net: add packet length to NetPacketSent callback
[qemu-kvm/fedora.git] / block / qcow2.c
blob9acbddf94306f296011a3912fb9b849ec8036b4e
1 /*
2 * Block driver for the QCOW version 2 format
4 * Copyright (c) 2004-2006 Fabrice Bellard
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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.
24 #include "qemu-common.h"
25 #include "block_int.h"
26 #include "module.h"
27 #include <zlib.h>
28 #include "aes.h"
29 #include "block/qcow2.h"
32 Differences with QCOW:
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.
38 - Size of compressed clusters is stored in sectors to reduce bit usage
39 in the cluster offsets.
40 - Support for storing additional data (such as the VM state) in the
41 snapshots.
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.
47 //#define DEBUG_ALLOC
48 //#define DEBUG_ALLOC2
49 //#define DEBUG_EXT
52 typedef struct {
53 uint32_t magic;
54 uint32_t len;
55 } QCowExtension;
56 #define QCOW_EXT_MAGIC_END 0
57 #define QCOW_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
61 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
63 const QCowHeader *cow_header = (const void *)buf;
65 if (buf_size >= sizeof(QCowHeader) &&
66 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
67 be32_to_cpu(cow_header->version) == QCOW_VERSION)
68 return 100;
69 else
70 return 0;
74 /*
75 * read qcow2 extension and fill bs
76 * start reading from start_offset
77 * finish reading upon magic of value 0 or when end_offset reached
78 * unknown magic is skipped (future extension this version knows nothing about)
79 * return 0 upon success, non-0 otherwise
81 static int qcow_read_extensions(BlockDriverState *bs, uint64_t start_offset,
82 uint64_t end_offset)
84 BDRVQcowState *s = bs->opaque;
85 QCowExtension ext;
86 uint64_t offset;
88 #ifdef DEBUG_EXT
89 printf("qcow_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
90 #endif
91 offset = start_offset;
92 while (offset < end_offset) {
94 #ifdef DEBUG_EXT
95 /* Sanity check */
96 if (offset > s->cluster_size)
97 printf("qcow_handle_extension: suspicious offset %lu\n", offset);
99 printf("attemting to read extended header in offset %lu\n", offset);
100 #endif
102 if (bdrv_pread(s->hd, offset, &ext, sizeof(ext)) != sizeof(ext)) {
103 fprintf(stderr, "qcow_handle_extension: ERROR: pread fail from offset %llu\n",
104 (unsigned long long)offset);
105 return 1;
107 be32_to_cpus(&ext.magic);
108 be32_to_cpus(&ext.len);
109 offset += sizeof(ext);
110 #ifdef DEBUG_EXT
111 printf("ext.magic = 0x%x\n", ext.magic);
112 #endif
113 switch (ext.magic) {
114 case QCOW_EXT_MAGIC_END:
115 return 0;
117 case QCOW_EXT_MAGIC_BACKING_FORMAT:
118 if (ext.len >= sizeof(bs->backing_format)) {
119 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
120 " (>=%zu)\n",
121 ext.len, sizeof(bs->backing_format));
122 return 2;
124 if (bdrv_pread(s->hd, offset , bs->backing_format,
125 ext.len) != ext.len)
126 return 3;
127 bs->backing_format[ext.len] = '\0';
128 #ifdef DEBUG_EXT
129 printf("Qcow2: Got format extension %s\n", bs->backing_format);
130 #endif
131 offset += ((ext.len + 7) & ~7);
132 break;
134 default:
135 /* unknown magic -- just skip it */
136 offset += ((ext.len + 7) & ~7);
137 break;
141 return 0;
145 static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
147 BDRVQcowState *s = bs->opaque;
148 int len, i, shift, ret;
149 QCowHeader header;
150 uint64_t ext_end;
152 /* Performance is terrible right now with cache=writethrough due mainly
153 * to reference count updates. If the user does not explicitly specify
154 * a caching type, force to writeback caching.
