Merge remote branch 'qemu-kvm/uq/master' into pulls
[qemu/aliguori-queue.git] / block / qcow2.c
blobbf8170ed02bee983b2da37ffe3b5a279397ac5d0
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
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 acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
433 &acb->hd_qiov, acb->cur_nr_sectors,
434 qcow_aio_read_cb, acb);
435 if (acb->hd_aiocb == NULL)
436 goto done;
437 } else {
438 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
439 if (ret < 0)
440 goto done;
442 } else {
443 /* Note: in this case, no need to wait */
444 memset(acb->buf, 0, 512 * acb->cur_nr_sectors);
445 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
446 if (ret < 0)
447 goto done;
449 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
450 /* add AIO support for compressed blocks ? */
451 if (qcow2_decompress_cluster(s, acb->cluster_offset) < 0)
452 goto done;
453 memcpy(acb->buf, s->cluster_cache + index_in_cluster * 512,
454 512 * acb->cur_nr_sectors);
455 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
456 if (ret < 0)
457 goto done;
458 } else {
459 if ((acb->cluster_offset & 511) != 0) {
460 ret = -EIO;
461 goto done;
464 acb->hd_iov.iov_base = (void *)acb->buf;
465 acb->hd_iov.iov_len = acb->cur_nr_sectors * 512;
466 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
467 acb->hd_aiocb = bdrv_aio_readv(s->hd,
468 (acb->cluster_offset >> 9) + index_in_cluster,
469 &acb->hd_qiov, acb->cur_nr_sectors,
470 qcow_aio_read_cb, acb);
471 if (acb->hd_aiocb == NULL)
472 goto done;
475 return;
476 done:
477 if (acb->qiov->niov > 1) {
478 qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
479 qemu_vfree(acb->orig_buf);
481 acb->common.cb(acb->common.opaque, ret);
482 qemu_aio_release(acb);
485 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
486 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
487 BlockDriverCompletionFunc *cb, void *opaque, int is_write)
489 QCowAIOCB *acb;
491 acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
492 if (!acb)
493 return NULL;
494 acb->hd_aiocb = NULL;
495 acb->sector_num = sector_num;
496 acb->qiov = qiov;
497 if (qiov->niov > 1) {
498 acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
499 if (is_write)
500 qemu_iovec_to_buffer(qiov, acb->buf);
501 } else {
502 acb->buf = (uint8_t *)qiov->iov->iov_base;
504 acb->remaining_sectors = nb_sectors;
505 acb->cur_nr_sectors = 0;
506 acb->cluster_offset = 0;
507 acb->l2meta.nb_clusters = 0;
508 QLIST_INIT(&acb->l2meta.dependent_requests);
509 return acb;
512 static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
513 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
514 BlockDriverCompletionFunc *cb, void *opaque)
516 QCowAIOCB *acb;
518 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
519 if (!acb)
520 return NULL;
522 qcow_aio_read_cb(acb, 0);
523 return &acb->common;
526 static void qcow_aio_write_cb(void *opaque, int ret);
528 static void run_dependent_requests(QCowL2Meta *m)
530 QCowAIOCB *req;
531 QCowAIOCB *next;
533 /* Take the request off the list of running requests */
534 if (m->nb_clusters != 0) {
535 QLIST_REMOVE(m, next_in_flight);
539 * Restart all dependent requests.
