]> git.proxmox.com Git - mirror_qemu.git/blame - block/qcow2.c
qcow2: Fix metadata preallocation
[mirror_qemu.git] / block / qcow2.c
CommitLineData
585f8587
FB
1/*
2 * Block driver for the QCOW version 2 format
5fafdf24 3 *
585f8587 4 * Copyright (c) 2004-2006 Fabrice Bellard
5fafdf24 5 *
585f8587
FB
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
faf07963 24#include "qemu-common.h"
585f8587 25#include "block_int.h"
5efa9d5a 26#include "module.h"
585f8587
FB
27#include <zlib.h>
28#include "aes.h"
f7d0fe02 29#include "block/qcow2.h"
585f8587
FB
30
31/*
32 Differences with QCOW:
33
34 - Support for multiple incremental snapshots.
35 - Memory management by reference counts.
36 - Clusters which have a reference count of one have the bit
37 QCOW_OFLAG_COPIED to optimize write performance.
5fafdf24 38 - Size of compressed clusters is stored in sectors to reduce bit usage
585f8587
FB
39 in the cluster offsets.
40 - Support for storing additional data (such as the VM state) in the
3b46e624 41 snapshots.
585f8587
FB
42 - If a backing store is used, the cluster size is not constrained
43 (could be backported to QCOW).
44 - L2 tables have always a size of one cluster.
45*/
46
9b80ddf3
AL
47
48typedef struct {
49 uint32_t magic;
50 uint32_t len;
51} QCowExtension;
52#define QCOW_EXT_MAGIC_END 0
f965509c 53#define QCOW_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
9b80ddf3
AL
54
55
585f8587
FB
56
57static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
58{
59 const QCowHeader *cow_header = (const void *)buf;
3b46e624 60
585f8587
FB
61 if (buf_size >= sizeof(QCowHeader) &&
62 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
5fafdf24 63 be32_to_cpu(cow_header->version) == QCOW_VERSION)
585f8587
FB
64 return 100;
65 else
66 return 0;
67}
68
9b80ddf3
AL
69
70/*
71 * read qcow2 extension and fill bs
72 * start reading from start_offset
73 * finish reading upon magic of value 0 or when end_offset reached
74 * unknown magic is skipped (future extension this version knows nothing about)
75 * return 0 upon success, non-0 otherwise
76 */
77static int qcow_read_extensions(BlockDriverState *bs, uint64_t start_offset,
78 uint64_t end_offset)
79{
80 BDRVQcowState *s = bs->opaque;
81 QCowExtension ext;
82 uint64_t offset;
83
84#ifdef DEBUG_EXT
85 printf("qcow_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
86#endif
87 offset = start_offset;
88 while (offset < end_offset) {
89
90#ifdef DEBUG_EXT
91 /* Sanity check */
92 if (offset > s->cluster_size)
93 printf("qcow_handle_extension: suspicious offset %lu\n", offset);
94
95 printf("attemting to read extended header in offset %lu\n", offset);
96#endif
97
98 if (bdrv_pread(s->hd, offset, &ext, sizeof(ext)) != sizeof(ext)) {
4c978075
AL
99 fprintf(stderr, "qcow_handle_extension: ERROR: pread fail from offset %llu\n",
100 (unsigned long long)offset);
9b80ddf3
AL
101 return 1;
102 }
103 be32_to_cpus(&ext.magic);
104 be32_to_cpus(&ext.len);
105 offset += sizeof(ext);
106#ifdef DEBUG_EXT
107 printf("ext.magic = 0x%x\n", ext.magic);
108#endif
109 switch (ext.magic) {
110 case QCOW_EXT_MAGIC_END:
111 return 0;
f965509c
AL
112
113 case QCOW_EXT_MAGIC_BACKING_FORMAT:
114 if (ext.len >= sizeof(bs->backing_format)) {
115 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
4c978075 116 " (>=%zu)\n",
f965509c
AL
117 ext.len, sizeof(bs->backing_format));
118 return 2;
119 }
120 if (bdrv_pread(s->hd, offset , bs->backing_format,
121 ext.len) != ext.len)
122 return 3;
123 bs->backing_format[ext.len] = '\0';
124#ifdef DEBUG_EXT
125 printf("Qcow2: Got format extension %s\n", bs->backing_format);
126#endif
127 offset += ((ext.len + 7) & ~7);
128 break;
129
9b80ddf3
AL
130 default:
131 /* unknown magic -- just skip it */
132 offset += ((ext.len + 7) & ~7);
133 break;
134 }
135 }
136
137 return 0;
138}
139
140
585f8587
FB
141static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
142{
143 BDRVQcowState *s = bs->opaque;
144 int len, i, shift, ret;
145 QCowHeader header;
9b80ddf3 146 uint64_t ext_end;
585f8587 147
b5eff355 148 ret = bdrv_file_open(&s->hd, filename, flags);
585f8587
FB
