?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
qcow2: Bring synchronous read/write back to life
[qemu-kvm/fedora.git] / block / qcow2.c
blobced257e9f57c2c8a1e8b6715453a2da5fd08fc7a
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
31 /*
32 Differences with QCOW:
33
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.
45 */
46
47
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
54
55
56
57 static int qcow_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 qcow_read_extensions(BlockDriverState *bs, uint64_t start_offset,
78 uint64_t end_offset)
79 {
80 BDRVQcowState *s = bs->opaque;
81 QCowExtension ext;
82 uint64_t offset;
83
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) {
89
90 #ifdef DEBUG_EXT
91 /* Sanity check */
92 if (offset > s->cluster_size)
93 printf("qcow_handle_extension: suspicious offset %lu\n", offset);
94
95 printf("attemting to read extended header in offset %lu\n", offset);
96 #endif
97
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;
102 }
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;
112
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;
119 }
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 += ((ext.len + 7) & ~7);
128 break;
129
130 default:
131 /* unknown magic -- just skip it */
132 offset += ((ext.len + 7) & ~7);
133 break;
134 }
135 }
136
137 return 0;
138 }
139
140
141 static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
142 {
143 BDRVQcowState *s = bs->opaque;
144 int len, i, shift, ret;
145 QCowHeader header;
146 uint64_t ext_end;
147
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);
166
167 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
168 goto fail;
169 if (header.size <= 1 ||
170 header.cluster_bits < MIN_CLUSTER_BITS ||
171 header.cluster_bits > MAX_CLUSTER_BITS)
172 goto fail;
173 if (header.crypt_method > QCOW_CRYPT_AES)
174 goto fail;
175 s->crypt_method_header = header.crypt_method;
176 if (s->crypt_method_header)
177 bs->encrypted = 1;
178 s->cluster_bits = header.cluster_bits;
179 s->cluster_size = 1 << s->cluster_bits;
180 s->cluster_sectors = 1 << (s->cluster_bits - 9);
181 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
182 s->l2_size = 1 << s->l2_bits;
183 bs->total_sectors = header.size / 512;
184 s->csize_shift = (62 - (s->cluster_bits - 8));
185 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
186 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
187 s->refcount_table_offset = header.refcount_table_offset;
188 s->refcount_table_size =
189 header.refcount_table_clusters << (s->cluster_bits - 3);
190
191 s->snapshots_offset = header.snapshots_offset;
192 s->nb_snapshots = header.nb_snapshots;
193
194 /* read the level 1 table */
195 s->l1_size = header.l1_size;
196 shift = s->cluster_bits + s->l2_bits;
197 s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
198 /* the L1 table must contain at least enough entries to put
199 header.size bytes */
200 if (s->l1_size < s->l1_vm_state_index)
201 goto fail;
202 s->l1_table_offset = header.l1_table_offset;
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]);
210 }
211 /* alloc L2 cache */
212 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
213 s->cluster_cache = qemu_malloc(s->cluster_size);
214 /* one more sector for decompressed data alignment */
215 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
216 + 512);
217 s->cluster_cache_offset = -1;
218
219 if (qcow2_refcount_init(bs) < 0)
220 goto fail;
221
222 LIST_INIT(&s->cluster_allocs);
223
224 /* read qcow2 extensions */
225 if (header.backing_file_offset)
226 ext_end = header.backing_file_offset;
227 else
228 ext_end = s->cluster_size;
229 if (qcow_read_extensions(bs, sizeof(header), ext_end))
230 goto fail;
231
232 /* read the backing file name */
233 if (header.backing_file_offset != 0) {
234 len = header.backing_file_size;
235 if (len > 1023)
236 len = 1023;