540 * Can't use QLIST_FOREACH here - the next link might not be the same
541 * any more after the callback (request could depend on a different
542 * request now)
544 for (req = m->dependent_requests.lh_first; req != NULL; req = next) {
545 next = req->next_depend.le_next;
546 qcow_aio_write_cb(req, 0);
549 /* Empty the list for the next part of the request */
550 QLIST_INIT(&m->dependent_requests);
553 static void qcow_aio_write_cb(void *opaque, int ret)
555 QCowAIOCB *acb = opaque;
556 BlockDriverState *bs = acb->common.bs;
557 BDRVQcowState *s = bs->opaque;
558 int index_in_cluster;
559 const uint8_t *src_buf;
560 int n_end;
562 acb->hd_aiocb = NULL;
564 if (ret >= 0) {
565 ret = qcow2_alloc_cluster_link_l2(bs, &acb->l2meta);
568 run_dependent_requests(&acb->l2meta);
570 if (ret < 0)
571 goto done;
573 acb->remaining_sectors -= acb->cur_nr_sectors;
574 acb->sector_num += acb->cur_nr_sectors;
575 acb->buf += acb->cur_nr_sectors * 512;
577 if (acb->remaining_sectors == 0) {
578 /* request completed */
579 ret = 0;
580 goto done;
583 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
584 n_end = index_in_cluster + acb->remaining_sectors;
585 if (s->crypt_method &&
586 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
587 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
589 ret = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9,
590 index_in_cluster, n_end, &acb->cur_nr_sectors, &acb->l2meta);
591 if (ret < 0) {
592 goto done;
595 acb->cluster_offset = acb->l2meta.cluster_offset;
597 /* Need to wait for another request? If so, we are done for now. */
598 if (acb->l2meta.nb_clusters == 0 && acb->l2meta.depends_on != NULL) {
599 QLIST_INSERT_HEAD(&acb->l2meta.depends_on->dependent_requests,
600 acb, next_depend);
601 return;
604 assert((acb->cluster_offset & 511) == 0);
606 if (s->crypt_method) {
607 if (!acb->cluster_data) {
608 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
609 s->cluster_size);
611 qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
612 acb->cur_nr_sectors, 1, &s->aes_encrypt_key);
613 src_buf = acb->cluster_data;
614 } else {
615 src_buf = acb->buf;
617 acb->hd_iov.iov_base = (void *)src_buf;
618 acb->hd_iov.iov_len = acb->cur_nr_sectors * 512;
619 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
620 acb->hd_aiocb = bdrv_aio_writev(s->hd,
621 (acb->cluster_offset >> 9) + index_in_cluster,
622 &acb->hd_qiov, acb->cur_nr_sectors,
623 qcow_aio_write_cb, acb);
624 if (acb->hd_aiocb == NULL)
625 goto done;
627 return;
629 done:
630 if (acb->qiov->niov > 1)
631 qemu_vfree(acb->orig_buf);
632 acb->common.cb(acb->common.opaque, ret);
633 qemu_aio_release(acb);
636 static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
637 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
638 BlockDriverCompletionFunc *cb, void *opaque)
640 BDRVQcowState *s = bs->opaque;
641 QCowAIOCB *acb;
643 s->cluster_cache_offset = -1; /* disable compressed cache */
645 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
646 if (!acb)
647 return NULL;
649 qcow_aio_write_cb(acb, 0);
650 return &acb->common;
653 static void qcow_close(BlockDriverState *bs)
655 BDRVQcowState *s = bs->opaque;
656 qemu_free(s->l1_table);
657 qemu_free(s->l2_cache);
658 qemu_free(s->cluster_cache);
659 qemu_free(s->cluster_data);
660 qcow2_refcount_close(bs);
661 bdrv_delete(s->hd);
665 * Updates the variable length parts of the qcow2 header, i.e. the backing file
666 * name and all extensions. qcow2 was not designed to allow such changes, so if
667 * we run out of space (we can only use the first cluster) this function may
668 * fail.
670 * Returns 0 on success, -errno in error cases.
672 static int qcow2_update_ext_header(BlockDriverState *bs,
673 const char *backing_file, const char *backing_fmt)
675 size_t backing_file_len = 0;
676 size_t backing_fmt_len = 0;
677 BDRVQcowState *s = bs->opaque;
678 QCowExtension ext_backing_fmt = {0, 0};
679 int ret;
681 /* Backing file format doesn't make sense without a backing file */
682 if (backing_fmt && !backing_file) {