149 if (ret < 0)
150 return ret;
151 if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
152 goto fail;
153 be32_to_cpus(&header.magic);
154 be32_to_cpus(&header.version);
155 be64_to_cpus(&header.backing_file_offset);
156 be32_to_cpus(&header.backing_file_size);
157 be64_to_cpus(&header.size);
158 be32_to_cpus(&header.cluster_bits);
159 be32_to_cpus(&header.crypt_method);
160 be64_to_cpus(&header.l1_table_offset);
161 be32_to_cpus(&header.l1_size);
162 be64_to_cpus(&header.refcount_table_offset);
163 be32_to_cpus(&header.refcount_table_clusters);
164 be64_to_cpus(&header.snapshots_offset);
165 be32_to_cpus(&header.nb_snapshots);
3b46e624 166
585f8587
FB
167 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
168 goto fail;
5fafdf24 169 if (header.size <= 1 ||
73c632ed
KW
170 header.cluster_bits < MIN_CLUSTER_BITS ||
171 header.cluster_bits > MAX_CLUSTER_BITS)
585f8587
FB
172 goto fail;
173 if (header.crypt_method > QCOW_CRYPT_AES)
174 goto fail;
175 s->crypt_method_header = header.crypt_method;
176 if (s->crypt_method_header)
177 bs->encrypted = 1;
178 s->cluster_bits = header.cluster_bits;
179 s->cluster_size = 1 << s->cluster_bits;
180 s->cluster_sectors = 1 << (s->cluster_bits - 9);
181 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
182 s->l2_size = 1 << s->l2_bits;
183 bs->total_sectors = header.size / 512;
184 s->csize_shift = (62 - (s->cluster_bits - 8));
185 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
186 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
187 s->refcount_table_offset = header.refcount_table_offset;
5fafdf24 188 s->refcount_table_size =
585f8587
FB
189 header.refcount_table_clusters << (s->cluster_bits - 3);
190
191 s->snapshots_offset = header.snapshots_offset;
192 s->nb_snapshots = header.nb_snapshots;
193
194 /* read the level 1 table */
195 s->l1_size = header.l1_size;
196 shift = s->cluster_bits + s->l2_bits;
197 s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
198 /* the L1 table must contain at least enough entries to put
199 header.size bytes */
200 if (s->l1_size < s->l1_vm_state_index)
201 goto fail;
202 s->l1_table_offset = header.l1_table_offset;
3f6a3ee5
KW
203 s->l1_table = qemu_mallocz(
204 align_offset(s->l1_size * sizeof(uint64_t), 512));
5fafdf24 205 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
585f8587
FB
206 s->l1_size * sizeof(uint64_t))
207 goto fail;
208 for(i = 0;i < s->l1_size; i++) {
209 be64_to_cpus(&s->l1_table[i]);
210 }
211 /* alloc L2 cache */
212 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
585f8587 213 s->cluster_cache = qemu_malloc(s->cluster_size);
585f8587 214 /* one more sector for decompressed data alignment */
095a9c58
AL
215 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
216 + 512);
585f8587 217 s->cluster_cache_offset = -1;
3b46e624 218
ed6ccf0f 219 if (qcow2_refcount_init(bs) < 0)
585f8587
FB
220 goto fail;
221
9b80ddf3
AL
222 /* read qcow2 extensions */
223 if (header.backing_file_offset)
224 ext_end = header.backing_file_offset;
225 else
226 ext_end = s->cluster_size;
227 if (qcow_read_extensions(bs, sizeof(header), ext_end))
228 goto fail;
229
585f8587
FB
230 /* read the backing file name */
231 if (header.backing_file_offset != 0) {
232 len = header.backing_file_size;
233 if (len > 1023)
234 len = 1023;
235 if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
236 goto fail;
237 bs->backing_file[len] = '\0';
238 }
ed6ccf0f 239 if (qcow2_read_snapshots(bs) < 0)
585f8587
FB
240 goto fail;
241
242#ifdef DEBUG_ALLOC
14899cdf 243 qcow2_check_refcounts(bs);
585f8587
FB
244#endif
245 return 0;
246
247 fail:
ed6ccf0f
KW
248 qcow2_free_snapshots(bs);
249 qcow2_refcount_close(bs);
585f8587
FB
250 qemu_free(s->l1_table);
251 qemu_free(s->l2_cache);
252 qemu_free(s->cluster_cache);
253 qemu_free(s->cluster_data);
254 bdrv_delete(s->hd);
255 return -1;
256}
257
258static int qcow_set_key(BlockDriverState *bs, const char *key)
259{
260 BDRVQcowState *s = bs->opaque;
261 uint8_t keybuf[16];
262 int len, i;
3b46e624 263
585f8587
FB
264 memset(keybuf, 0, 16);
265 len = strlen(key);
266 if (len > 16)
267 len = 16;
