qcow2: Use QLIST_FOREACH_SAFE macro
[qemu.git] / block / qcow2.c
blobf3e3cba71c560f0b3904ab28e09c4f72a90073e1
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.
48 typedef struct {
49 uint32_t magic;
50 uint32_t len;
51 } QCowExtension;
52 #define QCOW_EXT_MAGIC_END 0
53 #define QCOW_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
55 static BlockDriver bdrv_qcow2;
57 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
59 const QCowHeader *cow_header = (const void *)buf;
61 if (buf_size >= sizeof(QCowHeader) &&
62 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
63 be32_to_cpu(cow_header->version) == QCOW_VERSION)
64 return 100;
65 else
66 return 0;
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
77 static int qcow_read_extensions(BlockDriverState *bs, uint64_t start_offset,
78 uint64_t end_offset)
80 BDRVQcowState *s = bs->opaque;
81 QCowExtension ext;
82 uint64_t offset;
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) {
90 #ifdef DEBUG_EXT
91 /* Sanity check */
92 if (offset > s->cluster_size)
93 printf("qcow_handle_extension: suspicious offset %lu\n", offset);
95 printf("attemting to read extended header in offset %lu\n", offset);
96 #endif
98 if (bdrv_pread(s->hd, offset, &ext, sizeof(ext)) != sizeof(ext)) {
99 fprintf(stderr, "qcow_handle_extension: ERROR: pread fail from offset %llu\n",
100 (unsigned long long)offset);
101 return 1;
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;
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"
116 " (>=%zu)\n",
117 ext.len, sizeof(bs->backing_format));
118 return 2;
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 = ((offset + ext.len + 7) & ~7);
128 break;
130 default:
131 /* unknown magic -- just skip it */
132 offset = ((offset + ext.len + 7) & ~7);
133 break;
137 return 0;
141 static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
143 BDRVQcowState *s = bs->opaque;
144 int len, i, shift, ret;
145 QCowHeader header;
146 uint64_t ext_end;
148 ret = bdrv_file_open(&s->hd, filename, flags);
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);
167 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
168 goto fail;
169 if (header.cluster_bits < MIN_CLUSTER_BITS ||
170 header.cluster_bits > MAX_CLUSTER_BITS)
171 goto fail;
172 if (header.crypt_method > QCOW_CRYPT_AES)
173 goto fail;
174 s->crypt_method_header = header.crypt_method;
175 if (s->crypt_method_header)
176 bs->encrypted = 1;
177 s->cluster_bits = header.cluster_bits;
178 s->cluster_size = 1 << s->cluster_bits;
179 s->cluster_sectors = 1 << (s->cluster_bits - 9);
180 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
181 s->l2_size = 1 << s->l2_bits;
182 bs->total_sectors = header.size / 512;
183 s->csize_shift = (62 - (s->cluster_bits - 8));
184 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
185 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
186 s->refcount_table_offset = header.refcount_table_offset;
187 s->refcount_table_size =
188 header.refcount_table_clusters << (s->cluster_bits - 3);
190 s->snapshots_offset = header.snapshots_offset;
191 s->nb_snapshots = header.nb_snapshots;
193 /* read the level 1 table */
194 s->l1_size = header.l1_size;
195 shift = s->cluster_bits + s->l2_bits;
196 s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
197 /* the L1 table must contain at least enough entries to put
198 header.size bytes */
199 if (s->l1_size < s->l1_vm_state_index)
200 goto fail;
201 s->l1_table_offset = header.l1_table_offset;
202 if (s->l1_size > 0) {
203 s->l1_table = qemu_mallocz(
204 align_offset(s->l1_size * sizeof(uint64_t), 512));
205 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
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]);
212 /* alloc L2 cache */
213 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
214 s->cluster_cache = qemu_malloc(s->cluster_size);
215 /* one more sector for decompressed data alignment */
216 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
217 + 512);
218 s->cluster_cache_offset = -1;
220 if (qcow2_refcount_init(bs) < 0)
221 goto fail;
223 QLIST_INIT(&s->cluster_allocs);
225 /* read qcow2 extensions */
226 if (header.backing_file_offset)
227 ext_end = header.backing_file_offset;
