debuggers.hg

view tools/ioemu/block-qcow.c @ 10969:3ea0a6847198

[qemu] Make sure disk writes really made it to disk before we report I/O
completion to the guest domain. The DMA_MULTI_THREAD functionality
from the qemu-dm IDE emulation should make the performance overhead
of synchronous writes bearable, or at least comparable to native
hardware.

Signed-off-by: Rik van Riel <riel@redhat.com>
author chris@kneesaa.uk.xensource.com
date Fri Aug 04 10:28:51 2006 +0100 (2006-08-04)
parents b450f21472a0
children 08a11694b109
line source
1 /*
2 * Block driver for the QCOW format
3 *
4 * Copyright (c) 2004 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 "vl.h"
25 #include "block_int.h"
26 #include <zlib.h>
27 #include "aes.h"
29 /**************************************************************/
30 /* QEMU COW block driver with compression and encryption support */
32 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
33 #define QCOW_VERSION 1
35 #define QCOW_CRYPT_NONE 0
36 #define QCOW_CRYPT_AES 1
38 #define QCOW_OFLAG_COMPRESSED (1LL << 63)
40 typedef struct QCowHeader {
41 uint32_t magic;
42 uint32_t version;
43 uint64_t backing_file_offset;
44 uint32_t backing_file_size;
45 uint32_t mtime;
46 uint64_t size; /* in bytes */
47 uint8_t cluster_bits;
48 uint8_t l2_bits;
49 uint32_t crypt_method;
50 uint64_t l1_table_offset;
51 } QCowHeader;
53 #define L2_CACHE_SIZE 16
55 typedef struct BDRVQcowState {
56 int fd;
57 int cluster_bits;
58 int cluster_size;
59 int cluster_sectors;
60 int l2_bits;
61 int l2_size;
62 int l1_size;
63 uint64_t cluster_offset_mask;
64 uint64_t l1_table_offset;
65 uint64_t *l1_table;
66 uint64_t *l2_cache;
67 uint64_t l2_cache_offsets[L2_CACHE_SIZE];
68 uint32_t l2_cache_counts[L2_CACHE_SIZE];
69 uint8_t *cluster_cache;
70 uint8_t *cluster_data;
71 uint64_t cluster_cache_offset;
72 uint32_t crypt_method; /* current crypt method, 0 if no key yet */
73 uint32_t crypt_method_header;
74 AES_KEY aes_encrypt_key;
75 AES_KEY aes_decrypt_key;
76 } BDRVQcowState;
78 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
80 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
81 {
82 const QCowHeader *cow_header = (const void *)buf;
84 if (buf_size >= sizeof(QCowHeader) &&
85 be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
86 be32_to_cpu(cow_header->version) == QCOW_VERSION)
87 return 100;
88 else
89 return 0;
90 }
92 static int qcow_open(BlockDriverState *bs, const char *filename)
93 {
94 BDRVQcowState *s = bs->opaque;
95 int fd, len, i, shift;
96 QCowHeader header;
98 fd = open(filename, O_RDWR | O_BINARY | O_LARGEFILE | O_SYNC);
99 if (fd < 0) {
100 fd = open(filename, O_RDONLY | O_BINARY | O_LARGEFILE);
101 if (fd < 0)
102 return -1;
103 }
104 s->fd = fd;
105 if (read(fd, &header, sizeof(header)) != sizeof(header))
106 goto fail;
107 be32_to_cpus(&header.magic);
108 be32_to_cpus(&header.version);
109 be64_to_cpus(&header.backing_file_offset);
110 be32_to_cpus(&header.backing_file_size);
111 be32_to_cpus(&header.mtime);
112 be64_to_cpus(&header.size);
113 be32_to_cpus(&header.crypt_method);
114 be64_to_cpus(&header.l1_table_offset);
116 if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
117 goto fail;
118 if (header.size <= 1 || header.cluster_bits < 9)
119 goto fail;
120 if (header.crypt_method > QCOW_CRYPT_AES)
121 goto fail;
122 s->crypt_method_header = header.crypt_method;
