blob: 2e9c57e269e77e10ae649edd24d96bad4feb58fc [file] [log] [blame]
bellard585f8582006-08-05 21:14:20 +00001/*
2 * Block driver for the QCOW version 2 format
ths5fafdf22007-09-16 21:08:06 +00003 *
bellard585f8582006-08-05 21:14:20 +00004 * Copyright (c) 2004-2006 Fabrice Bellard
ths5fafdf22007-09-16 21:08:06 +00005 *
bellard585f8582006-08-05 21:14:20 +00006 * 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 */
Markus Armbrustere688df62018-02-01 12:18:31 +010024
Peter Maydell80c71a22016-01-18 18:01:42 +000025#include "qemu/osdep.h"
Vladimir Sementsov-Ogievskiy2714f132018-06-20 17:48:36 +030026
Max Reitz609f45e2018-06-14 21:14:28 +020027#include "block/qdict.h"
Kevin Wolf23588792016-03-08 15:57:05 +010028#include "sysemu/block-backend.h"
Markus Armbrusterdb725812019-08-12 07:23:50 +020029#include "qemu/main-loop.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010030#include "qemu/module.h"
Michael S. Tsirkin0d8c41d2018-05-03 22:50:20 +030031#include "qcow2.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010032#include "qemu/error-report.h"
Markus Armbrustere688df62018-02-01 12:18:31 +010033#include "qapi/error.h"
Markus Armbruster9af23982018-02-11 10:36:01 +010034#include "qapi/qapi-events-block-core.h"
Markus Armbruster6b673952018-02-01 12:18:35 +010035#include "qapi/qmp/qdict.h"
36#include "qapi/qmp/qstring.h"
Kevin Wolf3cce16f2012-03-01 18:36:21 +010037#include "trace.h"
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +080038#include "qemu/option_int.h"
Veronia Bahaaf348b6d2016-03-20 19:16:19 +020039#include "qemu/cutils.h"
Paolo Bonzini58369e22016-03-15 17:22:36 +010040#include "qemu/bswap.h"
Peter Maydell5df022c2022-02-26 18:07:23 +000041#include "qemu/memalign.h"
Kevin Wolfb76b4f62018-01-11 16:18:08 +010042#include "qapi/qobject-input-visitor.h"
43#include "qapi/qapi-visit-block-core.h"
Michael S. Tsirkin0d8c41d2018-05-03 22:50:20 +030044#include "crypto.h"
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +030045#include "block/aio_task.h"
Markus Armbrustere2c1c342022-12-21 14:35:49 +010046#include "block/dirty-bitmap.h"
bellard585f8582006-08-05 21:14:20 +000047
48/*
49 Differences with QCOW:
50
51 - Support for multiple incremental snapshots.
52 - Memory management by reference counts.
53 - Clusters which have a reference count of one have the bit
54 QCOW_OFLAG_COPIED to optimize write performance.
ths5fafdf22007-09-16 21:08:06 +000055 - Size of compressed clusters is stored in sectors to reduce bit usage
bellard585f8582006-08-05 21:14:20 +000056 in the cluster offsets.
57 - Support for storing additional data (such as the VM state) in the
ths3b46e622007-09-17 08:09:54 +000058 snapshots.
bellard585f8582006-08-05 21:14:20 +000059 - If a backing store is used, the cluster size is not constrained
60 (could be backported to QCOW).
61 - L2 tables have always a size of one cluster.
62*/
63
aliguori9b80ddf2009-03-28 17:55:06 +000064
65typedef struct {
66 uint32_t magic;
67 uint32_t len;
Jeff Codyc4217f62013-09-25 12:08:50 -040068} QEMU_PACKED QCowExtension;
Jeff Cody21d82ac2012-09-20 15:13:28 -040069
Jes Sorensen7c80ab32010-12-17 16:02:39 +010070#define QCOW2_EXT_MAGIC_END 0
Andrey Shinkevich80989692020-07-17 11:14:49 +030071#define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xe2792aca
Kevin Wolfcfcc4c62012-04-12 15:20:27 +020072#define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +010073#define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +030074#define QCOW2_EXT_MAGIC_BITMAPS 0x23852875
Kevin Wolf93c24932019-01-14 16:48:25 +010075#define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441
aliguori9b80ddf2009-03-28 17:55:06 +000076
Vladimir Sementsov-Ogievskiyc3c10f72018-11-01 21:27:37 +030077static int coroutine_fn
78qcow2_co_preadv_compressed(BlockDriverState *bs,
Vladimir Sementsov-Ogievskiy9a3978a2021-09-14 15:24:46 +030079 uint64_t l2_entry,
Vladimir Sementsov-Ogievskiyc3c10f72018-11-01 21:27:37 +030080 uint64_t offset,
81 uint64_t bytes,
Vladimir Sementsov-Ogievskiydf893d22019-06-04 19:15:13 +030082 QEMUIOVector *qiov,
83 size_t qiov_offset);
Vladimir Sementsov-Ogievskiyc3c10f72018-11-01 21:27:37 +030084
Jes Sorensen7c80ab32010-12-17 16:02:39 +010085static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
bellard585f8582006-08-05 21:14:20 +000086{
87 const QCowHeader *cow_header = (const void *)buf;
ths3b46e622007-09-17 08:09:54 +000088
bellard585f8582006-08-05 21:14:20 +000089 if (buf_size >= sizeof(QCowHeader) &&
90 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
Kevin Wolf6744cba2011-12-15 12:20:58 +010091 be32_to_cpu(cow_header->version) >= 2)
bellard585f8582006-08-05 21:14:20 +000092 return 100;
93 else
94 return 0;
95}
96
aliguori9b80ddf2009-03-28 17:55:06 +000097
Alberto Faria757dda52022-06-09 16:27:38 +010098static int qcow2_crypto_hdr_read_func(QCryptoBlock *block, size_t offset,
99 uint8_t *buf, size_t buflen,
100 void *opaque, Error **errp)
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100101{
102 BlockDriverState *bs = opaque;
103 BDRVQcow2State *s = bs->opaque;
104 ssize_t ret;
105
106 if ((offset + buflen) > s->crypto_header.length) {
107 error_setg(errp, "Request for data outside of extension header");
108 return -1;
109 }
110
Alberto Faria32cc71d2022-06-09 16:27:36 +0100111 ret = bdrv_pread(bs->file, s->crypto_header.offset + offset, buflen, buf,
Alberto Faria53fb7842022-06-09 16:27:35 +0100112 0);
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100113 if (ret < 0) {
114 error_setg_errno(errp, -ret, "Could not read encryption header");
115 return -1;
116 }
Alberto Faria757dda52022-06-09 16:27:38 +0100117 return 0;
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100118}
119
120
Alberto Faria757dda52022-06-09 16:27:38 +0100121static int qcow2_crypto_hdr_init_func(QCryptoBlock *block, size_t headerlen,
122 void *opaque, Error **errp)
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100123{
124 BlockDriverState *bs = opaque;
125 BDRVQcow2State *s = bs->opaque;
126 int64_t ret;
127 int64_t clusterlen;
128
129 ret = qcow2_alloc_clusters(bs, headerlen);
130 if (ret < 0) {
131 error_setg_errno(errp, -ret,
132 "Cannot allocate cluster for LUKS header size %zu",
133 headerlen);
134 return -1;
135 }
136
137 s->crypto_header.length = headerlen;
138 s->crypto_header.offset = ret;
139
Daniel P. Berrangé087ab8e2020-02-07 13:55:20 +0000140 /*
141 * Zero fill all space in cluster so it has predictable
142 * content, as we may not initialize some regions of the
143 * header (eg only 1 out of 8 key slots will be initialized)
144 */
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100145 clusterlen = size_to_clusters(s, headerlen) * s->cluster_size;
Kevin Wolf966b0002019-01-15 20:39:06 +0100146 assert(qcow2_pre_write_overlap_check(bs, 0, ret, clusterlen, false) == 0);
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100147 ret = bdrv_pwrite_zeroes(bs->file,
Daniel P. Berrangé087ab8e2020-02-07 13:55:20 +0000148 ret,
149 clusterlen, 0);
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100150 if (ret < 0) {
151 error_setg_errno(errp, -ret, "Could not zero fill encryption header");
152 return -1;
153 }
154
Alberto Faria757dda52022-06-09 16:27:38 +0100155 return 0;
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100156}
157
158
Alberto Faria757dda52022-06-09 16:27:38 +0100159static int qcow2_crypto_hdr_write_func(QCryptoBlock *block, size_t offset,
160 const uint8_t *buf, size_t buflen,
161 void *opaque, Error **errp)
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100162{
163 BlockDriverState *bs = opaque;
164 BDRVQcow2State *s = bs->opaque;
165 ssize_t ret;
166
167 if ((offset + buflen) > s->crypto_header.length) {
168 error_setg(errp, "Request for data outside of extension header");
169 return -1;
170 }
171
Alberto Faria32cc71d2022-06-09 16:27:36 +0100172 ret = bdrv_pwrite(bs->file, s->crypto_header.offset + offset, buflen, buf,
Alberto Faria53fb7842022-06-09 16:27:35 +0100173 0);
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100174 if (ret < 0) {
175 error_setg_errno(errp, -ret, "Could not read encryption header");
176 return -1;
177 }
Alberto Faria757dda52022-06-09 16:27:38 +0100178 return 0;
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100179}
180
Maxim Levitsky90766d92020-06-25 14:55:43 +0200181static QDict*
182qcow2_extract_crypto_opts(QemuOpts *opts, const char *fmt, Error **errp)
183{
184 QDict *cryptoopts_qdict;
185 QDict *opts_qdict;
186
187 /* Extract "encrypt." options into a qdict */
188 opts_qdict = qemu_opts_to_qdict(opts, NULL);
189 qdict_extract_subqdict(opts_qdict, &cryptoopts_qdict, "encrypt.");
190 qobject_unref(opts_qdict);
191 qdict_put_str(cryptoopts_qdict, "format", fmt);
192 return cryptoopts_qdict;
193}
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100194
Eric Blakea951a632020-03-24 12:42:30 -0500195/*
aliguori9b80ddf2009-03-28 17:55:06 +0000196 * read qcow2 extension and fill bs
197 * start reading from start_offset
198 * finish reading upon magic of value 0 or when end_offset reached
199 * unknown magic is skipped (future extension this version knows nothing about)
200 * return 0 upon success, non-0 otherwise
201 */
Jes Sorensen7c80ab32010-12-17 16:02:39 +0100202static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
Max Reitz3ef6c402013-09-05 09:40:43 +0200203 uint64_t end_offset, void **p_feature_table,
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +0300204 int flags, bool *need_update_header,
205 Error **errp)
aliguori9b80ddf2009-03-28 17:55:06 +0000206{
Kevin Wolfff991292015-09-07 17:12:56 +0200207 BDRVQcow2State *s = bs->opaque;
aliguori9b80ddf2009-03-28 17:55:06 +0000208 QCowExtension ext;
209 uint64_t offset;
Kevin Wolf75bab852012-02-02 14:52:08 +0100210 int ret;
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +0300211 Qcow2BitmapHeaderExt bitmaps_ext;
212
213 if (need_update_header != NULL) {
214 *need_update_header = false;
215 }
aliguori9b80ddf2009-03-28 17:55:06 +0000216
217#ifdef DEBUG_EXT
Jes Sorensen7c80ab32010-12-17 16:02:39 +0100218 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
aliguori9b80ddf2009-03-28 17:55:06 +0000219#endif
220 offset = start_offset;
221 while (offset < end_offset) {
222
223#ifdef DEBUG_EXT
224 /* Sanity check */
225 if (offset > s->cluster_size)
Jes Sorensen7c80ab32010-12-17 16:02:39 +0100226 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
aliguori9b80ddf2009-03-28 17:55:06 +0000227
Dong Xu Wang9b2260c2011-11-22 18:06:25 +0800228 printf("attempting to read extended header in offset %lu\n", offset);
aliguori9b80ddf2009-03-28 17:55:06 +0000229#endif
230
Alberto Faria32cc71d2022-06-09 16:27:36 +0100231 ret = bdrv_pread(bs->file, offset, sizeof(ext), &ext, 0);
Max Reitz3ef6c402013-09-05 09:40:43 +0200232 if (ret < 0) {
233 error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
234 "pread fail from offset %" PRIu64, offset);
aliguori9b80ddf2009-03-28 17:55:06 +0000235 return 1;
236 }
Peter Maydell3b698f52018-10-09 18:24:59 +0100237 ext.magic = be32_to_cpu(ext.magic);
238 ext.len = be32_to_cpu(ext.len);
aliguori9b80ddf2009-03-28 17:55:06 +0000239 offset += sizeof(ext);
240#ifdef DEBUG_EXT
241 printf("ext.magic = 0x%x\n", ext.magic);
242#endif
Kevin Wolf2ebafc82014-11-25 18:12:40 +0100243 if (offset > end_offset || ext.len > end_offset - offset) {
Max Reitz3ef6c402013-09-05 09:40:43 +0200244 error_setg(errp, "Header extension too large");
Kevin Wolf64ca6ae2012-02-22 12:37:13 +0100245 return -EINVAL;
246 }
247
aliguori9b80ddf2009-03-28 17:55:06 +0000248 switch (ext.magic) {
Jes Sorensen7c80ab32010-12-17 16:02:39 +0100249 case QCOW2_EXT_MAGIC_END:
aliguori9b80ddf2009-03-28 17:55:06 +0000250 return 0;
aliguorif9655092009-03-28 17:55:14 +0000251
Jes Sorensen7c80ab32010-12-17 16:02:39 +0100252 case QCOW2_EXT_MAGIC_BACKING_FORMAT:
aliguorif9655092009-03-28 17:55:14 +0000253 if (ext.len >= sizeof(bs->backing_format)) {
Max Reitz521b2b52014-04-29 19:03:12 +0200254 error_setg(errp, "ERROR: ext_backing_format: len=%" PRIu32
255 " too large (>=%zu)", ext.len,
256 sizeof(bs->backing_format));
aliguorif9655092009-03-28 17:55:14 +0000257 return 2;
258 }
Alberto Faria32cc71d2022-06-09 16:27:36 +0100259 ret = bdrv_pread(bs->file, offset, ext.len, bs->backing_format, 0);
Max Reitz3ef6c402013-09-05 09:40:43 +0200260 if (ret < 0) {
261 error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
262 "Could not read format name");
aliguorif9655092009-03-28 17:55:14 +0000263 return 3;
Max Reitz3ef6c402013-09-05 09:40:43 +0200264 }
aliguorif9655092009-03-28 17:55:14 +0000265 bs->backing_format[ext.len] = '\0';
Kevin Wolfe4603fe2015-04-07 15:03:16 +0200266 s->image_backing_format = g_strdup(bs->backing_format);
aliguorif9655092009-03-28 17:55:14 +0000267#ifdef DEBUG_EXT
268 printf("Qcow2: Got format extension %s\n", bs->backing_format);
269#endif
aliguorif9655092009-03-28 17:55:14 +0000270 break;
271
Kevin Wolfcfcc4c62012-04-12 15:20:27 +0200272 case QCOW2_EXT_MAGIC_FEATURE_TABLE:
273 if (p_feature_table != NULL) {
shiliyang5f14f312020-10-30 11:35:12 +0800274 void *feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature));
Alberto Faria32cc71d2022-06-09 16:27:36 +0100275 ret = bdrv_pread(bs->file, offset, ext.len, feature_table, 0);
Kevin Wolfcfcc4c62012-04-12 15:20:27 +0200276 if (ret < 0) {
Max Reitz3ef6c402013-09-05 09:40:43 +0200277 error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
278 "Could not read table");
lu zhipeng38f034e2022-09-21 22:45:15 +0800279 g_free(feature_table);
Kevin Wolfcfcc4c62012-04-12 15:20:27 +0200280 return ret;
281 }
282
283 *p_feature_table = feature_table;
284 }
285 break;
286
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100287 case QCOW2_EXT_MAGIC_CRYPTO_HEADER: {
288 unsigned int cflags = 0;
289 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
290 error_setg(errp, "CRYPTO header extension only "
291 "expected with LUKS encryption method");
292 return -EINVAL;
293 }
294 if (ext.len != sizeof(Qcow2CryptoHeaderExtension)) {
295 error_setg(errp, "CRYPTO header extension size %u, "
296 "but expected size %zu", ext.len,
297 sizeof(Qcow2CryptoHeaderExtension));
298 return -EINVAL;
299 }
300
Alberto Faria32cc71d2022-06-09 16:27:36 +0100301 ret = bdrv_pread(bs->file, offset, ext.len, &s->crypto_header, 0);
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100302 if (ret < 0) {
303 error_setg_errno(errp, -ret,
304 "Unable to read CRYPTO header extension");
305 return ret;
306 }
Peter Maydell3b698f52018-10-09 18:24:59 +0100307 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
308 s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100309
310 if ((s->crypto_header.offset % s->cluster_size) != 0) {
311 error_setg(errp, "Encryption header offset '%" PRIu64 "' is "
312 "not a multiple of cluster size '%u'",
313 s->crypto_header.offset, s->cluster_size);
314 return -EINVAL;
315 }
316
317 if (flags & BDRV_O_NO_IO) {
318 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
319 }
Daniel P. Berrange1cd9a782017-06-23 17:24:17 +0100320 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100321 qcow2_crypto_hdr_read_func,
Vladimir Sementsov-Ogievskiy8ac0f152019-05-06 17:27:41 +0300322 bs, cflags, QCOW2_MAX_THREADS, errp);
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100323 if (!s->crypto) {
324 return -EINVAL;
325 }
326 } break;
327
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +0300328 case QCOW2_EXT_MAGIC_BITMAPS:
329 if (ext.len != sizeof(bitmaps_ext)) {
330 error_setg_errno(errp, -ret, "bitmaps_ext: "
331 "Invalid extension length");
332 return -EINVAL;
333 }
334
335 if (!(s->autoclear_features & QCOW2_AUTOCLEAR_BITMAPS)) {
Max Reitzc9ceb3e2017-11-23 03:08:17 +0100336 if (s->qcow_version < 3) {
337 /* Let's be a bit more specific */
338 warn_report("This qcow2 v2 image contains bitmaps, but "
339 "they may have been modified by a program "
340 "without persistent bitmap support; so now "
341 "they must all be considered inconsistent");
342 } else {
343 warn_report("a program lacking bitmap support "
344 "modified this file, so all bitmaps are now "
345 "considered inconsistent");
346 }
Alistair Francis55d527a2017-09-11 12:52:46 -0700347 error_printf("Some clusters may be leaked, "
348 "run 'qemu-img check -r' on the image "
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +0300349 "file to fix.");
350 if (need_update_header != NULL) {
351 /* Updating is needed to drop invalid bitmap extension. */
352 *need_update_header = true;
353 }
354 break;
355 }
356
Alberto Faria32cc71d2022-06-09 16:27:36 +0100357 ret = bdrv_pread(bs->file, offset, ext.len, &bitmaps_ext, 0);
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +0300358 if (ret < 0) {
359 error_setg_errno(errp, -ret, "bitmaps_ext: "
360 "Could not read ext header");
361 return ret;
362 }
363
364 if (bitmaps_ext.reserved32 != 0) {
365 error_setg_errno(errp, -ret, "bitmaps_ext: "
366 "Reserved field is not zero");
367 return -EINVAL;
368 }
369
Peter Maydell3b698f52018-10-09 18:24:59 +0100370 bitmaps_ext.nb_bitmaps = be32_to_cpu(bitmaps_ext.nb_bitmaps);
371 bitmaps_ext.bitmap_directory_size =
372 be64_to_cpu(bitmaps_ext.bitmap_directory_size);
373 bitmaps_ext.bitmap_directory_offset =
374 be64_to_cpu(bitmaps_ext.bitmap_directory_offset);
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +0300375
376 if (bitmaps_ext.nb_bitmaps > QCOW2_MAX_BITMAPS) {
377 error_setg(errp,
378 "bitmaps_ext: Image has %" PRIu32 " bitmaps, "
379 "exceeding the QEMU supported maximum of %d",
380 bitmaps_ext.nb_bitmaps, QCOW2_MAX_BITMAPS);
381 return -EINVAL;
382 }
383
384 if (bitmaps_ext.nb_bitmaps == 0) {
385 error_setg(errp, "found bitmaps extension with zero bitmaps");
386 return -EINVAL;
387 }
388
Alberto Garcia74e60fb2019-12-12 12:01:21 +0200389 if (offset_into_cluster(s, bitmaps_ext.bitmap_directory_offset)) {
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +0300390 error_setg(errp, "bitmaps_ext: "
391 "invalid bitmap directory offset");
392 return -EINVAL;
393 }
394
395 if (bitmaps_ext.bitmap_directory_size >
396 QCOW2_MAX_BITMAP_DIRECTORY_SIZE) {
397 error_setg(errp, "bitmaps_ext: "
398 "bitmap directory size (%" PRIu64 ") exceeds "
399 "the maximum supported size (%d)",
400 bitmaps_ext.bitmap_directory_size,
401 QCOW2_MAX_BITMAP_DIRECTORY_SIZE);
402 return -EINVAL;
403 }
404
405 s->nb_bitmaps = bitmaps_ext.nb_bitmaps;
406 s->bitmap_directory_offset =
407 bitmaps_ext.bitmap_directory_offset;
408 s->bitmap_directory_size =
409 bitmaps_ext.bitmap_directory_size;
410
411#ifdef DEBUG_EXT
412 printf("Qcow2: Got bitmaps extension: "
413 "offset=%" PRIu64 " nb_bitmaps=%" PRIu32 "\n",
414 s->bitmap_directory_offset, s->nb_bitmaps);
415#endif
416 break;
417
Kevin Wolf9b890bd2019-01-15 19:02:40 +0100418 case QCOW2_EXT_MAGIC_DATA_FILE:
419 {
420 s->image_data_file = g_malloc0(ext.len + 1);
Alberto Faria32cc71d2022-06-09 16:27:36 +0100421 ret = bdrv_pread(bs->file, offset, ext.len, s->image_data_file, 0);
Kevin Wolf9b890bd2019-01-15 19:02:40 +0100422 if (ret < 0) {
423 error_setg_errno(errp, -ret,
424 "ERROR: Could not read data file name");
425 return ret;
426 }
427#ifdef DEBUG_EXT
428 printf("Qcow2: Got external data file %s\n", s->image_data_file);
429#endif
430 break;
431 }
432
aliguori9b80ddf2009-03-28 17:55:06 +0000433 default:
Kevin Wolf75bab852012-02-02 14:52:08 +0100434 /* unknown magic - save it in case we need to rewrite the header */
Eric Blake40969742017-11-17 10:47:47 -0600435 /* If you add a new feature, make sure to also update the fast
436 * path of qcow2_make_empty() to deal with it. */
Kevin Wolf75bab852012-02-02 14:52:08 +0100437 {
438 Qcow2UnknownHeaderExtension *uext;
439
440 uext = g_malloc0(sizeof(*uext) + ext.len);
441 uext->magic = ext.magic;
442 uext->len = ext.len;
443 QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next);
444
Alberto Faria32cc71d2022-06-09 16:27:36 +0100445 ret = bdrv_pread(bs->file, offset, uext->len, uext->data, 0);
Kevin Wolf75bab852012-02-02 14:52:08 +0100446 if (ret < 0) {
Max Reitz3ef6c402013-09-05 09:40:43 +0200447 error_setg_errno(errp, -ret, "ERROR: unknown extension: "
448 "Could not read data");
Kevin Wolf75bab852012-02-02 14:52:08 +0100449 return ret;
450 }
Kevin Wolf75bab852012-02-02 14:52:08 +0100451 }
aliguori9b80ddf2009-03-28 17:55:06 +0000452 break;
453 }
Kevin Wolffd29b4b2012-02-22 12:31:47 +0100454
455 offset += ((ext.len + 7) & ~7);
aliguori9b80ddf2009-03-28 17:55:06 +0000456 }
457
458 return 0;
459}
460
Kevin Wolf75bab852012-02-02 14:52:08 +0100461static void cleanup_unknown_header_ext(BlockDriverState *bs)
462{
Kevin Wolfff991292015-09-07 17:12:56 +0200463 BDRVQcow2State *s = bs->opaque;
Kevin Wolf75bab852012-02-02 14:52:08 +0100464 Qcow2UnknownHeaderExtension *uext, *next;
465
466 QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
467 QLIST_REMOVE(uext, next);
468 g_free(uext);
469 }
470}
aliguori9b80ddf2009-03-28 17:55:06 +0000471
Max Reitza55448b2016-03-16 19:54:33 +0100472static void report_unsupported_feature(Error **errp, Qcow2Feature *table,
473 uint64_t mask)
Kevin Wolfcfcc4c62012-04-12 15:20:27 +0200474{
Alberto Garcia7cdca2e2020-01-15 14:56:26 +0100475 g_autoptr(GString) features = g_string_sized_new(60);
Kevin Wolf12ac6d32014-07-17 11:41:53 +0200476
Kevin Wolfcfcc4c62012-04-12 15:20:27 +0200477 while (table && table->name[0] != '\0') {
478 if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
Kevin Wolf12ac6d32014-07-17 11:41:53 +0200479 if (mask & (1ULL << table->bit)) {
Alberto Garcia7cdca2e2020-01-15 14:56:26 +0100480 if (features->len > 0) {
481 g_string_append(features, ", ");
482 }
483 g_string_append_printf(features, "%.46s", table->name);
Kevin Wolf12ac6d32014-07-17 11:41:53 +0200484 mask &= ~(1ULL << table->bit);
Kevin Wolfcfcc4c62012-04-12 15:20:27 +0200485 }
486 }
487 table++;
488 }
489
490 if (mask) {
Alberto Garcia7cdca2e2020-01-15 14:56:26 +0100491 if (features->len > 0) {
492 g_string_append(features, ", ");
493 }
494 g_string_append_printf(features,
495 "Unknown incompatible feature: %" PRIx64, mask);
Kevin Wolfcfcc4c62012-04-12 15:20:27 +0200496 }
Kevin Wolf12ac6d32014-07-17 11:41:53 +0200497
Alberto Garcia7cdca2e2020-01-15 14:56:26 +0100498 error_setg(errp, "Unsupported qcow2 feature(s): %s", features->str);
Kevin Wolfcfcc4c62012-04-12 15:20:27 +0200499}
500
Stefan Hajnoczic61d0002012-07-27 09:05:19 +0100501/*
Stefan Hajnoczibfe80432012-07-27 09:05:22 +0100502 * Sets the dirty bit and flushes afterwards if necessary.
503 *
504 * The incompatible_features bit is only set if the image file header was
505 * updated successfully. Therefore it is not required to check the return
506 * value of this function.
507 */
Kevin Wolf280d3732012-12-07 18:08:47 +0100508int qcow2_mark_dirty(BlockDriverState *bs)
Stefan Hajnoczibfe80432012-07-27 09:05:22 +0100509{
Kevin Wolfff991292015-09-07 17:12:56 +0200510 BDRVQcow2State *s = bs->opaque;
Stefan Hajnoczibfe80432012-07-27 09:05:22 +0100511 uint64_t val;
512 int ret;
513
514 assert(s->qcow_version >= 3);
515
516 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
517 return 0; /* already dirty */
518 }
519
520 val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
Alberto Faria86da4322022-06-09 16:27:44 +0100521 ret = bdrv_pwrite_sync(bs->file,
522 offsetof(QCowHeader, incompatible_features),
523 sizeof(val), &val, 0);
Stefan Hajnoczibfe80432012-07-27 09:05:22 +0100524 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
533/*
Stefan Hajnoczic61d0002012-07-27 09:05:19 +0100534 * 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 */
538static int qcow2_mark_clean(BlockDriverState *bs)
539{
Kevin Wolfff991292015-09-07 17:12:56 +0200540 BDRVQcow2State *s = bs->opaque;
Stefan Hajnoczic61d0002012-07-27 09:05:19 +0100541
542 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
Kevin Wolf4c2e5f82014-04-03 13:47:50 +0200543 int ret;
544
545 s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
546
Paolo Bonzini8b220eb2018-03-01 17:36:14 +0100547 ret = qcow2_flush_caches(bs);
Stefan Hajnoczic61d0002012-07-27 09:05:19 +0100548 if (ret < 0) {
549 return ret;
550 }
551
Stefan Hajnoczic61d0002012-07-27 09:05:19 +0100552 return qcow2_update_header(bs);
553 }
554 return 0;
555}
556
Max Reitz69c98722013-08-30 14:34:24 +0200557/*
558 * Marks the image as corrupt.
559 */
560int qcow2_mark_corrupt(BlockDriverState *bs)
561{
Kevin Wolfff991292015-09-07 17:12:56 +0200562 BDRVQcow2State *s = bs->opaque;
Max Reitz69c98722013-08-30 14:34:24 +0200563
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 */
572int qcow2_mark_consistent(BlockDriverState *bs)
573{
Kevin Wolfff991292015-09-07 17:12:56 +0200574 BDRVQcow2State *s = bs->opaque;
Max Reitz69c98722013-08-30 14:34:24 +0200575
576 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
Paolo Bonzini8b220eb2018-03-01 17:36:14 +0100577 int ret = qcow2_flush_caches(bs);
Max Reitz69c98722013-08-30 14:34:24 +0200578 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
Max Reitz8bc584f2019-10-11 17:28:06 +0200588static 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
Paolo Bonzini2fd61632018-03-01 17:36:19 +0100604static int coroutine_fn qcow2_co_check_locked(BlockDriverState *bs,
605 BdrvCheckResult *result,
606 BdrvCheckMode fix)
Stefan Hajnocziacbe5982012-08-09 13:05:55 +0100607{
Max Reitz8bc584f2019-10-11 17:28:06 +0200608 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);
Max Reitz8bc584f2019-10-11 17:28:06 +0200615 if (ret < 0) {
Max Reitzfe446b52019-10-11 17:28:07 +0200616 qcow2_add_check_result(result, &snapshot_res, false);
Max Reitz8bc584f2019-10-11 17:28:06 +0200617 return ret;
618 }
619
620 ret = qcow2_check_refcounts(bs, &refcount_res, fix);
621 qcow2_add_check_result(result, &refcount_res, true);
Stefan Hajnocziacbe5982012-08-09 13:05:55 +0100622 if (ret < 0) {
Max Reitzfe446b52019-10-11 17:28:07 +0200623 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);
629 if (ret < 0) {
Stefan Hajnocziacbe5982012-08-09 13:05:55 +0100630 return ret;
631 }
632
633 if (fix && result->check_errors == 0 && result->corruptions == 0) {
Max Reitz24530f32013-08-30 14:34:30 +0200634 ret = qcow2_mark_clean(bs);
635 if (ret < 0) {
636 return ret;
637 }
638 return qcow2_mark_consistent(bs);
Stefan Hajnocziacbe5982012-08-09 13:05:55 +0100639 }
640 return ret;
641}
642
Paolo Bonzini2fd61632018-03-01 17:36:19 +0100643static 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
Alberto Garcia0cf0e592018-03-06 18:14:06 +0200656int 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)
Kevin Wolf8c7de282014-03-26 13:05:44 +0100660{
Kevin Wolfff991292015-09-07 17:12:56 +0200661 BDRVQcow2State *s = bs->opaque;
Alberto Garcia0cf0e592018-03-06 18:14:06 +0200662
663 if (entries > max_size_bytes / entry_len) {
664 error_setg(errp, "%s too large", table_name);
665 return -EFBIG;
666 }
Kevin Wolf8c7de282014-03-26 13:05:44 +0100667
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. */
Alberto Garcia0cf0e592018-03-06 18:14:06 +0200670 if ((INT64_MAX - entries * entry_len < offset) ||
671 (offset_into_cluster(s, offset) != 0)) {
672 error_setg(errp, "%s offset invalid", table_name);
Kevin Wolf8c7de282014-03-26 13:05:44 +0100673 return -EINVAL;
674 }
675
676 return 0;
677}
678
Alberto Garcia8a2ce0b2019-03-12 18:48:48 +0200679static 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
Kevin Wolf74c45102013-03-15 10:35:08 +0100703static QemuOptsList qcow2_runtime_opts = {
704 .name = "qcow2",
705 .head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head),
706 .desc = {
707 {
Kevin Wolf64aa99d2013-07-17 14:45:34 +0200708 .name = QCOW2_OPT_LAZY_REFCOUNTS,
Kevin Wolf74c45102013-03-15 10:35:08 +0100709 .type = QEMU_OPT_BOOL,
710 .help = "Postpone refcount updates",
711 },
Kevin Wolf67af6742013-06-19 13:44:19 +0200712 {
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 },
Max Reitz05de7e82013-10-10 11:09:25 +0200728 {
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 },
734 {
Max Reitzee42b5c2014-08-20 19:59:35 +0200735 .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 },
740 {
Max Reitz05de7e82013-10-10 11:09:25 +0200741 .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 },
Max Reitz6c1c8d52014-08-18 22:07:33 +0200780 {
Vladimir Sementsov-Ogievskiy0e4e4312018-07-05 18:15:15 +0300781 .name = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY,
782 .type = QEMU_OPT_BOOL,
783 .help = "Check for unintended writes into the bitmap directory",
784 },
785 {
Max Reitz6c1c8d52014-08-18 22:07:33 +0200786 .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 },
796 {
Alberto Garcia1221fe62018-02-05 16:33:36 +0200797 .name = QCOW2_OPT_L2_CACHE_ENTRY_SIZE,
798 .type = QEMU_OPT_SIZE,
799 .help = "Size of each entry in the L2 cache",
800 },
801 {
Max Reitz6c1c8d52014-08-18 22:07:33 +0200802 .name = QCOW2_OPT_REFCOUNT_CACHE_SIZE,
803 .type = QEMU_OPT_SIZE,
804 .help = "Maximum refcount block cache size",
805 },
Alberto Garcia279621c2015-08-04 15:14:40 +0300806 {
807 .name = QCOW2_OPT_CACHE_CLEAN_INTERVAL,
808 .type = QEMU_OPT_NUMBER,
809 .help = "Clean unused cache entries after this time (in seconds)",
810 },
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +0100811 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
812 "ID of secret providing qcow2 AES key or LUKS passphrase"),
Kevin Wolf74c45102013-03-15 10:35:08 +0100813 { /* end of list */ }
814 },
815};
816
Max Reitz4092e992013-10-10 11:09:26 +0200817static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = {
Vladimir Sementsov-Ogievskiy0e4e4312018-07-05 18:15:15 +0300818 [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,
Max Reitz4092e992013-10-10 11:09:26 +0200827};
828
Alberto Garcia279621c2015-08-04 15:14:40 +0300829static void cache_clean_timer_cb(void *opaque)
830{
831 BlockDriverState *bs = opaque;
Kevin Wolfff991292015-09-07 17:12:56 +0200832 BDRVQcow2State *s = bs->opaque;
Alberto Garciab2f68bf2018-02-05 16:33:09 +0200833 qcow2_cache_clean_unused(s->l2_table_cache);
834 qcow2_cache_clean_unused(s->refcount_block_cache);
Alberto Garcia279621c2015-08-04 15:14:40 +0300835 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
836 (int64_t) s->cache_clean_interval * 1000);
837}
838
839static void cache_clean_timer_init(BlockDriverState *bs, AioContext *context)
840{
Kevin Wolfff991292015-09-07 17:12:56 +0200841 BDRVQcow2State *s = bs->opaque;
Alberto Garcia279621c2015-08-04 15:14:40 +0300842 if (s->cache_clean_interval > 0) {
Pavel Dovgalyukad0ce642021-03-29 11:06:03 +0300843 s->cache_clean_timer =
844 aio_timer_new_with_attrs(context, QEMU_CLOCK_VIRTUAL,
845 SCALE_MS, QEMU_TIMER_ATTR_EXTERNAL,
846 cache_clean_timer_cb, bs);
Alberto Garcia279621c2015-08-04 15:14:40 +0300847 timer_mod(s->cache_clean_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
848 (int64_t) s->cache_clean_interval * 1000);
849 }
850}
851
852static void cache_clean_timer_del(BlockDriverState *bs)
853{
Kevin Wolfff991292015-09-07 17:12:56 +0200854 BDRVQcow2State *s = bs->opaque;
Alberto Garcia279621c2015-08-04 15:14:40 +0300855 if (s->cache_clean_timer) {
Alberto Garcia279621c2015-08-04 15:14:40 +0300856 timer_free(s->cache_clean_timer);
857 s->cache_clean_timer = NULL;
858 }
859}
860
861static void qcow2_detach_aio_context(BlockDriverState *bs)
862{
863 cache_clean_timer_del(bs);
864}
865
866static void qcow2_attach_aio_context(BlockDriverState *bs,
867 AioContext *new_context)
868{
869 cache_clean_timer_init(bs, new_context);
870}
871
Vladimir Sementsov-Ogievskiy772c4ca2021-02-02 15:49:53 +0300872static bool read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
Max Reitzbc85ef22015-06-01 18:09:19 +0200873 uint64_t *l2_cache_size,
Alberto Garcia1221fe62018-02-05 16:33:36 +0200874 uint64_t *l2_cache_entry_size,
Max Reitz6c1c8d52014-08-18 22:07:33 +0200875 uint64_t *refcount_cache_size, Error **errp)
876{
Kevin Wolfff991292015-09-07 17:12:56 +0200877 BDRVQcow2State *s = bs->opaque;
Leonid Blochb7495622018-09-26 19:04:43 +0300878 uint64_t combined_cache_size, l2_cache_max_setting;
Max Reitz6c1c8d52014-08-18 22:07:33 +0200879 bool l2_cache_size_set, refcount_cache_size_set, combined_cache_size_set;
Alberto Garciaaf39bd02019-02-13 18:48:53 +0200880 bool l2_cache_entry_size_set;
Alberto Garcia7af5eea2018-05-28 17:01:28 +0200881 int min_refcount_cache = MIN_REFCOUNT_CACHE_SIZE * s->cluster_size;
Leonid Blochb7495622018-09-26 19:04:43 +0300882 uint64_t virtual_disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
Alberto Garciab70d0822019-08-16 15:17:42 +0300883 uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size);
884 /* An L2 table is always one cluster in size so the max cache size
885 * should be a multiple of the cluster size. */
Alberto Garciac8fd8552020-07-10 18:12:54 +0200886 uint64_t max_l2_cache = ROUND_UP(max_l2_entries * l2_entry_size(s),
Alberto Garciab70d0822019-08-16 15:17:42 +0300887 s->cluster_size);
Max Reitz6c1c8d52014-08-18 22:07:33 +0200888
889 combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
890 l2_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_SIZE);
891 refcount_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
Alberto Garciaaf39bd02019-02-13 18:48:53 +0200892 l2_cache_entry_size_set = qemu_opt_get(opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE);
Max Reitz6c1c8d52014-08-18 22:07:33 +0200893
894 combined_cache_size = qemu_opt_get_size(opts, QCOW2_OPT_CACHE_SIZE, 0);
Leonid Blochb7495622018-09-26 19:04:43 +0300895 l2_cache_max_setting = qemu_opt_get_size(opts, QCOW2_OPT_L2_CACHE_SIZE,
896 DEFAULT_L2_CACHE_MAX_SIZE);
Max Reitz6c1c8d52014-08-18 22:07:33 +0200897 *refcount_cache_size = qemu_opt_get_size(opts,
898 QCOW2_OPT_REFCOUNT_CACHE_SIZE, 0);
899
Alberto Garcia1221fe62018-02-05 16:33:36 +0200900 *l2_cache_entry_size = qemu_opt_get_size(
901 opts, QCOW2_OPT_L2_CACHE_ENTRY_SIZE, s->cluster_size);
902
Leonid Blochb7495622018-09-26 19:04:43 +0300903 *l2_cache_size = MIN(max_l2_cache, l2_cache_max_setting);
904
Max Reitz6c1c8d52014-08-18 22:07:33 +0200905 if (combined_cache_size_set) {
906 if (l2_cache_size_set && refcount_cache_size_set) {
907 error_setg(errp, QCOW2_OPT_CACHE_SIZE ", " QCOW2_OPT_L2_CACHE_SIZE
908 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not be set "
Leonid Bloch308999e2018-07-25 17:27:55 +0300909 "at the same time");
Vladimir Sementsov-Ogievskiy772c4ca2021-02-02 15:49:53 +0300910 return false;
Leonid Blochb7495622018-09-26 19:04:43 +0300911 } else if (l2_cache_size_set &&
912 (l2_cache_max_setting > combined_cache_size)) {
Max Reitz6c1c8d52014-08-18 22:07:33 +0200913 error_setg(errp, QCOW2_OPT_L2_CACHE_SIZE " may not exceed "
914 QCOW2_OPT_CACHE_SIZE);
Vladimir Sementsov-Ogievskiy772c4ca2021-02-02 15:49:53 +0300915 return false;
Max Reitz6c1c8d52014-08-18 22:07:33 +0200916 } else if (*refcount_cache_size > combined_cache_size) {
917 error_setg(errp, QCOW2_OPT_REFCOUNT_CACHE_SIZE " may not exceed "
918 QCOW2_OPT_CACHE_SIZE);
Vladimir Sementsov-Ogievskiy772c4ca2021-02-02 15:49:53 +0300919 return false;
Max Reitz6c1c8d52014-08-18 22:07:33 +0200920 }
921
922 if (l2_cache_size_set) {
923 *refcount_cache_size = combined_cache_size - *l2_cache_size;
924 } else if (refcount_cache_size_set) {
925 *l2_cache_size = combined_cache_size - *refcount_cache_size;
926 } else {
Alberto Garcia52253992018-04-17 15:37:04 +0300927 /* Assign as much memory as possible to the L2 cache, and
928 * use the remainder for the refcount cache */
929 if (combined_cache_size >= max_l2_cache + min_refcount_cache) {
930 *l2_cache_size = max_l2_cache;
931 *refcount_cache_size = combined_cache_size - *l2_cache_size;
932 } else {
933 *refcount_cache_size =
934 MIN(combined_cache_size, min_refcount_cache);
935 *l2_cache_size = combined_cache_size - *refcount_cache_size;
936 }
Max Reitz6c1c8d52014-08-18 22:07:33 +0200937 }
Max Reitz6c1c8d52014-08-18 22:07:33 +0200938 }
Alberto Garciaaf39bd02019-02-13 18:48:53 +0200939
940 /*
941 * If the L2 cache is not enough to cover the whole disk then
942 * default to 4KB entries. Smaller entries reduce the cost of
943 * loads and evictions and increase I/O performance.