228 else
229 ext_end = s->cluster_size;
230 if (qcow_read_extensions(bs, sizeof(header), ext_end))
231 goto fail;
233 /* read the backing file name */
234 if (header.backing_file_offset != 0) {
235 len = header.backing_file_size;
236 if (len > 1023)
237 len = 1023;
238 if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
239 goto fail;
240 bs->backing_file[len] = '\0';
242 if (qcow2_read_snapshots(bs) < 0)
243 goto fail;
245 #ifdef DEBUG_ALLOC
246 qcow2_check_refcounts(bs);
247 #endif
248 return 0;
250 fail:
251 qcow2_free_snapshots(bs);
252 qcow2_refcount_close(bs);
253 qemu_free(s->l1_table);
254 qemu_free(s->l2_cache);
255 qemu_free(s->cluster_cache);
256 qemu_free(s->cluster_data);
257 bdrv_delete(s->hd);
258 return -1;
261 static int qcow_set_key(BlockDriverState *bs, const char *key)
263 BDRVQcowState *s = bs->opaque;
264 uint8_t keybuf[16];
265 int len, i;
267 memset(keybuf, 0, 16);
268 len = strlen(key);
269 if (len > 16)
270 len = 16;
271 /* XXX: we could compress the chars to 7 bits to increase
272 entropy */
273 for(i = 0;i < len;i++) {
274 keybuf[i] = key[i];
276 s->crypt_method = s->crypt_method_header;
278 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
279 return -1;
280 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
281 return -1;
282 #if 0
283 /* test */
285 uint8_t in[16];
286 uint8_t out[16];
287 uint8_t tmp[16];
288 for(i=0;i<16;i++)
289 in[i] = i;
290 AES_encrypt(in, tmp, &s->aes_encrypt_key);
291 AES_decrypt(tmp, out, &s->aes_decrypt_key);
292 for(i = 0; i < 16; i++)
293 printf(" %02x", tmp[i]);
294 printf("\n");
295 for(i = 0; i < 16; i++)
296 printf(" %02x", out[i]);
297 printf("\n");
299 #endif
300 return 0;
303 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
304 int nb_sectors, int *pnum)
306 uint64_t cluster_offset;
308 *pnum = nb_sectors;
309 cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, pnum);
311 return (cluster_offset != 0);
314 /* handle reading after the end of the backing file */
315 int qcow2_backing_read1(BlockDriverState *bs,
316 int64_t sector_num, uint8_t *buf, int nb_sectors)
318 int n1;
319 if ((sector_num + nb_sectors) <= bs->total_sectors)
320 return nb_sectors;
321 if (sector_num >= bs->total_sectors)
322 n1 = 0;
323 else
324 n1 = bs->total_sectors - sector_num;
325 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
326 return n1;
329 typedef struct QCowAIOCB {
330 BlockDriverAIOCB common;
331 int64_t sector_num;
332 QEMUIOVector *qiov;
333 uint8_t *buf;
334 void *orig_buf;
335 int remaining_sectors;
336 int cur_nr_sectors; /* number of sectors in current iteration */
337 uint64_t cluster_offset;
338 uint8_t *cluster_data;
339 BlockDriverAIOCB *hd_aiocb;
340 struct iovec hd_iov;
341 QEMUIOVector hd_qiov;
342 QEMUBH *bh;
343 QCowL2Meta l2meta;
344 QLIST_ENTRY(QCowAIOCB) next_depend;
345 } QCowAIOCB;
347 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
349 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
350 if (acb->hd_aiocb)
351 bdrv_aio_cancel(acb->hd_aiocb);
352 qemu_aio_release(acb);
355 static AIOPool qcow_aio_pool = {
356 .aiocb_size = sizeof(QCowAIOCB),
357 .cancel = qcow_aio_cancel,
360 static void qcow_aio_read_cb(void *opaque, int ret);
361 static void qcow_aio_read_bh(void *opaque)
363 QCowAIOCB *acb = opaque;
364 qemu_bh_delete(acb->bh);
365 acb->bh = NULL;
366 qcow_aio_read_cb(opaque, 0);
369 static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
371 if (acb->bh)
372 return -EIO;
374 acb->bh = qemu_bh_new(cb, acb);
375 if (!acb->bh)
376 return -EIO;
378 qemu_bh_schedule(acb->bh);
380 return 0;
383 static void qcow_aio_read_cb(void *opaque, int ret)
385 QCowAIOCB *acb = opaque;
386 BlockDriverState *bs = acb->common.bs;
387 BDRVQcowState *s = bs->opaque;
388 int index_in_cluster, n1;
390 acb->hd_aiocb = NULL;
391 if (ret < 0)
392 goto done;
394 /* post process the read buffer */
395 if (!acb->cluster_offset) {
396 /* nothing to do */
397 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
398 /* nothing to do */
399 } else {
400 if (s->crypt_method) {
401 qcow2_encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
402 acb->cur_nr_sectors, 0,
403 &s->aes_decrypt_key);