237 if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
238 goto fail;
239 bs->backing_file[len] = '\0';
240 }
241 if (qcow2_read_snapshots(bs) < 0)
242 goto fail;
243
244 #ifdef DEBUG_ALLOC
245 qcow2_check_refcounts(bs);
246 #endif
247 return 0;
248
249 fail:
250 qcow2_free_snapshots(bs);
251 qcow2_refcount_close(bs);
252 qemu_free(s->l1_table);
253 qemu_free(s->l2_cache);
254 qemu_free(s->cluster_cache);
255 qemu_free(s->cluster_data);
256 bdrv_delete(s->hd);
257 return -1;
258 }
259
260 static int qcow_set_key(BlockDriverState *bs, const char *key)
261 {
262 BDRVQcowState *s = bs->opaque;
263 uint8_t keybuf[16];
264 int len, i;
265
266 memset(keybuf, 0, 16);
267 len = strlen(key);
268 if (len > 16)
269 len = 16;
270 /* XXX: we could compress the chars to 7 bits to increase
271 entropy */
272 for(i = 0;i < len;i++) {
273 keybuf[i] = key[i];
274 }
275 s->crypt_method = s->crypt_method_header;
276
277 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
278 return -1;
279 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
280 return -1;
281 #if 0
282 /* test */
283 {
284 uint8_t in[16];
285 uint8_t out[16];
286 uint8_t tmp[16];
287 for(i=0;i<16;i++)
288 in[i] = i;
289 AES_encrypt(in, tmp, &s->aes_encrypt_key);
290 AES_decrypt(tmp, out, &s->aes_decrypt_key);
291 for(i = 0; i < 16; i++)
292 printf(" %02x", tmp[i]);
293 printf("\n");
294 for(i = 0; i < 16; i++)
295 printf(" %02x", out[i]);
296 printf("\n");
297 }
298 #endif
299 return 0;
300 }
301
302 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
303 int nb_sectors, int *pnum)
304 {
305 uint64_t cluster_offset;
306
307 *pnum = nb_sectors;
308 cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, pnum);
309
310 return (cluster_offset != 0);
311 }
312
313 /* handle reading after the end of the backing file */
314 int qcow2_backing_read1(BlockDriverState *bs,
315 int64_t sector_num, uint8_t *buf, int nb_sectors)
316 {
317 int n1;
318 if ((sector_num + nb_sectors) <= bs->total_sectors)
319 return nb_sectors;
320 if (sector_num >= bs->total_sectors)
321 n1 = 0;
322 else
323 n1 = bs->total_sectors - sector_num;
324 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
325 return n1;
326 }
327
328 typedef struct QCowAIOCB {
329 BlockDriverAIOCB common;
330 int64_t sector_num;
331 QEMUIOVector *qiov;
332 uint8_t *buf;
333 void *orig_buf;
334 int nb_sectors;
335 int n;
336 uint64_t cluster_offset;
337 uint8_t *cluster_data;
338 BlockDriverAIOCB *hd_aiocb;
339 struct iovec hd_iov;
340 QEMUIOVector hd_qiov;
341 QEMUBH *bh;
342 QCowL2Meta l2meta;
343 LIST_ENTRY(QCowAIOCB) next_depend;
344 } QCowAIOCB;
345
346 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
347 {
348 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
349 if (acb->hd_aiocb)
350 bdrv_aio_cancel(acb->hd_aiocb);
351 qemu_aio_release(acb);
352 }
353
354 static AIOPool qcow_aio_pool = {
355 .aiocb_size = sizeof(QCowAIOCB),
356 .cancel = qcow_aio_cancel,
357 };
358
359 static void qcow_aio_read_cb(void *opaque, int ret);
360 static void qcow_aio_read_bh(void *opaque)
361 {
362 QCowAIOCB *acb = opaque;
363 qemu_bh_delete(acb->bh);
364 acb->bh = NULL;
365 qcow_aio_read_cb(opaque, 0);
366 }
367
368 static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
369 {
370 if (acb->bh)
371 return -EIO;
372
373 acb->bh = qemu_bh_new(cb, acb);
374 if (!acb->bh)
375 return -EIO;
376
377 qemu_bh_schedule(acb->bh);
378
379 return 0;
380 }
381
382 static void qcow_aio_read_cb(void *opaque, int ret)
383 {
384 QCowAIOCB *acb = opaque;
385 BlockDriverState *bs = acb->common.bs;
386 BDRVQcowState *s = bs->opaque;
387 int index_in_cluster, n1;
388
389 acb->hd_aiocb = NULL;
390 if (ret < 0)
391 goto done;
392
393 /* post process the read buffer */
394 if (!acb->cluster_offset) {
395 /* nothing to do */
396 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
397 /* nothing to do */
398 } else {
