |
7be20252 AG |
5102 | .has_extended_l2 = true, |
5103 | .extended_l2 = has_subclusters(s), | |
0709c5a1 | 5104 | .refcount_bits = s->refcount_bits, |
b8968c87 AS |
5105 | .has_bitmaps = !!bitmaps, |
5106 | .bitmaps = bitmaps, | |
9b890bdc KW |
5107 | .has_data_file = !!s->image_data_file, |
5108 | .data_file = g_strdup(s->image_data_file), | |
6c3944dc KW |
5109 | .has_data_file_raw = has_data_file(bs), |
5110 | .data_file_raw = data_file_is_raw(bs), | |
572ad978 | 5111 | .compression_type = s->compression_type, |
37764dfb | 5112 | }; |
b1fc8f93 DL |
5113 | } else { |
5114 | /* if this assertion fails, this probably means a new version was | |
5115 | * added without having it covered here */ | |
5116 | assert(false); | |
37764dfb HR |
5117 | } |
5118 | ||
0a12f6f8 DB |
5119 | if (encrypt_info) { |
5120 | ImageInfoSpecificQCow2Encryption *qencrypt = | |
5121 | g_new(ImageInfoSpecificQCow2Encryption, 1); | |
5122 | switch (encrypt_info->format) { | |
5123 | case Q_CRYPTO_BLOCK_FORMAT_QCOW: | |
5124 | qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES; | |
0a12f6f8 DB |
5125 | break; |
5126 | case Q_CRYPTO_BLOCK_FORMAT_LUKS: | |
5127 | qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS; | |
5128 | qencrypt->u.luks = encrypt_info->u.luks; | |
5129 | break; | |
5130 | default: | |
5131 | abort(); | |
5132 | } | |
5133 | /* Since we did shallow copy above, erase any pointers | |
5134 | * in the original info */ | |
5135 | memset(&encrypt_info->u, 0, sizeof(encrypt_info->u)); | |
5136 | qapi_free_QCryptoBlockInfo(encrypt_info); | |
5137 | ||
5138 | spec_info->u.qcow2.data->has_encrypt = true; | |
5139 | spec_info->u.qcow2.data->encrypt = qencrypt; | |
5140 | } | |
5141 | ||
37764dfb HR |
5142 | return spec_info; |
5143 | } | |
5144 | ||
38841dcd HR |
5145 | static int qcow2_has_zero_init(BlockDriverState *bs) |
5146 | { | |
5147 | BDRVQcow2State *s = bs->opaque; | |
5148 | bool preallocated; | |
5149 | ||
5150 | if (qemu_in_coroutine()) { | |
5151 | qemu_co_mutex_lock(&s->lock); | |
5152 | } | |
5153 | /* | |
5154 | * Check preallocation status: Preallocated images have all L2 | |
5155 | * tables allocated, nonpreallocated images have none. It is | |
5156 | * therefore enough to check the first one. | |
5157 | */ | |
5158 | preallocated = s->l1_size > 0 && s->l1_table[0] != 0; | |
5159 | if (qemu_in_coroutine()) { | |
5160 | qemu_co_mutex_unlock(&s->lock); | |
5161 | } | |
5162 | ||
5163 | if (!preallocated) { | |
5164 | return 1; | |
5165 | } else if (bs->encrypted) { | |
5166 | return 0; | |
5167 | } else { | |
5168 | return bdrv_has_zero_init(s->data_file->bs); | |
5169 | } | |
5170 | } | |
5171 | ||
cf8074b3 KW |
5172 | static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, |
5173 | int64_t pos) | |
20d97356 | 5174 | { |
ff99129a | 5175 | BDRVQcow2State *s = bs->opaque; |
20d97356 | 5176 | |
66f82cee | 5177 | BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); |
5396234b VSO |
5178 | return bs->drv->bdrv_co_pwritev_part(bs, qcow2_vm_state_offset(s) + pos, |
5179 | qiov->size, qiov, 0, 0); | |
20d97356 BS |
5180 | } |
5181 | ||
5ddda0b8 KW |
5182 | static int qcow2_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, |
5183 | int64_t pos) | |
20d97356 | 5184 | { |
ff99129a | 5185 | BDRVQcow2State *s = bs->opaque; |
20d97356 | 5186 | |
66f82cee | 5187 | BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); |
df893d25 VSO |
5188 | return bs->drv->bdrv_co_preadv_part(bs, qcow2_vm_state_offset(s) + pos, |
5189 | qiov->size, qiov, 0, 0); | |
20d97356 BS |
5190 | } |
5191 | ||
9296b3ed HR |
5192 | /* |
5193 | * Downgrades an image's version. To achieve this, any incompatible features | |
5194 | * have to be removed. | |
5195 | */ | |
4057a2b2 | 5196 | static int qcow2_downgrade(BlockDriverState *bs, int target_version, |
d1402b50 HR |
5197 | BlockDriverAmendStatusCB *status_cb, void *cb_opaque, |
5198 | Error **errp) | |
9296b3ed | 5199 | { |
ff99129a | 5200 | BDRVQcow2State *s = bs->opaque; |
9296b3ed HR |
5201 | int current_version = s->qcow_version; |
5202 | int ret; | |
7fa140ab | 5203 | int i; |
9296b3ed | 5204 | |
d1402b50 HR |
5205 | /* This is qcow2_downgrade(), not qcow2_upgrade() */ |
5206 | assert(target_version < current_version); | |
5207 | ||
5208 | /* There are no other versions (now) that you can downgrade to */ | |
5209 | assert(target_version == 2); | |