407 acb->remaining_sectors -= acb->cur_nr_sectors;
408 acb->sector_num += acb->cur_nr_sectors;
409 acb->buf += acb->cur_nr_sectors * 512;
411 if (acb->remaining_sectors == 0) {
412 /* request completed */
413 ret = 0;
414 goto done;
417 /* prepare next AIO request */
418 acb->cur_nr_sectors = acb->remaining_sectors;
419 acb->cluster_offset = qcow2_get_cluster_offset(bs, acb->sector_num << 9,
420 &acb->cur_nr_sectors);
421 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
423 if (!acb->cluster_offset) {
424 if (bs->backing_hd) {
425 /* read from the base image */
426 n1 = qcow2_backing_read1(bs->backing_hd, acb->sector_num,
427 acb->buf, acb->cur_nr_sectors);
428 if (n1 > 0) {
429 acb->hd_iov.iov_base = (void *)acb->buf;
430 acb->hd_iov.iov_len = acb->cur_nr_sectors * 512;
431 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
432 BLKDBG_EVENT(s->hd, BLKDBG_READ_BACKING_AIO);
433 acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
434 &acb->hd_qiov, acb->cur_nr_sectors,
435 qcow_aio_read_cb, acb);
436 if (acb->hd_aiocb == NULL)
437 goto done;
438 } else {
439 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
440 if (ret < 0)
441 goto done;
443 } else {
444 /* Note: in this case, no need to wait */
445 memset(acb->buf, 0, 512 * acb->cur_nr_sectors);
446 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
447 if (ret < 0)
448 goto done;
450 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
451 /* add AIO support for compressed blocks ? */
452 if (qcow2_decompress_cluster(s, acb->cluster_offset) < 0)
453 goto done;
454 memcpy(acb->buf, s->cluster_cache + index_in_cluster * 512,
455 512 * acb->cur_nr_sectors);
456 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
457 if (ret < 0)
458 goto done;
459 } else {
460 if ((acb->cluster_offset & 511) != 0) {
461 ret = -EIO;
462 goto done;
465 acb->hd_iov.iov_base = (void *)acb->buf;
466 acb->hd_iov.iov_len = acb->cur_nr_sectors * 512;
467 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
468 BLKDBG_EVENT(s->hd, BLKDBG_READ_AIO);
469 acb->hd_aiocb = bdrv_aio_readv(s->hd,
470 (acb->cluster_offset >> 9) + index_in_cluster,
471 &acb->hd_qiov, acb->cur_nr_sectors,
472 qcow_aio_read_cb, acb);
473 if (acb->hd_aiocb == NULL) {
474 ret = -EIO;
475 goto done;
479 return;
480 done:
481 if (acb->qiov->niov > 1) {
482 qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
483 qemu_vfree(acb->orig_buf);
485 acb->common.cb(acb->common.opaque, ret);
486 qemu_aio_release(acb);
489 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
490 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
491 BlockDriverCompletionFunc *cb, void *opaque, int is_write)
493 QCowAIOCB *acb;
495 acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
496 if (!acb)
497 return NULL;
498 acb->hd_aiocb = NULL;
499 acb->sector_num = sector_num;
500 acb->qiov = qiov;
501 if (qiov->niov > 1) {
502 acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
503 if (is_write)
504 qemu_iovec_to_buffer(qiov, acb->buf);
505 } else {
506 acb->buf = (uint8_t *)qiov->iov->iov_base;
508 acb->remaining_sectors = nb_sectors;
509 acb->cur_nr_sectors = 0;
510 acb->cluster_offset = 0;
511 acb->l2meta.nb_clusters = 0;
512 QLIST_INIT(&acb->l2meta.dependent_requests);
513 return acb;
516 static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
517 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
518 BlockDriverCompletionFunc *cb, void *opaque)
520 QCowAIOCB *acb;
522 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
523 if (!acb)
524 return NULL;
526 qcow_aio_read_cb(acb, 0);
527 return &acb->common;
530 static void qcow_aio_write_cb(void *opaque, int ret);
532 static void run_dependent_requests(QCowL2Meta *m)
534 QCowAIOCB *req;
535 QCowAIOCB *next;
537 /* Take the request off the list of running requests */
538 if (m->nb_clusters != 0) {
539 QLIST_REMOVE(m, next_in_flight);
542 /* Restart all dependent requests */
543 QLIST_FOREACH_SAFE(req, &m->dependent_requests, next_depend, next) {
544 qcow_aio_write_cb(req, 0);
547 /* Empty the list for the next part of the request */
548 QLIST_INIT(&m->dependent_requests);
551 static void qcow_aio_write_cb(void *opaque, int ret)
553 QCowAIOCB *acb = opaque;