944 */
945 if (*l2_cache_size < max_l2_cache && !l2_cache_entry_size_set) {
946 *l2_cache_entry_size = MIN(s->cluster_size, 4096);
947 }
948
Leonid Bloch657ada52018-09-26 19:04:42 +0300949 /* l2_cache_size and refcount_cache_size are ensured to have at least
950 * their minimum values in qcow2_update_options_prepare() */
Alberto Garcia1221fe62018-02-05 16:33:36 +0200951
952 if (*l2_cache_entry_size < (1 << MIN_CLUSTER_BITS) ||
953 *l2_cache_entry_size > s->cluster_size ||
954 !is_power_of_2(*l2_cache_entry_size)) {
955 error_setg(errp, "L2 cache entry size must be a power of two "
956 "between %d and the cluster size (%d)",
957 1 << MIN_CLUSTER_BITS, s->cluster_size);
Vladimir Sementsov-Ogievskiy772c4ca2021-02-02 15:49:53 +0300958 return false;
Alberto Garcia1221fe62018-02-05 16:33:36 +0200959 }
Vladimir Sementsov-Ogievskiy772c4ca2021-02-02 15:49:53 +0300960
961 return true;
Max Reitz6c1c8d52014-08-18 22:07:33 +0200962}
963
Kevin Wolfee55b172015-04-16 16:16:02 +0200964typedef struct Qcow2ReopenState {
965 Qcow2Cache *l2_table_cache;
966 Qcow2Cache *refcount_block_cache;
Alberto Garcia3c2e5112018-02-05 16:33:13 +0200967 int l2_slice_size; /* Number of entries in a slice of the L2 table */
Kevin Wolfee55b172015-04-16 16:16:02 +0200968 bool use_lazy_refcounts;
969 int overlap_check;
970 bool discard_passthrough[QCOW2_DISCARD_MAX];
971 uint64_t cache_clean_interval;
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +0100972 QCryptoBlockOpenOptions *crypto_opts; /* Disk encryption runtime options */
Kevin Wolfee55b172015-04-16 16:16:02 +0200973} Qcow2ReopenState;
974
975static int qcow2_update_options_prepare(BlockDriverState *bs,
976 Qcow2ReopenState *r,
977 QDict *options, int flags,
978 Error **errp)
Kevin Wolf4c75d1a2015-04-16 11:29:27 +0200979{
980 BDRVQcow2State *s = bs->opaque;
Kevin Wolf94edf3f2015-04-16 11:44:26 +0200981 QemuOpts *opts = NULL;
Kevin Wolf4c75d1a2015-04-16 11:29:27 +0200982 const char *opt_overlap_check, *opt_overlap_check_template;
983 int overlap_check_template = 0;
Alberto Garcia1221fe62018-02-05 16:33:36 +0200984 uint64_t l2_cache_size, l2_cache_entry_size, refcount_cache_size;
Kevin Wolf4c75d1a2015-04-16 11:29:27 +0200985 int i;
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +0100986 const char *encryptfmt;
987 QDict *encryptopts = NULL;
Kevin Wolf4c75d1a2015-04-16 11:29:27 +0200988 int ret;
989
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +0100990 qdict_extract_subqdict(options, &encryptopts, "encrypt.");
991 encryptfmt = qdict_get_try_str(encryptopts, "format");
992
Kevin Wolf94edf3f2015-04-16 11:44:26 +0200993 opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort);
Markus Armbrusteraf175e82020-07-07 18:06:03 +0200994 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
Kevin Wolf94edf3f2015-04-16 11:44:26 +0200995 ret = -EINVAL;
996 goto fail;
997 }
998
999 /* get L2 table/refcount block cache size from command line options */
Vladimir Sementsov-Ogievskiy772c4ca2021-02-02 15:49:53 +03001000 if (!read_cache_sizes(bs, opts, &l2_cache_size, &l2_cache_entry_size,
1001 &refcount_cache_size, errp)) {
Kevin Wolf94edf3f2015-04-16 11:44:26 +02001002 ret = -EINVAL;
1003 goto fail;
1004 }
1005
Alberto Garcia1221fe62018-02-05 16:33:36 +02001006 l2_cache_size /= l2_cache_entry_size;
Kevin Wolf94edf3f2015-04-16 11:44:26 +02001007 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
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001026 /* 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
Alberto Garciac8fd8552020-07-10 18:12:54 +02001044 r->l2_slice_size = l2_cache_entry_size / l2_entry_size(s);
Alberto Garcia1221fe62018-02-05 16:33:36 +02001045 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);
Kevin Wolfee55b172015-04-16 16:16:02 +02001049 if (r->l2_table_cache == NULL || r->refcount_block_cache == NULL) {
Kevin Wolf94edf3f2015-04-16 11:44:26 +02001050 error_setg(errp, "Could not allocate metadata caches");
1051 ret = -ENOMEM;
1052 goto fail;
1053 }
1054
1055 /* New interval for cache cleanup timer */
Kevin Wolfee55b172015-04-16 16:16:02 +02001056 r->cache_clean_interval =
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001057 qemu_opt_get_number(opts, QCOW2_OPT_CACHE_CLEAN_INTERVAL,
Leonid Bloche957b502018-09-26 19:04:46 +03001058 DEFAULT_CACHE_CLEAN_INTERVAL);
Alberto Garcia91203f02016-11-25 13:27:44 +02001059#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
Kevin Wolfee55b172015-04-16 16:16:02 +02001067 if (r->cache_clean_interval > UINT_MAX) {
Kevin Wolf94edf3f2015-04-16 11:44:26 +02001068 error_setg(errp, "Cache clean interval too big");
1069 ret = -EINVAL;
1070 goto fail;
1071 }
Kevin Wolf94edf3f2015-04-16 11:44:26 +02001072
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001073 /* lazy-refcounts; flush if going from enabled to disabled */
Kevin Wolfee55b172015-04-16 16:16:02 +02001074 r->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS,
Kevin Wolf4c75d1a2015-04-16 11:29:27 +02001075 (s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS));
Kevin Wolfee55b172015-04-16 16:16:02 +02001076 if (r->use_lazy_refcounts && s->qcow_version < 3) {
Kevin Wolf007dbc32015-04-16 13:11:39 +02001077 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 }
Kevin Wolf4c75d1a2015-04-16 11:29:27 +02001082
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001083 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
Kevin Wolf007dbc32015-04-16 13:11:39 +02001091 /* Overlap check options */
Kevin Wolf4c75d1a2015-04-16 11:29:27 +02001092 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
Kevin Wolfee55b172015-04-16 16:16:02 +02001123 r->overlap_check = 0;
Kevin Wolf4c75d1a2015-04-16 11:29:27 +02001124 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 */
Kevin Wolfee55b172015-04-16 16:16:02 +02001127 r->overlap_check |=
Kevin Wolf4c75d1a2015-04-16 11:29:27 +02001128 qemu_opt_get_bool(opts, overlap_bool_option_names[i],
1129 overlap_check_template & (1 << i)) << i;
1130 }
1131
Kevin Wolfee55b172015-04-16 16:16:02 +02001132 r->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
1133 r->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
1134 r->discard_passthrough[QCOW2_DISCARD_REQUEST] =
Kevin Wolf007dbc32015-04-16 13:11:39 +02001135 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST,
1136 flags & BDRV_O_UNMAP);
Kevin Wolfee55b172015-04-16 16:16:02 +02001137 r->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] =
Kevin Wolf007dbc32015-04-16 13:11:39 +02001138 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
Kevin Wolfee55b172015-04-16 16:16:02 +02001139 r->discard_passthrough[QCOW2_DISCARD_OTHER] =
Kevin Wolf007dbc32015-04-16 13:11:39 +02001140 qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
1141
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01001142 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 }
Markus Armbruster796d3232018-06-26 19:41:19 +02001160 qdict_put_str(encryptopts, "format", "qcow");
1161 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
Vladimir Sementsov-Ogievskiy1184b412021-02-02 15:49:56 +03001162 if (!r->crypto_opts) {
1163 ret = -EINVAL;
1164 goto fail;
1165 }
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01001166 break;
1167
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +01001168 case QCOW_CRYPT_LUKS:
1169 if (encryptfmt && !g_str_equal(encryptfmt, "luks")) {
1170 error_setg(errp,
1171 "Header reported 'luks' encryption format but "
1172 "options specify '%s'", encryptfmt);
1173 ret = -EINVAL;
1174 goto fail;
1175 }
Markus Armbruster796d3232018-06-26 19:41:19 +02001176 qdict_put_str(encryptopts, "format", "luks");
1177 r->crypto_opts = block_crypto_open_opts_init(encryptopts, errp);
Vladimir Sementsov-Ogievskiy1184b412021-02-02 15:49:56 +03001178 if (!r->crypto_opts) {
1179 ret = -EINVAL;
1180 goto fail;
1181 }
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +01001182 break;
1183
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01001184 default:
1185 error_setg(errp, "Unsupported encryption method %d",
1186 s->crypt_method_header);
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01001187 ret = -EINVAL;
1188 goto fail;
1189 }
1190
Kevin Wolf4c75d1a2015-04-16 11:29:27 +02001191 ret = 0;
1192fail:
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02001193 qobject_unref(encryptopts);
Kevin Wolf94edf3f2015-04-16 11:44:26 +02001194 qemu_opts_del(opts);
1195 opts = NULL;
Kevin Wolfee55b172015-04-16 16:16:02 +02001196 return ret;
1197}
1198
1199static void qcow2_update_options_commit(BlockDriverState *bs,
1200 Qcow2ReopenState *r)
1201{
1202 BDRVQcow2State *s = bs->opaque;
1203 int i;
1204
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001205 if (s->l2_table_cache) {
Alberto Garciae64d4072018-02-05 16:33:08 +02001206 qcow2_cache_destroy(s->l2_table_cache);
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001207 }
1208 if (s->refcount_block_cache) {
Alberto Garciae64d4072018-02-05 16:33:08 +02001209 qcow2_cache_destroy(s->refcount_block_cache);
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001210 }
Kevin Wolfee55b172015-04-16 16:16:02 +02001211 s->l2_table_cache = r->l2_table_cache;
1212 s->refcount_block_cache = r->refcount_block_cache;
Alberto Garcia3c2e5112018-02-05 16:33:13 +02001213 s->l2_slice_size = r->l2_slice_size;
Kevin Wolfee55b172015-04-16 16:16:02 +02001214
1215 s->overlap_check = r->overlap_check;
1216 s->use_lazy_refcounts = r->use_lazy_refcounts;
1217
1218 for (i = 0; i < QCOW2_DISCARD_MAX; i++) {
1219 s->discard_passthrough[i] = r->discard_passthrough[i];
1220 }
1221
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001222 if (s->cache_clean_interval != r->cache_clean_interval) {
1223 cache_clean_timer_del(bs);
1224 s->cache_clean_interval = r->cache_clean_interval;
1225 cache_clean_timer_init(bs, bdrv_get_aio_context(bs));
1226 }
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01001227
1228 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
1229 s->crypto_opts = r->crypto_opts;
Kevin Wolfee55b172015-04-16 16:16:02 +02001230}
1231
1232static void qcow2_update_options_abort(BlockDriverState *bs,
1233 Qcow2ReopenState *r)
1234{
1235 if (r->l2_table_cache) {
Alberto Garciae64d4072018-02-05 16:33:08 +02001236 qcow2_cache_destroy(r->l2_table_cache);
Kevin Wolfee55b172015-04-16 16:16:02 +02001237 }
1238 if (r->refcount_block_cache) {
Alberto Garciae64d4072018-02-05 16:33:08 +02001239 qcow2_cache_destroy(r->refcount_block_cache);
Kevin Wolfee55b172015-04-16 16:16:02 +02001240 }
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01001241 qapi_free_QCryptoBlockOpenOptions(r->crypto_opts);
Kevin Wolfee55b172015-04-16 16:16:02 +02001242}
1243
1244static int qcow2_update_options(BlockDriverState *bs, QDict *options,
1245 int flags, Error **errp)
1246{
1247 Qcow2ReopenState r = {};
1248 int ret;
1249
1250 ret = qcow2_update_options_prepare(bs, &r, options, flags, errp);
1251 if (ret >= 0) {
1252 qcow2_update_options_commit(bs, &r);
1253 } else {
1254 qcow2_update_options_abort(bs, &r);
1255 }
Kevin Wolf94edf3f2015-04-16 11:44:26 +02001256
Kevin Wolf4c75d1a2015-04-16 11:29:27 +02001257 return ret;
1258}
1259
Denis Plotnikov572ad972020-05-07 11:25:18 +03001260static int validate_compression_type(BDRVQcow2State *s, Error **errp)
1261{
1262 switch (s->compression_type) {
1263 case QCOW2_COMPRESSION_TYPE_ZLIB:
Denis Plotnikovd298ac12020-05-07 11:25:20 +03001264#ifdef CONFIG_ZSTD
1265 case QCOW2_COMPRESSION_TYPE_ZSTD:
1266#endif
Denis Plotnikov572ad972020-05-07 11:25:18 +03001267 break;
1268
1269 default:
1270 error_setg(errp, "qcow2: unknown compression type: %u",
1271 s->compression_type);
1272 return -ENOTSUP;
1273 }
1274
1275 /*
1276 * if the compression type differs from QCOW2_COMPRESSION_TYPE_ZLIB
1277 * the incompatible feature flag must be set
1278 */
1279 if (s->compression_type == QCOW2_COMPRESSION_TYPE_ZLIB) {
1280 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) {
1281 error_setg(errp, "qcow2: Compression type incompatible feature "
1282 "bit must not be set");
1283 return -EINVAL;
1284 }
1285 } else {
1286 if (!(s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION)) {
1287 error_setg(errp, "qcow2: Compression type incompatible feature "
1288 "bit must be set");
1289 return -EINVAL;
1290 }
1291 }
1292
1293 return 0;
1294}
1295
Paolo Bonzini1fafcd92018-03-01 17:36:16 +01001296/* Called with s->lock held. */
1297static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
Hanna Reitz06e9cd12022-04-27 13:40:55 +02001298 int flags, bool open_data_file,
1299 Error **errp)
bellard585f8582006-08-05 21:14:20 +00001300{
Vladimir Sementsov-Ogievskiybc520242021-02-02 15:49:45 +03001301 ERRP_GUARD();
Kevin Wolfff991292015-09-07 17:12:56 +02001302 BDRVQcow2State *s = bs->opaque;
Kevin Wolf6d33e8e2014-03-26 13:05:47 +01001303 unsigned int len, i;
1304 int ret = 0;
bellard585f8582006-08-05 21:14:20 +00001305 QCowHeader header;
aliguori9b80ddf2009-03-28 17:55:06 +00001306 uint64_t ext_end;
Kevin Wolf2cf7cfa2013-05-14 16:14:33 +02001307 uint64_t l1_vm_state_index;
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +03001308 bool update_header = false;
bellard585f8582006-08-05 21:14:20 +00001309
Alberto Faria38505e22022-10-13 14:37:06 +02001310 ret = bdrv_co_pread(bs->file, 0, sizeof(header), &header, 0);
Jes Sorensen6d85a572010-12-17 16:02:40 +01001311 if (ret < 0) {
Max Reitz3ef6c402013-09-05 09:40:43 +02001312 error_setg_errno(errp, -ret, "Could not read qcow2 header");
bellard585f8582006-08-05 21:14:20 +00001313 goto fail;
Jes Sorensen6d85a572010-12-17 16:02:40 +01001314 }
Peter Maydell3b698f52018-10-09 18:24:59 +01001315 header.magic = be32_to_cpu(header.magic);
1316 header.version = be32_to_cpu(header.version);
1317 header.backing_file_offset = be64_to_cpu(header.backing_file_offset);
1318 header.backing_file_size = be32_to_cpu(header.backing_file_size);
1319 header.size = be64_to_cpu(header.size);
1320 header.cluster_bits = be32_to_cpu(header.cluster_bits);
1321 header.crypt_method = be32_to_cpu(header.crypt_method);
1322 header.l1_table_offset = be64_to_cpu(header.l1_table_offset);
1323 header.l1_size = be32_to_cpu(header.l1_size);
1324 header.refcount_table_offset = be64_to_cpu(header.refcount_table_offset);
1325 header.refcount_table_clusters =
1326 be32_to_cpu(header.refcount_table_clusters);
1327 header.snapshots_offset = be64_to_cpu(header.snapshots_offset);
1328 header.nb_snapshots = be32_to_cpu(header.nb_snapshots);
ths3b46e622007-09-17 08:09:54 +00001329
Kevin Wolfe8cdcec2011-02-09 11:11:07 +01001330 if (header.magic != QCOW_MAGIC) {
Max Reitz3ef6c402013-09-05 09:40:43 +02001331 error_setg(errp, "Image is not in qcow2 format");
Paolo Bonzini76abe402014-02-17 14:44:06 +01001332 ret = -EINVAL;
bellard585f8582006-08-05 21:14:20 +00001333 goto fail;
Jes Sorensen6d85a572010-12-17 16:02:40 +01001334 }
Kevin Wolf6744cba2011-12-15 12:20:58 +01001335 if (header.version < 2 || header.version > 3) {
Max Reitza55448b2016-03-16 19:54:33 +01001336 error_setg(errp, "Unsupported qcow2 version %" PRIu32, header.version);
Kevin Wolfe8cdcec2011-02-09 11:11:07 +01001337 ret = -ENOTSUP;
1338 goto fail;
1339 }
Kevin Wolf6744cba2011-12-15 12:20:58 +01001340
1341 s->qcow_version = header.version;
1342
Kevin Wolf24342f22014-03-26 13:05:41 +01001343 /* Initialise cluster size */
1344 if (header.cluster_bits < MIN_CLUSTER_BITS ||
1345 header.cluster_bits > MAX_CLUSTER_BITS) {
Max Reitz521b2b52014-04-29 19:03:12 +02001346 error_setg(errp, "Unsupported cluster size: 2^%" PRIu32,
1347 header.cluster_bits);
Kevin Wolf24342f22014-03-26 13:05:41 +01001348 ret = -EINVAL;
1349 goto fail;
1350 }
1351
1352 s->cluster_bits = header.cluster_bits;
1353 s->cluster_size = 1 << s->cluster_bits;
Kevin Wolf24342f22014-03-26 13:05:41 +01001354
Kevin Wolf6744cba2011-12-15 12:20:58 +01001355 /* Initialise version 3 header fields */
1356 if (header.version == 2) {
1357 header.incompatible_features = 0;
1358 header.compatible_features = 0;
1359 header.autoclear_features = 0;
1360 header.refcount_order = 4;
1361 header.header_length = 72;
1362 } else {
Peter Maydell3b698f52018-10-09 18:24:59 +01001363 header.incompatible_features =
1364 be64_to_cpu(header.incompatible_features);
1365 header.compatible_features = be64_to_cpu(header.compatible_features);
1366 header.autoclear_features = be64_to_cpu(header.autoclear_features);
1367 header.refcount_order = be32_to_cpu(header.refcount_order);
1368 header.header_length = be32_to_cpu(header.header_length);
Kevin Wolf24342f22014-03-26 13:05:41 +01001369
1370 if (header.header_length < 104) {
1371 error_setg(errp, "qcow2 header too short");
1372 ret = -EINVAL;
1373 goto fail;
1374 }
1375 }
1376
1377 if (header.header_length > s->cluster_size) {
1378 error_setg(errp, "qcow2 header exceeds cluster size");
1379 ret = -EINVAL;
1380 goto fail;
Kevin Wolf6744cba2011-12-15 12:20:58 +01001381 }
1382
1383 if (header.header_length > sizeof(header)) {
1384 s->unknown_header_fields_size = header.header_length - sizeof(header);
1385 s->unknown_header_fields = g_malloc(s->unknown_header_fields_size);
Alberto Faria38505e22022-10-13 14:37:06 +02001386 ret = bdrv_co_pread(bs->file, sizeof(header),
1387 s->unknown_header_fields_size,
1388 s->unknown_header_fields, 0);
Kevin Wolf6744cba2011-12-15 12:20:58 +01001389 if (ret < 0) {
Max Reitz3ef6c402013-09-05 09:40:43 +02001390 error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
1391 "fields");
Kevin Wolf6744cba2011-12-15 12:20:58 +01001392 goto fail;
1393 }
1394 }
1395
Kevin Wolfa1b3955c2014-03-26 13:05:42 +01001396 if (header.backing_file_offset > s->cluster_size) {
1397 error_setg(errp, "Invalid backing file offset");
1398 ret = -EINVAL;
1399 goto fail;
1400 }
1401
Kevin Wolfcfcc4c62012-04-12 15:20:27 +02001402 if (header.backing_file_offset) {
1403 ext_end = header.backing_file_offset;
1404 } else {
1405 ext_end = 1 << header.cluster_bits;
1406 }
1407
Kevin Wolf6744cba2011-12-15 12:20:58 +01001408 /* Handle feature bits */
1409 s->incompatible_features = header.incompatible_features;
1410 s->compatible_features = header.compatible_features;
1411 s->autoclear_features = header.autoclear_features;
1412
Denis Plotnikov572ad972020-05-07 11:25:18 +03001413 /*
1414 * Handle compression type
1415 * Older qcow2 images don't contain the compression type header.
1416 * Distinguish them by the header length and use
1417 * the only valid (default) compression type in that case
1418 */
1419 if (header.header_length > offsetof(QCowHeader, compression_type)) {
1420 s->compression_type = header.compression_type;
1421 } else {
1422 s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
1423 }
1424
1425 ret = validate_compression_type(s, errp);
1426 if (ret) {
1427 goto fail;
1428 }
1429
Stefan Hajnoczic61d0002012-07-27 09:05:19 +01001430 if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
Kevin Wolfcfcc4c62012-04-12 15:20:27 +02001431 void *feature_table = NULL;
1432 qcow2_read_extensions(bs, header.header_length, ext_end,
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +03001433 &feature_table, flags, NULL, NULL);
Max Reitza55448b2016-03-16 19:54:33 +01001434 report_unsupported_feature(errp, feature_table,
Stefan Hajnoczic61d0002012-07-27 09:05:19 +01001435 s->incompatible_features &
1436 ~QCOW2_INCOMPAT_MASK);
Kevin Wolf6744cba2011-12-15 12:20:58 +01001437 ret = -ENOTSUP;
Prasad Joshic5a33ee2014-03-28 23:08:58 +05301438 g_free(feature_table);
Kevin Wolf6744cba2011-12-15 12:20:58 +01001439 goto fail;
1440 }
1441
Max Reitz69c98722013-08-30 14:34:24 +02001442 if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
1443 /* Corrupt images may not be written to unless they are being repaired
1444 */
1445 if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
Max Reitz3ef6c402013-09-05 09:40:43 +02001446 error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
1447 "read/write");
Max Reitz69c98722013-08-30 14:34:24 +02001448 ret = -EACCES;
1449 goto fail;
1450 }
1451 }
1452
Alberto Garciad0346b52020-07-10 18:12:51 +02001453 s->subclusters_per_cluster =
1454 has_subclusters(s) ? QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER : 1;
1455 s->subcluster_size = s->cluster_size / s->subclusters_per_cluster;
1456 s->subcluster_bits = ctz32(s->subcluster_size);
1457
Alberto Garcia7be20252020-07-10 18:13:13 +02001458 if (s->subcluster_size < (1 << MIN_CLUSTER_BITS)) {
1459 error_setg(errp, "Unsupported subcluster size: %d", s->subcluster_size);
1460 ret = -EINVAL;
1461 goto fail;
1462 }
1463
Kevin Wolf6744cba2011-12-15 12:20:58 +01001464 /* Check support for various header values */
Max Reitzb72faf92015-02-10 15:28:52 -05001465 if (header.refcount_order > 6) {
1466 error_setg(errp, "Reference count entry width too large; may not "
1467 "exceed 64 bits");
1468 ret = -EINVAL;
Kevin Wolf6744cba2011-12-15 12:20:58 +01001469 goto fail;
1470 }
Max Reitzb6481f32013-09-03 10:09:53 +02001471 s->refcount_order = header.refcount_order;
Max Reitz346a53d2015-02-10 15:28:43 -05001472 s->refcount_bits = 1 << s->refcount_order;
1473 s->refcount_max = UINT64_C(1) << (s->refcount_bits - 1);
1474 s->refcount_max += s->refcount_max - 1;
Kevin Wolf6744cba2011-12-15 12:20:58 +01001475
bellard585f8582006-08-05 21:14:20 +00001476 s->crypt_method_header = header.crypt_method;
Jes Sorensen6d85a572010-12-17 16:02:40 +01001477 if (s->crypt_method_header) {
Daniel P. Berrangee6ff69b2016-03-21 14:11:48 +00001478 if (bdrv_uses_whitelist() &&
1479 s->crypt_method_header == QCOW_CRYPT_AES) {
Daniel P. Berrange8c0dcbc2016-06-13 12:30:09 +01001480 error_setg(errp,
1481 "Use of AES-CBC encrypted qcow2 images is no longer "
1482 "supported in system emulators");
1483 error_append_hint(errp,
1484 "You can use 'qemu-img convert' to convert your "
1485 "image to an alternative supported format, such "
1486 "as unencrypted qcow2, or raw with the LUKS "
1487 "format instead.\n");
1488 ret = -ENOSYS;
1489 goto fail;
Daniel P. Berrangee6ff69b2016-03-21 14:11:48 +00001490 }
1491
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +01001492 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1493 s->crypt_physical_offset = false;
1494 } else {
1495 /* Assuming LUKS and any future crypt methods we
1496 * add will all use physical offsets, due to the
1497 * fact that the alternative is insecure... */
1498 s->crypt_physical_offset = true;
1499 }
1500
Eric Blake54115412016-06-23 16:37:26 -06001501 bs->encrypted = true;
Jes Sorensen6d85a572010-12-17 16:02:40 +01001502 }
Kevin Wolf24342f22014-03-26 13:05:41 +01001503
Alberto Garciac8fd8552020-07-10 18:12:54 +02001504 s->l2_bits = s->cluster_bits - ctz32(l2_entry_size(s));
bellard585f8582006-08-05 21:14:20 +00001505 s->l2_size = 1 << s->l2_bits;
Max Reitz1d13d652014-10-22 14:09:28 +02001506 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1507 s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
1508 s->refcount_block_size = 1 << s->refcount_block_bits;
Leonid Blochbd016b92018-09-26 19:04:47 +03001509 bs->total_sectors = header.size / BDRV_SECTOR_SIZE;
bellard585f8582006-08-05 21:14:20 +00001510 s->csize_shift = (62 - (s->cluster_bits - 8));
1511 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
1512 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
Kevin Wolf5dab2fa2014-03-26 13:05:43 +01001513
bellard585f8582006-08-05 21:14:20 +00001514 s->refcount_table_offset = header.refcount_table_offset;
ths5fafdf22007-09-16 21:08:06 +00001515 s->refcount_table_size =
bellard585f8582006-08-05 21:14:20 +00001516 header.refcount_table_clusters << (s->cluster_bits - 3);
1517
Alberto Garcia951053a2017-11-03 16:18:53 +02001518 if (header.refcount_table_clusters == 0 && !(flags & BDRV_O_CHECK)) {
1519 error_setg(errp, "Image does not contain a reference count table");
1520 ret = -EINVAL;
1521 goto fail;
1522 }
1523
Alberto Garcia0cf0e592018-03-06 18:14:06 +02001524 ret = qcow2_validate_table(bs, s->refcount_table_offset,
1525 header.refcount_table_clusters,
1526 s->cluster_size, QCOW_MAX_REFTABLE_SIZE,
1527 "Reference count table", errp);
Kevin Wolf8c7de282014-03-26 13:05:44 +01001528 if (ret < 0) {
Kevin Wolf8c7de282014-03-26 13:05:44 +01001529 goto fail;
1530 }
1531
Max Reitz8bc584f2019-10-11 17:28:06 +02001532 if (!(flags & BDRV_O_CHECK)) {
1533 /*
1534 * The total size in bytes of the snapshot table is checked in
1535 * qcow2_read_snapshots() because the size of each snapshot is
1536 * variable and we don't know it yet.
1537 * Here we only check the offset and number of snapshots.
1538 */
1539 ret = qcow2_validate_table(bs, header.snapshots_offset,
1540 header.nb_snapshots,
1541 sizeof(QCowSnapshotHeader),
1542 sizeof(QCowSnapshotHeader) *
1543 QCOW_MAX_SNAPSHOTS,
1544 "Snapshot table", errp);
1545 if (ret < 0) {
1546 goto fail;
1547 }
Kevin Wolfce48f2f42014-03-26 13:05:45 +01001548 }
1549
bellard585f8582006-08-05 21:14:20 +00001550 /* read the level 1 table */
Alberto Garcia0cf0e592018-03-06 18:14:06 +02001551 ret = qcow2_validate_table(bs, header.l1_table_offset,
Alberto Garcia02b1ecf2020-08-28 13:08:28 +02001552 header.l1_size, L1E_SIZE,
Alberto Garcia0cf0e592018-03-06 18:14:06 +02001553 QCOW_MAX_L1_SIZE, "Active L1 table", errp);
1554 if (ret < 0) {
Kevin Wolf2d51c322014-03-26 13:05:46 +01001555 goto fail;
1556 }
bellard585f8582006-08-05 21:14:20 +00001557 s->l1_size = header.l1_size;
Alberto Garcia0cf0e592018-03-06 18:14:06 +02001558 s->l1_table_offset = header.l1_table_offset;
Kevin Wolf2cf7cfa2013-05-14 16:14:33 +02001559
1560 l1_vm_state_index = size_to_l1(s, header.size);
1561 if (l1_vm_state_index > INT_MAX) {
Max Reitz3ef6c402013-09-05 09:40:43 +02001562 error_setg(errp, "Image is too big");
Kevin Wolf2cf7cfa2013-05-14 16:14:33 +02001563 ret = -EFBIG;
1564 goto fail;
1565 }
1566 s->l1_vm_state_index = l1_vm_state_index;
1567
bellard585f8582006-08-05 21:14:20 +00001568 /* the L1 table must contain at least enough entries to put
1569 header.size bytes */
Jes Sorensen6d85a572010-12-17 16:02:40 +01001570 if (s->l1_size < s->l1_vm_state_index) {
Max Reitz3ef6c402013-09-05 09:40:43 +02001571 error_setg(errp, "L1 table is too small");
Jes Sorensen6d85a572010-12-17 16:02:40 +01001572 ret = -EINVAL;
bellard585f8582006-08-05 21:14:20 +00001573 goto fail;
Jes Sorensen6d85a572010-12-17 16:02:40 +01001574 }
Kevin Wolf2d51c322014-03-26 13:05:46 +01001575
Stefan Weild191d122009-10-26 16:11:16 +01001576 if (s->l1_size > 0) {
Alberto Garcia02b1ecf2020-08-28 13:08:28 +02001577 s->l1_table = qemu_try_blockalign(bs->file->bs, s->l1_size * L1E_SIZE);
Kevin Wolfde828152014-05-20 17:12:47 +02001578 if (s->l1_table == NULL) {
1579 error_setg(errp, "Could not allocate L1 table");
1580 ret = -ENOMEM;
1581 goto fail;
1582 }
Alberto Faria38505e22022-10-13 14:37:06 +02001583 ret = bdrv_co_pread(bs->file, s->l1_table_offset, s->l1_size * L1E_SIZE,
1584 s->l1_table, 0);
Jes Sorensen6d85a572010-12-17 16:02:40 +01001585 if (ret < 0) {
Max Reitz3ef6c402013-09-05 09:40:43 +02001586 error_setg_errno(errp, -ret, "Could not read L1 table");
Stefan Weild191d122009-10-26 16:11:16 +01001587 goto fail;
Jes Sorensen6d85a572010-12-17 16:02:40 +01001588 }
Stefan Weild191d122009-10-26 16:11:16 +01001589 for(i = 0;i < s->l1_size; i++) {
Peter Maydell3b698f52018-10-09 18:24:59 +01001590 s->l1_table[i] = be64_to_cpu(s->l1_table[i]);
Stefan Weild191d122009-10-26 16:11:16 +01001591 }
bellard585f8582006-08-05 21:14:20 +00001592 }
Kevin Wolf29c1a732011-01-10 17:17:28 +01001593
Kevin Wolf94edf3f2015-04-16 11:44:26 +02001594 /* Parse driver-specific options */
1595 ret = qcow2_update_options(bs, options, flags, errp);
Kevin Wolf90efa0e2015-04-16 11:36:10 +02001596 if (ret < 0) {
1597 goto fail;
1598 }
1599
Anthony Liguori06d92602011-11-14 15:09:46 -06001600 s->flags = flags;
ths3b46e622007-09-17 08:09:54 +00001601
Jes Sorensen6d85a572010-12-17 16:02:40 +01001602 ret = qcow2_refcount_init(bs);
1603 if (ret != 0) {
Max Reitz3ef6c402013-09-05 09:40:43 +02001604 error_setg_errno(errp, -ret, "Could not initialize refcount handling");
bellard585f8582006-08-05 21:14:20 +00001605 goto fail;
Jes Sorensen6d85a572010-12-17 16:02:40 +01001606 }
bellard585f8582006-08-05 21:14:20 +00001607
Blue Swirl72cf2d42009-09-12 07:36:22 +00001608 QLIST_INIT(&s->cluster_allocs);
Kevin Wolf0b919fa2013-06-19 13:44:20 +02001609 QTAILQ_INIT(&s->discards);
Kevin Wolff2149782009-08-31 16:48:49 +02001610
aliguori9b80ddf2009-03-28 17:55:06 +00001611 /* read qcow2 extensions */
Max Reitz3ef6c402013-09-05 09:40:43 +02001612 if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL,
Markus Armbrusteraf175e82020-07-07 18:06:03 +02001613 flags, &update_header, errp)) {
Jes Sorensen6d85a572010-12-17 16:02:40 +01001614 ret = -EINVAL;
aliguori9b80ddf2009-03-28 17:55:06 +00001615 goto fail;
Jes Sorensen6d85a572010-12-17 16:02:40 +01001616 }
aliguori9b80ddf2009-03-28 17:55:06 +00001617
Hanna Reitz06e9cd12022-04-27 13:40:55 +02001618 if (open_data_file) {
1619 /* Open external data file */
1620 s->data_file = bdrv_open_child(NULL, options, "data-file", bs,
1621 &child_of_bds, BDRV_CHILD_DATA,
1622 true, errp);
1623 if (*errp) {
1624 ret = -EINVAL;
1625 goto fail;
1626 }
Kevin Wolf0e8c08b2019-01-29 17:13:57 +01001627
Hanna Reitz06e9cd12022-04-27 13:40:55 +02001628 if (s->incompatible_features & QCOW2_INCOMPAT_DATA_FILE) {
1629 if (!s->data_file && s->image_data_file) {
1630 s->data_file = bdrv_open_child(s->image_data_file, options,
1631 "data-file", bs, &child_of_bds,
1632 BDRV_CHILD_DATA, false, errp);
1633 if (!s->data_file) {
1634 ret = -EINVAL;
1635 goto fail;
1636 }
1637 }
Kevin Wolf9b890bd2019-01-15 19:02:40 +01001638 if (!s->data_file) {
Hanna Reitz06e9cd12022-04-27 13:40:55 +02001639 error_setg(errp, "'data-file' is required for this image");
Kevin Wolf9b890bd2019-01-15 19:02:40 +01001640 ret = -EINVAL;
1641 goto fail;
1642 }
Max Reitz8b1869d2020-05-13 13:05:35 +02001643
Hanna Reitz06e9cd12022-04-27 13:40:55 +02001644 /* No data here */
1645 bs->file->role &= ~BDRV_CHILD_DATA;
Max Reitz8b1869d2020-05-13 13:05:35 +02001646
Hanna Reitz06e9cd12022-04-27 13:40:55 +02001647 /* Must succeed because we have given up permissions if anything */
1648 bdrv_child_refresh_perms(bs, bs->file, &error_abort);
1649 } else {
1650 if (s->data_file) {
1651 error_setg(errp, "'data-file' can only be set for images with "
1652 "an external data file");
1653 ret = -EINVAL;
1654 goto fail;
1655 }
Kevin Wolf6c3944d2019-02-22 14:29:38 +01001656
Hanna Reitz06e9cd12022-04-27 13:40:55 +02001657 s->data_file = bs->file;
Kevin Wolf6c3944d2019-02-22 14:29:38 +01001658
Hanna Reitz06e9cd12022-04-27 13:40:55 +02001659 if (data_file_is_raw(bs)) {
1660 error_setg(errp, "data-file-raw requires a data file");
1661 ret = -EINVAL;
1662 goto fail;
1663 }
Kevin Wolf0e8c08b2019-01-29 17:13:57 +01001664 }
1665 }
Kevin Wolf93c24932019-01-14 16:48:25 +01001666
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +01001667 /* qcow2_read_extension may have set up the crypto context
1668 * if the crypt method needs a header region, some methods
1669 * don't need header extensions, so must check here
1670 */
1671 if (s->crypt_method_header && !s->crypto) {
1672 if (s->crypt_method_header == QCOW_CRYPT_AES) {
1673 unsigned int cflags = 0;
1674 if (flags & BDRV_O_NO_IO) {
1675 cflags |= QCRYPTO_BLOCK_OPEN_NO_IO;
1676 }
Daniel P. Berrange1cd9a782017-06-23 17:24:17 +01001677 s->crypto = qcrypto_block_open(s->crypto_opts, "encrypt.",
Vladimir Sementsov-Ogievskiy8ac0f152019-05-06 17:27:41 +03001678 NULL, NULL, cflags,
1679 QCOW2_MAX_THREADS, errp);
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +01001680 if (!s->crypto) {
1681 ret = -EINVAL;
1682 goto fail;
1683 }
1684 } else if (!(flags & BDRV_O_NO_IO)) {
1685 error_setg(errp, "Missing CRYPTO header for crypt method %d",
1686 s->crypt_method_header);
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01001687 ret = -EINVAL;
1688 goto fail;
1689 }
1690 }
1691
bellard585f8582006-08-05 21:14:20 +00001692 /* read the backing file name */
1693 if (header.backing_file_offset != 0) {
1694 len = header.backing_file_size;
Jeff Cody9a29e182015-01-22 08:03:30 -05001695 if (len > MIN(1023, s->cluster_size - header.backing_file_offset) ||
Jeff Codye729fa62015-01-27 08:33:55 -05001696 len >= sizeof(bs->backing_file)) {
Kevin Wolf6d33e8e2014-03-26 13:05:47 +01001697 error_setg(errp, "Backing file name too long");
1698 ret = -EINVAL;
1699 goto fail;
Jes Sorensen6d85a572010-12-17 16:02:40 +01001700 }
Hanna Reitzec64b1c2022-08-03 16:44:44 +02001701
1702 s->image_backing_file = g_malloc(len + 1);
Alberto Faria38505e22022-10-13 14:37:06 +02001703 ret = bdrv_co_pread(bs->file, header.backing_file_offset, len,
1704 s->image_backing_file, 0);
Jes Sorensen6d85a572010-12-17 16:02:40 +01001705 if (ret < 0) {
Max Reitz3ef6c402013-09-05 09:40:43 +02001706 error_setg_errno(errp, -ret, "Could not read backing file name");
bellard585f8582006-08-05 21:14:20 +00001707 goto fail;
Jes Sorensen6d85a572010-12-17 16:02:40 +01001708 }
Hanna Reitzec64b1c2022-08-03 16:44:44 +02001709 s->image_backing_file[len] = '\0';
1710
1711 /*
1712 * Update only when something has changed. This function is called by
1713 * qcow2_co_invalidate_cache(), and we do not want to reset
1714 * auto_backing_file unless necessary.