156 if ((flags & BDRV_O_CACHE_DEF)) {
157 flags |= BDRV_O_CACHE_WB;
158 flags &= ~BDRV_O_CACHE_DEF;
160 ret = bdrv_file_open(&s->hd, filename, flags);
161 if (ret < 0)
162 return ret;
163 if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
164 goto fail;
165 be32_to_cpus(&header.magic);
166 be32_to_cpus(&header.version);
167 be64_to_cpus(&header.backing_file_offset);
168 be32_to_cpus(&header.backing_file_size);
169 be64_to_cpus(&header.size);
170 be32_to_cpus(&header.cluster_bits);
171 be32_to_cpus(&header.crypt_method);
172 be64_to_cpus(&header.l1_table_offset);
173 be32_to_cpus(&header.l1_size);
174 be64_to_cpus(&header.refcount_table_offset);
175 be32_to_cpus(&header.refcount_table_clusters);
176 be64_to_cpus(&header.snapshots_offset);
177 be32_to_cpus(&header.nb_snapshots);
179 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
180 goto fail;
181 if (header.size <= 1 ||
182 header.cluster_bits < MIN_CLUSTER_BITS ||
183 header.cluster_bits > MAX_CLUSTER_BITS)
184 goto fail;
185 if (header.crypt_method > QCOW_CRYPT_AES)
186 goto fail;
187 s->crypt_method_header = header.crypt_method;
188 if (s->crypt_method_header)
189 bs->encrypted = 1;
190 s->cluster_bits = header.cluster_bits;
191 s->cluster_size = 1 << s->cluster_bits;
192 s->cluster_sectors = 1 << (s->cluster_bits - 9);
193 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
194 s->l2_size = 1 << s->l2_bits;
195 bs->total_sectors = header.size / 512;
196 s->csize_shift = (62 - (s->cluster_bits - 8));
197 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
198 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
199 s->refcount_table_offset = header.refcount_table_offset;
200 s->refcount_table_size =
201 header.refcount_table_clusters << (s->cluster_bits - 3);
203 s->snapshots_offset = header.snapshots_offset;
204 s->nb_snapshots = header.nb_snapshots;
206 /* read the level 1 table */
207 s->l1_size = header.l1_size;
208 shift = s->cluster_bits + s->l2_bits;
209 s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
210 /* the L1 table must contain at least enough entries to put
211 header.size bytes */
212 if (s->l1_size < s->l1_vm_state_index)
213 goto fail;
214 s->l1_table_offset = header.l1_table_offset;
215 s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
216 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
217 s->l1_size * sizeof(uint64_t))
218 goto fail;
219 for(i = 0;i < s->l1_size; i++) {
220 be64_to_cpus(&s->l1_table[i]);
222 /* alloc L2 cache */
223 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
224 s->cluster_cache = qemu_malloc(s->cluster_size);
225 /* one more sector for decompressed data alignment */
226 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
227 + 512);
228 s->cluster_cache_offset = -1;
230 if (qcow2_refcount_init(bs) < 0)
231 goto fail;
233 /* read qcow2 extensions */
234 if (header.backing_file_offset)
235 ext_end = header.backing_file_offset;
236 else
237 ext_end = s->cluster_size;
238 if (qcow_read_extensions(bs, sizeof(header), ext_end))
239 goto fail;
241 /* read the backing file name */
242 if (header.backing_file_offset != 0) {
243 len = header.backing_file_size;
244 if (len > 1023)
245 len = 1023;
246 if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
247 goto fail;
248 bs->backing_file[len] = '\0';
250 if (qcow2_read_snapshots(bs) < 0)
251 goto fail;
253 #ifdef DEBUG_ALLOC
254 check_refcounts(bs);
255 #endif
256 return 0;
258 fail:
259 qcow2_free_snapshots(bs);
260 qcow2_refcount_close(bs);
261 qemu_free(s->l1_table);
262 qemu_free(s->l2_cache);
263 qemu_free(s->cluster_cache);
264 qemu_free(s->cluster_data);
265 bdrv_delete(s->hd);
266 return -1;
269 static int qcow_set_key(BlockDriverState *bs, const char *key)
271 BDRVQcowState *s = bs->opaque;
272 uint8_t keybuf[16];
273 int len, i;
275 memset(keybuf, 0, 16);
276 len = strlen(key);
277 if (len > 16)
278 len = 16;