683 return -EINVAL;
686 /* Prepare the backing file format extension if needed */
687 if (backing_fmt) {
688 ext_backing_fmt.len = cpu_to_be32(strlen(backing_fmt));
689 ext_backing_fmt.magic = cpu_to_be32(QCOW_EXT_MAGIC_BACKING_FORMAT);
690 backing_fmt_len = ((sizeof(ext_backing_fmt)
691 + strlen(backing_fmt) + 7) & ~7);
694 /* Check if we can fit the new header into the first cluster */
695 if (backing_file) {
696 backing_file_len = strlen(backing_file);
699 size_t header_size = sizeof(QCowHeader) + backing_file_len
700 + backing_fmt_len;
702 if (header_size > s->cluster_size) {
703 return -ENOSPC;
706 /* Rewrite backing file name and qcow2 extensions */
707 size_t ext_size = header_size - sizeof(QCowHeader);
708 uint8_t buf[ext_size];
709 size_t offset = 0;
710 size_t backing_file_offset = 0;
712 if (backing_file) {
713 if (backing_fmt) {
714 int padding = backing_fmt_len -
715 (sizeof(ext_backing_fmt) + strlen(backing_fmt));
717 memcpy(buf + offset, &ext_backing_fmt, sizeof(ext_backing_fmt));
718 offset += sizeof(ext_backing_fmt);
720 memcpy(buf + offset, backing_fmt, strlen(backing_fmt));
721 offset += strlen(backing_fmt);
723 memset(buf + offset, 0, padding);
724 offset += padding;
727 memcpy(buf + offset, backing_file, backing_file_len);
728 backing_file_offset = sizeof(QCowHeader) + offset;
731 ret = bdrv_pwrite(s->hd, sizeof(QCowHeader), buf, ext_size);
732 if (ret < 0) {
733 goto fail;
736 /* Update header fields */
737 uint64_t be_backing_file_offset = cpu_to_be64(backing_file_offset);
738 uint32_t be_backing_file_size = cpu_to_be32(backing_file_len);
740 ret = bdrv_pwrite(s->hd, offsetof(QCowHeader, backing_file_offset),
741 &be_backing_file_offset, sizeof(uint64_t));
742 if (ret < 0) {
743 goto fail;
746 ret = bdrv_pwrite(s->hd, offsetof(QCowHeader, backing_file_size),
747 &be_backing_file_size, sizeof(uint32_t));
748 if (ret < 0) {
749 goto fail;
752 ret = 0;
753 fail:
754 return ret;
757 static int qcow2_change_backing_file(BlockDriverState *bs,
758 const char *backing_file, const char *backing_fmt)
760 return qcow2_update_ext_header(bs, backing_file, backing_fmt);
763 static int get_bits_from_size(size_t size)
765 int res = 0;
767 if (size == 0) {
768 return -1;
771 while (size != 1) {
772 /* Not a power of two */
773 if (size & 1) {
774 return -1;
777 size >>= 1;
778 res++;
781 return res;
785 static int preallocate(BlockDriverState *bs)
787 BDRVQcowState *s = bs->opaque;
788 uint64_t nb_sectors;
789 uint64_t offset;
790 int num;
791 int ret;
792 QCowL2Meta meta;
794 nb_sectors = bdrv_getlength(bs) >> 9;
795 offset = 0;
796 QLIST_INIT(&meta.dependent_requests);
797 meta.cluster_offset = 0;
799 while (nb_sectors) {
800 num = MIN(nb_sectors, INT_MAX >> 9);
801 ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta);
803 if (ret < 0) {
804 return -1;
807 if (qcow2_alloc_cluster_link_l2(bs, &meta) < 0) {
808 qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters);
809 return -1;
812 /* There are no dependent requests, but we need to remove our request
813 * from the list of in-flight requests */
814 run_dependent_requests(&meta);
816 /* TODO Preallocate data if requested */
818 nb_sectors -= num;
819 offset += num << 9;
823 * It is expected that the image file is large enough to actually contain
824 * all of the allocated clusters (otherwise we get failing reads after
825 * EOF). Extend the image to the last allocated sector.
827 if (meta.cluster_offset != 0) {
828 uint8_t buf[512];
829 memset(buf, 0, 512);
830 bdrv_write(s->hd, (meta.cluster_offset >> 9) + num - 1, buf, 1);
833 return 0;
836 static int qcow_create2(const char *filename, int64_t total_size,
837 const char *backing_file, const char *backing_format,
838 int flags, size_t cluster_size, int prealloc)
841 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
842 int ref_clusters, backing_format_len = 0;
843 int rounded_ext_bf_len = 0;
844 QCowHeader header;
845 uint64_t tmp, offset;
846 QCowCreateState s1, *s = &s1;
847 QCowExtension ext_bf = {0, 0};
848 int ret;
850 memset(s, 0, sizeof(*s));