123 if (s->crypt_method_header)
124 bs->encrypted = 1;
125 s->cluster_bits = header.cluster_bits;
126 s->cluster_size = 1 << s->cluster_bits;
127 s->cluster_sectors = 1 << (s->cluster_bits - 9);
128 s->l2_bits = header.l2_bits;
129 s->l2_size = 1 << s->l2_bits;
130 bs->total_sectors = header.size / 512;
131 s->cluster_offset_mask = (1LL << (63 - s->cluster_bits)) - 1;
133 /* read the level 1 table */
134 shift = s->cluster_bits + s->l2_bits;
135 s->l1_size = (header.size + (1LL << shift) - 1) >> shift;
137 s->l1_table_offset = header.l1_table_offset;
138 s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
139 if (!s->l1_table)
140 goto fail;
141 lseek(fd, s->l1_table_offset, SEEK_SET);
142 if (read(fd, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
143 s->l1_size * sizeof(uint64_t))
144 goto fail;
145 for(i = 0;i < s->l1_size; i++) {
146 be64_to_cpus(&s->l1_table[i]);
147 }
148 /* alloc L2 cache */
149 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
150 if (!s->l2_cache)
151 goto fail;
152 s->cluster_cache = qemu_malloc(s->cluster_size);
153 if (!s->cluster_cache)
154 goto fail;
155 s->cluster_data = qemu_malloc(s->cluster_size);
156 if (!s->cluster_data)
157 goto fail;
158 s->cluster_cache_offset = -1;
160 /* read the backing file name */
161 if (header.backing_file_offset != 0) {
162 len = header.backing_file_size;
163 if (len > 1023)
164 len = 1023;
165 lseek(fd, header.backing_file_offset, SEEK_SET);
166 if (read(fd, bs->backing_file, len) != len)
167 goto fail;
168 bs->backing_file[len] = '\0';
169 }
170 return 0;
172 fail:
173 qemu_free(s->l1_table);
174 qemu_free(s->l2_cache);
175 qemu_free(s->cluster_cache);
176 qemu_free(s->cluster_data);
177 close(fd);
178 return -1;
179 }
181 static int qcow_set_key(BlockDriverState *bs, const char *key)
182 {
183 BDRVQcowState *s = bs->opaque;
184 uint8_t keybuf[16];
185 int len, i;
187 memset(keybuf, 0, 16);
188 len = strlen(key);
189 if (len > 16)
190 len = 16;
191 /* XXX: we could compress the chars to 7 bits to increase
192 entropy */
193 for(i = 0;i < len;i++) {
194 keybuf[i] = key[i];
195 }
196 s->crypt_method = s->crypt_method_header;
198 if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
199 return -1;
200 if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
201 return -1;
202 #if 0
203 /* test */
204 {
205 uint8_t in[16];
206 uint8_t out[16];
207 uint8_t tmp[16];
208 for(i=0;i<16;i++)
209 in[i] = i;
210 AES_encrypt(in, tmp, &s->aes_encrypt_key);
211 AES_decrypt(tmp, out, &s->aes_decrypt_key);
212 for(i = 0; i < 16; i++)
213 printf(" %02x", tmp[i]);
214 printf("\n");
215 for(i = 0; i < 16; i++)
216 printf(" %02x", out[i]);
217 printf("\n");
218 }
219 #endif
220 return 0;
221 }
223 /* The crypt function is compatible with the linux cryptoloop
224 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
225 supported */
226 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
227 uint8_t *out_buf, const uint8_t *in_buf,
228 int nb_sectors, int enc,
229 const AES_KEY *key)
230 {
231 union {
232 uint64_t ll[2];
233 uint8_t b[16];
234 } ivec;
235 int i;
237 for(i = 0; i < nb_sectors; i++) {
238 ivec.ll[0] = cpu_to_le64(sector_num);
239 ivec.ll[1] = 0;
240 AES_cbc_encrypt(in_buf, out_buf, 512, key,
241 ivec.b, enc);
242 sector_num++;
243 in_buf += 512;
244 out_buf += 512;
245 }
246 }
248 /* 'allocate' is:
249 *
250 * 0 to not allocate.