279 /* XXX: we could compress the chars to 7 bits to increase
280 entropy */
281 for(i = 0;i < len;i++) {
282 keybuf[i] = key[i];
284 s->crypt_method = s->crypt_method_header;
286 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
287 return -1;
288 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
289 return -1;
290 #if 0
291 /* test */
293 uint8_t in[16];
294 uint8_t out[16];
295 uint8_t tmp[16];
296 for(i=0;i<16;i++)
297 in[i] = i;
298 AES_encrypt(in, tmp, &s->aes_encrypt_key);
299 AES_decrypt(tmp, out, &s->aes_decrypt_key);
300 for(i = 0; i < 16; i++)
301 printf(" %02x", tmp[i]);
302 printf("\n");
303 for(i = 0; i < 16; i++)
304 printf(" %02x", out[i]);
305 printf("\n");
307 #endif
308 return 0;
311 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
312 int nb_sectors, int *pnum)
314 uint64_t cluster_offset;
316 *pnum = nb_sectors;
317 cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, pnum);
319 return (cluster_offset != 0);
322 /* handle reading after the end of the backing file */
323 int qcow2_backing_read1(BlockDriverState *bs,
324 int64_t sector_num, uint8_t *buf, int nb_sectors)
326 int n1;
327 if ((sector_num + nb_sectors) <= bs->total_sectors)
328 return nb_sectors;
329 if (sector_num >= bs->total_sectors)
330 n1 = 0;
331 else
332 n1 = bs->total_sectors - sector_num;
333 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
334 return n1;
337 typedef struct QCowAIOCB {
338 BlockDriverAIOCB common;
339 int64_t sector_num;
340 QEMUIOVector *qiov;
341 uint8_t *buf;
342 void *orig_buf;
343 int nb_sectors;
344 int n;
345 uint64_t cluster_offset;
346 uint8_t *cluster_data;
347 BlockDriverAIOCB *hd_aiocb;
348 struct iovec hd_iov;
349 QEMUIOVector hd_qiov;
350 QEMUBH *bh;
351 QCowL2Meta l2meta;
352 } QCowAIOCB;
354 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
356 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
357 if (acb->hd_aiocb)
358 bdrv_aio_cancel(acb->hd_aiocb);
359 qemu_aio_release(acb);
362 static AIOPool qcow_aio_pool = {
363 .aiocb_size = sizeof(QCowAIOCB),
364 .cancel = qcow_aio_cancel,
367 static void qcow_aio_read_cb(void *opaque, int ret);
368 static void qcow_aio_read_bh(void *opaque)
370 QCowAIOCB *acb = opaque;
371 qemu_bh_delete(acb->bh);
372 acb->bh = NULL;
373 qcow_aio_read_cb(opaque, 0);
376 static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
378 if (acb->bh)
379 return -EIO;
381 acb->bh = qemu_bh_new(cb, acb);
382 if (!acb->bh)
383 return -EIO;
385 qemu_bh_schedule(acb->bh);
387 return 0;
390 static void qcow_aio_read_cb(void *opaque, int ret)
392 QCowAIOCB *acb = opaque;
393 BlockDriverState *bs = acb->common.bs;
394 BDRVQcowState *s = bs->opaque;
395 int index_in_cluster, n1;
397 acb->hd_aiocb = NULL;
398 if (ret < 0)
399 goto done;
401 /* post process the read buffer */
402 if (!acb->cluster_offset) {
403 /* nothing to do */
404 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
405 /* nothing to do */
406 } else {
407 if (s->crypt_method) {
408 qcow2_encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
409 acb->n, 0,
410 &s->aes_decrypt_key);
414 acb->nb_sectors -= acb->n;
415 acb->sector_num += acb->n;
416 acb->buf += acb->n * 512;
418 if (acb->nb_sectors == 0) {
419 /* request completed */
420 ret = 0;
421 goto done;
424 /* prepare next AIO request */
425 acb->n = acb->nb_sectors;
426 acb->cluster_offset =
427 qcow2_get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
428 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
430 if (!acb->cluster_offset) {
431 if (bs->backing_hd) {
432 /* read from the base image */
433 n1 = qcow2_backing_read1(bs->backing_hd, acb->sector_num,
434 acb->buf, acb->n);
435 if (n1 > 0) {
436 acb->hd_iov.iov_base = (void *)acb->buf;
437 acb->hd_iov.iov_len = acb->n * 512;
438 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
439 acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