9296b3ed HR |
5210 | |
5211 | if (s->refcount_order != 4) { | |
d1402b50 | 5212 | error_setg(errp, "compat=0.10 requires refcount_bits=16"); |
9296b3ed HR |
5213 | return -ENOTSUP; |
5214 | } | |
5215 | ||
966b000f KW |
5216 | if (has_data_file(bs)) { |
5217 | error_setg(errp, "Cannot downgrade an image with a data file"); | |
5218 | return -ENOTSUP; | |
5219 | } | |
5220 | ||
7fa140ab EB |
5221 | /* |
5222 | * If any internal snapshot has a different size than the current | |
5223 | * image size, or VM state size that exceeds 32 bits, downgrading | |
5224 | * is unsafe. Even though we would still use v3-compliant output | |
5225 | * to preserve that data, other v2 programs might not realize | |
5226 | * those optional fields are important. | |
5227 | */ | |
5228 | for (i = 0; i < s->nb_snapshots; i++) { | |
5229 | if (s->snapshots[i].vm_state_size > UINT32_MAX || | |
5230 | s->snapshots[i].disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) { | |
5231 | error_setg(errp, "Internal snapshots prevent downgrade of image"); | |
5232 | return -ENOTSUP; | |
5233 | } | |
5234 | } | |
5235 | ||
9296b3ed HR |
5236 | /* clear incompatible features */ |
5237 | if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) { | |
5238 | ret = qcow2_mark_clean(bs); | |
5239 | if (ret < 0) { | |
d1402b50 | 5240 | error_setg_errno(errp, -ret, "Failed to make the image clean"); |
9296b3ed HR |
5241 | return ret; |
5242 | } | |
5243 | } | |
5244 | ||
5245 | /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in | |
5246 | * the first place; if that happens nonetheless, returning -ENOTSUP is the | |
5247 | * best thing to do anyway */ | |
5248 | ||
5249 | if (s->incompatible_features) { | |
d1402b50 HR |
5250 | error_setg(errp, "Cannot downgrade an image with incompatible features " |
5251 | "%#" PRIx64 " set", s->incompatible_features); | |
9296b3ed HR |
5252 | return -ENOTSUP; |
5253 | } | |
5254 | ||
5255 | /* since we can ignore compatible features, we can set them to 0 as well */ | |
5256 | s->compatible_features = 0; | |
5257 | /* if lazy refcounts have been used, they have already been fixed through | |
5258 | * clearing the dirty flag */ | |
5259 | ||
5260 | /* clearing autoclear features is trivial */ | |
5261 | s->autoclear_features = 0; | |
5262 | ||
8b13976d | 5263 | ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque); |
9296b3ed | 5264 | if (ret < 0) { |
d1402b50 | 5265 | error_setg_errno(errp, -ret, "Failed to turn zero into data clusters"); |
9296b3ed HR |
5266 | return ret; |
5267 | } | |
5268 | ||
5269 | s->qcow_version = target_version; | |
5270 | ret = qcow2_update_header(bs); | |
5271 | if (ret < 0) { | |
5272 | s->qcow_version = current_version; | |
d1402b50 | 5273 | error_setg_errno(errp, -ret, "Failed to update the image header"); |
9296b3ed HR |
5274 | return ret; |
5275 | } | |
5276 | return 0; | |
5277 | } | |
5278 | ||
722efb0c HR |
5279 | /* |
5280 | * Upgrades an image's version. While newer versions encompass all | |
5281 | * features of older versions, some things may have to be presented | |
5282 | * differently. | |
5283 | */ | |
5284 | static int qcow2_upgrade(BlockDriverState *bs, int target_version, | |
5285 | BlockDriverAmendStatusCB *status_cb, void *cb_opaque, | |
5286 | Error **errp) | |
5287 | { | |
5288 | BDRVQcow2State *s = bs->opaque; | |
0a85af35 | 5289 | bool need_snapshot_update; |
722efb0c | 5290 | int current_version = s->qcow_version; |
0a85af35 | 5291 | int i; |
722efb0c HR |
5292 | int ret; |
5293 | ||
5294 | /* This is qcow2_upgrade(), not qcow2_downgrade() */ | |
5295 | assert(target_version > current_version); | |
5296 | ||
5297 | /* There are no other versions (yet) that you can upgrade to */ | |
5298 | assert(target_version == 3); | |
5299 | ||
0a85af35 HR |
5300 | status_cb(bs, 0, 2, cb_opaque); |
5301 | ||
5302 | /* | |
5303 | * In v2, snapshots do not need to have extra data. v3 requires | |
5304 | * the 64-bit VM state size and the virtual disk size to be | |
5305 | * present. | |
5306 | * qcow2_write_snapshots() will always write the list in the | |
5307 | * v3-compliant format. | |
5308 | */ | |
5309 | need_snapshot_update = false; | |
5310 | for (i = 0; i < s->nb_snapshots; i++) { | |
5311 | if (s->snapshots[i].extra_data_size < | |
5312 | sizeof_field(QCowSnapshotExtraData, vm_state_size_large) + | |