251 *
252 * 1 to allocate a normal cluster (for sector indexes 'n_start' to
253 * 'n_end')
254 *
255 * 2 to allocate a compressed cluster of size
256 * 'compressed_size'. 'compressed_size' must be > 0 and <
257 * cluster_size
258 *
259 * return 0 if not allocated.
260 */
261 static uint64_t get_cluster_offset(BlockDriverState *bs,
262 uint64_t offset, int allocate,
263 int compressed_size,
264 int n_start, int n_end)
265 {
266 BDRVQcowState *s = bs->opaque;
267 int min_index, i, j, l1_index, l2_index;
268 uint64_t l2_offset, *l2_table, cluster_offset, tmp;
269 uint32_t min_count;
270 int new_l2_table;
272 l1_index = offset >> (s->l2_bits + s->cluster_bits);
273 l2_offset = s->l1_table[l1_index];
274 new_l2_table = 0;
275 if (!l2_offset) {
276 if (!allocate)
277 return 0;
278 /* allocate a new l2 entry */
279 l2_offset = lseek(s->fd, 0, SEEK_END);
280 /* round to cluster size */
281 l2_offset = (l2_offset + s->cluster_size - 1) & ~(s->cluster_size - 1);
282 /* update the L1 entry */
283 s->l1_table[l1_index] = l2_offset;
284 tmp = cpu_to_be64(l2_offset);
285 lseek(s->fd, s->l1_table_offset + l1_index * sizeof(tmp), SEEK_SET);
286 if (write(s->fd, &tmp, sizeof(tmp)) != sizeof(tmp))
287 return 0;
288 new_l2_table = 1;
289 }
290 for(i = 0; i < L2_CACHE_SIZE; i++) {
291 if (l2_offset == s->l2_cache_offsets[i]) {
292 /* increment the hit count */
293 if (++s->l2_cache_counts[i] == 0xffffffff) {
294 for(j = 0; j < L2_CACHE_SIZE; j++) {
295 s->l2_cache_counts[j] >>= 1;
296 }
297 }
298 l2_table = s->l2_cache + (i << s->l2_bits);
299 goto found;
300 }
301 }
302 /* not found: load a new entry in the least used one */
303 min_index = 0;
304 min_count = 0xffffffff;
305 for(i = 0; i < L2_CACHE_SIZE; i++) {
306 if (s->l2_cache_counts[i] < min_count) {
307 min_count = s->l2_cache_counts[i];
308 min_index = i;
309 }
310 }
311 l2_table = s->l2_cache + (min_index << s->l2_bits);
312 lseek(s->fd, l2_offset, SEEK_SET);
313 if (new_l2_table) {
314 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
315 if (write(s->fd, l2_table, s->l2_size * sizeof(uint64_t)) !=
316 s->l2_size * sizeof(uint64_t))
317 return 0;
318 } else {
319 if (read(s->fd, l2_table, s->l2_size * sizeof(uint64_t)) !=
320 s->l2_size * sizeof(uint64_t))
321 return 0;
322 }
323 s->l2_cache_offsets[min_index] = l2_offset;
324 s->l2_cache_counts[min_index] = 1;
325 found:
326 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
327 cluster_offset = be64_to_cpu(l2_table[l2_index]);
328 if (!cluster_offset ||
329 ((cluster_offset & QCOW_OFLAG_COMPRESSED) && allocate == 1)) {
330 if (!allocate)
331 return 0;
332 /* allocate a new cluster */
333 if ((cluster_offset & QCOW_OFLAG_COMPRESSED) &&
334 (n_end - n_start) < s->cluster_sectors) {
335 /* if the cluster is already compressed, we must
336 decompress it in the case it is not completely
337 overwritten */
338 if (decompress_cluster(s, cluster_offset) < 0)
339 return 0;
340 cluster_offset = lseek(s->fd, 0, SEEK_END);
341 cluster_offset = (cluster_offset + s->cluster_size - 1) &
