blob: 1b9c91128ac02440c4a5ca4661d51376fd1e2798 [file] [log] [blame]
bellardea2384d2004-08-01 21:59:26 +00001/*
2 * Block driver for the QCOW format
ths5fafdf22007-09-16 21:08:06 +00003 *
bellard83f64092006-08-01 16:21:11 +00004 * Copyright (c) 2004-2006 Fabrice Bellard
ths5fafdf22007-09-16 21:08:06 +00005 *
bellardea2384d2004-08-01 21:59:26 +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 */
Peter Maydell80c71a22016-01-18 18:01:42 +000024#include "qemu/osdep.h"
Markus Armbrusterda34e652016-03-14 09:01:28 +010025#include "qapi/error.h"
pbrookfaf07962007-11-11 02:51:17 +000026#include "qemu-common.h"
Daniel P. Berrangee6ff69b2016-03-21 14:11:48 +000027#include "qemu/error-report.h"
Paolo Bonzini737e1502012-12-17 18:19:44 +010028#include "block/block_int.h"
Kevin Wolf6af40162016-03-08 15:57:05 +010029#include "sysemu/block-backend.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010030#include "qemu/module.h"
Paolo Bonzini58369e22016-03-15 17:22:36 +010031#include "qemu/bswap.h"
bellard28d34b82004-09-29 21:30:43 +000032#include <zlib.h>
Markus Armbrustercc7a8ea2015-03-17 17:22:46 +010033#include "qapi/qmp/qerror.h"
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +010034#include "crypto/cipher.h"
Paolo Bonzinicaf71f82012-12-17 18:19:50 +010035#include "migration/migration.h"
bellardea2384d2004-08-01 21:59:26 +000036
37/**************************************************************/
38/* QEMU COW block driver with compression and encryption support */
39
40#define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
41#define QCOW_VERSION 1
42
43#define QCOW_CRYPT_NONE 0
44#define QCOW_CRYPT_AES 1
45
46#define QCOW_OFLAG_COMPRESSED (1LL << 63)
47
48typedef struct QCowHeader {
49 uint32_t magic;
50 uint32_t version;
51 uint64_t backing_file_offset;
52 uint32_t backing_file_size;
53 uint32_t mtime;
54 uint64_t size; /* in bytes */
55 uint8_t cluster_bits;
56 uint8_t l2_bits;
Kevin Wolfea54fef2014-05-07 16:56:10 +020057 uint16_t padding;
bellardea2384d2004-08-01 21:59:26 +000058 uint32_t crypt_method;
59 uint64_t l1_table_offset;
Kevin Wolfea54fef2014-05-07 16:56:10 +020060} QEMU_PACKED QCowHeader;
bellardea2384d2004-08-01 21:59:26 +000061
62#define L2_CACHE_SIZE 16
63
64typedef struct BDRVQcowState {
bellardea2384d2004-08-01 21:59:26 +000065 int cluster_bits;
66 int cluster_size;
67 int cluster_sectors;
68 int l2_bits;
69 int l2_size;
Kevin Wolf46485de2014-05-08 13:08:20 +020070 unsigned int l1_size;
bellardea2384d2004-08-01 21:59:26 +000071 uint64_t cluster_offset_mask;
72 uint64_t l1_table_offset;
73 uint64_t *l1_table;
74 uint64_t *l2_cache;
75 uint64_t l2_cache_offsets[L2_CACHE_SIZE];
76 uint32_t l2_cache_counts[L2_CACHE_SIZE];
77 uint8_t *cluster_cache;
78 uint8_t *cluster_data;
79 uint64_t cluster_cache_offset;
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +010080 QCryptoCipher *cipher; /* NULL if no key yet */
bellardea2384d2004-08-01 21:59:26 +000081 uint32_t crypt_method_header;
Kevin Wolf52b8eb62011-07-15 16:27:42 +020082 CoMutex lock;
Kevin Wolffd9f1022011-11-22 16:44:45 +010083 Error *migration_blocker;
bellardea2384d2004-08-01 21:59:26 +000084} BDRVQcowState;
85
Kevin Wolf66f82ce2010-04-14 14:17:38 +020086static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
bellardea2384d2004-08-01 21:59:26 +000087
88static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
89{
90 const QCowHeader *cow_header = (const void *)buf;
ths3b46e622007-09-17 08:09:54 +000091
bellard712e7872005-04-28 21:09:32 +000092 if (buf_size >= sizeof(QCowHeader) &&
93 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
ths5fafdf22007-09-16 21:08:06 +000094 be32_to_cpu(cow_header->version) == QCOW_VERSION)
bellardea2384d2004-08-01 21:59:26 +000095 return 100;
96 else
97 return 0;
98}
99
Max Reitz015a1032013-09-05 14:22:29 +0200100static int qcow_open(BlockDriverState *bs, QDict *options, int flags,
101 Error **errp)
bellardea2384d2004-08-01 21:59:26 +0000102{
103 BDRVQcowState *s = bs->opaque;
Kevin Wolfd66e5ce2014-05-08 13:35:09 +0200104 unsigned int len, i, shift;
105 int ret;
bellardea2384d2004-08-01 21:59:26 +0000106 QCowHeader header;
bellard83f64092006-08-01 16:21:11 +0000107
Kevin Wolfcf2ab8f2016-06-20 18:24:02 +0200108 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800109 if (ret < 0) {
bellardea2384d2004-08-01 21:59:26 +0000110 goto fail;
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800111 }
bellardea2384d2004-08-01 21:59:26 +0000112 be32_to_cpus(&header.magic);
113 be32_to_cpus(&header.version);
114 be64_to_cpus(&header.backing_file_offset);
115 be32_to_cpus(&header.backing_file_size);
116 be32_to_cpus(&header.mtime);
117 be64_to_cpus(&header.size);
118 be32_to_cpus(&header.crypt_method);
119 be64_to_cpus(&header.l1_table_offset);
ths3b46e622007-09-17 08:09:54 +0000120
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800121 if (header.magic != QCOW_MAGIC) {
Paolo Bonzini76abe402014-02-17 14:44:06 +0100122 error_setg(errp, "Image not in qcow format");
123 ret = -EINVAL;
bellardea2384d2004-08-01 21:59:26 +0000124 goto fail;
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800125 }
126 if (header.version != QCOW_VERSION) {
Max Reitza55448b2016-03-16 19:54:33 +0100127 error_setg(errp, "Unsupported qcow version %" PRIu32, header.version);
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800128 ret = -ENOTSUP;
bellardea2384d2004-08-01 21:59:26 +0000129 goto fail;
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800130 }
131
Kevin Wolf7159a452014-05-07 17:30:30 +0200132 if (header.size <= 1) {
133 error_setg(errp, "Image size is too small (must be at least 2 bytes)");
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800134 ret = -EINVAL;
bellardea2384d2004-08-01 21:59:26 +0000135 goto fail;