268 /* XXX: we could compress the chars to 7 bits to increase
269 entropy */
270 for(i = 0;i < len;i++) {
271 keybuf[i] = key[i];
272 }
273 s->crypt_method = s->crypt_method_header;
274
275 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
276 return -1;
277 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
278 return -1;
279#if 0
280 /* test */
281 {
282 uint8_t in[16];
283 uint8_t out[16];
284 uint8_t tmp[16];
285 for(i=0;i<16;i++)
286 in[i] = i;
287 AES_encrypt(in, tmp, &s->aes_encrypt_key);
288 AES_decrypt(tmp, out, &s->aes_decrypt_key);
289 for(i = 0; i < 16; i++)
290 printf(" %02x", tmp[i]);
291 printf("\n");
292 for(i = 0; i < 16; i++)
293 printf(" %02x", out[i]);
294 printf("\n");
295 }
296#endif
297 return 0;
298}
299
5fafdf24 300static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
585f8587
FB
301 int nb_sectors, int *pnum)
302{
585f8587
FB
303 uint64_t cluster_offset;
304
095a9c58 305 *pnum = nb_sectors;
ed6ccf0f 306 cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, pnum);
095a9c58 307
585f8587
FB
308 return (cluster_offset != 0);
309}
310
a9465922 311/* handle reading after the end of the backing file */
ed6ccf0f 312int qcow2_backing_read1(BlockDriverState *bs,
45aba42f 313 int64_t sector_num, uint8_t *buf, int nb_sectors)
a9465922
FB
314{
315 int n1;
316 if ((sector_num + nb_sectors) <= bs->total_sectors)
317 return nb_sectors;
318 if (sector_num >= bs->total_sectors)
319 n1 = 0;
320 else
321 n1 = bs->total_sectors - sector_num;
322 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
323 return n1;
324}
325
ce1a14dc
PB
326typedef struct QCowAIOCB {
327 BlockDriverAIOCB common;
585f8587 328 int64_t sector_num;
f141eafe 329 QEMUIOVector *qiov;
585f8587 330 uint8_t *buf;
f141eafe 331 void *orig_buf;
585f8587
FB
332 int nb_sectors;
333 int n;
334 uint64_t cluster_offset;
5fafdf24 335 uint8_t *cluster_data;
585f8587 336 BlockDriverAIOCB *hd_aiocb;
c87c0672
AL
337 struct iovec hd_iov;
338 QEMUIOVector hd_qiov;
1490791f 339 QEMUBH *bh;
e976c6a1 340 QCowL2Meta l2meta;
585f8587
FB
341} QCowAIOCB;
342
c16b5a2c
CH
343static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
344{
345 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
346 if (acb->hd_aiocb)
347 bdrv_aio_cancel(acb->hd_aiocb);
348 qemu_aio_release(acb);
349}
350
351static AIOPool qcow_aio_pool = {
352 .aiocb_size = sizeof(QCowAIOCB),
353 .cancel = qcow_aio_cancel,
354};
355
1490791f
AL
356static void qcow_aio_read_cb(void *opaque, int ret);
357static void qcow_aio_read_bh(void *opaque)
358{
359 QCowAIOCB *acb = opaque;
360 qemu_bh_delete(acb->bh);
361 acb->bh = NULL;
362 qcow_aio_read_cb(opaque, 0);
363}
364
a32ef786
AL
365static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
366{
367 if (acb->bh)
368 return -EIO;
369
370 acb->bh = qemu_bh_new(cb, acb);
371 if (!acb->bh)
372 return -EIO;
373
374 qemu_bh_schedule(acb->bh);
375
376 return 0;
377}
378
585f8587
FB
379static void qcow_aio_read_cb(void *opaque, int ret)
380{
ce1a14dc
PB
381 QCowAIOCB *acb = opaque;
382 BlockDriverState *bs = acb->common.bs;
585f8587 383 BDRVQcowState *s = bs->opaque;
a9465922 384 int index_in_cluster, n1;
585f8587 385
ce1a14dc 386 acb->hd_aiocb = NULL;
f141eafe
AL
387 if (ret < 0)
388 goto done;
585f8587 389
585f8587 390 /* post process the read buffer */
ce1a14dc 391 if (!acb->cluster_offset) {
585f8587 392 /* nothing to do */
ce1a14dc 393 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
585f8587
FB
394 /* nothing to do */
395 } else {
396 if (s->crypt_method) {
ed6ccf0f 397 qcow2_encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
5fafdf24 398 acb->n, 0,
585f8587
FB
399 &s->aes_decrypt_key);
400 }
401 }
402
ce1a14dc
PB
403 acb->nb_sectors -= acb->n;
404 acb->sector_num += acb->n;
405 acb->buf += acb->n * 512;
585f8587 406
ce1a14dc 407 if (acb->nb_sectors == 0) {
585f8587 408 /* request completed */
f141eafe
AL
409 ret = 0;
410 goto done;
585f8587 411 }
3b46e624 412
585f8587 413 /* prepare next AIO request */
095a9c58 414 acb->n = acb->nb_sectors;
ed6ccf0f
KW
415 acb->cluster_offset =
416 qcow2_get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
