?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Provide marshalling mechanism for json
[qemu/aliguori-queue.git] / block / qcow2.c
blob3954cf13984c3a0c9df0143132fd8369f33c7e36
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.cluster_bits < MIN_CLUSTER_BITS ||
170 header.cluster_bits > MAX_CLUSTER_BITS)
171 goto fail;
172 if (header.crypt_method > QCOW_CRYPT_AES)
173 goto fail;
174 s->crypt_method_header = header.crypt_method;
175 if (s->crypt_method_header)
176 bs->encrypted = 1;
177 s->cluster_bits = header.cluster_bits;
178 s->cluster_size = 1 << s->cluster_bits;
179 s->cluster_sectors = 1 << (s->cluster_bits - 9);
180 s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
181 s->l2_size = 1 << s->l2_bits;
182 bs->total_sectors = header.size / 512;
183 s->csize_shift = (62 - (s->cluster_bits - 8));
184 s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
185 s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
186 s->refcount_table_offset = header.refcount_table_offset;
187 s->refcount_table_size =
188 header.refcount_table_clusters << (s->cluster_bits - 3);
189
190 s->snapshots_offset = header.snapshots_offset;
191 s->nb_snapshots = header.nb_snapshots;
192
193 /* read the level 1 table */
194 s->l1_size = header.l1_size;
195 shift = s->cluster_bits + s->l2_bits;
196 s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
197 /* the L1 table must contain at least enough entries to put
198 header.size bytes */
199 if (s->l1_size < s->l1_vm_state_index)
200 goto fail;
201 s->l1_table_offset = header.l1_table_offset;
202 if (s->l1_size > 0) {
203 s->l1_table = qemu_mallocz(
204 align_offset(s->l1_size * sizeof(uint64_t), 512));
205 if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
206 s->l1_size * sizeof(uint64_t))
207 goto fail;
208 for(i = 0;i < s->l1_size; i++) {
209 be64_to_cpus(&s->l1_table[i]);
210 }
211 }
212 /* alloc L2 cache */
213 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
214 s->cluster_cache = qemu_malloc(s->cluster_size);
215 /* one more sector for decompressed data alignment */
216 s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
217 + 512);
218 s->cluster_cache_offset = -1;
219
220 if (qcow2_refcount_init(bs) < 0)
221 goto fail;
222
223 QLIST_INIT(&s->cluster_allocs);
224
225 /* read qcow2 extensions */
226 if (header.backing_file_offset)
227 ext_end = header.backing_file_offset;
228 else
229 ext_end = s->cluster_size;
230 if (qcow_read_extensions(bs, sizeof(header), ext_end))
231 goto fail;
232
233 /* read the backing file name */
234 if (header.backing_file_offset != 0) {
235 len = header.backing_file_size;
236 if (len > 1023)
237 len = 1023;
238 if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
239 goto fail;
240 bs->backing_file[len] = '\0';
241 }
242 if (qcow2_read_snapshots(bs) < 0)
243 goto fail;
244
245 #ifdef DEBUG_ALLOC
246 qcow2_check_refcounts(bs);
247 #endif
248 return 0;
249
250 fail:
251 qcow2_free_snapshots(bs);
252 qcow2_refcount_close(bs);
253 qemu_free(s->l1_table);
254 qemu_free(s->l2_cache);
255 qemu_free(s->cluster_cache);
256 qemu_free(s->cluster_data);
257 bdrv_delete(s->hd);
258 return -1;
259 }
260
261 static int qcow_set_key(BlockDriverState *bs, const char *key)
262 {
263 BDRVQcowState *s = bs->opaque;
264 uint8_t keybuf[16];
265 int len, i;
266
267 memset(keybuf, 0, 16);
268 len = strlen(key);
269 if (len > 16)
270 len = 16;
271 /* XXX: we could compress the chars to 7 bits to increase
272 entropy */
273 for(i = 0;i < len;i++) {