1715 */
1716 if (!g_str_equal(s->image_backing_file, bs->backing_file)) {
1717 pstrcpy(bs->backing_file, sizeof(bs->backing_file),
1718 s->image_backing_file);
1719 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
1720 s->image_backing_file);
1721 }
bellard585f8582006-08-05 21:14:20 +00001722 }
Kevin Wolf42deb292011-11-16 11:43:28 +01001723
Max Reitz8bc584f2019-10-11 17:28:06 +02001724 /*
1725 * Internal snapshots; skip reading them in check mode, because
1726 * we do not need them then, and we do not want to abort because
1727 * of a broken table.
1728 */
1729 if (!(flags & BDRV_O_CHECK)) {
1730 s->snapshots_offset = header.snapshots_offset;
1731 s->nb_snapshots = header.nb_snapshots;
Kevin Wolf11b128f2014-03-26 13:06:04 +01001732
Max Reitz8bc584f2019-10-11 17:28:06 +02001733 ret = qcow2_read_snapshots(bs, errp);
1734 if (ret < 0) {
1735 goto fail;
1736 }
Jes Sorensen6d85a572010-12-17 16:02:40 +01001737 }
bellard585f8582006-08-05 21:14:20 +00001738
Stefan Hajnocziaf7b7082012-06-14 11:42:23 +01001739 /* Clear unknown autoclear feature bits */
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +03001740 update_header |= s->autoclear_features & ~QCOW2_AUTOCLEAR_MASK;
Vladimir Sementsov-Ogievskiy307261b2021-05-27 18:40:54 +03001741 update_header = update_header && bdrv_is_writable(bs);
Vladimir Sementsov-Ogievskiyd1258dd2017-06-28 15:05:11 +03001742 if (update_header) {
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +03001743 s->autoclear_features &= QCOW2_AUTOCLEAR_MASK;
Vladimir Sementsov-Ogievskiyd1258dd2017-06-28 15:05:11 +03001744 }
1745
Vladimir Sementsov-Ogievskiy9c98f142018-10-29 16:23:17 -04001746 /* == Handle persistent dirty bitmaps ==
1747 *
1748 * We want load dirty bitmaps in three cases:
1749 *
1750 * 1. Normal open of the disk in active mode, not related to invalidation
1751 * after migration.
1752 *
1753 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1754 * bitmaps are _not_ migrating through migration channel, i.e.
1755 * 'dirty-bitmaps' capability is disabled.
1756 *
1757 * 3. Invalidation of source vm after failed or canceled migration.
1758 * This is a very interesting case. There are two possible types of
1759 * bitmaps:
1760 *
1761 * A. Stored on inactivation and removed. They should be loaded from the
1762 * image.
1763 *
1764 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1765 * the migration channel (with dirty-bitmaps capability).
1766 *
1767 * On the other hand, there are two possible sub-cases:
1768 *
1769 * 3.1 disk was changed by somebody else while were inactive. In this
1770 * case all in-RAM dirty bitmaps (both persistent and not) are
1771 * definitely invalid. And we don't have any method to determine
1772 * this.
1773 *
1774 * Simple and safe thing is to just drop all the bitmaps of type B on
1775 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1776 *
1777 * On the other hand, resuming source vm, if disk was already changed
1778 * is a bad thing anyway: not only bitmaps, the whole vm state is
1779 * out of sync with disk.
1780 *
1781 * This means, that user or management tool, who for some reason
1782 * decided to resume source vm, after disk was already changed by
1783 * target vm, should at least drop all dirty bitmaps by hand.
1784 *
1785 * So, we can ignore this case for now, but TODO: "generation"
1786 * extension for qcow2, to determine, that image was changed after
1787 * last inactivation. And if it is changed, we will drop (or at least
1788 * mark as 'invalid' all the bitmaps of type B, both persistent
1789 * and not).
1790 *
1791 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1792 * to disk ('dirty-bitmaps' capability disabled), or not saved
1793 * ('dirty-bitmaps' capability enabled), but we don't need to care
1794 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1795 * and not stored has flag IN_USE=1 in the image and will be skipped
1796 * on loading.
1797 *
1798 * One remaining possible case when we don't want load bitmaps:
1799 *
1800 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1801 * will be loaded on invalidation, no needs try loading them before)
1802 */
1803
1804 if (!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)) {
1805 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
Vladimir Sementsov-Ogievskiy0c1e9d22021-02-02 15:49:51 +03001806 bool header_updated;
1807 if (!qcow2_load_dirty_bitmaps(bs, &header_updated, errp)) {
Tuguoyi66be5c32019-12-18 11:53:35 +00001808 ret = -EINVAL;
1809 goto fail;
1810 }
Vladimir Sementsov-Ogievskiy9c98f142018-10-29 16:23:17 -04001811
1812 update_header = update_header && !header_updated;
Vladimir Sementsov-Ogievskiyd1258dd2017-06-28 15:05:11 +03001813 }
Vladimir Sementsov-Ogievskiyd1258dd2017-06-28 15:05:11 +03001814
1815 if (update_header) {
Stefan Hajnocziaf7b7082012-06-14 11:42:23 +01001816 ret = qcow2_update_header(bs);
1817 if (ret < 0) {
Max Reitz3ef6c402013-09-05 09:40:43 +02001818 error_setg_errno(errp, -ret, "Could not update qcow2 header");
Stefan Hajnocziaf7b7082012-06-14 11:42:23 +01001819 goto fail;
1820 }
1821 }
1822
Kevin Wolf3b650812019-11-22 16:57:48 +01001823 bs->supported_zero_flags = header.version >= 3 ?
1824 BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK : 0;
Kevin Wolff01643f2020-04-24 14:54:42 +02001825 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
Kevin Wolf68d100e2011-06-30 17:42:09 +02001826
Stefan Hajnoczic61d0002012-07-27 09:05:19 +01001827 /* Repair image if dirty */
Vladimir Sementsov-Ogievskiy307261b2021-05-27 18:40:54 +03001828 if (!(flags & BDRV_O_CHECK) && bdrv_is_writable(bs) &&
Stefan Hajnoczi058f8f12012-08-09 13:05:56 +01001829 (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
Stefan Hajnoczic61d0002012-07-27 09:05:19 +01001830 BdrvCheckResult result = {0};
1831
Paolo Bonzini2fd61632018-03-01 17:36:19 +01001832 ret = qcow2_co_check_locked(bs, &result,
1833 BDRV_FIX_ERRORS | BDRV_FIX_LEAKS);
Max Reitz791fff52017-11-10 21:31:07 +01001834 if (ret < 0 || result.check_errors) {
1835 if (ret >= 0) {
1836 ret = -EIO;
1837 }
Max Reitz3ef6c402013-09-05 09:40:43 +02001838 error_setg_errno(errp, -ret, "Could not repair dirty image");
Stefan Hajnoczic61d0002012-07-27 09:05:19 +01001839 goto fail;
1840 }
1841 }
1842
bellard585f8582006-08-05 21:14:20 +00001843#ifdef DEBUG_ALLOC
Philipp Hahn6cbc3032011-08-04 19:22:10 +02001844 {
1845 BdrvCheckResult result = {0};
Stefan Hajnoczib35278f2012-06-15 16:41:07 +01001846 qcow2_check_refcounts(bs, &result, 0);
Philipp Hahn6cbc3032011-08-04 19:22:10 +02001847 }
bellard585f8582006-08-05 21:14:20 +00001848#endif
Vladimir Sementsov-Ogievskiyceb029c2018-06-20 17:48:37 +03001849
Vladimir Sementsov-Ogievskiy6f13a312019-05-06 17:27:38 +03001850 qemu_co_queue_init(&s->thread_task_queue);
Vladimir Sementsov-Ogievskiyceb029c2018-06-20 17:48:37 +03001851
Jes Sorensen6d85a572010-12-17 16:02:40 +01001852 return ret;
bellard585f8582006-08-05 21:14:20 +00001853
1854 fail:
Kevin Wolf9b890bd2019-01-15 19:02:40 +01001855 g_free(s->image_data_file);
Hanna Reitz06e9cd12022-04-27 13:40:55 +02001856 if (open_data_file && has_data_file(bs)) {
Kevin Wolf0e8c08b2019-01-29 17:13:57 +01001857 bdrv_unref_child(bs, s->data_file);
Vladimir Sementsov-Ogievskiy808cf3c2020-03-16 09:06:31 +03001858 s->data_file = NULL;
Kevin Wolf0e8c08b2019-01-29 17:13:57 +01001859 }
Kevin Wolf6744cba2011-12-15 12:20:58 +01001860 g_free(s->unknown_header_fields);
Kevin Wolf75bab852012-02-02 14:52:08 +01001861 cleanup_unknown_header_ext(bs);
Kevin Wolfed6ccf02009-05-28 16:07:07 +02001862 qcow2_free_snapshots(bs);
1863 qcow2_refcount_close(bs);
Kevin Wolfde828152014-05-20 17:12:47 +02001864 qemu_vfree(s->l1_table);
Max Reitzcf939802013-08-30 14:34:26 +02001865 /* else pre-write overlap checks in cache_destroy may crash */
1866 s->l1_table = NULL;
Alberto Garcia279621c2015-08-04 15:14:40 +03001867 cache_clean_timer_del(bs);
Kevin Wolf29c1a732011-01-10 17:17:28 +01001868 if (s->l2_table_cache) {
Alberto Garciae64d4072018-02-05 16:33:08 +02001869 qcow2_cache_destroy(s->l2_table_cache);
Kevin Wolf29c1a732011-01-10 17:17:28 +01001870 }
Prasad Joshic5a33ee2014-03-28 23:08:58 +05301871 if (s->refcount_block_cache) {
Alberto Garciae64d4072018-02-05 16:33:08 +02001872 qcow2_cache_destroy(s->refcount_block_cache);
Prasad Joshic5a33ee2014-03-28 23:08:58 +05301873 }
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01001874 qcrypto_block_free(s->crypto);
1875 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
Jes Sorensen6d85a572010-12-17 16:02:40 +01001876 return ret;
bellard585f8582006-08-05 21:14:20 +00001877}
1878
Paolo Bonzini1fafcd92018-03-01 17:36:16 +01001879typedef struct QCow2OpenCo {
1880 BlockDriverState *bs;
1881 QDict *options;
1882 int flags;
1883 Error **errp;
1884 int ret;
1885} QCow2OpenCo;
1886
1887static void coroutine_fn qcow2_open_entry(void *opaque)
1888{
1889 QCow2OpenCo *qoc = opaque;
1890 BDRVQcow2State *s = qoc->bs->opaque;
1891
1892 qemu_co_mutex_lock(&s->lock);
Hanna Reitz06e9cd12022-04-27 13:40:55 +02001893 qoc->ret = qcow2_do_open(qoc->bs, qoc->options, qoc->flags, true,
1894 qoc->errp);
Paolo Bonzini1fafcd92018-03-01 17:36:16 +01001895 qemu_co_mutex_unlock(&s->lock);
1896}
1897
Kevin Wolf4e4bf5c2016-12-16 18:52:37 +01001898static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
1899 Error **errp)
1900{
Paolo Bonzini1fafcd92018-03-01 17:36:16 +01001901 BDRVQcow2State *s = bs->opaque;
1902 QCow2OpenCo qoc = {
1903 .bs = bs,
1904 .options = options,
1905 .flags = flags,
1906 .errp = errp,
1907 .ret = -EINPROGRESS
1908 };
Vladimir Sementsov-Ogievskiy83930782022-07-26 23:11:21 +03001909 int ret;
Paolo Bonzini1fafcd92018-03-01 17:36:16 +01001910
Vladimir Sementsov-Ogievskiy83930782022-07-26 23:11:21 +03001911 ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
1912 if (ret < 0) {
1913 return ret;
Kevin Wolf4e4bf5c2016-12-16 18:52:37 +01001914 }
1915
Paolo Bonzini1fafcd92018-03-01 17:36:16 +01001916 /* Initialise locks */
1917 qemu_co_mutex_init(&s->lock);
1918
1919 if (qemu_in_coroutine()) {
1920 /* From bdrv_co_create. */
1921 qcow2_open_entry(&qoc);
1922 } else {
Kevin Wolf4720cbe2019-01-07 13:02:48 +01001923 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
Paolo Bonzini1fafcd92018-03-01 17:36:16 +01001924 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc));
1925 BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
1926 }
1927 return qoc.ret;
Kevin Wolf4e4bf5c2016-12-16 18:52:37 +01001928}
1929
Kevin Wolf3baca892014-07-16 17:48:16 +02001930static void qcow2_refresh_limits(BlockDriverState *bs, Error **errp)
Kevin Wolfd34682c2013-12-11 19:26:16 +01001931{
Kevin Wolfff991292015-09-07 17:12:56 +02001932 BDRVQcow2State *s = bs->opaque;
Kevin Wolfd34682c2013-12-11 19:26:16 +01001933
Eric Blakea84178c2016-06-23 16:37:15 -06001934 if (bs->encrypted) {
1935 /* Encryption works on a sector granularity */
Alberto Garcia6f8f0152018-10-11 13:58:02 +03001936 bs->bl.request_alignment = qcrypto_block_get_sector_size(s->crypto);
Eric Blakea84178c2016-06-23 16:37:15 -06001937 }
Alberto Garciaa6841a22020-07-10 18:13:10 +02001938 bs->bl.pwrite_zeroes_alignment = s->subcluster_size;
Eric Blakeecdbead2016-11-17 14:13:55 -06001939 bs->bl.pdiscard_alignment = s->cluster_size;
Kevin Wolfd34682c2013-12-11 19:26:16 +01001940}
1941
Jeff Cody21d82ac2012-09-20 15:13:28 -04001942static int qcow2_reopen_prepare(BDRVReopenState *state,
1943 BlockReopenQueue *queue, Error **errp)
1944{
Kevin Wolfbcfd86d2021-07-08 13:47:04 +02001945 BDRVQcow2State *s = state->bs->opaque;
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001946 Qcow2ReopenState *r;
Kevin Wolf4c2e5f82014-04-03 13:47:50 +02001947 int ret;
1948
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001949 r = g_new0(Qcow2ReopenState, 1);
1950 state->opaque = r;
1951
1952 ret = qcow2_update_options_prepare(state->bs, r, state->options,
1953 state->flags, errp);
1954 if (ret < 0) {
1955 goto fail;
1956 }
1957
1958 /* We need to write out any unwritten data if we reopen read-only. */
Kevin Wolf4c2e5f82014-04-03 13:47:50 +02001959 if ((state->flags & BDRV_O_RDWR) == 0) {
Vladimir Sementsov-Ogievskiy169b8792017-06-28 15:05:20 +03001960 ret = qcow2_reopen_bitmaps_ro(state->bs, errp);
1961 if (ret < 0) {
1962 goto fail;
1963 }
1964
Kevin Wolf4c2e5f82014-04-03 13:47:50 +02001965 ret = bdrv_flush(state->bs);
1966 if (ret < 0) {
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001967 goto fail;
Kevin Wolf4c2e5f82014-04-03 13:47:50 +02001968 }
1969
1970 ret = qcow2_mark_clean(state->bs);
1971 if (ret < 0) {
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001972 goto fail;
Kevin Wolf4c2e5f82014-04-03 13:47:50 +02001973 }
1974 }
1975
Kevin Wolfbcfd86d2021-07-08 13:47:04 +02001976 /*
1977 * Without an external data file, s->data_file points to the same BdrvChild
1978 * as bs->file. It needs to be resynced after reopen because bs->file may
1979 * be changed. We can't use it in the meantime.
1980 */
1981 if (!has_data_file(state->bs)) {
1982 assert(s->data_file == state->bs->file);
1983 s->data_file = NULL;
1984 }
1985
Jeff Cody21d82ac2012-09-20 15:13:28 -04001986 return 0;
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001987
1988fail:
1989 qcow2_update_options_abort(state->bs, r);
1990 g_free(r);
1991 return ret;
1992}
1993
1994static void qcow2_reopen_commit(BDRVReopenState *state)
1995{
Kevin Wolfbcfd86d2021-07-08 13:47:04 +02001996 BDRVQcow2State *s = state->bs->opaque;
1997
Kevin Wolf5b0959a2015-04-16 13:42:27 +02001998 qcow2_update_options_commit(state->bs, state->opaque);
Kevin Wolfbcfd86d2021-07-08 13:47:04 +02001999 if (!s->data_file) {
2000 /*
2001 * If we don't have an external data file, s->data_file was cleared by
2002 * qcow2_reopen_prepare() and needs to be updated.
2003 */
2004 s->data_file = state->bs->file;
2005 }
Peter Krempa65eb7c82020-02-28 13:44:47 +01002006 g_free(state->opaque);
2007}
2008
2009static void qcow2_reopen_commit_post(BDRVReopenState *state)
2010{
Vladimir Sementsov-Ogievskiy4dd09f62019-09-27 15:23:55 +03002011 if (state->flags & BDRV_O_RDWR) {
2012 Error *local_err = NULL;
2013
2014 if (qcow2_reopen_bitmaps_rw(state->bs, &local_err) < 0) {
2015 /*
2016 * This is not fatal, bitmaps just left read-only, so all following
2017 * writes will fail. User can remove read-only bitmaps to unblock
2018 * writes or retry reopen.
2019 */
2020 error_reportf_err(local_err,
2021 "%s: Failed to make dirty bitmaps writable: ",
2022 bdrv_get_node_name(state->bs));
2023 }
2024 }
Kevin Wolf5b0959a2015-04-16 13:42:27 +02002025}
2026
2027static void qcow2_reopen_abort(BDRVReopenState *state)
2028{
Kevin Wolfbcfd86d2021-07-08 13:47:04 +02002029 BDRVQcow2State *s = state->bs->opaque;
2030
2031 if (!s->data_file) {
2032 /*
2033 * If we don't have an external data file, s->data_file was cleared by
2034 * qcow2_reopen_prepare() and needs to be restored.
2035 */
2036 s->data_file = state->bs->file;
2037 }
Kevin Wolf5b0959a2015-04-16 13:42:27 +02002038 qcow2_update_options_abort(state->bs, state->opaque);
2039 g_free(state->opaque);
Jeff Cody21d82ac2012-09-20 15:13:28 -04002040}
2041
Kevin Wolf5365f442015-11-16 15:34:59 +01002042static void qcow2_join_options(QDict *options, QDict *old_options)
2043{
2044 bool has_new_overlap_template =
2045 qdict_haskey(options, QCOW2_OPT_OVERLAP) ||
2046 qdict_haskey(options, QCOW2_OPT_OVERLAP_TEMPLATE);
2047 bool has_new_total_cache_size =
2048 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE);
2049 bool has_all_cache_options;
2050
2051 /* New overlap template overrides all old overlap options */
2052 if (has_new_overlap_template) {
2053 qdict_del(old_options, QCOW2_OPT_OVERLAP);
2054 qdict_del(old_options, QCOW2_OPT_OVERLAP_TEMPLATE);
2055 qdict_del(old_options, QCOW2_OPT_OVERLAP_MAIN_HEADER);
2056 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L1);
2057 qdict_del(old_options, QCOW2_OPT_OVERLAP_ACTIVE_L2);
2058 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE);
2059 qdict_del(old_options, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK);
2060 qdict_del(old_options, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE);
2061 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L1);
2062 qdict_del(old_options, QCOW2_OPT_OVERLAP_INACTIVE_L2);
2063 }
2064
2065 /* New total cache size overrides all old options */
2066 if (qdict_haskey(options, QCOW2_OPT_CACHE_SIZE)) {
2067 qdict_del(old_options, QCOW2_OPT_L2_CACHE_SIZE);
2068 qdict_del(old_options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2069 }
2070
2071 qdict_join(options, old_options, false);
2072
2073 /*
2074 * If after merging all cache size options are set, an old total size is
2075 * overwritten. Do keep all options, however, if all three are new. The
2076 * resulting error message is what we want to happen.
2077 */
2078 has_all_cache_options =
2079 qdict_haskey(options, QCOW2_OPT_CACHE_SIZE) ||
2080 qdict_haskey(options, QCOW2_OPT_L2_CACHE_SIZE) ||
2081 qdict_haskey(options, QCOW2_OPT_REFCOUNT_CACHE_SIZE);
2082
2083 if (has_all_cache_options && !has_new_total_cache_size) {
2084 qdict_del(options, QCOW2_OPT_CACHE_SIZE);
2085 }
2086}
2087
Eric Blakea320fb02018-02-13 14:26:52 -06002088static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
2089 bool want_zero,
2090 int64_t offset, int64_t count,
2091 int64_t *pnum, int64_t *map,
2092 BlockDriverState **file)
bellard585f8582006-08-05 21:14:20 +00002093{
Kevin Wolfff991292015-09-07 17:12:56 +02002094 BDRVQcow2State *s = bs->opaque;
Alberto Garcia388e5812020-07-10 18:12:44 +02002095 uint64_t host_offset;
Kevin Wolfecfe1862016-05-31 16:13:07 +02002096 unsigned int bytes;
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002097 QCow2SubclusterType type;
Alberto Garcia74e60fb2019-12-12 12:01:21 +02002098 int ret, status = 0;
bellard585f8582006-08-05 21:14:20 +00002099
Kevin Wolf5e978552019-10-24 16:26:58 +02002100 qemu_co_mutex_lock(&s->lock);
2101
Vladimir Sementsov-Ogievskiy69f47502019-04-08 19:26:17 +03002102 if (!s->metadata_preallocation_checked) {
2103 ret = qcow2_detect_metadata_preallocation(bs);
2104 s->metadata_preallocation = (ret == 1);
2105 s->metadata_preallocation_checked = true;
2106 }
2107
Eric Blakea320fb02018-02-13 14:26:52 -06002108 bytes = MIN(INT_MAX, count);
Alberto Garciaca4a0bb2020-07-10 18:12:59 +02002109 ret = qcow2_get_host_offset(bs, offset, &bytes, &host_offset, &type);
Stefan Hajnoczif8a2e5e2011-11-14 12:44:21 +00002110 qemu_co_mutex_unlock(&s->lock);
Kevin Wolf1c46efa2010-05-21 17:59:36 +02002111 if (ret < 0) {
Paolo Bonzinid6636402013-09-04 19:00:25 +02002112 return ret;
Kevin Wolf1c46efa2010-05-21 17:59:36 +02002113 }
aliguori095a9c52008-08-14 18:10:28 +00002114
Eric Blakea320fb02018-02-13 14:26:52 -06002115 *pnum = bytes;
Kevin Wolfecfe1862016-05-31 16:13:07 +02002116
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002117 if ((type == QCOW2_SUBCLUSTER_NORMAL ||
Alberto Garcia97490a12020-07-10 18:13:01 +02002118 type == QCOW2_SUBCLUSTER_ZERO_ALLOC ||
2119 type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) && !s->crypto) {
Alberto Garcia388e5812020-07-10 18:12:44 +02002120 *map = host_offset;
Kevin Wolf37be1402019-02-27 13:22:56 +01002121 *file = s->data_file->bs;
Eric Blakea320fb02018-02-13 14:26:52 -06002122 status |= BDRV_BLOCK_OFFSET_VALID;
Paolo Bonzini4bc74be2013-09-04 19:00:30 +02002123 }
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002124 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN ||
2125 type == QCOW2_SUBCLUSTER_ZERO_ALLOC) {
Paolo Bonzini4bc74be2013-09-04 19:00:30 +02002126 status |= BDRV_BLOCK_ZERO;
Alberto Garcia97490a12020-07-10 18:13:01 +02002127 } else if (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN &&
2128 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC) {
Paolo Bonzini4bc74be2013-09-04 19:00:30 +02002129 status |= BDRV_BLOCK_DATA;
2130 }
Vladimir Sementsov-Ogievskiy69f47502019-04-08 19:26:17 +03002131 if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) &&
2132 (status & BDRV_BLOCK_OFFSET_VALID))
2133 {
2134 status |= BDRV_BLOCK_RECURSE;
2135 }
Paolo Bonzini4bc74be2013-09-04 19:00:30 +02002136 return status;
bellard585f8582006-08-05 21:14:20 +00002137}
2138
Fam Zhengfd9fcd32018-06-01 17:26:42 +08002139static coroutine_fn int qcow2_handle_l2meta(BlockDriverState *bs,
2140 QCowL2Meta **pl2meta,
2141 bool link_l2)
2142{
2143 int ret = 0;
2144 QCowL2Meta *l2meta = *pl2meta;
2145
2146 while (l2meta != NULL) {
2147 QCowL2Meta *next;
2148
Fam Zheng354d9302018-06-27 11:57:51 +08002149 if (link_l2) {
Fam Zhengfd9fcd32018-06-01 17:26:42 +08002150 ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
2151 if (ret) {
2152 goto out;
2153 }
Kevin Wolf8b24cd12018-06-28 17:05:45 +02002154 } else {
2155 qcow2_alloc_cluster_abort(bs, l2meta);
Fam Zhengfd9fcd32018-06-01 17:26:42 +08002156 }
2157
2158 /* Take the request off the list of running requests */
Alberto Garciaf7bd5bb2020-09-03 18:37:48 +02002159 QLIST_REMOVE(l2meta, next_in_flight);
Fam Zhengfd9fcd32018-06-01 17:26:42 +08002160
2161 qemu_co_queue_restart_all(&l2meta->dependent_requests);
2162
2163 next = l2meta->next;
2164 g_free(l2meta);
2165 l2meta = next;
2166 }
2167out:
2168 *pl2meta = l2meta;
2169 return ret;
2170}
2171
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002172static coroutine_fn int
2173qcow2_co_preadv_encrypted(BlockDriverState *bs,
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002174 uint64_t host_offset,
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002175 uint64_t offset,
2176 uint64_t bytes,
2177 QEMUIOVector *qiov,
2178 uint64_t qiov_offset)
2179{
2180 int ret;
2181 BDRVQcow2State *s = bs->opaque;
2182 uint8_t *buf;
2183
2184 assert(bs->encrypted && s->crypto);
2185 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2186
2187 /*
2188 * For encrypted images, read everything into a temporary
2189 * contiguous buffer on which the AES functions can work.
2190 * Also, decryption in a separate buffer is better as it
2191 * prevents the guest from learning information about the
2192 * encrypted nature of the virtual disk.
2193 */
2194
2195 buf = qemu_try_blockalign(s->data_file->bs, bytes);
2196 if (buf == NULL) {
2197 return -ENOMEM;
2198 }
2199
2200 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002201 ret = bdrv_co_pread(s->data_file, host_offset, bytes, buf, 0);
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002202 if (ret < 0) {
2203 goto fail;
2204 }
2205
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002206 if (qcow2_co_decrypt(bs, host_offset, offset, buf, bytes) < 0)
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002207 {
2208 ret = -EIO;
2209 goto fail;
2210 }
2211 qemu_iovec_from_buf(qiov, qiov_offset, buf, bytes);
2212
2213fail:
2214 qemu_vfree(buf);
2215
2216 return ret;
2217}
2218
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002219typedef struct Qcow2AioTask {
2220 AioTask task;
2221
2222 BlockDriverState *bs;
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002223 QCow2SubclusterType subcluster_type; /* only for read */
Vladimir Sementsov-Ogievskiy9a3978a2021-09-14 15:24:46 +03002224 uint64_t host_offset; /* or l2_entry for compressed read */
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002225 uint64_t offset;
2226 uint64_t bytes;
2227 QEMUIOVector *qiov;
2228 uint64_t qiov_offset;
2229 QCowL2Meta *l2meta; /* only for write */
2230} Qcow2AioTask;
2231
2232static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task);
2233static coroutine_fn int qcow2_add_task(BlockDriverState *bs,
2234 AioTaskPool *pool,
2235 AioTaskFunc func,
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002236 QCow2SubclusterType subcluster_type,
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002237 uint64_t host_offset,
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002238 uint64_t offset,
2239 uint64_t bytes,
2240 QEMUIOVector *qiov,
2241 size_t qiov_offset,
2242 QCowL2Meta *l2meta)
2243{
2244 Qcow2AioTask local_task;
2245 Qcow2AioTask *task = pool ? g_new(Qcow2AioTask, 1) : &local_task;
2246
2247 *task = (Qcow2AioTask) {
2248 .task.func = func,
2249 .bs = bs,
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002250 .subcluster_type = subcluster_type,
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002251 .qiov = qiov,
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002252 .host_offset = host_offset,
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002253 .offset = offset,
2254 .bytes = bytes,
2255 .qiov_offset = qiov_offset,
2256 .l2meta = l2meta,
2257 };
2258
2259 trace_qcow2_add_task(qemu_coroutine_self(), bs, pool,
2260 func == qcow2_co_preadv_task_entry ? "read" : "write",
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002261 subcluster_type, host_offset, offset, bytes,
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002262 qiov, qiov_offset);
2263
2264 if (!pool) {
2265 return func(&task->task);
2266 }
2267
2268 aio_task_pool_start_task(pool, &task->task);
2269
2270 return 0;
2271}
2272
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002273static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs,
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002274 QCow2SubclusterType subc_type,
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002275 uint64_t host_offset,
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002276 uint64_t offset, uint64_t bytes,
2277 QEMUIOVector *qiov,
2278 size_t qiov_offset)
2279{
2280 BDRVQcow2State *s = bs->opaque;
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002281
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002282 switch (subc_type) {
2283 case QCOW2_SUBCLUSTER_ZERO_PLAIN:
2284 case QCOW2_SUBCLUSTER_ZERO_ALLOC:
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002285 /* Both zero types are handled in qcow2_co_preadv_part */
2286 g_assert_not_reached();
2287
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002288 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
Alberto Garcia97490a12020-07-10 18:13:01 +02002289 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002290 assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */
2291
2292 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
2293 return bdrv_co_preadv_part(bs->backing, offset, bytes,
2294 qiov, qiov_offset, 0);
2295
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002296 case QCOW2_SUBCLUSTER_COMPRESSED:
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002297 return qcow2_co_preadv_compressed(bs, host_offset,
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002298 offset, bytes, qiov, qiov_offset);
2299
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002300 case QCOW2_SUBCLUSTER_NORMAL:
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002301 if (bs->encrypted) {
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002302 return qcow2_co_preadv_encrypted(bs, host_offset,
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002303 offset, bytes, qiov, qiov_offset);
2304 }
2305
2306 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002307 return bdrv_co_preadv_part(s->data_file, host_offset,
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002308 bytes, qiov, qiov_offset, 0);
2309
2310 default:
2311 g_assert_not_reached();
2312 }
2313
2314 g_assert_not_reached();
2315}
2316
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002317static coroutine_fn int qcow2_co_preadv_task_entry(AioTask *task)
2318{
2319 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2320
2321 assert(!t->l2meta);
2322
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002323 return qcow2_co_preadv_task(t->bs, t->subcluster_type,
2324 t->host_offset, t->offset, t->bytes,
2325 t->qiov, t->qiov_offset);
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002326}
2327
Vladimir Sementsov-Ogievskiydf893d22019-06-04 19:15:13 +03002328static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs,
Vladimir Sementsov-Ogievskiyf7ef38d2021-09-03 13:27:59 +03002329 int64_t offset, int64_t bytes,
Vladimir Sementsov-Ogievskiydf893d22019-06-04 19:15:13 +03002330 QEMUIOVector *qiov,
Vladimir Sementsov-Ogievskiyf7ef38d2021-09-03 13:27:59 +03002331 size_t qiov_offset,
2332 BdrvRequestFlags flags)
aliguori14907912008-10-31 17:28:00 +00002333{
Kevin Wolfff991292015-09-07 17:12:56 +02002334 BDRVQcow2State *s = bs->opaque;
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002335 int ret = 0;
Kevin Wolfecfe1862016-05-31 16:13:07 +02002336 unsigned int cur_bytes; /* number of bytes in current iteration */
Alberto Garcia388e5812020-07-10 18:12:44 +02002337 uint64_t host_offset = 0;
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002338 QCow2SubclusterType type;
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002339 AioTaskPool *aio = NULL;
bellard585f8582006-08-05 21:14:20 +00002340
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002341 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
Frediano Ziglio5ebaa272011-08-23 15:21:18 +02002342 /* prepare next request */
Kevin Wolfecfe1862016-05-31 16:13:07 +02002343 cur_bytes = MIN(bytes, INT_MAX);
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01002344 if (s->crypto) {
Kevin Wolfecfe1862016-05-31 16:13:07 +02002345 cur_bytes = MIN(cur_bytes,
2346 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
Kevin Wolfbd28f832010-09-13 18:08:52 +02002347 }
2348
Vladimir Sementsov-Ogievskiyf24196d2019-05-06 17:27:39 +03002349 qemu_co_mutex_lock(&s->lock);
Alberto Garciaca4a0bb2020-07-10 18:12:59 +02002350 ret = qcow2_get_host_offset(bs, offset, &cur_bytes,
2351 &host_offset, &type);
Vladimir Sementsov-Ogievskiyf24196d2019-05-06 17:27:39 +03002352 qemu_co_mutex_unlock(&s->lock);
Kevin Wolf68d100e2011-06-30 17:42:09 +02002353 if (ret < 0) {
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002354 goto out;
Kevin Wolf171e3d62010-04-06 15:30:09 +02002355 }
Frediano Ziglio5ebaa272011-08-23 15:21:18 +02002356
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002357 if (type == QCOW2_SUBCLUSTER_ZERO_PLAIN ||
2358 type == QCOW2_SUBCLUSTER_ZERO_ALLOC ||
Alberto Garcia97490a12020-07-10 18:13:01 +02002359 (type == QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN && !bs->backing) ||
2360 (type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC && !bs->backing))
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002361 {
Vladimir Sementsov-Ogievskiydf893d22019-06-04 19:15:13 +03002362 qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes);
Vladimir Sementsov-Ogievskiy88f468e2019-09-16 20:53:22 +03002363 } else {
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002364 if (!aio && cur_bytes != bytes) {
2365 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2366 }
Alberto Garciaca4a0bb2020-07-10 18:12:59 +02002367 ret = qcow2_add_task(bs, aio, qcow2_co_preadv_task_entry, type,
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002368 host_offset, offset, cur_bytes,
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002369 qiov, qiov_offset, NULL);
Frediano Ziglio5ebaa272011-08-23 15:21:18 +02002370 if (ret < 0) {
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002371 goto out;
Frediano Ziglio5ebaa272011-08-23 15:21:18 +02002372 }
Frediano Zigliofaf575c2011-08-23 15:21:14 +02002373 }
Frediano Ziglio5ebaa272011-08-23 15:21:18 +02002374
Kevin Wolfecfe1862016-05-31 16:13:07 +02002375 bytes -= cur_bytes;
2376 offset += cur_bytes;
Vladimir Sementsov-Ogievskiydf893d22019-06-04 19:15:13 +03002377 qiov_offset += cur_bytes;
bellard585f8582006-08-05 21:14:20 +00002378 }
aliguorif141eaf2009-04-07 18:43:24 +00002379
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002380out:
2381 if (aio) {
2382 aio_task_pool_wait_all(aio);
2383 if (ret == 0) {
2384 ret = aio_task_pool_status(aio);
2385 }
2386 g_free(aio);
2387 }
2388
2389 return ret;
bellard585f8582006-08-05 21:14:20 +00002390}
2391
Alberto Garciaee22a9d2017-06-19 16:40:08 +03002392/* Check if it's possible to merge a write request with the writing of
2393 * the data from the COW regions */
2394static bool merge_cow(uint64_t offset, unsigned bytes,
Vladimir Sementsov-Ogievskiy53962342019-06-04 19:15:14 +03002395 QEMUIOVector *qiov, size_t qiov_offset,
2396 QCowL2Meta *l2meta)
Alberto Garciaee22a9d2017-06-19 16:40:08 +03002397{
2398 QCowL2Meta *m;
2399
2400 for (m = l2meta; m != NULL; m = m->next) {
2401 /* If both COW regions are empty then there's nothing to merge */
2402 if (m->cow_start.nb_bytes == 0 && m->cow_end.nb_bytes == 0) {
2403 continue;
2404 }
2405
Anton Nefedovc8bb23c2019-05-16 17:27:49 +03002406 /* If COW regions are handled already, skip this too */
2407 if (m->skip_cow) {
2408 continue;
2409 }
2410
Alberto Garcia3441ad42020-10-07 18:13:23 +02002411 /*
2412 * The write request should start immediately after the first
2413 * COW region. This does not always happen because the area
2414 * touched by the request can be larger than the one defined
2415 * by @m (a single request can span an area consisting of a
2416 * mix of previously unallocated and allocated clusters, that
2417 * is why @l2meta is a list).
2418 */
Alberto Garciaee22a9d2017-06-19 16:40:08 +03002419 if (l2meta_cow_start(m) + m->cow_start.nb_bytes != offset) {
Alberto Garcia3441ad42020-10-07 18:13:23 +02002420 /* In this case the request starts before this region */
2421 assert(offset < l2meta_cow_start(m));
2422 assert(m->cow_start.nb_bytes == 0);
Alberto Garciaee22a9d2017-06-19 16:40:08 +03002423 continue;
2424 }
2425
Alberto Garcia3441ad42020-10-07 18:13:23 +02002426 /* The write request should end immediately before the second
2427 * COW region (see above for why it does not always happen) */
Alberto Garciaee22a9d2017-06-19 16:40:08 +03002428 if (m->offset + m->cow_end.offset != offset + bytes) {
Alberto Garcia3441ad42020-10-07 18:13:23 +02002429 assert(offset + bytes > m->offset + m->cow_end.offset);
2430 assert(m->cow_end.nb_bytes == 0);
Alberto Garciaee22a9d2017-06-19 16:40:08 +03002431 continue;
2432 }
2433
2434 /* Make sure that adding both COW regions to the QEMUIOVector
2435 * does not exceed IOV_MAX */
Vladimir Sementsov-Ogievskiy53962342019-06-04 19:15:14 +03002436 if (qemu_iovec_subvec_niov(qiov, qiov_offset, bytes) > IOV_MAX - 2) {
Alberto Garciaee22a9d2017-06-19 16:40:08 +03002437 continue;
2438 }
2439
Vladimir Sementsov-Ogievskiy53962342019-06-04 19:15:14 +03002440 m->data_qiov = qiov;
2441 m->data_qiov_offset = qiov_offset;
Alberto Garciaee22a9d2017-06-19 16:40:08 +03002442 return true;
2443 }
2444
2445 return false;
2446}
2447
Alberto Garcia46cd1e82020-10-26 17:58:27 +01002448/*
2449 * Return 1 if the COW regions read as zeroes, 0 if not, < 0 on error.
2450 * Note that returning 0 does not guarantee non-zero data.
2451 */
Paolo Bonzini050ed2e2022-09-22 10:49:12 +02002452static int coroutine_fn is_zero_cow(BlockDriverState *bs, QCowL2Meta *m)
Anton Nefedovc8bb23c2019-05-16 17:27:49 +03002453{
2454 /*
2455 * This check is designed for optimization shortcut so it must be
2456 * efficient.