ce1a14dc 417 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
ce1a14dc
PB
418
419 if (!acb->cluster_offset) {
585f8587
FB
420 if (bs->backing_hd) {
421 /* read from the base image */
ed6ccf0f 422 n1 = qcow2_backing_read1(bs->backing_hd, acb->sector_num,
ce1a14dc 423 acb->buf, acb->n);
a9465922 424 if (n1 > 0) {
3f4cb3d3 425 acb->hd_iov.iov_base = (void *)acb->buf;
c87c0672
AL
426 acb->hd_iov.iov_len = acb->n * 512;
427 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
428 acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
429 &acb->hd_qiov, acb->n,
430 qcow_aio_read_cb, acb);
ce1a14dc 431 if (acb->hd_aiocb == NULL)
f141eafe 432 goto done;
a9465922 433 } else {
a32ef786
AL
434 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
435 if (ret < 0)
f141eafe 436 goto done;
a9465922 437 }
585f8587
FB
438 } else {
439 /* Note: in this case, no need to wait */
ce1a14dc 440 memset(acb->buf, 0, 512 * acb->n);
a32ef786
AL
441 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
442 if (ret < 0)
f141eafe 443 goto done;
585f8587 444 }
ce1a14dc 445 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
585f8587 446 /* add AIO support for compressed blocks ? */
ed6ccf0f 447 if (qcow2_decompress_cluster(s, acb->cluster_offset) < 0)
f141eafe 448 goto done;
5fafdf24 449 memcpy(acb->buf,
ce1a14dc 450 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
a32ef786
AL
451 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
452 if (ret < 0)
f141eafe 453 goto done;
585f8587 454 } else {
ce1a14dc 455 if ((acb->cluster_offset & 511) != 0) {
585f8587 456 ret = -EIO;
f141eafe 457 goto done;
585f8587 458 }
c87c0672 459
3f4cb3d3 460 acb->hd_iov.iov_base = (void *)acb->buf;
c87c0672
AL
461 acb->hd_iov.iov_len = acb->n * 512;
462 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
463 acb->hd_aiocb = bdrv_aio_readv(s->hd,
5fafdf24 464 (acb->cluster_offset >> 9) + index_in_cluster,
c87c0672 465 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
ce1a14dc 466 if (acb->hd_aiocb == NULL)
f141eafe
AL
467 goto done;
468 }
469
470 return;
471done:
472 if (acb->qiov->niov > 1) {
473 qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
474 qemu_vfree(acb->orig_buf);
585f8587 475 }
f141eafe
AL
476 acb->common.cb(acb->common.opaque, ret);
477 qemu_aio_release(acb);
585f8587
FB
478}
479
ce1a14dc 480static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
f141eafe
AL
481 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
482 BlockDriverCompletionFunc *cb, void *opaque, int is_write)
585f8587 483{
ce1a14dc
PB
484 QCowAIOCB *acb;
485
c16b5a2c 486 acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
ce1a14dc
PB
487 if (!acb)
488 return NULL;
489 acb->hd_aiocb = NULL;
490 acb->sector_num = sector_num;
f141eafe
AL
491 acb->qiov = qiov;
492 if (qiov->niov > 1) {
e268ca52 493 acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
f141eafe
AL
494 if (is_write)
495 qemu_iovec_to_buffer(qiov, acb->buf);
3f4cb3d3
BS
496 } else {
497 acb->buf = (uint8_t *)qiov->iov->iov_base;
498 }
ce1a14dc
PB
499 acb->nb_sectors = nb_sectors;
500 acb->n = 0;
501 acb->cluster_offset = 0;
e976c6a1 502 acb->l2meta.nb_clusters = 0;
ce1a14dc
PB
503 return acb;
504}
505
f141eafe
AL
506static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
507 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
ce1a14dc
PB
508 BlockDriverCompletionFunc *cb, void *opaque)
509{
510 QCowAIOCB *acb;
511
f141eafe 512 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
ce1a14dc
PB
513 if (!acb)
514 return NULL;
585f8587
FB
515
516 qcow_aio_read_cb(acb, 0);
ce1a14dc 517 return &acb->common;
585f8587
FB
518}
519
520static void qcow_aio_write_cb(void *opaque, int ret)
521{
ce1a14dc
PB
522 QCowAIOCB *acb = opaque;
523 BlockDriverState *bs = acb->common.bs;
585f8587 524 BDRVQcowState *s = bs->opaque;
585f8587 525 int index_in_cluster;
585f8587 526 const uint8_t *src_buf;
095a9c58 527 int n_end;
ce1a14dc
PB
528
529 acb->hd_aiocb = NULL;
530
f141eafe
AL
531 if (ret < 0)
532 goto done;
585f8587 533
ed6ccf0f
KW
534 if (qcow2_alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) {