852 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
853 if (fd < 0)
854 return -1;
855 memset(&header, 0, sizeof(header));
856 header.magic = cpu_to_be32(QCOW_MAGIC);
857 header.version = cpu_to_be32(QCOW_VERSION);
858 header.size = cpu_to_be64(total_size * 512);
859 header_size = sizeof(header);
860 backing_filename_len = 0;
861 if (backing_file) {
862 if (backing_format) {
863 ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
864 backing_format_len = strlen(backing_format);
865 ext_bf.len = backing_format_len;
866 rounded_ext_bf_len = (sizeof(ext_bf) + ext_bf.len + 7) & ~7;
867 header_size += rounded_ext_bf_len;
869 header.backing_file_offset = cpu_to_be64(header_size);
870 backing_filename_len = strlen(backing_file);
871 header.backing_file_size = cpu_to_be32(backing_filename_len);
872 header_size += backing_filename_len;
875 /* Cluster size */
876 s->cluster_bits = get_bits_from_size(cluster_size);
877 if (s->cluster_bits < MIN_CLUSTER_BITS ||
878 s->cluster_bits > MAX_CLUSTER_BITS)
880 fprintf(stderr, "Cluster size must be a power of two between "
881 "%d and %dk\n",
882 1 << MIN_CLUSTER_BITS,
883 1 << (MAX_CLUSTER_BITS - 10));
884 return -EINVAL;
886 s->cluster_size = 1 << s->cluster_bits;
888 header.cluster_bits = cpu_to_be32(s->cluster_bits);
889 header_size = (header_size + 7) & ~7;
890 if (flags & BLOCK_FLAG_ENCRYPT) {
891 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
892 } else {
893 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
895 l2_bits = s->cluster_bits - 3;
896 shift = s->cluster_bits + l2_bits;
897 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
898 offset = align_offset(header_size, s->cluster_size);
899 s->l1_table_offset = offset;
900 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
901 header.l1_size = cpu_to_be32(l1_size);
902 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
904 s->refcount_table = qemu_mallocz(s->cluster_size);
906 s->refcount_table_offset = offset;
907 header.refcount_table_offset = cpu_to_be64(offset);
908 header.refcount_table_clusters = cpu_to_be32(1);
909 offset += s->cluster_size;
910 s->refcount_block_offset = offset;
912 /* count how many refcount blocks needed */
913 tmp = offset >> s->cluster_bits;
914 ref_clusters = (tmp >> (s->cluster_bits - REFCOUNT_SHIFT)) + 1;
915 for (i=0; i < ref_clusters; i++) {
916 s->refcount_table[i] = cpu_to_be64(offset);
917 offset += s->cluster_size;
920 s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
922 /* update refcounts */
923 qcow2_create_refcount_update(s, 0, header_size);
924 qcow2_create_refcount_update(s, s->l1_table_offset,
925 l1_size * sizeof(uint64_t));
926 qcow2_create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
927 qcow2_create_refcount_update(s, s->refcount_block_offset,
928 ref_clusters * s->cluster_size);
930 /* write all the data */
931 ret = qemu_write_full(fd, &header, sizeof(header));
932 if (ret != sizeof(header)) {
933 ret = -1;
934 goto exit;
936 if (backing_file) {
937 if (backing_format_len) {
938 char zero[16];
939 int padding = rounded_ext_bf_len - (ext_bf.len + sizeof(ext_bf));
941 memset(zero, 0, sizeof(zero));
942 cpu_to_be32s(&ext_bf.magic);
943 cpu_to_be32s(&ext_bf.len);
944 ret = qemu_write_full(fd, &ext_bf, sizeof(ext_bf));
945 if (ret != sizeof(ext_bf)) {
946 ret = -1;
947 goto exit;
949 ret = qemu_write_full(fd, backing_format, backing_format_len);
950 if (ret != backing_format_len) {
951 ret = -1;
952 goto exit;
954 if (padding > 0) {
955 ret = qemu_write_full(fd, zero, padding);
956 if (ret != padding) {
957 ret = -1;
958 goto exit;
962 ret = qemu_write_full(fd, backing_file, backing_filename_len);
963 if (ret != backing_filename_len) {
964 ret = -1;
965 goto exit;
968 lseek(fd, s->l1_table_offset, SEEK_SET);
969 tmp = 0;
970 for(i = 0;i < l1_size; i++) {
971 ret = qemu_write_full(fd, &tmp, sizeof(tmp));
972 if (ret != sizeof(tmp)) {
973 ret = -1;
974 goto exit;
977 lseek(fd, s->refcount_table_offset, SEEK_SET);
978 ret = qemu_write_full(fd, s->refcount_table, s->cluster_size);
