memory: crack wide ioport accesses into smaller ones when needed
[qemu/mdroth.git] / block / qcow2.c
blobbfff6cd9630f54d1def691b3c3a6fbc06a6b9720
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"
30 #include "qemu-error.h"
31 #include "qerror.h"
34 Differences with QCOW:
36 - Support for multiple incremental snapshots.
37 - Memory management by reference counts.
38 - Clusters which have a reference count of one have the bit
39 QCOW_OFLAG_COPIED to optimize write performance.
40 - Size of compressed clusters is stored in sectors to reduce bit usage
41 in the cluster offsets.
42 - Support for storing additional data (such as the VM state) in the
43 snapshots.
44 - If a backing store is used, the cluster size is not constrained
45 (could be backported to QCOW).
46 - L2 tables have always a size of one cluster.
50 typedef struct {
51 uint32_t magic;
52 uint32_t len;
53 } QCowExtension;
54 #define QCOW2_EXT_MAGIC_END 0
55 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
57 static int qcow2_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 qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
78 uint64_t end_offset)
80 QCowExtension ext;
81 uint64_t offset;
83 #ifdef DEBUG_EXT
84 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
85 #endif
86 offset = start_offset;
87 while (offset < end_offset) {
89 #ifdef DEBUG_EXT
90 /* Sanity check */
91 if (offset > s->cluster_size)
92 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
94 printf("attemting to read extended header in offset %lu\n", offset);
95 #endif
97 if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) {
98 fprintf(stderr, "qcow2_read_extension: ERROR: "
99 "pread fail from offset %" PRIu64 "\n",
100 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 QCOW2_EXT_MAGIC_END:
111 return 0;
113 case QCOW2_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(bs->file, 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 qcow2_open(BlockDriverState *bs, int flags)
143 BDRVQcowState *s = bs->opaque;
144 int len, i, ret = 0;
145 QCowHeader header;
146 uint64_t ext_end;
147 bool writethrough;
149 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
150 if (ret < 0) {
151 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) {
168 ret = -EINVAL;
169 goto fail;
171 if (header.version != QCOW_VERSION) {
172 char version[64];
173 snprintf(version, sizeof(version), "QCOW version %d", header.version);
174 qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
175 bs->device_name, "qcow2", version);
176 ret = -ENOTSUP;
177 goto fail;
179 if (header.cluster_bits < MIN_CLUSTER_BITS ||
180 header.cluster_bits > MAX_CLUSTER_BITS) {
181 ret = -EINVAL;
182 goto fail;
184 if (header.crypt_method > QCOW_CRYPT_AES) {
185 ret = -EINVAL;
186 goto fail;
188 s->crypt_method_header = header.crypt_method;
189 if (s->crypt_method_header) {
190 bs->encrypted = 1;
192 s->cluster_bits = header.cluster_bits;
193 s->cluster_size = 1 << s->cluster_bits;
194 s->cluster_sectors = 1 << (s->cluster_bits - 9);
195 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
196 s->l2_size = 1 << s->l2_bits;
197 bs->total_sectors = header.size / 512;
198 s->csize_shift = (62 - (s->cluster_bits - 8));
199 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
200 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
201 s->refcount_table_offset = header.refcount_table_offset;
202 s->refcount_table_size =
203 header.refcount_table_clusters << (s->cluster_bits - 3);
205 s->snapshots_offset = header.snapshots_offset;
206 s->nb_snapshots = header.nb_snapshots;
208 /* read the level 1 table */
209 s->l1_size = header.l1_size;
210 s->l1_vm_state_index = size_to_l1(s, header.size);
211 /* the L1 table must contain at least enough entries to put
212 header.size bytes */
213 if (s->l1_size < s->l1_vm_state_index) {
214 ret = -EINVAL;
215 goto fail;
217 s->l1_table_offset = header.l1_table_offset;
218 if (s->l1_size > 0) {
219 s->l1_table = g_malloc0(
220 align_offset(s->l1_size * sizeof(uint64_t), 512));
221 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
222 s->l1_size * sizeof(uint64_t));
223 if (ret < 0) {
224 goto fail;
226 for(i = 0;i < s->l1_size; i++) {
227 be64_to_cpus(&s->l1_table[i]);