342 ~(s->cluster_size - 1);
343 /* write the cluster content */
344 lseek(s->fd, cluster_offset, SEEK_SET);
345 if (write(s->fd, s->cluster_cache, s->cluster_size) !=
346 s->cluster_size)
347 return -1;
348 } else {
349 cluster_offset = lseek(s->fd, 0, SEEK_END);
350 if (allocate == 1) {
351 /* round to cluster size */
352 cluster_offset = (cluster_offset + s->cluster_size - 1) &
353 ~(s->cluster_size - 1);
354 ftruncate(s->fd, cluster_offset + s->cluster_size);
355 /* if encrypted, we must initialize the cluster
356 content which won't be written */
357 if (s->crypt_method &&
358 (n_end - n_start) < s->cluster_sectors) {
359 uint64_t start_sect;
360 start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
361 memset(s->cluster_data + 512, 0xaa, 512);
362 for(i = 0; i < s->cluster_sectors; i++) {
363 if (i < n_start || i >= n_end) {
364 encrypt_sectors(s, start_sect + i,
365 s->cluster_data,
366 s->cluster_data + 512, 1, 1,
367 &s->aes_encrypt_key);
368 lseek(s->fd, cluster_offset + i * 512, SEEK_SET);
369 if (write(s->fd, s->cluster_data, 512) != 512)
370 return -1;
371 }
372 }
373 }
374 } else {
375 cluster_offset |= QCOW_OFLAG_COMPRESSED |
376 (uint64_t)compressed_size << (63 - s->cluster_bits);
377 }
378 }
379 /* update L2 table */
380 tmp = cpu_to_be64(cluster_offset);
381 l2_table[l2_index] = tmp;
382 lseek(s->fd, l2_offset + l2_index * sizeof(tmp), SEEK_SET);
383 if (write(s->fd, &tmp, sizeof(tmp)) != sizeof(tmp))
384 return 0;
385 }
386 return cluster_offset;
387 }
389 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
390 int nb_sectors, int *pnum)
391 {
392 BDRVQcowState *s = bs->opaque;
393 int index_in_cluster, n;
394 uint64_t cluster_offset;
396 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
397 index_in_cluster = sector_num & (s->cluster_sectors - 1);
398 n = s->cluster_sectors - index_in_cluster;
399 if (n > nb_sectors)
400 n = nb_sectors;
401 *pnum = n;
402 return (cluster_offset != 0);
403 }
405 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
406 const uint8_t *buf, int buf_size)
407 {
408 z_stream strm1, *strm = &strm1;
409 int ret, out_len;
411 memset(strm, 0, sizeof(*strm));
413 strm->next_in = (uint8_t *)buf;
414 strm->avail_in = buf_size;
415 strm->next_out = out_buf;
416 strm->avail_out = out_buf_size;
418 ret = inflateInit2(strm, -12);
419 if (ret != Z_OK)
420 return -1;
421 ret = inflate(strm, Z_FINISH);
422 out_len = strm->next_out - out_buf;
423 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
424 out_len != out_buf_size) {
425 inflateEnd(strm);
426 return -1;
427 }
428 inflateEnd(strm);
429 return 0;
430 }
432 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
433 {
434 int ret, csize;
435 uint64_t coffset;
437 coffset = cluster_offset & s->cluster_offset_mask;
438 if (s->cluster_cache_offset != coffset) {
439 csize = cluster_offset >> (63 - s->cluster_bits);
440 csize &= (s->cluster_size - 1);
441 lseek(s->fd, coffset, SEEK_SET);
442 ret = read(s->fd, s->cluster_data, csize);
443 if (ret != csize)
444 return -1;