274 keybuf[i] = key[i];
275 }
276 s->crypt_method = s->crypt_method_header;
277
278 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
279 return -1;
280 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
281 return -1;
282 #if 0
283 /* test */
284 {
285 uint8_t in[16];
286 uint8_t out[16];
287 uint8_t tmp[16];
288 for(i=0;i<16;i++)
289 in[i] = i;
290 AES_encrypt(in, tmp, &s->aes_encrypt_key);
291 AES_decrypt(tmp, out, &s->aes_decrypt_key);
292 for(i = 0; i < 16; i++)
293 printf(" %02x", tmp[i]);
294 printf("\n");
295 for(i = 0; i < 16; i++)
296 printf(" %02x", out[i]);
297 printf("\n");
298 }
299 #endif
300 return 0;
301 }
302
303 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
304 int nb_sectors, int *pnum)
305 {
306 uint64_t cluster_offset;
307
308 *pnum = nb_sectors;
309 cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, pnum);
310
311 return (cluster_offset != 0);
312 }
313
314 /* handle reading after the end of the backing file */
315 int qcow2_backing_read1(BlockDriverState *bs,
316 int64_t sector_num, uint8_t *buf, int nb_sectors)
317 {
318 int n1;
319 if ((sector_num + nb_sectors) <= bs->total_sectors)
320 return nb_sectors;
321 if (sector_num >= bs->total_sectors)
322 n1 = 0;
323 else
324 n1 = bs->total_sectors - sector_num;
325 memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
326 return n1;
327 }
328
329 typedef struct QCowAIOCB {
330 BlockDriverAIOCB common;
331 int64_t sector_num;
332 QEMUIOVector *qiov;
333 uint8_t *buf;
334 void *orig_buf;
335 int nb_sectors;
336 int n;
337 uint64_t cluster_offset;
338 uint8_t *cluster_data;
339 BlockDriverAIOCB *hd_aiocb;
340 struct iovec hd_iov;
341 QEMUIOVector hd_qiov;
342 QEMUBH *bh;
343 QCowL2Meta l2meta;
344 QLIST_ENTRY(QCowAIOCB) next_depend;
345 } QCowAIOCB;
346
347 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
348 {
349 QCowAIOCB *acb = (QCowAIOCB *)blockacb;
350 if (acb->hd_aiocb)
351 bdrv_aio_cancel(acb->hd_aiocb);
352 qemu_aio_release(acb);
353 }
354
355 static AIOPool qcow_aio_pool = {
356 .aiocb_size = sizeof(QCowAIOCB),
357 .cancel = qcow_aio_cancel,
358 };
359
360 static void qcow_aio_read_cb(void *opaque, int ret);
361 static void qcow_aio_read_bh(void *opaque)
362 {
363 QCowAIOCB *acb = opaque;
364 qemu_bh_delete(acb->bh);
365 acb->bh = NULL;
366 qcow_aio_read_cb(opaque, 0);
367 }
368
369 static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb)
370 {
371 if (acb->bh)
372 return -EIO;
373
374 acb->bh = qemu_bh_new(cb, acb);
375 if (!acb->bh)
376 return -EIO;
377
378 qemu_bh_schedule(acb->bh);
379
380 return 0;
381 }
382
383 static void qcow_aio_read_cb(void *opaque, int ret)
384 {
385 QCowAIOCB *acb = opaque;
386 BlockDriverState *bs = acb->common.bs;
387 BDRVQcowState *s = bs->opaque;
388 int index_in_cluster, n1;
389
390 acb->hd_aiocb = NULL;
391 if (ret < 0)
392 goto done;
393
394 /* post process the read buffer */
395 if (!acb->cluster_offset) {
396 /* nothing to do */
397 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
398 /* nothing to do */
399 } else {
400 if (s->crypt_method) {
401 qcow2_encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
402 acb->n, 0,
403 &s->aes_decrypt_key);
404 }
405 }
406
407 acb->nb_sectors -= acb->n;
408 acb->sector_num += acb->n;
409 acb->buf += acb->n * 512;
410
411 if (acb->nb_sectors == 0) {
412 /* request completed */
413 ret = 0;
414 goto done;
415 }
416
417 /* prepare next AIO request */
418 acb->n = acb->nb_sectors;
419 acb->cluster_offset =