554 BlockDriverState *bs = acb->common.bs;
555 BDRVQcowState *s = bs->opaque;
556 int index_in_cluster;
557 const uint8_t *src_buf;
558 int n_end;
560 acb->hd_aiocb = NULL;
562 if (ret >= 0) {
563 ret = qcow2_alloc_cluster_link_l2(bs, &acb->l2meta);
566 run_dependent_requests(&acb->l2meta);
568 if (ret < 0)
569 goto done;
571 acb->remaining_sectors -= acb->cur_nr_sectors;
572 acb->sector_num += acb->cur_nr_sectors;
573 acb->buf += acb->cur_nr_sectors * 512;
575 if (acb->remaining_sectors == 0) {
576 /* request completed */
577 ret = 0;
578 goto done;
581 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
582 n_end = index_in_cluster + acb->remaining_sectors;
583 if (s->crypt_method &&
584 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
585 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
587 ret = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9,
588 index_in_cluster, n_end, &acb->cur_nr_sectors, &acb->l2meta);
589 if (ret < 0) {
590 goto done;
593 acb->cluster_offset = acb->l2meta.cluster_offset;
595 /* Need to wait for another request? If so, we are done for now. */
596 if (acb->l2meta.nb_clusters == 0 && acb->l2meta.depends_on != NULL) {
597 QLIST_INSERT_HEAD(&acb->l2meta.depends_on->dependent_requests,
598 acb, next_depend);
599 return;
602 assert((acb->cluster_offset & 511) == 0);
604 if (s->crypt_method) {
605 if (!acb->cluster_data) {
606 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
607 s->cluster_size);
609 qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
610 acb->cur_nr_sectors, 1, &s->aes_encrypt_key);
611 src_buf = acb->cluster_data;
612 } else {
613 src_buf = acb->buf;
615 acb->hd_iov.iov_base = (void *)src_buf;
616 acb->hd_iov.iov_len = acb->cur_nr_sectors * 512;
617 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
618 BLKDBG_EVENT(s->hd, BLKDBG_WRITE_AIO);
619 acb->hd_aiocb = bdrv_aio_writev(s->hd,
620 (acb->cluster_offset >> 9) + index_in_cluster,
621 &acb->hd_qiov, acb->cur_nr_sectors,
622 qcow_aio_write_cb, acb);
623 if (acb->hd_aiocb == NULL) {
624 ret = -EIO;
625 goto fail;
628 return;
630 fail:
631 if (acb->l2meta.nb_clusters != 0) {
632 QLIST_REMOVE(&acb->l2meta, next_in_flight);
634 done:
635 if (acb->qiov->niov > 1)
636 qemu_vfree(acb->orig_buf);
637 acb->common.cb(acb->common.opaque, ret);
638 qemu_aio_release(acb);
641 static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
642 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
643 BlockDriverCompletionFunc *cb, void *opaque)
645 BDRVQcowState *s = bs->opaque;
646 QCowAIOCB *acb;
648 s->cluster_cache_offset = -1; /* disable compressed cache */
650 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
651 if (!acb)
652 return NULL;
654 qcow_aio_write_cb(acb, 0);
655 return &acb->common;
658 static void qcow_close(BlockDriverState *bs)
660 BDRVQcowState *s = bs->opaque;
661 qemu_free(s->l1_table);
662 qemu_free(s->l2_cache);
663 qemu_free(s->cluster_cache);
664 qemu_free(s->cluster_data);
665 qcow2_refcount_close(bs);
666 bdrv_delete(s->hd);
670 * Updates the variable length parts of the qcow2 header, i.e. the backing file
671 * name and all extensions. qcow2 was not designed to allow such changes, so if
672 * we run out of space (we can only use the first cluster) this function may
673 * fail.
675 * Returns 0 on success, -errno in error cases.
677 static int qcow2_update_ext_header(BlockDriverState *bs,
678 const char *backing_file, const char *backing_fmt)
680 size_t backing_file_len = 0;
681 size_t backing_fmt_len = 0;
682 BDRVQcowState *s = bs->opaque;
683 QCowExtension ext_backing_fmt = {0, 0};
684 int ret;
686 /* Backing file format doesn't make sense without a backing file */
687 if (backing_fmt && !backing_file) {
688 return -EINVAL;
691 /* Prepare the backing file format extension if needed */
692 if (backing_fmt) {
693 ext_backing_fmt.len = cpu_to_be32(strlen(backing_fmt));
694 ext_backing_fmt.magic = cpu_to_be32(QCOW_EXT_MAGIC_BACKING_FORMAT);
695 backing_fmt_len = ((sizeof(ext_backing_fmt)
696 + strlen(backing_fmt) + 7) & ~7);
699 /* Check if we can fit the new header into the first cluster */
700 if (backing_file) {
701 backing_file_len = strlen(backing_file);
