PPC: e500mc: add missing IVORs to bitmap
[qemu.git] / block / qcow2.c
1 /*
2 * Block driver for the QCOW version 2 format
3 *
4 * Copyright (c) 2004-2006 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
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"
32
33 /*
34 Differences with QCOW:
35
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.
47 */
48
49
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
56
57 static int qcow2_probe(const uint8_t *buf, int buf_size, const char *filename)
58 {
59 const QCowHeader *cow_header = (const void *)buf;
60
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;
67 }
68
69
70 /*
71 * read qcow2 extension and fill bs
72 * start reading from start_offset
73 * finish reading upon magic of value 0 or when end_offset reached
74 * unknown magic is skipped (future extension this version knows nothing about)
75 * return 0 upon success, non-0 otherwise
76 */
77 static int qcow2_read_extensions(BlockDriverState *bs, uint64_t start_offset,
78 uint64_t end_offset)
79 {
80 QCowExtension ext;
81 uint64_t offset;
82
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) {
88
89 #ifdef DEBUG_EXT
90 BDRVQcowState *s = bs->opaque;
91 /* Sanity check */
92 if (offset > s->cluster_size)
93 printf("qcow2_read_extension: suspicious offset %lu\n", offset);
94
95 printf("attempting to read extended header in offset %lu\n", offset);
96 #endif
97
98 if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) {
99 fprintf(stderr, "qcow2_read_extension: ERROR: "
100 "pread fail from offset %" PRIu64 "\n",
101 offset);
102 return 1;
103 }
104 be32_to_cpus(&ext.magic);
105 be32_to_cpus(&ext.len);
106 offset += sizeof(ext);
107 #ifdef DEBUG_EXT
108 printf("ext.magic = 0x%x\n", ext.magic);
109 #endif
110 switch (ext.magic) {
111 case QCOW2_EXT_MAGIC_END:
112 return 0;
113
114 case QCOW2_EXT_MAGIC_BACKING_FORMAT:
115 if (ext.len >= sizeof(bs->backing_format)) {
116 fprintf(stderr, "ERROR: ext_backing_format: len=%u too large"
117 " (>=%zu)\n",
118 ext.len, sizeof(bs->backing_format));
119 return 2;
120 }
121 if (bdrv_pread(bs->file, offset , bs->backing_format,
122 ext.len) != ext.len)
123 return 3;
124 bs->backing_format[ext.len] = '\0';
125 #ifdef DEBUG_EXT
126 printf("Qcow2: Got format extension %s\n", bs->backing_format);
127 #endif
128 offset = ((offset + ext.len + 7) & ~7);
129 break;
130
131 default:
132 /* unknown magic -- just skip it */
133 offset = ((offset + ext.len + 7) & ~7);
134 break;
135 }
136 }
137
138 return 0;
139 }
140
141
142 static int qcow2_open(BlockDriverState *bs, int flags)
143 {
144 BDRVQcowState *s = bs->opaque;
145 int len, i, ret = 0;
146 QCowHeader header;
147 uint64_t ext_end;
148 bool writethrough;
149
150 ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
151 if (ret < 0) {
152 goto fail;
153 }
154 be32_to_cpus(&header.magic);
155 be32_to_cpus(&header.version);
156 be64_to_cpus(&header.backing_file_offset);
157 be32_to_cpus(&header.backing_file_size);
158 be64_to_cpus(&header.size);
159 be32_to_cpus(&header.cluster_bits);
160 be32_to_cpus(&header.crypt_method);
161 be64_to_cpus(&header.l1_table_offset);
162 be32_to_cpus(&header.l1_size);
163 be64_to_cpus(&header.refcount_table_offset);
164 be32_to_cpus(&header.refcount_table_clusters);
165 be64_to_cpus(&header.snapshots_offset);
166 be32_to_cpus(&header.nb_snapshots);
167
168 if (header.magic != QCOW_MAGIC) {
169 ret = -EINVAL;
170 goto fail;
171 }
172 if (header.version != QCOW_VERSION) {
173 char version[64];
174 snprintf(version, sizeof(version), "QCOW version %d", header.version);
175 qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
176 bs->device_name, "qcow2", version);
177 ret = -ENOTSUP;
178 goto fail;
179 }
180 if (header.cluster_bits < MIN_CLUSTER_BITS ||
181 header.cluster_bits > MAX_CLUSTER_BITS) {
182 ret = -EINVAL;
183 goto fail;
184 }
185 if (header.crypt_method > QCOW_CRYPT_AES) {
186 ret = -EINVAL;
187 goto fail;
188 }
189 s->crypt_method_header = header.crypt_method;
190 if (s->crypt_method_header) {
191 bs->encrypted = 1;
192 }
193 s->cluster_bits = header.cluster_bits;
194 s->cluster_size = 1 << s->cluster_bits;
195 s->cluster_sectors = 1 << (s->cluster_bits - 9);
196 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
197 s->l2_size = 1 << s->l2_bits;
198 bs->total_sectors = header.size / 512;