420 qcow2_get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
421 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
422
423 if (!acb->cluster_offset) {
424 if (bs->backing_hd) {
425 /* read from the base image */
426 n1 = qcow2_backing_read1(bs->backing_hd, acb->sector_num,
427 acb->buf, acb->n);
428 if (n1 > 0) {
429 acb->hd_iov.iov_base = (void *)acb->buf;
430 acb->hd_iov.iov_len = acb->n * 512;
431 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
432 acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num,
433 &acb->hd_qiov, acb->n,
434 qcow_aio_read_cb, acb);
435 if (acb->hd_aiocb == NULL)
436 goto done;
437 } else {
438 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
439 if (ret < 0)
440 goto done;
441 }
442 } else {
443 /* Note: in this case, no need to wait */
444 memset(acb->buf, 0, 512 * acb->n);
445 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
446 if (ret < 0)
447 goto done;
448 }
449 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
450 /* add AIO support for compressed blocks ? */
451 if (qcow2_decompress_cluster(s, acb->cluster_offset) < 0)
452 goto done;
453 memcpy(acb->buf,
454 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
455 ret = qcow_schedule_bh(qcow_aio_read_bh, acb);
456 if (ret < 0)
457 goto done;
458 } else {
459 if ((acb->cluster_offset & 511) != 0) {
460 ret = -EIO;
461 goto done;
462 }
463
464 acb->hd_iov.iov_base = (void *)acb->buf;
465 acb->hd_iov.iov_len = acb->n * 512;
466 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
467 acb->hd_aiocb = bdrv_aio_readv(s->hd,
468 (acb->cluster_offset >> 9) + index_in_cluster,
469 &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb);
470 if (acb->hd_aiocb == NULL)
471 goto done;
472 }
473
474 return;
475 done:
476 if (acb->qiov->niov > 1) {
477 qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size);
478 qemu_vfree(acb->orig_buf);
479 }
480 acb->common.cb(acb->common.opaque, ret);
481 qemu_aio_release(acb);
482 }
483
484 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
485 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
486 BlockDriverCompletionFunc *cb, void *opaque, int is_write)
487 {
488 QCowAIOCB *acb;
489
490 acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque);
491 if (!acb)
492 return NULL;
493 acb->hd_aiocb = NULL;
494 acb->sector_num = sector_num;
495 acb->qiov = qiov;
496 if (qiov->niov > 1) {
497 acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size);
498 if (is_write)
499 qemu_iovec_to_buffer(qiov, acb->buf);
500 } else {
501 acb->buf = (uint8_t *)qiov->iov->iov_base;
502 }
503 acb->nb_sectors = nb_sectors;
504 acb->n = 0;
505 acb->cluster_offset = 0;
506 acb->l2meta.nb_clusters = 0;
507 QLIST_INIT(&acb->l2meta.dependent_requests);
508 return acb;
509 }
510
511 static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs,
512 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
513 BlockDriverCompletionFunc *cb, void *opaque)
514 {
515 QCowAIOCB *acb;
516
517 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
518 if (!acb)
519 return NULL;
520
521 qcow_aio_read_cb(acb, 0);
522 return &acb->common;
523 }
524
525 static void qcow_aio_write_cb(void *opaque, int ret);
526
527 static void run_dependent_requests(QCowL2Meta *m)
528 {
529 QCowAIOCB *req;
530 QCowAIOCB *next;
531
532 /* Take the request off the list of running requests */
533 if (m->nb_clusters != 0) {
534 QLIST_REMOVE(m, next_in_flight);
535 }
536
537 /*
538 * Restart all dependent requests.