535 qcow2_free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters);
f141eafe 536 goto done;
e976c6a1
AL
537 }
538
ce1a14dc
PB
539 acb->nb_sectors -= acb->n;
540 acb->sector_num += acb->n;
541 acb->buf += acb->n * 512;
585f8587 542
ce1a14dc 543 if (acb->nb_sectors == 0) {
585f8587 544 /* request completed */
f141eafe
AL
545 ret = 0;
546 goto done;
585f8587 547 }
3b46e624 548
ce1a14dc 549 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
095a9c58
AL
550 n_end = index_in_cluster + acb->nb_sectors;
551 if (s->crypt_method &&
552 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
553 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
554
ed6ccf0f 555 acb->cluster_offset = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9,
05203524 556 index_in_cluster,
e976c6a1
AL
557 n_end, &acb->n, &acb->l2meta);
558 if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
585f8587 559 ret = -EIO;
f141eafe 560 goto done;
585f8587
FB
561 }
562 if (s->crypt_method) {
ce1a14dc 563 if (!acb->cluster_data) {
095a9c58
AL
564 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
565 s->cluster_size);
585f8587 566 }
ed6ccf0f 567 qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
ce1a14dc
PB
568 acb->n, 1, &s->aes_encrypt_key);
569 src_buf = acb->cluster_data;
585f8587 570 } else {
ce1a14dc 571 src_buf = acb->buf;
585f8587 572 }
c87c0672
AL
573 acb->hd_iov.iov_base = (void *)src_buf;
574 acb->hd_iov.iov_len = acb->n * 512;
575 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
576 acb->hd_aiocb = bdrv_aio_writev(s->hd,
577 (acb->cluster_offset >> 9) + index_in_cluster,
578 &acb->hd_qiov, acb->n,
579 qcow_aio_write_cb, acb);
ce1a14dc 580 if (acb->hd_aiocb == NULL)
f141eafe
AL
581 goto done;
582
583 return;
584
585done:
586 if (acb->qiov->niov > 1)
587 qemu_vfree(acb->orig_buf);
588 acb->common.cb(acb->common.opaque, ret);
589 qemu_aio_release(acb);
585f8587
FB
590}
591
f141eafe
AL
592static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
593 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
ce1a14dc 594 BlockDriverCompletionFunc *cb, void *opaque)
585f8587 595{
585f8587 596 BDRVQcowState *s = bs->opaque;
ce1a14dc 597 QCowAIOCB *acb;
3b46e624 598
585f8587
FB
599 s->cluster_cache_offset = -1; /* disable compressed cache */
600
f141eafe 601 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
ce1a14dc
PB
602 if (!acb)
603 return NULL;
3b46e624 604
585f8587 605 qcow_aio_write_cb(acb, 0);
ce1a14dc 606 return &acb->common;
585f8587
FB
607}
608
585f8587
FB
609static void qcow_close(BlockDriverState *bs)
610{
611 BDRVQcowState *s = bs->opaque;
612 qemu_free(s->l1_table);
613 qemu_free(s->l2_cache);
614 qemu_free(s->cluster_cache);
615 qemu_free(s->cluster_data);
ed6ccf0f 616 qcow2_refcount_close(bs);
585f8587
FB
617 bdrv_delete(s->hd);
618}
619
73c632ed
KW
620static int get_bits_from_size(size_t size)
621{
622 int res = 0;
623
624 if (size == 0) {
625 return -1;
626 }
627
628 while (size != 1) {
629 /* Not a power of two */
630 if (size & 1) {
631 return -1;
632 }
633
634 size >>= 1;
635 res++;
636 }
637
638 return res;
639}
640
a35e1c17
KW
641
642static int preallocate(BlockDriverState *bs)
643{
644 BDRVQcowState *s = bs->opaque;
2000cbc5 645 uint64_t cluster_offset = 0;
a35e1c17
KW
646 uint64_t nb_sectors;
647 uint64_t offset;
648 int num;
649 QCowL2Meta meta;
650
651 nb_sectors = bdrv_getlength(bs) >> 9;
652 offset = 0;
653
654 while (nb_sectors) {
655 num = MIN(nb_sectors, INT_MAX >> 9);
656 cluster_offset = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num,
657 &meta);
658
659 if (cluster_offset == 0) {
660 return -1;
661 }
662
663 if (qcow2_alloc_cluster_link_l2(bs, cluster_offset, &meta) < 0) {
664 qcow2_free_any_clusters(bs, cluster_offset, meta.nb_clusters);
665 return -1;
666 }
667
668 /* TODO Preallocate data if requested */
669
670 nb_sectors -= num;
671 offset += num << 9;
672 }
673
674 /*
675 * It is expected that the image file is large enough to actually contain
676 * all of the allocated clusters (otherwise we get failing reads after
677 * EOF). Extend the image to the last allocated sector.