440 &acb->hd_qiov, acb->n,
441 qcow_aio_read_cb, acb);
442 if (acb->hd_aiocb == NULL)
443 goto done;
444 } else {
445 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
446 if (ret < 0)
447 goto done;
449 } else {
450 /* Note: in this case, no need to wait */
451 memset(acb->buf, 0, 512 * acb->n);
452 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
453 if (ret < 0)
454 goto done;
456 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
457 /* add AIO support for compressed blocks ? */
458 if (qcow2_decompress_cluster(s, acb->cluster_offset) < 0)
459 goto done;
460 memcpy(acb->buf,
461 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
462 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
463 if (ret < 0)
464 goto done;
465 } else {
466 if ((acb->cluster_offset & 511) != 0) {
467 ret = -EIO;
468 goto done;
471 acb->hd_iov.iov_base = (void *)acb->buf;
472 acb->hd_iov.iov_len = acb->n * 512;
473 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
474 acb->hd_aiocb = bdrv_aio_readv(s->hd,
475 (acb->cluster_offset >> 9) + index_in_cluster,
476 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
477 if (acb->hd_aiocb == NULL)
478 goto done;
481 return;
482 done:
483 if (acb->qiov->niov > 1) {
484 qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
485 qemu_vfree(acb->orig_buf);
487 acb->common.cb(acb->common.opaque, ret);
488 qemu_aio_release(acb);
491 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
492 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
493 BlockDriverCompletionFunc *cb, void *opaque, int is_write)
495 QCowAIOCB *acb;
497 acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
498 if (!acb)
499 return NULL;
500 acb->hd_aiocb = NULL;
501 acb->sector_num = sector_num;
502 acb->qiov = qiov;
503 if (qiov->niov > 1) {
504 acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
505 if (is_write)
506 qemu_iovec_to_buffer(qiov, acb->buf);
507 } else {
508 acb->buf = (uint8_t *)qiov->iov->iov_base;
510 acb->nb_sectors = nb_sectors;
511 acb->n = 0;
512 acb->cluster_offset = 0;
513 acb->l2meta.nb_clusters = 0;
514 return acb;
517 static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
518 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
519 BlockDriverCompletionFunc *cb, void *opaque)
521 QCowAIOCB *acb;
523 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
524 if (!acb)
525 return NULL;
527 qcow_aio_read_cb(acb, 0);
528 return &acb->common;
531 static void qcow_aio_write_cb(void *opaque, int ret)
533 QCowAIOCB *acb = opaque;
534 BlockDriverState *bs = acb->common.bs;
535 BDRVQcowState *s = bs->opaque;
536 int index_in_cluster;
537 const uint8_t *src_buf;
538 int n_end;
540 acb->hd_aiocb = NULL;
542 if (ret < 0)
543 goto done;
545 if (qcow2_alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) {
546 qcow2_free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters);
547 goto done;
550 acb->nb_sectors -= acb->n;
551 acb->sector_num += acb->n;
552 acb->buf += acb->n * 512;
554 if (acb->nb_sectors == 0) {
555 /* request completed */
556 ret = 0;
557 goto done;
560 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
561 n_end = index_in_cluster + acb->nb_sectors;
562 if (s->crypt_method &&
563 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
564 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
566 acb->cluster_offset = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9,
567 index_in_cluster,
568 n_end, &acb->n, &acb->l2meta);
569 if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
570 ret = -EIO;
571 goto done;
573 if (s->crypt_method) {
574 if (!acb->cluster_data) {
575 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
576 s->cluster_size);
578 qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
579 acb->n, 1, &s->aes_encrypt_key);
580 src_buf = acb->cluster_data;
581 } else {
582 src_buf = acb->buf;
584 acb->hd_iov.iov_base = (void *)src_buf;