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800136 }
Kevin Wolf7159a452014-05-07 17:30:30 +0200137 if (header.cluster_bits < 9 || header.cluster_bits > 16) {
138 error_setg(errp, "Cluster size must be between 512 and 64k");
139 ret = -EINVAL;
140 goto fail;
141 }
142
Kevin Wolf42eb5812014-05-15 16:10:11 +0200143 /* l2_bits specifies number of entries; storing a uint64_t in each entry,
144 * so bytes = num_entries << 3. */
145 if (header.l2_bits < 9 - 3 || header.l2_bits > 16 - 3) {
146 error_setg(errp, "L2 table size must be between 512 and 64k");
147 ret = -EINVAL;
148 goto fail;
149 }
150
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800151 if (header.crypt_method > QCOW_CRYPT_AES) {
Paolo Bonzinib6d50662014-02-17 14:43:58 +0100152 error_setg(errp, "invalid encryption method in qcow header");
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800153 ret = -EINVAL;
154 goto fail;
155 }
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100156 if (!qcrypto_cipher_supports(QCRYPTO_CIPHER_ALG_AES_128)) {
157 error_setg(errp, "AES cipher not available");
158 ret = -EINVAL;
159 goto fail;
160 }
bellardea2384d2004-08-01 21:59:26 +0000161 s->crypt_method_header = header.crypt_method;
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800162 if (s->crypt_method_header) {
Daniel P. Berrangee6ff69b2016-03-21 14:11:48 +0000163 if (bdrv_uses_whitelist() &&
164 s->crypt_method_header == QCOW_CRYPT_AES) {
Daniel P. Berrange8c0dcbc2016-06-13 12:30:09 +0100165 error_setg(errp,
166 "Use of AES-CBC encrypted qcow images is no longer "
167 "supported in system emulators");
168 error_append_hint(errp,
169 "You can use 'qemu-img convert' to convert your "
170 "image to an alternative supported format, such "
171 "as unencrypted qcow, or raw with the LUKS "
172 "format instead.\n");
173 ret = -ENOSYS;
174 goto fail;
Daniel P. Berrangee6ff69b2016-03-21 14:11:48 +0000175 }
176
Eric Blake54115412016-06-23 16:37:26 -0600177 bs->encrypted = true;
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800178 }
bellardea2384d2004-08-01 21:59:26 +0000179 s->cluster_bits = header.cluster_bits;
180 s->cluster_size = 1 << s->cluster_bits;
181 s->cluster_sectors = 1 << (s->cluster_bits - 9);
182 s->l2_bits = header.l2_bits;
183 s->l2_size = 1 << s->l2_bits;
184 bs->total_sectors = header.size / 512;
185 s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
186
187 /* read the level 1 table */
188 shift = s->cluster_bits + s->l2_bits;
Kevin Wolf46485de2014-05-08 13:08:20 +0200189 if (header.size > UINT64_MAX - (1LL << shift)) {
190 error_setg(errp, "Image too large");
191 ret = -EINVAL;
192 goto fail;
193 } else {
194 uint64_t l1_size = (header.size + (1LL << shift) - 1) >> shift;
195 if (l1_size > INT_MAX / sizeof(uint64_t)) {
196 error_setg(errp, "Image too large");
197 ret = -EINVAL;
198 goto fail;
199 }
200 s->l1_size = l1_size;
201 }
bellardea2384d2004-08-01 21:59:26 +0000202
203 s->l1_table_offset = header.l1_table_offset;
Markus Armbruster5839e532014-08-19 10:31:08 +0200204 s->l1_table = g_try_new(uint64_t, s->l1_size);
Kevin Wolf0df93302014-05-20 13:36:05 +0200205 if (s->l1_table == NULL) {
206 error_setg(errp, "Could not allocate memory for L1 table");
207 ret = -ENOMEM;
208 goto fail;
209 }
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800210
Kevin Wolfcf2ab8f2016-06-20 18:24:02 +0200211 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800212 s->l1_size * sizeof(uint64_t));
213 if (ret < 0) {
bellardea2384d2004-08-01 21:59:26 +0000214 goto fail;
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800215 }
216
bellardea2384d2004-08-01 21:59:26 +0000217 for(i = 0;i < s->l1_size; i++) {
218 be64_to_cpus(&s->l1_table[i]);
219 }
Kevin Wolf0df93302014-05-20 13:36:05 +0200220
221 /* alloc L2 cache (max. 64k * 16 * 8 = 8 MB) */
222 s->l2_cache =
Kevin Wolf9a4f4c32015-06-16 14:19:22 +0200223 qemu_try_blockalign(bs->file->bs,
Kevin Wolf0df93302014-05-20 13:36:05 +0200224 s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
225 if (s->l2_cache == NULL) {
226 error_setg(errp, "Could not allocate L2 table cache");
227 ret = -ENOMEM;
228 goto fail;
229 }
Anthony Liguori7267c092011-08-20 22:09:37 -0500230 s->cluster_cache = g_malloc(s->cluster_size);
Anthony Liguori7267c092011-08-20 22:09:37 -0500231 s->cluster_data = g_malloc(s->cluster_size);
bellardea2384d2004-08-01 21:59:26 +0000232 s->cluster_cache_offset = -1;
ths3b46e622007-09-17 08:09:54 +0000233
bellardea2384d2004-08-01 21:59:26 +0000234 /* read the backing file name */
235 if (header.backing_file_offset != 0) {
236 len = header.backing_file_size;
Jeff Codye729fa62015-01-27 08:33:55 -0500237 if (len > 1023 || len >= sizeof(bs->backing_file)) {
Kevin Wolfd66e5ce2014-05-08 13:35:09 +0200238 error_setg(errp, "Backing file name too long");
239 ret = -EINVAL;
240 goto fail;
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800241 }
Kevin Wolfcf2ab8f2016-06-20 18:24:02 +0200242 ret = bdrv_pread(bs->file, header.backing_file_offset,
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800243 bs->backing_file, len);
244 if (ret < 0) {
bellardea2384d2004-08-01 21:59:26 +0000245 goto fail;
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800246 }
bellardea2384d2004-08-01 21:59:26 +0000247 bs->backing_file[len] = '\0';
248 }
Scott Woodde33b1f2011-08-11 16:27:15 -0500249
Kevin Wolffd9f1022011-11-22 16:44:45 +0100250 /* Disable migration when qcow images are used */
Alberto Garcia81e5f782015-04-08 12:29:19 +0300251 error_setg(&s->migration_blocker, "The qcow format used by node '%s' "
252 "does not support live migration",
253 bdrv_get_device_or_node_name(bs));
Kevin Wolffd9f1022011-11-22 16:44:45 +0100254 migrate_add_blocker(s->migration_blocker);
255