5313 | sizeof_field(QCowSnapshotExtraData, disk_size)) | |
5314 | { | |
5315 | need_snapshot_update = true; | |
5316 | break; | |
5317 | } | |
5318 | } | |
5319 | if (need_snapshot_update) { | |
5320 | ret = qcow2_write_snapshots(bs); | |
5321 | if (ret < 0) { | |
5322 | error_setg_errno(errp, -ret, "Failed to update the snapshot table"); | |
5323 | return ret; | |
5324 | } | |
5325 | } | |
5326 | status_cb(bs, 1, 2, cb_opaque); | |
722efb0c HR |
5327 | |
5328 | s->qcow_version = target_version; | |
5329 | ret = qcow2_update_header(bs); | |
5330 | if (ret < 0) { | |
5331 | s->qcow_version = current_version; | |
5332 | error_setg_errno(errp, -ret, "Failed to update the image header"); | |
5333 | return ret; | |
5334 | } | |
0a85af35 | 5335 | status_cb(bs, 2, 2, cb_opaque); |
722efb0c HR |
5336 | |
5337 | return 0; | |
5338 | } | |
5339 | ||
c293a809 HR |
5340 | typedef enum Qcow2AmendOperation { |
5341 | /* This is the value Qcow2AmendHelperCBInfo::last_operation will be | |
5342 | * statically initialized to so that the helper CB can discern the first | |
5343 | * invocation from an operation change */ | |
5344 | QCOW2_NO_OPERATION = 0, | |
5345 | ||
722efb0c | 5346 | QCOW2_UPGRADING, |
90766d9d | 5347 | QCOW2_UPDATING_ENCRYPTION, |
61ce55fc | 5348 | QCOW2_CHANGING_REFCOUNT_ORDER, |
c293a809 HR |
5349 | QCOW2_DOWNGRADING, |
5350 | } Qcow2AmendOperation; | |
5351 | ||
5352 | typedef struct Qcow2AmendHelperCBInfo { | |
5353 | /* The code coordinating the amend operations should only modify | |
5354 | * these four fields; the rest will be managed by the CB */ | |
5355 | BlockDriverAmendStatusCB *original_status_cb; | |
5356 | void *original_cb_opaque; | |
5357 | ||
5358 | Qcow2AmendOperation current_operation; | |
5359 | ||
5360 | /* Total number of operations to perform (only set once) */ | |
5361 | int total_operations; | |
5362 | ||
5363 | /* The following fields are managed by the CB */ | |
5364 | ||
5365 | /* Number of operations completed */ | |
5366 | int operations_completed; | |
5367 | ||
5368 | /* Cumulative offset of all completed operations */ | |
5369 | int64_t offset_completed; | |
5370 | ||
5371 | Qcow2AmendOperation last_operation; | |
5372 | int64_t last_work_size; | |
5373 | } Qcow2AmendHelperCBInfo; | |
5374 | ||
5375 | static void qcow2_amend_helper_cb(BlockDriverState *bs, | |
5376 | int64_t operation_offset, | |
5377 | int64_t operation_work_size, void *opaque) | |
5378 | { | |
5379 | Qcow2AmendHelperCBInfo *info = opaque; | |
5380 | int64_t current_work_size; | |
5381 | int64_t projected_work_size; | |
5382 | ||
5383 | if (info->current_operation != info->last_operation) { | |
5384 | if (info->last_operation != QCOW2_NO_OPERATION) { | |
5385 | info->offset_completed += info->last_work_size; | |
5386 | info->operations_completed++; | |
5387 | } | |
5388 | ||
5389 | info->last_operation = info->current_operation; | |
5390 | } | |
5391 | ||
5392 | assert(info->total_operations > 0); | |
5393 | assert(info->operations_completed < info->total_operations); | |
5394 | ||
5395 | info->last_work_size = operation_work_size; | |
5396 | ||
5397 | current_work_size = info->offset_completed + operation_work_size; | |
5398 | ||
5399 | /* current_work_size is the total work size for (operations_completed + 1) | |
5400 | * operations (which includes this one), so multiply it by the number of | |
5401 | * operations not covered and divide it by the number of operations | |
5402 | * covered to get a projection for the operations not covered */ | |
5403 | projected_work_size = current_work_size * (info->total_operations - | |
5404 | info->operations_completed - 1) | |
5405 | / (info->operations_completed + 1); | |
5406 | ||
5407 | info->original_status_cb(bs, info->offset_completed + operation_offset, | |
5408 | current_work_size + projected_work_size, | |
5409 | info->original_cb_opaque); | |
5410 | } | |
5411 | ||
77485434 | 5412 | static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts, |
8b13976d | 5413 | BlockDriverAmendStatusCB *status_cb, |
d1402b50 | 5414 | void *cb_opaque, |
a3579bfa | 5415 | bool force, |
d1402b50 | 5416 | Error **errp) |
9296b3ed | 5417 | { |
ff99129a | 5418 | BDRVQcow2State *s = bs->opaque; |
9296b3ed HR |
5419 | int old_version = s->qcow_version, new_version = old_version; |
5420 | uint64_t new_size = 0; | |
9b890bdc | 5421 | const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL; |