199 s->csize_shift = (62 - (s->cluster_bits - 8));
200 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
201 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
202 s->refcount_table_offset = header.refcount_table_offset;
203 s->refcount_table_size =
204 header.refcount_table_clusters << (s->cluster_bits - 3);
205
206 s->snapshots_offset = header.snapshots_offset;
207 s->nb_snapshots = header.nb_snapshots;
208
209 /* read the level 1 table */
210 s->l1_size = header.l1_size;
211 s->l1_vm_state_index = size_to_l1(s, header.size);
212 /* the L1 table must contain at least enough entries to put
213 header.size bytes */
214 if (s->l1_size < s->l1_vm_state_index) {
215 ret = -EINVAL;
216 goto fail;
217 }
218 s->l1_table_offset = header.l1_table_offset;
219 if (s->l1_size > 0) {
220 s->l1_table = g_malloc0(
221 align_offset(s->l1_size * sizeof(uint64_t), 512));
222 ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table,
223 s->l1_size * sizeof(uint64_t));
224 if (ret < 0) {
225 goto fail;
226 }
227 for(i = 0;i < s->l1_size; i++) {
228 be64_to_cpus(&s->l1_table[i]);
229 }
230 }
231
232 /* alloc L2 table/refcount block cache */
233 writethrough = ((flags & BDRV_O_CACHE_WB) == 0);
234 s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE, writethrough);
235 s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE,
236 writethrough);
237
238 s->cluster_cache = g_malloc(s->cluster_size);
239 /* one more sector for decompressed data alignment */
240 s->cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
241 + 512);
242 s->cluster_cache_offset = -1;
243 s->flags = flags;
244
245 ret = qcow2_refcount_init(bs);
246 if (ret != 0) {
247 goto fail;
248 }
249
250 QLIST_INIT(&s->cluster_allocs);
251
252 /* read qcow2 extensions */
253 if (header.backing_file_offset) {
254 ext_end = header.backing_file_offset;
255 } else {
256 ext_end = s->cluster_size;
257 }
258 if (qcow2_read_extensions(bs, sizeof(header), ext_end)) {
259 ret = -EINVAL;
260 goto fail;
261 }
262
263 /* read the backing file name */
264 if (header.backing_file_offset != 0) {
265 len = header.backing_file_size;
266 if (len > 1023) {
267 len = 1023;
268 }
269 ret = bdrv_pread(bs->file, header.backing_file_offset,
270 bs->backing_file, len);
271 if (ret < 0) {
272 goto fail;
273 }
274 bs->backing_file[len] = '\0';
275 }
276
277 ret = qcow2_read_snapshots(bs);
278 if (ret < 0) {
279 goto fail;
280 }
281
282 /* Initialise locks */
283 qemu_co_mutex_init(&s->lock);
284
285 #ifdef DEBUG_ALLOC
286 {
287 BdrvCheckResult result = {0};
288 qcow2_check_refcounts(bs, &result);
289 }
290 #endif
291 return ret;
292
293 fail:
294 qcow2_free_snapshots(bs);
295 qcow2_refcount_close(bs);
296 g_free(s->l1_table);
297 if (s->l2_table_cache) {
298 qcow2_cache_destroy(bs, s->l2_table_cache);
299 }
300 g_free(s->cluster_cache);
301 qemu_vfree(s->cluster_data);
302 return ret;
303 }
304
305 static int qcow2_set_key(BlockDriverState *bs, const char *key)
306 {
307 BDRVQcowState *s = bs->opaque;
308 uint8_t keybuf[16];
309 int len, i;
310
311 memset(keybuf, 0, 16);
312 len = strlen(key);
313 if (len > 16)
314 len = 16;
315 /* XXX: we could compress the chars to 7 bits to increase
316 entropy */
317 for(i = 0;i < len;i++) {
318 keybuf[i] = key[i];
319 }
320 s->crypt_method = s->crypt_method_header;
321
322 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
323 return -1;
324 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
325 return -1;
326 #if 0
327 /* test */
328 {
329 uint8_t in[16];
330 uint8_t out[16];
331 uint8_t tmp[16];
332 for(i=0;i<16;i++)
333 in[i] = i;
334 AES_encrypt(in, tmp, &s->aes_encrypt_key);
335 AES_decrypt(tmp, out, &s->aes_decrypt_key);
336 for(i = 0; i < 16; i++)
337 printf(" %02x", tmp[i]);
338 printf("\n");
339 for(i = 0; i < 16; i++)
340 printf(" %02x", out[i]);
341 printf("\n");
342 }
343 #endif
344 return 0;
345 }
346
347 static int coroutine_fn qcow2_co_is_allocated(BlockDriverState *bs,
348 int64_t sector_num, int nb_sectors, int *pnum)
349 {
350 BDRVQcowState *s = bs->opaque;
351 uint64_t cluster_offset;
352 int ret;
353
354 *pnum = nb_sectors;
355 /* FIXME We can get errors here, but the bdrv_co_is_allocated interface
356 * can't pass them on today */
357 qemu_co_mutex_lock(&s->lock);
358 ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
359 qemu_co_mutex_unlock(&s->lock);
360 if (ret < 0) {