Scott Woodde33b1f2011-08-11 16:27:15 -0500256 qemu_co_mutex_init(&s->lock);
bellardea2384d2004-08-01 21:59:26 +0000257 return 0;
258
259 fail:
Anthony Liguori7267c092011-08-20 22:09:37 -0500260 g_free(s->l1_table);
Kevin Wolf0df93302014-05-20 13:36:05 +0200261 qemu_vfree(s->l2_cache);
Anthony Liguori7267c092011-08-20 22:09:37 -0500262 g_free(s->cluster_cache);
263 g_free(s->cluster_data);
Li Zhi Hui84b0ec02011-12-15 18:14:00 +0800264 return ret;
bellardea2384d2004-08-01 21:59:26 +0000265}
266
Jeff Codyd1776922012-09-20 15:13:29 -0400267
268/* We have nothing to do for QCOW reopen, stubs just return
269 * success */
270static int qcow_reopen_prepare(BDRVReopenState *state,
271 BlockReopenQueue *queue, Error **errp)
272{
273 return 0;
274}
275
bellardea2384d2004-08-01 21:59:26 +0000276static int qcow_set_key(BlockDriverState *bs, const char *key)
277{
278 BDRVQcowState *s = bs->opaque;
279 uint8_t keybuf[16];
280 int len, i;
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100281 Error *err;
ths3b46e622007-09-17 08:09:54 +0000282
bellardea2384d2004-08-01 21:59:26 +0000283 memset(keybuf, 0, 16);
284 len = strlen(key);
285 if (len > 16)
286 len = 16;
287 /* XXX: we could compress the chars to 7 bits to increase
288 entropy */
289 for(i = 0;i < len;i++) {
290 keybuf[i] = key[i];
291 }
Daniel P. Berrange8336aaf2015-05-12 17:09:18 +0100292 assert(bs->encrypted);
bellardea2384d2004-08-01 21:59:26 +0000293
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100294 qcrypto_cipher_free(s->cipher);
295 s->cipher = qcrypto_cipher_new(
296 QCRYPTO_CIPHER_ALG_AES_128,
297 QCRYPTO_CIPHER_MODE_CBC,
298 keybuf, G_N_ELEMENTS(keybuf),
299 &err);
300
301 if (!s->cipher) {
302 /* XXX would be nice if errors in this method could
303 * be properly propagate to the caller. Would need
304 * the bdrv_set_key() API signature to be fixed. */
305 error_free(err);
bellardea2384d2004-08-01 21:59:26 +0000306 return -1;
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100307 }
bellardea2384d2004-08-01 21:59:26 +0000308 return 0;
309}
310
311/* The crypt function is compatible with the linux cryptoloop
312 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
313 supported */
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100314static int encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
315 uint8_t *out_buf, const uint8_t *in_buf,
316 int nb_sectors, bool enc, Error **errp)
bellardea2384d2004-08-01 21:59:26 +0000317{
318 union {
319 uint64_t ll[2];
320 uint8_t b[16];
321 } ivec;
322 int i;
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100323 int ret;
bellardea2384d2004-08-01 21:59:26 +0000324
325 for(i = 0; i < nb_sectors; i++) {
326 ivec.ll[0] = cpu_to_le64(sector_num);
327 ivec.ll[1] = 0;
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100328 if (qcrypto_cipher_setiv(s->cipher,
329 ivec.b, G_N_ELEMENTS(ivec.b),
330 errp) < 0) {
331 return -1;
332 }
333 if (enc) {
334 ret = qcrypto_cipher_encrypt(s->cipher,
335 in_buf,
336 out_buf,
337 512,
338 errp);
339 } else {
340 ret = qcrypto_cipher_decrypt(s->cipher,
341 in_buf,
342 out_buf,
343 512,
344 errp);
345 }
346 if (ret < 0) {
347 return -1;
348 }
bellardea2384d2004-08-01 21:59:26 +0000349 sector_num++;
350 in_buf += 512;
351 out_buf += 512;
352 }
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100353 return 0;
bellardea2384d2004-08-01 21:59:26 +0000354}
355
356/* 'allocate' is:
357 *
358 * 0 to not allocate.
359 *
360 * 1 to allocate a normal cluster (for sector indexes 'n_start' to
361 * 'n_end')
362 *
363 * 2 to allocate a compressed cluster of size
364 * 'compressed_size'. 'compressed_size' must be > 0 and <
ths5fafdf22007-09-16 21:08:06 +0000365 * cluster_size
bellardea2384d2004-08-01 21:59:26 +0000366 *
367 * return 0 if not allocated.
368 */
369static uint64_t get_cluster_offset(BlockDriverState *bs,
370 uint64_t offset, int allocate,
371 int compressed_size,
372 int n_start, int n_end)
373{
374 BDRVQcowState *s = bs->opaque;
375 int min_index, i, j, l1_index, l2_index;
376 uint64_t l2_offset, *l2_table, cluster_offset, tmp;
377 uint32_t min_count;
378 int new_l2_table;
ths3b46e622007-09-17 08:09:54 +0000379
bellardea2384d2004-08-01 21:59:26 +0000380 l1_index = offset >> (s->l2_bits + s->cluster_bits);
381 l2_offset = s->l1_table[l1_index];
382 new_l2_table = 0;
383 if (!l2_offset) {
384 if (!allocate)
385 return 0;
386 /* allocate a new l2 entry */
Kevin Wolf9a4f4c32015-06-16 14:19:22 +0200387 l2_offset = bdrv_getlength(bs->file->bs);
bellardea2384d2004-08-01 21:59:26 +0000388 /* round to cluster size */
389 l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
390 /* update the L1 entry */
391 s->l1_table[l1_index] = l2_offset;
392 tmp = cpu_to_be64(l2_offset);
Kevin Wolfd9ca2ea2016-06-20 20:09:15 +0200393 if (bdrv_pwrite_sync(bs->file,
Kevin Wolf5e5557d2010-06-18 16:11:53 +0200394 s->l1_table_offset + l1_index * sizeof(tmp),
395 &tmp, sizeof(tmp)) < 0)
bellardea2384d2004-08-01 21:59:26 +0000396 return 0;
397 new_l2_table = 1;
398 }
399 for(i = 0; i < L2_CACHE_SIZE; i++) {
400 if (l2_offset == s->l2_cache_offsets[i]) {
401 /* increment the hit count */
402 if (++s->l2_cache_counts[i] == 0xffffffff) {
403 for(j = 0; j < L2_CACHE_SIZE; j++) {
404 s->l2_cache_counts[j] >>= 1;
405 }
406 }
407 l2_table = s->l2_cache + (i << s->l2_bits);
408 goto found;
409 }
410 }
411 /* not found: load a new entry in the least used one */
412 min_index = 0;
413 min_count = 0xffffffff;
414 for(i = 0; i < L2_CACHE_SIZE; i++) {
415 if (s->l2_cache_counts[i] < min_count) {
416 min_count = s->l2_cache_counts[i];
417 min_index = i;
418 }
419 }
420 l2_table = s->l2_cache + (min_index << s->l2_bits);
bellardea2384d2004-08-01 21:59:26 +0000421 if (new_l2_table) {
422 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