539 * Can't use QLIST_FOREACH here - the next link might not be the same
540 * any more after the callback (request could depend on a different
541 * request now)
542 */
543 for (req = m->dependent_requests.lh_first; req != NULL; req = next) {
544 next = req->next_depend.le_next;
545 qcow_aio_write_cb(req, 0);
546 }
547
548 /* Empty the list for the next part of the request */
549 QLIST_INIT(&m->dependent_requests);
550 }
551
552 static void qcow_aio_write_cb(void *opaque, int ret)
553 {
554 QCowAIOCB *acb = opaque;
555 BlockDriverState *bs = acb->common.bs;
556 BDRVQcowState *s = bs->opaque;
557 int index_in_cluster;
558 const uint8_t *src_buf;
559 int n_end;
560
561 acb->hd_aiocb = NULL;
562
563 if (ret >= 0) {
564 ret = qcow2_alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta);
565 }
566
567 run_dependent_requests(&acb->l2meta);
568
569 if (ret < 0)
570 goto done;
571
572 acb->nb_sectors -= acb->n;
573 acb->sector_num += acb->n;
574 acb->buf += acb->n * 512;
575
576 if (acb->nb_sectors == 0) {
577 /* request completed */
578 ret = 0;
579 goto done;
580 }
581
582 index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
583 n_end = index_in_cluster + acb->nb_sectors;
584 if (s->crypt_method &&
585 n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
586 n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
587
588 acb->cluster_offset = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9,
589 index_in_cluster,
590 n_end, &acb->n, &acb->l2meta);
591
592 /* Need to wait for another request? If so, we are done for now. */
593 if (!acb->cluster_offset && acb->l2meta.depends_on != NULL) {
594 QLIST_INSERT_HEAD(&acb->l2meta.depends_on->dependent_requests,
595 acb, next_depend);
596 return;
597 }
598
599 if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) {
600 ret = -EIO;
601 goto done;
602 }
603 if (s->crypt_method) {
604 if (!acb->cluster_data) {
605 acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
606 s->cluster_size);
607 }
608 qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
609 acb->n, 1, &s->aes_encrypt_key);
610 src_buf = acb->cluster_data;
611 } else {
612 src_buf = acb->buf;
613 }
614 acb->hd_iov.iov_base = (void *)src_buf;
615 acb->hd_iov.iov_len = acb->n * 512;
616 qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1);
617 acb->hd_aiocb = bdrv_aio_writev(s->hd,
618 (acb->cluster_offset >> 9) + index_in_cluster,
619 &acb->hd_qiov, acb->n,
620 qcow_aio_write_cb, acb);
621 if (acb->hd_aiocb == NULL)
622 goto done;
623
624 return;
625
626 done:
627 if (acb->qiov->niov > 1)
628 qemu_vfree(acb->orig_buf);
629 acb->common.cb(acb->common.opaque, ret);
630 qemu_aio_release(acb);
631 }
632
633 static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs,
634 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
635 BlockDriverCompletionFunc *cb, void *opaque)
636 {
637 BDRVQcowState *s = bs->opaque;
638 QCowAIOCB *acb;
639
640 s->cluster_cache_offset = -1; /* disable compressed cache */
641
642 acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
643 if (!acb)
644 return NULL;
645
646 qcow_aio_write_cb(acb, 0);
647 return &acb->common;
648 }
649
650 static void qcow_close(BlockDriverState *bs)
651 {
652 BDRVQcowState *s = bs->opaque;
653 qemu_free(s->l1_table);
654 qemu_free(s->l2_cache);
655 qemu_free(s->cluster_cache);
656 qemu_free(s->cluster_data);
657 qcow2_refcount_close(bs);
658 bdrv_delete(s->hd);
659 }
660