361 *pnum = 0;
362 }
363
364 return (cluster_offset != 0);
365 }
366
367 /* handle reading after the end of the backing file */
368 int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
369 int64_t sector_num, int nb_sectors)
370 {
371 int n1;
372 if ((sector_num + nb_sectors) <= bs->total_sectors)
373 return nb_sectors;
374 if (sector_num >= bs->total_sectors)
375 n1 = 0;
376 else
377 n1 = bs->total_sectors - sector_num;
378
379 qemu_iovec_memset_skip(qiov, 0, 512 * (nb_sectors - n1), 512 * n1);
380
381 return n1;
382 }
383
384 static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
385 int remaining_sectors, QEMUIOVector *qiov)
386 {
387 BDRVQcowState *s = bs->opaque;
388 int index_in_cluster, n1;
389 int ret;
390 int cur_nr_sectors; /* number of sectors in current iteration */
391 uint64_t cluster_offset = 0;
392 uint64_t bytes_done = 0;
393 QEMUIOVector hd_qiov;
394 uint8_t *cluster_data = NULL;
395
396 qemu_iovec_init(&hd_qiov, qiov->niov);
397
398 qemu_co_mutex_lock(&s->lock);
399
400 while (remaining_sectors != 0) {
401
402 /* prepare next request */
403 cur_nr_sectors = remaining_sectors;
404 if (s->crypt_method) {
405 cur_nr_sectors = MIN(cur_nr_sectors,
406 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
407 }
408
409 ret = qcow2_get_cluster_offset(bs, sector_num << 9,
410 &cur_nr_sectors, &cluster_offset);
411 if (ret < 0) {
412 goto fail;
413 }
414
415 index_in_cluster = sector_num & (s->cluster_sectors - 1);
416
417 qemu_iovec_reset(&hd_qiov);
418 qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
419 cur_nr_sectors * 512);
420
421 if (!cluster_offset) {
422
423 if (bs->backing_hd) {
424 /* read from the base image */
425 n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov,
426 sector_num, cur_nr_sectors);
427 if (n1 > 0) {
428 BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
429 qemu_co_mutex_unlock(&s->lock);
430 ret = bdrv_co_readv(bs->backing_hd, sector_num,
431 n1, &hd_qiov);
432 qemu_co_mutex_lock(&s->lock);
433 if (ret < 0) {
434 goto fail;
435 }
436 }
437 } else {
438 /* Note: in this case, no need to wait */
439 qemu_iovec_memset(&hd_qiov, 0, 512 * cur_nr_sectors);
440 }
441 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
442 /* add AIO support for compressed blocks ? */
443 ret = qcow2_decompress_cluster(bs, cluster_offset);
444 if (ret < 0) {
445 goto fail;
446 }
447
448 qemu_iovec_from_buffer(&hd_qiov,
449 s->cluster_cache + index_in_cluster * 512,
450 512 * cur_nr_sectors);
451 } else {
452 if ((cluster_offset & 511) != 0) {
453 ret = -EIO;
454 goto fail;
455 }
456
457 if (s->crypt_method) {
458 /*
459 * For encrypted images, read everything into a temporary
460 * contiguous buffer on which the AES functions can work.
461 */
462 if (!cluster_data) {
463 cluster_data =
464 qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
465 }
466
467 assert(cur_nr_sectors <=
468 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors);
469 qemu_iovec_reset(&hd_qiov);
470 qemu_iovec_add(&hd_qiov, cluster_data,
471 512 * cur_nr_sectors);
472 }
473
474 BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
475 qemu_co_mutex_unlock(&s->lock);
476 ret = bdrv_co_readv(bs->file,
477 (cluster_offset >> 9) + index_in_cluster,
478 cur_nr_sectors, &hd_qiov);
479 qemu_co_mutex_lock(&s->lock);
480 if (ret < 0) {
481 goto fail;
482 }
483 if (s->crypt_method) {
484 qcow2_encrypt_sectors(s, sector_num, cluster_data,
485 cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
486 qemu_iovec_reset(&hd_qiov);
487 qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
488 cur_nr_sectors * 512);
489 qemu_iovec_from_buffer(&hd_qiov, cluster_data,
490 512 * cur_nr_sectors);
491 }
492 }
493
494 remaining_sectors -= cur_nr_sectors;
495 sector_num += cur_nr_sectors;
496 bytes_done += cur_nr_sectors * 512;
497 }
498 ret = 0;
499
500 fail:
501 qemu_co_mutex_unlock(&s->lock);
502
503 qemu_iovec_destroy(&hd_qiov);
504 qemu_vfree(cluster_data);
505
506 return ret;
507 }
508
509 static void run_dependent_requests(BDRVQcowState *s, QCowL2Meta *m)
510 {
511 /* Take the request off the list of running requests */
512 if (m->nb_clusters != 0) {
513 QLIST_REMOVE(m, next_in_flight);
514 }
515
516 /* Restart all dependent requests */
517 if (!qemu_co_queue_empty(&m->dependent_requests)) {
518 qemu_co_mutex_unlock(&s->lock);