Alberto Garcia46cd1e82020-10-26 17:58:27 +01002457 * Instead of is_zero(), use bdrv_co_is_zero_fast() as it is
2458 * faster (but not as accurate and can result in false negatives).
Anton Nefedovc8bb23c2019-05-16 17:27:49 +03002459 */
Alberto Garcia46cd1e82020-10-26 17:58:27 +01002460 int ret = bdrv_co_is_zero_fast(bs, m->offset + m->cow_start.offset,
2461 m->cow_start.nb_bytes);
2462 if (ret <= 0) {
2463 return ret;
2464 }
2465
2466 return bdrv_co_is_zero_fast(bs, m->offset + m->cow_end.offset,
2467 m->cow_end.nb_bytes);
Anton Nefedovc8bb23c2019-05-16 17:27:49 +03002468}
2469
Paolo Bonzini050ed2e2022-09-22 10:49:12 +02002470static int coroutine_fn handle_alloc_space(BlockDriverState *bs,
2471 QCowL2Meta *l2meta)
Anton Nefedovc8bb23c2019-05-16 17:27:49 +03002472{
2473 BDRVQcow2State *s = bs->opaque;
2474 QCowL2Meta *m;
2475
2476 if (!(s->data_file->bs->supported_zero_flags & BDRV_REQ_NO_FALLBACK)) {
2477 return 0;
2478 }
2479
2480 if (bs->encrypted) {
2481 return 0;
2482 }
2483
2484 for (m = l2meta; m != NULL; m = m->next) {
2485 int ret;
Alberto Garciabf4a66e2020-07-10 18:13:09 +02002486 uint64_t start_offset = m->alloc_offset + m->cow_start.offset;
2487 unsigned nb_bytes = m->cow_end.offset + m->cow_end.nb_bytes -
2488 m->cow_start.offset;
Anton Nefedovc8bb23c2019-05-16 17:27:49 +03002489
2490 if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
2491 continue;
2492 }
2493
Alberto Garcia46cd1e82020-10-26 17:58:27 +01002494 ret = is_zero_cow(bs, m);
2495 if (ret < 0) {
2496 return ret;
2497 } else if (ret == 0) {
Anton Nefedovc8bb23c2019-05-16 17:27:49 +03002498 continue;
2499 }
2500
2501 /*
2502 * instead of writing zero COW buffers,
2503 * efficiently zero out the whole clusters
2504 */
2505
Alberto Garciabf4a66e2020-07-10 18:13:09 +02002506 ret = qcow2_pre_write_overlap_check(bs, 0, start_offset, nb_bytes,
Anton Nefedovc8bb23c2019-05-16 17:27:49 +03002507 true);
2508 if (ret < 0) {
2509 return ret;
2510 }
2511
2512 BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE);
Alberto Garciabf4a66e2020-07-10 18:13:09 +02002513 ret = bdrv_co_pwrite_zeroes(s->data_file, start_offset, nb_bytes,
Anton Nefedovc8bb23c2019-05-16 17:27:49 +03002514 BDRV_REQ_NO_FALLBACK);
2515 if (ret < 0) {
2516 if (ret != -ENOTSUP && ret != -EAGAIN) {
2517 return ret;
2518 }
2519 continue;
2520 }
2521
2522 trace_qcow2_skip_cow(qemu_coroutine_self(), m->offset, m->nb_clusters);
2523 m->skip_cow = true;
2524 }
2525 return 0;
2526}
2527
Vladimir Sementsov-Ogievskiy6aa7a262019-09-16 20:53:23 +03002528/*
2529 * qcow2_co_pwritev_task
2530 * Called with s->lock unlocked
2531 * l2meta - if not NULL, qcow2_co_pwritev_task() will consume it. Caller must
2532 * not use it somehow after qcow2_co_pwritev_task() call
2533 */
2534static coroutine_fn int qcow2_co_pwritev_task(BlockDriverState *bs,
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002535 uint64_t host_offset,
Vladimir Sementsov-Ogievskiy6aa7a262019-09-16 20:53:23 +03002536 uint64_t offset, uint64_t bytes,
2537 QEMUIOVector *qiov,
2538 uint64_t qiov_offset,
2539 QCowL2Meta *l2meta)
2540{
2541 int ret;
2542 BDRVQcow2State *s = bs->opaque;
2543 void *crypt_buf = NULL;
Vladimir Sementsov-Ogievskiy6aa7a262019-09-16 20:53:23 +03002544 QEMUIOVector encrypted_qiov;
2545
2546 if (bs->encrypted) {
2547 assert(s->crypto);
2548 assert(bytes <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
2549 crypt_buf = qemu_try_blockalign(bs->file->bs, bytes);
2550 if (crypt_buf == NULL) {
2551 ret = -ENOMEM;
2552 goto out_unlocked;
2553 }
2554 qemu_iovec_to_buf(qiov, qiov_offset, crypt_buf, bytes);
2555
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002556 if (qcow2_co_encrypt(bs, host_offset, offset, crypt_buf, bytes) < 0) {
Vladimir Sementsov-Ogievskiy6aa7a262019-09-16 20:53:23 +03002557 ret = -EIO;
2558 goto out_unlocked;
2559 }
2560
2561 qemu_iovec_init_buf(&encrypted_qiov, crypt_buf, bytes);
2562 qiov = &encrypted_qiov;
2563 qiov_offset = 0;
2564 }
2565
2566 /* Try to efficiently initialize the physical space with zeroes */
2567 ret = handle_alloc_space(bs, l2meta);
2568 if (ret < 0) {
2569 goto out_unlocked;
2570 }
2571
2572 /*
2573 * If we need to do COW, check if it's possible to merge the
2574 * writing of the guest data together with that of the COW regions.
2575 * If it's not possible (or not necessary) then write the
2576 * guest data now.
2577 */
2578 if (!merge_cow(offset, bytes, qiov, qiov_offset, l2meta)) {
2579 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002580 trace_qcow2_writev_data(qemu_coroutine_self(), host_offset);
2581 ret = bdrv_co_pwritev_part(s->data_file, host_offset,
Vladimir Sementsov-Ogievskiy6aa7a262019-09-16 20:53:23 +03002582 bytes, qiov, qiov_offset, 0);
2583 if (ret < 0) {
2584 goto out_unlocked;
2585 }
2586 }
2587
2588 qemu_co_mutex_lock(&s->lock);
2589
2590 ret = qcow2_handle_l2meta(bs, &l2meta, true);
2591 goto out_locked;
2592
2593out_unlocked:
2594 qemu_co_mutex_lock(&s->lock);
2595
2596out_locked:
2597 qcow2_handle_l2meta(bs, &l2meta, false);
2598 qemu_co_mutex_unlock(&s->lock);
2599
2600 qemu_vfree(crypt_buf);
2601
2602 return ret;
2603}
2604
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002605static coroutine_fn int qcow2_co_pwritev_task_entry(AioTask *task)
2606{
2607 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
2608
Alberto Garcia10dabdc52020-07-10 18:13:00 +02002609 assert(!t->subcluster_type);
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002610
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002611 return qcow2_co_pwritev_task(t->bs, t->host_offset,
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002612 t->offset, t->bytes, t->qiov, t->qiov_offset,
2613 t->l2meta);
2614}
2615
Vladimir Sementsov-Ogievskiy53962342019-06-04 19:15:14 +03002616static coroutine_fn int qcow2_co_pwritev_part(
Vladimir Sementsov-Ogievskiye75abed2021-09-03 13:28:00 +03002617 BlockDriverState *bs, int64_t offset, int64_t bytes,
2618 QEMUIOVector *qiov, size_t qiov_offset, BdrvRequestFlags flags)
bellard585f8582006-08-05 21:14:20 +00002619{
Kevin Wolfff991292015-09-07 17:12:56 +02002620 BDRVQcow2State *s = bs->opaque;
Kevin Wolfd46a0bb2016-06-01 16:55:05 +02002621 int offset_in_cluster;
Kevin Wolf68d100e2011-06-30 17:42:09 +02002622 int ret;
Kevin Wolfd46a0bb2016-06-01 16:55:05 +02002623 unsigned int cur_bytes; /* number of sectors in current iteration */
Alberto Garciabfd09892020-09-11 16:09:42 +02002624 uint64_t host_offset;
Kevin Wolf8d2497c2013-01-14 17:31:31 +01002625 QCowL2Meta *l2meta = NULL;
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002626 AioTaskPool *aio = NULL;
Frediano Ziglioc2271402011-08-23 15:21:15 +02002627
Kevin Wolfd46a0bb2016-06-01 16:55:05 +02002628 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset, bytes);
Kevin Wolf3cce16f2012-03-01 18:36:21 +01002629
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002630 while (bytes != 0 && aio_task_pool_status(aio) == 0) {
Frediano Ziglio3fc48d02011-08-23 15:21:19 +02002631
Kevin Wolff50f88b2012-12-07 18:08:46 +01002632 l2meta = NULL;
Kevin Wolfcf5c1a22012-12-07 18:08:44 +01002633
Kevin Wolf3cce16f2012-03-01 18:36:21 +01002634 trace_qcow2_writev_start_part(qemu_coroutine_self());
Kevin Wolfd46a0bb2016-06-01 16:55:05 +02002635 offset_in_cluster = offset_into_cluster(s, offset);
2636 cur_bytes = MIN(bytes, INT_MAX);
2637 if (bs->encrypted) {
2638 cur_bytes = MIN(cur_bytes,
2639 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
2640 - offset_in_cluster);
bellard585f8582006-08-05 21:14:20 +00002641 }
Kevin Wolf6f5f0602010-09-13 18:24:10 +02002642
Vladimir Sementsov-Ogievskiy6aa7a262019-09-16 20:53:23 +03002643 qemu_co_mutex_lock(&s->lock);
2644
Alberto Garciabfd09892020-09-11 16:09:42 +02002645 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes,
2646 &host_offset, &l2meta);
Frediano Ziglio5ebaa272011-08-23 15:21:18 +02002647 if (ret < 0) {
Vladimir Sementsov-Ogievskiy5447c3a2019-05-06 17:27:40 +03002648 goto out_locked;
Frediano Ziglio5ebaa272011-08-23 15:21:18 +02002649 }
Kevin Wolf6f5f0602010-09-13 18:24:10 +02002650
Alberto Garciabfd09892020-09-11 16:09:42 +02002651 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset,
Vladimir Sementsov-Ogievskiy5447c3a2019-05-06 17:27:40 +03002652 cur_bytes, true);
2653 if (ret < 0) {
2654 goto out_locked;
2655 }
2656
2657 qemu_co_mutex_unlock(&s->lock);
2658
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002659 if (!aio && cur_bytes != bytes) {
2660 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
2661 }
2662 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_task_entry, 0,
Alberto Garciabfd09892020-09-11 16:09:42 +02002663 host_offset, offset,
Alberto Garcia9c4269d2020-07-10 18:12:43 +02002664 cur_bytes, qiov, qiov_offset, l2meta);
Vladimir Sementsov-Ogievskiy6aa7a262019-09-16 20:53:23 +03002665 l2meta = NULL; /* l2meta is consumed by qcow2_co_pwritev_task() */
Anton Nefedovc8bb23c2019-05-16 17:27:49 +03002666 if (ret < 0) {
Vladimir Sementsov-Ogievskiy6aa7a262019-09-16 20:53:23 +03002667 goto fail_nometa;
Kevin Wolff50f88b2012-12-07 18:08:46 +01002668 }
Kevin Wolf0fa91312011-09-01 15:02:13 +02002669
Kevin Wolfd46a0bb2016-06-01 16:55:05 +02002670 bytes -= cur_bytes;
2671 offset += cur_bytes;
Vladimir Sementsov-Ogievskiy6aa7a262019-09-16 20:53:23 +03002672 qiov_offset += cur_bytes;
Kevin Wolfd46a0bb2016-06-01 16:55:05 +02002673 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes);
bellard585f8582006-08-05 21:14:20 +00002674 }
Frediano Ziglio3fc48d02011-08-23 15:21:19 +02002675 ret = 0;
Kevin Wolf6f5f0602010-09-13 18:24:10 +02002676
Vladimir Sementsov-Ogievskiy5447c3a2019-05-06 17:27:40 +03002677 qemu_co_mutex_lock(&s->lock);
2678
2679out_locked:
Fam Zhengfd9fcd32018-06-01 17:26:42 +08002680 qcow2_handle_l2meta(bs, &l2meta, false);
Kevin Wolf0fa91312011-09-01 15:02:13 +02002681
Paolo Bonzinia8c57402017-06-29 15:27:39 +02002682 qemu_co_mutex_unlock(&s->lock);
2683
Vladimir Sementsov-Ogievskiy6aa7a262019-09-16 20:53:23 +03002684fail_nometa:
Vladimir Sementsov-Ogievskiyd710cf52019-09-16 20:53:24 +03002685 if (aio) {
2686 aio_task_pool_wait_all(aio);
2687 if (ret == 0) {
2688 ret = aio_task_pool_status(aio);
2689 }
2690 g_free(aio);
2691 }
2692
Kevin Wolf3cce16f2012-03-01 18:36:21 +01002693 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
Kevin Wolf42496d62011-06-07 15:04:32 +02002694
Kevin Wolf68d100e2011-06-30 17:42:09 +02002695 return ret;
bellard585f8582006-08-05 21:14:20 +00002696}
2697
Kevin Wolfec6d8912015-12-22 16:04:57 +01002698static int qcow2_inactivate(BlockDriverState *bs)
2699{
2700 BDRVQcow2State *s = bs->opaque;
2701 int ret, result = 0;
Vladimir Sementsov-Ogievskiy5f728262017-06-28 15:05:19 +03002702 Error *local_err = NULL;
Kevin Wolfec6d8912015-12-22 16:04:57 +01002703
Vladimir Sementsov-Ogievskiy644ddbb2019-09-27 15:23:52 +03002704 qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err);
Pavel Butsykin83a8c772017-09-04 13:18:00 +03002705 if (local_err != NULL) {
2706 result = -EINVAL;
Vladimir Sementsov-Ogievskiy132adb62018-10-29 16:23:15 -04002707 error_reportf_err(local_err, "Lost persistent bitmaps during "
2708 "inactivation of node '%s': ",
2709 bdrv_get_device_or_node_name(bs));
Pavel Butsykin83a8c772017-09-04 13:18:00 +03002710 }
2711
Kevin Wolfec6d8912015-12-22 16:04:57 +01002712 ret = qcow2_cache_flush(bs, s->l2_table_cache);
2713 if (ret) {
2714 result = ret;
2715 error_report("Failed to flush the L2 table cache: %s",
2716 strerror(-ret));
2717 }
2718
2719 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
2720 if (ret) {
2721 result = ret;
2722 error_report("Failed to flush the refcount block cache: %s",
2723 strerror(-ret));
2724 }
2725
2726 if (result == 0) {
2727 qcow2_mark_clean(bs);
2728 }
2729
2730 return result;
2731}
2732
Hanna Reitz06e9cd12022-04-27 13:40:55 +02002733static void qcow2_do_close(BlockDriverState *bs, bool close_data_file)
bellard585f8582006-08-05 21:14:20 +00002734{
Kevin Wolfff991292015-09-07 17:12:56 +02002735 BDRVQcow2State *s = bs->opaque;
Kevin Wolfde828152014-05-20 17:12:47 +02002736 qemu_vfree(s->l1_table);
Max Reitzcf939802013-08-30 14:34:26 +02002737 /* else pre-write overlap checks in cache_destroy may crash */
2738 s->l1_table = NULL;
Kevin Wolf29c1a732011-01-10 17:17:28 +01002739
Kevin Wolf140fd5a2015-12-22 16:10:32 +01002740 if (!(s->flags & BDRV_O_INACTIVE)) {
Kevin Wolfec6d8912015-12-22 16:04:57 +01002741 qcow2_inactivate(bs);
Kevin Wolf27eb6c02014-03-11 15:15:03 +01002742 }
Stefan Hajnoczic61d0002012-07-27 09:05:19 +01002743
Alberto Garcia279621c2015-08-04 15:14:40 +03002744 cache_clean_timer_del(bs);
Alberto Garciae64d4072018-02-05 16:33:08 +02002745 qcow2_cache_destroy(s->l2_table_cache);
2746 qcow2_cache_destroy(s->refcount_block_cache);
Kevin Wolf29c1a732011-01-10 17:17:28 +01002747
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01002748 qcrypto_block_free(s->crypto);
2749 s->crypto = NULL;
Pan Nengyuan4aebf0f2020-02-27 09:29:49 +08002750 qapi_free_QCryptoBlockOpenOptions(s->crypto_opts);
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +01002751
Kevin Wolf6744cba2011-12-15 12:20:58 +01002752 g_free(s->unknown_header_fields);
Kevin Wolf75bab852012-02-02 14:52:08 +01002753 cleanup_unknown_header_ext(bs);
Kevin Wolf6744cba2011-12-15 12:20:58 +01002754
Kevin Wolf9b890bd2019-01-15 19:02:40 +01002755 g_free(s->image_data_file);
Kevin Wolfe4603fe2015-04-07 15:03:16 +02002756 g_free(s->image_backing_file);
2757 g_free(s->image_backing_format);
2758
Hanna Reitz06e9cd12022-04-27 13:40:55 +02002759 if (close_data_file && has_data_file(bs)) {
Kevin Wolf0e8c08b2019-01-29 17:13:57 +01002760 bdrv_unref_child(bs, s->data_file);
Vladimir Sementsov-Ogievskiy808cf3c2020-03-16 09:06:31 +03002761 s->data_file = NULL;
Kevin Wolf0e8c08b2019-01-29 17:13:57 +01002762 }
2763
Kevin Wolfed6ccf02009-05-28 16:07:07 +02002764 qcow2_refcount_close(bs);
Li Zhi Hui28c12022011-12-07 17:25:48 +08002765 qcow2_free_snapshots(bs);
bellard585f8582006-08-05 21:14:20 +00002766}
2767
Hanna Reitz06e9cd12022-04-27 13:40:55 +02002768static void qcow2_close(BlockDriverState *bs)
2769{
2770 qcow2_do_close(bs, true);
2771}
2772
Paolo Bonzini2b148f32018-03-01 17:36:18 +01002773static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
2774 Error **errp)
Anthony Liguori06d92602011-11-14 15:09:46 -06002775{
Vladimir Sementsov-Ogievskiye6247c92021-02-02 15:49:54 +03002776 ERRP_GUARD();
Kevin Wolfff991292015-09-07 17:12:56 +02002777 BDRVQcow2State *s = bs->opaque;
Hanna Reitz06e9cd12022-04-27 13:40:55 +02002778 BdrvChild *data_file;
Anthony Liguori06d92602011-11-14 15:09:46 -06002779 int flags = s->flags;
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01002780 QCryptoBlock *crypto = NULL;
Kevin Wolfacdfb482013-03-18 13:08:10 +01002781 QDict *options;
Kevin Wolf5a8a30d2014-03-12 15:59:16 +01002782 int ret;
Anthony Liguori06d92602011-11-14 15:09:46 -06002783
2784 /*
2785 * Backing files are read-only which makes all of their metadata immutable,
2786 * that means we don't have to worry about reopening them here.
2787 */
2788
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01002789 crypto = s->crypto;
2790 s->crypto = NULL;
Anthony Liguori06d92602011-11-14 15:09:46 -06002791
Hanna Reitz06e9cd12022-04-27 13:40:55 +02002792 /*
2793 * Do not reopen s->data_file (i.e., have qcow2_do_close() not close it,
2794 * and then prevent qcow2_do_open() from opening it), because this function
2795 * runs in the I/O path and as such we must not invoke global-state
2796 * functions like bdrv_unref_child() and bdrv_open_child().
2797 */
Anthony Liguori06d92602011-11-14 15:09:46 -06002798
Hanna Reitz06e9cd12022-04-27 13:40:55 +02002799 qcow2_do_close(bs, false);
2800
2801 data_file = s->data_file;
Kevin Wolfff991292015-09-07 17:12:56 +02002802 memset(s, 0, sizeof(BDRVQcow2State));
Hanna Reitz06e9cd12022-04-27 13:40:55 +02002803 s->data_file = data_file;
2804
Kevin Wolfd475e5a2014-03-11 17:42:41 +01002805 options = qdict_clone_shallow(bs->options);
Kevin Wolf5a8a30d2014-03-12 15:59:16 +01002806
Kevin Wolf140fd5a2015-12-22 16:10:32 +01002807 flags &= ~BDRV_O_INACTIVE;
Paolo Bonzini2b148f32018-03-01 17:36:18 +01002808 qemu_co_mutex_lock(&s->lock);
Hanna Reitz06e9cd12022-04-27 13:40:55 +02002809 ret = qcow2_do_open(bs, options, flags, false, errp);
Paolo Bonzini2b148f32018-03-01 17:36:18 +01002810 qemu_co_mutex_unlock(&s->lock);
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02002811 qobject_unref(options);
Vladimir Sementsov-Ogievskiye6247c92021-02-02 15:49:54 +03002812 if (ret < 0) {
2813 error_prepend(errp, "Could not reopen qcow2 layer: ");
Kevin Wolf191fb112015-12-22 16:14:10 +01002814 bs->drv = NULL;
Kevin Wolf5a8a30d2014-03-12 15:59:16 +01002815 return;
2816 }
Kevin Wolfacdfb482013-03-18 13:08:10 +01002817
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01002818 s->crypto = crypto;
Anthony Liguori06d92602011-11-14 15:09:46 -06002819}
2820
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002821static size_t header_ext_add(char *buf, uint32_t magic, const void *s,
2822 size_t len, size_t buflen)
Kevin Wolf756e6732010-01-12 12:55:17 +01002823{
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002824 QCowExtension *ext_backing_fmt = (QCowExtension*) buf;
2825 size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7);
Kevin Wolf756e6732010-01-12 12:55:17 +01002826
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002827 if (buflen < ext_len) {
Kevin Wolf756e6732010-01-12 12:55:17 +01002828 return -ENOSPC;
2829 }
2830
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002831 *ext_backing_fmt = (QCowExtension) {
2832 .magic = cpu_to_be32(magic),
2833 .len = cpu_to_be32(len),
2834 };
Stefan Hajnoczi0647d472016-09-13 09:56:27 +01002835
2836 if (len) {
2837 memcpy(buf + sizeof(QCowExtension), s, len);
2838 }
Kevin Wolf756e6732010-01-12 12:55:17 +01002839
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002840 return ext_len;
2841}
Kevin Wolf756e6732010-01-12 12:55:17 +01002842
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002843/*
2844 * Updates the qcow2 header, including the variable length parts of it, i.e.
2845 * the backing file name and all extensions. qcow2 was not designed to allow
2846 * such changes, so if we run out of space (we can only use the first cluster)
2847 * this function may fail.
2848 *
2849 * Returns 0 on success, -errno in error cases.
2850 */
2851int qcow2_update_header(BlockDriverState *bs)
2852{
Kevin Wolfff991292015-09-07 17:12:56 +02002853 BDRVQcow2State *s = bs->opaque;
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002854 QCowHeader *header;
2855 char *buf;
2856 size_t buflen = s->cluster_size;
2857 int ret;
2858 uint64_t total_size;
2859 uint32_t refcount_table_clusters;
Kevin Wolf6744cba2011-12-15 12:20:58 +01002860 size_t header_length;
Kevin Wolf75bab852012-02-02 14:52:08 +01002861 Qcow2UnknownHeaderExtension *uext;
Kevin Wolf756e6732010-01-12 12:55:17 +01002862
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002863 buf = qemu_blockalign(bs, buflen);
Kevin Wolf756e6732010-01-12 12:55:17 +01002864
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002865 /* Header structure */
2866 header = (QCowHeader*) buf;
2867
2868 if (buflen < sizeof(*header)) {
2869 ret = -ENOSPC;
2870 goto fail;
2871 }
2872
Kevin Wolf6744cba2011-12-15 12:20:58 +01002873 header_length = sizeof(*header) + s->unknown_header_fields_size;
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002874 total_size = bs->total_sectors * BDRV_SECTOR_SIZE;
2875 refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2876
Denis Plotnikov572ad972020-05-07 11:25:18 +03002877 ret = validate_compression_type(s, NULL);
2878 if (ret) {
2879 goto fail;
2880 }
2881
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002882 *header = (QCowHeader) {
Kevin Wolf6744cba2011-12-15 12:20:58 +01002883 /* Version 2 fields */
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002884 .magic = cpu_to_be32(QCOW_MAGIC),
Kevin Wolf6744cba2011-12-15 12:20:58 +01002885 .version = cpu_to_be32(s->qcow_version),
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002886 .backing_file_offset = 0,
2887 .backing_file_size = 0,
2888 .cluster_bits = cpu_to_be32(s->cluster_bits),
2889 .size = cpu_to_be64(total_size),
2890 .crypt_method = cpu_to_be32(s->crypt_method_header),
2891 .l1_size = cpu_to_be32(s->l1_size),
2892 .l1_table_offset = cpu_to_be64(s->l1_table_offset),
2893 .refcount_table_offset = cpu_to_be64(s->refcount_table_offset),
2894 .refcount_table_clusters = cpu_to_be32(refcount_table_clusters),
2895 .nb_snapshots = cpu_to_be32(s->nb_snapshots),
2896 .snapshots_offset = cpu_to_be64(s->snapshots_offset),
Kevin Wolf6744cba2011-12-15 12:20:58 +01002897
2898 /* Version 3 fields */
2899 .incompatible_features = cpu_to_be64(s->incompatible_features),
2900 .compatible_features = cpu_to_be64(s->compatible_features),
2901 .autoclear_features = cpu_to_be64(s->autoclear_features),
Max Reitzb6481f32013-09-03 10:09:53 +02002902 .refcount_order = cpu_to_be32(s->refcount_order),
Kevin Wolf6744cba2011-12-15 12:20:58 +01002903 .header_length = cpu_to_be32(header_length),
Denis Plotnikov572ad972020-05-07 11:25:18 +03002904 .compression_type = s->compression_type,
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002905 };
2906
Kevin Wolf6744cba2011-12-15 12:20:58 +01002907 /* For older versions, write a shorter header */
2908 switch (s->qcow_version) {
2909 case 2:
2910 ret = offsetof(QCowHeader, incompatible_features);
2911 break;
2912 case 3:
2913 ret = sizeof(*header);
2914 break;
2915 default:
Jim Meyeringb6c14762012-05-21 13:06:54 +02002916 ret = -EINVAL;
2917 goto fail;
Kevin Wolf6744cba2011-12-15 12:20:58 +01002918 }
2919
2920 buf += ret;
2921 buflen -= ret;
2922 memset(buf, 0, buflen);
2923
2924 /* Preserve any unknown field in the header */
2925 if (s->unknown_header_fields_size) {
2926 if (buflen < s->unknown_header_fields_size) {
2927 ret = -ENOSPC;
2928 goto fail;
2929 }
2930
2931 memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size);
2932 buf += s->unknown_header_fields_size;
2933 buflen -= s->unknown_header_fields_size;
2934 }
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002935
2936 /* Backing file format header extension */
Kevin Wolfe4603fe2015-04-07 15:03:16 +02002937 if (s->image_backing_format) {
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002938 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT,
Kevin Wolfe4603fe2015-04-07 15:03:16 +02002939 s->image_backing_format,
2940 strlen(s->image_backing_format),
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002941 buflen);
2942 if (ret < 0) {
2943 goto fail;
Kevin Wolf756e6732010-01-12 12:55:17 +01002944 }
2945
Kevin Wolfe24e49e2012-02-02 12:32:31 +01002946 buf += ret;
2947 buflen -= ret;
Kevin Wolf756e6732010-01-12 12:55:17 +01002948 }
2949
Kevin Wolf9b890bd2019-01-15 19:02:40 +01002950 /* External data file header extension */
2951 if (has_data_file(bs) && s->image_data_file) {
2952 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_DATA_FILE,
2953 s->image_data_file, strlen(s->image_data_file),
2954 buflen);
2955 if (ret < 0) {
2956 goto fail;
2957 }
2958
2959 buf += ret;
2960 buflen -= ret;
2961 }
2962
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +01002963 /* Full disk encryption header pointer extension */
2964 if (s->crypto_header.offset != 0) {
Peter Maydell3b698f52018-10-09 18:24:59 +01002965 s->crypto_header.offset = cpu_to_be64(s->crypto_header.offset);
2966 s->crypto_header.length = cpu_to_be64(s->crypto_header.length);
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +01002967 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_CRYPTO_HEADER,
2968 &s->crypto_header, sizeof(s->crypto_header),
2969 buflen);
Peter Maydell3b698f52018-10-09 18:24:59 +01002970 s->crypto_header.offset = be64_to_cpu(s->crypto_header.offset);
2971 s->crypto_header.length = be64_to_cpu(s->crypto_header.length);
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +01002972 if (ret < 0) {
2973 goto fail;
2974 }
2975 buf += ret;
2976 buflen -= ret;
2977 }
2978
Eric Blakee7be13a2020-03-24 12:42:32 -05002979 /*
2980 * Feature table. A mere 8 feature names occupies 392 bytes, and
2981 * when coupled with the v3 minimum header of 104 bytes plus the
2982 * 8-byte end-of-extension marker, that would leave only 8 bytes
2983 * for a backing file name in an image with 512-byte clusters.
2984 * Thus, we choose to omit this header for cluster sizes 4k and
2985 * smaller.
2986 */
2987 if (s->qcow_version >= 3 && s->cluster_size > 4096) {
Eric Blakebb40ebc2020-03-24 12:42:31 -05002988 static const Qcow2Feature features[] = {
Kevin Wolf1a4828c2015-12-02 19:11:04 +01002989 {
2990 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2991 .bit = QCOW2_INCOMPAT_DIRTY_BITNR,
2992 .name = "dirty bit",
2993 },
2994 {
2995 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
2996 .bit = QCOW2_INCOMPAT_CORRUPT_BITNR,
2997 .name = "corrupt bit",
2998 },
2999 {
Kevin Wolf93c24932019-01-14 16:48:25 +01003000 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3001 .bit = QCOW2_INCOMPAT_DATA_FILE_BITNR,
3002 .name = "external data file",
3003 },
3004 {
Denis Plotnikov572ad972020-05-07 11:25:18 +03003005 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3006 .bit = QCOW2_INCOMPAT_COMPRESSION_BITNR,
3007 .name = "compression type",
3008 },
3009 {
Alberto Garcia7be20252020-07-10 18:13:13 +02003010 .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
3011 .bit = QCOW2_INCOMPAT_EXTL2_BITNR,
3012 .name = "extended L2 entries",
3013 },
3014 {
Kevin Wolf1a4828c2015-12-02 19:11:04 +01003015 .type = QCOW2_FEAT_TYPE_COMPATIBLE,
3016 .bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
3017 .name = "lazy refcounts",
3018 },
Eric Blakebb40ebc2020-03-24 12:42:31 -05003019 {
3020 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
3021 .bit = QCOW2_AUTOCLEAR_BITMAPS_BITNR,
3022 .name = "bitmaps",
3023 },
3024 {
3025 .type = QCOW2_FEAT_TYPE_AUTOCLEAR,
3026 .bit = QCOW2_AUTOCLEAR_DATA_FILE_RAW_BITNR,
3027 .name = "raw external data",
3028 },
Kevin Wolf1a4828c2015-12-02 19:11:04 +01003029 };
Kevin Wolfcfcc4c62012-04-12 15:20:27 +02003030
Kevin Wolf1a4828c2015-12-02 19:11:04 +01003031 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
3032 features, sizeof(features), buflen);
3033 if (ret < 0) {
3034 goto fail;
3035 }
3036 buf += ret;
3037 buflen -= ret;
Kevin Wolfcfcc4c62012-04-12 15:20:27 +02003038 }
Kevin Wolfcfcc4c62012-04-12 15:20:27 +02003039
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +03003040 /* Bitmap extension */
3041 if (s->nb_bitmaps > 0) {
3042 Qcow2BitmapHeaderExt bitmaps_header = {
3043 .nb_bitmaps = cpu_to_be32(s->nb_bitmaps),
3044 .bitmap_directory_size =
3045 cpu_to_be64(s->bitmap_directory_size),
3046 .bitmap_directory_offset =
3047 cpu_to_be64(s->bitmap_directory_offset)
3048 };
3049 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BITMAPS,
3050 &bitmaps_header, sizeof(bitmaps_header),
3051 buflen);
3052 if (ret < 0) {
3053 goto fail;
3054 }
3055 buf += ret;
3056 buflen -= ret;
3057 }
3058
Kevin Wolf75bab852012-02-02 14:52:08 +01003059 /* Keep unknown header extensions */
3060 QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
3061 ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen);
3062 if (ret < 0) {
3063 goto fail;
3064 }
3065
3066 buf += ret;
3067 buflen -= ret;
3068 }
3069
Kevin Wolfe24e49e2012-02-02 12:32:31 +01003070 /* End of header extensions */
3071 ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen);
Kevin Wolf756e6732010-01-12 12:55:17 +01003072 if (ret < 0) {
3073 goto fail;
3074 }
3075
Kevin Wolfe24e49e2012-02-02 12:32:31 +01003076 buf += ret;
3077 buflen -= ret;
Kevin Wolf756e6732010-01-12 12:55:17 +01003078
Kevin Wolfe24e49e2012-02-02 12:32:31 +01003079 /* Backing file name */
Kevin Wolfe4603fe2015-04-07 15:03:16 +02003080 if (s->image_backing_file) {
3081 size_t backing_file_len = strlen(s->image_backing_file);
Kevin Wolfe24e49e2012-02-02 12:32:31 +01003082
3083 if (buflen < backing_file_len) {
3084 ret = -ENOSPC;
3085 goto fail;
3086 }
3087
Jim Meyering00ea1882012-10-04 13:10:01 +02003088 /* Using strncpy is ok here, since buf is not NUL-terminated. */
Kevin Wolfe4603fe2015-04-07 15:03:16 +02003089 strncpy(buf, s->image_backing_file, buflen);
Kevin Wolfe24e49e2012-02-02 12:32:31 +01003090
3091 header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
3092 header->backing_file_size = cpu_to_be32(backing_file_len);
Kevin Wolf756e6732010-01-12 12:55:17 +01003093 }
3094
Kevin Wolfe24e49e2012-02-02 12:32:31 +01003095 /* Write the new header */
Alberto Faria32cc71d2022-06-09 16:27:36 +01003096 ret = bdrv_pwrite(bs->file, 0, s->cluster_size, header, 0);
Kevin Wolf756e6732010-01-12 12:55:17 +01003097 if (ret < 0) {
3098 goto fail;
3099 }
3100
3101 ret = 0;
3102fail:
Kevin Wolfe24e49e2012-02-02 12:32:31 +01003103 qemu_vfree(header);
Kevin Wolf756e6732010-01-12 12:55:17 +01003104 return ret;
3105}
3106
3107static int qcow2_change_backing_file(BlockDriverState *bs,
3108 const char *backing_file, const char *backing_fmt)
3109{
Kevin Wolfff991292015-09-07 17:12:56 +02003110 BDRVQcow2State *s = bs->opaque;
Kevin Wolfe4603fe2015-04-07 15:03:16 +02003111
Kevin Wolf6c3944d2019-02-22 14:29:38 +01003112 /* Adding a backing file means that the external data file alone won't be
3113 * enough to make sense of the content */
3114 if (backing_file && data_file_is_raw(bs)) {
3115 return -EINVAL;
3116 }
3117
Max Reitz4e876bc2016-04-06 18:32:48 +02003118 if (backing_file && strlen(backing_file) > 1023) {
3119 return -EINVAL;
3120 }
3121
Max Reitz998c2012019-02-01 20:29:08 +01003122 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3123 backing_file ?: "");
Kevin Wolfe24e49e2012-02-02 12:32:31 +01003124 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3125 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3126
Kevin Wolfe4603fe2015-04-07 15:03:16 +02003127 g_free(s->image_backing_file);
3128 g_free(s->image_backing_format);
3129
3130 s->image_backing_file = backing_file ? g_strdup(bs->backing_file) : NULL;
3131 s->image_backing_format = backing_fmt ? g_strdup(bs->backing_format) : NULL;
3132
Kevin Wolfe24e49e2012-02-02 12:32:31 +01003133 return qcow2_update_header(bs);
Kevin Wolf756e6732010-01-12 12:55:17 +01003134}
3135
Kevin Wolf60900b72018-01-10 17:55:16 +01003136static int qcow2_set_up_encryption(BlockDriverState *bs,
3137 QCryptoBlockCreateOptions *cryptoopts,
3138 Error **errp)
3139{
3140 BDRVQcow2State *s = bs->opaque;
3141 QCryptoBlock *crypto = NULL;
3142 int fmt, ret;
3143
3144 switch (cryptoopts->format) {
3145 case Q_CRYPTO_BLOCK_FORMAT_LUKS:
3146 fmt = QCOW_CRYPT_LUKS;
3147 break;
3148 case Q_CRYPTO_BLOCK_FORMAT_QCOW:
3149 fmt = QCOW_CRYPT_AES;
3150 break;
3151 default:
3152 error_setg(errp, "Crypto format not supported in qcow2");
3153 return -EINVAL;
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01003154 }
Kevin Wolf60900b72018-01-10 17:55:16 +01003155
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +01003156 s->crypt_method_header = fmt;
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01003157
Daniel P. Berrange1cd9a782017-06-23 17:24:17 +01003158 crypto = qcrypto_block_create(cryptoopts, "encrypt.",
Daniel P. Berrange4652b8f2017-06-23 17:24:12 +01003159 qcow2_crypto_hdr_init_func,
3160 qcow2_crypto_hdr_write_func,
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01003161 bs, errp);
3162 if (!crypto) {
Kevin Wolf60900b72018-01-10 17:55:16 +01003163 return -EINVAL;
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01003164 }
3165
3166 ret = qcow2_update_header(bs);
3167 if (ret < 0) {
3168 error_setg_errno(errp, -ret, "Could not write encryption header");
3169 goto out;
3170 }
3171
Kevin Wolf60900b72018-01-10 17:55:16 +01003172 ret = 0;
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01003173 out:
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01003174 qcrypto_block_free(crypto);
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01003175 return ret;
3176}
3177
Max Reitz7bc45dc2017-06-13 22:21:00 +02003178/**
3179 * Preallocates metadata structures for data clusters between @offset (in the
3180 * guest disk) and @new_length (which is thus generally the new guest disk
3181 * size).
3182 *
3183 * Returns: 0 on success, -errno on failure.
3184 */
Kevin Wolf47e86b82018-06-26 15:52:13 +02003185static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset,
Kevin Wolf718c0fc2019-04-15 16:34:30 +02003186 uint64_t new_length, PreallocMode mode,
3187 Error **errp)
Kevin Wolfa35e1c12009-08-17 15:50:10 +02003188{
Kevin Wolf93e32b32019-04-15 17:54:50 +02003189 BDRVQcow2State *s = bs->opaque;
Kevin Wolfd46a0bb2016-06-01 16:55:05 +02003190 uint64_t bytes;
Kevin Wolf060bee82012-12-07 18:08:45 +01003191 uint64_t host_offset = 0;
Kevin Wolf718c0fc2019-04-15 16:34:30 +02003192 int64_t file_length;
Kevin Wolfd46a0bb2016-06-01 16:55:05 +02003193 unsigned int cur_bytes;
Kevin Wolf148da7e2010-01-20 15:03:01 +01003194 int ret;
Alberto Garcia1a52b732020-09-08 16:08:27 +02003195 QCowL2Meta *meta = NULL, *m;
Kevin Wolfa35e1c12009-08-17 15:50:10 +02003196
Max Reitz7bc45dc2017-06-13 22:21:00 +02003197 assert(offset <= new_length);
3198 bytes = new_length - offset;
Kevin Wolfa35e1c12009-08-17 15:50:10 +02003199
Kevin Wolfd46a0bb2016-06-01 16:55:05 +02003200 while (bytes) {
Kevin Wolff29fbf72019-04-15 16:25:01 +02003201 cur_bytes = MIN(bytes, QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size));
Alberto Garciabfd09892020-09-11 16:09:42 +02003202 ret = qcow2_alloc_host_offset(bs, offset, &cur_bytes,
3203 &host_offset, &meta);
Kevin Wolf148da7e2010-01-20 15:03:01 +01003204 if (ret < 0) {
Kevin Wolf360bd072019-04-15 16:56:07 +02003205 error_setg_errno(errp, -ret, "Allocating clusters failed");
Alberto Garcia1a52b732020-09-08 16:08:27 +02003206 goto out;
Kevin Wolfa35e1c12009-08-17 15:50:10 +02003207 }
3208
Alberto Garcia1a52b732020-09-08 16:08:27 +02003209 for (m = meta; m != NULL; m = m->next) {
3210 m->prealloc = true;
3211 }
Stefan Hajnoczic7927072014-04-01 11:12:57 +02003212
Alberto Garcia1a52b732020-09-08 16:08:27 +02003213 ret = qcow2_handle_l2meta(bs, &meta, true);
3214 if (ret < 0) {
3215 error_setg_errno(errp, -ret, "Mapping clusters failed");
3216 goto out;
Kevin Wolff50f88b2012-12-07 18:08:46 +01003217 }
Kevin Wolff2149782009-08-31 16:48:49 +02003218
Kevin Wolfa35e1c12009-08-17 15:50:10 +02003219 /* TODO Preallocate data if requested */
3220
Kevin Wolfd46a0bb2016-06-01 16:55:05 +02003221 bytes -= cur_bytes;
3222 offset += cur_bytes;
Kevin Wolfa35e1c12009-08-17 15:50:10 +02003223 }
3224
3225 /*
3226 * It is expected that the image file is large enough to actually contain
3227 * all of the allocated clusters (otherwise we get failing reads after
3228 * EOF). Extend the image to the last allocated sector.