399 if (s->crypt_method) {
400 qcow2_encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
401 acb->n, 0,
402 &s->aes_decrypt_key);
403 }
404 }
405
406 acb->nb_sectors -= acb->n;
407 acb->sector_num += acb->n;
408 acb->buf += acb->n * 512;
409
410 if (acb->nb_sectors == 0) {
411 /* request completed */
412 ret = 0;
413 goto done;
414 }
415
416 /* prepare next AIO request */
417 acb->n = acb->nb_sectors;
418 acb->cluster_offset =
419 qcow2_get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
420 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
421
422 if (!acb->cluster_offset) {
423 if (bs->backing_hd) {
424 /* read from the base image */
425 n1 = qcow2_backing_read1(bs->backing_hd, acb->sector_num,
426 acb->buf, acb->n);
427 if (n1 > 0) {
428 acb->hd_iov.iov_base = (void *)acb->buf;
429 acb->hd_iov.iov_len = acb->n * 512;
430 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
431 acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
432 &acb->hd_qiov, acb->n,
433 qcow_aio_read_cb, acb);
434 if (acb->hd_aiocb == NULL)
435 goto done;
436 } else {
437 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
438 if (ret < 0)
439 goto done;
440 }
441 } else {
442 /* Note: in this case, no need to wait */
443 memset(acb->buf, 0, 512 * acb->n);
444 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
445 if (ret < 0)
446 goto done;
447 }
448 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
449 /* add AIO support for compressed blocks ? */
450 if (qcow2_decompress_cluster(s, acb->cluster_offset) < 0)
451 goto done;
452 memcpy(acb->buf,
453 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
454 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
455 if (ret < 0)
456 goto done;
457 } else {
458 if ((acb->cluster_offset & 511) != 0) {
459 ret = -EIO;
460 goto done;
461 }
462
463 acb->hd_iov.iov_base = (void *)acb->buf;
464 acb->hd_iov.iov_len = acb->n * 512;
465 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
466 acb->hd_aiocb = bdrv_aio_readv(s->hd,
467 (acb->cluster_offset >> 9) + index_in_cluster,
468 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
469 if (acb->hd_aiocb == NULL)
470 goto done;
471 }
472
473 return;
474 done:
475 if (acb->qiov->niov > 1) {
476 qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
477 qemu_vfree(acb->orig_buf);
478 }
479 acb->common.cb(acb->common.opaque, ret);
480 qemu_aio_release(acb);
481 }
482
483 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
484 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
485 BlockDriverCompletionFunc *cb, void *opaque, int is_write)
486 {
487 QCowAIOCB *acb;
488
489 acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
490 if (!acb)
491 return NULL;
492 acb->hd_aiocb = NULL;
493 acb->sector_num = sector_num;
494 acb->qiov = qiov;
495 if (qiov->niov > 1) {
496 acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
497 if (is_write)
498 qemu_iovec_to_buffer(qiov, acb->buf);
499 } else {
500 acb->buf = (uint8_t *)qiov->iov->iov_base;
501 }
502 acb->nb_sectors = nb_sectors;
503 acb->n = 0;
504 acb->cluster_offset = 0;
505 acb->l2meta.nb_clusters = 0;
506 LIST_INIT(&acb->l2meta.dependent_requests);
507 return acb;
508 }
509
510 static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
511 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
512 BlockDriverCompletionFunc *cb, void *opaque)
513 {
514 QCowAIOCB *acb;
515
516 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
517 if (!acb)
518 return NULL;
519
520 qcow_aio_read_cb(acb, 0);
521 return &acb->common;
522 }
523
524 static void qcow_aio_write_cb(void *opaque, int ret);
525
526 static void run_dependent_requests(QCowL2Meta *m)
527 {
528 QCowAIOCB *req;
529 QCowAIOCB *next;
530
531 /* Take the request off the list of running requests */
532 if (m->nb_clusters != 0) {
533 LIST_REMOVE(m, next_in_flight);
534 }
535
536 /*
537 * Restart all dependent requests.