519 qemu_co_queue_restart_all(&m->dependent_requests);
520 qemu_co_mutex_lock(&s->lock);
521 }
522 }
523
524 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
525 int64_t sector_num,
526 int remaining_sectors,
527 QEMUIOVector *qiov)
528 {
529 BDRVQcowState *s = bs->opaque;
530 int index_in_cluster;
531 int n_end;
532 int ret;
533 int cur_nr_sectors; /* number of sectors in current iteration */
534 uint64_t cluster_offset;
535 QEMUIOVector hd_qiov;
536 uint64_t bytes_done = 0;
537 uint8_t *cluster_data = NULL;
538 QCowL2Meta l2meta = {
539 .nb_clusters = 0,
540 };
541
542 qemu_co_queue_init(&l2meta.dependent_requests);
543
544 qemu_iovec_init(&hd_qiov, qiov->niov);
545
546 s->cluster_cache_offset = -1; /* disable compressed cache */
547
548 qemu_co_mutex_lock(&s->lock);
549
550 while (remaining_sectors != 0) {
551
552 index_in_cluster = sector_num & (s->cluster_sectors - 1);
553 n_end = index_in_cluster + remaining_sectors;
554 if (s->crypt_method &&
555 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) {
556 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
557 }
558
559 ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
560 index_in_cluster, n_end, &cur_nr_sectors, &l2meta);
561 if (ret < 0) {
562 goto fail;
563 }
564
565 cluster_offset = l2meta.cluster_offset;
566 assert((cluster_offset & 511) == 0);
567
568 qemu_iovec_reset(&hd_qiov);
569 qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
570 cur_nr_sectors * 512);
571
572 if (s->crypt_method) {
573 if (!cluster_data) {
574 cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS *
575 s->cluster_size);
576 }
577
578 assert(hd_qiov.size <=
579 QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
580 qemu_iovec_to_buffer(&hd_qiov, cluster_data);
581
582 qcow2_encrypt_sectors(s, sector_num, cluster_data,
583 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
584
585 qemu_iovec_reset(&hd_qiov);
586 qemu_iovec_add(&hd_qiov, cluster_data,
587 cur_nr_sectors * 512);
588 }
589
590 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
591 qemu_co_mutex_unlock(&s->lock);
592 ret = bdrv_co_writev(bs->file,
593 (cluster_offset >> 9) + index_in_cluster,
594 cur_nr_sectors, &hd_qiov);
595 qemu_co_mutex_lock(&s->lock);
596 if (ret < 0) {
597 goto fail;
598 }
599
600 ret = qcow2_alloc_cluster_link_l2(bs, &l2meta);
601 if (ret < 0) {
602 goto fail;
603 }
604
605 run_dependent_requests(s, &l2meta);
606
607 remaining_sectors -= cur_nr_sectors;
608 sector_num += cur_nr_sectors;
609 bytes_done += cur_nr_sectors * 512;
610 }
611 ret = 0;
612
613 fail:
614 run_dependent_requests(s, &l2meta);
615
616 qemu_co_mutex_unlock(&s->lock);
617
618 qemu_iovec_destroy(&hd_qiov);
619 qemu_vfree(cluster_data);
620
621 return ret;
622 }
623
624 static void qcow2_close(BlockDriverState *bs)
625 {
626 BDRVQcowState *s = bs->opaque;
627 g_free(s->l1_table);
628
629 qcow2_cache_flush(bs, s->l2_table_cache);
630 qcow2_cache_flush(bs, s->refcount_block_cache);
631
632 qcow2_cache_destroy(bs, s->l2_table_cache);
633 qcow2_cache_destroy(bs, s->refcount_block_cache);
634
635 g_free(s->cluster_cache);
636 qemu_vfree(s->cluster_data);
637 qcow2_refcount_close(bs);
638 qcow2_free_snapshots(bs);
639 }
640
641 static void qcow2_invalidate_cache(BlockDriverState *bs)
642 {
643 BDRVQcowState *s = bs->opaque;
644 int flags = s->flags;
645 AES_KEY aes_encrypt_key;
646 AES_KEY aes_decrypt_key;
647 uint32_t crypt_method = 0;
648
649 /*
650 * Backing files are read-only which makes all of their metadata immutable,
651 * that means we don't have to worry about reopening them here.
652 */
653
654 if (s->crypt_method) {
655 crypt_method = s->crypt_method;
656 memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key));
657 memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key));
658 }
659
660 qcow2_close(bs);
661
662 memset(s, 0, sizeof(BDRVQcowState));
663 qcow2_open(bs, flags);
664
665 if (crypt_method) {
666 s->crypt_method = crypt_method;
667 memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key));
668 memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key));
669 }
670 }
671
672 /*
673 * Updates the variable length parts of the qcow2 header, i.e. the backing file
674 * name and all extensions. qcow2 was not designed to allow such changes, so if
675 * we run out of space (we can only use the first cluster) this function may
676 * fail.