Kevin Wolfd9ca2ea2016-06-20 20:09:15 +0200423 if (bdrv_pwrite_sync(bs->file, l2_offset, l2_table,
Kevin Wolf5e5557d2010-06-18 16:11:53 +0200424 s->l2_size * sizeof(uint64_t)) < 0)
bellardea2384d2004-08-01 21:59:26 +0000425 return 0;
426 } else {
Kevin Wolfcf2ab8f2016-06-20 18:24:02 +0200427 if (bdrv_pread(bs->file, l2_offset, l2_table,
Kevin Wolf9a4f4c32015-06-16 14:19:22 +0200428 s->l2_size * sizeof(uint64_t)) !=
bellardea2384d2004-08-01 21:59:26 +0000429 s->l2_size * sizeof(uint64_t))
430 return 0;
431 }
432 s->l2_cache_offsets[min_index] = l2_offset;
433 s->l2_cache_counts[min_index] = 1;
434 found:
435 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
436 cluster_offset = be64_to_cpu(l2_table[l2_index]);
ths5fafdf22007-09-16 21:08:06 +0000437 if (!cluster_offset ||
bellardea2384d2004-08-01 21:59:26 +0000438 ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
439 if (!allocate)
440 return 0;
441 /* allocate a new cluster */
442 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
443 (n_end - n_start) < s->cluster_sectors) {
444 /* if the cluster is already compressed, we must
445 decompress it in the case it is not completely
446 overwritten */
Kevin Wolf66f82ce2010-04-14 14:17:38 +0200447 if (decompress_cluster(bs, cluster_offset) < 0)
bellardea2384d2004-08-01 21:59:26 +0000448 return 0;
Kevin Wolf9a4f4c32015-06-16 14:19:22 +0200449 cluster_offset = bdrv_getlength(bs->file->bs);
ths5fafdf22007-09-16 21:08:06 +0000450 cluster_offset = (cluster_offset + s->cluster_size - 1) &
bellardea2384d2004-08-01 21:59:26 +0000451 ~(s->cluster_size - 1);
452 /* write the cluster content */
Kevin Wolfd9ca2ea2016-06-20 20:09:15 +0200453 if (bdrv_pwrite(bs->file, cluster_offset, s->cluster_cache,
Kevin Wolf9a4f4c32015-06-16 14:19:22 +0200454 s->cluster_size) !=
bellardea2384d2004-08-01 21:59:26 +0000455 s->cluster_size)
456 return -1;
457 } else {
Kevin Wolf9a4f4c32015-06-16 14:19:22 +0200458 cluster_offset = bdrv_getlength(bs->file->bs);
aliguori7f48fa12009-01-08 19:29:03 +0000459 if (allocate == 1) {
460 /* round to cluster size */
461 cluster_offset = (cluster_offset + s->cluster_size - 1) &
462 ~(s->cluster_size - 1);
Kevin Wolf9a4f4c32015-06-16 14:19:22 +0200463 bdrv_truncate(bs->file->bs, cluster_offset + s->cluster_size);
aliguori7f48fa12009-01-08 19:29:03 +0000464 /* if encrypted, we must initialize the cluster
465 content which won't be written */
Daniel P. Berrange8336aaf2015-05-12 17:09:18 +0100466 if (bs->encrypted &&
aliguori7f48fa12009-01-08 19:29:03 +0000467 (n_end - n_start) < s->cluster_sectors) {
468 uint64_t start_sect;
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100469 assert(s->cipher);
aliguori7f48fa12009-01-08 19:29:03 +0000470 start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
471 memset(s->cluster_data + 512, 0x00, 512);
472 for(i = 0; i < s->cluster_sectors; i++) {
473 if (i < n_start || i >= n_end) {
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100474 Error *err = NULL;
475 if (encrypt_sectors(s, start_sect + i,
476 s->cluster_data,
477 s->cluster_data + 512, 1,
478 true, &err) < 0) {
479 error_free(err);
480 errno = EIO;
481 return -1;
482 }
Kevin Wolfd9ca2ea2016-06-20 20:09:15 +0200483 if (bdrv_pwrite(bs->file,
Kevin Wolf9a4f4c32015-06-16 14:19:22 +0200484 cluster_offset + i * 512,
aliguori7f48fa12009-01-08 19:29:03 +0000485 s->cluster_data, 512) != 512)
486 return -1;
487 }
bellardea2384d2004-08-01 21:59:26 +0000488 }
489 }
aliguori7f48fa12009-01-08 19:29:03 +0000490 } else if (allocate == 2) {
491 cluster_offset |= QCOW_OFLAG_COMPRESSED |
492 (uint64_t)compressed_size << (63 - s->cluster_bits);
bellardea2384d2004-08-01 21:59:26 +0000493 }
494 }
495 /* update L2 table */
496 tmp = cpu_to_be64(cluster_offset);
497 l2_table[l2_index] = tmp;
Kevin Wolfd9ca2ea2016-06-20 20:09:15 +0200498 if (bdrv_pwrite_sync(bs->file, l2_offset + l2_index * sizeof(tmp),
Kevin Wolf5e5557d2010-06-18 16:11:53 +0200499 &tmp, sizeof(tmp)) < 0)
bellardea2384d2004-08-01 21:59:26 +0000500 return 0;
501 }
502 return cluster_offset;
503}
504
Paolo Bonzinib6b8a332013-09-04 19:00:28 +0200505static int64_t coroutine_fn qcow_co_get_block_status(BlockDriverState *bs,
Fam Zheng67a0fd22016-01-26 11:58:48 +0800506 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
bellardea2384d2004-08-01 21:59:26 +0000507{
508 BDRVQcowState *s = bs->opaque;
509 int index_in_cluster, n;
510 uint64_t cluster_offset;
511
Stefan Hajnoczif8a2e5e2011-11-14 12:44:21 +0000512 qemu_co_mutex_lock(&s->lock);
bellardea2384d2004-08-01 21:59:26 +0000513 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
Stefan Hajnoczif8a2e5e2011-11-14 12:44:21 +0000514 qemu_co_mutex_unlock(&s->lock);
bellardea2384d2004-08-01 21:59:26 +0000515 index_in_cluster = sector_num & (s->cluster_sectors - 1);
516 n = s->cluster_sectors - index_in_cluster;
517 if (n > nb_sectors)
518 n = nb_sectors;
519 *pnum = n;
Paolo Bonzini4bc74be2013-09-04 19:00:30 +0200520 if (!cluster_offset) {
521 return 0;
522 }
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100523 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) || s->cipher) {
Paolo Bonzini4bc74be2013-09-04 19:00:30 +0200524 return BDRV_BLOCK_DATA;
525 }
526 cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS);
Fam Zheng3064bf62016-01-26 11:58:49 +0800527 *file = bs->file->bs;
Paolo Bonzini4bc74be2013-09-04 19:00:30 +0200528 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | cluster_offset;
bellardea2384d2004-08-01 21:59:26 +0000529}
530
531static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
532 const uint8_t *buf, int buf_size)
533{
534 z_stream strm1, *strm = &strm1;
535 int ret, out_len;
536
537 memset(strm, 0, sizeof(*strm));
538
539 strm->next_in = (uint8_t *)buf;