231 /* alloc L2 table/refcount block cache */
232 writethrough = ((flags & BDRV_O_CACHE_WB) == 0);
233 s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE, writethrough);
234 s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE,
235 writethrough);
237 s->cluster_cache = g_malloc(s->cluster_size);
238 /* one more sector for decompressed data alignment */
239 s->cluster_data = g_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
240 + 512);
241 s->cluster_cache_offset = -1;
243 ret = qcow2_refcount_init(bs);
244 if (ret != 0) {
245 goto fail;
248 QLIST_INIT(&s->cluster_allocs);
250 /* read qcow2 extensions */
251 if (header.backing_file_offset) {
252 ext_end = header.backing_file_offset;
253 } else {
254 ext_end = s->cluster_size;
256 if (qcow2_read_extensions(bs, sizeof(header), ext_end)) {
257 ret = -EINVAL;
258 goto fail;
261 /* read the backing file name */
262 if (header.backing_file_offset != 0) {
263 len = header.backing_file_size;
264 if (len > 1023) {
265 len = 1023;
267 ret = bdrv_pread(bs->file, header.backing_file_offset,
268 bs->backing_file, len);
269 if (ret < 0) {
270 goto fail;
272 bs->backing_file[len] = '\0';
274 if (qcow2_read_snapshots(bs) < 0) {
275 ret = -EINVAL;
276 goto fail;
279 /* Initialise locks */
280 qemu_co_mutex_init(&s->lock);
282 #ifdef DEBUG_ALLOC
283 qcow2_check_refcounts(bs);
284 #endif
285 return ret;
287 fail:
288 qcow2_free_snapshots(bs);
289 qcow2_refcount_close(bs);
290 g_free(s->l1_table);
291 if (s->l2_table_cache) {
292 qcow2_cache_destroy(bs, s->l2_table_cache);
294 g_free(s->cluster_cache);
295 g_free(s->cluster_data);
296 return ret;
299 static int qcow2_set_key(BlockDriverState *bs, const char *key)
301 BDRVQcowState *s = bs->opaque;
302 uint8_t keybuf[16];
303 int len, i;
305 memset(keybuf, 0, 16);
306 len = strlen(key);
307 if (len > 16)
308 len = 16;
309 /* XXX: we could compress the chars to 7 bits to increase
310 entropy */
311 for(i = 0;i < len;i++) {
312 keybuf[i] = key[i];
314 s->crypt_method = s->crypt_method_header;
316 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
317 return -1;
318 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
319 return -1;
320 #if 0
321 /* test */
323 uint8_t in[16];
324 uint8_t out[16];
325 uint8_t tmp[16];
326 for(i=0;i<16;i++)
327 in[i] = i;
328 AES_encrypt(in, tmp, &s->aes_encrypt_key);
329 AES_decrypt(tmp, out, &s->aes_decrypt_key);
330 for(i = 0; i < 16; i++)
331 printf(" %02x", tmp[i]);
332 printf("\n");
333 for(i = 0; i < 16; i++)
334 printf(" %02x", out[i]);
335 printf("\n");
337 #endif
338 return 0;
341 static int qcow2_is_allocated(BlockDriverState *bs, int64_t sector_num,
342 int nb_sectors, int *pnum)
344 uint64_t cluster_offset;
345 int ret;
347 *pnum = nb_sectors;
348 /* FIXME We can get errors here, but the bdrv_is_allocated interface can't
349 * pass them on today */
350 ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
351 if (ret < 0) {
352 *pnum = 0;
355 return (cluster_offset != 0);
358 /* handle reading after the end of the backing file */
359 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
360 int64_t sector_num, int nb_sectors)
362 int n1;
363 if ((sector_num + nb_sectors) <= bs->total_sectors)
364 return nb_sectors;
365 if (sector_num >= bs->total_sectors)
366 n1 = 0;
367 else
368 n1 = bs->total_sectors - sector_num;
370 qemu_iovec_memset_skip(qiov, 0, 512 * (nb_sectors - n1), 512 * n1);
372 return n1;
375 typedef struct QCowAIOCB {
376 BlockDriverAIOCB common;
377 int64_t sector_num;
378 QEMUIOVector *qiov;
379 int remaining_sectors;
380 int cur_nr_sectors; /* number of sectors in current iteration */
381 uint64_t bytes_done;
382 uint64_t cluster_offset;
383 uint8_t *cluster_data;
384 bool is_write;
385 QEMUIOVector hd_qiov;
386 QEMUBH *bh;
387 QCowL2Meta l2meta;
388 QLIST_ENTRY(QCowAIOCB) next_depend;
389 } QCowAIOCB;
391 static void qcow2_aio_cancel(BlockDriverAIOCB *blockacb)
393 QCowAIOCB *acb = container_of(blockacb, QCowAIOCB, common);
394 qemu_aio_release(acb);
397 static AIOPool qcow2_aio_pool = {
398 .aiocb_size = sizeof(QCowAIOCB),
399 .cancel = qcow2_aio_cancel,
403 * Returns 0 when the request is completed successfully, 1 when there is still
404 * a part left to do and -errno in error cases.