3229 */
Kevin Wolf718c0fc2019-04-15 16:34:30 +02003230 file_length = bdrv_getlength(s->data_file->bs);
3231 if (file_length < 0) {
3232 error_setg_errno(errp, -file_length, "Could not get file size");
Alberto Garcia1a52b732020-09-08 16:08:27 +02003233 ret = file_length;
3234 goto out;
Kevin Wolf718c0fc2019-04-15 16:34:30 +02003235 }
3236
3237 if (host_offset + cur_bytes > file_length) {
3238 if (mode == PREALLOC_MODE_METADATA) {
3239 mode = PREALLOC_MODE_OFF;
3240 }
Max Reitzc80d8b02019-09-18 11:51:40 +02003241 ret = bdrv_co_truncate(s->data_file, host_offset + cur_bytes, false,
Kevin Wolf7b8e4852020-04-24 14:54:40 +02003242 mode, 0, errp);
Kevin Wolf19dbcbf2010-06-22 16:59:46 +02003243 if (ret < 0) {
Alberto Garcia1a52b732020-09-08 16:08:27 +02003244 goto out;
Kevin Wolf19dbcbf2010-06-22 16:59:46 +02003245 }
Kevin Wolfa35e1c12009-08-17 15:50:10 +02003246 }
3247
Alberto Garcia1a52b732020-09-08 16:08:27 +02003248 ret = 0;
3249
3250out:
3251 qcow2_handle_l2meta(bs, &meta, false);
3252 return ret;
Kevin Wolfa35e1c12009-08-17 15:50:10 +02003253}
3254
Stefan Hajnoczi7c5bcc42017-07-05 13:57:33 +01003255/* qcow2_refcount_metadata_size:
3256 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
3257 * @cluster_size: size of a cluster, in bytes
3258 * @refcount_order: refcount bits power-of-2 exponent
Max Reitz12cc30a2017-06-13 22:21:03 +02003259 * @generous_increase: allow for the refcount table to be 1.5x as large as it
3260 * needs to be
Stefan Hajnoczi7c5bcc42017-07-05 13:57:33 +01003261 *
3262 * Returns: Number of bytes required for refcount blocks and table metadata.
3263 */
Max Reitz12cc30a2017-06-13 22:21:03 +02003264int64_t qcow2_refcount_metadata_size(int64_t clusters, size_t cluster_size,
3265 int refcount_order, bool generous_increase,
3266 uint64_t *refblock_count)
Stefan Hajnoczi7c5bcc42017-07-05 13:57:33 +01003267{
3268 /*
3269 * Every host cluster is reference-counted, including metadata (even
3270 * refcount metadata is recursively included).
3271 *
3272 * An accurate formula for the size of refcount metadata size is difficult
3273 * to derive. An easier method of calculation is finding the fixed point
3274 * where no further refcount blocks or table clusters are required to
3275 * reference count every cluster.
3276 */
Alberto Garcia02b1ecf2020-08-28 13:08:28 +02003277 int64_t blocks_per_table_cluster = cluster_size / REFTABLE_ENTRY_SIZE;
Stefan Hajnoczi7c5bcc42017-07-05 13:57:33 +01003278 int64_t refcounts_per_block = cluster_size * 8 / (1 << refcount_order);
3279 int64_t table = 0; /* number of refcount table clusters */
3280 int64_t blocks = 0; /* number of refcount block clusters */
3281 int64_t last;
3282 int64_t n = 0;
3283
3284 do {
3285 last = n;
3286 blocks = DIV_ROUND_UP(clusters + table + blocks, refcounts_per_block);
3287 table = DIV_ROUND_UP(blocks, blocks_per_table_cluster);
3288 n = clusters + blocks + table;
Max Reitz12cc30a2017-06-13 22:21:03 +02003289
3290 if (n == last && generous_increase) {
3291 clusters += DIV_ROUND_UP(table, 2);
3292 n = 0; /* force another loop */
3293 generous_increase = false;
3294 }
Stefan Hajnoczi7c5bcc42017-07-05 13:57:33 +01003295 } while (n != last);
3296
Max Reitz12cc30a2017-06-13 22:21:03 +02003297 if (refblock_count) {
3298 *refblock_count = blocks;
3299 }
3300
Stefan Hajnoczi7c5bcc42017-07-05 13:57:33 +01003301 return (blocks + table) * cluster_size;
3302}
3303
Stefan Hajnoczi95c67e32017-07-05 13:57:32 +01003304/**
3305 * qcow2_calc_prealloc_size:
3306 * @total_size: virtual disk size in bytes
3307 * @cluster_size: cluster size in bytes
3308 * @refcount_order: refcount bits power-of-2 exponent
Alberto Garcia0dd07b22020-07-10 18:13:11 +02003309 * @extended_l2: true if the image has extended L2 entries
Stefan Hajnoczi95c67e32017-07-05 13:57:32 +01003310 *
3311 * Returns: Total number of bytes required for the fully allocated image
3312 * (including metadata).
3313 */
3314static int64_t qcow2_calc_prealloc_size(int64_t total_size,
3315 size_t cluster_size,
Alberto Garcia0dd07b22020-07-10 18:13:11 +02003316 int refcount_order,
3317 bool extended_l2)
Stefan Hajnoczi95c67e32017-07-05 13:57:32 +01003318{
Stefan Hajnoczi95c67e32017-07-05 13:57:32 +01003319 int64_t meta_size = 0;
Stefan Hajnoczi7c5bcc42017-07-05 13:57:33 +01003320 uint64_t nl1e, nl2e;
Alberto Garcia9e029682018-02-15 15:10:08 +02003321 int64_t aligned_total_size = ROUND_UP(total_size, cluster_size);
Alberto Garcia0dd07b22020-07-10 18:13:11 +02003322 size_t l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL;
Stefan Hajnoczi95c67e32017-07-05 13:57:32 +01003323
3324 /* header: 1 cluster */
3325 meta_size += cluster_size;
3326
3327 /* total size of L2 tables */
3328 nl2e = aligned_total_size / cluster_size;
Alberto Garcia0dd07b22020-07-10 18:13:11 +02003329 nl2e = ROUND_UP(nl2e, cluster_size / l2e_size);
3330 meta_size += nl2e * l2e_size;
Stefan Hajnoczi95c67e32017-07-05 13:57:32 +01003331
3332 /* total size of L1 tables */
Alberto Garcia0dd07b22020-07-10 18:13:11 +02003333 nl1e = nl2e * l2e_size / cluster_size;
Alberto Garcia02b1ecf2020-08-28 13:08:28 +02003334 nl1e = ROUND_UP(nl1e, cluster_size / L1E_SIZE);
3335 meta_size += nl1e * L1E_SIZE;
Stefan Hajnoczi95c67e32017-07-05 13:57:32 +01003336
Stefan Hajnoczi7c5bcc42017-07-05 13:57:33 +01003337 /* total size of refcount table and blocks */
3338 meta_size += qcow2_refcount_metadata_size(
3339 (meta_size + aligned_total_size) / cluster_size,
Max Reitz12cc30a2017-06-13 22:21:03 +02003340 cluster_size, refcount_order, false, NULL);
Stefan Hajnoczi95c67e32017-07-05 13:57:32 +01003341
3342 return meta_size + aligned_total_size;
3343}
3344
Alberto Garcia7be20252020-07-10 18:13:13 +02003345static bool validate_cluster_size(size_t cluster_size, bool extended_l2,
3346 Error **errp)
Kevin Wolfa9420732010-06-11 21:37:37 +02003347{
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003348 int cluster_bits = ctz32(cluster_size);
Kevin Wolfa9420732010-06-11 21:37:37 +02003349 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
3350 (1 << cluster_bits) != cluster_size)
3351 {
Max Reitz3ef6c402013-09-05 09:40:43 +02003352 error_setg(errp, "Cluster size must be a power of two between %d and "
3353 "%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003354 return false;
3355 }
Alberto Garcia7be20252020-07-10 18:13:13 +02003356
3357 if (extended_l2) {
3358 unsigned min_cluster_size =
3359 (1 << MIN_CLUSTER_BITS) * QCOW_EXTL2_SUBCLUSTERS_PER_CLUSTER;
3360 if (cluster_size < min_cluster_size) {
3361 error_setg(errp, "Extended L2 entries are only supported with "
3362 "cluster sizes of at least %u bytes", min_cluster_size);
3363 return false;
3364 }
3365 }
3366
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003367 return true;
3368}
3369
Alberto Garcia7be20252020-07-10 18:13:13 +02003370static size_t qcow2_opt_get_cluster_size_del(QemuOpts *opts, bool extended_l2,
3371 Error **errp)
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003372{
3373 size_t cluster_size;
3374
3375 cluster_size = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE,
3376 DEFAULT_CLUSTER_SIZE);
Alberto Garcia7be20252020-07-10 18:13:13 +02003377 if (!validate_cluster_size(cluster_size, extended_l2, errp)) {
Stefan Hajnoczi0eb4a8c2017-07-05 13:57:34 +01003378 return 0;
Kevin Wolfa9420732010-06-11 21:37:37 +02003379 }
Stefan Hajnoczi0eb4a8c2017-07-05 13:57:34 +01003380 return cluster_size;
3381}
3382
3383static int qcow2_opt_get_version_del(QemuOpts *opts, Error **errp)
3384{
3385 char *buf;
3386 int ret;
3387
3388 buf = qemu_opt_get_del(opts, BLOCK_OPT_COMPAT_LEVEL);
3389 if (!buf) {
3390 ret = 3; /* default */
3391 } else if (!strcmp(buf, "0.10")) {
3392 ret = 2;
3393 } else if (!strcmp(buf, "1.1")) {
3394 ret = 3;
3395 } else {
3396 error_setg(errp, "Invalid compatibility level: '%s'", buf);
3397 ret = -EINVAL;
3398 }
3399 g_free(buf);
3400 return ret;
3401}
3402
3403static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts *opts, int version,
3404 Error **errp)
3405{
3406 uint64_t refcount_bits;
3407
3408 refcount_bits = qemu_opt_get_number_del(opts, BLOCK_OPT_REFCOUNT_BITS, 16);
3409 if (refcount_bits > 64 || !is_power_of_2(refcount_bits)) {
3410 error_setg(errp, "Refcount width must be a power of two and may not "
3411 "exceed 64 bits");
3412 return 0;
3413 }
3414
3415 if (version < 3 && refcount_bits != 16) {
3416 error_setg(errp, "Different refcount widths than 16 bits require "
3417 "compatibility level 1.1 or above (use compat=1.1 or "
3418 "greater)");
3419 return 0;
3420 }
3421
3422 return refcount_bits;
3423}
3424
Stefan Hajnoczic2743932018-01-18 13:43:46 +01003425static int coroutine_fn
Kevin Wolf60900b72018-01-10 17:55:16 +01003426qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
Stefan Hajnoczi0eb4a8c2017-07-05 13:57:34 +01003427{
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003428 BlockdevCreateOptionsQcow2 *qcow2_opts;
Stefan Hajnoczi0eb4a8c2017-07-05 13:57:34 +01003429 QDict *options;
Kevin Wolfa9420732010-06-11 21:37:37 +02003430
3431 /*
3432 * Open the image file and write a minimal qcow2 header.
3433 *
3434 * We keep things simple and start with a zero-sized image. We also
3435 * do without refcount blocks or a L1 table for now. We'll fix the
3436 * inconsistency later.
3437 *
3438 * We do need a refcount table because growing the refcount table means
Eric Blakea951a632020-03-24 12:42:30 -05003439 * allocating two new refcount blocks - the second of which would be at
Kevin Wolfa9420732010-06-11 21:37:37 +02003440 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
3441 * size for any qcow2 image.
3442 */
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003443 BlockBackend *blk = NULL;
3444 BlockDriverState *bs = NULL;
Kevin Wolfdcc98682019-01-14 16:57:27 +01003445 BlockDriverState *data_bs = NULL;
Kevin Wolff8413b32013-12-04 11:06:36 +01003446 QCowHeader *header;
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003447 size_t cluster_size;
3448 int version;
3449 int refcount_order;
shiliyang5f14f312020-10-30 11:35:12 +08003450 uint64_t *refcount_table;
Kevin Wolfa9420732010-06-11 21:37:37 +02003451 int ret;
Denis Plotnikov572ad972020-05-07 11:25:18 +03003452 uint8_t compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
Kevin Wolfa9420732010-06-11 21:37:37 +02003453
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003454 assert(create_options->driver == BLOCKDEV_DRIVER_QCOW2);
3455 qcow2_opts = &create_options->u.qcow2;
3456
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003457 bs = bdrv_open_blockdev_ref(qcow2_opts->file, errp);
3458 if (bs == NULL) {
3459 return -EIO;
3460 }
3461
3462 /* Validate options and set default values */
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003463 if (!QEMU_IS_ALIGNED(qcow2_opts->size, BDRV_SECTOR_SIZE)) {
Alberto Garcia3afea402020-01-18 20:09:30 +01003464 error_setg(errp, "Image size must be a multiple of %u bytes",
3465 (unsigned) BDRV_SECTOR_SIZE);
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003466 ret = -EINVAL;
3467 goto out;
3468 }
3469
3470 if (qcow2_opts->has_version) {
3471 switch (qcow2_opts->version) {
3472 case BLOCKDEV_QCOW2_VERSION_V2:
3473 version = 2;
3474 break;
3475 case BLOCKDEV_QCOW2_VERSION_V3:
3476 version = 3;
3477 break;
3478 default:
3479 g_assert_not_reached();
3480 }
3481 } else {
3482 version = 3;
3483 }
3484
3485 if (qcow2_opts->has_cluster_size) {
3486 cluster_size = qcow2_opts->cluster_size;
3487 } else {
3488 cluster_size = DEFAULT_CLUSTER_SIZE;
3489 }
3490
Alberto Garcia7be20252020-07-10 18:13:13 +02003491 if (!qcow2_opts->has_extended_l2) {
3492 qcow2_opts->extended_l2 = false;
3493 }
3494 if (qcow2_opts->extended_l2) {
3495 if (version < 3) {
3496 error_setg(errp, "Extended L2 entries are only supported with "
3497 "compatibility level 1.1 and above (use version=v3 or "
3498 "greater)");
3499 ret = -EINVAL;
3500 goto out;
3501 }
3502 }
3503
3504 if (!validate_cluster_size(cluster_size, qcow2_opts->extended_l2, errp)) {
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003505 ret = -EINVAL;
3506 goto out;
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003507 }
3508
3509 if (!qcow2_opts->has_preallocation) {
3510 qcow2_opts->preallocation = PREALLOC_MODE_OFF;
3511 }
Markus Armbruster54fde4f2022-11-04 17:06:52 +01003512 if (qcow2_opts->backing_file &&
Alberto Garcia21187712020-07-10 18:13:14 +02003513 qcow2_opts->preallocation != PREALLOC_MODE_OFF &&
3514 !qcow2_opts->extended_l2)
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003515 {
Alberto Garcia21187712020-07-10 18:13:14 +02003516 error_setg(errp, "Backing file and preallocation can only be used at "
3517 "the same time if extended_l2 is on");
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003518 ret = -EINVAL;
3519 goto out;
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003520 }
Markus Armbruster54fde4f2022-11-04 17:06:52 +01003521 if (qcow2_opts->has_backing_fmt && !qcow2_opts->backing_file) {
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003522 error_setg(errp, "Backing format cannot be used without backing file");
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003523 ret = -EINVAL;
3524 goto out;
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003525 }
3526
3527 if (!qcow2_opts->has_lazy_refcounts) {
3528 qcow2_opts->lazy_refcounts = false;
3529 }
3530 if (version < 3 && qcow2_opts->lazy_refcounts) {
3531 error_setg(errp, "Lazy refcounts only supported with compatibility "
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003532 "level 1.1 and above (use version=v3 or greater)");
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003533 ret = -EINVAL;
3534 goto out;
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003535 }
3536
3537 if (!qcow2_opts->has_refcount_bits) {
3538 qcow2_opts->refcount_bits = 16;
3539 }
3540 if (qcow2_opts->refcount_bits > 64 ||
3541 !is_power_of_2(qcow2_opts->refcount_bits))
3542 {
3543 error_setg(errp, "Refcount width must be a power of two and may not "
3544 "exceed 64 bits");
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003545 ret = -EINVAL;
3546 goto out;
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003547 }
3548 if (version < 3 && qcow2_opts->refcount_bits != 16) {
3549 error_setg(errp, "Different refcount widths than 16 bits require "
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003550 "compatibility level 1.1 or above (use version=v3 or "
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003551 "greater)");
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003552 ret = -EINVAL;
3553 goto out;
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003554 }
3555 refcount_order = ctz32(qcow2_opts->refcount_bits);
3556
Kevin Wolf6c3944d2019-02-22 14:29:38 +01003557 if (qcow2_opts->data_file_raw && !qcow2_opts->data_file) {
3558 error_setg(errp, "data-file-raw requires data-file");
3559 ret = -EINVAL;
3560 goto out;
3561 }
Markus Armbruster54fde4f2022-11-04 17:06:52 +01003562 if (qcow2_opts->data_file_raw && qcow2_opts->backing_file) {
Kevin Wolf6c3944d2019-02-22 14:29:38 +01003563 error_setg(errp, "Backing file and data-file-raw cannot be used at "
3564 "the same time");
3565 ret = -EINVAL;
3566 goto out;
3567 }
Max Reitz48410822021-03-26 15:55:08 +01003568 if (qcow2_opts->data_file_raw &&
3569 qcow2_opts->preallocation == PREALLOC_MODE_OFF)
3570 {
3571 /*
3572 * data-file-raw means that "the external data file can be
3573 * read as a consistent standalone raw image without looking
3574 * at the qcow2 metadata." It does not say that the metadata
3575 * must be ignored, though (and the qcow2 driver in fact does
3576 * not ignore it), so the L1/L2 tables must be present and
3577 * give a 1:1 mapping, so you get the same result regardless
3578 * of whether you look at the metadata or whether you ignore
3579 * it.
3580 */
3581 qcow2_opts->preallocation = PREALLOC_MODE_METADATA;
3582
3583 /*
3584 * Cannot use preallocation with backing files, but giving a
3585 * backing file when specifying data_file_raw is an error
3586 * anyway.
3587 */
Markus Armbruster54fde4f2022-11-04 17:06:52 +01003588 assert(!qcow2_opts->backing_file);
Max Reitz48410822021-03-26 15:55:08 +01003589 }
Kevin Wolf6c3944d2019-02-22 14:29:38 +01003590
Kevin Wolfdcc98682019-01-14 16:57:27 +01003591 if (qcow2_opts->data_file) {
3592 if (version < 3) {
3593 error_setg(errp, "External data files are only supported with "
3594 "compatibility level 1.1 and above (use version=v3 or "
3595 "greater)");
3596 ret = -EINVAL;
3597 goto out;
3598 }
3599 data_bs = bdrv_open_blockdev_ref(qcow2_opts->data_file, errp);
Kevin Wolfa0cf8362019-03-13 15:22:38 +01003600 if (data_bs == NULL) {
Kevin Wolfdcc98682019-01-14 16:57:27 +01003601 ret = -EIO;
3602 goto out;
3603 }
3604 }
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003605
Denis Plotnikov572ad972020-05-07 11:25:18 +03003606 if (qcow2_opts->has_compression_type &&
3607 qcow2_opts->compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) {
3608
3609 ret = -EINVAL;
3610
3611 if (version < 3) {
3612 error_setg(errp, "Non-zlib compression type is only supported with "
3613 "compatibility level 1.1 and above (use version=v3 or "
3614 "greater)");
3615 goto out;
3616 }
3617
3618 switch (qcow2_opts->compression_type) {
Denis Plotnikovd298ac12020-05-07 11:25:20 +03003619#ifdef CONFIG_ZSTD
3620 case QCOW2_COMPRESSION_TYPE_ZSTD:
3621 break;
3622#endif
Denis Plotnikov572ad972020-05-07 11:25:18 +03003623 default:
3624 error_setg(errp, "Unknown compression type");
3625 goto out;
3626 }
3627
3628 compression_type = qcow2_opts->compression_type;
3629 }
3630
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003631 /* Create BlockBackend to write to the image */
Eric Blakea3aeeab2020-04-28 14:26:46 -05003632 blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL,
3633 errp);
3634 if (!blk) {
3635 ret = -EPERM;
Kevin Wolfcbf2b7c2018-01-09 17:35:26 +01003636 goto out;
Kevin Wolfa9420732010-06-11 21:37:37 +02003637 }
Kevin Wolf23588792016-03-08 15:57:05 +01003638 blk_set_allow_write_beyond_eof(blk, true);
3639
Kevin Wolfa9420732010-06-11 21:37:37 +02003640 /* Write the header */
Kevin Wolff8413b32013-12-04 11:06:36 +01003641 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header));
3642 header = g_malloc0(cluster_size);
3643 *header = (QCowHeader) {
3644 .magic = cpu_to_be32(QCOW_MAGIC),
3645 .version = cpu_to_be32(version),
Stefan Hajnoczi0eb4a8c2017-07-05 13:57:34 +01003646 .cluster_bits = cpu_to_be32(ctz32(cluster_size)),
Kevin Wolff8413b32013-12-04 11:06:36 +01003647 .size = cpu_to_be64(0),
3648 .l1_table_offset = cpu_to_be64(0),
3649 .l1_size = cpu_to_be32(0),
3650 .refcount_table_offset = cpu_to_be64(cluster_size),
3651 .refcount_table_clusters = cpu_to_be32(1),
Max Reitzbd4b1672015-02-18 17:40:46 -05003652 .refcount_order = cpu_to_be32(refcount_order),
Denis Plotnikov572ad972020-05-07 11:25:18 +03003653 /* don't deal with endianness since compression_type is 1 byte long */
3654 .compression_type = compression_type,
Kevin Wolff8413b32013-12-04 11:06:36 +01003655 .header_length = cpu_to_be32(sizeof(*header)),
3656 };
Kevin Wolfa9420732010-06-11 21:37:37 +02003657
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01003658 /* We'll update this to correct value later */
3659 header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
Kevin Wolfa9420732010-06-11 21:37:37 +02003660
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003661 if (qcow2_opts->lazy_refcounts) {
Kevin Wolff8413b32013-12-04 11:06:36 +01003662 header->compatible_features |=
Stefan Hajnoczibfe80432012-07-27 09:05:22 +01003663 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
3664 }
Kevin Wolfdcc98682019-01-14 16:57:27 +01003665 if (data_bs) {
3666 header->incompatible_features |=
3667 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE);
3668 }
Kevin Wolf6c3944d2019-02-22 14:29:38 +01003669 if (qcow2_opts->data_file_raw) {
3670 header->autoclear_features |=
3671 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW);
3672 }
Denis Plotnikov572ad972020-05-07 11:25:18 +03003673 if (compression_type != QCOW2_COMPRESSION_TYPE_ZLIB) {
3674 header->incompatible_features |=
3675 cpu_to_be64(QCOW2_INCOMPAT_COMPRESSION);
3676 }
Stefan Hajnoczibfe80432012-07-27 09:05:22 +01003677
Alberto Garcia7be20252020-07-10 18:13:13 +02003678 if (qcow2_opts->extended_l2) {
3679 header->incompatible_features |=
3680 cpu_to_be64(QCOW2_INCOMPAT_EXTL2);
3681 }
3682
Alberto Faria38505e22022-10-13 14:37:06 +02003683 ret = blk_co_pwrite(blk, 0, cluster_size, header, 0);
Kevin Wolff8413b32013-12-04 11:06:36 +01003684 g_free(header);
Kevin Wolfa9420732010-06-11 21:37:37 +02003685 if (ret < 0) {
Max Reitz3ef6c402013-09-05 09:40:43 +02003686 error_setg_errno(errp, -ret, "Could not write qcow2 header");
Kevin Wolfa9420732010-06-11 21:37:37 +02003687 goto out;
3688 }
3689
Kevin Wolfb106ad92014-03-28 18:06:31 +01003690 /* Write a refcount table with one refcount block */
3691 refcount_table = g_malloc0(2 * cluster_size);
3692 refcount_table[0] = cpu_to_be64(2 * cluster_size);
Alberto Faria38505e22022-10-13 14:37:06 +02003693 ret = blk_co_pwrite(blk, cluster_size, 2 * cluster_size, refcount_table, 0);
Anthony Liguori7267c092011-08-20 22:09:37 -05003694 g_free(refcount_table);
Kevin Wolfa9420732010-06-11 21:37:37 +02003695
3696 if (ret < 0) {
Max Reitz3ef6c402013-09-05 09:40:43 +02003697 error_setg_errno(errp, -ret, "Could not write refcount table");
Kevin Wolfa9420732010-06-11 21:37:37 +02003698 goto out;
3699 }
3700
Kevin Wolf23588792016-03-08 15:57:05 +01003701 blk_unref(blk);
3702 blk = NULL;
Kevin Wolfa9420732010-06-11 21:37:37 +02003703
3704 /*
3705 * And now open the image and make it consistent first (i.e. increase the
3706 * refcount of the cluster that is occupied by the header and the refcount
3707 * table)
3708 */
Max Reitze6641712015-08-26 19:47:48 +02003709 options = qdict_new();
Eric Blake46f5ac22017-04-27 16:58:17 -05003710 qdict_put_str(options, "driver", "qcow2");
Kevin Wolfcbf2b7c2018-01-09 17:35:26 +01003711 qdict_put_str(options, "file", bs->node_name);
Kevin Wolfdcc98682019-01-14 16:57:27 +01003712 if (data_bs) {
3713 qdict_put_str(options, "data-file", data_bs->node_name);
3714 }
Kevin Wolfcbf2b7c2018-01-09 17:35:26 +01003715 blk = blk_new_open(NULL, NULL, options,
Kevin Wolf55880602017-02-17 15:07:38 +01003716 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_NO_FLUSH,
Markus Armbrusteraf175e82020-07-07 18:06:03 +02003717 errp);
Kevin Wolf23588792016-03-08 15:57:05 +01003718 if (blk == NULL) {
Kevin Wolf23588792016-03-08 15:57:05 +01003719 ret = -EIO;
Kevin Wolfa9420732010-06-11 21:37:37 +02003720 goto out;
3721 }
3722
Kevin Wolf23588792016-03-08 15:57:05 +01003723 ret = qcow2_alloc_clusters(blk_bs(blk), 3 * cluster_size);
Kevin Wolfa9420732010-06-11 21:37:37 +02003724 if (ret < 0) {
Max Reitz3ef6c402013-09-05 09:40:43 +02003725 error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
3726 "header and refcount table");
Kevin Wolfa9420732010-06-11 21:37:37 +02003727 goto out;
3728
3729 } else if (ret != 0) {
3730 error_report("Huh, first cluster in empty image is already in use?");
3731 abort();
3732 }
3733
Kevin Wolf9b890bd2019-01-15 19:02:40 +01003734 /* Set the external data file if necessary */
3735 if (data_bs) {
3736 BDRVQcow2State *s = blk_bs(blk)->opaque;
3737 s->image_data_file = g_strdup(data_bs->filename);
3738 }
3739
Kevin Wolfb527c9b2015-12-02 18:34:39 +01003740 /* Create a full header (including things like feature table) */
Kevin Wolf23588792016-03-08 15:57:05 +01003741 ret = qcow2_update_header(blk_bs(blk));
Kevin Wolfb527c9b2015-12-02 18:34:39 +01003742 if (ret < 0) {
3743 error_setg_errno(errp, -ret, "Could not update qcow2 header");
3744 goto out;
3745 }
3746
Kevin Wolfa9420732010-06-11 21:37:37 +02003747 /* Okay, now that we have a valid image, let's give it the right size */
Alberto Faria38505e22022-10-13 14:37:06 +02003748 ret = blk_co_truncate(blk, qcow2_opts->size, false,
3749 qcow2_opts->preallocation, 0, errp);
Kevin Wolfa9420732010-06-11 21:37:37 +02003750 if (ret < 0) {
Max Reitzed3d2ec2017-03-28 22:51:27 +02003751 error_prepend(errp, "Could not resize image: ");
Kevin Wolfa9420732010-06-11 21:37:37 +02003752 goto out;
3753 }
3754
Eric Blakea951a632020-03-24 12:42:30 -05003755 /* Want a backing file? There you go. */
Markus Armbruster54fde4f2022-11-04 17:06:52 +01003756 if (qcow2_opts->backing_file) {
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003757 const char *backing_format = NULL;
3758
3759 if (qcow2_opts->has_backing_fmt) {
3760 backing_format = BlockdevDriver_str(qcow2_opts->backing_fmt);
3761 }
3762
3763 ret = bdrv_change_backing_file(blk_bs(blk), qcow2_opts->backing_file,
Eric Blakee54ee1b2020-07-06 15:39:53 -05003764 backing_format, false);
Kevin Wolfa9420732010-06-11 21:37:37 +02003765 if (ret < 0) {
Max Reitz3ef6c402013-09-05 09:40:43 +02003766 error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003767 "with format '%s'", qcow2_opts->backing_file,
3768 backing_format);
Kevin Wolfa9420732010-06-11 21:37:37 +02003769 goto out;
3770 }
3771 }
3772
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01003773 /* Want encryption? There you go. */
Markus Armbruster54fde4f2022-11-04 17:06:52 +01003774 if (qcow2_opts->encrypt) {
Kevin Wolf60900b72018-01-10 17:55:16 +01003775 ret = qcow2_set_up_encryption(blk_bs(blk), qcow2_opts->encrypt, errp);
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01003776 if (ret < 0) {
3777 goto out;
3778 }
3779 }
3780
Kevin Wolf23588792016-03-08 15:57:05 +01003781 blk_unref(blk);
3782 blk = NULL;
Max Reitzba2ab2f2013-10-24 20:35:06 +02003783
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01003784 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3785 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3786 * have to setup decryption context. We're not doing any I/O on the top
3787 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3788 * not have effect.
3789 */
Max Reitze6641712015-08-26 19:47:48 +02003790 options = qdict_new();
Eric Blake46f5ac22017-04-27 16:58:17 -05003791 qdict_put_str(options, "driver", "qcow2");
Kevin Wolfcbf2b7c2018-01-09 17:35:26 +01003792 qdict_put_str(options, "file", bs->node_name);
Kevin Wolfdcc98682019-01-14 16:57:27 +01003793 if (data_bs) {
3794 qdict_put_str(options, "data-file", data_bs->node_name);
3795 }
Kevin Wolfcbf2b7c2018-01-09 17:35:26 +01003796 blk = blk_new_open(NULL, NULL, options,
Daniel P. Berrangeb25b3872017-06-23 17:24:10 +01003797 BDRV_O_RDWR | BDRV_O_NO_BACKING | BDRV_O_NO_IO,
Markus Armbrusteraf175e82020-07-07 18:06:03 +02003798 errp);
Kevin Wolf23588792016-03-08 15:57:05 +01003799 if (blk == NULL) {
Kevin Wolf23588792016-03-08 15:57:05 +01003800 ret = -EIO;
Max Reitzba2ab2f2013-10-24 20:35:06 +02003801 goto out;
3802 }
3803
Kevin Wolfa9420732010-06-11 21:37:37 +02003804 ret = 0;
3805out:
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003806 blk_unref(blk);
3807 bdrv_unref(bs);
Kevin Wolfdcc98682019-01-14 16:57:27 +01003808 bdrv_unref(data_bs);
Kevin Wolfa9420732010-06-11 21:37:37 +02003809 return ret;
3810}
Kevin Wolfde5f3f42010-05-07 12:43:45 +02003811
Maxim Levitskyb92902d2020-03-26 03:12:17 +02003812static int coroutine_fn qcow2_co_create_opts(BlockDriver *drv,
3813 const char *filename,
3814 QemuOpts *opts,
Stefan Hajnocziefc75e22018-01-18 13:43:45 +01003815 Error **errp)
Kevin Wolfde5f3f42010-05-07 12:43:45 +02003816{
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003817 BlockdevCreateOptions *create_options = NULL;
Markus Armbruster92adf9d2018-06-14 21:14:32 +02003818 QDict *qdict;
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003819 Visitor *v;
Kevin Wolfcbf2b7c2018-01-09 17:35:26 +01003820 BlockDriverState *bs = NULL;
Kevin Wolf9b890bd2019-01-15 19:02:40 +01003821 BlockDriverState *data_bs = NULL;
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003822 const char *val;
Max Reitz3ef6c402013-09-05 09:40:43 +02003823 int ret;
Kevin Wolfde5f3f42010-05-07 12:43:45 +02003824
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003825 /* Only the keyval visitor supports the dotted syntax needed for
3826 * encryption, so go through a QDict before getting a QAPI type. Ignore
3827 * options meant for the protocol layer so that the visitor doesn't
3828 * complain. */
3829 qdict = qemu_opts_to_qdict_filtered(opts, NULL, bdrv_qcow2.create_opts,
3830 true);
3831
3832 /* Handle encryption options */
3833 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT);
3834 if (val && !strcmp(val, "on")) {
3835 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT, "qcow");
3836 } else if (val && !strcmp(val, "off")) {
3837 qdict_del(qdict, BLOCK_OPT_ENCRYPT);
3838 }
3839
3840 val = qdict_get_try_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT);
3841 if (val && !strcmp(val, "aes")) {
3842 qdict_put_str(qdict, BLOCK_OPT_ENCRYPT_FORMAT, "qcow");
3843 }
3844
3845 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3846 * version=v2/v3 below. */
3847 val = qdict_get_try_str(qdict, BLOCK_OPT_COMPAT_LEVEL);
3848 if (val && !strcmp(val, "0.10")) {
3849 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v2");
3850 } else if (val && !strcmp(val, "1.1")) {
3851 qdict_put_str(qdict, BLOCK_OPT_COMPAT_LEVEL, "v3");
3852 }
3853
3854 /* Change legacy command line options into QMP ones */
3855 static const QDictRenames opt_renames[] = {
3856 { BLOCK_OPT_BACKING_FILE, "backing-file" },
3857 { BLOCK_OPT_BACKING_FMT, "backing-fmt" },
3858 { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" },
3859 { BLOCK_OPT_LAZY_REFCOUNTS, "lazy-refcounts" },
Alberto Garcia7be20252020-07-10 18:13:13 +02003860 { BLOCK_OPT_EXTL2, "extended-l2" },
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003861 { BLOCK_OPT_REFCOUNT_BITS, "refcount-bits" },
3862 { BLOCK_OPT_ENCRYPT, BLOCK_OPT_ENCRYPT_FORMAT },
3863 { BLOCK_OPT_COMPAT_LEVEL, "version" },
Kevin Wolf6c3944d2019-02-22 14:29:38 +01003864 { BLOCK_OPT_DATA_FILE_RAW, "data-file-raw" },
Denis Plotnikov572ad972020-05-07 11:25:18 +03003865 { BLOCK_OPT_COMPRESSION_TYPE, "compression-type" },
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003866 { NULL, NULL },
3867 };
3868
3869 if (!qdict_rename_keys(qdict, opt_renames, errp)) {
Kevin Wolf29ca9e42018-01-09 19:44:33 +01003870 ret = -EINVAL;
3871 goto finish;
3872 }
Kevin Wolf60900b72018-01-10 17:55:16 +01003873
Kevin Wolfcbf2b7c2018-01-09 17:35:26 +01003874 /* Create and open the file (protocol layer) */
Emanuele Giuseppe Esposito2475a0d2022-11-28 09:23:31 -05003875 ret = bdrv_co_create_file(filename, opts, errp);
Kevin Wolfcbf2b7c2018-01-09 17:35:26 +01003876 if (ret < 0) {
3877 goto finish;
3878 }
3879
3880 bs = bdrv_open(filename, NULL, NULL,
3881 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
3882 if (bs == NULL) {
3883 ret = -EIO;
3884 goto finish;
3885 }
3886
Kevin Wolf9b890bd2019-01-15 19:02:40 +01003887 /* Create and open an external data file (protocol layer) */
3888 val = qdict_get_try_str(qdict, BLOCK_OPT_DATA_FILE);
3889 if (val) {
Emanuele Giuseppe Esposito2475a0d2022-11-28 09:23:31 -05003890 ret = bdrv_co_create_file(val, opts, errp);
Kevin Wolf9b890bd2019-01-15 19:02:40 +01003891 if (ret < 0) {
3892 goto finish;
3893 }
3894
3895 data_bs = bdrv_open(val, NULL, NULL,
3896 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
3897 errp);
3898 if (data_bs == NULL) {
3899 ret = -EIO;
3900 goto finish;
3901 }
3902
3903 qdict_del(qdict, BLOCK_OPT_DATA_FILE);
3904 qdict_put_str(qdict, "data-file", data_bs->node_name);
3905 }
3906
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003907 /* Set 'driver' and 'node' options */
3908 qdict_put_str(qdict, "driver", "qcow2");
3909 qdict_put_str(qdict, "file", bs->node_name);
3910
3911 /* Now get the QAPI type BlockdevCreateOptions */
Markus Armbrusteraf910622018-06-14 21:14:33 +02003912 v = qobject_input_visitor_new_flat_confused(qdict, errp);
3913 if (!v) {
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003914 ret = -EINVAL;
3915 goto finish;
3916 }
3917
Markus Armbrusterb11a0932020-07-07 18:06:07 +02003918 visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003919 visit_free(v);
Markus Armbrusterb11a0932020-07-07 18:06:07 +02003920 if (!create_options) {
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003921 ret = -EINVAL;
3922 goto finish;
3923 }
3924
3925 /* Silently round up size */
3926 create_options->u.qcow2.size = ROUND_UP(create_options->u.qcow2.size,
3927 BDRV_SECTOR_SIZE);
3928
Kevin Wolfcbf2b7c2018-01-09 17:35:26 +01003929 /* Create the qcow2 image (format layer) */
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003930 ret = qcow2_co_create(create_options, errp);
Maxim Levitsky6094cbe2020-12-17 19:09:04 +02003931finish:
Kevin Wolfcbf2b7c2018-01-09 17:35:26 +01003932 if (ret < 0) {
Maxim Levitsky6094cbe2020-12-17 19:09:04 +02003933 bdrv_co_delete_file_noerr(bs);
3934 bdrv_co_delete_file_noerr(data_bs);
3935 } else {
3936 ret = 0;
Kevin Wolfcbf2b7c2018-01-09 17:35:26 +01003937 }
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +08003938
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003939 qobject_unref(qdict);
Kevin Wolfcbf2b7c2018-01-09 17:35:26 +01003940 bdrv_unref(bs);
Kevin Wolf9b890bd2019-01-15 19:02:40 +01003941 bdrv_unref(data_bs);
Kevin Wolfb76b4f62018-01-11 16:18:08 +01003942 qapi_free_BlockdevCreateOptions(create_options);
Max Reitz3ef6c402013-09-05 09:40:43 +02003943 return ret;
Kevin Wolfde5f3f42010-05-07 12:43:45 +02003944}
3945
Denis V. Lunev2928abc2016-05-11 10:00:14 +03003946
Eric Blakef06f6b62017-10-11 22:47:00 -05003947static bool is_zero(BlockDriverState *bs, int64_t offset, int64_t bytes)
Denis V. Lunev2928abc2016-05-11 10:00:14 +03003948{
Eric Blake31826642017-10-11 22:47:08 -05003949 int64_t nr;
3950 int res;
Eric Blakef06f6b62017-10-11 22:47:00 -05003951
3952 /* Clamp to image length, before checking status of underlying sectors */
Eric Blake8cbf74b2017-10-11 22:47:19 -05003953 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
3954 bytes = bs->total_sectors * BDRV_SECTOR_SIZE - offset;
Eric Blakefbaa6bb2017-05-06 19:05:50 -05003955 }
3956
Eric Blakef06f6b62017-10-11 22:47:00 -05003957 if (!bytes) {
Eric Blakeebb718a2016-05-25 21:48:49 -06003958 return true;
3959 }
Vladimir Sementsov-Ogievskiy67c095c2020-09-24 22:39:59 +03003960
3961 /*
3962 * bdrv_block_status_above doesn't merge different types of zeros, for
3963 * example, zeros which come from the region which is unallocated in
3964 * the whole backing chain, and zeros which come because of a short
3965 * backing file. So, we need a loop.