678 */
679 if (cluster_offset != 0) {
ea80b906
KW
680 uint8_t buf[512];
681 memset(buf, 0, 512);
682 bdrv_write(s->hd, (cluster_offset >> 9) + num - 1, buf, 1);
a35e1c17
KW
683 }
684
685 return 0;
686}
687
f965509c
AL
688static int qcow_create2(const char *filename, int64_t total_size,
689 const char *backing_file, const char *backing_format,
a35e1c17 690 int flags, size_t cluster_size, int prealloc)
585f8587 691{
f965509c 692
585f8587 693 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
2d2431f0 694 int ref_clusters, backing_format_len = 0;
585f8587
FB
695 QCowHeader header;
696 uint64_t tmp, offset;
697 QCowCreateState s1, *s = &s1;
f965509c
AL
698 QCowExtension ext_bf = {0, 0};
699
3b46e624 700
585f8587
FB
701 memset(s, 0, sizeof(*s));
702
703 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
704 if (fd < 0)
705 return -1;
706 memset(&header, 0, sizeof(header));
707 header.magic = cpu_to_be32(QCOW_MAGIC);
708 header.version = cpu_to_be32(QCOW_VERSION);
709 header.size = cpu_to_be64(total_size * 512);
710 header_size = sizeof(header);
711 backing_filename_len = 0;
712 if (backing_file) {
f965509c
AL
713 if (backing_format) {
714 ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
715 backing_format_len = strlen(backing_format);
716 ext_bf.len = (backing_format_len + 7) & ~7;
717 header_size += ((sizeof(ext_bf) + ext_bf.len + 7) & ~7);
718 }
585f8587
FB
719 header.backing_file_offset = cpu_to_be64(header_size);
720 backing_filename_len = strlen(backing_file);
721 header.backing_file_size = cpu_to_be32(backing_filename_len);
722 header_size += backing_filename_len;
723 }
73c632ed
KW
724
725 /* Cluster size */
726 s->cluster_bits = get_bits_from_size(cluster_size);
727 if (s->cluster_bits < MIN_CLUSTER_BITS ||
728 s->cluster_bits > MAX_CLUSTER_BITS)
729 {
730 fprintf(stderr, "Cluster size must be a power of two between "
731 "%d and %dk\n",
732 1 << MIN_CLUSTER_BITS,
733 1 << (MAX_CLUSTER_BITS - 10));
734 return -EINVAL;
735 }
585f8587 736 s->cluster_size = 1 << s->cluster_bits;
73c632ed 737
585f8587
FB
738 header.cluster_bits = cpu_to_be32(s->cluster_bits);
739 header_size = (header_size + 7) & ~7;
ec36ba14 740 if (flags & BLOCK_FLAG_ENCRYPT) {
585f8587
FB
741 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
742 } else {
743 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
744 }
745 l2_bits = s->cluster_bits - 3;
746 shift = s->cluster_bits + l2_bits;
747 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
748 offset = align_offset(header_size, s->cluster_size);
749 s->l1_table_offset = offset;
750 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
751 header.l1_size = cpu_to_be32(l1_size);
15e6690a 752 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
585f8587
FB
753
754 s->refcount_table = qemu_mallocz(s->cluster_size);
3b46e624 755
585f8587
FB
756 s->refcount_table_offset = offset;
757 header.refcount_table_offset = cpu_to_be64(offset);
758 header.refcount_table_clusters = cpu_to_be32(1);
759 offset += s->cluster_size;
585f8587 760 s->refcount_block_offset = offset;
2d2431f0
AL
761
762 /* count how many refcount blocks needed */
763 tmp = offset >> s->cluster_bits;
764 ref_clusters = (tmp >> (s->cluster_bits - REFCOUNT_SHIFT)) + 1;
765 for (i=0; i < ref_clusters; i++) {
766 s->refcount_table[i] = cpu_to_be64(offset);
767 offset += s->cluster_size;
768 }
769
770 s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
585f8587
FB
771
772 /* update refcounts */
ed6ccf0f
KW
773 qcow2_create_refcount_update(s, 0, header_size);
774 qcow2_create_refcount_update(s, s->l1_table_offset,
775 l1_size * sizeof(uint64_t));
776 qcow2_create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
777 qcow2_create_refcount_update(s, s->refcount_block_offset,
778 ref_clusters * s->cluster_size);
3b46e624 779
585f8587
FB
780 /* write all the data */
781 write(fd, &header, sizeof(header));
782 if (backing_file) {
f965509c
AL
783 if (backing_format_len) {
784 char zero[16];
785 int d = ext_bf.len - backing_format_len;
786
787 memset(zero, 0, sizeof(zero));
788 cpu_to_be32s(&ext_bf.magic);
789 cpu_to_be32s(&ext_bf.len);
790 write(fd, &ext_bf, sizeof(ext_bf));
791 write(fd, backing_format, backing_format_len);
792 if (d>0) {
793 write(fd, zero, d);
794 }
795 }
585f8587
FB
796 write(fd, backing_file, backing_filename_len);
797 }
798 lseek(fd, s->l1_table_offset, SEEK_SET);
799 tmp = 0;
800 for(i = 0;i < l1_size; i++) {