406 static int qcow2_aio_read_cb(QCowAIOCB *acb)
408 BlockDriverState *bs = acb->common.bs;
409 BDRVQcowState *s = bs->opaque;
410 int index_in_cluster, n1;
411 int ret;
413 /* post process the read buffer */
414 if (!acb->cluster_offset) {
415 /* nothing to do */
416 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
417 /* nothing to do */
418 } else {
419 if (s->crypt_method) {
420 qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data,
421 acb->cluster_data, acb->cur_nr_sectors, 0, &s->aes_decrypt_key);
422 qemu_iovec_reset(&acb->hd_qiov);
423 qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done,
424 acb->cur_nr_sectors * 512);
425 qemu_iovec_from_buffer(&acb->hd_qiov, acb->cluster_data,
426 512 * acb->cur_nr_sectors);
430 acb->remaining_sectors -= acb->cur_nr_sectors;
431 acb->sector_num += acb->cur_nr_sectors;
432 acb->bytes_done += acb->cur_nr_sectors * 512;
434 if (acb->remaining_sectors == 0) {
435 /* request completed */
436 return 0;
439 /* prepare next AIO request */
440 acb->cur_nr_sectors = acb->remaining_sectors;
441 if (s->crypt_method) {
442 acb->cur_nr_sectors = MIN(acb->cur_nr_sectors,
443 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
446 ret = qcow2_get_cluster_offset(bs, acb->sector_num << 9,
447 &acb->cur_nr_sectors, &acb->cluster_offset);
448 if (ret < 0) {
449 return ret;
452 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
454 qemu_iovec_reset(&acb->hd_qiov);
455 qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done,
456 acb->cur_nr_sectors * 512);
458 if (!acb->cluster_offset) {
460 if (bs->backing_hd) {
461 /* read from the base image */
462 n1 = qcow2_backing_read1(bs->backing_hd, &acb->hd_qiov,
463 acb->sector_num, acb->cur_nr_sectors);
464 if (n1 > 0) {
465 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
466 qemu_co_mutex_unlock(&s->lock);
467 ret = bdrv_co_readv(bs->backing_hd, acb->sector_num,
468 n1, &acb->hd_qiov);
469 qemu_co_mutex_lock(&s->lock);
470 if (ret < 0) {
471 return ret;
474 return 1;
475 } else {
476 /* Note: in this case, no need to wait */
477 qemu_iovec_memset(&acb->hd_qiov, 0, 512 * acb->cur_nr_sectors);
478 return 1;
480 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
481 /* add AIO support for compressed blocks ? */
482 ret = qcow2_decompress_cluster(bs, acb->cluster_offset);
483 if (ret < 0) {
484 return ret;
487 qemu_iovec_from_buffer(&acb->hd_qiov,
488 s->cluster_cache + index_in_cluster * 512,
489 512 * acb->cur_nr_sectors);
491 return 1;
492 } else {
493 if ((acb->cluster_offset & 511) != 0) {
494 return -EIO;
497 if (s->crypt_method) {
499 * For encrypted images, read everything into a temporary
500 * contiguous buffer on which the AES functions can work.
502 if (!acb->cluster_data) {
503 acb->cluster_data =
504 g_malloc0(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
507 assert(acb->cur_nr_sectors <=
508 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
509 qemu_iovec_reset(&acb->hd_qiov);
510 qemu_iovec_add(&acb->hd_qiov, acb->cluster_data,
511 512 * acb->cur_nr_sectors);
514 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
515 qemu_co_mutex_unlock(&s->lock);
516 ret = bdrv_co_readv(bs->file,
517 (acb->cluster_offset >> 9) + index_in_cluster,
518 acb->cur_nr_sectors, &acb->hd_qiov);
519 qemu_co_mutex_lock(&s->lock);
520 if (ret < 0) {
521 return ret;
525 return 1;
528 static QCowAIOCB *qcow2_aio_setup(BlockDriverState *bs, int64_t sector_num,
529 QEMUIOVector *qiov, int nb_sectors,
530 BlockDriverCompletionFunc *cb,
531 void *opaque, int is_write)
533 QCowAIOCB *acb;
535 acb = qemu_aio_get(&qcow2_aio_pool, bs, cb, opaque);
536 if (!acb)
537 return NULL;
538 acb->sector_num = sector_num;
539 acb->qiov = qiov;
540 acb->is_write = is_write;
542 qemu_iovec_init(&acb->hd_qiov, qiov->niov);
544 acb->bytes_done = 0;
545 acb->remaining_sectors = nb_sectors;
546 acb->cur_nr_sectors = 0;
547 acb->cluster_offset = 0;
548 acb->l2meta.nb_clusters = 0;
549 qemu_co_queue_init(&acb->l2meta.dependent_requests);
550 return acb;
553 static int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
554 int nb_sectors, QEMUIOVector *qiov)
556 BDRVQcowState *s = bs->opaque;