3966 */
3967 do {
3968 res = bdrv_block_status_above(bs, NULL, offset, bytes, &nr, NULL, NULL);
3969 offset += nr;
3970 bytes -= nr;
3971 } while (res >= 0 && (res & BDRV_BLOCK_ZERO) && nr && bytes);
3972
3973 return res >= 0 && (res & BDRV_BLOCK_ZERO) && bytes == 0;
Denis V. Lunev2928abc2016-05-11 10:00:14 +03003974}
3975
Eric Blake5544b592016-06-01 15:10:06 -06003976static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
Vladimir Sementsov-Ogievskiyf34b2bc2021-09-03 13:28:03 +03003977 int64_t offset, int64_t bytes, BdrvRequestFlags flags)
Kevin Wolf621f0582012-03-20 15:12:58 +01003978{
3979 int ret;
Kevin Wolfff991292015-09-07 17:12:56 +02003980 BDRVQcow2State *s = bs->opaque;
Kevin Wolf621f0582012-03-20 15:12:58 +01003981
Alberto Garciaa6841a22020-07-10 18:13:10 +02003982 uint32_t head = offset_into_subcluster(s, offset);
3983 uint32_t tail = ROUND_UP(offset + bytes, s->subcluster_size) -
3984 (offset + bytes);
Denis V. Lunev2928abc2016-05-11 10:00:14 +03003985
Manos Pitsidianakisf5a5ca72017-06-09 13:18:08 +03003986 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset, bytes);
3987 if (offset + bytes == bs->total_sectors * BDRV_SECTOR_SIZE) {
Eric Blakefbaa6bb2017-05-06 19:05:50 -05003988 tail = 0;
3989 }
Denis V. Lunev5a64e942016-05-25 21:48:47 -06003990
Eric Blakeebb718a2016-05-25 21:48:49 -06003991 if (head || tail) {
Eric Blakeebb718a2016-05-25 21:48:49 -06003992 uint64_t off;
Kevin Wolfecfe1862016-05-31 16:13:07 +02003993 unsigned int nr;
Alberto Garcia10dabdc52020-07-10 18:13:00 +02003994 QCow2SubclusterType type;
Denis V. Lunev2928abc2016-05-11 10:00:14 +03003995
Alberto Garciaa6841a22020-07-10 18:13:10 +02003996 assert(head + bytes + tail <= s->subcluster_size);
Denis V. Lunev2928abc2016-05-11 10:00:14 +03003997
Eric Blakeebb718a2016-05-25 21:48:49 -06003998 /* check whether remainder of cluster already reads as zero */
Eric Blakef06f6b62017-10-11 22:47:00 -05003999 if (!(is_zero(bs, offset - head, head) &&
Alberto Garciaa6841a22020-07-10 18:13:10 +02004000 is_zero(bs, offset + bytes, tail))) {
Denis V. Lunev2928abc2016-05-11 10:00:14 +03004001 return -ENOTSUP;
4002 }
4003
Denis V. Lunev2928abc2016-05-11 10:00:14 +03004004 qemu_co_mutex_lock(&s->lock);
4005 /* We can have new write after previous check */
Alberto Garciaa6841a22020-07-10 18:13:10 +02004006 offset -= head;
4007 bytes = s->subcluster_size;
4008 nr = s->subcluster_size;
Alberto Garciaca4a0bb2020-07-10 18:12:59 +02004009 ret = qcow2_get_host_offset(bs, offset, &nr, &off, &type);
4010 if (ret < 0 ||
Alberto Garcia10dabdc52020-07-10 18:13:00 +02004011 (type != QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN &&
Alberto Garcia97490a12020-07-10 18:13:01 +02004012 type != QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC &&
Alberto Garcia10dabdc52020-07-10 18:13:00 +02004013 type != QCOW2_SUBCLUSTER_ZERO_PLAIN &&
4014 type != QCOW2_SUBCLUSTER_ZERO_ALLOC)) {
Denis V. Lunev2928abc2016-05-11 10:00:14 +03004015 qemu_co_mutex_unlock(&s->lock);
Alberto Garcia580384d2020-09-09 14:37:39 +02004016 return ret < 0 ? ret : -ENOTSUP;
Denis V. Lunev2928abc2016-05-11 10:00:14 +03004017 }
4018 } else {
4019 qemu_co_mutex_lock(&s->lock);
Kevin Wolf621f0582012-03-20 15:12:58 +01004020 }
4021
Manos Pitsidianakisf5a5ca72017-06-09 13:18:08 +03004022 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset, bytes);
Denis V. Lunev5a64e942016-05-25 21:48:47 -06004023
Alberto Garciaa6841a22020-07-10 18:13:10 +02004024 /* Whatever is left can use real zero subclusters */
4025 ret = qcow2_subcluster_zeroize(bs, offset, bytes, flags);
Kevin Wolf621f0582012-03-20 15:12:58 +01004026 qemu_co_mutex_unlock(&s->lock);
4027
4028 return ret;
4029}
4030
Eric Blake82e8a782016-07-15 17:23:03 -06004031static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
Vladimir Sementsov-Ogievskiy0c802282021-09-03 13:28:06 +03004032 int64_t offset, int64_t bytes)
Kevin Wolf5ea929e2011-01-26 16:56:48 +01004033{
Paolo Bonzini6db39ae2011-10-20 13:16:25 +02004034 int ret;
Kevin Wolfff991292015-09-07 17:12:56 +02004035 BDRVQcow2State *s = bs->opaque;
Paolo Bonzini6db39ae2011-10-20 13:16:25 +02004036
Alberto Garcia80f5c012020-03-31 13:43:45 +02004037 /* If the image does not support QCOW_OFLAG_ZERO then discarding
4038 * clusters could expose stale data from the backing file. */
4039 if (s->qcow_version < 3 && bs->backing) {
4040 return -ENOTSUP;
4041 }
4042
Manos Pitsidianakisf5a5ca72017-06-09 13:18:08 +03004043 if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
4044 assert(bytes < s->cluster_size);
Eric Blake048c5fd2017-04-06 20:37:09 -05004045 /* Ignore partial clusters, except for the special case of the
4046 * complete partial cluster at the end of an unaligned file */
4047 if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
Manos Pitsidianakisf5a5ca72017-06-09 13:18:08 +03004048 offset + bytes != bs->total_sectors * BDRV_SECTOR_SIZE) {
Eric Blake048c5fd2017-04-06 20:37:09 -05004049 return -ENOTSUP;
4050 }
Eric Blake49228d12016-11-17 14:13:57 -06004051 }
4052
Paolo Bonzini6db39ae2011-10-20 13:16:25 +02004053 qemu_co_mutex_lock(&s->lock);
Manos Pitsidianakisf5a5ca72017-06-09 13:18:08 +03004054 ret = qcow2_cluster_discard(bs, offset, bytes, QCOW2_DISCARD_REQUEST,
Eric Blaked2cb36a2017-05-06 19:05:52 -05004055 false);
Paolo Bonzini6db39ae2011-10-20 13:16:25 +02004056 qemu_co_mutex_unlock(&s->lock);
4057 return ret;
Kevin Wolf5ea929e2011-01-26 16:56:48 +01004058}
4059
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004060static int coroutine_fn
4061qcow2_co_copy_range_from(BlockDriverState *bs,
Vladimir Sementsov-Ogievskiy48535042021-09-03 13:28:01 +03004062 BdrvChild *src, int64_t src_offset,
4063 BdrvChild *dst, int64_t dst_offset,
4064 int64_t bytes, BdrvRequestFlags read_flags,
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03004065 BdrvRequestFlags write_flags)
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004066{
4067 BDRVQcow2State *s = bs->opaque;
4068 int ret;
4069 unsigned int cur_bytes; /* number of bytes in current iteration */
4070 BdrvChild *child = NULL;
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03004071 BdrvRequestFlags cur_write_flags;
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004072
4073 assert(!bs->encrypted);
4074 qemu_co_mutex_lock(&s->lock);
4075
4076 while (bytes != 0) {
4077 uint64_t copy_offset = 0;
Alberto Garcia10dabdc52020-07-10 18:13:00 +02004078 QCow2SubclusterType type;
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004079 /* prepare next request */
4080 cur_bytes = MIN(bytes, INT_MAX);
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03004081 cur_write_flags = write_flags;
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004082
Alberto Garciaca4a0bb2020-07-10 18:12:59 +02004083 ret = qcow2_get_host_offset(bs, src_offset, &cur_bytes,
4084 &copy_offset, &type);
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004085 if (ret < 0) {
4086 goto out;
4087 }
4088
Alberto Garciaca4a0bb2020-07-10 18:12:59 +02004089 switch (type) {
Alberto Garcia10dabdc52020-07-10 18:13:00 +02004090 case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
Alberto Garcia97490a12020-07-10 18:13:01 +02004091 case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004092 if (bs->backing && bs->backing->bs) {
4093 int64_t backing_length = bdrv_getlength(bs->backing->bs);
4094 if (src_offset >= backing_length) {
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03004095 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004096 } else {
4097 child = bs->backing;
4098 cur_bytes = MIN(cur_bytes, backing_length - src_offset);
4099 copy_offset = src_offset;
4100 }
4101 } else {
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03004102 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004103 }
4104 break;
4105
Alberto Garcia10dabdc52020-07-10 18:13:00 +02004106 case QCOW2_SUBCLUSTER_ZERO_PLAIN:
4107 case QCOW2_SUBCLUSTER_ZERO_ALLOC:
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03004108 cur_write_flags |= BDRV_REQ_ZERO_WRITE;
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004109 break;
4110
Alberto Garcia10dabdc52020-07-10 18:13:00 +02004111 case QCOW2_SUBCLUSTER_COMPRESSED:
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004112 ret = -ENOTSUP;
4113 goto out;
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004114
Alberto Garcia10dabdc52020-07-10 18:13:00 +02004115 case QCOW2_SUBCLUSTER_NORMAL:
Kevin Wolf966b0002019-01-15 20:39:06 +01004116 child = s->data_file;
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004117 break;
4118
4119 default:
4120 abort();
4121 }
4122 qemu_co_mutex_unlock(&s->lock);
4123 ret = bdrv_co_copy_range_from(child,
4124 copy_offset,
4125 dst, dst_offset,
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03004126 cur_bytes, read_flags, cur_write_flags);
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004127 qemu_co_mutex_lock(&s->lock);
4128 if (ret < 0) {
4129 goto out;
4130 }
4131
4132 bytes -= cur_bytes;
4133 src_offset += cur_bytes;
4134 dst_offset += cur_bytes;
4135 }
4136 ret = 0;
4137
4138out:
4139 qemu_co_mutex_unlock(&s->lock);
4140 return ret;
4141}
4142
4143static int coroutine_fn
4144qcow2_co_copy_range_to(BlockDriverState *bs,
Vladimir Sementsov-Ogievskiy48535042021-09-03 13:28:01 +03004145 BdrvChild *src, int64_t src_offset,
4146 BdrvChild *dst, int64_t dst_offset,
4147 int64_t bytes, BdrvRequestFlags read_flags,
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03004148 BdrvRequestFlags write_flags)
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004149{
4150 BDRVQcow2State *s = bs->opaque;
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004151 int ret;
4152 unsigned int cur_bytes; /* number of sectors in current iteration */
Alberto Garciabfd09892020-09-11 16:09:42 +02004153 uint64_t host_offset;
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004154 QCowL2Meta *l2meta = NULL;
4155
4156 assert(!bs->encrypted);
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004157
4158 qemu_co_mutex_lock(&s->lock);
4159
4160 while (bytes != 0) {
4161
4162 l2meta = NULL;
4163
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004164 cur_bytes = MIN(bytes, INT_MAX);
4165
4166 /* TODO:
4167 * If src->bs == dst->bs, we could simply copy by incrementing
4168 * the refcnt, without copying user data.
4169 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
Alberto Garciabfd09892020-09-11 16:09:42 +02004170 ret = qcow2_alloc_host_offset(bs, dst_offset, &cur_bytes,
4171 &host_offset, &l2meta);
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004172 if (ret < 0) {
4173 goto fail;
4174 }
4175
Alberto Garciabfd09892020-09-11 16:09:42 +02004176 ret = qcow2_pre_write_overlap_check(bs, 0, host_offset, cur_bytes,
4177 true);
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004178 if (ret < 0) {
4179 goto fail;
4180 }
4181
4182 qemu_co_mutex_unlock(&s->lock);
Alberto Garciabfd09892020-09-11 16:09:42 +02004183 ret = bdrv_co_copy_range_to(src, src_offset, s->data_file, host_offset,
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03004184 cur_bytes, read_flags, write_flags);
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004185 qemu_co_mutex_lock(&s->lock);
4186 if (ret < 0) {
4187 goto fail;
4188 }
4189
4190 ret = qcow2_handle_l2meta(bs, &l2meta, true);
4191 if (ret) {
4192 goto fail;
4193 }
4194
4195 bytes -= cur_bytes;
Fam Zhenge06f4632018-06-29 14:03:26 +08004196 src_offset += cur_bytes;
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004197 dst_offset += cur_bytes;
4198 }
4199 ret = 0;
4200
4201fail:
4202 qcow2_handle_l2meta(bs, &l2meta, false);
4203
4204 qemu_co_mutex_unlock(&s->lock);
4205
Fam Zhengfd9fcd32018-06-01 17:26:42 +08004206 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
4207
4208 return ret;
4209}
4210
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004211static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,
Max Reitzc80d8b02019-09-18 11:51:40 +02004212 bool exact, PreallocMode prealloc,
Kevin Wolf92b92792020-04-24 14:54:39 +02004213 BdrvRequestFlags flags, Error **errp)
Stefan Hajnoczi419b19d2010-04-28 11:36:11 +01004214{
Kevin Wolfff991292015-09-07 17:12:56 +02004215 BDRVQcow2State *s = bs->opaque;
Max Reitz95b98f32017-06-13 22:21:02 +02004216 uint64_t old_length;
Kevin Wolf2cf7cfa2013-05-14 16:14:33 +02004217 int64_t new_l1_size;
4218 int ret;
Leonid Bloch45b49492018-09-26 19:04:45 +03004219 QDict *options;
Stefan Hajnoczi419b19d2010-04-28 11:36:11 +01004220
Max Reitz772d1f92017-06-13 22:21:05 +02004221 if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_METADATA &&
4222 prealloc != PREALLOC_MODE_FALLOC && prealloc != PREALLOC_MODE_FULL)
4223 {
Max Reitz8243ccb2017-06-13 22:20:52 +02004224 error_setg(errp, "Unsupported preallocation mode '%s'",
Markus Armbruster977c7362017-08-24 10:46:08 +02004225 PreallocMode_str(prealloc));
Max Reitz8243ccb2017-06-13 22:20:52 +02004226 return -ENOTSUP;
4227 }
4228
Alberto Garcia3afea402020-01-18 20:09:30 +01004229 if (!QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)) {
4230 error_setg(errp, "The new size must be a multiple of %u",
4231 (unsigned) BDRV_SECTOR_SIZE);
Stefan Hajnoczi419b19d2010-04-28 11:36:11 +01004232 return -EINVAL;
4233 }
4234
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004235 qemu_co_mutex_lock(&s->lock);
4236
Eric Blake7fa140a2020-04-28 14:26:47 -05004237 /*
4238 * Even though we store snapshot size for all images, it was not
4239 * required until v3, so it is not safe to proceed for v2.
4240 */
4241 if (s->nb_snapshots && s->qcow_version < 3) {
4242 error_setg(errp, "Can't resize a v2 image which has snapshots");
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004243 ret = -ENOTSUP;
4244 goto fail;
Stefan Hajnoczi419b19d2010-04-28 11:36:11 +01004245 }
4246
Eric Blakeee1244a2020-04-28 14:26:48 -05004247 /* See qcow2-bitmap.c for which bitmap scenarios prevent a resize. */
John Snowd19c6b32019-03-11 21:51:46 +03004248 if (qcow2_truncate_bitmaps_check(bs, errp)) {
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004249 ret = -ENOTSUP;
4250 goto fail;
Vladimir Sementsov-Ogievskiy88ddffa2017-06-28 15:05:08 +03004251 }
4252
Leonid Blochbd016b92018-09-26 19:04:47 +03004253 old_length = bs->total_sectors * BDRV_SECTOR_SIZE;
Stefan Hajnoczi419b19d2010-04-28 11:36:11 +01004254 new_l1_size = size_to_l1(s, offset);
Pavel Butsykin46b732c2017-09-18 15:42:29 +03004255
4256 if (offset < old_length) {
Pavel Butsykin163bc392017-09-29 15:16:13 +03004257 int64_t last_cluster, old_file_size;
Pavel Butsykin46b732c2017-09-18 15:42:29 +03004258 if (prealloc != PREALLOC_MODE_OFF) {
4259 error_setg(errp,
4260 "Preallocation can't be used for shrinking an image");
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004261 ret = -EINVAL;
4262 goto fail;
Pavel Butsykin46b732c2017-09-18 15:42:29 +03004263 }
4264
4265 ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size),
4266 old_length - ROUND_UP(offset,
4267 s->cluster_size),
4268 QCOW2_DISCARD_ALWAYS, true);
4269 if (ret < 0) {
4270 error_setg_errno(errp, -ret, "Failed to discard cropped clusters");
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004271 goto fail;
Pavel Butsykin46b732c2017-09-18 15:42:29 +03004272 }
4273
4274 ret = qcow2_shrink_l1_table(bs, new_l1_size);
4275 if (ret < 0) {
4276 error_setg_errno(errp, -ret,
4277 "Failed to reduce the number of L2 tables");
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004278 goto fail;
Pavel Butsykin46b732c2017-09-18 15:42:29 +03004279 }
4280
4281 ret = qcow2_shrink_reftable(bs);
4282 if (ret < 0) {
4283 error_setg_errno(errp, -ret,
4284 "Failed to discard unused refblocks");
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004285 goto fail;
Pavel Butsykin46b732c2017-09-18 15:42:29 +03004286 }
Pavel Butsykin163bc392017-09-29 15:16:13 +03004287
4288 old_file_size = bdrv_getlength(bs->file->bs);
4289 if (old_file_size < 0) {
4290 error_setg_errno(errp, -old_file_size,
4291 "Failed to inquire current file length");
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004292 ret = old_file_size;
4293 goto fail;
Pavel Butsykin163bc392017-09-29 15:16:13 +03004294 }
4295 last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4296 if (last_cluster < 0) {
4297 error_setg_errno(errp, -last_cluster,
4298 "Failed to find the last cluster");
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004299 ret = last_cluster;
4300 goto fail;
Pavel Butsykin163bc392017-09-29 15:16:13 +03004301 }
4302 if ((last_cluster + 1) * s->cluster_size < old_file_size) {
Max Reitz233521b2017-10-09 17:54:31 +02004303 Error *local_err = NULL;
4304
Max Reitze61a28a2019-09-18 11:51:42 +02004305 /*
4306 * Do not pass @exact here: It will not help the user if
4307 * we get an error here just because they wanted to shrink
4308 * their qcow2 image (on a block device) with qemu-img.
4309 * (And on the qcow2 layer, the @exact requirement is
4310 * always fulfilled, so there is no need to pass it on.)
4311 */
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004312 bdrv_co_truncate(bs->file, (last_cluster + 1) * s->cluster_size,
Kevin Wolf7b8e4852020-04-24 14:54:40 +02004313 false, PREALLOC_MODE_OFF, 0, &local_err);
Max Reitz233521b2017-10-09 17:54:31 +02004314 if (local_err) {
4315 warn_reportf_err(local_err,
4316 "Failed to truncate the tail of the image: ");
Pavel Butsykin163bc392017-09-29 15:16:13 +03004317 }
4318 }
Pavel Butsykin46b732c2017-09-18 15:42:29 +03004319 } else {
4320 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
4321 if (ret < 0) {
4322 error_setg_errno(errp, -ret, "Failed to grow the L1 table");
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004323 goto fail;
Pavel Butsykin46b732c2017-09-18 15:42:29 +03004324 }
Max Reitz48410822021-03-26 15:55:08 +01004325
4326 if (data_file_is_raw(bs) && prealloc == PREALLOC_MODE_OFF) {
4327 /*
4328 * When creating a qcow2 image with data-file-raw, we enforce
4329 * at least prealloc=metadata, so that the L1/L2 tables are
4330 * fully allocated and reading from the data file will return
4331 * the same data as reading from the qcow2 image. When the
4332 * image is grown, we must consequently preallocate the
4333 * metadata structures to cover the added area.
4334 */
4335 prealloc = PREALLOC_MODE_METADATA;
4336 }
Stefan Hajnoczi419b19d2010-04-28 11:36:11 +01004337 }
4338
Max Reitz95b98f32017-06-13 22:21:02 +02004339 switch (prealloc) {
4340 case PREALLOC_MODE_OFF:
Kevin Wolf718c0fc2019-04-15 16:34:30 +02004341 if (has_data_file(bs)) {
Max Reitze61a28a2019-09-18 11:51:42 +02004342 /*
4343 * If the caller wants an exact resize, the external data
4344 * file should be resized to the exact target size, too,
4345 * so we pass @exact here.
4346 */
Kevin Wolf7b8e4852020-04-24 14:54:40 +02004347 ret = bdrv_co_truncate(s->data_file, offset, exact, prealloc, 0,
4348 errp);
Kevin Wolf718c0fc2019-04-15 16:34:30 +02004349 if (ret < 0) {
4350 goto fail;
4351 }
4352 }
Max Reitz95b98f32017-06-13 22:21:02 +02004353 break;
4354
4355 case PREALLOC_MODE_METADATA:
Kevin Wolf718c0fc2019-04-15 16:34:30 +02004356 ret = preallocate_co(bs, old_length, offset, prealloc, errp);
Max Reitz95b98f32017-06-13 22:21:02 +02004357 if (ret < 0) {
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004358 goto fail;
Max Reitz95b98f32017-06-13 22:21:02 +02004359 }
4360 break;
4361
Max Reitz772d1f92017-06-13 22:21:05 +02004362 case PREALLOC_MODE_FALLOC:
4363 case PREALLOC_MODE_FULL:
4364 {
4365 int64_t allocation_start, host_offset, guest_offset;
4366 int64_t clusters_allocated;
Max Reitz4b96fa32020-05-05 16:18:01 +02004367 int64_t old_file_size, last_cluster, new_file_size;
Max Reitz772d1f92017-06-13 22:21:05 +02004368 uint64_t nb_new_data_clusters, nb_new_l2_tables;
Alberto Garcia40dee942020-07-10 18:13:12 +02004369 bool subclusters_need_allocation = false;
Max Reitz772d1f92017-06-13 22:21:05 +02004370
Kevin Wolf966b0002019-01-15 20:39:06 +01004371 /* With a data file, preallocation means just allocating the metadata
4372 * and forwarding the truncate request to the data file */
4373 if (has_data_file(bs)) {
Kevin Wolf718c0fc2019-04-15 16:34:30 +02004374 ret = preallocate_co(bs, old_length, offset, prealloc, errp);
Kevin Wolf966b0002019-01-15 20:39:06 +01004375 if (ret < 0) {
Kevin Wolf966b0002019-01-15 20:39:06 +01004376 goto fail;
4377 }
4378 break;
4379 }
4380
Max Reitz772d1f92017-06-13 22:21:05 +02004381 old_file_size = bdrv_getlength(bs->file->bs);
4382 if (old_file_size < 0) {
4383 error_setg_errno(errp, -old_file_size,
4384 "Failed to inquire current file length");
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004385 ret = old_file_size;
4386 goto fail;
Max Reitz772d1f92017-06-13 22:21:05 +02004387 }
Max Reitz4b96fa32020-05-05 16:18:01 +02004388
4389 last_cluster = qcow2_get_last_cluster(bs, old_file_size);
4390 if (last_cluster >= 0) {
4391 old_file_size = (last_cluster + 1) * s->cluster_size;
4392 } else {
4393 old_file_size = ROUND_UP(old_file_size, s->cluster_size);
4394 }
Max Reitz772d1f92017-06-13 22:21:05 +02004395
Alberto Garciaa5675f32020-06-17 16:00:36 +02004396 nb_new_data_clusters = (ROUND_UP(offset, s->cluster_size) -
4397 start_of_cluster(s, old_length)) >> s->cluster_bits;
Max Reitz772d1f92017-06-13 22:21:05 +02004398
4399 /* This is an overestimation; we will not actually allocate space for
4400 * these in the file but just make sure the new refcount structures are
4401 * able to cover them so we will not have to allocate new refblocks
4402 * while entering the data blocks in the potentially new L2 tables.
4403 * (We do not actually care where the L2 tables are placed. Maybe they
4404 * are already allocated or they can be placed somewhere before
4405 * @old_file_size. It does not matter because they will be fully
4406 * allocated automatically, so they do not need to be covered by the
4407 * preallocation. All that matters is that we will not have to allocate
4408 * new refcount structures for them.) */
4409 nb_new_l2_tables = DIV_ROUND_UP(nb_new_data_clusters,
Alberto Garciac8fd8552020-07-10 18:12:54 +02004410 s->cluster_size / l2_entry_size(s));
Max Reitz772d1f92017-06-13 22:21:05 +02004411 /* The cluster range may not be aligned to L2 boundaries, so add one L2
4412 * table for a potential head/tail */
4413 nb_new_l2_tables++;
4414
4415 allocation_start = qcow2_refcount_area(bs, old_file_size,
4416 nb_new_data_clusters +
4417 nb_new_l2_tables,
4418 true, 0, 0);
4419 if (allocation_start < 0) {
4420 error_setg_errno(errp, -allocation_start,
4421 "Failed to resize refcount structures");
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004422 ret = allocation_start;
4423 goto fail;
Max Reitz772d1f92017-06-13 22:21:05 +02004424 }
4425
4426 clusters_allocated = qcow2_alloc_clusters_at(bs, allocation_start,
4427 nb_new_data_clusters);
4428 if (clusters_allocated < 0) {
4429 error_setg_errno(errp, -clusters_allocated,
4430 "Failed to allocate data clusters");
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004431 ret = clusters_allocated;
4432 goto fail;
Max Reitz772d1f92017-06-13 22:21:05 +02004433 }
4434
4435 assert(clusters_allocated == nb_new_data_clusters);
4436
4437 /* Allocate the data area */
4438 new_file_size = allocation_start +
4439 nb_new_data_clusters * s->cluster_size;
Kevin Wolfeb8a0cf2020-04-24 16:27:01 +02004440 /*
4441 * Image file grows, so @exact does not matter.
4442 *
4443 * If we need to zero out the new area, try first whether the protocol
4444 * driver can already take care of this.
4445 */
4446 if (flags & BDRV_REQ_ZERO_WRITE) {
4447 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc,
4448 BDRV_REQ_ZERO_WRITE, NULL);
4449 if (ret >= 0) {
4450 flags &= ~BDRV_REQ_ZERO_WRITE;
Alberto Garcia40dee942020-07-10 18:13:12 +02004451 /* Ensure that we read zeroes and not backing file data */
4452 subclusters_need_allocation = true;
Kevin Wolfeb8a0cf2020-04-24 16:27:01 +02004453 }
4454 } else {
4455 ret = -1;
4456 }
4457 if (ret < 0) {
4458 ret = bdrv_co_truncate(bs->file, new_file_size, false, prealloc, 0,
4459 errp);
4460 }
Max Reitz772d1f92017-06-13 22:21:05 +02004461 if (ret < 0) {
4462 error_prepend(errp, "Failed to resize underlying file: ");
4463 qcow2_free_clusters(bs, allocation_start,
4464 nb_new_data_clusters * s->cluster_size,
4465 QCOW2_DISCARD_OTHER);
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004466 goto fail;
Max Reitz772d1f92017-06-13 22:21:05 +02004467 }
4468
4469 /* Create the necessary L2 entries */
4470 host_offset = allocation_start;
4471 guest_offset = old_length;
4472 while (nb_new_data_clusters) {
Alberto Garcia13bec222018-02-05 16:33:31 +02004473 int64_t nb_clusters = MIN(
4474 nb_new_data_clusters,
4475 s->l2_slice_size - offset_to_l2_slice_index(s, guest_offset));
Alberto Garciaa5675f32020-06-17 16:00:36 +02004476 unsigned cow_start_length = offset_into_cluster(s, guest_offset);
4477 QCowL2Meta allocation;
4478 guest_offset = start_of_cluster(s, guest_offset);
4479 allocation = (QCowL2Meta) {
Max Reitz772d1f92017-06-13 22:21:05 +02004480 .offset = guest_offset,
4481 .alloc_offset = host_offset,
4482 .nb_clusters = nb_clusters,
Alberto Garciaa5675f32020-06-17 16:00:36 +02004483 .cow_start = {
4484 .offset = 0,
4485 .nb_bytes = cow_start_length,
4486 },
4487 .cow_end = {
4488 .offset = nb_clusters << s->cluster_bits,
4489 .nb_bytes = 0,
4490 },
Alberto Garcia40dee942020-07-10 18:13:12 +02004491 .prealloc = !subclusters_need_allocation,
Max Reitz772d1f92017-06-13 22:21:05 +02004492 };
4493 qemu_co_queue_init(&allocation.dependent_requests);
4494
4495 ret = qcow2_alloc_cluster_link_l2(bs, &allocation);
4496 if (ret < 0) {
4497 error_setg_errno(errp, -ret, "Failed to update L2 tables");
4498 qcow2_free_clusters(bs, host_offset,
4499 nb_new_data_clusters * s->cluster_size,
4500 QCOW2_DISCARD_OTHER);
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004501 goto fail;
Max Reitz772d1f92017-06-13 22:21:05 +02004502 }
4503
4504 guest_offset += nb_clusters * s->cluster_size;
4505 host_offset += nb_clusters * s->cluster_size;
4506 nb_new_data_clusters -= nb_clusters;
4507 }
4508 break;
4509 }
4510
Max Reitz95b98f32017-06-13 22:21:02 +02004511 default:
4512 g_assert_not_reached();
4513 }
4514
Kevin Wolff01643f2020-04-24 14:54:42 +02004515 if ((flags & BDRV_REQ_ZERO_WRITE) && offset > old_length) {
Alberto Garciaa6841a22020-07-10 18:13:10 +02004516 uint64_t zero_start = QEMU_ALIGN_UP(old_length, s->subcluster_size);
Kevin Wolff01643f2020-04-24 14:54:42 +02004517
4518 /*
Alberto Garciaa6841a22020-07-10 18:13:10 +02004519 * Use zero clusters as much as we can. qcow2_subcluster_zeroize()
4520 * requires a subcluster-aligned start. The end may be unaligned if
4521 * it is at the end of the image (which it is here).