9296b3ed | 5422 | bool lazy_refcounts = s->use_lazy_refcounts; |
6c3944dc | 5423 | bool data_file_raw = data_file_is_raw(bs); |
1bd0e2d1 | 5424 | const char *compat = NULL; |
61ce55fc | 5425 | int refcount_bits = s->refcount_bits; |
9296b3ed | 5426 | int ret; |
1bd0e2d1 | 5427 | QemuOptDesc *desc = opts->list->desc; |
c293a809 | 5428 | Qcow2AmendHelperCBInfo helper_cb_info; |
90766d9d | 5429 | bool encryption_update = false; |
9296b3ed | 5430 | |
1bd0e2d1 CL |
5431 | while (desc && desc->name) { |
5432 | if (!qemu_opt_find(opts, desc->name)) { | |
9296b3ed | 5433 | /* only change explicitly defined options */ |
1bd0e2d1 | 5434 | desc++; |
9296b3ed HR |
5435 | continue; |
5436 | } | |
5437 | ||
8a17b83c HR |
5438 | if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) { |
5439 | compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL); | |
1bd0e2d1 | 5440 | if (!compat) { |
9296b3ed | 5441 | /* preserve default */ |
f7077c98 | 5442 | } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) { |
9296b3ed | 5443 | new_version = 2; |
f7077c98 | 5444 | } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) { |
9296b3ed HR |
5445 | new_version = 3; |
5446 | } else { | |
d1402b50 | 5447 | error_setg(errp, "Unknown compatibility level %s", compat); |
9296b3ed HR |
5448 | return -EINVAL; |
5449 | } | |
8a17b83c HR |
5450 | } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) { |
5451 | new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); | |
5452 | } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) { | |
5453 | backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); | |
5454 | } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) { | |
5455 | backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); | |
90766d9d ML |
5456 | } else if (g_str_has_prefix(desc->name, "encrypt.")) { |
5457 | if (!s->crypto) { | |
5458 | error_setg(errp, | |
5459 | "Can't amend encryption options - encryption not present"); | |
5460 | return -EINVAL; | |
5461 | } | |
5462 | if (s->crypt_method_header != QCOW_CRYPT_LUKS) { | |
5463 | error_setg(errp, | |
5464 | "Only LUKS encryption options can be amended"); | |
5465 | return -ENOTSUP; | |
5466 | } | |
5467 | encryption_update = true; | |
8a17b83c HR |
5468 | } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) { |
5469 | lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS, | |
1bd0e2d1 | 5470 | lazy_refcounts); |
06d05fa7 | 5471 | } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) { |
61ce55fc HR |
5472 | refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS, |
5473 | refcount_bits); | |
5474 | ||
5475 | if (refcount_bits <= 0 || refcount_bits > 64 || | |
5476 | !is_power_of_2(refcount_bits)) | |
5477 | { | |
d1402b50 HR |
5478 | error_setg(errp, "Refcount width must be a power of two and " |
5479 | "may not exceed 64 bits"); | |
61ce55fc HR |
5480 | return -EINVAL; |
5481 | } | |
9b890bdc KW |
5482 | } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) { |
5483 | data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE); | |
5484 | if (data_file && !has_data_file(bs)) { | |
5485 | error_setg(errp, "data-file can only be set for images that " | |
5486 | "use an external data file"); | |
5487 | return -EINVAL; | |
5488 | } | |
6c3944dc KW |
5489 | } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) { |
5490 | data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW, | |
5491 | data_file_raw); | |
5492 | if (data_file_raw && !data_file_is_raw(bs)) { | |
5493 | error_setg(errp, "data-file-raw cannot be set on existing " | |
5494 | "images"); | |
5495 | return -EINVAL; | |
5496 | } | |
9296b3ed | 5497 | } else { |
164e0f89 | 5498 | /* if this point is reached, this probably means a new option was |
9296b3ed | 5499 | * added without having it covered here */ |
164e0f89 | 5500 | abort(); |
9296b3ed | 5501 | } |
1bd0e2d1 CL |
5502 | |
5503 | desc++; | |
9296b3ed HR |
5504 | } |
5505 | ||
c293a809 HR |
5506 | helper_cb_info = (Qcow2AmendHelperCBInfo){ |
5507 | .original_status_cb = status_cb, | |
5508 | .original_cb_opaque = cb_opaque, | |
722efb0c | 5509 | .total_operations = (new_version != old_version) |
90766d9d ML |
5510 | + (s->refcount_bits != refcount_bits) + |
5511 | (encryption_update == true) | |
c293a809 HR |
5512 | }; |
5513 | ||
1038bbb8 HR |