979 if (ret != s->cluster_size) {
980 ret = -1;
981 goto exit;
984 lseek(fd, s->refcount_block_offset, SEEK_SET);
985 ret = qemu_write_full(fd, s->refcount_block,
986 ref_clusters * s->cluster_size);
987 if (ret != ref_clusters * s->cluster_size) {
988 ret = -1;
989 goto exit;
992 ret = 0;
993 exit:
994 qemu_free(s->refcount_table);
995 qemu_free(s->refcount_block);
996 close(fd);
998 /* Preallocate metadata */
999 if (ret == 0 && prealloc) {
1000 BlockDriverState *bs;
1001 bs = bdrv_new("");
1002 bdrv_open(bs, filename, BDRV_O_CACHE_WB | BDRV_O_RDWR);
1003 preallocate(bs);
1004 bdrv_close(bs);
1007 return ret;
1010 static int qcow_create(const char *filename, QEMUOptionParameter *options)
1012 const char *backing_file = NULL;
1013 const char *backing_fmt = NULL;
1014 uint64_t sectors = 0;
1015 int flags = 0;
1016 size_t cluster_size = 65536;
1017 int prealloc = 0;
1019 /* Read out options */
1020 while (options && options->name) {
1021 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1022 sectors = options->value.n / 512;
1023 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1024 backing_file = options->value.s;
1025 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
1026 backing_fmt = options->value.s;
1027 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
1028 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
1029 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
1030 if (options->value.n) {
1031 cluster_size = options->value.n;
1033 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
1034 if (!options->value.s || !strcmp(options->value.s, "off")) {
1035 prealloc = 0;
1036 } else if (!strcmp(options->value.s, "metadata")) {
1037 prealloc = 1;
1038 } else {
1039 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
1040 options->value.s);
1041 return -EINVAL;
1044 options++;
1047 if (backing_file && prealloc) {
1048 fprintf(stderr, "Backing file and preallocation cannot be used at "
1049 "the same time\n");
1050 return -EINVAL;
1053 return qcow_create2(filename, sectors, backing_file, backing_fmt, flags,
1054 cluster_size, prealloc);
1057 static int qcow_make_empty(BlockDriverState *bs)
1059 #if 0
1060 /* XXX: not correct */
1061 BDRVQcowState *s = bs->opaque;
1062 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1063 int ret;
1065 memset(s->l1_table, 0, l1_length);
1066 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
1067 return -1;
1068 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1069 if (ret < 0)
1070 return ret;
1072 l2_cache_reset(bs);
1073 #endif
1074 return 0;
1077 /* XXX: put compressed sectors first, then all the cluster aligned
1078 tables to avoid losing bytes in alignment */
1079 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1080 const uint8_t *buf, int nb_sectors)
1082 BDRVQcowState *s = bs->opaque;
1083 z_stream strm;
1084 int ret, out_len;
1085 uint8_t *out_buf;
1086 uint64_t cluster_offset;
1088 if (nb_sectors == 0) {
1089 /* align end of file to a sector boundary to ease reading with
1090 sector based I/Os */
1091 cluster_offset = bdrv_getlength(s->hd);
1092 cluster_offset = (cluster_offset + 511) & ~511;
1093 bdrv_truncate(s->hd, cluster_offset);
1094 return 0;
1097 if (nb_sectors != s->cluster_sectors)
1098 return -EINVAL;
1100 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1102 /* best compression, small window, no zlib header */
1103 memset(&strm, 0, sizeof(strm));
1104 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1105 Z_DEFLATED, -12,
1106 9, Z_DEFAULT_STRATEGY);
1107 if (ret != 0) {
1108 qemu_free(out_buf);
1109 return -1;
1112 strm.avail_in = s->cluster_size;
1113 strm.next_in = (uint8_t *)buf;
1114 strm.avail_out = s->cluster_size;
1115 strm.next_out = out_buf;
1117 ret = deflate(&strm, Z_FINISH);
1118 if (ret != Z_STREAM_END && ret != Z_OK) {
1119 qemu_free(out_buf);
1120 deflateEnd(&strm);
1121 return -1;
1123 out_len = strm.next_out - out_buf;
1125 deflateEnd(&strm);
1127 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1128 /* could not compress: write normal cluster */