Kevin Wolff01643f2020-04-24 14:54:42 +02004522 */
Alberto Garciae4d70192020-05-04 17:52:17 +02004523 if (offset > zero_start) {
Alberto Garciaa6841a22020-07-10 18:13:10 +02004524 ret = qcow2_subcluster_zeroize(bs, zero_start, offset - zero_start,
4525 0);
Alberto Garciae4d70192020-05-04 17:52:17 +02004526 if (ret < 0) {
4527 error_setg_errno(errp, -ret, "Failed to zero out new clusters");
4528 goto fail;
4529 }
Kevin Wolff01643f2020-04-24 14:54:42 +02004530 }
4531
4532 /* Write explicit zeros for the unaligned head */
4533 if (zero_start > old_length) {
Alberto Garciae4d70192020-05-04 17:52:17 +02004534 uint64_t len = MIN(zero_start, offset) - old_length;
Kevin Wolff01643f2020-04-24 14:54:42 +02004535 uint8_t *buf = qemu_blockalign0(bs, len);
4536 QEMUIOVector qiov;
4537 qemu_iovec_init_buf(&qiov, buf, len);
4538
4539 qemu_co_mutex_unlock(&s->lock);
4540 ret = qcow2_co_pwritev_part(bs, old_length, len, &qiov, 0, 0);
4541 qemu_co_mutex_lock(&s->lock);
4542
4543 qemu_vfree(buf);
4544 if (ret < 0) {
4545 error_setg_errno(errp, -ret, "Failed to zero out the new area");
4546 goto fail;
4547 }
4548 }
4549 }
4550
Max Reitz95b98f32017-06-13 22:21:02 +02004551 if (prealloc != PREALLOC_MODE_OFF) {
4552 /* Flush metadata before actually changing the image size */
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004553 ret = qcow2_write_caches(bs);
Max Reitz95b98f32017-06-13 22:21:02 +02004554 if (ret < 0) {
4555 error_setg_errno(errp, -ret,
4556 "Failed to flush the preallocated area to disk");
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004557 goto fail;
Max Reitz95b98f32017-06-13 22:21:02 +02004558 }
4559 }
4560
Leonid Bloch45b49492018-09-26 19:04:45 +03004561 bs->total_sectors = offset / BDRV_SECTOR_SIZE;
4562
Stefan Hajnoczi419b19d2010-04-28 11:36:11 +01004563 /* write updated header.size */
4564 offset = cpu_to_be64(offset);
Alberto Fariaa8f0e832022-06-09 16:27:43 +01004565 ret = bdrv_co_pwrite_sync(bs->file, offsetof(QCowHeader, size),
4566 sizeof(offset), &offset, 0);
Stefan Hajnoczi419b19d2010-04-28 11:36:11 +01004567 if (ret < 0) {
Max Reitzf59adb32017-03-28 22:51:29 +02004568 error_setg_errno(errp, -ret, "Failed to update the image size");
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004569 goto fail;
Stefan Hajnoczi419b19d2010-04-28 11:36:11 +01004570 }
4571
4572 s->l1_vm_state_index = new_l1_size;
Leonid Bloch45b49492018-09-26 19:04:45 +03004573
4574 /* Update cache sizes */
4575 options = qdict_clone_shallow(bs->options);
4576 ret = qcow2_update_options(bs, options, s->flags, errp);
4577 qobject_unref(options);
4578 if (ret < 0) {
4579 goto fail;
4580 }
Kevin Wolf061ca8a2018-06-21 17:54:35 +02004581 ret = 0;
4582fail:
4583 qemu_co_mutex_unlock(&s->lock);
4584 return ret;
Stefan Hajnoczi419b19d2010-04-28 11:36:11 +01004585}
4586
Pavel Butsykinfcccefc2016-07-22 11:17:43 +03004587static coroutine_fn int
Andrey Shinkevich0d483dc2019-12-02 15:15:05 +03004588qcow2_co_pwritev_compressed_task(BlockDriverState *bs,
Vladimir Sementsov-Ogievskiy53962342019-06-04 19:15:14 +03004589 uint64_t offset, uint64_t bytes,
4590 QEMUIOVector *qiov, size_t qiov_offset)
Blue Swirl20d97352010-04-23 20:19:47 +00004591{
Kevin Wolfff991292015-09-07 17:12:56 +02004592 BDRVQcow2State *s = bs->opaque;
Vladimir Sementsov-Ogievskiy2714f132018-06-20 17:48:36 +03004593 int ret;
Alberto Garciae1f4a372019-04-30 13:08:02 +03004594 ssize_t out_len;
Pavel Butsykinfcccefc2016-07-22 11:17:43 +03004595 uint8_t *buf, *out_buf;
Kevin Wolf77e023f2019-02-27 10:26:24 +01004596 uint64_t cluster_offset;
Blue Swirl20d97352010-04-23 20:19:47 +00004597
Andrey Shinkevich0d483dc2019-12-02 15:15:05 +03004598 assert(bytes == s->cluster_size || (bytes < s->cluster_size &&
4599 (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS)));
Anton Nefedov3e3b8382017-11-14 13:16:49 +03004600
Pavel Butsykinfcccefc2016-07-22 11:17:43 +03004601 buf = qemu_blockalign(bs, s->cluster_size);
Andrey Shinkevich0d483dc2019-12-02 15:15:05 +03004602 if (bytes < s->cluster_size) {
Pavel Butsykina2c0ca62016-07-22 11:17:44 +03004603 /* Zero-pad last write if image size is not cluster aligned */
4604 memset(buf + bytes, 0, s->cluster_size - bytes);
4605 }
Vladimir Sementsov-Ogievskiy53962342019-06-04 19:15:14 +03004606 qemu_iovec_to_buf(qiov, qiov_offset, buf, bytes);
Blue Swirl20d97352010-04-23 20:19:47 +00004607
Vladimir Sementsov-Ogievskiyebf7bba2016-07-14 19:59:25 +03004608 out_buf = g_malloc(s->cluster_size);
Blue Swirl20d97352010-04-23 20:19:47 +00004609
Vladimir Sementsov-Ogievskiy6994fd72018-11-01 21:27:33 +03004610 out_len = qcow2_co_compress(bs, out_buf, s->cluster_size - 1,
4611 buf, s->cluster_size);
Alberto Garciae1f4a372019-04-30 13:08:02 +03004612 if (out_len == -ENOMEM) {
Blue Swirl20d97352010-04-23 20:19:47 +00004613 /* could not compress: write normal cluster */
Vladimir Sementsov-Ogievskiy53962342019-06-04 19:15:14 +03004614 ret = qcow2_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, 0);
Kevin Wolf8f1efd02011-10-18 17:12:44 +02004615 if (ret < 0) {
4616 goto fail;
4617 }
Pavel Butsykinfcccefc2016-07-22 11:17:43 +03004618 goto success;
Alberto Garciae1f4a372019-04-30 13:08:02 +03004619 } else if (out_len < 0) {
4620 ret = -EINVAL;
4621 goto fail;
Blue Swirl20d97352010-04-23 20:19:47 +00004622 }
4623
Pavel Butsykinfcccefc2016-07-22 11:17:43 +03004624 qemu_co_mutex_lock(&s->lock);
Kevin Wolf77e023f2019-02-27 10:26:24 +01004625 ret = qcow2_alloc_compressed_cluster_offset(bs, offset, out_len,
4626 &cluster_offset);
4627 if (ret < 0) {
Pavel Butsykinfcccefc2016-07-22 11:17:43 +03004628 qemu_co_mutex_unlock(&s->lock);
Pavel Butsykinfcccefc2016-07-22 11:17:43 +03004629 goto fail;
4630 }
Pavel Butsykinfcccefc2016-07-22 11:17:43 +03004631
Kevin Wolf966b0002019-01-15 20:39:06 +01004632 ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len, true);
Pavel Butsykinfcccefc2016-07-22 11:17:43 +03004633 qemu_co_mutex_unlock(&s->lock);
4634 if (ret < 0) {
4635 goto fail;
4636 }
4637
Kevin Wolf966b0002019-01-15 20:39:06 +01004638 BLKDBG_EVENT(s->data_file, BLKDBG_WRITE_COMPRESSED);
Vladimir Sementsov-Ogievskiyb00cb152019-04-22 17:58:31 +03004639 ret = bdrv_co_pwrite(s->data_file, cluster_offset, out_len, out_buf, 0);
Pavel Butsykinfcccefc2016-07-22 11:17:43 +03004640 if (ret < 0) {
4641 goto fail;
4642 }
4643success:
Kevin Wolf8f1efd02011-10-18 17:12:44 +02004644 ret = 0;
4645fail:
Pavel Butsykinfcccefc2016-07-22 11:17:43 +03004646 qemu_vfree(buf);
Anthony Liguori7267c092011-08-20 22:09:37 -05004647 g_free(out_buf);
Kevin Wolf8f1efd02011-10-18 17:12:44 +02004648 return ret;
Blue Swirl20d97352010-04-23 20:19:47 +00004649}
4650
Andrey Shinkevich0d483dc2019-12-02 15:15:05 +03004651static coroutine_fn int qcow2_co_pwritev_compressed_task_entry(AioTask *task)
4652{
4653 Qcow2AioTask *t = container_of(task, Qcow2AioTask, task);
4654
Alberto Garcia10dabdc52020-07-10 18:13:00 +02004655 assert(!t->subcluster_type && !t->l2meta);
Andrey Shinkevich0d483dc2019-12-02 15:15:05 +03004656
4657 return qcow2_co_pwritev_compressed_task(t->bs, t->offset, t->bytes, t->qiov,
4658 t->qiov_offset);
4659}
4660
4661/*
4662 * XXX: put compressed sectors first, then all the cluster aligned
4663 * tables to avoid losing bytes in alignment
4664 */
4665static coroutine_fn int
4666qcow2_co_pwritev_compressed_part(BlockDriverState *bs,
Vladimir Sementsov-Ogievskiye75abed2021-09-03 13:28:00 +03004667 int64_t offset, int64_t bytes,
Andrey Shinkevich0d483dc2019-12-02 15:15:05 +03004668 QEMUIOVector *qiov, size_t qiov_offset)
4669{
4670 BDRVQcow2State *s = bs->opaque;
4671 AioTaskPool *aio = NULL;
4672 int ret = 0;
4673
4674 if (has_data_file(bs)) {
4675 return -ENOTSUP;
4676 }
4677
4678 if (bytes == 0) {
4679 /*
4680 * align end of file to a sector boundary to ease reading with
4681 * sector based I/Os
4682 */
4683 int64_t len = bdrv_getlength(bs->file->bs);
4684 if (len < 0) {
4685 return len;
4686 }
Kevin Wolf7b8e4852020-04-24 14:54:40 +02004687 return bdrv_co_truncate(bs->file, len, false, PREALLOC_MODE_OFF, 0,
4688 NULL);
Andrey Shinkevich0d483dc2019-12-02 15:15:05 +03004689 }
4690
4691 if (offset_into_cluster(s, offset)) {
4692 return -EINVAL;
4693 }
4694
Alberto Garciafb43d2d2020-04-06 16:34:01 +02004695 if (offset_into_cluster(s, bytes) &&
4696 (offset + bytes) != (bs->total_sectors << BDRV_SECTOR_BITS)) {
4697 return -EINVAL;
4698 }
4699
Andrey Shinkevich0d483dc2019-12-02 15:15:05 +03004700 while (bytes && aio_task_pool_status(aio) == 0) {
4701 uint64_t chunk_size = MIN(bytes, s->cluster_size);
4702
4703 if (!aio && chunk_size != bytes) {
4704 aio = aio_task_pool_new(QCOW2_MAX_WORKERS);
4705 }
4706
4707 ret = qcow2_add_task(bs, aio, qcow2_co_pwritev_compressed_task_entry,
4708 0, 0, offset, chunk_size, qiov, qiov_offset, NULL);
4709 if (ret < 0) {
4710 break;
4711 }
4712 qiov_offset += chunk_size;
4713 offset += chunk_size;
4714 bytes -= chunk_size;
4715 }
4716
4717 if (aio) {
4718 aio_task_pool_wait_all(aio);
4719 if (ret == 0) {
4720 ret = aio_task_pool_status(aio);
4721 }
4722 g_free(aio);
4723 }
4724
4725 return ret;
4726}
4727
Vladimir Sementsov-Ogievskiyc3c10f72018-11-01 21:27:37 +03004728static int coroutine_fn
4729qcow2_co_preadv_compressed(BlockDriverState *bs,
Vladimir Sementsov-Ogievskiy9a3978a2021-09-14 15:24:46 +03004730 uint64_t l2_entry,
Vladimir Sementsov-Ogievskiyc3c10f72018-11-01 21:27:37 +03004731 uint64_t offset,
4732 uint64_t bytes,
Vladimir Sementsov-Ogievskiydf893d22019-06-04 19:15:13 +03004733 QEMUIOVector *qiov,
4734 size_t qiov_offset)
Vladimir Sementsov-Ogievskiyf4b3e2a2018-11-01 21:27:34 +03004735{
4736 BDRVQcow2State *s = bs->opaque;
Vladimir Sementsov-Ogievskiya6e09842021-09-14 15:24:47 +03004737 int ret = 0, csize;
Vladimir Sementsov-Ogievskiyf4b3e2a2018-11-01 21:27:34 +03004738 uint64_t coffset;
Vladimir Sementsov-Ogievskiyc3c10f72018-11-01 21:27:37 +03004739 uint8_t *buf, *out_buf;
Vladimir Sementsov-Ogievskiyc3c10f72018-11-01 21:27:37 +03004740 int offset_in_cluster = offset_into_cluster(s, offset);
Vladimir Sementsov-Ogievskiyf4b3e2a2018-11-01 21:27:34 +03004741
Vladimir Sementsov-Ogievskiya6e09842021-09-14 15:24:47 +03004742 qcow2_parse_compressed_l2_entry(bs, l2_entry, &coffset, &csize);
Vladimir Sementsov-Ogievskiyf4b3e2a2018-11-01 21:27:34 +03004743
Vladimir Sementsov-Ogievskiyc3c10f72018-11-01 21:27:37 +03004744 buf = g_try_malloc(csize);
4745 if (!buf) {
4746 return -ENOMEM;
Vladimir Sementsov-Ogievskiyf4b3e2a2018-11-01 21:27:34 +03004747 }
Vladimir Sementsov-Ogievskiyc3c10f72018-11-01 21:27:37 +03004748
4749 out_buf = qemu_blockalign(bs, s->cluster_size);
4750
4751 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
Vladimir Sementsov-Ogievskiyb00cb152019-04-22 17:58:31 +03004752 ret = bdrv_co_pread(bs->file, coffset, csize, buf, 0);
Vladimir Sementsov-Ogievskiyc3c10f72018-11-01 21:27:37 +03004753 if (ret < 0) {
4754 goto fail;
4755 }
4756
Vladimir Sementsov-Ogievskiye23c9d72018-11-01 21:27:38 +03004757 if (qcow2_co_decompress(bs, out_buf, s->cluster_size, buf, csize) < 0) {
Vladimir Sementsov-Ogievskiyc3c10f72018-11-01 21:27:37 +03004758 ret = -EIO;
4759 goto fail;
4760 }
4761
Vladimir Sementsov-Ogievskiydf893d22019-06-04 19:15:13 +03004762 qemu_iovec_from_buf(qiov, qiov_offset, out_buf + offset_in_cluster, bytes);
Vladimir Sementsov-Ogievskiyc3c10f72018-11-01 21:27:37 +03004763
4764fail:
4765 qemu_vfree(out_buf);
4766 g_free(buf);
4767
4768 return ret;
Vladimir Sementsov-Ogievskiyf4b3e2a2018-11-01 21:27:34 +03004769}
4770
Max Reitz94054182014-10-24 15:57:32 +02004771static int make_completely_empty(BlockDriverState *bs)
4772{
Kevin Wolfff991292015-09-07 17:12:56 +02004773 BDRVQcow2State *s = bs->opaque;
Max Reitzed3d2ec2017-03-28 22:51:27 +02004774 Error *local_err = NULL;
Max Reitz94054182014-10-24 15:57:32 +02004775 int ret, l1_clusters;
4776 int64_t offset;
4777 uint64_t *new_reftable = NULL;
4778 uint64_t rt_entry, l1_size2;
4779 struct {
4780 uint64_t l1_offset;
4781 uint64_t reftable_offset;
4782 uint32_t reftable_clusters;
4783 } QEMU_PACKED l1_ofs_rt_ofs_cls;
4784
4785 ret = qcow2_cache_empty(bs, s->l2_table_cache);
4786 if (ret < 0) {
4787 goto fail;
4788 }
4789
4790 ret = qcow2_cache_empty(bs, s->refcount_block_cache);
4791 if (ret < 0) {
4792 goto fail;
4793 }
4794
4795 /* Refcounts will be broken utterly */
4796 ret = qcow2_mark_dirty(bs);
4797 if (ret < 0) {
4798 goto fail;
4799 }
4800
4801 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4802
Alberto Garcia02b1ecf2020-08-28 13:08:28 +02004803 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE);
4804 l1_size2 = (uint64_t)s->l1_size * L1E_SIZE;
Max Reitz94054182014-10-24 15:57:32 +02004805
4806 /* After this call, neither the in-memory nor the on-disk refcount
4807 * information accurately describe the actual references */
4808
Kevin Wolf720ff282016-06-16 15:13:15 +02004809 ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset,
Eric Blake74021bc2016-06-01 15:10:04 -06004810 l1_clusters * s->cluster_size, 0);
Max Reitz94054182014-10-24 15:57:32 +02004811 if (ret < 0) {
4812 goto fail_broken_refcounts;
4813 }
4814 memset(s->l1_table, 0, l1_size2);
4815
4816 BLKDBG_EVENT(bs->file, BLKDBG_EMPTY_IMAGE_PREPARE);
4817
4818 /* Overwrite enough clusters at the beginning of the sectors to place
4819 * the refcount table, a refcount block and the L1 table in; this may
4820 * overwrite parts of the existing refcount and L1 table, which is not
4821 * an issue because the dirty flag is set, complete data loss is in fact
4822 * desired and partial data loss is consequently fine as well */
Kevin Wolf720ff282016-06-16 15:13:15 +02004823 ret = bdrv_pwrite_zeroes(bs->file, s->cluster_size,
Eric Blake74021bc2016-06-01 15:10:04 -06004824 (2 + l1_clusters) * s->cluster_size, 0);
Max Reitz94054182014-10-24 15:57:32 +02004825 /* This call (even if it failed overall) may have overwritten on-disk
4826 * refcount structures; in that case, the in-memory refcount information
4827 * will probably differ from the on-disk information which makes the BDS
4828 * unusable */
4829 if (ret < 0) {
4830 goto fail_broken_refcounts;
4831 }
4832
4833 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
4834 BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
4835
4836 /* "Create" an empty reftable (one cluster) directly after the image
4837 * header and an empty L1 table three clusters after the image header;
4838 * the cluster between those two will be used as the first refblock */
Peter Maydellf1f7a1d2016-06-16 17:06:17 +01004839 l1_ofs_rt_ofs_cls.l1_offset = cpu_to_be64(3 * s->cluster_size);
4840 l1_ofs_rt_ofs_cls.reftable_offset = cpu_to_be64(s->cluster_size);
4841 l1_ofs_rt_ofs_cls.reftable_clusters = cpu_to_be32(1);
Kevin Wolfd9ca2ea2016-06-20 20:09:15 +02004842 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_table_offset),
Alberto Faria32cc71d2022-06-09 16:27:36 +01004843 sizeof(l1_ofs_rt_ofs_cls), &l1_ofs_rt_ofs_cls, 0);
Max Reitz94054182014-10-24 15:57:32 +02004844 if (ret < 0) {
4845 goto fail_broken_refcounts;
4846 }
4847
4848 s->l1_table_offset = 3 * s->cluster_size;
4849
Alberto Garcia02b1ecf2020-08-28 13:08:28 +02004850 new_reftable = g_try_new0(uint64_t, s->cluster_size / REFTABLE_ENTRY_SIZE);
Max Reitz94054182014-10-24 15:57:32 +02004851 if (!new_reftable) {
4852 ret = -ENOMEM;
4853 goto fail_broken_refcounts;
4854 }
4855
4856 s->refcount_table_offset = s->cluster_size;
Alberto Garcia02b1ecf2020-08-28 13:08:28 +02004857 s->refcount_table_size = s->cluster_size / REFTABLE_ENTRY_SIZE;
Alberto Garcia7061a072017-02-01 14:38:28 +02004858 s->max_refcount_table_index = 0;
Max Reitz94054182014-10-24 15:57:32 +02004859
4860 g_free(s->refcount_table);
4861 s->refcount_table = new_reftable;
4862 new_reftable = NULL;
4863
4864 /* Now the in-memory refcount information again corresponds to the on-disk
4865 * information (reftable is empty and no refblocks (the refblock cache is
4866 * empty)); however, this means some clusters (e.g. the image header) are
4867 * referenced, but not refcounted, but the normal qcow2 code assumes that
4868 * the in-memory information is always correct */
4869
4870 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_ALLOC);
4871
4872 /* Enter the first refblock into the reftable */
4873 rt_entry = cpu_to_be64(2 * s->cluster_size);
Alberto Faria32cc71d2022-06-09 16:27:36 +01004874 ret = bdrv_pwrite_sync(bs->file, s->cluster_size, sizeof(rt_entry),
4875 &rt_entry, 0);
Max Reitz94054182014-10-24 15:57:32 +02004876 if (ret < 0) {
4877 goto fail_broken_refcounts;
4878 }
4879 s->refcount_table[0] = 2 * s->cluster_size;
4880
4881 s->free_cluster_index = 0;
4882 assert(3 + l1_clusters <= s->refcount_block_size);
4883 offset = qcow2_alloc_clusters(bs, 3 * s->cluster_size + l1_size2);
4884 if (offset < 0) {
4885 ret = offset;
4886 goto fail_broken_refcounts;
4887 } else if (offset > 0) {
4888 error_report("First cluster in emptied image is in use");
4889 abort();
4890 }
4891
4892 /* Now finally the in-memory information corresponds to the on-disk
4893 * structures and is correct */
4894 ret = qcow2_mark_clean(bs);
4895 if (ret < 0) {
4896 goto fail;
4897 }
4898
Max Reitzc80d8b02019-09-18 11:51:40 +02004899 ret = bdrv_truncate(bs->file, (3 + l1_clusters) * s->cluster_size, false,
Kevin Wolf7b8e4852020-04-24 14:54:40 +02004900 PREALLOC_MODE_OFF, 0, &local_err);
Max Reitz94054182014-10-24 15:57:32 +02004901 if (ret < 0) {
Max Reitzed3d2ec2017-03-28 22:51:27 +02004902 error_report_err(local_err);
Max Reitz94054182014-10-24 15:57:32 +02004903 goto fail;
4904 }
4905
4906 return 0;
4907
4908fail_broken_refcounts:
4909 /* The BDS is unusable at this point. If we wanted to make it usable, we
4910 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4911 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4912 * again. However, because the functions which could have caused this error
4913 * path to be taken are used by those functions as well, it's very likely
4914 * that that sequence will fail as well. Therefore, just eject the BDS. */
4915 bs->drv = NULL;
4916
4917fail:
4918 g_free(new_reftable);
4919 return ret;
4920}
4921
Max Reitz491d27e2014-10-24 15:57:31 +02004922static int qcow2_make_empty(BlockDriverState *bs)
4923{
Kevin Wolfff991292015-09-07 17:12:56 +02004924 BDRVQcow2State *s = bs->opaque;
Eric Blaked2cb36a2017-05-06 19:05:52 -05004925 uint64_t offset, end_offset;
4926 int step = QEMU_ALIGN_DOWN(INT_MAX, s->cluster_size);
Max Reitz94054182014-10-24 15:57:32 +02004927 int l1_clusters, ret = 0;
Max Reitz491d27e2014-10-24 15:57:31 +02004928
Alberto Garcia02b1ecf2020-08-28 13:08:28 +02004929 l1_clusters = DIV_ROUND_UP(s->l1_size, s->cluster_size / L1E_SIZE);
Max Reitz94054182014-10-24 15:57:32 +02004930
Eric Blake40969742017-11-17 10:47:47 -06004931 if (s->qcow_version >= 3 && !s->snapshots && !s->nb_bitmaps &&
Daniel P. Berrangef0603322017-11-17 11:29:13 +00004932 3 + l1_clusters <= s->refcount_block_size &&
Kevin Wolfdb045242019-04-29 12:52:21 +02004933 s->crypt_method_header != QCOW_CRYPT_LUKS &&
4934 !has_data_file(bs)) {
Eric Blake40969742017-11-17 10:47:47 -06004935 /* The following function only works for qcow2 v3 images (it
4936 * requires the dirty flag) and only as long as there are no
4937 * features that reserve extra clusters (such as snapshots,
4938 * LUKS header, or persistent bitmaps), because it completely
4939 * empties the image. Furthermore, the L1 table and three
4940 * additional clusters (image header, refcount table, one
Kevin Wolfdb045242019-04-29 12:52:21 +02004941 * refcount block) have to fit inside one refcount block. It
4942 * only resets the image file, i.e. does not work with an
4943 * external data file. */
Max Reitz94054182014-10-24 15:57:32 +02004944 return make_completely_empty(bs);
4945 }
4946
4947 /* This fallback code simply discards every active cluster; this is slow,
4948 * but works in all cases */
Eric Blaked2cb36a2017-05-06 19:05:52 -05004949 end_offset = bs->total_sectors * BDRV_SECTOR_SIZE;
4950 for (offset = 0; offset < end_offset; offset += step) {
Max Reitz491d27e2014-10-24 15:57:31 +02004951 /* As this function is generally used after committing an external
4952 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4953 * default action for this kind of discard is to pass the discard,
4954 * which will ideally result in an actually smaller image file, as
4955 * is probably desired. */
Eric Blaked2cb36a2017-05-06 19:05:52 -05004956 ret = qcow2_cluster_discard(bs, offset, MIN(step, end_offset - offset),
4957 QCOW2_DISCARD_SNAPSHOT, true);
Max Reitz491d27e2014-10-24 15:57:31 +02004958 if (ret < 0) {
4959 break;
4960 }
4961 }
4962
4963 return ret;
4964}
4965
Dong Xu Wanga9681682011-11-10 16:23:22 +08004966static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
Blue Swirl20d97352010-04-23 20:19:47 +00004967{
Kevin Wolfff991292015-09-07 17:12:56 +02004968 BDRVQcow2State *s = bs->opaque;
Kevin Wolf29c1a732011-01-10 17:17:28 +01004969 int ret;
4970
Paolo Bonzini8b94ff82011-10-20 13:16:24 +02004971 qemu_co_mutex_lock(&s->lock);
Paolo Bonzini8b220eb2018-03-01 17:36:14 +01004972 ret = qcow2_write_caches(bs);
Paolo Bonzini8b94ff82011-10-20 13:16:24 +02004973 qemu_co_mutex_unlock(&s->lock);
Kevin Wolf29c1a732011-01-10 17:17:28 +01004974
Paolo Bonzini8b220eb2018-03-01 17:36:14 +01004975 return ret;
Kevin Wolfeb489bb2011-11-10 18:10:11 +01004976}
4977
Stefan Hajnoczic501c352017-07-05 13:57:35 +01004978static BlockMeasureInfo *qcow2_measure(QemuOpts *opts, BlockDriverState *in_bs,
4979 Error **errp)
4980{
4981 Error *local_err = NULL;
4982 BlockMeasureInfo *info;
4983 uint64_t required = 0; /* bytes that contribute to required size */
4984 uint64_t virtual_size; /* disk size as seen by guest */
4985 uint64_t refcount_bits;
4986 uint64_t l2_tables;
Stefan Hajnoczi61914f82019-02-18 10:45:24 +00004987 uint64_t luks_payload_size = 0;
Stefan Hajnoczic501c352017-07-05 13:57:35 +01004988 size_t cluster_size;
4989 int version;
4990 char *optstr;
4991 PreallocMode prealloc;
4992 bool has_backing_file;
Stefan Hajnoczi61914f82019-02-18 10:45:24 +00004993 bool has_luks;
Alberto Garcia7be20252020-07-10 18:13:13 +02004994 bool extended_l2;
Alberto Garcia0dd07b22020-07-10 18:13:11 +02004995 size_t l2e_size;
Stefan Hajnoczic501c352017-07-05 13:57:35 +01004996
4997 /* Parse image creation options */
Alberto Garcia7be20252020-07-10 18:13:13 +02004998 extended_l2 = qemu_opt_get_bool_del(opts, BLOCK_OPT_EXTL2, false);
4999
5000 cluster_size = qcow2_opt_get_cluster_size_del(opts, extended_l2,
5001 &local_err);
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005002 if (local_err) {
5003 goto err;
5004 }
5005
5006 version = qcow2_opt_get_version_del(opts, &local_err);
5007 if (local_err) {
5008 goto err;
5009 }
5010
5011 refcount_bits = qcow2_opt_get_refcount_bits_del(opts, version, &local_err);
5012 if (local_err) {
5013 goto err;
5014 }
5015
5016 optstr = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
Marc-André Lureauf7abe0e2017-08-24 10:46:10 +02005017 prealloc = qapi_enum_parse(&PreallocMode_lookup, optstr,
Markus Armbruster06c60b62017-08-24 10:45:57 +02005018 PREALLOC_MODE_OFF, &local_err);
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005019 g_free(optstr);
5020 if (local_err) {
5021 goto err;
5022 }
5023
5024 optstr = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
5025 has_backing_file = !!optstr;
5026 g_free(optstr);
5027
Stefan Hajnoczi61914f82019-02-18 10:45:24 +00005028 optstr = qemu_opt_get_del(opts, BLOCK_OPT_ENCRYPT_FORMAT);
5029 has_luks = optstr && strcmp(optstr, "luks") == 0;
5030 g_free(optstr);
5031
5032 if (has_luks) {
Stefan Hajnoczi6d49d3a2020-02-21 11:25:19 +00005033 g_autoptr(QCryptoBlockCreateOptions) create_opts = NULL;
Maxim Levitsky90766d92020-06-25 14:55:43 +02005034 QDict *cryptoopts = qcow2_extract_crypto_opts(opts, "luks", errp);
Stefan Hajnoczi61914f82019-02-18 10:45:24 +00005035 size_t headerlen;
5036
Stefan Hajnoczi6d49d3a2020-02-21 11:25:19 +00005037 create_opts = block_crypto_create_opts_init(cryptoopts, errp);
5038 qobject_unref(cryptoopts);
5039 if (!create_opts) {
5040 goto err;
5041 }
5042
5043 if (!qcrypto_block_calculate_payload_offset(create_opts,
5044 "encrypt.",
5045 &headerlen,
5046 &local_err)) {
Stefan Hajnoczi61914f82019-02-18 10:45:24 +00005047 goto err;
5048 }
5049
5050 luks_payload_size = ROUND_UP(headerlen, cluster_size);
5051 }
5052
Alberto Garcia9e029682018-02-15 15:10:08 +02005053 virtual_size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
5054 virtual_size = ROUND_UP(virtual_size, cluster_size);
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005055
5056 /* Check that virtual disk size is valid */
Alberto Garcia0dd07b22020-07-10 18:13:11 +02005057 l2e_size = extended_l2 ? L2E_SIZE_EXTENDED : L2E_SIZE_NORMAL;
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005058 l2_tables = DIV_ROUND_UP(virtual_size / cluster_size,
Alberto Garcia0dd07b22020-07-10 18:13:11 +02005059 cluster_size / l2e_size);
Alberto Garcia02b1ecf2020-08-28 13:08:28 +02005060 if (l2_tables * L1E_SIZE > QCOW_MAX_L1_SIZE) {
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005061 error_setg(&local_err, "The image size is too large "
5062 "(try using a larger cluster size)");
5063 goto err;
5064 }
5065
5066 /* Account for input image */
5067 if (in_bs) {
5068 int64_t ssize = bdrv_getlength(in_bs);
5069 if (ssize < 0) {
5070 error_setg_errno(&local_err, -ssize,
5071 "Unable to get image virtual_size");
5072 goto err;
5073 }
5074
Alberto Garcia9e029682018-02-15 15:10:08 +02005075 virtual_size = ROUND_UP(ssize, cluster_size);
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005076
5077 if (has_backing_file) {
5078 /* We don't how much of the backing chain is shared by the input
5079 * image and the new image file. In the worst case the new image's
5080 * backing file has nothing in common with the input image. Be
5081 * conservative and assume all clusters need to be written.
5082 */
5083 required = virtual_size;
5084 } else {
Eric Blakeb85ee452017-09-25 09:55:22 -05005085 int64_t offset;
Eric Blake31826642017-10-11 22:47:08 -05005086 int64_t pnum = 0;
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005087
Eric Blake31826642017-10-11 22:47:08 -05005088 for (offset = 0; offset < ssize; offset += pnum) {
5089 int ret;
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005090
Eric Blake31826642017-10-11 22:47:08 -05005091 ret = bdrv_block_status_above(in_bs, NULL, offset,
5092 ssize - offset, &pnum, NULL,
5093 NULL);
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005094 if (ret < 0) {
5095 error_setg_errno(&local_err, -ret,
5096 "Unable to get block status");
5097 goto err;
5098 }
5099
5100 if (ret & BDRV_BLOCK_ZERO) {
5101 /* Skip zero regions (safe with no backing file) */
5102 } else if ((ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) ==
5103 (BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED)) {
5104 /* Extend pnum to end of cluster for next iteration */
Eric Blake31826642017-10-11 22:47:08 -05005105 pnum = ROUND_UP(offset + pnum, cluster_size) - offset;
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005106
5107 /* Count clusters we've seen */
Eric Blake31826642017-10-11 22:47:08 -05005108 required += offset % cluster_size + pnum;
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005109 }
5110 }
5111 }
5112 }
5113
5114 /* Take into account preallocation. Nothing special is needed for
5115 * PREALLOC_MODE_METADATA since metadata is always counted.
5116 */
5117 if (prealloc == PREALLOC_MODE_FULL || prealloc == PREALLOC_MODE_FALLOC) {
5118 required = virtual_size;
5119 }
5120
Eric Blake5d72c682020-05-21 14:21:34 -05005121 info = g_new0(BlockMeasureInfo, 1);
Alberto Garcia0dd07b22020-07-10 18:13:11 +02005122 info->fully_allocated = luks_payload_size +
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005123 qcow2_calc_prealloc_size(virtual_size, cluster_size,
Alberto Garcia0dd07b22020-07-10 18:13:11 +02005124 ctz32(refcount_bits), extended_l2);
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005125
Eric Blake5d72c682020-05-21 14:21:34 -05005126 /*
5127 * Remove data clusters that are not required. This overestimates the
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005128 * required size because metadata needed for the fully allocated file is
Eric Blake5d72c682020-05-21 14:21:34 -05005129 * still counted. Show bitmaps only if both source and destination
5130 * would support them.
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005131 */
5132 info->required = info->fully_allocated - virtual_size + required;
Eric Blake5d72c682020-05-21 14:21:34 -05005133 info->has_bitmaps = version >= 3 && in_bs &&
5134 bdrv_supports_persistent_dirty_bitmap(in_bs);
5135 if (info->has_bitmaps) {
5136 info->bitmaps = qcow2_get_persistent_dirty_bitmap_size(in_bs,
5137 cluster_size);
5138 }
Stefan Hajnoczic501c352017-07-05 13:57:35 +01005139 return info;
5140
5141err:
5142 error_propagate(errp, local_err);
5143 return NULL;
5144}
5145
Jes Sorensen7c80ab32010-12-17 16:02:39 +01005146static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
Blue Swirl20d97352010-04-23 20:19:47 +00005147{
Kevin Wolfff991292015-09-07 17:12:56 +02005148 BDRVQcow2State *s = bs->opaque;
Blue Swirl20d97352010-04-23 20:19:47 +00005149 bdi->cluster_size = s->cluster_size;
Jes Sorensen7c80ab32010-12-17 16:02:39 +01005150 bdi->vm_state_offset = qcow2_vm_state_offset(s);
Vladimir Sementsov-Ogievskiy38b44092021-05-04 19:06:56 +03005151 bdi->is_dirty = s->incompatible_features & QCOW2_INCOMPAT_DIRTY;
Blue Swirl20d97352010-04-23 20:19:47 +00005152 return 0;
5153}
5154
Andrey Shinkevich1bf6e9c2019-02-08 18:06:06 +03005155static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs,
5156 Error **errp)
Max Reitz37764df2013-10-09 10:46:18 +02005157{
Kevin Wolfff991292015-09-07 17:12:56 +02005158 BDRVQcow2State *s = bs->opaque;
Daniel P. Berrange0a12f6f2017-06-23 17:24:18 +01005159 ImageInfoSpecific *spec_info;
5160 QCryptoBlockInfo *encrypt_info = NULL;
Max Reitz37764df2013-10-09 10:46:18 +02005161
Daniel P. Berrange0a12f6f2017-06-23 17:24:18 +01005162 if (s->crypto != NULL) {
Vladimir Sementsov-Ogievskiy83bad8c2021-02-02 15:49:50 +03005163 encrypt_info = qcrypto_block_get_info(s->crypto, errp);
5164 if (!encrypt_info) {
Andrey Shinkevich1bf6e9c2019-02-08 18:06:06 +03005165 return NULL;
5166 }
Daniel P. Berrange0a12f6f2017-06-23 17:24:18 +01005167 }
5168
5169 spec_info = g_new(ImageInfoSpecific, 1);
Max Reitz37764df2013-10-09 10:46:18 +02005170 *spec_info = (ImageInfoSpecific){
Eric Blake6a8f9662015-10-26 16:34:54 -06005171 .type = IMAGE_INFO_SPECIFIC_KIND_QCOW2,
Andrey Shinkevichb8968c82019-02-08 18:06:07 +03005172 .u.qcow2.data = g_new0(ImageInfoSpecificQCow2, 1),
Max Reitz37764df2013-10-09 10:46:18 +02005173 };
5174 if (s->qcow_version == 2) {
Eric Blake32bafa82016-03-17 16:48:37 -06005175 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
Max Reitz0709c5a2015-02-10 15:28:44 -05005176 .compat = g_strdup("0.10"),
5177 .refcount_bits = s->refcount_bits,
Max Reitz37764df2013-10-09 10:46:18 +02005178 };
5179 } else if (s->qcow_version == 3) {
Andrey Shinkevichb8968c82019-02-08 18:06:07 +03005180 Qcow2BitmapInfoList *bitmaps;
Vladimir Sementsov-Ogievskiy83bad8c2021-02-02 15:49:50 +03005181 if (!qcow2_get_bitmap_info_list(bs, &bitmaps, errp)) {
Andrey Shinkevichb8968c82019-02-08 18:06:07 +03005182 qapi_free_ImageInfoSpecific(spec_info);
Eric Blake71eaec22020-03-20 13:36:20 -05005183 qapi_free_QCryptoBlockInfo(encrypt_info);
Andrey Shinkevichb8968c82019-02-08 18:06:07 +03005184 return NULL;
5185 }
Eric Blake32bafa82016-03-17 16:48:37 -06005186 *spec_info->u.qcow2.data = (ImageInfoSpecificQCow2){
Max Reitz37764df2013-10-09 10:46:18 +02005187 .compat = g_strdup("1.1"),
5188 .lazy_refcounts = s->compatible_features &
5189 QCOW2_COMPAT_LAZY_REFCOUNTS,
5190 .has_lazy_refcounts = true,
Max Reitz9009b192014-09-30 21:31:28 +02005191 .corrupt = s->incompatible_features &
5192 QCOW2_INCOMPAT_CORRUPT,
5193 .has_corrupt = true,
Alberto Garcia7be20252020-07-10 18:13:13 +02005194 .has_extended_l2 = true,
5195 .extended_l2 = has_subclusters(s),
Max Reitz0709c5a2015-02-10 15:28:44 -05005196 .refcount_bits = s->refcount_bits,
Andrey Shinkevichb8968c82019-02-08 18:06:07 +03005197 .has_bitmaps = !!bitmaps,
5198 .bitmaps = bitmaps,
Kevin Wolf9b890bd2019-01-15 19:02:40 +01005199 .data_file = g_strdup(s->image_data_file),
Kevin Wolf6c3944d2019-02-22 14:29:38 +01005200 .has_data_file_raw = has_data_file(bs),
5201 .data_file_raw = data_file_is_raw(bs),
Denis Plotnikov572ad972020-05-07 11:25:18 +03005202 .compression_type = s->compression_type,
Max Reitz37764df2013-10-09 10:46:18 +02005203 };
Denis V. Lunevb1fc8f92015-12-10 12:55:48 +03005204 } else {
5205 /* if this assertion fails, this probably means a new version was
5206 * added without having it covered here */
5207 assert(false);
Max Reitz37764df2013-10-09 10:46:18 +02005208 }
5209
Daniel P. Berrange0a12f6f2017-06-23 17:24:18 +01005210 if (encrypt_info) {
5211 ImageInfoSpecificQCow2Encryption *qencrypt =
5212 g_new(ImageInfoSpecificQCow2Encryption, 1);
5213 switch (encrypt_info->format) {
5214 case Q_CRYPTO_BLOCK_FORMAT_QCOW:
5215 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES;
Daniel P. Berrange0a12f6f2017-06-23 17:24:18 +01005216 break;
5217 case Q_CRYPTO_BLOCK_FORMAT_LUKS:
5218 qencrypt->format = BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS;
5219 qencrypt->u.luks = encrypt_info->u.luks;
5220 break;
5221 default:
5222 abort();
5223 }
5224 /* Since we did shallow copy above, erase any pointers
5225 * in the original info */
5226 memset(&encrypt_info->u, 0, sizeof(encrypt_info->u));
5227 qapi_free_QCryptoBlockInfo(encrypt_info);
5228
Daniel P. Berrange0a12f6f2017-06-23 17:24:18 +01005229 spec_info->u.qcow2.data->encrypt = qencrypt;
5230 }
5231
Max Reitz37764df2013-10-09 10:46:18 +02005232 return spec_info;
5233}
5234
Max Reitz38841dc2019-07-24 19:12:34 +02005235static int qcow2_has_zero_init(BlockDriverState *bs)
5236{
5237 BDRVQcow2State *s = bs->opaque;
5238 bool preallocated;
5239
5240 if (qemu_in_coroutine()) {
5241 qemu_co_mutex_lock(&s->lock);
5242 }
5243 /*
5244 * Check preallocation status: Preallocated images have all L2
5245 * tables allocated, nonpreallocated images have none. It is
5246 * therefore enough to check the first one.
5247 */
5248 preallocated = s->l1_size > 0 && s->l1_table[0] != 0;
5249 if (qemu_in_coroutine()) {
5250 qemu_co_mutex_unlock(&s->lock);
5251 }
5252
5253 if (!preallocated) {
5254 return 1;
5255 } else if (bs->encrypted) {
5256 return 0;
5257 } else {
5258 return bdrv_has_zero_init(s->data_file->bs);
5259 }
5260}
5261
Vladimir Sementsov-Ogievskiy558902c2021-09-03 13:27:58 +03005262/*
5263 * Check the request to vmstate. On success return
5264 * qcow2_vm_state_offset(bs) + @pos
5265 */
5266static int64_t qcow2_check_vmstate_request(BlockDriverState *bs,
5267 QEMUIOVector *qiov, int64_t pos)
5268{
5269 BDRVQcow2State *s = bs->opaque;
5270 int64_t vmstate_offset = qcow2_vm_state_offset(s);
5271 int ret;
5272
5273 /* Incoming requests must be OK */
5274 bdrv_check_qiov_request(pos, qiov->size, qiov, 0, &error_abort);
5275
5276 if (INT64_MAX - pos < vmstate_offset) {
5277 return -EIO;
5278 }
5279
5280 pos += vmstate_offset;
5281 ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL);
5282 if (ret < 0) {
5283 return ret;
5284 }
5285
5286 return pos;
5287}
5288
Alberto Faria014688a2022-10-13 14:36:56 +02005289static coroutine_fn int qcow2_save_vmstate(BlockDriverState *bs,
5290 QEMUIOVector *qiov, int64_t pos)
Blue Swirl20d97352010-04-23 20:19:47 +00005291{
Vladimir Sementsov-Ogievskiy558902c2021-09-03 13:27:58 +03005292 int64_t offset = qcow2_check_vmstate_request(bs, qiov, pos);
5293 if (offset < 0) {
5294 return offset;
5295 }
Blue Swirl20d97352010-04-23 20:19:47 +00005296
Kevin Wolf66f82ce2010-04-14 14:17:38 +02005297 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
Vladimir Sementsov-Ogievskiy558902c2021-09-03 13:27:58 +03005298 return bs->drv->bdrv_co_pwritev_part(bs, offset, qiov->size, qiov, 0, 0);
Blue Swirl20d97352010-04-23 20:19:47 +00005299}
5300
Alberto Faria014688a2022-10-13 14:36:56 +02005301static coroutine_fn int qcow2_load_vmstate(BlockDriverState *bs,
5302 QEMUIOVector *qiov, int64_t pos)
Blue Swirl20d97352010-04-23 20:19:47 +00005303{
Vladimir Sementsov-Ogievskiy558902c2021-09-03 13:27:58 +03005304 int64_t offset = qcow2_check_vmstate_request(bs, qiov, pos);
5305 if (offset < 0) {
5306 return offset;
5307 }
Blue Swirl20d97352010-04-23 20:19:47 +00005308
Kevin Wolf66f82ce2010-04-14 14:17:38 +02005309 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
Vladimir Sementsov-Ogievskiy558902c2021-09-03 13:27:58 +03005310 return bs->drv->bdrv_co_preadv_part(bs, offset, qiov->size, qiov, 0, 0);
Blue Swirl20d97352010-04-23 20:19:47 +00005311}
5312
Vladimir Sementsov-Ogievskiy083c2452021-12-23 17:01:37 +01005313static int qcow2_has_compressed_clusters(BlockDriverState *bs)
5314{
5315 int64_t offset = 0;
5316 int64_t bytes = bdrv_getlength(bs);
5317
5318 if (bytes < 0) {
5319 return bytes;
5320 }
5321
5322 while (bytes != 0) {
5323 int ret;
5324 QCow2SubclusterType type;
5325 unsigned int cur_bytes = MIN(INT_MAX, bytes);
5326 uint64_t host_offset;
5327
5328 ret = qcow2_get_host_offset(bs, offset, &cur_bytes, &host_offset,
5329 &type);
5330 if (ret < 0) {
5331 return ret;
5332 }
5333
5334 if (type == QCOW2_SUBCLUSTER_COMPRESSED) {
5335 return 1;
5336 }
5337
5338 offset += cur_bytes;
5339 bytes -= cur_bytes;
5340 }
5341
5342 return 0;
5343}
5344
Max Reitz9296b3e2013-09-03 10:09:54 +02005345/*
5346 * Downgrades an image's version. To achieve this, any incompatible features
5347 * have to be removed.
5348 */
Max Reitz4057a2b2014-10-27 11:12:53 +01005349static int qcow2_downgrade(BlockDriverState *bs, int target_version,
Max Reitzd1402b52018-05-09 23:00:18 +02005350 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5351 Error **errp)
Max Reitz9296b3e2013-09-03 10:09:54 +02005352{
Kevin Wolfff991292015-09-07 17:12:56 +02005353 BDRVQcow2State *s = bs->opaque;
Max Reitz9296b3e2013-09-03 10:09:54 +02005354 int current_version = s->qcow_version;
5355 int ret;
Eric Blake7fa140a2020-04-28 14:26:47 -05005356 int i;
Max Reitz9296b3e2013-09-03 10:09:54 +02005357
Max Reitzd1402b52018-05-09 23:00:18 +02005358 /* This is qcow2_downgrade(), not qcow2_upgrade() */
5359 assert(target_version < current_version);
5360
5361 /* There are no other versions (now) that you can downgrade to */
5362 assert(target_version == 2);
Max Reitz9296b3e2013-09-03 10:09:54 +02005363
5364 if (s->refcount_order != 4) {
Max Reitzd1402b52018-05-09 23:00:18 +02005365 error_setg(errp, "compat=0.10 requires refcount_bits=16");
Max Reitz9296b3e2013-09-03 10:09:54 +02005366 return -ENOTSUP;
5367 }
5368
Kevin Wolf966b0002019-01-15 20:39:06 +01005369 if (has_data_file(bs)) {
5370 error_setg(errp, "Cannot downgrade an image with a data file");
5371 return -ENOTSUP;
5372 }
5373
Eric Blake7fa140a2020-04-28 14:26:47 -05005374 /*
5375 * If any internal snapshot has a different size than the current
5376 * image size, or VM state size that exceeds 32 bits, downgrading
5377 * is unsafe. Even though we would still use v3-compliant output
5378 * to preserve that data, other v2 programs might not realize
5379 * those optional fields are important.