5514 | /* Upgrade first (some features may require compat=1.1) */ |
5515 | if (new_version > old_version) { | |
722efb0c HR |
5516 | helper_cb_info.current_operation = QCOW2_UPGRADING; |
5517 | ret = qcow2_upgrade(bs, new_version, &qcow2_amend_helper_cb, | |
5518 | &helper_cb_info, errp); | |
1038bbb8 | 5519 | if (ret < 0) { |
1038bbb8 | 5520 | return ret; |
9296b3ed HR |
5521 | } |
5522 | } | |
5523 | ||
90766d9d ML |
5524 | if (encryption_update) { |
5525 | QDict *amend_opts_dict; | |
5526 | QCryptoBlockAmendOptions *amend_opts; | |
5527 | ||
5528 | helper_cb_info.current_operation = QCOW2_UPDATING_ENCRYPTION; | |
5529 | amend_opts_dict = qcow2_extract_crypto_opts(opts, "luks", errp); | |
5530 | if (!amend_opts_dict) { | |
5531 | return -EINVAL; | |
5532 | } | |
5533 | amend_opts = block_crypto_amend_opts_init(amend_opts_dict, errp); | |
5534 | qobject_unref(amend_opts_dict); | |
5535 | if (!amend_opts) { | |
5536 | return -EINVAL; | |
5537 | } | |
5538 | ret = qcrypto_block_amend_options(s->crypto, | |
5539 | qcow2_crypto_hdr_read_func, | |
5540 | qcow2_crypto_hdr_write_func, | |
5541 | bs, | |
5542 | amend_opts, | |
5543 | force, | |
5544 | errp); | |
5545 | qapi_free_QCryptoBlockAmendOptions(amend_opts); | |
5546 | if (ret < 0) { | |
5547 | return ret; | |
5548 | } | |
5549 | } | |
5550 | ||
61ce55fc HR |
5551 | if (s->refcount_bits != refcount_bits) { |
5552 | int refcount_order = ctz32(refcount_bits); | |
61ce55fc HR |
5553 | |
5554 | if (new_version < 3 && refcount_bits != 16) { | |
d1402b50 HR |
5555 | error_setg(errp, "Refcount widths other than 16 bits require " |
5556 | "compatibility level 1.1 or above (use compat=1.1 or " | |
5557 | "greater)"); | |
61ce55fc HR |
5558 | return -EINVAL; |
5559 | } | |
5560 | ||
5561 | helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER; | |
5562 | ret = qcow2_change_refcount_order(bs, refcount_order, | |
5563 | &qcow2_amend_helper_cb, | |
d1402b50 | 5564 | &helper_cb_info, errp); |
61ce55fc | 5565 | if (ret < 0) { |
61ce55fc HR |
5566 | return ret; |
5567 | } | |
5568 | } | |
5569 | ||
6c3944dc KW |
5570 | /* data-file-raw blocks backing files, so clear it first if requested */ |
5571 | if (data_file_raw) { | |
5572 | s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW; | |
5573 | } else { | |
5574 | s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW; | |
5575 | } | |
5576 | ||
9b890bdc KW |
5577 | if (data_file) { |
5578 | g_free(s->image_data_file); | |
5579 | s->image_data_file = *data_file ? g_strdup(data_file) : NULL; | |
5580 | } | |
5581 | ||
5582 | ret = qcow2_update_header(bs); | |
5583 | if (ret < 0) { | |
5584 | error_setg_errno(errp, -ret, "Failed to update the image header"); | |
5585 | return ret; | |
5586 | } | |
5587 | ||
9296b3ed | 5588 | if (backing_file || backing_format) { |
bc5ee6da EB |
5589 | if (g_strcmp0(backing_file, s->image_backing_file) || |
5590 | g_strcmp0(backing_format, s->image_backing_format)) { | |
5591 | warn_report("Deprecated use of amend to alter the backing file; " | |
5592 | "use qemu-img rebase instead"); | |
5593 | } | |
e4603fe1 KW |
5594 | ret = qcow2_change_backing_file(bs, |
5595 | backing_file ?: s->image_backing_file, | |
5596 | backing_format ?: s->image_backing_format); | |
9296b3ed | 5597 | if (ret < 0) { |
d1402b50 | 5598 | error_setg_errno(errp, -ret, "Failed to change the backing file"); |
9296b3ed HR |
5599 | return ret; |
5600 | } | |
5601 | } | |
5602 | ||
5603 | if (s->use_lazy_refcounts != lazy_refcounts) { | |
5604 | if (lazy_refcounts) { | |
1038bbb8 | 5605 | if (new_version < 3) { |
d1402b50 HR |
5606 | error_setg(errp, "Lazy refcounts only supported with " |
5607 | "compatibility level 1.1 and above (use compat=1.1 " | |
5608 | "or greater)"); | |
9296b3ed HR |
5609 | return -EINVAL; |
5610 | } | |
5611 | s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; | |
5612 | ret = qcow2_update_header(bs); | |
5613 | if (ret < 0) { | |
5614 | s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; | |
d1402b50 | 5615 | error_setg_errno(errp, -ret, "Failed to update the image header"); |
9296b3ed HR |
5616 | return ret; |
5617 | } | |
5618 | s->use_lazy_refcounts = true; | |
5619 | } else { | |
5620 | /* make image clean first */ | |
5621 | ret = qcow2_mark_clean(bs); | |
5622 | if (ret < 0) { | |
d1402b50 | 5623 | error_setg_errno(errp, -ret, "Failed to make the image clean"); |