557 QCowAIOCB *acb;
558 int ret;
560 acb = qcow2_aio_setup(bs, sector_num, qiov, nb_sectors, NULL, NULL, 0);
562 qemu_co_mutex_lock(&s->lock);
563 do {
564 ret = qcow2_aio_read_cb(acb);
565 } while (ret > 0);
566 qemu_co_mutex_unlock(&s->lock);
568 qemu_iovec_destroy(&acb->hd_qiov);
569 qemu_aio_release(acb);
571 return ret;
574 static void run_dependent_requests(BDRVQcowState *s, QCowL2Meta *m)
576 /* Take the request off the list of running requests */
577 if (m->nb_clusters != 0) {
578 QLIST_REMOVE(m, next_in_flight);
581 /* Restart all dependent requests */
582 if (!qemu_co_queue_empty(&m->dependent_requests)) {
583 qemu_co_mutex_unlock(&s->lock);
584 while(qemu_co_queue_next(&m->dependent_requests));
585 qemu_co_mutex_lock(&s->lock);
590 * Returns 0 when the request is completed successfully, 1 when there is still
591 * a part left to do and -errno in error cases.
593 static int qcow2_aio_write_cb(QCowAIOCB *acb)
595 BlockDriverState *bs = acb->common.bs;
596 BDRVQcowState *s = bs->opaque;
597 int index_in_cluster;
598 int n_end;
599 int ret;
601 ret = qcow2_alloc_cluster_link_l2(bs, &acb->l2meta);
603 run_dependent_requests(s, &acb->l2meta);
605 if (ret < 0) {
606 return ret;
609 acb->remaining_sectors -= acb->cur_nr_sectors;
610 acb->sector_num += acb->cur_nr_sectors;
611 acb->bytes_done += acb->cur_nr_sectors * 512;
613 if (acb->remaining_sectors == 0) {
614 /* request completed */
615 return 0;
618 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
619 n_end = index_in_cluster + acb->remaining_sectors;
620 if (s->crypt_method &&
621 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
622 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
624 ret = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9,
625 index_in_cluster, n_end, &acb->cur_nr_sectors, &acb->l2meta);
626 if (ret < 0) {
627 return ret;
630 acb->cluster_offset = acb->l2meta.cluster_offset;
631 assert((acb->cluster_offset & 511) == 0);
633 qemu_iovec_reset(&acb->hd_qiov);
634 qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done,
635 acb->cur_nr_sectors * 512);
637 if (s->crypt_method) {
638 if (!acb->cluster_data) {
639 acb->cluster_data = g_malloc0(QCOW_MAX_CRYPT_CLUSTERS *
640 s->cluster_size);
643 assert(acb->hd_qiov.size <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
644 qemu_iovec_to_buffer(&acb->hd_qiov, acb->cluster_data);
646 qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data,
647 acb->cluster_data, acb->cur_nr_sectors, 1, &s->aes_encrypt_key);
649 qemu_iovec_reset(&acb->hd_qiov);
650 qemu_iovec_add(&acb->hd_qiov, acb->cluster_data,
651 acb->cur_nr_sectors * 512);
654 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
655 qemu_co_mutex_unlock(&s->lock);
656 ret = bdrv_co_writev(bs->file,
657 (acb->cluster_offset >> 9) + index_in_cluster,
658 acb->cur_nr_sectors, &acb->hd_qiov);
659 qemu_co_mutex_lock(&s->lock);
660 if (ret < 0) {
661 return ret;
664 return 1;
667 static int qcow2_co_writev(BlockDriverState *bs,
668 int64_t sector_num,
669 int nb_sectors,
670 QEMUIOVector *qiov)
672 BDRVQcowState *s = bs->opaque;
673 QCowAIOCB *acb;
674 int ret;
676 acb = qcow2_aio_setup(bs, sector_num, qiov, nb_sectors, NULL, NULL, 1);
677 s->cluster_cache_offset = -1; /* disable compressed cache */
679 qemu_co_mutex_lock(&s->lock);
680 do {
681 ret = qcow2_aio_write_cb(acb);
682 } while (ret > 0);
683 qemu_co_mutex_unlock(&s->lock);
685 qemu_iovec_destroy(&acb->hd_qiov);
686 qemu_aio_release(acb);
688 return ret;
691 static void qcow2_close(BlockDriverState *bs)
693 BDRVQcowState *s = bs->opaque;
694 g_free(s->l1_table);
696 qcow2_cache_flush(bs, s->l2_table_cache);
697 qcow2_cache_flush(bs, s->refcount_block_cache);
699 qcow2_cache_destroy(bs, s->l2_table_cache);
700 qcow2_cache_destroy(bs, s->refcount_block_cache);
702 g_free(s->cluster_cache);
703 g_free(s->cluster_data);
704 qcow2_refcount_close(bs);
708 * Updates the variable length parts of the qcow2 header, i.e. the backing file
709 * name and all extensions. qcow2 was not designed to allow such changes, so if
710 * we run out of space (we can only use the first cluster) this function may
711 * fail.