801 write(fd, &tmp, sizeof(tmp));
802 }
803 lseek(fd, s->refcount_table_offset, SEEK_SET);
804 write(fd, s->refcount_table, s->cluster_size);
3b46e624 805
585f8587 806 lseek(fd, s->refcount_block_offset, SEEK_SET);
2d2431f0 807 write(fd, s->refcount_block, ref_clusters * s->cluster_size);
585f8587
FB
808
809 qemu_free(s->refcount_table);
810 qemu_free(s->refcount_block);
811 close(fd);
a35e1c17
KW
812
813 /* Preallocate metadata */
814 if (prealloc) {
815 BlockDriverState *bs;
816 bs = bdrv_new("");
817 bdrv_open(bs, filename, BDRV_O_CACHE_WB);
818 preallocate(bs);
819 bdrv_close(bs);
820 }
821
585f8587 822 return 0;
585f8587
FB
823}
824
0e7e1989
KW
825static int qcow_create(const char *filename, QEMUOptionParameter *options)
826{
827 const char *backing_file = NULL;
828 const char *backing_fmt = NULL;
829 uint64_t sectors = 0;
830 int flags = 0;
9ccb258e 831 size_t cluster_size = 65536;
a35e1c17 832 int prealloc = 0;
0e7e1989
KW
833
834 /* Read out options */
835 while (options && options->name) {
836 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
837 sectors = options->value.n / 512;
838 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
839 backing_file = options->value.s;
840 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
841 backing_fmt = options->value.s;
842 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
843 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
73c632ed
KW
844 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
845 if (options->value.n) {
846 cluster_size = options->value.n;
847 }
a35e1c17
KW
848 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
849 if (!options->value.s || !strcmp(options->value.s, "off")) {
850 prealloc = 0;
851 } else if (!strcmp(options->value.s, "metadata")) {
852 prealloc = 1;
853 } else {
854 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
855 options->value.s);
856 return -EINVAL;
857 }
0e7e1989
KW
858 }
859 options++;
860 }
861
a35e1c17
KW
862 if (backing_file && prealloc) {
863 fprintf(stderr, "Backing file and preallocation cannot be used at "
864 "the same time\n");
865 return -EINVAL;
866 }
867
73c632ed 868 return qcow_create2(filename, sectors, backing_file, backing_fmt, flags,
a35e1c17 869 cluster_size, prealloc);
f965509c
AL
870}
871
585f8587
FB
872static int qcow_make_empty(BlockDriverState *bs)
873{
874#if 0
875 /* XXX: not correct */
876 BDRVQcowState *s = bs->opaque;
877 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
878 int ret;
879
880 memset(s->l1_table, 0, l1_length);
881 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
ac674887 882 return -1;
585f8587
FB
883 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
884 if (ret < 0)
885 return ret;
3b46e624 886
585f8587
FB
887 l2_cache_reset(bs);
888#endif
889 return 0;
890}
891
892/* XXX: put compressed sectors first, then all the cluster aligned
893 tables to avoid losing bytes in alignment */
5fafdf24 894static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
585f8587
FB
895 const uint8_t *buf, int nb_sectors)
896{
897 BDRVQcowState *s = bs->opaque;
898 z_stream strm;
899 int ret, out_len;
900 uint8_t *out_buf;
901 uint64_t cluster_offset;
902
903 if (nb_sectors == 0) {
904 /* align end of file to a sector boundary to ease reading with
905 sector based I/Os */
906 cluster_offset = bdrv_getlength(s->hd);
907 cluster_offset = (cluster_offset + 511) & ~511;
908 bdrv_truncate(s->hd, cluster_offset);
909 return 0;
910 }
911
912 if (nb_sectors != s->cluster_sectors)
913 return -EINVAL;
914
915 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
585f8587
FB
916
917 /* best compression, small window, no zlib header */
918 memset(&strm, 0, sizeof(strm));
919 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
5fafdf24 920 Z_DEFLATED, -12,
585f8587
FB
921 9, Z_DEFAULT_STRATEGY);
922 if (ret != 0) {
923 qemu_free(out_buf);
924 return -1;
925 }
926
927 strm.avail_in = s->cluster_size;
928 strm.next_in = (uint8_t *)buf;
929 strm.avail_out = s->cluster_size;
930 strm.next_out = out_buf;
931
932 ret = deflate(&strm, Z_FINISH);
933 if (ret != Z_STREAM_END && ret != Z_OK) {
934 qemu_free(out_buf);
935 deflateEnd(&strm);
936 return -1;
937 }
938 out_len = strm.next_out - out_buf;
939
940 deflateEnd(&strm);
941
942 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
943 /* could not compress: write normal cluster */
ade40677 944 bdrv_write(bs, sector_num, buf, s->cluster_sectors);
585f8587 945 } else {