9296b3ed HR |
5624 | return ret; |
5625 | } | |
5626 | /* now disallow lazy refcounts */ | |
5627 | s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; | |
5628 | ret = qcow2_update_header(bs); | |
5629 | if (ret < 0) { | |
5630 | s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; | |
d1402b50 | 5631 | error_setg_errno(errp, -ret, "Failed to update the image header"); |
9296b3ed HR |
5632 | return ret; |
5633 | } | |
5634 | s->use_lazy_refcounts = false; | |
5635 | } | |
5636 | } | |
5637 | ||
5638 | if (new_size) { | |
a3aeeab5 EB |
5639 | BlockBackend *blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, |
5640 | errp); | |
5641 | if (!blk) { | |
5642 | return -EPERM; | |
d7086422 KW |
5643 | } |
5644 | ||
e8d04f92 HR |
5645 | /* |
5646 | * Amending image options should ensure that the image has | |
5647 | * exactly the given new values, so pass exact=true here. | |
5648 | */ | |
8c6242b6 | 5649 | ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, 0, errp); |
70b27f36 | 5650 | blk_unref(blk); |
9296b3ed HR |
5651 | if (ret < 0) { |
5652 | return ret; | |
5653 | } | |
5654 | } | |
5655 | ||
1038bbb8 HR |
5656 | /* Downgrade last (so unsupported features can be removed before) */ |
5657 | if (new_version < old_version) { | |
c293a809 HR |
5658 | helper_cb_info.current_operation = QCOW2_DOWNGRADING; |
5659 | ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb, | |
d1402b50 | 5660 | &helper_cb_info, errp); |
1038bbb8 HR |
5661 | if (ret < 0) { |
5662 | return ret; | |
5663 | } | |
5664 | } | |
5665 | ||
9296b3ed HR |
5666 | return 0; |
5667 | } | |
5668 | ||
8ea1613d ML |
5669 | static int coroutine_fn qcow2_co_amend(BlockDriverState *bs, |
5670 | BlockdevAmendOptions *opts, | |
5671 | bool force, | |
5672 | Error **errp) | |
5673 | { | |
5674 | BlockdevAmendOptionsQcow2 *qopts = &opts->u.qcow2; | |
5675 | BDRVQcow2State *s = bs->opaque; | |
5676 | int ret = 0; | |
5677 | ||
5678 | if (qopts->has_encrypt) { | |
5679 | if (!s->crypto) { | |
5680 | error_setg(errp, "image is not encrypted, can't amend"); | |
5681 | return -EOPNOTSUPP; | |
5682 | } | |
5683 | ||
5684 | if (qopts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_LUKS) { | |
5685 | error_setg(errp, | |
5686 | "Amend can't be used to change the qcow2 encryption format"); | |
5687 | return -EOPNOTSUPP; | |
5688 | } | |
5689 | ||
5690 | if (s->crypt_method_header != QCOW_CRYPT_LUKS) { | |
5691 | error_setg(errp, | |
5692 | "Only LUKS encryption options can be amended for qcow2 with blockdev-amend"); | |
5693 | return -EOPNOTSUPP; | |
5694 | } | |
5695 | ||
5696 | ret = qcrypto_block_amend_options(s->crypto, | |
5697 | qcow2_crypto_hdr_read_func, | |
5698 | qcow2_crypto_hdr_write_func, | |
5699 | bs, | |
5700 | qopts->encrypt, | |
5701 | force, | |
5702 | errp); | |
5703 | } | |
5704 | return ret; | |
5705 | } | |
5706 | ||
85186ebd HR |
5707 | /* |
5708 | * If offset or size are negative, respectively, they will not be included in | |
5709 | * the BLOCK_IMAGE_CORRUPTED event emitted. | |
5710 | * fatal will be ignored for read-only BDS; corruptions found there will always | |
5711 | * be considered non-fatal. | |
5712 | */ | |
5713 | void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset, | |
5714 | int64_t size, const char *message_format, ...) | |
5715 | { | |
ff99129a | 5716 | BDRVQcow2State *s = bs->opaque; |
dc881b44 | 5717 | const char *node_name; |
85186ebd HR |
5718 | char *message; |
5719 | va_list ap; | |
5720 | ||
ddf3b47e | 5721 | fatal = fatal && bdrv_is_writable(bs); |
85186ebd HR |
5722 | |
5723 | if (s->signaled_corruption && | |
5724 | (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT))) | |
5725 | { | |
5726 | return; | |
5727 | } | |
5728 | ||
5729 | va_start(ap, message_format); | |
5730 | message = g_strdup_vprintf(message_format, ap); | |
5731 | va_end(ap); | |
5732 | ||
5733 | if (fatal) { | |
5734 | fprintf(stderr, "qcow2: Marking image as corrupt: %s; further " | |
5735 | "corruption events will be suppressed\n", message); | |
5736 | } else { | |
5737 | fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal " | |
5738 | "corruption events will be suppressed\n", message); | |
5739 | } | |
5740 | ||
dc881b44 AG |
5741 | node_name = bdrv_get_node_name(bs); |
5742 | qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs), | |
5743 | *node_name != '\0', node_name, | |