713 * Returns 0 on success, -errno in error cases.
715 static int qcow2_update_ext_header(BlockDriverState *bs,
716 const char *backing_file, const char *backing_fmt)
718 size_t backing_file_len = 0;
719 size_t backing_fmt_len = 0;
720 BDRVQcowState *s = bs->opaque;
721 QCowExtension ext_backing_fmt = {0, 0};
722 int ret;
724 /* Backing file format doesn't make sense without a backing file */
725 if (backing_fmt && !backing_file) {
726 return -EINVAL;
729 /* Prepare the backing file format extension if needed */
730 if (backing_fmt) {
731 ext_backing_fmt.len = cpu_to_be32(strlen(backing_fmt));
732 ext_backing_fmt.magic = cpu_to_be32(QCOW2_EXT_MAGIC_BACKING_FORMAT);
733 backing_fmt_len = ((sizeof(ext_backing_fmt)
734 + strlen(backing_fmt) + 7) & ~7);
737 /* Check if we can fit the new header into the first cluster */
738 if (backing_file) {
739 backing_file_len = strlen(backing_file);
742 size_t header_size = sizeof(QCowHeader) + backing_file_len
743 + backing_fmt_len;
745 if (header_size > s->cluster_size) {
746 return -ENOSPC;
749 /* Rewrite backing file name and qcow2 extensions */
750 size_t ext_size = header_size - sizeof(QCowHeader);
751 uint8_t buf[ext_size];
752 size_t offset = 0;
753 size_t backing_file_offset = 0;
755 if (backing_file) {
756 if (backing_fmt) {
757 int padding = backing_fmt_len -
758 (sizeof(ext_backing_fmt) + strlen(backing_fmt));
760 memcpy(buf + offset, &ext_backing_fmt, sizeof(ext_backing_fmt));
761 offset += sizeof(ext_backing_fmt);
763 memcpy(buf + offset, backing_fmt, strlen(backing_fmt));
764 offset += strlen(backing_fmt);
766 memset(buf + offset, 0, padding);
767 offset += padding;
770 memcpy(buf + offset, backing_file, backing_file_len);
771 backing_file_offset = sizeof(QCowHeader) + offset;
774 ret = bdrv_pwrite_sync(bs->file, sizeof(QCowHeader), buf, ext_size);
775 if (ret < 0) {
776 goto fail;
779 /* Update header fields */
780 uint64_t be_backing_file_offset = cpu_to_be64(backing_file_offset);
781 uint32_t be_backing_file_size = cpu_to_be32(backing_file_len);
783 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_offset),
784 &be_backing_file_offset, sizeof(uint64_t));
785 if (ret < 0) {
786 goto fail;
789 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_size),
790 &be_backing_file_size, sizeof(uint32_t));
791 if (ret < 0) {
792 goto fail;
795 ret = 0;
796 fail:
797 return ret;
800 static int qcow2_change_backing_file(BlockDriverState *bs,
801 const char *backing_file, const char *backing_fmt)
803 return qcow2_update_ext_header(bs, backing_file, backing_fmt);
806 static int preallocate(BlockDriverState *bs)
808 uint64_t nb_sectors;
809 uint64_t offset;
810 int num;
811 int ret;
812 QCowL2Meta meta;
814 nb_sectors = bdrv_getlength(bs) >> 9;
815 offset = 0;
816 qemu_co_queue_init(&meta.dependent_requests);
817 meta.cluster_offset = 0;
819 while (nb_sectors) {
820 num = MIN(nb_sectors, INT_MAX >> 9);
821 ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta);
822 if (ret < 0) {
823 return ret;
826 ret = qcow2_alloc_cluster_link_l2(bs, &meta);
827 if (ret < 0) {
828 qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters);
829 return ret;
832 /* There are no dependent requests, but we need to remove our request
833 * from the list of in-flight requests */
834 run_dependent_requests(bs->opaque, &meta);
836 /* TODO Preallocate data if requested */
838 nb_sectors -= num;
839 offset += num << 9;
843 * It is expected that the image file is large enough to actually contain
844 * all of the allocated clusters (otherwise we get failing reads after
845 * EOF). Extend the image to the last allocated sector.