ed6ccf0f
KW
946 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
947 sector_num << 9, out_len);
52d893ec
AL
948 if (!cluster_offset)
949 return -1;
585f8587
FB
950 cluster_offset &= s->cluster_offset_mask;
951 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
952 qemu_free(out_buf);
953 return -1;
954 }
955 }
3b46e624 956
585f8587
FB
957 qemu_free(out_buf);
958 return 0;
959}
960
961static void qcow_flush(BlockDriverState *bs)
962{
963 BDRVQcowState *s = bs->opaque;
964 bdrv_flush(s->hd);
965}
966
45566e9c
CH
967static int64_t qcow_vm_state_offset(BDRVQcowState *s)
968{
969 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
970}
971
585f8587
FB
972static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
973{
974 BDRVQcowState *s = bs->opaque;
975 bdi->cluster_size = s->cluster_size;
45566e9c 976 bdi->vm_state_offset = qcow_vm_state_offset(s);
585f8587
FB
977 return 0;
978}
979
585f8587 980
e97fc193
AL
981static int qcow_check(BlockDriverState *bs)
982{
ed6ccf0f 983 return qcow2_check_refcounts(bs);
585f8587
FB
984}
985
986#if 0
987static void dump_refcounts(BlockDriverState *bs)
988{
989 BDRVQcowState *s = bs->opaque;
990 int64_t nb_clusters, k, k1, size;
991 int refcount;
992
993 size = bdrv_getlength(s->hd);
6db6c638 994 nb_clusters = size_to_clusters(s, size);
585f8587
FB
995 for(k = 0; k < nb_clusters;) {
996 k1 = k;
997 refcount = get_refcount(bs, k);
998 k++;
999 while (k < nb_clusters && get_refcount(bs, k) == refcount)
1000 k++;
1001 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
1002 }
1003}
1004#endif
585f8587 1005
45566e9c 1006static int qcow_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
178e08a5
AL
1007 int64_t pos, int size)
1008{
45566e9c 1009 BDRVQcowState *s = bs->opaque;
178e08a5
AL
1010 int growable = bs->growable;
1011
1012 bs->growable = 1;
45566e9c 1013 bdrv_pwrite(bs, qcow_vm_state_offset(s) + pos, buf, size);
178e08a5
AL
1014 bs->growable = growable;
1015
1016 return size;
1017}
1018
45566e9c 1019static int qcow_load_vmstate(BlockDriverState *bs, uint8_t *buf,
178e08a5
AL
1020 int64_t pos, int size)
1021{
45566e9c 1022 BDRVQcowState *s = bs->opaque;
178e08a5
AL
1023 int growable = bs->growable;
1024 int ret;
1025
1026 bs->growable = 1;
45566e9c 1027 ret = bdrv_pread(bs, qcow_vm_state_offset(s) + pos, buf, size);
178e08a5
AL
1028 bs->growable = growable;
1029
1030 return ret;
1031}
1032
0e7e1989 1033static QEMUOptionParameter qcow_create_options[] = {
db08adf5
KW
1034 {
1035 .name = BLOCK_OPT_SIZE,
1036 .type = OPT_SIZE,
1037 .help = "Virtual disk size"
1038 },
1039 {
1040 .name = BLOCK_OPT_BACKING_FILE,
1041 .type = OPT_STRING,
1042 .help = "File name of a base image"
1043 },
1044 {
1045 .name = BLOCK_OPT_BACKING_FMT,
1046 .type = OPT_STRING,
1047 .help = "Image format of the base image"
1048 },
1049 {
1050 .name = BLOCK_OPT_ENCRYPT,
1051 .type = OPT_FLAG,
1052 .help = "Encrypt the image"
1053 },
1054 {
1055 .name = BLOCK_OPT_CLUSTER_SIZE,
1056 .type = OPT_SIZE,
1057 .help = "qcow2 cluster size"
1058 },
a35e1c17
KW
1059 {
1060 .name = BLOCK_OPT_PREALLOC,
1061 .type = OPT_STRING,
1062 .help = "Preallocation mode (allowed values: off, metadata)"
1063 },
0e7e1989
KW
1064 { NULL }
1065};
1066
5efa9d5a 1067static BlockDriver bdrv_qcow2 = {
e60f469c
AJ
1068 .format_name = "qcow2",
1069 .instance_size = sizeof(BDRVQcowState),
1070 .bdrv_probe = qcow_probe,
1071 .bdrv_open = qcow_open,
1072 .bdrv_close = qcow_close,
1073 .bdrv_create = qcow_create,
1074 .bdrv_flush = qcow_flush,
1075 .bdrv_is_allocated = qcow_is_allocated,
1076 .bdrv_set_key = qcow_set_key,
1077 .bdrv_make_empty = qcow_make_empty,
1078
f141eafe
AL
1079 .bdrv_aio_readv = qcow_aio_readv,
1080 .bdrv_aio_writev = qcow_aio_writev,
585f8587
FB
1081 .bdrv_write_compressed = qcow_write_compressed,
1082
ed6ccf0f
KW
1083 .bdrv_snapshot_create = qcow2_snapshot_create,
1084 .bdrv_snapshot_goto = qcow2_snapshot_goto,
1085 .bdrv_snapshot_delete = qcow2_snapshot_delete,
1086 .bdrv_snapshot_list = qcow2_snapshot_list,
e60f469c 1087 .bdrv_get_info = qcow_get_info,
f965509c 1088
45566e9c
CH
1089 .bdrv_save_vmstate = qcow_save_vmstate,
1090 .bdrv_load_vmstate = qcow_load_vmstate,
178e08a5 1091
0e7e1989 1092 .create_options = qcow_create_options,
e97fc193 1093 .bdrv_check = qcow_check,
585f8587 1094};
5efa9d5a
AL
1095
1096static void bdrv_qcow2_init(void)
1097{
1098 bdrv_register(&bdrv_qcow2);
1099}
1100
1101block_init(bdrv_qcow2_init);