704 size_t header_size = sizeof(QCowHeader) + backing_file_len
705 + backing_fmt_len;
707 if (header_size > s->cluster_size) {
708 return -ENOSPC;
711 /* Rewrite backing file name and qcow2 extensions */
712 size_t ext_size = header_size - sizeof(QCowHeader);
713 uint8_t buf[ext_size];
714 size_t offset = 0;
715 size_t backing_file_offset = 0;
717 if (backing_file) {
718 if (backing_fmt) {
719 int padding = backing_fmt_len -
720 (sizeof(ext_backing_fmt) + strlen(backing_fmt));
722 memcpy(buf + offset, &ext_backing_fmt, sizeof(ext_backing_fmt));
723 offset += sizeof(ext_backing_fmt);
725 memcpy(buf + offset, backing_fmt, strlen(backing_fmt));
726 offset += strlen(backing_fmt);
728 memset(buf + offset, 0, padding);
729 offset += padding;
732 memcpy(buf + offset, backing_file, backing_file_len);
733 backing_file_offset = sizeof(QCowHeader) + offset;
736 ret = bdrv_pwrite(s->hd, sizeof(QCowHeader), buf, ext_size);
737 if (ret < 0) {
738 goto fail;
741 /* Update header fields */
742 uint64_t be_backing_file_offset = cpu_to_be64(backing_file_offset);
743 uint32_t be_backing_file_size = cpu_to_be32(backing_file_len);
745 ret = bdrv_pwrite(s->hd, offsetof(QCowHeader, backing_file_offset),
746 &be_backing_file_offset, sizeof(uint64_t));
747 if (ret < 0) {
748 goto fail;
751 ret = bdrv_pwrite(s->hd, offsetof(QCowHeader, backing_file_size),
752 &be_backing_file_size, sizeof(uint32_t));
753 if (ret < 0) {
754 goto fail;
757 ret = 0;
758 fail:
759 return ret;
762 static int qcow2_change_backing_file(BlockDriverState *bs,
763 const char *backing_file, const char *backing_fmt)
765 return qcow2_update_ext_header(bs, backing_file, backing_fmt);
768 static int get_bits_from_size(size_t size)
770 int res = 0;
772 if (size == 0) {
773 return -1;
776 while (size != 1) {
777 /* Not a power of two */
778 if (size & 1) {
779 return -1;
782 size >>= 1;
783 res++;
786 return res;
790 static int preallocate(BlockDriverState *bs)
792 BDRVQcowState *s = bs->opaque;
793 uint64_t nb_sectors;
794 uint64_t offset;
795 int num;
796 int ret;
797 QCowL2Meta meta;
799 nb_sectors = bdrv_getlength(bs) >> 9;
800 offset = 0;
801 QLIST_INIT(&meta.dependent_requests);
802 meta.cluster_offset = 0;
804 while (nb_sectors) {
805 num = MIN(nb_sectors, INT_MAX >> 9);
806 ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta);
808 if (ret < 0) {
809 return -1;
812 if (qcow2_alloc_cluster_link_l2(bs, &meta) < 0) {
813 qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters);
814 return -1;
817 /* There are no dependent requests, but we need to remove our request
818 * from the list of in-flight requests */
819 run_dependent_requests(&meta);
821 /* TODO Preallocate data if requested */
823 nb_sectors -= num;
824 offset += num << 9;
828 * It is expected that the image file is large enough to actually contain
829 * all of the allocated clusters (otherwise we get failing reads after
830 * EOF). Extend the image to the last allocated sector.
832 if (meta.cluster_offset != 0) {
833 uint8_t buf[512];
834 memset(buf, 0, 512);
835 bdrv_write(s->hd, (meta.cluster_offset >> 9) + num - 1, buf, 1);
838 return 0;
841 static int qcow_create2(const char *filename, int64_t total_size,
842 const char *backing_file, const char *backing_format,
843 int flags, size_t cluster_size, int prealloc)
846 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
847 int ref_clusters, reftable_clusters, backing_format_len = 0;
848 int rounded_ext_bf_len = 0;
849 QCowHeader header;
850 uint64_t tmp, offset;
851 uint64_t old_ref_clusters;
852 QCowCreateState s1, *s = &s1;
853 QCowExtension ext_bf = {0, 0};
854 int ret;
856 memset(s, 0, sizeof(*s));
858 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
859 if (fd < 0)
860 return -errno;
861 memset(&header, 0, sizeof(header));
862 header.magic = cpu_to_be32(QCOW_MAGIC);
863 header.version = cpu_to_be32(QCOW_VERSION);
864 header.size = cpu_to_be64(total_size * 512);
865 header_size = sizeof(header);
866 backing_filename_len = 0;
867 if (backing_file) {
868 if (backing_format) {
869 ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
870 backing_format_len = strlen(backing_format);
871 ext_bf.len = backing_format_len;