538 * Can't use LIST_FOREACH here - the next link might not be the same
539 * any more after the callback (request could depend on a different
540 * request now)
541 */
542 for (req = m->dependent_requests.lh_first; req != NULL; req = next) {
543 next = req->next_depend.le_next;
544 qcow_aio_write_cb(req, 0);
545 }
546
547 /* Empty the list for the next part of the request */
548 LIST_INIT(&m->dependent_requests);
549 }
550
551 static void qcow_aio_write_cb(void *opaque, int ret)
552 {
553 QCowAIOCB *acb = opaque;
554 BlockDriverState *bs = acb->common.bs;
555 BDRVQcowState *s = bs->opaque;
556 int index_in_cluster;
557 const uint8_t *src_buf;
558 int n_end;
559
560 acb->hd_aiocb = NULL;
561
562 if (ret >= 0) {
563 ret = qcow2_alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta);
564 }
565
566 run_dependent_requests(&acb->l2meta);
567
568 if (ret < 0)
569 goto done;
570
571 acb->nb_sectors -= acb->n;
572 acb->sector_num += acb->n;
573 acb->buf += acb->n * 512;
574
575 if (acb->nb_sectors == 0) {
576 /* request completed */
577 ret = 0;
578 goto done;
579 }
580
581 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
582 n_end = index_in_cluster + acb->nb_sectors;
583 if (s->crypt_method &&
584 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
585 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
586
587 acb->cluster_offset = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9,
588 index_in_cluster,
589 n_end, &acb->n, &acb->l2meta);
590
591 /* Need to wait for another request? If so, we are done for now. */
592 if (!acb->cluster_offset && acb->l2meta.depends_on != NULL) {
593 LIST_INSERT_HEAD(&acb->l2meta.depends_on->dependent_requests,
594 acb, next_depend);
595 return;
596 }
597
598 if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
599 ret = -EIO;
600 goto done;
601 }
602 if (s->crypt_method) {
603 if (!acb->cluster_data) {
604 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
605 s->cluster_size);
606 }
607 qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
608 acb->n, 1, &s->aes_encrypt_key);
609 src_buf = acb->cluster_data;
610 } else {
611 src_buf = acb->buf;
612 }
613 acb->hd_iov.iov_base = (void *)src_buf;
614 acb->hd_iov.iov_len = acb->n * 512;
615 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
616 acb->hd_aiocb = bdrv_aio_writev(s->hd,
617 (acb->cluster_offset >> 9) + index_in_cluster,
618 &acb->hd_qiov, acb->n,
619 qcow_aio_write_cb, acb);
620 if (acb->hd_aiocb == NULL)
621 goto done;
622
623 return;
624
625 done:
626 if (acb->qiov->niov > 1)
627 qemu_vfree(acb->orig_buf);
628 acb->common.cb(acb->common.opaque, ret);
629 qemu_aio_release(acb);
630 }
631
632 static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
633 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
634 BlockDriverCompletionFunc *cb, void *opaque)
635 {
636 BDRVQcowState *s = bs->opaque;
637 QCowAIOCB *acb;
638
639 s->cluster_cache_offset = -1; /* disable compressed cache */
640
641 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
642 if (!acb)
643 return NULL;
644
645 qcow_aio_write_cb(acb, 0);
646 return &acb->common;
647 }
648
649 static void qcow_close(BlockDriverState *bs)
650 {
651 BDRVQcowState *s = bs->opaque;
652 qemu_free(s->l1_table);
653 qemu_free(s->l2_cache);