585 acb->hd_iov.iov_len = acb->n * 512;
586 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
587 acb->hd_aiocb = bdrv_aio_writev(s->hd,
588 (acb->cluster_offset >> 9) + index_in_cluster,
589 &acb->hd_qiov, acb->n,
590 qcow_aio_write_cb, acb);
591 if (acb->hd_aiocb == NULL)
592 goto done;
594 return;
596 done:
597 if (acb->qiov->niov > 1)
598 qemu_vfree(acb->orig_buf);
599 acb->common.cb(acb->common.opaque, ret);
600 qemu_aio_release(acb);
603 static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
604 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
605 BlockDriverCompletionFunc *cb, void *opaque)
607 BDRVQcowState *s = bs->opaque;
608 QCowAIOCB *acb;
610 s->cluster_cache_offset = -1; /* disable compressed cache */
612 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
613 if (!acb)
614 return NULL;
616 qcow_aio_write_cb(acb, 0);
617 return &acb->common;
620 static void qcow_close(BlockDriverState *bs)
622 BDRVQcowState *s = bs->opaque;
623 qemu_free(s->l1_table);
624 qemu_free(s->l2_cache);
625 qemu_free(s->cluster_cache);
626 qemu_free(s->cluster_data);
627 qcow2_refcount_close(bs);
628 bdrv_delete(s->hd);
631 static int get_bits_from_size(size_t size)
633 int res = 0;
635 if (size == 0) {
636 return -1;
639 while (size != 1) {
640 /* Not a power of two */
641 if (size & 1) {
642 return -1;
645 size >>= 1;
646 res++;
649 return res;
652 static int qcow_create2(const char *filename, int64_t total_size,
653 const char *backing_file, const char *backing_format,
654 int flags, size_t cluster_size)
657 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
658 int ref_clusters, backing_format_len = 0;
659 QCowHeader header;
660 uint64_t tmp, offset;
661 QCowCreateState s1, *s = &s1;
662 QCowExtension ext_bf = {0, 0};
665 memset(s, 0, sizeof(*s));
667 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
668 if (fd < 0)
669 return -1;
670 memset(&header, 0, sizeof(header));
671 header.magic = cpu_to_be32(QCOW_MAGIC);
672 header.version = cpu_to_be32(QCOW_VERSION);
673 header.size = cpu_to_be64(total_size * 512);
674 header_size = sizeof(header);
675 backing_filename_len = 0;
676 if (backing_file) {
677 if (backing_format) {
678 ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
679 backing_format_len = strlen(backing_format);
680 ext_bf.len = (backing_format_len + 7) & ~7;
681 header_size += ((sizeof(ext_bf) + ext_bf.len + 7) & ~7);
683 header.backing_file_offset = cpu_to_be64(header_size);
684 backing_filename_len = strlen(backing_file);
685 header.backing_file_size = cpu_to_be32(backing_filename_len);
686 header_size += backing_filename_len;
689 /* Cluster size */
690 s->cluster_bits = get_bits_from_size(cluster_size);
691 if (s->cluster_bits < MIN_CLUSTER_BITS ||
692 s->cluster_bits > MAX_CLUSTER_BITS)
694 fprintf(stderr, "Cluster size must be a power of two between "
695 "%d and %dk\n",
696 1 << MIN_CLUSTER_BITS,
697 1 << (MAX_CLUSTER_BITS - 10));
698 return -EINVAL;
700 s->cluster_size = 1 << s->cluster_bits;
702 header.cluster_bits = cpu_to_be32(s->cluster_bits);
703 header_size = (header_size + 7) & ~7;
704 if (flags & BLOCK_FLAG_ENCRYPT) {
705 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
706 } else {
707 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
709 l2_bits = s->cluster_bits - 3;
710 shift = s->cluster_bits + l2_bits;
711 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
712 offset = align_offset(header_size, s->cluster_size);
713 s->l1_table_offset = offset;
714 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
715 header.l1_size = cpu_to_be32(l1_size);
716 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
718 s->refcount_table = qemu_mallocz(s->cluster_size);
720 s->refcount_table_offset = offset;
721 header.refcount_table_offset = cpu_to_be64(offset);
722 header.refcount_table_clusters = cpu_to_be32(1);
723 offset += s->cluster_size;