677 *
678 * Returns 0 on success, -errno in error cases.
679 */
680 static int qcow2_update_ext_header(BlockDriverState *bs,
681 const char *backing_file, const char *backing_fmt)
682 {
683 size_t backing_file_len = 0;
684 size_t backing_fmt_len = 0;
685 BDRVQcowState *s = bs->opaque;
686 QCowExtension ext_backing_fmt = {0, 0};
687 int ret;
688
689 /* Backing file format doesn't make sense without a backing file */
690 if (backing_fmt && !backing_file) {
691 return -EINVAL;
692 }
693
694 /* Prepare the backing file format extension if needed */
695 if (backing_fmt) {
696 ext_backing_fmt.len = cpu_to_be32(strlen(backing_fmt));
697 ext_backing_fmt.magic = cpu_to_be32(QCOW2_EXT_MAGIC_BACKING_FORMAT);
698 backing_fmt_len = ((sizeof(ext_backing_fmt)
699 + strlen(backing_fmt) + 7) & ~7);
700 }
701
702 /* Check if we can fit the new header into the first cluster */
703 if (backing_file) {
704 backing_file_len = strlen(backing_file);
705 }
706
707 size_t header_size = sizeof(QCowHeader) + backing_file_len
708 + backing_fmt_len;
709
710 if (header_size > s->cluster_size) {
711 return -ENOSPC;
712 }
713
714 /* Rewrite backing file name and qcow2 extensions */
715 size_t ext_size = header_size - sizeof(QCowHeader);
716 uint8_t buf[ext_size];
717 size_t offset = 0;
718 size_t backing_file_offset = 0;
719
720 if (backing_file) {
721 if (backing_fmt) {
722 int padding = backing_fmt_len -
723 (sizeof(ext_backing_fmt) + strlen(backing_fmt));
724
725 memcpy(buf + offset, &ext_backing_fmt, sizeof(ext_backing_fmt));
726 offset += sizeof(ext_backing_fmt);
727
728 memcpy(buf + offset, backing_fmt, strlen(backing_fmt));
729 offset += strlen(backing_fmt);
730
731 memset(buf + offset, 0, padding);
732 offset += padding;
733 }
734
735 memcpy(buf + offset, backing_file, backing_file_len);
736 backing_file_offset = sizeof(QCowHeader) + offset;
737 }
738
739 ret = bdrv_pwrite_sync(bs->file, sizeof(QCowHeader), buf, ext_size);
740 if (ret < 0) {
741 goto fail;
742 }
743
744 /* Update header fields */
745 uint64_t be_backing_file_offset = cpu_to_be64(backing_file_offset);
746 uint32_t be_backing_file_size = cpu_to_be32(backing_file_len);
747
748 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_offset),
749 &be_backing_file_offset, sizeof(uint64_t));
750 if (ret < 0) {
751 goto fail;
752 }
753
754 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_size),
755 &be_backing_file_size, sizeof(uint32_t));
756 if (ret < 0) {
757 goto fail;
758 }
759
760 ret = 0;
761 fail:
762 return ret;
763 }
764
765 static int qcow2_change_backing_file(BlockDriverState *bs,
766 const char *backing_file, const char *backing_fmt)
767 {
768 return qcow2_update_ext_header(bs, backing_file, backing_fmt);
769 }
770
771 static int preallocate(BlockDriverState *bs)
772 {
773 uint64_t nb_sectors;
774 uint64_t offset;
775 int num;
776 int ret;
777 QCowL2Meta meta;
778
779 nb_sectors = bdrv_getlength(bs) >> 9;
780 offset = 0;
781 qemu_co_queue_init(&meta.dependent_requests);
782 meta.cluster_offset = 0;
783
784 while (nb_sectors) {
785 num = MIN(nb_sectors, INT_MAX >> 9);
786 ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta);
787 if (ret < 0) {
788 return ret;
789 }
790
791 ret = qcow2_alloc_cluster_link_l2(bs, &meta);
792 if (ret < 0) {
793 qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters);
794 return ret;
795 }
796
797 /* There are no dependent requests, but we need to remove our request
798 * from the list of in-flight requests */
799 run_dependent_requests(bs->opaque, &meta);
800
801 /* TODO Preallocate data if requested */
802
803 nb_sectors -= num;
804 offset += num << 9;
805 }
806
807 /*
808 * It is expected that the image file is large enough to actually contain
809 * all of the allocated clusters (otherwise we get failing reads after
810 * EOF). Extend the image to the last allocated sector.
811 */
812 if (meta.cluster_offset != 0) {
813 uint8_t buf[512];
814 memset(buf, 0, 512);
815 ret = bdrv_write(bs->file, (meta.cluster_offset >> 9) + num - 1, buf, 1);
816 if (ret < 0) {
817 return ret;
818 }
819 }
820
821 return 0;
822 }
823
824 static int qcow2_create2(const char *filename, int64_t total_size,
825 const char *backing_file, const char *backing_format,
826 int flags, size_t cluster_size, int prealloc,
827 QEMUOptionParameter *options)
828 {
829 /* Calculate cluster_bits */
830 int cluster_bits;
831 cluster_bits = ffs(cluster_size) - 1;
832 if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
833 (1 << cluster_bits) != cluster_size)
834 {
835 error_report(
836 "Cluster size must be a power of two between %d and %dk",
837 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10));
838 return -EINVAL;
839 }
840
841 /*
842 * Open the image file and write a minimal qcow2 header.