661 static int get_bits_from_size(size_t size)
662 {
663 int res = 0;
664
665 if (size == 0) {
666 return -1;
667 }
668
669 while (size != 1) {
670 /* Not a power of two */
671 if (size & 1) {
672 return -1;
673 }
674
675 size >>= 1;
676 res++;
677 }
678
679 return res;
680 }
681
682
683 static int preallocate(BlockDriverState *bs)
684 {
685 BDRVQcowState *s = bs->opaque;
686 uint64_t cluster_offset = 0;
687 uint64_t nb_sectors;
688 uint64_t offset;
689 int num;
690 QCowL2Meta meta;
691
692 nb_sectors = bdrv_getlength(bs) >> 9;
693 offset = 0;
694 QLIST_INIT(&meta.dependent_requests);
695
696 while (nb_sectors) {
697 num = MIN(nb_sectors, INT_MAX >> 9);
698 cluster_offset = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num,
699 &meta);
700
701 if (cluster_offset == 0) {
702 return -1;
703 }
704
705 if (qcow2_alloc_cluster_link_l2(bs, cluster_offset, &meta) < 0) {
706 qcow2_free_any_clusters(bs, cluster_offset, meta.nb_clusters);
707 return -1;
708 }
709
710 /* There are no dependent requests, but we need to remove our request
711 * from the list of in-flight requests */
712 run_dependent_requests(&meta);
713
714 /* TODO Preallocate data if requested */
715
716 nb_sectors -= num;
717 offset += num << 9;
718 }
719
720 /*
721 * It is expected that the image file is large enough to actually contain
722 * all of the allocated clusters (otherwise we get failing reads after
723 * EOF). Extend the image to the last allocated sector.
724 */
725 if (cluster_offset != 0) {
726 uint8_t buf[512];
727 memset(buf, 0, 512);
728 bdrv_write(s->hd, (cluster_offset >> 9) + num - 1, buf, 1);
729 }
730
731 return 0;
732 }
733
734 static int qcow_create2(const char *filename, int64_t total_size,
735 const char *backing_file, const char *backing_format,
736 int flags, size_t cluster_size, int prealloc)
737 {
738
739 int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
740 int ref_clusters, backing_format_len = 0;
741 QCowHeader header;
742 uint64_t tmp, offset;
743 QCowCreateState s1, *s = &s1;
744 QCowExtension ext_bf = {0, 0};
745
746
747 memset(s, 0, sizeof(*s));
748
749 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
750 if (fd < 0)
751 return -1;
752 memset(&header, 0, sizeof(header));
753 header.magic = cpu_to_be32(QCOW_MAGIC);
754 header.version = cpu_to_be32(QCOW_VERSION);
755 header.size = cpu_to_be64(total_size * 512);
756 header_size = sizeof(header);
757 backing_filename_len = 0;
758 if (backing_file) {
759 if (backing_format) {
760 ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT;
761 backing_format_len = strlen(backing_format);
762 ext_bf.len = (backing_format_len + 7) & ~7;
763 header_size += ((sizeof(ext_bf) + ext_bf.len + 7) & ~7);
764 }
765 header.backing_file_offset = cpu_to_be64(header_size);
766 backing_filename_len = strlen(backing_file);
767 header.backing_file_size = cpu_to_be32(backing_filename_len);
768 header_size += backing_filename_len;
769 }
770
771 /* Cluster size */
772 s->cluster_bits = get_bits_from_size(cluster_size);
773 if (s->cluster_bits < MIN_CLUSTER_BITS ||
774 s->cluster_bits > MAX_CLUSTER_BITS)
775 {
776 fprintf(stderr, "Cluster size must be a power of two between "
777 "%d and %dk\n",
778 1 << MIN_CLUSTER_BITS,
779 1 << (MAX_CLUSTER_BITS - 10));
780 return -EINVAL;
781 }
782 s->cluster_size = 1 << s->cluster_bits;
783
784 header.cluster_bits = cpu_to_be32(s->cluster_bits);
785 header_size = (header_size + 7) & ~7;
786 if (flags & BLOCK_FLAG_ENCRYPT) {