847 if (meta.cluster_offset != 0) {
848 uint8_t buf[512];
849 memset(buf, 0, 512);
850 ret = bdrv_write(bs->file, (meta.cluster_offset >> 9) + num - 1, buf, 1);
851 if (ret < 0) {
852 return ret;
856 return 0;
859 static int qcow2_create2(const char *filename, int64_t total_size,
860 const char *backing_file, const char *backing_format,
861 int flags, size_t cluster_size, int prealloc,
862 QEMUOptionParameter *options)
864 /* Calulate cluster_bits */
865 int cluster_bits;
866 cluster_bits = ffs(cluster_size) - 1;
867 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
868 (1 << cluster_bits) != cluster_size)
870 error_report(
871 "Cluster size must be a power of two between %d and %dk",
872 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
873 return -EINVAL;
877 * Open the image file and write a minimal qcow2 header.
879 * We keep things simple and start with a zero-sized image. We also
880 * do without refcount blocks or a L1 table for now. We'll fix the
881 * inconsistency later.
883 * We do need a refcount table because growing the refcount table means
884 * allocating two new refcount blocks - the seconds of which would be at
885 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
886 * size for any qcow2 image.
888 BlockDriverState* bs;
889 QCowHeader header;
890 uint8_t* refcount_table;
891 int ret;
893 ret = bdrv_create_file(filename, options);
894 if (ret < 0) {
895 return ret;
898 ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR);
899 if (ret < 0) {
900 return ret;
903 /* Write the header */
904 memset(&header, 0, sizeof(header));
905 header.magic = cpu_to_be32(QCOW_MAGIC);
906 header.version = cpu_to_be32(QCOW_VERSION);
907 header.cluster_bits = cpu_to_be32(cluster_bits);
908 header.size = cpu_to_be64(0);
909 header.l1_table_offset = cpu_to_be64(0);
910 header.l1_size = cpu_to_be32(0);
911 header.refcount_table_offset = cpu_to_be64(cluster_size);
912 header.refcount_table_clusters = cpu_to_be32(1);
914 if (flags & BLOCK_FLAG_ENCRYPT) {
915 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
916 } else {
917 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
920 ret = bdrv_pwrite(bs, 0, &header, sizeof(header));
921 if (ret < 0) {
922 goto out;
925 /* Write an empty refcount table */
926 refcount_table = g_malloc0(cluster_size);
927 ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size);
928 g_free(refcount_table);
930 if (ret < 0) {
931 goto out;
934 bdrv_close(bs);
937 * And now open the image and make it consistent first (i.e. increase the
938 * refcount of the cluster that is occupied by the header and the refcount
939 * table)
941 BlockDriver* drv = bdrv_find_format("qcow2");
942 assert(drv != NULL);
943 ret = bdrv_open(bs, filename,
944 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv);
945 if (ret < 0) {
946 goto out;
949 ret = qcow2_alloc_clusters(bs, 2 * cluster_size);
950 if (ret < 0) {
951 goto out;
953 } else if (ret != 0) {
954 error_report("Huh, first cluster in empty image is already in use?");
955 abort();
958 /* Okay, now that we have a valid image, let's give it the right size */
959 ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE);
960 if (ret < 0) {
961 goto out;
964 /* Want a backing file? There you go.*/
965 if (backing_file) {
966 ret = bdrv_change_backing_file(bs, backing_file, backing_format);
967 if (ret < 0) {
968 goto out;
972 /* And if we're supposed to preallocate metadata, do that now */
973 if (prealloc) {
974 ret = preallocate(bs);
975 if (ret < 0) {
976 goto out;
980 ret = 0;
981 out:
982 bdrv_delete(bs);
983 return ret;
986 static int qcow2_create(const char *filename, QEMUOptionParameter *options)
988 const char *backing_file = NULL;
989 const char *backing_fmt = NULL;
990 uint64_t sectors = 0;
991 int flags = 0;
992 size_t cluster_size = DEFAULT_CLUSTER_SIZE;
993 int prealloc = 0;
995 /* Read out options */
996 while (options && options->name) {
997 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
998 sectors = options->value.n / 512;
999 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1000 backing_file = options->value.s;
1001 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
1002 backing_fmt = options->value.s;
1003 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
1004 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