872 rounded_ext_bf_len = (sizeof(ext_bf) + ext_bf.len + 7) & ~7;
873 header_size += rounded_ext_bf_len;
875 header.backing_file_offset = cpu_to_be64(header_size);
876 backing_filename_len = strlen(backing_file);
877 header.backing_file_size = cpu_to_be32(backing_filename_len);
878 header_size += backing_filename_len;
881 /* Cluster size */
882 s->cluster_bits = get_bits_from_size(cluster_size);
883 if (s->cluster_bits < MIN_CLUSTER_BITS ||
884 s->cluster_bits > MAX_CLUSTER_BITS)
886 fprintf(stderr, "Cluster size must be a power of two between "
887 "%d and %dk\n",
888 1 << MIN_CLUSTER_BITS,
889 1 << (MAX_CLUSTER_BITS - 10));
890 return -EINVAL;
892 s->cluster_size = 1 << s->cluster_bits;
894 header.cluster_bits = cpu_to_be32(s->cluster_bits);
895 header_size = (header_size + 7) & ~7;
896 if (flags & BLOCK_FLAG_ENCRYPT) {
897 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
898 } else {
899 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
901 l2_bits = s->cluster_bits - 3;
902 shift = s->cluster_bits + l2_bits;
903 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
904 offset = align_offset(header_size, s->cluster_size);
905 s->l1_table_offset = offset;
906 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
907 header.l1_size = cpu_to_be32(l1_size);
908 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
910 /* count how many refcount blocks needed */
912 #define NUM_CLUSTERS(bytes) \
913 (((bytes) + (s->cluster_size) - 1) / (s->cluster_size))
915 ref_clusters = NUM_CLUSTERS(NUM_CLUSTERS(offset) * sizeof(uint16_t));
917 do {
918 uint64_t image_clusters;
919 old_ref_clusters = ref_clusters;
921 /* Number of clusters used for the refcount table */
922 reftable_clusters = NUM_CLUSTERS(ref_clusters * sizeof(uint64_t));
924 /* Number of clusters that the whole image will have */
925 image_clusters = NUM_CLUSTERS(offset) + ref_clusters
926 + reftable_clusters;
928 /* Number of refcount blocks needed for the image */
929 ref_clusters = NUM_CLUSTERS(image_clusters * sizeof(uint16_t));
931 } while (ref_clusters != old_ref_clusters);
933 s->refcount_table = qemu_mallocz(reftable_clusters * s->cluster_size);
935 s->refcount_table_offset = offset;
936 header.refcount_table_offset = cpu_to_be64(offset);
937 header.refcount_table_clusters = cpu_to_be32(reftable_clusters);
938 offset += (reftable_clusters * s->cluster_size);
939 s->refcount_block_offset = offset;
941 for (i=0; i < ref_clusters; i++) {
942 s->refcount_table[i] = cpu_to_be64(offset);
943 offset += s->cluster_size;
946 s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
948 /* update refcounts */
949 qcow2_create_refcount_update(s, 0, header_size);
950 qcow2_create_refcount_update(s, s->l1_table_offset,
951 l1_size * sizeof(uint64_t));
952 qcow2_create_refcount_update(s, s->refcount_table_offset,
953 reftable_clusters * s->cluster_size);
954 qcow2_create_refcount_update(s, s->refcount_block_offset,
955 ref_clusters * s->cluster_size);
957 /* write all the data */
958 ret = qemu_write_full(fd, &header, sizeof(header));
959 if (ret != sizeof(header)) {
960 ret = -errno;
961 goto exit;
963 if (backing_file) {
964 if (backing_format_len) {
965 char zero[16];
966 int padding = rounded_ext_bf_len - (ext_bf.len + sizeof(ext_bf));
968 memset(zero, 0, sizeof(zero));
969 cpu_to_be32s(&ext_bf.magic);
970 cpu_to_be32s(&ext_bf.len);
971 ret = qemu_write_full(fd, &ext_bf, sizeof(ext_bf));
972 if (ret != sizeof(ext_bf)) {
973 ret = -errno;
974 goto exit;
976 ret = qemu_write_full(fd, backing_format, backing_format_len);
977 if (ret != backing_format_len) {
978 ret = -errno;
979 goto exit;
981 if (padding > 0) {
982 ret = qemu_write_full(fd, zero, padding);
983 if (ret != padding) {
984 ret = -errno;
985 goto exit;
989 ret = qemu_write_full(fd, backing_file, backing_filename_len);
990 if (ret != backing_filename_len) {
991 ret = -errno;
992 goto exit;
995 lseek(fd, s->l1_table_offset, SEEK_SET);
996 tmp = 0;
997 for(i = 0;i < l1_size; i++) {
998 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
999 if (ret != sizeof(tmp)) {
1000 ret = -errno;
1001 goto exit;
1004 lseek(fd, s->refcount_table_offset, SEEK_SET);
1005 ret = qemu_write_full(fd, s->refcount_table,