540 strm->avail_in = buf_size;
541 strm->next_out = out_buf;
542 strm->avail_out = out_buf_size;
543
544 ret = inflateInit2(strm, -12);
545 if (ret != Z_OK)
546 return -1;
547 ret = inflate(strm, Z_FINISH);
548 out_len = strm->next_out - out_buf;
549 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
550 out_len != out_buf_size) {
551 inflateEnd(strm);
552 return -1;
553 }
554 inflateEnd(strm);
555 return 0;
556}
ths3b46e622007-09-17 08:09:54 +0000557
Kevin Wolf66f82ce2010-04-14 14:17:38 +0200558static int decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
bellardea2384d2004-08-01 21:59:26 +0000559{
Kevin Wolf66f82ce2010-04-14 14:17:38 +0200560 BDRVQcowState *s = bs->opaque;
bellardea2384d2004-08-01 21:59:26 +0000561 int ret, csize;
562 uint64_t coffset;
563
564 coffset = cluster_offset & s->cluster_offset_mask;
565 if (s->cluster_cache_offset != coffset) {
566 csize = cluster_offset >> (63 - s->cluster_bits);
567 csize &= (s->cluster_size - 1);
Kevin Wolfcf2ab8f2016-06-20 18:24:02 +0200568 ret = bdrv_pread(bs->file, coffset, s->cluster_data, csize);
ths5fafdf22007-09-16 21:08:06 +0000569 if (ret != csize)
bellardea2384d2004-08-01 21:59:26 +0000570 return -1;
571 if (decompress_buffer(s->cluster_cache, s->cluster_size,
572 s->cluster_data, csize) < 0) {
573 return -1;
574 }
575 s->cluster_cache_offset = coffset;
576 }
577 return 0;
578}
579
Dong Xu Wanga9681682011-11-10 16:23:22 +0800580static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
Kevin Wolf52b8eb62011-07-15 16:27:42 +0200581 int nb_sectors, QEMUIOVector *qiov)
582{
583 BDRVQcowState *s = bs->opaque;
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200584 int index_in_cluster;
585 int ret = 0, n;
586 uint64_t cluster_offset;
587 struct iovec hd_iov;
588 QEMUIOVector hd_qiov;
589 uint8_t *buf;
590 void *orig_buf;
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100591 Error *err = NULL;
Kevin Wolf52b8eb62011-07-15 16:27:42 +0200592
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200593 if (qiov->niov > 1) {
Kevin Wolf0df93302014-05-20 13:36:05 +0200594 buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
595 if (buf == NULL) {
596 return -ENOMEM;
597 }
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200598 } else {
599 orig_buf = NULL;
600 buf = (uint8_t *)qiov->iov->iov_base;
601 }
Kevin Wolf52b8eb62011-07-15 16:27:42 +0200602
603 qemu_co_mutex_lock(&s->lock);
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200604
605 while (nb_sectors != 0) {
606 /* prepare next request */
607 cluster_offset = get_cluster_offset(bs, sector_num << 9,
608 0, 0, 0, 0);
609 index_in_cluster = sector_num & (s->cluster_sectors - 1);
610 n = s->cluster_sectors - index_in_cluster;
611 if (n > nb_sectors) {
612 n = nb_sectors;
613 }
614
615 if (!cluster_offset) {
Kevin Wolf760e0062015-06-17 14:55:21 +0200616 if (bs->backing) {
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200617 /* read from the base image */
618 hd_iov.iov_base = (void *)buf;
619 hd_iov.iov_len = n * 512;
620 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
621 qemu_co_mutex_unlock(&s->lock);
Kevin Wolf28b04a82016-05-24 17:21:22 +0200622 ret = bdrv_co_readv(bs->backing, sector_num, n, &hd_qiov);
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200623 qemu_co_mutex_lock(&s->lock);
624 if (ret < 0) {
625 goto fail;
626 }
627 } else {
628 /* Note: in this case, no need to wait */
629 memset(buf, 0, 512 * n);
630 }
631 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
632 /* add AIO support for compressed blocks ? */
633 if (decompress_cluster(bs, cluster_offset) < 0) {
634 goto fail;
635 }
636 memcpy(buf,
637 s->cluster_cache + index_in_cluster * 512, 512 * n);
638 } else {
639 if ((cluster_offset & 511) != 0) {
640 goto fail;
641 }
642 hd_iov.iov_base = (void *)buf;
643 hd_iov.iov_len = n * 512;
644 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
645 qemu_co_mutex_unlock(&s->lock);
Kevin Wolf28b04a82016-05-24 17:21:22 +0200646 ret = bdrv_co_readv(bs->file,
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200647 (cluster_offset >> 9) + index_in_cluster,
648 n, &hd_qiov);
649 qemu_co_mutex_lock(&s->lock);
650 if (ret < 0) {
651 break;
652 }
Daniel P. Berrange8336aaf2015-05-12 17:09:18 +0100653 if (bs->encrypted) {
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100654 assert(s->cipher);
655 if (encrypt_sectors(s, sector_num, buf, buf,
656 n, false, &err) < 0) {
657 goto fail;
658 }
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200659 }
660 }
661 ret = 0;
662
663 nb_sectors -= n;
664 sector_num += n;
665 buf += n * 512;
666 }
667
668done:
Kevin Wolf52b8eb62011-07-15 16:27:42 +0200669 qemu_co_mutex_unlock(&s->lock);
670
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200671 if (qiov->niov > 1) {
Michael Tokarev03396142012-06-07 20:17:55 +0400672 qemu_iovec_from_buf(qiov, 0, orig_buf, qiov->size);
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200673 qemu_vfree(orig_buf);
aliguorif141eaf2009-04-07 18:43:24 +0000674 }
Kevin Wolf52b8eb62011-07-15 16:27:42 +0200675
676 return ret;
bellard83f64092006-08-01 16:21:11 +0000677
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200678fail:
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100679 error_free(err);
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200680 ret = -EIO;
681 goto done;
bellard83f64092006-08-01 16:21:11 +0000682}
683
Dong Xu Wanga9681682011-11-10 16:23:22 +0800684static coroutine_fn int qcow_co_writev(BlockDriverState *bs, int64_t sector_num,
Kevin Wolf52b8eb62011-07-15 16:27:42 +0200685 int nb_sectors, QEMUIOVector *qiov)
bellard83f64092006-08-01 16:21:11 +0000686{
bellard83f64092006-08-01 16:21:11 +0000687 BDRVQcowState *s = bs->opaque;