5744 | message, offset >= 0, offset, | |
5745 | size >= 0, size, | |
3ab72385 | 5746 | fatal); |
85186ebd HR |
5747 | g_free(message); |
5748 | ||
5749 | if (fatal) { | |
5750 | qcow2_mark_corrupt(bs); | |
5751 | bs->drv = NULL; /* make BDS unusable */ | |
5752 | } | |
5753 | ||
5754 | s->signaled_corruption = true; | |
5755 | } | |
5756 | ||
df373fb0 ML |
5757 | #define QCOW_COMMON_OPTIONS \ |
5758 | { \ | |
5759 | .name = BLOCK_OPT_SIZE, \ | |
5760 | .type = QEMU_OPT_SIZE, \ | |
5761 | .help = "Virtual disk size" \ | |
5762 | }, \ | |
5763 | { \ | |
5764 | .name = BLOCK_OPT_COMPAT_LEVEL, \ | |
5765 | .type = QEMU_OPT_STRING, \ | |
5766 | .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \ | |
5767 | }, \ | |
5768 | { \ | |
5769 | .name = BLOCK_OPT_BACKING_FILE, \ | |
5770 | .type = QEMU_OPT_STRING, \ | |
5771 | .help = "File name of a base image" \ | |
5772 | }, \ | |
5773 | { \ | |
5774 | .name = BLOCK_OPT_BACKING_FMT, \ | |
5775 | .type = QEMU_OPT_STRING, \ | |
5776 | .help = "Image format of the base image" \ | |
5777 | }, \ | |
5778 | { \ | |
5779 | .name = BLOCK_OPT_DATA_FILE, \ | |
5780 | .type = QEMU_OPT_STRING, \ | |
5781 | .help = "File name of an external data file" \ | |
5782 | }, \ | |
5783 | { \ | |
5784 | .name = BLOCK_OPT_DATA_FILE_RAW, \ | |
5785 | .type = QEMU_OPT_BOOL, \ | |
5786 | .help = "The external data file must stay valid " \ | |
5787 | "as a raw image" \ | |
5788 | }, \ | |
df373fb0 ML |
5789 | { \ |
5790 | .name = BLOCK_OPT_LAZY_REFCOUNTS, \ | |
5791 | .type = QEMU_OPT_BOOL, \ | |
5792 | .help = "Postpone refcount updates", \ | |
5793 | .def_value_str = "off" \ | |
5794 | }, \ | |
5795 | { \ | |
5796 | .name = BLOCK_OPT_REFCOUNT_BITS, \ | |
5797 | .type = QEMU_OPT_NUMBER, \ | |
5798 | .help = "Width of a reference count entry in bits", \ | |
5799 | .def_value_str = "16" \ | |
df373fb0 ML |
5800 | } |
5801 | ||
1bd0e2d1 CL |
5802 | static QemuOptsList qcow2_create_opts = { |
5803 | .name = "qcow2-create-opts", | |
5804 | .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head), | |
5805 | .desc = { | |
0b6786a9 ML |
5806 | { \ |
5807 | .name = BLOCK_OPT_ENCRYPT, \ | |
5808 | .type = QEMU_OPT_BOOL, \ | |
5809 | .help = "Encrypt the image with format 'aes'. (Deprecated " \ | |
5810 | "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", \ | |
5811 | }, \ | |
5812 | { \ | |
5813 | .name = BLOCK_OPT_ENCRYPT_FORMAT, \ | |
5814 | .type = QEMU_OPT_STRING, \ | |
5815 | .help = "Encrypt the image, format choices: 'aes', 'luks'", \ | |
5816 | }, \ | |
5817 | BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \ | |
5818 | "ID of secret providing qcow AES key or LUKS passphrase"), \ | |
5819 | BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \ | |
5820 | BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \ | |
5821 | BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \ | |
5822 | BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \ | |
5823 | BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \ | |
5824 | BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \ | |
5825 | { \ | |
5826 | .name = BLOCK_OPT_CLUSTER_SIZE, \ | |
5827 | .type = QEMU_OPT_SIZE, \ | |
5828 | .help = "qcow2 cluster size", \ | |
5829 | .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) \ | |
5830 | }, \ | |
7be20252 AG |
5831 | { \ |
5832 | .name = BLOCK_OPT_EXTL2, \ | |
5833 | .type = QEMU_OPT_BOOL, \ | |
5834 | .help = "Extended L2 tables", \ | |
5835 | .def_value_str = "off" \ | |
5836 | }, \ | |
0b6786a9 ML |
5837 | { \ |
5838 | .name = BLOCK_OPT_PREALLOC, \ | |
5839 | .type = QEMU_OPT_STRING, \ | |
5840 | .help = "Preallocation mode (allowed values: off, " \ | |
5841 | "metadata, falloc, full)" \ | |
5842 | }, \ | |
5843 | { \ | |
5844 | .name = BLOCK_OPT_COMPRESSION_TYPE, \ | |
5845 | .type = QEMU_OPT_STRING, \ | |
5846 | .help = "Compression method used for image cluster " \ | |
5847 | "compression", \ | |
5848 | .def_value_str = "zlib" \ | |
5849 | }, | |
df373fb0 ML |
5850 | QCOW_COMMON_OPTIONS, |
5851 | { /* end of list */ } | |
5852 | } | |
5853 | }; | |
5854 | ||
5855 | static QemuOptsList qcow2_amend_opts = { | |
5856 | .name = "qcow2-amend-opts", | |
5857 | .head = QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts.head), | |
5858 | .desc = { | |
90766d9d ML |
5859 | BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."), |