843 *
844 * We keep things simple and start with a zero-sized image. We also
845 * do without refcount blocks or a L1 table for now. We'll fix the
846 * inconsistency later.
847 *
848 * We do need a refcount table because growing the refcount table means
849 * allocating two new refcount blocks - the seconds of which would be at
850 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
851 * size for any qcow2 image.
852 */
853 BlockDriverState* bs;
854 QCowHeader header;
855 uint8_t* refcount_table;
856 int ret;
857
858 ret = bdrv_create_file(filename, options);
859 if (ret < 0) {
860 return ret;
861 }
862
863 ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR);
864 if (ret < 0) {
865 return ret;
866 }
867
868 /* Write the header */
869 memset(&header, 0, sizeof(header));
870 header.magic = cpu_to_be32(QCOW_MAGIC);
871 header.version = cpu_to_be32(QCOW_VERSION);
872 header.cluster_bits = cpu_to_be32(cluster_bits);
873 header.size = cpu_to_be64(0);
874 header.l1_table_offset = cpu_to_be64(0);
875 header.l1_size = cpu_to_be32(0);
876 header.refcount_table_offset = cpu_to_be64(cluster_size);
877 header.refcount_table_clusters = cpu_to_be32(1);
878
879 if (flags & BLOCK_FLAG_ENCRYPT) {
880 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
881 } else {
882 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
883 }
884
885 ret = bdrv_pwrite(bs, 0, &header, sizeof(header));
886 if (ret < 0) {
887 goto out;
888 }
889
890 /* Write an empty refcount table */
891 refcount_table = g_malloc0(cluster_size);
892 ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size);
893 g_free(refcount_table);
894
895 if (ret < 0) {
896 goto out;
897 }
898
899 bdrv_close(bs);
900
901 /*
902 * And now open the image and make it consistent first (i.e. increase the
903 * refcount of the cluster that is occupied by the header and the refcount
904 * table)
905 */
906 BlockDriver* drv = bdrv_find_format("qcow2");
907 assert(drv != NULL);
908 ret = bdrv_open(bs, filename,
909 BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv);
910 if (ret < 0) {
911 goto out;
912 }
913
914 ret = qcow2_alloc_clusters(bs, 2 * cluster_size);
915 if (ret < 0) {
916 goto out;
917
918 } else if (ret != 0) {
919 error_report("Huh, first cluster in empty image is already in use?");
920 abort();
921 }
922
923 /* Okay, now that we have a valid image, let's give it the right size */
924 ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE);
925 if (ret < 0) {
926 goto out;
927 }
928
929 /* Want a backing file? There you go.*/
930 if (backing_file) {
931 ret = bdrv_change_backing_file(bs, backing_file, backing_format);
932 if (ret < 0) {
933 goto out;
934 }
935 }
936
937 /* And if we're supposed to preallocate metadata, do that now */
938 if (prealloc) {
939 ret = preallocate(bs);
940 if (ret < 0) {
941 goto out;
942 }
943 }
944
945 ret = 0;
946 out:
947 bdrv_delete(bs);
948 return ret;
949 }
950
951 static int qcow2_create(const char *filename, QEMUOptionParameter *options)
952 {
953 const char *backing_file = NULL;
954 const char *backing_fmt = NULL;
955 uint64_t sectors = 0;
956 int flags = 0;
957 size_t cluster_size = DEFAULT_CLUSTER_SIZE;
958 int prealloc = 0;
959
960 /* Read out options */
961 while (options && options->name) {
962 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
963 sectors = options->value.n / 512;
964 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
965 backing_file = options->value.s;
966 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
967 backing_fmt = options->value.s;
968 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
969 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
970 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
971 if (options->value.n) {
972 cluster_size = options->value.n;
973 }
974 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
975 if (!options->value.s || !strcmp(options->value.s, "off")) {
976 prealloc = 0;
977 } else if (!strcmp(options->value.s, "metadata")) {
978 prealloc = 1;
979 } else {
980 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
981 options->value.s);
982 return -EINVAL;
983 }
984 }
985 options++;
986 }
987
988 if (backing_file && prealloc) {
989 fprintf(stderr, "Backing file and preallocation cannot be used at "
990 "the same time\n");
991 return -EINVAL;
992 }
993
994 return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags,