1006 reftable_clusters * s->cluster_size);
1007 if (ret != reftable_clusters * s->cluster_size) {
1008 ret = -errno;
1009 goto exit;
1012 lseek(fd, s->refcount_block_offset, SEEK_SET);
1013 ret = qemu_write_full(fd, s->refcount_block,
1014 ref_clusters * s->cluster_size);
1015 if (ret != ref_clusters * s->cluster_size) {
1016 ret = -errno;
1017 goto exit;
1020 ret = 0;
1021 exit:
1022 qemu_free(s->refcount_table);
1023 qemu_free(s->refcount_block);
1024 close(fd);
1026 /* Preallocate metadata */
1027 if (ret == 0 && prealloc) {
1028 BlockDriverState *bs;
1029 bs = bdrv_new("");
1030 bdrv_open(bs, filename, BDRV_O_CACHE_WB | BDRV_O_RDWR, &bdrv_qcow2);
1031 preallocate(bs);
1032 bdrv_close(bs);
1035 return ret;
1038 static int qcow_create(const char *filename, QEMUOptionParameter *options)
1040 const char *backing_file = NULL;
1041 const char *backing_fmt = NULL;
1042 uint64_t sectors = 0;
1043 int flags = 0;
1044 size_t cluster_size = 65536;
1045 int prealloc = 0;
1047 /* Read out options */
1048 while (options && options->name) {
1049 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1050 sectors = options->value.n / 512;
1051 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1052 backing_file = options->value.s;
1053 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
1054 backing_fmt = options->value.s;
1055 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
1056 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
1057 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
1058 if (options->value.n) {
1059 cluster_size = options->value.n;
1061 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
1062 if (!options->value.s || !strcmp(options->value.s, "off")) {
1063 prealloc = 0;
1064 } else if (!strcmp(options->value.s, "metadata")) {
1065 prealloc = 1;
1066 } else {
1067 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
1068 options->value.s);
1069 return -EINVAL;
1072 options++;
1075 if (backing_file && prealloc) {
1076 fprintf(stderr, "Backing file and preallocation cannot be used at "
1077 "the same time\n");
1078 return -EINVAL;
1081 return qcow_create2(filename, sectors, backing_file, backing_fmt, flags,
1082 cluster_size, prealloc);
1085 static int qcow_make_empty(BlockDriverState *bs)
1087 #if 0
1088 /* XXX: not correct */
1089 BDRVQcowState *s = bs->opaque;
1090 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1091 int ret;
1093 memset(s->l1_table, 0, l1_length);
1094 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
1095 return -1;
1096 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1097 if (ret < 0)
1098 return ret;
1100 l2_cache_reset(bs);
1101 #endif
1102 return 0;
1105 /* XXX: put compressed sectors first, then all the cluster aligned
1106 tables to avoid losing bytes in alignment */
1107 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1108 const uint8_t *buf, int nb_sectors)
1110 BDRVQcowState *s = bs->opaque;
1111 z_stream strm;
1112 int ret, out_len;
1113 uint8_t *out_buf;
1114 uint64_t cluster_offset;
1116 if (nb_sectors == 0) {
1117 /* align end of file to a sector boundary to ease reading with
1118 sector based I/Os */
1119 cluster_offset = bdrv_getlength(s->hd);
1120 cluster_offset = (cluster_offset + 511) & ~511;
1121 bdrv_truncate(s->hd, cluster_offset);
1122 return 0;
1125 if (nb_sectors != s->cluster_sectors)
1126 return -EINVAL;
1128 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1130 /* best compression, small window, no zlib header */
1131 memset(&strm, 0, sizeof(strm));
1132 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1133 Z_DEFLATED, -12,
1134 9, Z_DEFAULT_STRATEGY);
1135 if (ret != 0) {
1136 qemu_free(out_buf);
1137 return -1;
1140 strm.avail_in = s->cluster_size;
1141 strm.next_in = (uint8_t *)buf;
1142 strm.avail_out = s->cluster_size;
1143 strm.next_out = out_buf;
1145 ret = deflate(&strm, Z_FINISH);
1146 if (ret != Z_STREAM_END && ret != Z_OK) {
1147 qemu_free(out_buf);
1148 deflateEnd(&strm);
1149 return -1;
1151 out_len = strm.next_out - out_buf;
1153 deflateEnd(&strm);
1155 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1156 /* could not compress: write normal cluster */