654 qemu_free(s->cluster_cache);
655 qemu_free(s->cluster_data);
656 qcow2_refcount_close(bs);
657 bdrv_delete(s->hd);
658 }
659
660 static int get_bits_from_size(size_t size)
661 {
662 int res = 0;
663
664 if (size == 0) {
665 return -1;
666 }
667
668 while (size != 1) {
669 /* Not a power of two */
670 if (size & 1) {
671 return -1;
672 }
673
674 size >>= 1;
675 res++;
676 }
677
678 return res;
679 }
680
681 static int qcow_create2(const char *filename, int64_t total_size,
682 const char *backing_file, const char *backing_format,
683 int flags, size_t cluster_size)
684 {
685
686 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
687 int ref_clusters, backing_format_len = 0;
688 QCowHeader header;
689 uint64_t tmp, offset;
690 QCowCreateState s1, *s = &s1;
691 QCowExtension ext_bf = {0, 0};
692
693
694 memset(s, 0, sizeof(*s));
695
696 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
697 if (fd < 0)
698 return -1;
699 memset(&header, 0, sizeof(header));
700 header.magic = cpu_to_be32(QCOW_MAGIC);
701 header.version = cpu_to_be32(QCOW_VERSION);
702 header.size = cpu_to_be64(total_size * 512);
703 header_size = sizeof(header);
704 backing_filename_len = 0;
705 if (backing_file) {
706 if (backing_format) {
707 ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
708 backing_format_len = strlen(backing_format);
709 ext_bf.len = (backing_format_len + 7) & ~7;
710 header_size += ((sizeof(ext_bf) + ext_bf.len + 7) & ~7);
711 }
712 header.backing_file_offset = cpu_to_be64(header_size);
713 backing_filename_len = strlen(backing_file);
714 header.backing_file_size = cpu_to_be32(backing_filename_len);
715 header_size += backing_filename_len;
716 }
717
718 /* Cluster size */
719 s->cluster_bits = get_bits_from_size(cluster_size);
720 if (s->cluster_bits < MIN_CLUSTER_BITS ||
721 s->cluster_bits > MAX_CLUSTER_BITS)
722 {
723 fprintf(stderr, "Cluster size must be a power of two between "
724 "%d and %dk\n",
725 1 << MIN_CLUSTER_BITS,
726 1 << (MAX_CLUSTER_BITS - 10));
727 return -EINVAL;
728 }
729 s->cluster_size = 1 << s->cluster_bits;
730
731 header.cluster_bits = cpu_to_be32(s->cluster_bits);
732 header_size = (header_size + 7) & ~7;
733 if (flags & BLOCK_FLAG_ENCRYPT) {
734 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
735 } else {
736 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
737 }
738 l2_bits = s->cluster_bits - 3;
739 shift = s->cluster_bits + l2_bits;
740 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
741 offset = align_offset(header_size, s->cluster_size);
742 s->l1_table_offset = offset;
743 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
744 header.l1_size = cpu_to_be32(l1_size);
745 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
746
747 s->refcount_table = qemu_mallocz(s->cluster_size);
748
749 s->refcount_table_offset = offset;
750 header.refcount_table_offset = cpu_to_be64(offset);
751 header.refcount_table_clusters = cpu_to_be32(1);
752 offset += s->cluster_size;
753 s->refcount_block_offset = offset;
754
755 /* count how many refcount blocks needed */
756 tmp = offset >> s->cluster_bits;
757 ref_clusters = (tmp >> (s->cluster_bits - REFCOUNT_SHIFT)) + 1;
758 for (i=0; i < ref_clusters; i++) {
759 s->refcount_table[i] = cpu_to_be64(offset);