995 cluster_size, prealloc, options);
996 }
997
998 static int qcow2_make_empty(BlockDriverState *bs)
999 {
1000 #if 0
1001 /* XXX: not correct */
1002 BDRVQcowState *s = bs->opaque;
1003 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1004 int ret;
1005
1006 memset(s->l1_table, 0, l1_length);
1007 if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
1008 return -1;
1009 ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
1010 if (ret < 0)
1011 return ret;
1012
1013 l2_cache_reset(bs);
1014 #endif
1015 return 0;
1016 }
1017
1018 static coroutine_fn int qcow2_co_discard(BlockDriverState *bs,
1019 int64_t sector_num, int nb_sectors)
1020 {
1021 int ret;
1022 BDRVQcowState *s = bs->opaque;
1023
1024 qemu_co_mutex_lock(&s->lock);
1025 ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS,
1026 nb_sectors);
1027 qemu_co_mutex_unlock(&s->lock);
1028 return ret;
1029 }
1030
1031 static int qcow2_truncate(BlockDriverState *bs, int64_t offset)
1032 {
1033 BDRVQcowState *s = bs->opaque;
1034 int ret, new_l1_size;
1035
1036 if (offset & 511) {
1037 return -EINVAL;
1038 }
1039
1040 /* cannot proceed if image has snapshots */
1041 if (s->nb_snapshots) {
1042 return -ENOTSUP;
1043 }
1044
1045 /* shrinking is currently not supported */
1046 if (offset < bs->total_sectors * 512) {
1047 return -ENOTSUP;
1048 }
1049
1050 new_l1_size = size_to_l1(s, offset);
1051 ret = qcow2_grow_l1_table(bs, new_l1_size, true);
1052 if (ret < 0) {
1053 return ret;
1054 }
1055
1056 /* write updated header.size */
1057 offset = cpu_to_be64(offset);
1058 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size),
1059 &offset, sizeof(uint64_t));
1060 if (ret < 0) {
1061 return ret;
1062 }
1063
1064 s->l1_vm_state_index = new_l1_size;
1065 return 0;
1066 }
1067
1068 /* XXX: put compressed sectors first, then all the cluster aligned
1069 tables to avoid losing bytes in alignment */
1070 static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num,
1071 const uint8_t *buf, int nb_sectors)
1072 {
1073 BDRVQcowState *s = bs->opaque;
1074 z_stream strm;
1075 int ret, out_len;
1076 uint8_t *out_buf;
1077 uint64_t cluster_offset;
1078
1079 if (nb_sectors == 0) {
1080 /* align end of file to a sector boundary to ease reading with
1081 sector based I/Os */
1082 cluster_offset = bdrv_getlength(bs->file);
1083 cluster_offset = (cluster_offset + 511) & ~511;
1084 bdrv_truncate(bs->file, cluster_offset);
1085 return 0;
1086 }
1087
1088 if (nb_sectors != s->cluster_sectors)
1089 return -EINVAL;
1090
1091 out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1092
1093 /* best compression, small window, no zlib header */
1094 memset(&strm, 0, sizeof(strm));
1095 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1096 Z_DEFLATED, -12,
1097 9, Z_DEFAULT_STRATEGY);
1098 if (ret != 0) {
1099 ret = -EINVAL;
1100 goto fail;
1101 }
1102
1103 strm.avail_in = s->cluster_size;
1104 strm.next_in = (uint8_t *)buf;
1105 strm.avail_out = s->cluster_size;
1106 strm.next_out = out_buf;
1107
1108 ret = deflate(&strm, Z_FINISH);
1109 if (ret != Z_STREAM_END && ret != Z_OK) {
1110 deflateEnd(&strm);
1111 ret = -EINVAL;
1112 goto fail;
1113 }
1114 out_len = strm.next_out - out_buf;
1115
1116 deflateEnd(&strm);
1117
1118 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1119 /* could not compress: write normal cluster */
1120 ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors);
1121 if (ret < 0) {
1122 goto fail;
1123 }
1124 } else {
1125 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
1126 sector_num << 9, out_len);
1127 if (!cluster_offset) {
1128 ret = -EIO;
1129 goto fail;
1130 }
1131 cluster_offset &= s->cluster_offset_mask;
1132 BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED);
1133 ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len);
1134 if (ret < 0) {
1135 goto fail;
1136 }
1137 }
1138
1139 ret = 0;
1140 fail:
1141 g_free(out_buf);
1142 return ret;
1143 }
1144
1145 static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
1146 {
1147 BDRVQcowState *s = bs->opaque;
1148 int ret;
1149
1150 qemu_co_mutex_lock(&s->lock);
1151 ret = qcow2_cache_flush(bs, s->l2_table_cache);
1152 if (ret < 0) {
1153 qemu_co_mutex_unlock(&s->lock);
1154 return ret;
1155 }
1156
1157 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
1158 if (ret < 0) {
1159 qemu_co_mutex_unlock(&s->lock);