1129 bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1130 } else {
1131 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1132 sector_num << 9, out_len);
1133 if (!cluster_offset)
1134 return -1;
1135 cluster_offset &= s->cluster_offset_mask;
1136 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1137 qemu_free(out_buf);
1138 return -1;
1142 qemu_free(out_buf);
1143 return 0;
1146 static void qcow_flush(BlockDriverState *bs)
1148 BDRVQcowState *s = bs->opaque;
1149 bdrv_flush(s->hd);
1152 static BlockDriverAIOCB *qcow_aio_flush(BlockDriverState *bs,
1153 BlockDriverCompletionFunc *cb, void *opaque)
1155 BDRVQcowState *s = bs->opaque;
1157 return bdrv_aio_flush(s->hd, cb, opaque);
1160 static int64_t qcow_vm_state_offset(BDRVQcowState *s)
1162 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1165 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1167 BDRVQcowState *s = bs->opaque;
1168 bdi->cluster_size = s->cluster_size;
1169 bdi->vm_state_offset = qcow_vm_state_offset(s);
1170 return 0;
1174 static int qcow_check(BlockDriverState *bs)
1176 return qcow2_check_refcounts(bs);
1179 #if 0
1180 static void dump_refcounts(BlockDriverState *bs)
1182 BDRVQcowState *s = bs->opaque;
1183 int64_t nb_clusters, k, k1, size;
1184 int refcount;
1186 size = bdrv_getlength(s->hd);
1187 nb_clusters = size_to_clusters(s, size);
1188 for(k = 0; k < nb_clusters;) {
1189 k1 = k;
1190 refcount = get_refcount(bs, k);
1191 k++;
1192 while (k < nb_clusters && get_refcount(bs, k) == refcount)
1193 k++;
1194 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
1197 #endif
1199 static int qcow_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1200 int64_t pos, int size)
1202 BDRVQcowState *s = bs->opaque;
1203 int growable = bs->growable;
1204 int ret;
1206 bs->growable = 1;
1207 ret = bdrv_pwrite(bs, qcow_vm_state_offset(s) + pos, buf, size);
1208 bs->growable = growable;
1210 return ret;
1213 static int qcow_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1214 int64_t pos, int size)
1216 BDRVQcowState *s = bs->opaque;
1217 int growable = bs->growable;
1218 int ret;
1220 bs->growable = 1;
1221 ret = bdrv_pread(bs, qcow_vm_state_offset(s) + pos, buf, size);
1222 bs->growable = growable;
1224 return ret;
1227 static QEMUOptionParameter qcow_create_options[] = {
1229 .name = BLOCK_OPT_SIZE,
1230 .type = OPT_SIZE,
1231 .help = "Virtual disk size"
1234 .name = BLOCK_OPT_BACKING_FILE,
1235 .type = OPT_STRING,
1236 .help = "File name of a base image"
1239 .name = BLOCK_OPT_BACKING_FMT,
1240 .type = OPT_STRING,
1241 .help = "Image format of the base image"
1244 .name = BLOCK_OPT_ENCRYPT,
1245 .type = OPT_FLAG,
1246 .help = "Encrypt the image"
1249 .name = BLOCK_OPT_CLUSTER_SIZE,
1250 .type = OPT_SIZE,
1251 .help = "qcow2 cluster size"
1254 .name = BLOCK_OPT_PREALLOC,
1255 .type = OPT_STRING,
1256 .help = "Preallocation mode (allowed values: off, metadata)"
1258 { NULL }
1261 static BlockDriver bdrv_qcow2 = {
1262 .format_name = "qcow2",
1263 .instance_size = sizeof(BDRVQcowState),
1264 .bdrv_probe = qcow_probe,
1265 .bdrv_open = qcow_open,
1266 .bdrv_close = qcow_close,
1267 .bdrv_create = qcow_create,
1268 .bdrv_flush = qcow_flush,
1269 .bdrv_is_allocated = qcow_is_allocated,
1270 .bdrv_set_key = qcow_set_key,
1271 .bdrv_make_empty = qcow_make_empty,
1273 .bdrv_aio_readv = qcow_aio_readv,
1274 .bdrv_aio_writev = qcow_aio_writev,
1275 .bdrv_aio_flush = qcow_aio_flush,
1276 .bdrv_write_compressed = qcow_write_compressed,
1278 .bdrv_snapshot_create = qcow2_snapshot_create,
1279 .bdrv_snapshot_goto = qcow2_snapshot_goto,
1280 .bdrv_snapshot_delete = qcow2_snapshot_delete,
1281 .bdrv_snapshot_list = qcow2_snapshot_list,
1282 .bdrv_get_info = qcow_get_info,
1284 .bdrv_save_vmstate = qcow_save_vmstate,
1285 .bdrv_load_vmstate = qcow_load_vmstate,
1287 .bdrv_change_backing_file = qcow2_change_backing_file,
1289 .create_options = qcow_create_options,
1290 .bdrv_check = qcow_check,
1293 static void bdrv_qcow2_init(void)
1295 bdrv_register(&bdrv_qcow2);
1298 block_init(bdrv_qcow2_init);