760 offset += s->cluster_size;
761 }
762
763 s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
764
765 /* update refcounts */
766 qcow2_create_refcount_update(s, 0, header_size);
767 qcow2_create_refcount_update(s, s->l1_table_offset,
768 l1_size * sizeof(uint64_t));
769 qcow2_create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
770 qcow2_create_refcount_update(s, s->refcount_block_offset,
771 ref_clusters * s->cluster_size);
772
773 /* write all the data */
774 write(fd, &header, sizeof(header));
775 if (backing_file) {
776 if (backing_format_len) {
777 char zero[16];
778 int d = ext_bf.len - backing_format_len;
779
780 memset(zero, 0, sizeof(zero));
781 cpu_to_be32s(&ext_bf.magic);
782 cpu_to_be32s(&ext_bf.len);
783 write(fd, &ext_bf, sizeof(ext_bf));
784 write(fd, backing_format, backing_format_len);
785 if (d>0) {
786 write(fd, zero, d);
787 }
788 }
789 write(fd, backing_file, backing_filename_len);
790 }
791 lseek(fd, s->l1_table_offset, SEEK_SET);
792 tmp = 0;
793 for(i = 0;i < l1_size; i++) {
794 write(fd, &tmp, sizeof(tmp));
795 }
796 lseek(fd, s->refcount_table_offset, SEEK_SET);
797 write(fd, s->refcount_table, s->cluster_size);
798
799 lseek(fd, s->refcount_block_offset, SEEK_SET);
800 write(fd, s->refcount_block, ref_clusters * s->cluster_size);
801
802 qemu_free(s->refcount_table);
803 qemu_free(s->refcount_block);
804 close(fd);
805 return 0;
806 }
807
808 static int qcow_create(const char *filename, QEMUOptionParameter *options)
809 {
810 const char *backing_file = NULL;
811 const char *backing_fmt = NULL;
812 uint64_t sectors = 0;
813 int flags = 0;
814 size_t cluster_size = 65536;
815
816 /* Read out options */
817 while (options && options->name) {
818 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
819 sectors = options->value.n / 512;
820 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
821 backing_file = options->value.s;
822 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
823 backing_fmt = options->value.s;
824 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
825 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
826 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
827 if (options->value.n) {
828 cluster_size = options->value.n;
829 }
830 }
831 options++;
832 }
833
834 return qcow_create2(filename, sectors, backing_file, backing_fmt, flags,
835 cluster_size);
836 }
837
838 static int qcow_make_empty(BlockDriverState *bs)
839 {
840 #if 0
841 /* XXX: not correct */
842 BDRVQcowState *s = bs->opaque;
843 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
844 int ret;
845
846 memset(s->l1_table, 0, l1_length);
847 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
848 return -1;
849 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
850 if (ret < 0)
851 return ret;
852
853 l2_cache_reset(bs);
854 #endif
855 return 0;
856 }
857
858 static int qcow2_write(BlockDriverState *bs, int64_t sector_num,
859 const uint8_t *buf, int nb_sectors)
860 {
861 BDRVQcowState *s = bs->opaque;
862 int ret, index_in_cluster, n;
863 uint64_t cluster_offset;
864 int n_end;
865 QCowL2Meta l2meta;
866
867 while (nb_sectors > 0) {
868 memset(&l2meta, 0, sizeof(l2meta));
869
870 index_in_cluster = sector_num & (s->cluster_sectors - 1);
871 n_end = index_in_cluster + nb_sectors;