1160 return ret;
1161 }
1162 qemu_co_mutex_unlock(&s->lock);
1163
1164 return 0;
1165 }
1166
1167 static coroutine_fn int qcow2_co_flush_to_disk(BlockDriverState *bs)
1168 {
1169 return bdrv_co_flush(bs->file);
1170 }
1171
1172 static int64_t qcow2_vm_state_offset(BDRVQcowState *s)
1173 {
1174 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1175 }
1176
1177 static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1178 {
1179 BDRVQcowState *s = bs->opaque;
1180 bdi->cluster_size = s->cluster_size;
1181 bdi->vm_state_offset = qcow2_vm_state_offset(s);
1182 return 0;
1183 }
1184
1185
1186 static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result)
1187 {
1188 return qcow2_check_refcounts(bs, result);
1189 }
1190
1191 #if 0
1192 static void dump_refcounts(BlockDriverState *bs)
1193 {
1194 BDRVQcowState *s = bs->opaque;
1195 int64_t nb_clusters, k, k1, size;
1196 int refcount;
1197
1198 size = bdrv_getlength(bs->file);
1199 nb_clusters = size_to_clusters(s, size);
1200 for(k = 0; k < nb_clusters;) {
1201 k1 = k;
1202 refcount = get_refcount(bs, k);
1203 k++;
1204 while (k < nb_clusters && get_refcount(bs, k) == refcount)
1205 k++;
1206 printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
1207 k - k1);
1208 }
1209 }
1210 #endif
1211
1212 static int qcow2_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1213 int64_t pos, int size)
1214 {
1215 BDRVQcowState *s = bs->opaque;
1216 int growable = bs->growable;
1217 int ret;
1218
1219 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE);
1220 bs->growable = 1;
1221 ret = bdrv_pwrite(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1222 bs->growable = growable;
1223
1224 return ret;
1225 }
1226
1227 static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1228 int64_t pos, int size)
1229 {
1230 BDRVQcowState *s = bs->opaque;
1231 int growable = bs->growable;
1232 int ret;
1233
1234 BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD);
1235 bs->growable = 1;
1236 ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size);
1237 bs->growable = growable;
1238
1239 return ret;
1240 }
1241
1242 static QEMUOptionParameter qcow2_create_options[] = {
1243 {
1244 .name = BLOCK_OPT_SIZE,
1245 .type = OPT_SIZE,
1246 .help = "Virtual disk size"
1247 },
1248 {
1249 .name = BLOCK_OPT_BACKING_FILE,
1250 .type = OPT_STRING,
1251 .help = "File name of a base image"
1252 },
1253 {
1254 .name = BLOCK_OPT_BACKING_FMT,
1255 .type = OPT_STRING,
1256 .help = "Image format of the base image"
1257 },
1258 {
1259 .name = BLOCK_OPT_ENCRYPT,
1260 .type = OPT_FLAG,
1261 .help = "Encrypt the image"
1262 },
1263 {
1264 .name = BLOCK_OPT_CLUSTER_SIZE,
1265 .type = OPT_SIZE,
1266 .help = "qcow2 cluster size",
1267 .value = { .n = DEFAULT_CLUSTER_SIZE },
1268 },
1269 {
1270 .name = BLOCK_OPT_PREALLOC,
1271 .type = OPT_STRING,
1272 .help = "Preallocation mode (allowed values: off, metadata)"
1273 },
1274 { NULL }
1275 };
1276
1277 static BlockDriver bdrv_qcow2 = {
1278 .format_name = "qcow2",
1279 .instance_size = sizeof(BDRVQcowState),
1280 .bdrv_probe = qcow2_probe,
1281 .bdrv_open = qcow2_open,
1282 .bdrv_close = qcow2_close,
1283 .bdrv_create = qcow2_create,
1284 .bdrv_co_is_allocated = qcow2_co_is_allocated,
1285 .bdrv_set_key = qcow2_set_key,
1286 .bdrv_make_empty = qcow2_make_empty,
1287
1288 .bdrv_co_readv = qcow2_co_readv,
1289 .bdrv_co_writev = qcow2_co_writev,
1290 .bdrv_co_flush_to_os = qcow2_co_flush_to_os,
1291 .bdrv_co_flush_to_disk = qcow2_co_flush_to_disk,
1292
1293 .bdrv_co_discard = qcow2_co_discard,
1294 .bdrv_truncate = qcow2_truncate,
1295 .bdrv_write_compressed = qcow2_write_compressed,
1296
1297 .bdrv_snapshot_create = qcow2_snapshot_create,
1298 .bdrv_snapshot_goto = qcow2_snapshot_goto,
1299 .bdrv_snapshot_delete = qcow2_snapshot_delete,
1300 .bdrv_snapshot_list = qcow2_snapshot_list,
1301 .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp,
1302 .bdrv_get_info = qcow2_get_info,
1303
1304 .bdrv_save_vmstate = qcow2_save_vmstate,
1305 .bdrv_load_vmstate = qcow2_load_vmstate,
1306
1307 .bdrv_change_backing_file = qcow2_change_backing_file,
1308
1309 .bdrv_invalidate_cache = qcow2_invalidate_cache,
1310
1311 .create_options = qcow2_create_options,
1312 .bdrv_check = qcow2_check,
1313 };
1314
1315 static void bdrv_qcow2_init(void)
1316 {
1317 bdrv_register(&bdrv_qcow2);
1318 }
1319
1320 block_init(bdrv_qcow2_init);