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200688 int index_in_cluster;
689 uint64_t cluster_offset;
690 const uint8_t *src_buf;
691 int ret = 0, n;
692 uint8_t *cluster_data = NULL;
693 struct iovec hd_iov;
694 QEMUIOVector hd_qiov;
695 uint8_t *buf;
696 void *orig_buf;
ths3b46e622007-09-17 08:09:54 +0000697
bellard83f64092006-08-01 16:21:11 +0000698 s->cluster_cache_offset = -1; /* disable compressed cache */
699
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200700 if (qiov->niov > 1) {
Kevin Wolf0df93302014-05-20 13:36:05 +0200701 buf = orig_buf = qemu_try_blockalign(bs, qiov->size);
702 if (buf == NULL) {
703 return -ENOMEM;
704 }
Michael Tokarevd5e6b162012-06-07 20:21:06 +0400705 qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200706 } else {
707 orig_buf = NULL;
708 buf = (uint8_t *)qiov->iov->iov_base;
709 }
Christoph Hellwigad530892009-05-25 15:45:37 +0200710
Kevin Wolf52b8eb62011-07-15 16:27:42 +0200711 qemu_co_mutex_lock(&s->lock);
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200712
713 while (nb_sectors != 0) {
714
715 index_in_cluster = sector_num & (s->cluster_sectors - 1);
716 n = s->cluster_sectors - index_in_cluster;
717 if (n > nb_sectors) {
718 n = nb_sectors;
719 }
720 cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
721 index_in_cluster,
722 index_in_cluster + n);
723 if (!cluster_offset || (cluster_offset & 511) != 0) {
724 ret = -EIO;
725 break;
726 }
Daniel P. Berrange8336aaf2015-05-12 17:09:18 +0100727 if (bs->encrypted) {
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100728 Error *err = NULL;
729 assert(s->cipher);
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200730 if (!cluster_data) {
731 cluster_data = g_malloc0(s->cluster_size);
732 }
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100733 if (encrypt_sectors(s, sector_num, cluster_data, buf,
734 n, true, &err) < 0) {
735 error_free(err);
736 ret = -EIO;
737 break;
738 }
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200739 src_buf = cluster_data;
740 } else {
741 src_buf = buf;
742 }
743
744 hd_iov.iov_base = (void *)src_buf;
745 hd_iov.iov_len = n * 512;
746 qemu_iovec_init_external(&hd_qiov, &hd_iov, 1);
747 qemu_co_mutex_unlock(&s->lock);
Kevin Wolf25ec1772016-05-24 17:21:22 +0200748 ret = bdrv_co_writev(bs->file,
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200749 (cluster_offset >> 9) + index_in_cluster,
750 n, &hd_qiov);
751 qemu_co_mutex_lock(&s->lock);
752 if (ret < 0) {
753 break;
754 }
755 ret = 0;
756
757 nb_sectors -= n;
758 sector_num += n;
759 buf += n * 512;
760 }
Kevin Wolf52b8eb62011-07-15 16:27:42 +0200761 qemu_co_mutex_unlock(&s->lock);
ths3b46e622007-09-17 08:09:54 +0000762
Frediano Ziglio27deebe2011-08-23 15:21:11 +0200763 if (qiov->niov > 1) {
764 qemu_vfree(orig_buf);
Kevin Wolfb11a24d2011-06-07 15:20:44 +0200765 }
Stefan Weiladd8d262011-10-07 07:32:47 +0200766 g_free(cluster_data);
Kevin Wolfb11a24d2011-06-07 15:20:44 +0200767
Kevin Wolf52b8eb62011-07-15 16:27:42 +0200768 return ret;
bellard83f64092006-08-01 16:21:11 +0000769}
770
bellarde2731ad2004-09-18 19:32:11 +0000771static void qcow_close(BlockDriverState *bs)
bellardea2384d2004-08-01 21:59:26 +0000772{
773 BDRVQcowState *s = bs->opaque;
Kevin Wolffd9f1022011-11-22 16:44:45 +0100774
Daniel P. Berrangef6fa64f2015-07-01 18:10:37 +0100775 qcrypto_cipher_free(s->cipher);
776 s->cipher = NULL;
Anthony Liguori7267c092011-08-20 22:09:37 -0500777 g_free(s->l1_table);
Kevin Wolf0df93302014-05-20 13:36:05 +0200778 qemu_vfree(s->l2_cache);
Anthony Liguori7267c092011-08-20 22:09:37 -0500779 g_free(s->cluster_cache);
780 g_free(s->cluster_data);
Kevin Wolffd9f1022011-11-22 16:44:45 +0100781
782 migrate_del_blocker(s->migration_blocker);
783 error_free(s->migration_blocker);
bellardea2384d2004-08-01 21:59:26 +0000784}
785
Chunyan Liu16d12152014-06-05 17:20:57 +0800786static int qcow_create(const char *filename, QemuOpts *opts, Error **errp)
bellardea2384d2004-08-01 21:59:26 +0000787{
Li Zhi Hui2b16c9f2011-11-21 15:40:39 +0800788 int header_size, backing_filename_len, l1_size, shift, i;
bellardea2384d2004-08-01 21:59:26 +0000789 QCowHeader header;
Li Zhi Hui2b16c9f2011-11-21 15:40:39 +0800790 uint8_t *tmp;
Kevin Wolf0e7e1982009-05-18 16:42:10 +0200791 int64_t total_size = 0;
Chunyan Liu16d12152014-06-05 17:20:57 +0800792 char *backing_file = NULL;
Kevin Wolf0e7e1982009-05-18 16:42:10 +0200793 int flags = 0;
Max Reitz34b5d2c2013-09-05 14:45:29 +0200794 Error *local_err = NULL;
Kirill A. Shutemov3e1a8132010-01-20 00:56:12 +0100795 int ret;
Kevin Wolf6af40162016-03-08 15:57:05 +0100796 BlockBackend *qcow_blk;
Kevin Wolf0e7e1982009-05-18 16:42:10 +0200797
798 /* Read out options */
Hu Tao180e9522014-09-10 17:05:46 +0800799 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
800 BDRV_SECTOR_SIZE);
Chunyan Liu16d12152014-06-05 17:20:57 +0800801 backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
802 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ENCRYPT, false)) {
803 flags |= BLOCK_FLAG_ENCRYPT;
Kevin Wolf0e7e1982009-05-18 16:42:10 +0200804 }
bellardea2384d2004-08-01 21:59:26 +0000805
Chunyan Liuc282e1f2014-06-05 17:21:11 +0800806 ret = bdrv_create_file(filename, opts, &local_err);
Li Zhi Hui2b16c9f2011-11-21 15:40:39 +0800807 if (ret < 0) {
Paolo Bonzinib6d50662014-02-17 14:43:58 +0100808 error_propagate(errp, local_err);
Chunyan Liu16d12152014-06-05 17:20:57 +0800809 goto cleanup;
Li Zhi Hui2b16c9f2011-11-21 15:40:39 +0800810 }
811
Max Reitzefaa7c42016-03-16 19:54:38 +0100812 qcow_blk = blk_new_open(filename, NULL, NULL,
Kevin Wolf72e775c2016-03-15 14:34:37 +0100813 BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
Kevin Wolf6af40162016-03-08 15:57:05 +0100814 if (qcow_blk == NULL) {