5860 | BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."), | |
5861 | BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."), | |
5862 | BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."), | |
5863 | BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), | |
df373fb0 | 5864 | QCOW_COMMON_OPTIONS, |
1bd0e2d1 CL |
5865 | { /* end of list */ } |
5866 | } | |
20d97356 BS |
5867 | }; |
5868 | ||
2654267c HR |
5869 | static const char *const qcow2_strong_runtime_opts[] = { |
5870 | "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET, | |
5871 | ||
5872 | NULL | |
5873 | }; | |
5874 | ||
5f535a94 | 5875 | BlockDriver bdrv_qcow2 = { |
7c80ab3f | 5876 | .format_name = "qcow2", |
ff99129a | 5877 | .instance_size = sizeof(BDRVQcow2State), |
7c80ab3f JS |
5878 | .bdrv_probe = qcow2_probe, |
5879 | .bdrv_open = qcow2_open, | |
5880 | .bdrv_close = qcow2_close, | |
21d82ac9 | 5881 | .bdrv_reopen_prepare = qcow2_reopen_prepare, |
5b0959a7 | 5882 | .bdrv_reopen_commit = qcow2_reopen_commit, |
65eb7c85 | 5883 | .bdrv_reopen_commit_post = qcow2_reopen_commit_post, |
5b0959a7 | 5884 | .bdrv_reopen_abort = qcow2_reopen_abort, |
5365f44d | 5885 | .bdrv_join_options = qcow2_join_options, |
69dca43d | 5886 | .bdrv_child_perm = bdrv_default_perms, |
efc75e2a | 5887 | .bdrv_co_create_opts = qcow2_co_create_opts, |
b0292b85 | 5888 | .bdrv_co_create = qcow2_co_create, |
38841dcd | 5889 | .bdrv_has_zero_init = qcow2_has_zero_init, |
a320fb04 | 5890 | .bdrv_co_block_status = qcow2_co_block_status, |
7c80ab3f | 5891 | |
df893d25 | 5892 | .bdrv_co_preadv_part = qcow2_co_preadv_part, |
5396234b | 5893 | .bdrv_co_pwritev_part = qcow2_co_pwritev_part, |
eb489bb1 | 5894 | .bdrv_co_flush_to_os = qcow2_co_flush_to_os, |
419b19d9 | 5895 | |
5544b59f | 5896 | .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes, |
82e8a788 | 5897 | .bdrv_co_pdiscard = qcow2_co_pdiscard, |
fd9fcd37 FZ |
5898 | .bdrv_co_copy_range_from = qcow2_co_copy_range_from, |
5899 | .bdrv_co_copy_range_to = qcow2_co_copy_range_to, | |
061ca8a3 | 5900 | .bdrv_co_truncate = qcow2_co_truncate, |
5396234b | 5901 | .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part, |
491d27e2 | 5902 | .bdrv_make_empty = qcow2_make_empty, |
20d97356 BS |
5903 | |
5904 | .bdrv_snapshot_create = qcow2_snapshot_create, | |
5905 | .bdrv_snapshot_goto = qcow2_snapshot_goto, | |
5906 | .bdrv_snapshot_delete = qcow2_snapshot_delete, | |
5907 | .bdrv_snapshot_list = qcow2_snapshot_list, | |
1bd0e2d1 | 5908 | .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, |
c501c352 | 5909 | .bdrv_measure = qcow2_measure, |
1bd0e2d1 | 5910 | .bdrv_get_info = qcow2_get_info, |
37764dfb | 5911 | .bdrv_get_specific_info = qcow2_get_specific_info, |
20d97356 | 5912 | |
7c80ab3f JS |
5913 | .bdrv_save_vmstate = qcow2_save_vmstate, |
5914 | .bdrv_load_vmstate = qcow2_load_vmstate, | |
20d97356 | 5915 | |
d67066d8 | 5916 | .is_format = true, |
8ee79e70 | 5917 | .supports_backing = true, |
20d97356 BS |
5918 | .bdrv_change_backing_file = qcow2_change_backing_file, |
5919 | ||
d34682cd | 5920 | .bdrv_refresh_limits = qcow2_refresh_limits, |
2b148f39 | 5921 | .bdrv_co_invalidate_cache = qcow2_co_invalidate_cache, |
ec6d8912 | 5922 | .bdrv_inactivate = qcow2_inactivate, |
06d9260f | 5923 | |
1bd0e2d1 | 5924 | .create_opts = &qcow2_create_opts, |
df373fb0 | 5925 | .amend_opts = &qcow2_amend_opts, |
2654267c | 5926 | .strong_runtime_opts = qcow2_strong_runtime_opts, |
8a2ce0bc | 5927 | .mutable_opts = mutable_opts, |
2fd61638 | 5928 | .bdrv_co_check = qcow2_co_check, |
c282e1fd | 5929 | .bdrv_amend_options = qcow2_amend_options, |
8ea1613d | 5930 | .bdrv_co_amend = qcow2_co_amend, |
279621c0 AG |
5931 | |
5932 | .bdrv_detach_aio_context = qcow2_detach_aio_context, | |
5933 | .bdrv_attach_aio_context = qcow2_attach_aio_context, | |
1b6b0562 | 5934 | |
ef893b5c EB |
5935 | .bdrv_supports_persistent_dirty_bitmap = |
5936 | qcow2_supports_persistent_dirty_bitmap, | |
d2c3080e VSO |
5937 | .bdrv_co_can_store_new_dirty_bitmap = qcow2_co_can_store_new_dirty_bitmap, |
5938 | .bdrv_co_remove_persistent_dirty_bitmap = | |
5939 | qcow2_co_remove_persistent_dirty_bitmap, | |
20d97356 BS |
5940 | }; |
5941 | ||
5efa9d5a AL |
5942 | static void bdrv_qcow2_init(void) |
5943 | { | |
5944 | bdrv_register(&bdrv_qcow2); | |
5945 | } | |
5946 | ||
5947 | block_init(bdrv_qcow2_init); |