872 if (s->crypt_method &&
873 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
874 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
875 cluster_offset = qcow2_alloc_cluster_offset(bs, sector_num << 9,
876 index_in_cluster,
877 n_end, &n, &l2meta);
878 if (!cluster_offset)
879 return -1;
880 if (s->crypt_method) {
881 qcow2_encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
882 &s->aes_encrypt_key);
883 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
884 s->cluster_data, n * 512);
885 } else {
886 ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
887 }
888 if (ret != n * 512 || qcow2_alloc_cluster_link_l2(bs, cluster_offset, &l2meta) < 0) {
889 qcow2_free_any_clusters(bs, cluster_offset, l2meta.nb_clusters);
890 return -1;
891 }
892 nb_sectors -= n;
893 sector_num += n;
894 buf += n * 512;
895 if (l2meta.nb_clusters != 0) {
896 LIST_REMOVE(&l2meta, next_in_flight);
897 }
898 }
899 s->cluster_cache_offset = -1; /* disable compressed cache */
900 return 0;
901 }
902
903 /* XXX: put compressed sectors first, then all the cluster aligned
904 tables to avoid losing bytes in alignment */
905 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
906 const uint8_t *buf, int nb_sectors)
907 {
908 BDRVQcowState *s = bs->opaque;
909 z_stream strm;
910 int ret, out_len;
911 uint8_t *out_buf;
912 uint64_t cluster_offset;
913
914 if (nb_sectors == 0) {
915 /* align end of file to a sector boundary to ease reading with
916 sector based I/Os */
917 cluster_offset = bdrv_getlength(s->hd);
918 cluster_offset = (cluster_offset + 511) & ~511;
919 bdrv_truncate(s->hd, cluster_offset);
920 return 0;
921 }
922
923 if (nb_sectors != s->cluster_sectors)
924 return -EINVAL;
925
926 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
927
928 /* best compression, small window, no zlib header */
929 memset(&strm, 0, sizeof(strm));
930 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
931 Z_DEFLATED, -12,
932 9, Z_DEFAULT_STRATEGY);
933 if (ret != 0) {
934 qemu_free(out_buf);
935 return -1;
936 }
937
938 strm.avail_in = s->cluster_size;
939 strm.next_in = (uint8_t *)buf;
940 strm.avail_out = s->cluster_size;
941 strm.next_out = out_buf;
942
943 ret = deflate(&strm, Z_FINISH);
944 if (ret != Z_STREAM_END && ret != Z_OK) {
945 qemu_free(out_buf);
946 deflateEnd(&strm);
947 return -1;
948 }
949 out_len = strm.next_out - out_buf;
950
951 deflateEnd(&strm);
952
953 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
954 /* could not compress: write normal cluster */
955 bdrv_write(bs, sector_num, buf, s->cluster_sectors);
956 } else {
957 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
958 sector_num << 9, out_len);
959 if (!cluster_offset)
960 return -1;
961 cluster_offset &= s->cluster_offset_mask;
962 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
963 qemu_free(out_buf);
964 return -1;
965 }
966 }
967
968 qemu_free(out_buf);
969 return 0;
970 }
971
972 static void qcow_flush(BlockDriverState *bs)
973 {
974 BDRVQcowState *s = bs->opaque;
975 bdrv_flush(s->hd);
976 }
977
978 static int64_t qcow_vm_state_offset(BDRVQcowState *s)
979 {
980 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
981 }
982
983 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
984 {