Paolo Bonzinib6d50662014-02-17 14:43:58 +0100815 error_propagate(errp, local_err);
Kevin Wolf6af40162016-03-08 15:57:05 +0100816 ret = -EIO;
Chunyan Liu16d12152014-06-05 17:20:57 +0800817 goto cleanup;
Li Zhi Hui2b16c9f2011-11-21 15:40:39 +0800818 }
819
Kevin Wolf6af40162016-03-08 15:57:05 +0100820 blk_set_allow_write_beyond_eof(qcow_blk, true);
821
822 ret = blk_truncate(qcow_blk, 0);
Li Zhi Hui2b16c9f2011-11-21 15:40:39 +0800823 if (ret < 0) {
824 goto exit;
825 }
826
bellardea2384d2004-08-01 21:59:26 +0000827 memset(&header, 0, sizeof(header));
828 header.magic = cpu_to_be32(QCOW_MAGIC);
829 header.version = cpu_to_be32(QCOW_VERSION);
Hu Tao180e9522014-09-10 17:05:46 +0800830 header.size = cpu_to_be64(total_size);
bellardea2384d2004-08-01 21:59:26 +0000831 header_size = sizeof(header);
832 backing_filename_len = 0;
833 if (backing_file) {
aurel327852e5d2008-03-18 06:52:48 +0000834 if (strcmp(backing_file, "fat:")) {
835 header.backing_file_offset = cpu_to_be64(header_size);
836 backing_filename_len = strlen(backing_file);
837 header.backing_file_size = cpu_to_be32(backing_filename_len);
838 header_size += backing_filename_len;
839 } else {
840 /* special backing file for vvfat */
841 backing_file = NULL;
842 }
bellardea2384d2004-08-01 21:59:26 +0000843 header.cluster_bits = 9; /* 512 byte cluster to avoid copying
Deepak Kathayatdc6fb732014-03-24 16:30:17 +0800844 unmodified sectors */
bellardea2384d2004-08-01 21:59:26 +0000845 header.l2_bits = 12; /* 32 KB L2 tables */
846 } else {
847 header.cluster_bits = 12; /* 4 KB clusters */
848 header.l2_bits = 9; /* 4 KB L2 tables */
849 }
850 header_size = (header_size + 7) & ~7;
851 shift = header.cluster_bits + header.l2_bits;
Hu Tao180e9522014-09-10 17:05:46 +0800852 l1_size = (total_size + (1LL << shift) - 1) >> shift;
bellardea2384d2004-08-01 21:59:26 +0000853
854 header.l1_table_offset = cpu_to_be64(header_size);
thsec36ba12007-09-16 21:59:02 +0000855 if (flags & BLOCK_FLAG_ENCRYPT) {
bellardea2384d2004-08-01 21:59:26 +0000856 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
857 } else {
858 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
859 }
ths3b46e622007-09-17 08:09:54 +0000860
bellardea2384d2004-08-01 21:59:26 +0000861 /* write all the data */
Eric Blake8341f002016-05-06 10:26:27 -0600862 ret = blk_pwrite(qcow_blk, 0, &header, sizeof(header), 0);
Kirill A. Shutemov3e1a8132010-01-20 00:56:12 +0100863 if (ret != sizeof(header)) {
Kirill A. Shutemov3e1a8132010-01-20 00:56:12 +0100864 goto exit;
865 }
866
bellardea2384d2004-08-01 21:59:26 +0000867 if (backing_file) {
Kevin Wolf6af40162016-03-08 15:57:05 +0100868 ret = blk_pwrite(qcow_blk, sizeof(header),
Eric Blake8341f002016-05-06 10:26:27 -0600869 backing_file, backing_filename_len, 0);
Kirill A. Shutemov3e1a8132010-01-20 00:56:12 +0100870 if (ret != backing_filename_len) {
Kirill A. Shutemov3e1a8132010-01-20 00:56:12 +0100871 goto exit;
872 }
bellardea2384d2004-08-01 21:59:26 +0000873 }
Kirill A. Shutemov3e1a8132010-01-20 00:56:12 +0100874
Li Zhi Hui2b16c9f2011-11-21 15:40:39 +0800875 tmp = g_malloc0(BDRV_SECTOR_SIZE);
Laurent Vivierd737b782016-05-31 18:35:52 +0200876 for (i = 0; i < DIV_ROUND_UP(sizeof(uint64_t) * l1_size, BDRV_SECTOR_SIZE);
877 i++) {
Eric Blake8341f002016-05-06 10:26:27 -0600878 ret = blk_pwrite(qcow_blk, header_size + BDRV_SECTOR_SIZE * i,
879 tmp, BDRV_SECTOR_SIZE, 0);
Li Zhi Hui2b16c9f2011-11-21 15:40:39 +0800880 if (ret != BDRV_SECTOR_SIZE) {
881 g_free(tmp);
882 goto exit;
883 }
884 }
885
886 g_free(tmp);
Kirill A. Shutemov3e1a8132010-01-20 00:56:12 +0100887 ret = 0;
888exit:
Kevin Wolf6af40162016-03-08 15:57:05 +0100889 blk_unref(qcow_blk);
Chunyan Liu16d12152014-06-05 17:20:57 +0800890cleanup:
891 g_free(backing_file);
Kirill A. Shutemov3e1a8132010-01-20 00:56:12 +0100892 return ret;
bellardea2384d2004-08-01 21:59:26 +0000893}
894
bellardc47c33b2006-08-05 21:32:10 +0000895static int qcow_make_empty(BlockDriverState *bs)
bellard95389c82005-12-18 18:28:15 +0000896{
897 BDRVQcowState *s = bs->opaque;
898 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
bellard83f64092006-08-01 16:21:11 +0000899 int ret;
bellard95389c82005-12-18 18:28:15 +0000900
901 memset(s->l1_table, 0, l1_length);
Kevin Wolfd9ca2ea2016-06-20 20:09:15 +0200902 if (bdrv_pwrite_sync(bs->file, s->l1_table_offset, s->l1_table,
Kevin Wolf5e5557d2010-06-18 16:11:53 +0200903 l1_length) < 0)
904 return -1;
Kevin Wolf9a4f4c32015-06-16 14:19:22 +0200905 ret = bdrv_truncate(bs->file->bs, s->l1_table_offset + l1_length);
bellard83f64092006-08-01 16:21:11 +0000906 if (ret < 0)
907 return ret;
bellard95389c82005-12-18 18:28:15 +0000908
909 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
910 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
911 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
912
913 return 0;
914}
915
bellardea2384d2004-08-01 21:59:26 +0000916/* XXX: put compressed sectors first, then all the cluster aligned
917 tables to avoid losing bytes in alignment */
Pavel Butsykinf2b95a12016-07-22 11:17:46 +0300918static coroutine_fn int
919qcow_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
920 uint64_t bytes, QEMUIOVector *qiov)
bellardea2384d2004-08-01 21:59:26 +0000921{
922 BDRVQcowState *s = bs->opaque;
Pavel Butsykinf2b95a12016-07-22 11:17:46 +0300923 QEMUIOVector hd_qiov;
924 struct iovec iov;
bellardea2384d2004-08-01 21:59:26 +0000925 z_stream strm;
926 int ret, out_len;
Pavel Butsykinf2b95a12016-07-22 11:17:46 +0300927 uint8_t *buf, *out_buf;
bellardea2384d2004-08-01 21:59:26 +0000928 uint64_t cluster_offset;
929
Pavel Butsykinf2b95a12016-07-22 11:17:46 +0300930 if (bytes != s->cluster_size) {
Stefan Hajnoczi16b3c5c2013-04-15 17:17:32 +0200931 ret = -EINVAL;
932
933 /* Zero-pad last write if image size is not cluster aligned */