787 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
788 } else {
789 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
790 }
791 l2_bits = s->cluster_bits - 3;
792 shift = s->cluster_bits + l2_bits;
793 l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
794 offset = align_offset(header_size, s->cluster_size);
795 s->l1_table_offset = offset;
796 header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
797 header.l1_size = cpu_to_be32(l1_size);
798 offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
799
800 s->refcount_table = qemu_mallocz(s->cluster_size);
801
802 s->refcount_table_offset = offset;
803 header.refcount_table_offset = cpu_to_be64(offset);
804 header.refcount_table_clusters = cpu_to_be32(1);
805 offset += s->cluster_size;
806 s->refcount_block_offset = offset;
807
808 /* count how many refcount blocks needed */
809 tmp = offset >> s->cluster_bits;
810 ref_clusters = (tmp >> (s->cluster_bits - REFCOUNT_SHIFT)) + 1;
811 for (i=0; i < ref_clusters; i++) {
812 s->refcount_table[i] = cpu_to_be64(offset);
813 offset += s->cluster_size;
814 }
815
816 s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size);
817
818 /* update refcounts */
819 qcow2_create_refcount_update(s, 0, header_size);
820 qcow2_create_refcount_update(s, s->l1_table_offset,
821 l1_size * sizeof(uint64_t));
822 qcow2_create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
823 qcow2_create_refcount_update(s, s->refcount_block_offset,
824 ref_clusters * s->cluster_size);
825
826 /* write all the data */
827 write(fd, &header, sizeof(header));
828 if (backing_file) {
829 if (backing_format_len) {
830 char zero[16];
831 int d = ext_bf.len - backing_format_len;
832
833 memset(zero, 0, sizeof(zero));
834 cpu_to_be32s(&ext_bf.magic);
835 cpu_to_be32s(&ext_bf.len);
836 write(fd, &ext_bf, sizeof(ext_bf));
837 write(fd, backing_format, backing_format_len);
838 if (d>0) {
839 write(fd, zero, d);
840 }
841 }
842 write(fd, backing_file, backing_filename_len);
843 }
844 lseek(fd, s->l1_table_offset, SEEK_SET);
845 tmp = 0;
846 for(i = 0;i < l1_size; i++) {
847 write(fd, &tmp, sizeof(tmp));
848 }
849 lseek(fd, s->refcount_table_offset, SEEK_SET);
850 write(fd, s->refcount_table, s->cluster_size);
851
852 lseek(fd, s->refcount_block_offset, SEEK_SET);
853 write(fd, s->refcount_block, ref_clusters * s->cluster_size);
854
855 qemu_free(s->refcount_table);
856 qemu_free(s->refcount_block);
857 close(fd);
858
859 /* Preallocate metadata */
860 if (prealloc) {
861 BlockDriverState *bs;
862 bs = bdrv_new("");
863 bdrv_open(bs, filename, BDRV_O_CACHE_WB);
864 preallocate(bs);
865 bdrv_close(bs);
866 }
867
868 return 0;
869 }
870
871 static int qcow_create(const char *filename, QEMUOptionParameter *options)
872 {
873 const char *backing_file = NULL;
874 const char *backing_fmt = NULL;
875 uint64_t sectors = 0;
876 int flags = 0;
877 size_t cluster_size = 65536;
878 int prealloc = 0;
879
880 /* Read out options */
881 while (options && options->name) {
882 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
883 sectors = options->value.n / 512;
884 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
885 backing_file = options->value.s;
886 } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) {
887 backing_fmt = options->value.s;
888 } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) {
889 flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