724 s->refcount_block_offset = offset;
726 /* count how many refcount blocks needed */
727 tmp = offset >> s->cluster_bits;
728 ref_clusters = (tmp >> (s->cluster_bits - REFCOUNT_SHIFT)) + 1;
729 for (i=0; i < ref_clusters; i++) {
730 s->refcount_table[i] = cpu_to_be64(offset);
731 offset += s->cluster_size;
734 s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
736 /* update refcounts */
737 qcow2_create_refcount_update(s, 0, header_size);
738 qcow2_create_refcount_update(s, s->l1_table_offset,
739 l1_size * sizeof(uint64_t));
740 qcow2_create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
741 qcow2_create_refcount_update(s, s->refcount_block_offset,
742 ref_clusters * s->cluster_size);
744 /* write all the data */
745 write(fd, &header, sizeof(header));
746 if (backing_file) {
747 if (backing_format_len) {
748 char zero[16];
749 int d = ext_bf.len - backing_format_len;
751 memset(zero, 0, sizeof(zero));
752 cpu_to_be32s(&ext_bf.magic);
753 cpu_to_be32s(&ext_bf.len);
754 write(fd, &ext_bf, sizeof(ext_bf));
755 write(fd, backing_format, backing_format_len);
756 if (d>0) {
757 write(fd, zero, d);
760 write(fd, backing_file, backing_filename_len);
762 lseek(fd, s->l1_table_offset, SEEK_SET);
763 tmp = 0;
764 for(i = 0;i < l1_size; i++) {
765 write(fd, &tmp, sizeof(tmp));
767 lseek(fd, s->refcount_table_offset, SEEK_SET);
768 write(fd, s->refcount_table, s->cluster_size);
770 lseek(fd, s->refcount_block_offset, SEEK_SET);
771 write(fd, s->refcount_block, ref_clusters * s->cluster_size);
773 qemu_free(s->refcount_table);
774 qemu_free(s->refcount_block);
775 close(fd);
776 return 0;
779 static int qcow_create(const char *filename, QEMUOptionParameter *options)
781 const char *backing_file = NULL;
782 const char *backing_fmt = NULL;
783 uint64_t sectors = 0;
784 int flags = 0;
785 size_t cluster_size = 65536;
787 /* Read out options */
788 while (options && options->name) {
789 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
790 sectors = options->value.n / 512;
791 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
792 backing_file = options->value.s;
793 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
794 backing_fmt = options->value.s;
795 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
796 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
797 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
798 if (options->value.n) {
799 cluster_size = options->value.n;
802 options++;
805 return qcow_create2(filename, sectors, backing_file, backing_fmt, flags,
806 cluster_size);
809 static int qcow_make_empty(BlockDriverState *bs)
811 #if 0
812 /* XXX: not correct */
813 BDRVQcowState *s = bs->opaque;
814 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
815 int ret;
817 memset(s->l1_table, 0, l1_length);
818 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
819 return -1;
820 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
821 if (ret < 0)
822 return ret;
824 l2_cache_reset(bs);
825 #endif
826 return 0;
829 /* XXX: put compressed sectors first, then all the cluster aligned
830 tables to avoid losing bytes in alignment */
831 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
832 const uint8_t *buf, int nb_sectors)
834 BDRVQcowState *s = bs->opaque;
835 z_stream strm;
836 int ret, out_len;
837 uint8_t *out_buf;
838 uint64_t cluster_offset;
840 if (nb_sectors == 0) {
841 /* align end of file to a sector boundary to ease reading with
842 sector based I/Os */
843 cluster_offset = bdrv_getlength(s->hd);
844 cluster_offset = (cluster_offset + 511) & ~511;
845 bdrv_truncate(s->hd, cluster_offset);
846 return 0;
849 if (nb_sectors != s->cluster_sectors)
850 return -EINVAL;
852 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
854 /* best compression, small window, no zlib header */
855 memset(&strm, 0, sizeof(strm));
856 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