1005 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
1006 if (options->value.n) {
1007 cluster_size = options->value.n;
1009 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
1010 if (!options->value.s || !strcmp(options->value.s, "off")) {
1011 prealloc = 0;
1012 } else if (!strcmp(options->value.s, "metadata")) {
1013 prealloc = 1;
1014 } else {
1015 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
1016 options->value.s);
1017 return -EINVAL;
1020 options++;
1023 if (backing_file && prealloc) {
1024 fprintf(stderr, "Backing file and preallocation cannot be used at "
1025 "the same time\n");
1026 return -EINVAL;
1029 return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags,
1030 cluster_size, prealloc, options);
1033 static int qcow2_make_empty(BlockDriverState *bs)
1035 #if 0
1036 /* XXX: not correct */
1037 BDRVQcowState *s = bs->opaque;
1038 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1039 int ret;
1041 memset(s->l1_table, 0, l1_length);
1042 if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
1043 return -1;
1044 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
1045 if (ret < 0)
1046 return ret;
1048 l2_cache_reset(bs);
1049 #endif
1050 return 0;
1053 static int qcow2_discard(BlockDriverState *bs, int64_t sector_num,
1054 int nb_sectors)
1056 return qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1057 nb_sectors);
1060 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
1062 BDRVQcowState *s = bs->opaque;
1063 int ret, new_l1_size;
1065 if (offset & 511) {
1066 return -EINVAL;
1069 /* cannot proceed if image has snapshots */
1070 if (s->nb_snapshots) {
1071 return -ENOTSUP;
1074 /* shrinking is currently not supported */
1075 if (offset < bs->total_sectors * 512) {
1076 return -ENOTSUP;
1079 new_l1_size = size_to_l1(s, offset);
1080 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
1081 if (ret < 0) {
1082 return ret;
1085 /* write updated header.size */
1086 offset = cpu_to_be64(offset);
1087 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
1088 &offset, sizeof(uint64_t));
1089 if (ret < 0) {
1090 return ret;
1093 s->l1_vm_state_index = new_l1_size;
1094 return 0;
1097 /* XXX: put compressed sectors first, then all the cluster aligned
1098 tables to avoid losing bytes in alignment */
1099 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
1100 const uint8_t *buf, int nb_sectors)
1102 BDRVQcowState *s = bs->opaque;
1103 z_stream strm;
1104 int ret, out_len;
1105 uint8_t *out_buf;
1106 uint64_t cluster_offset;
1108 if (nb_sectors == 0) {
1109 /* align end of file to a sector boundary to ease reading with
1110 sector based I/Os */
1111 cluster_offset = bdrv_getlength(bs->file);
1112 cluster_offset = (cluster_offset + 511) & ~511;
1113 bdrv_truncate(bs->file, cluster_offset);
1114 return 0;
1117 if (nb_sectors != s->cluster_sectors)
1118 return -EINVAL;
1120 out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1122 /* best compression, small window, no zlib header */
1123 memset(&strm, 0, sizeof(strm));
1124 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1125 Z_DEFLATED, -12,
1126 9, Z_DEFAULT_STRATEGY);
1127 if (ret != 0) {
1128 g_free(out_buf);
1129 return -1;
1132 strm.avail_in = s->cluster_size;
1133 strm.next_in = (uint8_t *)buf;
1134 strm.avail_out = s->cluster_size;
1135 strm.next_out = out_buf;
1137 ret = deflate(&strm, Z_FINISH);
1138 if (ret != Z_STREAM_END && ret != Z_OK) {
1139 g_free(out_buf);
1140 deflateEnd(&strm);
1141 return -1;
1143 out_len = strm.next_out - out_buf;
1145 deflateEnd(&strm);
1147 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1148 /* could not compress: write normal cluster */
1149 bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1150 } else {
1151 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1152 sector_num << 9, out_len);
1153 if (!cluster_offset)
1154 return -1;
1155 cluster_offset &= s->cluster_offset_mask;
1156 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1157 if (bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len) != out_len) {
1158 g_free(out_buf);
1159 return -1;
1163 g_free(out_buf);
1164 return 0;
1167 static int qcow2_flush(BlockDriverState *bs)
1169 BDRVQcowState *s = bs->opaque;
1170 int ret;
1172 ret = qcow2_cache_flush(bs, s->l2_table_cache);