890 } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
891 if (options->value.n) {
892 cluster_size = options->value.n;
893 }
894 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
895 if (!options->value.s || !strcmp(options->value.s, "off")) {
896 prealloc = 0;
897 } else if (!strcmp(options->value.s, "metadata")) {
898 prealloc = 1;
899 } else {
900 fprintf(stderr, "Invalid preallocation mode: '%s'\n",
901 options->value.s);
902 return -EINVAL;
903 }
904 }
905 options++;
906 }
907
908 if (backing_file && prealloc) {
909 fprintf(stderr, "Backing file and preallocation cannot be used at "
910 "the same time\n");
911 return -EINVAL;
912 }
913
914 return qcow_create2(filename, sectors, backing_file, backing_fmt, flags,
915 cluster_size, prealloc);
916 }
917
918 static int qcow_make_empty(BlockDriverState *bs)
919 {
920 #if 0
921 /* XXX: not correct */
922 BDRVQcowState *s = bs->opaque;
923 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
924 int ret;
925
926 memset(s->l1_table, 0, l1_length);
927 if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
928 return -1;
929 ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
930 if (ret < 0)
931 return ret;
932
933 l2_cache_reset(bs);
934 #endif
935 return 0;
936 }
937
938 /* XXX: put compressed sectors first, then all the cluster aligned
939 tables to avoid losing bytes in alignment */
940 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
941 const uint8_t *buf, int nb_sectors)
942 {
943 BDRVQcowState *s = bs->opaque;
944 z_stream strm;
945 int ret, out_len;
946 uint8_t *out_buf;
947 uint64_t cluster_offset;
948
949 if (nb_sectors == 0) {
950 /* align end of file to a sector boundary to ease reading with
951 sector based I/Os */
952 cluster_offset = bdrv_getlength(s->hd);
953 cluster_offset = (cluster_offset + 511) & ~511;
954 bdrv_truncate(s->hd, cluster_offset);
955 return 0;
956 }
957
958 if (nb_sectors != s->cluster_sectors)
959 return -EINVAL;
960
961 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
962
963 /* best compression, small window, no zlib header */
964 memset(&strm, 0, sizeof(strm));
965 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
966 Z_DEFLATED, -12,
967 9, Z_DEFAULT_STRATEGY);
968 if (ret != 0) {
969 qemu_free(out_buf);
970 return -1;
971 }
972
973 strm.avail_in = s->cluster_size;
974 strm.next_in = (uint8_t *)buf;
975 strm.avail_out = s->cluster_size;
976 strm.next_out = out_buf;
977
978 ret = deflate(&strm, Z_FINISH);
979 if (ret != Z_STREAM_END && ret != Z_OK) {
980 qemu_free(out_buf);
981 deflateEnd(&strm);
982 return -1;
983 }
984 out_len = strm.next_out - out_buf;
985
986 deflateEnd(&strm);
987
988 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
989 /* could not compress: write normal cluster */
990 bdrv_write(bs, sector_num, buf, s->cluster_sectors);
991 } else {
992 cluster_offset = qcow2_alloc_compressed_cluster_offset(bs,
993 sector_num << 9, out_len);
994 if (!cluster_offset)
995 return -1;
996 cluster_offset &= s->cluster_offset_mask;
997 if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
998 qemu_free(out_buf);
999 return -1;
1000 }
1001 }
1002
1003 qemu_free(out_buf);
1004 return 0;
1005 }
1006
1007 static void qcow_flush(BlockDriverState *bs)
1008 {
1009 BDRVQcowState *s = bs->opaque;
1010 bdrv_flush(s->hd);
1011 }
1012
1013 static int64_t qcow_vm_state_offset(BDRVQcowState *s)
1014 {
1015 return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
1016 }
1017