857 Z_DEFLATED, -12,
858 9, Z_DEFAULT_STRATEGY);
859 if (ret != 0) {
860 qemu_free(out_buf);
861 return -1;
864 strm.avail_in = s->cluster_size;
865 strm.next_in = (uint8_t *)buf;
866 strm.avail_out = s->cluster_size;
867 strm.next_out = out_buf;
869 ret = deflate(&strm, Z_FINISH);
870 if (ret != Z_STREAM_END && ret != Z_OK) {
871 qemu_free(out_buf);
872 deflateEnd(&strm);
873 return -1;
875 out_len = strm.next_out - out_buf;
877 deflateEnd(&strm);
879 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
880 /* could not compress: write normal cluster */
881 bdrv_write(bs, sector_num, buf, s->cluster_sectors);
882 } else {
883 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
884 sector_num << 9, out_len);
885 if (!cluster_offset)
886 return -1;
887 cluster_offset &= s->cluster_offset_mask;
888 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
889 qemu_free(out_buf);
890 return -1;
894 qemu_free(out_buf);
895 return 0;
898 static void qcow_flush(BlockDriverState *bs)
900 BDRVQcowState *s = bs->opaque;
901 bdrv_flush(s->hd);
904 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
906 BDRVQcowState *s = bs->opaque;
907 bdi->cluster_size = s->cluster_size;
908 bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
909 (s->cluster_bits + s->l2_bits);
910 return 0;
914 static int qcow_check(BlockDriverState *bs)
916 return qcow2_check_refcounts(bs);
919 #if 0
920 static void dump_refcounts(BlockDriverState *bs)
922 BDRVQcowState *s = bs->opaque;
923 int64_t nb_clusters, k, k1, size;
924 int refcount;
926 size = bdrv_getlength(s->hd);
927 nb_clusters = size_to_clusters(s, size);
928 for(k = 0; k < nb_clusters;) {
929 k1 = k;
930 refcount = get_refcount(bs, k);
931 k++;
932 while (k < nb_clusters && get_refcount(bs, k) == refcount)
933 k++;
934 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
937 #endif
939 static int qcow_put_buffer(BlockDriverState *bs, const uint8_t *buf,
940 int64_t pos, int size)
942 int growable = bs->growable;
944 bs->growable = 1;
945 bdrv_pwrite(bs, pos, buf, size);
946 bs->growable = growable;
948 return size;
951 static int qcow_get_buffer(BlockDriverState *bs, uint8_t *buf,
952 int64_t pos, int size)
954 int growable = bs->growable;
955 int ret;
957 bs->growable = 1;
958 ret = bdrv_pread(bs, pos, buf, size);
959 bs->growable = growable;
961 return ret;
964 static QEMUOptionParameter qcow_create_options[] = {
966 .name = BLOCK_OPT_SIZE,
967 .type = OPT_SIZE,
968 .help = "Virtual disk size"
971 .name = BLOCK_OPT_BACKING_FILE,
972 .type = OPT_STRING,
973 .help = "File name of a base image"
976 .name = BLOCK_OPT_BACKING_FMT,
977 .type = OPT_STRING,
978 .help = "Image format of the base image"
981 .name = BLOCK_OPT_ENCRYPT,
982 .type = OPT_FLAG,
983 .help = "Encrypt the image"
986 .name = BLOCK_OPT_CLUSTER_SIZE,
987 .type = OPT_SIZE,
988 .help = "qcow2 cluster size"
990 { NULL }
993 static BlockDriver bdrv_qcow2 = {
994 .format_name = "qcow2",
995 .instance_size = sizeof(BDRVQcowState),
996 .bdrv_probe = qcow_probe,
997 .bdrv_open = qcow_open,
998 .bdrv_close = qcow_close,
999 .bdrv_create = qcow_create,
1000 .bdrv_flush = qcow_flush,
1001 .bdrv_is_allocated = qcow_is_allocated,
1002 .bdrv_set_key = qcow_set_key,
1003 .bdrv_make_empty = qcow_make_empty,
1005 .bdrv_aio_readv = qcow_aio_readv,
1006 .bdrv_aio_writev = qcow_aio_writev,
1007 .bdrv_write_compressed = qcow_write_compressed,
1009 .bdrv_snapshot_create = qcow2_snapshot_create,
1010 .bdrv_snapshot_goto = qcow2_snapshot_goto,
1011 .bdrv_snapshot_delete = qcow2_snapshot_delete,
1012 .bdrv_snapshot_list = qcow2_snapshot_list,
1013 .bdrv_get_info = qcow_get_info,
1015 .bdrv_put_buffer = qcow_put_buffer,
1016 .bdrv_get_buffer = qcow_get_buffer,
1018 .create_options = qcow_create_options,
1019 .bdrv_check = qcow_check,
1022 static void bdrv_qcow2_init(void)
1024 bdrv_register(&bdrv_qcow2);
1027 block_init(bdrv_qcow2_init);