1173 if (ret < 0) {
1174 return ret;
1177 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1178 if (ret < 0) {
1179 return ret;
1182 return bdrv_flush(bs->file);
1185 static BlockDriverAIOCB *qcow2_aio_flush(BlockDriverState *bs,
1186 BlockDriverCompletionFunc *cb,
1187 void *opaque)
1189 BDRVQcowState *s = bs->opaque;
1190 int ret;
1192 ret = qcow2_cache_flush(bs, s->l2_table_cache);
1193 if (ret < 0) {
1194 return NULL;
1197 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1198 if (ret < 0) {
1199 return NULL;
1202 return bdrv_aio_flush(bs->file, cb, opaque);
1205 static int64_t qcow2_vm_state_offset(BDRVQcowState *s)
1207 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1210 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1212 BDRVQcowState *s = bs->opaque;
1213 bdi->cluster_size = s->cluster_size;
1214 bdi->vm_state_offset = qcow2_vm_state_offset(s);
1215 return 0;
1219 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result)
1221 return qcow2_check_refcounts(bs, result);
1224 #if 0
1225 static void dump_refcounts(BlockDriverState *bs)
1227 BDRVQcowState *s = bs->opaque;
1228 int64_t nb_clusters, k, k1, size;
1229 int refcount;
1231 size = bdrv_getlength(bs->file);
1232 nb_clusters = size_to_clusters(s, size);
1233 for(k = 0; k < nb_clusters;) {
1234 k1 = k;
1235 refcount = get_refcount(bs, k);
1236 k++;
1237 while (k < nb_clusters && get_refcount(bs, k) == refcount)
1238 k++;
1239 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
1240 k - k1);
1243 #endif
1245 static int qcow2_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1246 int64_t pos, int size)
1248 BDRVQcowState *s = bs->opaque;
1249 int growable = bs->growable;
1250 int ret;
1252 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
1253 bs->growable = 1;
1254 ret = bdrv_pwrite(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1255 bs->growable = growable;
1257 return ret;
1260 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1261 int64_t pos, int size)
1263 BDRVQcowState *s = bs->opaque;
1264 int growable = bs->growable;
1265 int ret;
1267 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
1268 bs->growable = 1;
1269 ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1270 bs->growable = growable;
1272 return ret;
1275 static QEMUOptionParameter qcow2_create_options[] = {
1277 .name = BLOCK_OPT_SIZE,
1278 .type = OPT_SIZE,
1279 .help = "Virtual disk size"
1282 .name = BLOCK_OPT_BACKING_FILE,
1283 .type = OPT_STRING,
1284 .help = "File name of a base image"
1287 .name = BLOCK_OPT_BACKING_FMT,
1288 .type = OPT_STRING,
1289 .help = "Image format of the base image"
1292 .name = BLOCK_OPT_ENCRYPT,
1293 .type = OPT_FLAG,
1294 .help = "Encrypt the image"
1297 .name = BLOCK_OPT_CLUSTER_SIZE,
1298 .type = OPT_SIZE,
1299 .help = "qcow2 cluster size",
1300 .value = { .n = DEFAULT_CLUSTER_SIZE },
1303 .name = BLOCK_OPT_PREALLOC,
1304 .type = OPT_STRING,
1305 .help = "Preallocation mode (allowed values: off, metadata)"
1307 { NULL }
1310 static BlockDriver bdrv_qcow2 = {
1311 .format_name = "qcow2",
1312 .instance_size = sizeof(BDRVQcowState),
1313 .bdrv_probe = qcow2_probe,
1314 .bdrv_open = qcow2_open,
1315 .bdrv_close = qcow2_close,
1316 .bdrv_create = qcow2_create,
1317 .bdrv_flush = qcow2_flush,
1318 .bdrv_is_allocated = qcow2_is_allocated,
1319 .bdrv_set_key = qcow2_set_key,
1320 .bdrv_make_empty = qcow2_make_empty,
1322 .bdrv_co_readv = qcow2_co_readv,
1323 .bdrv_co_writev = qcow2_co_writev,
1324 .bdrv_aio_flush = qcow2_aio_flush,
1326 .bdrv_discard = qcow2_discard,
1327 .bdrv_truncate = qcow2_truncate,
1328 .bdrv_write_compressed = qcow2_write_compressed,
1330 .bdrv_snapshot_create = qcow2_snapshot_create,
1331 .bdrv_snapshot_goto = qcow2_snapshot_goto,
1332 .bdrv_snapshot_delete = qcow2_snapshot_delete,
1333 .bdrv_snapshot_list = qcow2_snapshot_list,
1334 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
1335 .bdrv_get_info = qcow2_get_info,
1337 .bdrv_save_vmstate = qcow2_save_vmstate,
1338 .bdrv_load_vmstate = qcow2_load_vmstate,
1340 .bdrv_change_backing_file = qcow2_change_backing_file,
1342 .create_options = qcow2_create_options,
1343 .bdrv_check = qcow2_check,
1346 static void bdrv_qcow2_init(void)
1348 bdrv_register(&bdrv_qcow2);
1351 block_init(bdrv_qcow2_init);