1018 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1019 {
1020 BDRVQcowState *s = bs->opaque;
1021 bdi->cluster_size = s->cluster_size;
1022 bdi->vm_state_offset = qcow_vm_state_offset(s);
1023 return 0;
1024 }
1025
1026
1027 static int qcow_check(BlockDriverState *bs)
1028 {
1029 return qcow2_check_refcounts(bs);
1030 }
1031
1032 #if 0
1033 static void dump_refcounts(BlockDriverState *bs)
1034 {
1035 BDRVQcowState *s = bs->opaque;
1036 int64_t nb_clusters, k, k1, size;
1037 int refcount;
1038
1039 size = bdrv_getlength(s->hd);
1040 nb_clusters = size_to_clusters(s, size);
1041 for(k = 0; k < nb_clusters;) {
1042 k1 = k;
1043 refcount = get_refcount(bs, k);
1044 k++;
1045 while (k < nb_clusters && get_refcount(bs, k) == refcount)
1046 k++;
1047 printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
1048 }
1049 }
1050 #endif
1051
1052 static int qcow_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1053 int64_t pos, int size)
1054 {
1055 BDRVQcowState *s = bs->opaque;
1056 int growable = bs->growable;
1057
1058 bs->growable = 1;
1059 bdrv_pwrite(bs, qcow_vm_state_offset(s) + pos, buf, size);
1060 bs->growable = growable;
1061
1062 return size;
1063 }
1064
1065 static int qcow_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1066 int64_t pos, int size)
1067 {
1068 BDRVQcowState *s = bs->opaque;
1069 int growable = bs->growable;
1070 int ret;
1071
1072 bs->growable = 1;
1073 ret = bdrv_pread(bs, qcow_vm_state_offset(s) + pos, buf, size);
1074 bs->growable = growable;
1075
1076 return ret;
1077 }
1078
1079 static QEMUOptionParameter qcow_create_options[] = {
1080 {
1081 .name = BLOCK_OPT_SIZE,
1082 .type = OPT_SIZE,
1083 .help = "Virtual disk size"
1084 },
1085 {
1086 .name = BLOCK_OPT_BACKING_FILE,
1087 .type = OPT_STRING,
1088 .help = "File name of a base image"
1089 },
1090 {
1091 .name = BLOCK_OPT_BACKING_FMT,
1092 .type = OPT_STRING,
1093 .help = "Image format of the base image"
1094 },
1095 {
1096 .name = BLOCK_OPT_ENCRYPT,
1097 .type = OPT_FLAG,
1098 .help = "Encrypt the image"
1099 },
1100 {
1101 .name = BLOCK_OPT_CLUSTER_SIZE,
1102 .type = OPT_SIZE,
1103 .help = "qcow2 cluster size"
1104 },
1105 {
1106 .name = BLOCK_OPT_PREALLOC,
1107 .type = OPT_STRING,
1108 .help = "Preallocation mode (allowed values: off, metadata)"
1109 },
1110 { NULL }
1111 };
1112
1113 static BlockDriver bdrv_qcow2 = {
1114 .format_name = "qcow2",
1115 .instance_size = sizeof(BDRVQcowState),
1116 .bdrv_probe = qcow_probe,
1117 .bdrv_open = qcow_open,
1118 .bdrv_close = qcow_close,
1119 .bdrv_create = qcow_create,
1120 .bdrv_flush = qcow_flush,
1121 .bdrv_is_allocated = qcow_is_allocated,
1122 .bdrv_set_key = qcow_set_key,
1123 .bdrv_make_empty = qcow_make_empty,
1124
1125 .bdrv_aio_readv = qcow_aio_readv,
1126 .bdrv_aio_writev = qcow_aio_writev,
1127 .bdrv_write_compressed = qcow_write_compressed,
1128
1129 .bdrv_snapshot_create = qcow2_snapshot_create,
1130 .bdrv_snapshot_goto = qcow2_snapshot_goto,
1131 .bdrv_snapshot_delete = qcow2_snapshot_delete,
1132 .bdrv_snapshot_list = qcow2_snapshot_list,
1133 .bdrv_get_info = qcow_get_info,
1134
1135 .bdrv_save_vmstate = qcow_save_vmstate,
1136 .bdrv_load_vmstate = qcow_load_vmstate,
1137
1138 .create_options = qcow_create_options,
1139 .bdrv_check = qcow_check,
1140 };
1141
1142 static void bdrv_qcow2_init(void)
1143 {
1144 bdrv_register(&bdrv_qcow2);
1145 }
1146
1147 block_init(bdrv_qcow2_init);
Anthony Liguori's QEMU queue