985 BDRVQcowState *s = bs->opaque;
986 bdi->cluster_size = s->cluster_size;
987 bdi->vm_state_offset = qcow_vm_state_offset(s);
988 return 0;
989 }
990
991
992 static int qcow_check(BlockDriverState *bs)
993 {
994 return qcow2_check_refcounts(bs);
995 }
996
997 #if 0
998 static void dump_refcounts(BlockDriverState *bs)
999 {
1000 BDRVQcowState *s = bs->opaque;
1001 int64_t nb_clusters, k, k1, size;
1002 int refcount;
1003
1004 size = bdrv_getlength(s->hd);
1005 nb_clusters = size_to_clusters(s, size);
1006 for(k = 0; k < nb_clusters;) {
1007 k1 = k;
1008 refcount = get_refcount(bs, k);
1009 k++;
1010 while (k < nb_clusters && get_refcount(bs, k) == refcount)
1011 k++;
1012 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
1013 }
1014 }
1015 #endif
1016
1017 static int qcow_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1018 int64_t pos, int size)
1019 {
1020 BDRVQcowState *s = bs->opaque;
1021 int growable = bs->growable;
1022
1023 bs->growable = 1;
1024 bdrv_pwrite(bs, qcow_vm_state_offset(s) + pos, buf, size);
1025 bs->growable = growable;
1026
1027 return size;
1028 }
1029
1030 static int qcow_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1031 int64_t pos, int size)
1032 {
1033 BDRVQcowState *s = bs->opaque;
1034 int growable = bs->growable;
1035 int ret;
1036
1037 bs->growable = 1;
1038 ret = bdrv_pread(bs, qcow_vm_state_offset(s) + pos, buf, size);
1039 bs->growable = growable;
1040
1041 return ret;
1042 }
1043
1044 static QEMUOptionParameter qcow_create_options[] = {
1045 {
1046 .name = BLOCK_OPT_SIZE,
1047 .type = OPT_SIZE,
1048 .help = "Virtual disk size"
1049 },
1050 {
1051 .name = BLOCK_OPT_BACKING_FILE,
1052 .type = OPT_STRING,
1053 .help = "File name of a base image"
1054 },
1055 {
1056 .name = BLOCK_OPT_BACKING_FMT,
1057 .type = OPT_STRING,
1058 .help = "Image format of the base image"
1059 },
1060 {
1061 .name = BLOCK_OPT_ENCRYPT,
1062 .type = OPT_FLAG,
1063 .help = "Encrypt the image"
1064 },
1065 {
1066 .name = BLOCK_OPT_CLUSTER_SIZE,
1067 .type = OPT_SIZE,
1068 .help = "qcow2 cluster size"
1069 },
1070 { NULL }
1071 };
1072
1073 static BlockDriver bdrv_qcow2 = {
1074 .format_name = "qcow2",
1075 .instance_size = sizeof(BDRVQcowState),
1076 .bdrv_probe = qcow_probe,
1077 .bdrv_open = qcow_open,
1078 .bdrv_close = qcow_close,
1079 .bdrv_create = qcow_create,
1080 .bdrv_flush = qcow_flush,
1081 .bdrv_is_allocated = qcow_is_allocated,
1082 .bdrv_set_key = qcow_set_key,
1083 .bdrv_make_empty = qcow_make_empty,
1084
1085 .bdrv_read = qcow2_read,
1086 .bdrv_write = qcow2_write,
1087 .bdrv_aio_readv = qcow_aio_readv,
1088 .bdrv_aio_writev = qcow_aio_writev,
1089 .bdrv_write_compressed = qcow_write_compressed,
1090
1091 .bdrv_snapshot_create = qcow2_snapshot_create,
1092 .bdrv_snapshot_goto = qcow2_snapshot_goto,
1093 .bdrv_snapshot_delete = qcow2_snapshot_delete,
1094 .bdrv_snapshot_list = qcow2_snapshot_list,
1095 .bdrv_get_info = qcow_get_info,
1096
1097 .bdrv_save_vmstate = qcow_save_vmstate,
1098 .bdrv_load_vmstate = qcow_load_vmstate,
1099
1100 .create_options = qcow_create_options,
1101 .bdrv_check = qcow_check,
1102 };
1103
1104 static void bdrv_qcow2_init(void)
1105 {
1106 bdrv_register(&bdrv_qcow2);
1107 }
1108
1109 block_init(bdrv_qcow2_init);
Patches shipped by Fedora