1157 bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1158 } else {
1159 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1160 sector_num << 9, out_len);
1161 if (!cluster_offset)
1162 return -1;
1163 cluster_offset &= s->cluster_offset_mask;
1164 BLKDBG_EVENT(s->hd, BLKDBG_WRITE_COMPRESSED);
1165 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1166 qemu_free(out_buf);
1167 return -1;
1171 qemu_free(out_buf);
1172 return 0;
1175 static void qcow_flush(BlockDriverState *bs)
1177 BDRVQcowState *s = bs->opaque;
1178 bdrv_flush(s->hd);
1181 static BlockDriverAIOCB *qcow_aio_flush(BlockDriverState *bs,
1182 BlockDriverCompletionFunc *cb, void *opaque)
1184 BDRVQcowState *s = bs->opaque;
1186 return bdrv_aio_flush(s->hd, cb, opaque);
1189 static int64_t qcow_vm_state_offset(BDRVQcowState *s)
1191 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1194 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1196 BDRVQcowState *s = bs->opaque;
1197 bdi->cluster_size = s->cluster_size;
1198 bdi->vm_state_offset = qcow_vm_state_offset(s);
1199 return 0;
1203 static int qcow_check(BlockDriverState *bs)
1205 return qcow2_check_refcounts(bs);
1208 #if 0
1209 static void dump_refcounts(BlockDriverState *bs)
1211 BDRVQcowState *s = bs->opaque;
1212 int64_t nb_clusters, k, k1, size;
1213 int refcount;
1215 size = bdrv_getlength(s->hd);
1216 nb_clusters = size_to_clusters(s, size);
1217 for(k = 0; k < nb_clusters;) {
1218 k1 = k;
1219 refcount = get_refcount(bs, k);
1220 k++;
1221 while (k < nb_clusters && get_refcount(bs, k) == refcount)
1222 k++;
1223 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
1226 #endif
1228 static int qcow_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1229 int64_t pos, int size)
1231 BDRVQcowState *s = bs->opaque;
1232 int growable = bs->growable;
1233 int ret;
1235 BLKDBG_EVENT(s->hd, BLKDBG_VMSTATE_SAVE);
1236 bs->growable = 1;
1237 ret = bdrv_pwrite(bs, qcow_vm_state_offset(s) + pos, buf, size);
1238 bs->growable = growable;
1240 return ret;
1243 static int qcow_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1244 int64_t pos, int size)
1246 BDRVQcowState *s = bs->opaque;
1247 int growable = bs->growable;
1248 int ret;
1250 BLKDBG_EVENT(s->hd, BLKDBG_VMSTATE_LOAD);
1251 bs->growable = 1;
1252 ret = bdrv_pread(bs, qcow_vm_state_offset(s) + pos, buf, size);
1253 bs->growable = growable;
1255 return ret;
1258 static QEMUOptionParameter qcow_create_options[] = {
1260 .name = BLOCK_OPT_SIZE,
1261 .type = OPT_SIZE,
1262 .help = "Virtual disk size"
1265 .name = BLOCK_OPT_BACKING_FILE,
1266 .type = OPT_STRING,
1267 .help = "File name of a base image"
1270 .name = BLOCK_OPT_BACKING_FMT,
1271 .type = OPT_STRING,
1272 .help = "Image format of the base image"
1275 .name = BLOCK_OPT_ENCRYPT,
1276 .type = OPT_FLAG,
1277 .help = "Encrypt the image"
1280 .name = BLOCK_OPT_CLUSTER_SIZE,
1281 .type = OPT_SIZE,
1282 .help = "qcow2 cluster size"
1285 .name = BLOCK_OPT_PREALLOC,
1286 .type = OPT_STRING,
1287 .help = "Preallocation mode (allowed values: off, metadata)"
1289 { NULL }
1292 static BlockDriver bdrv_qcow2 = {
1293 .format_name = "qcow2",
1294 .instance_size = sizeof(BDRVQcowState),
1295 .bdrv_probe = qcow_probe,
1296 .bdrv_open = qcow_open,
1297 .bdrv_close = qcow_close,
1298 .bdrv_create = qcow_create,
1299 .bdrv_flush = qcow_flush,
1300 .bdrv_is_allocated = qcow_is_allocated,
1301 .bdrv_set_key = qcow_set_key,
1302 .bdrv_make_empty = qcow_make_empty,
1304 .bdrv_aio_readv = qcow_aio_readv,
1305 .bdrv_aio_writev = qcow_aio_writev,
1306 .bdrv_aio_flush = qcow_aio_flush,
1307 .bdrv_write_compressed = qcow_write_compressed,
1309 .bdrv_snapshot_create = qcow2_snapshot_create,
1310 .bdrv_snapshot_goto = qcow2_snapshot_goto,
1311 .bdrv_snapshot_delete = qcow2_snapshot_delete,
1312 .bdrv_snapshot_list = qcow2_snapshot_list,
1313 .bdrv_get_info = qcow_get_info,
1315 .bdrv_save_vmstate = qcow_save_vmstate,
1316 .bdrv_load_vmstate = qcow_load_vmstate,
1318 .bdrv_change_backing_file = qcow2_change_backing_file,
1320 .create_options = qcow_create_options,
1321 .bdrv_check = qcow_check,
1324 static void bdrv_qcow2_init(void)
1326 bdrv_register(&bdrv_qcow2);
1329 block_init(bdrv_qcow2_init);