Pavel Butsykinf2b95a12016-07-22 11:17:46 +0300934 if (offset + bytes == bs->total_sectors << BDRV_SECTOR_BITS &&
935 bytes < s->cluster_size)
936 {
Stefan Hajnoczi16b3c5c2013-04-15 17:17:32 +0200937 uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size);
938 memset(pad_buf, 0, s->cluster_size);
Pavel Butsykinf2b95a12016-07-22 11:17:46 +0300939 qemu_iovec_to_buf(qiov, 0, pad_buf, s->cluster_size);
940 iov = (struct iovec) {
941 .iov_base = pad_buf,
942 .iov_len = s->cluster_size,
943 };
944 qemu_iovec_init_external(&hd_qiov, &iov, 1);
945 ret = qcow_co_pwritev_compressed(bs, offset, bytes, &hd_qiov);
Stefan Hajnoczi16b3c5c2013-04-15 17:17:32 +0200946 qemu_vfree(pad_buf);
947 }
948 return ret;
949 }
Pavel Butsykinf2b95a12016-07-22 11:17:46 +0300950 buf = qemu_blockalign(bs, s->cluster_size);
951 qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
bellardea2384d2004-08-01 21:59:26 +0000952
Vladimir Sementsov-Ogievskiyebf7bba2016-07-14 19:59:25 +0300953 out_buf = g_malloc(s->cluster_size);
bellardea2384d2004-08-01 21:59:26 +0000954
955 /* best compression, small window, no zlib header */
956 memset(&strm, 0, sizeof(strm));
957 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
ths5fafdf22007-09-16 21:08:06 +0000958 Z_DEFLATED, -12,
bellardea2384d2004-08-01 21:59:26 +0000959 9, Z_DEFAULT_STRATEGY);
960 if (ret != 0) {
Kevin Wolf64ebe712011-10-26 11:21:50 +0200961 ret = -EINVAL;
962 goto fail;
bellardea2384d2004-08-01 21:59:26 +0000963 }
964
965 strm.avail_in = s->cluster_size;
966 strm.next_in = (uint8_t *)buf;
967 strm.avail_out = s->cluster_size;
968 strm.next_out = out_buf;
969
970 ret = deflate(&strm, Z_FINISH);
971 if (ret != Z_STREAM_END && ret != Z_OK) {
bellardea2384d2004-08-01 21:59:26 +0000972 deflateEnd(&strm);
Kevin Wolf64ebe712011-10-26 11:21:50 +0200973 ret = -EINVAL;
974 goto fail;
bellardea2384d2004-08-01 21:59:26 +0000975 }
976 out_len = strm.next_out - out_buf;
977
978 deflateEnd(&strm);
979
980 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
981 /* could not compress: write normal cluster */
Pavel Butsykinf2b95a12016-07-22 11:17:46 +0300982 ret = qcow_co_writev(bs, offset >> BDRV_SECTOR_BITS,
983 bytes >> BDRV_SECTOR_BITS, qiov);
Kevin Wolf64ebe712011-10-26 11:21:50 +0200984 if (ret < 0) {
985 goto fail;
986 }
Pavel Butsykinf2b95a12016-07-22 11:17:46 +0300987 goto success;
bellardea2384d2004-08-01 21:59:26 +0000988 }
Pavel Butsykinf2b95a12016-07-22 11:17:46 +0300989 qemu_co_mutex_lock(&s->lock);
990 cluster_offset = get_cluster_offset(bs, offset, 2, out_len, 0, 0);
991 qemu_co_mutex_unlock(&s->lock);
992 if (cluster_offset == 0) {
993 ret = -EIO;
994 goto fail;
995 }
996 cluster_offset &= s->cluster_offset_mask;
ths3b46e622007-09-17 08:09:54 +0000997
Pavel Butsykinf2b95a12016-07-22 11:17:46 +0300998 iov = (struct iovec) {
999 .iov_base = out_buf,
1000 .iov_len = out_len,
1001 };
1002 qemu_iovec_init_external(&hd_qiov, &iov, 1);
1003 ret = bdrv_co_pwritev(bs->file, cluster_offset, out_len, &hd_qiov, 0);
1004 if (ret < 0) {
1005 goto fail;
1006 }
1007success:
Kevin Wolf64ebe712011-10-26 11:21:50 +02001008 ret = 0;
1009fail:
Pavel Butsykinf2b95a12016-07-22 11:17:46 +03001010 qemu_vfree(buf);
Anthony Liguori7267c092011-08-20 22:09:37 -05001011 g_free(out_buf);
Kevin Wolf64ebe712011-10-26 11:21:50 +02001012 return ret;
bellardea2384d2004-08-01 21:59:26 +00001013}
1014
bellardc47c33b2006-08-05 21:32:10 +00001015static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1016{
1017 BDRVQcowState *s = bs->opaque;
1018 bdi->cluster_size = s->cluster_size;
1019 return 0;
1020}
1021
Chunyan Liu16d12152014-06-05 17:20:57 +08001022static QemuOptsList qcow_create_opts = {
1023 .name = "qcow-create-opts",
1024 .head = QTAILQ_HEAD_INITIALIZER(qcow_create_opts.head),
1025 .desc = {
1026 {
1027 .name = BLOCK_OPT_SIZE,
1028 .type = QEMU_OPT_SIZE,
1029 .help = "Virtual disk size"
1030 },
1031 {
1032 .name = BLOCK_OPT_BACKING_FILE,
1033 .type = QEMU_OPT_STRING,
1034 .help = "File name of a base image"
1035 },
1036 {
1037 .name = BLOCK_OPT_ENCRYPT,
1038 .type = QEMU_OPT_BOOL,
1039 .help = "Encrypt the image",
1040 .def_value_str = "off"
1041 },
1042 { /* end of list */ }
1043 }
Kevin Wolf0e7e1982009-05-18 16:42:10 +02001044};
1045
Anthony Liguori5efa9d52009-05-09 17:03:42 -05001046static BlockDriver bdrv_qcow = {
aurel32e60f4692009-03-07 22:00:29 +00001047 .format_name = "qcow",
1048 .instance_size = sizeof(BDRVQcowState),
1049 .bdrv_probe = qcow_probe,
1050 .bdrv_open = qcow_open,
1051 .bdrv_close = qcow_close,
Chunyan Liu16d12152014-06-05 17:20:57 +08001052 .bdrv_reopen_prepare = qcow_reopen_prepare,
Chunyan Liuc282e1f2014-06-05 17:21:11 +08001053 .bdrv_create = qcow_create,
Peter Lieven3ac21622013-06-28 12:47:42 +02001054 .bdrv_has_zero_init = bdrv_has_zero_init_1,
Kevin Wolf8ee79e72014-06-04 15:09:35 +02001055 .supports_backing = true,
Kevin Wolfc68b89a2011-11-10 17:25:44 +01001056
1057 .bdrv_co_readv = qcow_co_readv,
1058 .bdrv_co_writev = qcow_co_writev,
Paolo Bonzinib6b8a332013-09-04 19:00:28 +02001059 .bdrv_co_get_block_status = qcow_co_get_block_status,
Kevin Wolfc68b89a2011-11-10 17:25:44 +01001060
1061 .bdrv_set_key = qcow_set_key,
1062 .bdrv_make_empty = qcow_make_empty,
Pavel Butsykinf2b95a12016-07-22 11:17:46 +03001063 .bdrv_co_pwritev_compressed = qcow_co_pwritev_compressed,
Kevin Wolfc68b89a2011-11-10 17:25:44 +01001064 .bdrv_get_info = qcow_get_info,
Kevin Wolf0e7e1982009-05-18 16:42:10 +02001065
Chunyan Liu16d12152014-06-05 17:20:57 +08001066 .create_opts = &qcow_create_opts,
bellardea2384d2004-08-01 21:59:26 +00001067};
Anthony Liguori5efa9d52009-05-09 17:03:42 -05001068
1069static void bdrv_qcow_init(void)
1070{
1071 bdrv_register(&bdrv_qcow);
1072}
1073
1074block_init(bdrv_qcow_init);