5380 */
5381 for (i = 0; i < s->nb_snapshots; i++) {
5382 if (s->snapshots[i].vm_state_size > UINT32_MAX ||
5383 s->snapshots[i].disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) {
5384 error_setg(errp, "Internal snapshots prevent downgrade of image");
5385 return -ENOTSUP;
5386 }
5387 }
5388
Max Reitz9296b3e2013-09-03 10:09:54 +02005389 /* clear incompatible features */
5390 if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
5391 ret = qcow2_mark_clean(bs);
5392 if (ret < 0) {
Max Reitzd1402b52018-05-09 23:00:18 +02005393 error_setg_errno(errp, -ret, "Failed to make the image clean");
Max Reitz9296b3e2013-09-03 10:09:54 +02005394 return ret;
5395 }
5396 }
5397
5398 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
5399 * the first place; if that happens nonetheless, returning -ENOTSUP is the
5400 * best thing to do anyway */
5401
Vladimir Sementsov-Ogievskiy083c2452021-12-23 17:01:37 +01005402 if (s->incompatible_features & ~QCOW2_INCOMPAT_COMPRESSION) {
Max Reitzd1402b52018-05-09 23:00:18 +02005403 error_setg(errp, "Cannot downgrade an image with incompatible features "
Vladimir Sementsov-Ogievskiy083c2452021-12-23 17:01:37 +01005404 "0x%" PRIx64 " set",
5405 s->incompatible_features & ~QCOW2_INCOMPAT_COMPRESSION);
Max Reitz9296b3e2013-09-03 10:09:54 +02005406 return -ENOTSUP;
5407 }
5408
5409 /* since we can ignore compatible features, we can set them to 0 as well */
5410 s->compatible_features = 0;
5411 /* if lazy refcounts have been used, they have already been fixed through
5412 * clearing the dirty flag */
5413
5414 /* clearing autoclear features is trivial */
5415 s->autoclear_features = 0;
5416
Max Reitz8b139762015-07-27 17:51:32 +02005417 ret = qcow2_expand_zero_clusters(bs, status_cb, cb_opaque);
Max Reitz9296b3e2013-09-03 10:09:54 +02005418 if (ret < 0) {
Max Reitzd1402b52018-05-09 23:00:18 +02005419 error_setg_errno(errp, -ret, "Failed to turn zero into data clusters");
Max Reitz9296b3e2013-09-03 10:09:54 +02005420 return ret;
5421 }
5422
Vladimir Sementsov-Ogievskiy083c2452021-12-23 17:01:37 +01005423 if (s->incompatible_features & QCOW2_INCOMPAT_COMPRESSION) {
5424 ret = qcow2_has_compressed_clusters(bs);
5425 if (ret < 0) {
5426 error_setg(errp, "Failed to check block status");
5427 return -EINVAL;
5428 }
5429 if (ret) {
5430 error_setg(errp, "Cannot downgrade an image with zstd compression "
5431 "type and existing compressed clusters");
5432 return -ENOTSUP;
5433 }
5434 /*
5435 * No compressed clusters for now, so just chose default zlib
5436 * compression.
5437 */
5438 s->incompatible_features &= ~QCOW2_INCOMPAT_COMPRESSION;
5439 s->compression_type = QCOW2_COMPRESSION_TYPE_ZLIB;
5440 }
5441
5442 assert(s->incompatible_features == 0);
5443
Max Reitz9296b3e2013-09-03 10:09:54 +02005444 s->qcow_version = target_version;
5445 ret = qcow2_update_header(bs);
5446 if (ret < 0) {
5447 s->qcow_version = current_version;
Max Reitzd1402b52018-05-09 23:00:18 +02005448 error_setg_errno(errp, -ret, "Failed to update the image header");
Max Reitz9296b3e2013-09-03 10:09:54 +02005449 return ret;
5450 }
5451 return 0;
5452}
5453
Max Reitz722efb02019-10-11 17:28:04 +02005454/*
5455 * Upgrades an image's version. While newer versions encompass all
5456 * features of older versions, some things may have to be presented
5457 * differently.
5458 */
5459static int qcow2_upgrade(BlockDriverState *bs, int target_version,
5460 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5461 Error **errp)
5462{
5463 BDRVQcow2State *s = bs->opaque;
Max Reitz0a85af32019-10-11 17:28:05 +02005464 bool need_snapshot_update;
Max Reitz722efb02019-10-11 17:28:04 +02005465 int current_version = s->qcow_version;
Max Reitz0a85af32019-10-11 17:28:05 +02005466 int i;
Max Reitz722efb02019-10-11 17:28:04 +02005467 int ret;
5468
5469 /* This is qcow2_upgrade(), not qcow2_downgrade() */
5470 assert(target_version > current_version);
5471
5472 /* There are no other versions (yet) that you can upgrade to */
5473 assert(target_version == 3);
5474
Max Reitz0a85af32019-10-11 17:28:05 +02005475 status_cb(bs, 0, 2, cb_opaque);
5476
5477 /*
5478 * In v2, snapshots do not need to have extra data. v3 requires
5479 * the 64-bit VM state size and the virtual disk size to be
5480 * present.
5481 * qcow2_write_snapshots() will always write the list in the
5482 * v3-compliant format.
5483 */
5484 need_snapshot_update = false;
5485 for (i = 0; i < s->nb_snapshots; i++) {
5486 if (s->snapshots[i].extra_data_size <
5487 sizeof_field(QCowSnapshotExtraData, vm_state_size_large) +
5488 sizeof_field(QCowSnapshotExtraData, disk_size))
5489 {
5490 need_snapshot_update = true;
5491 break;
5492 }
5493 }
5494 if (need_snapshot_update) {
5495 ret = qcow2_write_snapshots(bs);
5496 if (ret < 0) {
5497 error_setg_errno(errp, -ret, "Failed to update the snapshot table");
5498 return ret;
5499 }
5500 }
5501 status_cb(bs, 1, 2, cb_opaque);
Max Reitz722efb02019-10-11 17:28:04 +02005502
5503 s->qcow_version = target_version;
5504 ret = qcow2_update_header(bs);
5505 if (ret < 0) {
5506 s->qcow_version = current_version;
5507 error_setg_errno(errp, -ret, "Failed to update the image header");
5508 return ret;
5509 }
Max Reitz0a85af32019-10-11 17:28:05 +02005510 status_cb(bs, 2, 2, cb_opaque);
Max Reitz722efb02019-10-11 17:28:04 +02005511
5512 return 0;
5513}
5514
Max Reitzc293a802015-07-27 17:51:36 +02005515typedef enum Qcow2AmendOperation {
5516 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
5517 * statically initialized to so that the helper CB can discern the first
5518 * invocation from an operation change */
5519 QCOW2_NO_OPERATION = 0,
5520
Max Reitz722efb02019-10-11 17:28:04 +02005521 QCOW2_UPGRADING,
Maxim Levitsky90766d92020-06-25 14:55:43 +02005522 QCOW2_UPDATING_ENCRYPTION,
Max Reitz61ce55f2015-07-27 17:51:38 +02005523 QCOW2_CHANGING_REFCOUNT_ORDER,
Max Reitzc293a802015-07-27 17:51:36 +02005524 QCOW2_DOWNGRADING,
5525} Qcow2AmendOperation;
5526
5527typedef struct Qcow2AmendHelperCBInfo {
5528 /* The code coordinating the amend operations should only modify
5529 * these four fields; the rest will be managed by the CB */
5530 BlockDriverAmendStatusCB *original_status_cb;
5531 void *original_cb_opaque;
5532
5533 Qcow2AmendOperation current_operation;
5534
5535 /* Total number of operations to perform (only set once) */
5536 int total_operations;
5537
5538 /* The following fields are managed by the CB */
5539
5540 /* Number of operations completed */
5541 int operations_completed;
5542
5543 /* Cumulative offset of all completed operations */
5544 int64_t offset_completed;
5545
5546 Qcow2AmendOperation last_operation;
5547 int64_t last_work_size;
5548} Qcow2AmendHelperCBInfo;
5549
5550static void qcow2_amend_helper_cb(BlockDriverState *bs,
5551 int64_t operation_offset,
5552 int64_t operation_work_size, void *opaque)
5553{
5554 Qcow2AmendHelperCBInfo *info = opaque;
5555 int64_t current_work_size;
5556 int64_t projected_work_size;
5557
5558 if (info->current_operation != info->last_operation) {
5559 if (info->last_operation != QCOW2_NO_OPERATION) {
5560 info->offset_completed += info->last_work_size;
5561 info->operations_completed++;
5562 }
5563
5564 info->last_operation = info->current_operation;
5565 }
5566
5567 assert(info->total_operations > 0);
5568 assert(info->operations_completed < info->total_operations);
5569
5570 info->last_work_size = operation_work_size;
5571
5572 current_work_size = info->offset_completed + operation_work_size;
5573
5574 /* current_work_size is the total work size for (operations_completed + 1)
5575 * operations (which includes this one), so multiply it by the number of
5576 * operations not covered and divide it by the number of operations
5577 * covered to get a projection for the operations not covered */
5578 projected_work_size = current_work_size * (info->total_operations -
5579 info->operations_completed - 1)
5580 / (info->operations_completed + 1);
5581
5582 info->original_status_cb(bs, info->offset_completed + operation_offset,
5583 current_work_size + projected_work_size,
5584 info->original_cb_opaque);
5585}
5586
Max Reitz77485432014-10-27 11:12:50 +01005587static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,
Max Reitz8b139762015-07-27 17:51:32 +02005588 BlockDriverAmendStatusCB *status_cb,
Max Reitzd1402b52018-05-09 23:00:18 +02005589 void *cb_opaque,
Maxim Levitskya3579bf2020-06-25 14:55:38 +02005590 bool force,
Max Reitzd1402b52018-05-09 23:00:18 +02005591 Error **errp)
Max Reitz9296b3e2013-09-03 10:09:54 +02005592{
Kevin Wolfff991292015-09-07 17:12:56 +02005593 BDRVQcow2State *s = bs->opaque;
Max Reitz9296b3e2013-09-03 10:09:54 +02005594 int old_version = s->qcow_version, new_version = old_version;
5595 uint64_t new_size = 0;
Kevin Wolf9b890bd2019-01-15 19:02:40 +01005596 const char *backing_file = NULL, *backing_format = NULL, *data_file = NULL;
Max Reitz9296b3e2013-09-03 10:09:54 +02005597 bool lazy_refcounts = s->use_lazy_refcounts;
Kevin Wolf6c3944d2019-02-22 14:29:38 +01005598 bool data_file_raw = data_file_is_raw(bs);
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +08005599 const char *compat = NULL;
Max Reitz61ce55f2015-07-27 17:51:38 +02005600 int refcount_bits = s->refcount_bits;
Max Reitz9296b3e2013-09-03 10:09:54 +02005601 int ret;
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +08005602 QemuOptDesc *desc = opts->list->desc;
Max Reitzc293a802015-07-27 17:51:36 +02005603 Qcow2AmendHelperCBInfo helper_cb_info;
Maxim Levitsky90766d92020-06-25 14:55:43 +02005604 bool encryption_update = false;
Max Reitz9296b3e2013-09-03 10:09:54 +02005605
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +08005606 while (desc && desc->name) {
5607 if (!qemu_opt_find(opts, desc->name)) {
Max Reitz9296b3e2013-09-03 10:09:54 +02005608 /* only change explicitly defined options */
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +08005609 desc++;
Max Reitz9296b3e2013-09-03 10:09:54 +02005610 continue;
5611 }
5612
Max Reitz8a17b832015-02-18 17:40:47 -05005613 if (!strcmp(desc->name, BLOCK_OPT_COMPAT_LEVEL)) {
5614 compat = qemu_opt_get(opts, BLOCK_OPT_COMPAT_LEVEL);
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +08005615 if (!compat) {
Max Reitz9296b3e2013-09-03 10:09:54 +02005616 /* preserve default */
Eric Blakef7077c92019-07-05 10:28:12 -05005617 } else if (!strcmp(compat, "0.10") || !strcmp(compat, "v2")) {
Max Reitz9296b3e2013-09-03 10:09:54 +02005618 new_version = 2;
Eric Blakef7077c92019-07-05 10:28:12 -05005619 } else if (!strcmp(compat, "1.1") || !strcmp(compat, "v3")) {
Max Reitz9296b3e2013-09-03 10:09:54 +02005620 new_version = 3;
5621 } else {
Max Reitzd1402b52018-05-09 23:00:18 +02005622 error_setg(errp, "Unknown compatibility level %s", compat);
Max Reitz9296b3e2013-09-03 10:09:54 +02005623 return -EINVAL;
5624 }
Max Reitz8a17b832015-02-18 17:40:47 -05005625 } else if (!strcmp(desc->name, BLOCK_OPT_SIZE)) {
5626 new_size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
5627 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FILE)) {
5628 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5629 } else if (!strcmp(desc->name, BLOCK_OPT_BACKING_FMT)) {
5630 backing_format = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
Maxim Levitsky90766d92020-06-25 14:55:43 +02005631 } else if (g_str_has_prefix(desc->name, "encrypt.")) {
5632 if (!s->crypto) {
5633 error_setg(errp,
5634 "Can't amend encryption options - encryption not present");
5635 return -EINVAL;
5636 }
5637 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
5638 error_setg(errp,
5639 "Only LUKS encryption options can be amended");
5640 return -ENOTSUP;
5641 }
5642 encryption_update = true;
Max Reitz8a17b832015-02-18 17:40:47 -05005643 } else if (!strcmp(desc->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
5644 lazy_refcounts = qemu_opt_get_bool(opts, BLOCK_OPT_LAZY_REFCOUNTS,
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +08005645 lazy_refcounts);
Max Reitz06d05fa2015-02-18 17:40:49 -05005646 } else if (!strcmp(desc->name, BLOCK_OPT_REFCOUNT_BITS)) {
Max Reitz61ce55f2015-07-27 17:51:38 +02005647 refcount_bits = qemu_opt_get_number(opts, BLOCK_OPT_REFCOUNT_BITS,
5648 refcount_bits);
5649
5650 if (refcount_bits <= 0 || refcount_bits > 64 ||
5651 !is_power_of_2(refcount_bits))
5652 {
Max Reitzd1402b52018-05-09 23:00:18 +02005653 error_setg(errp, "Refcount width must be a power of two and "
5654 "may not exceed 64 bits");
Max Reitz61ce55f2015-07-27 17:51:38 +02005655 return -EINVAL;
5656 }
Kevin Wolf9b890bd2019-01-15 19:02:40 +01005657 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE)) {
5658 data_file = qemu_opt_get(opts, BLOCK_OPT_DATA_FILE);
5659 if (data_file && !has_data_file(bs)) {
5660 error_setg(errp, "data-file can only be set for images that "
5661 "use an external data file");
5662 return -EINVAL;
5663 }
Kevin Wolf6c3944d2019-02-22 14:29:38 +01005664 } else if (!strcmp(desc->name, BLOCK_OPT_DATA_FILE_RAW)) {
5665 data_file_raw = qemu_opt_get_bool(opts, BLOCK_OPT_DATA_FILE_RAW,
5666 data_file_raw);
5667 if (data_file_raw && !data_file_is_raw(bs)) {
5668 error_setg(errp, "data-file-raw cannot be set on existing "
5669 "images");
5670 return -EINVAL;
5671 }
Max Reitz9296b3e2013-09-03 10:09:54 +02005672 } else {
Max Reitz164e0f82015-07-27 17:51:34 +02005673 /* if this point is reached, this probably means a new option was
Max Reitz9296b3e2013-09-03 10:09:54 +02005674 * added without having it covered here */
Max Reitz164e0f82015-07-27 17:51:34 +02005675 abort();
Max Reitz9296b3e2013-09-03 10:09:54 +02005676 }
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +08005677
5678 desc++;
Max Reitz9296b3e2013-09-03 10:09:54 +02005679 }
5680
Max Reitzc293a802015-07-27 17:51:36 +02005681 helper_cb_info = (Qcow2AmendHelperCBInfo){
5682 .original_status_cb = status_cb,
5683 .original_cb_opaque = cb_opaque,
Max Reitz722efb02019-10-11 17:28:04 +02005684 .total_operations = (new_version != old_version)
Maxim Levitsky90766d92020-06-25 14:55:43 +02005685 + (s->refcount_bits != refcount_bits) +
5686 (encryption_update == true)
Max Reitzc293a802015-07-27 17:51:36 +02005687 };
5688
Max Reitz1038bbb2015-07-27 17:51:35 +02005689 /* Upgrade first (some features may require compat=1.1) */
5690 if (new_version > old_version) {
Max Reitz722efb02019-10-11 17:28:04 +02005691 helper_cb_info.current_operation = QCOW2_UPGRADING;
5692 ret = qcow2_upgrade(bs, new_version, &qcow2_amend_helper_cb,
5693 &helper_cb_info, errp);
Max Reitz1038bbb2015-07-27 17:51:35 +02005694 if (ret < 0) {
Max Reitz1038bbb2015-07-27 17:51:35 +02005695 return ret;
Max Reitz9296b3e2013-09-03 10:09:54 +02005696 }
5697 }
5698
Maxim Levitsky90766d92020-06-25 14:55:43 +02005699 if (encryption_update) {
5700 QDict *amend_opts_dict;
5701 QCryptoBlockAmendOptions *amend_opts;
5702
5703 helper_cb_info.current_operation = QCOW2_UPDATING_ENCRYPTION;
5704 amend_opts_dict = qcow2_extract_crypto_opts(opts, "luks", errp);
5705 if (!amend_opts_dict) {
5706 return -EINVAL;
5707 }
5708 amend_opts = block_crypto_amend_opts_init(amend_opts_dict, errp);
5709 qobject_unref(amend_opts_dict);
5710 if (!amend_opts) {
5711 return -EINVAL;
5712 }
5713 ret = qcrypto_block_amend_options(s->crypto,
5714 qcow2_crypto_hdr_read_func,
5715 qcow2_crypto_hdr_write_func,
5716 bs,
5717 amend_opts,
5718 force,
5719 errp);
5720 qapi_free_QCryptoBlockAmendOptions(amend_opts);
5721 if (ret < 0) {
5722 return ret;
5723 }
5724 }
5725
Max Reitz61ce55f2015-07-27 17:51:38 +02005726 if (s->refcount_bits != refcount_bits) {
5727 int refcount_order = ctz32(refcount_bits);
Max Reitz61ce55f2015-07-27 17:51:38 +02005728
5729 if (new_version < 3 && refcount_bits != 16) {
Max Reitzd1402b52018-05-09 23:00:18 +02005730 error_setg(errp, "Refcount widths other than 16 bits require "
5731 "compatibility level 1.1 or above (use compat=1.1 or "
5732 "greater)");
Max Reitz61ce55f2015-07-27 17:51:38 +02005733 return -EINVAL;
5734 }
5735
5736 helper_cb_info.current_operation = QCOW2_CHANGING_REFCOUNT_ORDER;
5737 ret = qcow2_change_refcount_order(bs, refcount_order,
5738 &qcow2_amend_helper_cb,
Max Reitzd1402b52018-05-09 23:00:18 +02005739 &helper_cb_info, errp);
Max Reitz61ce55f2015-07-27 17:51:38 +02005740 if (ret < 0) {
Max Reitz61ce55f2015-07-27 17:51:38 +02005741 return ret;
5742 }
5743 }
5744
Kevin Wolf6c3944d2019-02-22 14:29:38 +01005745 /* data-file-raw blocks backing files, so clear it first if requested */
5746 if (data_file_raw) {
5747 s->autoclear_features |= QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5748 } else {
5749 s->autoclear_features &= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW;
5750 }
5751
Kevin Wolf9b890bd2019-01-15 19:02:40 +01005752 if (data_file) {
5753 g_free(s->image_data_file);
5754 s->image_data_file = *data_file ? g_strdup(data_file) : NULL;
5755 }
5756
5757 ret = qcow2_update_header(bs);
5758 if (ret < 0) {
5759 error_setg_errno(errp, -ret, "Failed to update the image header");
5760 return ret;
5761 }
5762
Max Reitz9296b3e2013-09-03 10:09:54 +02005763 if (backing_file || backing_format) {
Eric Blakebc5ee6d2020-07-06 15:39:51 -05005764 if (g_strcmp0(backing_file, s->image_backing_file) ||
5765 g_strcmp0(backing_format, s->image_backing_format)) {
Eric Blake5a385bf2021-05-03 14:35:59 -07005766 error_setg(errp, "Cannot amend the backing file");
5767 error_append_hint(errp,
5768 "You can use 'qemu-img rebase' instead.\n");
5769 return -EINVAL;
Max Reitz9296b3e2013-09-03 10:09:54 +02005770 }
5771 }
5772
5773 if (s->use_lazy_refcounts != lazy_refcounts) {
5774 if (lazy_refcounts) {
Max Reitz1038bbb2015-07-27 17:51:35 +02005775 if (new_version < 3) {
Max Reitzd1402b52018-05-09 23:00:18 +02005776 error_setg(errp, "Lazy refcounts only supported with "
5777 "compatibility level 1.1 and above (use compat=1.1 "
5778 "or greater)");
Max Reitz9296b3e2013-09-03 10:09:54 +02005779 return -EINVAL;
5780 }
5781 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
5782 ret = qcow2_update_header(bs);
5783 if (ret < 0) {
5784 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
Max Reitzd1402b52018-05-09 23:00:18 +02005785 error_setg_errno(errp, -ret, "Failed to update the image header");
Max Reitz9296b3e2013-09-03 10:09:54 +02005786 return ret;
5787 }
5788 s->use_lazy_refcounts = true;
5789 } else {
5790 /* make image clean first */
5791 ret = qcow2_mark_clean(bs);
5792 if (ret < 0) {
Max Reitzd1402b52018-05-09 23:00:18 +02005793 error_setg_errno(errp, -ret, "Failed to make the image clean");
Max Reitz9296b3e2013-09-03 10:09:54 +02005794 return ret;
5795 }
5796 /* now disallow lazy refcounts */
5797 s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS;
5798 ret = qcow2_update_header(bs);
5799 if (ret < 0) {
5800 s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS;
Max Reitzd1402b52018-05-09 23:00:18 +02005801 error_setg_errno(errp, -ret, "Failed to update the image header");
Max Reitz9296b3e2013-09-03 10:09:54 +02005802 return ret;
5803 }
5804 s->use_lazy_refcounts = false;
5805 }
5806 }
5807
5808 if (new_size) {
Eric Blakea3aeeab2020-04-28 14:26:46 -05005809 BlockBackend *blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL,
5810 errp);
5811 if (!blk) {
5812 return -EPERM;
Kevin Wolfd7086422017-01-13 19:02:32 +01005813 }
5814
Max Reitze8d04f92019-09-18 11:51:43 +02005815 /*
5816 * Amending image options should ensure that the image has
5817 * exactly the given new values, so pass exact=true here.
5818 */
Kevin Wolf8c6242b2020-04-24 14:54:41 +02005819 ret = blk_truncate(blk, new_size, true, PREALLOC_MODE_OFF, 0, errp);
Kevin Wolf70b27f32017-02-17 10:58:25 +01005820 blk_unref(blk);
Max Reitz9296b3e2013-09-03 10:09:54 +02005821 if (ret < 0) {
5822 return ret;
5823 }
5824 }
5825
Max Reitz1038bbb2015-07-27 17:51:35 +02005826 /* Downgrade last (so unsupported features can be removed before) */
5827 if (new_version < old_version) {
Max Reitzc293a802015-07-27 17:51:36 +02005828 helper_cb_info.current_operation = QCOW2_DOWNGRADING;
5829 ret = qcow2_downgrade(bs, new_version, &qcow2_amend_helper_cb,
Max Reitzd1402b52018-05-09 23:00:18 +02005830 &helper_cb_info, errp);
Max Reitz1038bbb2015-07-27 17:51:35 +02005831 if (ret < 0) {
5832 return ret;
5833 }
5834 }
5835
Max Reitz9296b3e2013-09-03 10:09:54 +02005836 return 0;
5837}
5838
Maxim Levitsky8ea16132020-06-25 14:55:47 +02005839static int coroutine_fn qcow2_co_amend(BlockDriverState *bs,
5840 BlockdevAmendOptions *opts,
5841 bool force,
5842 Error **errp)
5843{
5844 BlockdevAmendOptionsQcow2 *qopts = &opts->u.qcow2;
5845 BDRVQcow2State *s = bs->opaque;
5846 int ret = 0;
5847
Markus Armbruster54fde4f2022-11-04 17:06:52 +01005848 if (qopts->encrypt) {
Maxim Levitsky8ea16132020-06-25 14:55:47 +02005849 if (!s->crypto) {
5850 error_setg(errp, "image is not encrypted, can't amend");
5851 return -EOPNOTSUPP;
5852 }
5853
5854 if (qopts->encrypt->format != Q_CRYPTO_BLOCK_FORMAT_LUKS) {
5855 error_setg(errp,
5856 "Amend can't be used to change the qcow2 encryption format");
5857 return -EOPNOTSUPP;
5858 }
5859
5860 if (s->crypt_method_header != QCOW_CRYPT_LUKS) {
5861 error_setg(errp,
5862 "Only LUKS encryption options can be amended for qcow2 with blockdev-amend");
5863 return -EOPNOTSUPP;
5864 }
5865
5866 ret = qcrypto_block_amend_options(s->crypto,
5867 qcow2_crypto_hdr_read_func,
5868 qcow2_crypto_hdr_write_func,
5869 bs,
5870 qopts->encrypt,
5871 force,
5872 errp);
5873 }
5874 return ret;
5875}
5876
Max Reitz85186eb2014-09-05 16:07:16 +02005877/*
5878 * If offset or size are negative, respectively, they will not be included in
5879 * the BLOCK_IMAGE_CORRUPTED event emitted.
5880 * fatal will be ignored for read-only BDS; corruptions found there will always
5881 * be considered non-fatal.
5882 */
5883void qcow2_signal_corruption(BlockDriverState *bs, bool fatal, int64_t offset,
5884 int64_t size, const char *message_format, ...)
5885{
Kevin Wolfff991292015-09-07 17:12:56 +02005886 BDRVQcow2State *s = bs->opaque;
Alberto Garciadc881b42015-04-08 12:29:20 +03005887 const char *node_name;
Max Reitz85186eb2014-09-05 16:07:16 +02005888 char *message;
5889 va_list ap;
5890
Max Reitzddf3b472018-06-06 21:37:01 +02005891 fatal = fatal && bdrv_is_writable(bs);
Max Reitz85186eb2014-09-05 16:07:16 +02005892
5893 if (s->signaled_corruption &&
5894 (!fatal || (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT)))
5895 {
5896 return;
5897 }
5898
5899 va_start(ap, message_format);
5900 message = g_strdup_vprintf(message_format, ap);
5901 va_end(ap);
5902
5903 if (fatal) {
5904 fprintf(stderr, "qcow2: Marking image as corrupt: %s; further "
5905 "corruption events will be suppressed\n", message);
5906 } else {
5907 fprintf(stderr, "qcow2: Image is corrupt: %s; further non-fatal "
5908 "corruption events will be suppressed\n", message);
5909 }
5910
Alberto Garciadc881b42015-04-08 12:29:20 +03005911 node_name = bdrv_get_node_name(bs);
5912 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs),
Markus Armbruster54fde4f2022-11-04 17:06:52 +01005913 *node_name ? node_name : NULL,
Alberto Garciadc881b42015-04-08 12:29:20 +03005914 message, offset >= 0, offset,
5915 size >= 0, size,
Peter Xu3ab72382018-08-15 21:37:37 +08005916 fatal);
Max Reitz85186eb2014-09-05 16:07:16 +02005917 g_free(message);
5918
5919 if (fatal) {
5920 qcow2_mark_corrupt(bs);
5921 bs->drv = NULL; /* make BDS unusable */
5922 }
5923
5924 s->signaled_corruption = true;
5925}
5926
Maxim Levitskydf373fb2020-06-25 14:55:39 +02005927#define QCOW_COMMON_OPTIONS \
5928 { \
5929 .name = BLOCK_OPT_SIZE, \
5930 .type = QEMU_OPT_SIZE, \
5931 .help = "Virtual disk size" \
5932 }, \
5933 { \
5934 .name = BLOCK_OPT_COMPAT_LEVEL, \
5935 .type = QEMU_OPT_STRING, \
5936 .help = "Compatibility level (v2 [0.10] or v3 [1.1])" \
5937 }, \
5938 { \
5939 .name = BLOCK_OPT_BACKING_FILE, \
5940 .type = QEMU_OPT_STRING, \
5941 .help = "File name of a base image" \
5942 }, \
5943 { \
5944 .name = BLOCK_OPT_BACKING_FMT, \
5945 .type = QEMU_OPT_STRING, \
5946 .help = "Image format of the base image" \
5947 }, \
5948 { \
5949 .name = BLOCK_OPT_DATA_FILE, \
5950 .type = QEMU_OPT_STRING, \
5951 .help = "File name of an external data file" \
5952 }, \
5953 { \
5954 .name = BLOCK_OPT_DATA_FILE_RAW, \
5955 .type = QEMU_OPT_BOOL, \
5956 .help = "The external data file must stay valid " \
5957 "as a raw image" \
5958 }, \
5959 { \
Maxim Levitskydf373fb2020-06-25 14:55:39 +02005960 .name = BLOCK_OPT_LAZY_REFCOUNTS, \
5961 .type = QEMU_OPT_BOOL, \
5962 .help = "Postpone refcount updates", \
5963 .def_value_str = "off" \
5964 }, \
5965 { \
5966 .name = BLOCK_OPT_REFCOUNT_BITS, \
5967 .type = QEMU_OPT_NUMBER, \
5968 .help = "Width of a reference count entry in bits", \
5969 .def_value_str = "16" \
Maxim Levitskydf373fb2020-06-25 14:55:39 +02005970 }
5971
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +08005972static QemuOptsList qcow2_create_opts = {
5973 .name = "qcow2-create-opts",
5974 .head = QTAILQ_HEAD_INITIALIZER(qcow2_create_opts.head),
5975 .desc = {
Maxim Levitsky0b6786a2020-06-25 14:55:40 +02005976 { \
5977 .name = BLOCK_OPT_ENCRYPT, \
5978 .type = QEMU_OPT_BOOL, \
5979 .help = "Encrypt the image with format 'aes'. (Deprecated " \
5980 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT "=aes)", \
5981 }, \
5982 { \
5983 .name = BLOCK_OPT_ENCRYPT_FORMAT, \
5984 .type = QEMU_OPT_STRING, \
5985 .help = "Encrypt the image, format choices: 'aes', 'luks'", \
5986 }, \
5987 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.", \
5988 "ID of secret providing qcow AES key or LUKS passphrase"), \
5989 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."), \
5990 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."), \
5991 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."), \
5992 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."), \
5993 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."), \
5994 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."), \
5995 { \
5996 .name = BLOCK_OPT_CLUSTER_SIZE, \
5997 .type = QEMU_OPT_SIZE, \
5998 .help = "qcow2 cluster size", \
5999 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE) \
6000 }, \
6001 { \
Alberto Garcia7be20252020-07-10 18:13:13 +02006002 .name = BLOCK_OPT_EXTL2, \
6003 .type = QEMU_OPT_BOOL, \
6004 .help = "Extended L2 tables", \
6005 .def_value_str = "off" \
6006 }, \
6007 { \
Maxim Levitsky0b6786a2020-06-25 14:55:40 +02006008 .name = BLOCK_OPT_PREALLOC, \
6009 .type = QEMU_OPT_STRING, \
6010 .help = "Preallocation mode (allowed values: off, " \
6011 "metadata, falloc, full)" \
6012 }, \
6013 { \
6014 .name = BLOCK_OPT_COMPRESSION_TYPE, \
6015 .type = QEMU_OPT_STRING, \
6016 .help = "Compression method used for image cluster " \
6017 "compression", \
6018 .def_value_str = "zlib" \
6019 },
Maxim Levitskydf373fb2020-06-25 14:55:39 +02006020 QCOW_COMMON_OPTIONS,
6021 { /* end of list */ }
6022 }
6023};
6024
6025static QemuOptsList qcow2_amend_opts = {
6026 .name = "qcow2-amend-opts",
6027 .head = QTAILQ_HEAD_INITIALIZER(qcow2_amend_opts.head),
6028 .desc = {
Maxim Levitsky90766d92020-06-25 14:55:43 +02006029 BLOCK_CRYPTO_OPT_DEF_LUKS_STATE("encrypt."),
6030 BLOCK_CRYPTO_OPT_DEF_LUKS_KEYSLOT("encrypt."),
6031 BLOCK_CRYPTO_OPT_DEF_LUKS_OLD_SECRET("encrypt."),
6032 BLOCK_CRYPTO_OPT_DEF_LUKS_NEW_SECRET("encrypt."),
6033 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
Maxim Levitskydf373fb2020-06-25 14:55:39 +02006034 QCOW_COMMON_OPTIONS,
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +08006035 { /* end of list */ }
6036 }
Blue Swirl20d97352010-04-23 20:19:47 +00006037};
6038
Max Reitz26542672019-02-01 20:29:25 +01006039static const char *const qcow2_strong_runtime_opts[] = {
6040 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET,
6041
6042 NULL
6043};
6044
Max Reitz5f535a92014-12-02 18:32:41 +01006045BlockDriver bdrv_qcow2 = {
Jes Sorensen7c80ab32010-12-17 16:02:39 +01006046 .format_name = "qcow2",
Kevin Wolfff991292015-09-07 17:12:56 +02006047 .instance_size = sizeof(BDRVQcow2State),
Jes Sorensen7c80ab32010-12-17 16:02:39 +01006048 .bdrv_probe = qcow2_probe,
6049 .bdrv_open = qcow2_open,
6050 .bdrv_close = qcow2_close,
Jeff Cody21d82ac2012-09-20 15:13:28 -04006051 .bdrv_reopen_prepare = qcow2_reopen_prepare,
Kevin Wolf5b0959a2015-04-16 13:42:27 +02006052 .bdrv_reopen_commit = qcow2_reopen_commit,
Peter Krempa65eb7c82020-02-28 13:44:47 +01006053 .bdrv_reopen_commit_post = qcow2_reopen_commit_post,
Kevin Wolf5b0959a2015-04-16 13:42:27 +02006054 .bdrv_reopen_abort = qcow2_reopen_abort,
Kevin Wolf5365f442015-11-16 15:34:59 +01006055 .bdrv_join_options = qcow2_join_options,
Max Reitz69dca432020-05-13 13:05:39 +02006056 .bdrv_child_perm = bdrv_default_perms,
Stefan Hajnocziefc75e22018-01-18 13:43:45 +01006057 .bdrv_co_create_opts = qcow2_co_create_opts,
Kevin Wolfb0292b82018-01-09 16:50:57 +01006058 .bdrv_co_create = qcow2_co_create,
Max Reitz38841dc2019-07-24 19:12:34 +02006059 .bdrv_has_zero_init = qcow2_has_zero_init,
Eric Blakea320fb02018-02-13 14:26:52 -06006060 .bdrv_co_block_status = qcow2_co_block_status,
Blue Swirl20d97352010-04-23 20:19:47 +00006061
Vladimir Sementsov-Ogievskiydf893d22019-06-04 19:15:13 +03006062 .bdrv_co_preadv_part = qcow2_co_preadv_part,
Vladimir Sementsov-Ogievskiy53962342019-06-04 19:15:14 +03006063 .bdrv_co_pwritev_part = qcow2_co_pwritev_part,
Kevin Wolfeb489bb2011-11-10 18:10:11 +01006064 .bdrv_co_flush_to_os = qcow2_co_flush_to_os,
Stefan Hajnoczi419b19d2010-04-28 11:36:11 +01006065
Eric Blake5544b592016-06-01 15:10:06 -06006066 .bdrv_co_pwrite_zeroes = qcow2_co_pwrite_zeroes,
Eric Blake82e8a782016-07-15 17:23:03 -06006067 .bdrv_co_pdiscard = qcow2_co_pdiscard,
Fam Zhengfd9fcd32018-06-01 17:26:42 +08006068 .bdrv_co_copy_range_from = qcow2_co_copy_range_from,
6069 .bdrv_co_copy_range_to = qcow2_co_copy_range_to,
Kevin Wolf061ca8a2018-06-21 17:54:35 +02006070 .bdrv_co_truncate = qcow2_co_truncate,
Vladimir Sementsov-Ogievskiy53962342019-06-04 19:15:14 +03006071 .bdrv_co_pwritev_compressed_part = qcow2_co_pwritev_compressed_part,
Max Reitz491d27e2014-10-24 15:57:31 +02006072 .bdrv_make_empty = qcow2_make_empty,
Blue Swirl20d97352010-04-23 20:19:47 +00006073
6074 .bdrv_snapshot_create = qcow2_snapshot_create,
6075 .bdrv_snapshot_goto = qcow2_snapshot_goto,
6076 .bdrv_snapshot_delete = qcow2_snapshot_delete,
6077 .bdrv_snapshot_list = qcow2_snapshot_list,
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +08006078 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
Stefan Hajnoczic501c352017-07-05 13:57:35 +01006079 .bdrv_measure = qcow2_measure,
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +08006080 .bdrv_get_info = qcow2_get_info,
Max Reitz37764df2013-10-09 10:46:18 +02006081 .bdrv_get_specific_info = qcow2_get_specific_info,
Blue Swirl20d97352010-04-23 20:19:47 +00006082
Jes Sorensen7c80ab32010-12-17 16:02:39 +01006083 .bdrv_save_vmstate = qcow2_save_vmstate,
6084 .bdrv_load_vmstate = qcow2_load_vmstate,
Blue Swirl20d97352010-04-23 20:19:47 +00006085
Max Reitzd67066d2020-05-13 13:05:12 +02006086 .is_format = true,
Kevin Wolf8ee79e72014-06-04 15:09:35 +02006087 .supports_backing = true,
Blue Swirl20d97352010-04-23 20:19:47 +00006088 .bdrv_change_backing_file = qcow2_change_backing_file,
6089
Kevin Wolfd34682c2013-12-11 19:26:16 +01006090 .bdrv_refresh_limits = qcow2_refresh_limits,
Paolo Bonzini2b148f32018-03-01 17:36:18 +01006091 .bdrv_co_invalidate_cache = qcow2_co_invalidate_cache,
Kevin Wolfec6d8912015-12-22 16:04:57 +01006092 .bdrv_inactivate = qcow2_inactivate,
Anthony Liguori06d92602011-11-14 15:09:46 -06006093
Chunyan Liu1bd0e2d2014-06-05 17:20:59 +08006094 .create_opts = &qcow2_create_opts,
Maxim Levitskydf373fb2020-06-25 14:55:39 +02006095 .amend_opts = &qcow2_amend_opts,
Max Reitz26542672019-02-01 20:29:25 +01006096 .strong_runtime_opts = qcow2_strong_runtime_opts,
Alberto Garcia8a2ce0b2019-03-12 18:48:48 +02006097 .mutable_opts = mutable_opts,
Paolo Bonzini2fd61632018-03-01 17:36:19 +01006098 .bdrv_co_check = qcow2_co_check,
Chunyan Liuc282e1f2014-06-05 17:21:11 +08006099 .bdrv_amend_options = qcow2_amend_options,
Maxim Levitsky8ea16132020-06-25 14:55:47 +02006100 .bdrv_co_amend = qcow2_co_amend,
Alberto Garcia279621c2015-08-04 15:14:40 +03006101
6102 .bdrv_detach_aio_context = qcow2_detach_aio_context,
6103 .bdrv_attach_aio_context = qcow2_attach_aio_context,
Vladimir Sementsov-Ogievskiy1b6b0562017-06-28 15:05:14 +03006104
Eric Blakeef893b52020-05-12 20:16:42 -05006105 .bdrv_supports_persistent_dirty_bitmap =
6106 qcow2_supports_persistent_dirty_bitmap,
Vladimir Sementsov-Ogievskiyd2c30802019-09-20 11:25:43 +03006107 .bdrv_co_can_store_new_dirty_bitmap = qcow2_co_can_store_new_dirty_bitmap,
6108 .bdrv_co_remove_persistent_dirty_bitmap =
6109 qcow2_co_remove_persistent_dirty_bitmap,
Blue Swirl20d97352010-04-23 20:19:47 +00006110};
6111
Anthony Liguori5efa9d52009-05-09 17:03:42 -05006112static void bdrv_qcow2_init(void)
6113{
6114 bdrv_register(&bdrv_qcow2);
6115}
6116
6117block_init(bdrv_qcow2_init);