445 if (decompress_buffer(s->cluster_cache, s->cluster_size,
446 s->cluster_data, csize) < 0) {
447 return -1;
448 }
449 s->cluster_cache_offset = coffset;
450 }
451 return 0;
452 }
454 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
455 uint8_t *buf, int nb_sectors)
456 {
457 BDRVQcowState *s = bs->opaque;
458 int ret, index_in_cluster, n;
459 uint64_t cluster_offset;
461 while (nb_sectors > 0) {
462 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0, 0, 0, 0);
463 index_in_cluster = sector_num & (s->cluster_sectors - 1);
464 n = s->cluster_sectors - index_in_cluster;
465 if (n > nb_sectors)
466 n = nb_sectors;
467 if (!cluster_offset) {
468 memset(buf, 0, 512 * n);
469 } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
470 if (decompress_cluster(s, cluster_offset) < 0)
471 return -1;
472 memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
473 } else {
474 lseek(s->fd, cluster_offset + index_in_cluster * 512, SEEK_SET);
475 ret = read(s->fd, buf, n * 512);
476 if (ret != n * 512)
477 return -1;
478 if (s->crypt_method) {
479 encrypt_sectors(s, sector_num, buf, buf, n, 0,
480 &s->aes_decrypt_key);
481 }
482 }
483 nb_sectors -= n;
484 sector_num += n;
485 buf += n * 512;
486 }
487 return 0;
488 }
490 static int qcow_write(BlockDriverState *bs, int64_t sector_num,
491 const uint8_t *buf, int nb_sectors)
492 {
493 BDRVQcowState *s = bs->opaque;
494 int ret, index_in_cluster, n;
495 uint64_t cluster_offset;
497 while (nb_sectors > 0) {
498 index_in_cluster = sector_num & (s->cluster_sectors - 1);
499 n = s->cluster_sectors - index_in_cluster;
500 if (n > nb_sectors)
501 n = nb_sectors;
502 cluster_offset = get_cluster_offset(bs, sector_num << 9, 1, 0,
503 index_in_cluster,
504 index_in_cluster + n);
505 if (!cluster_offset)
506 return -1;
507 lseek(s->fd, cluster_offset + index_in_cluster * 512, SEEK_SET);
508 if (s->crypt_method) {
509 encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
510 &s->aes_encrypt_key);
511 ret = write(s->fd, s->cluster_data, n * 512);
512 } else {
513 ret = write(s->fd, buf, n * 512);
514 }
515 if (ret != n * 512)
516 return -1;
517 nb_sectors -= n;
518 sector_num += n;
519 buf += n * 512;
520 }
521 s->cluster_cache_offset = -1; /* disable compressed cache */
522 return 0;
523 }
525 static void qcow_close(BlockDriverState *bs)
526 {
527 BDRVQcowState *s = bs->opaque;
528 qemu_free(s->l1_table);
529 qemu_free(s->l2_cache);
530 qemu_free(s->cluster_cache);
531 qemu_free(s->cluster_data);
532 close(s->fd);
533 }
535 static int qcow_create(const char *filename, int64_t total_size,
536 const char *backing_file, int flags)
537 {
538 int fd, header_size, backing_filename_len, l1_size, i, shift;
539 QCowHeader header;
540 char backing_filename[1024];
541 uint64_t tmp;
542 struct stat st;
544 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
545 0644);
546 if (fd < 0)
547 return -1;
548 memset(&header, 0, sizeof(header));
549 header.magic = cpu_to_be32(QCOW_MAGIC);
550 header.version = cpu_to_be32(QCOW_VERSION);
551 header.size = cpu_to_be64(total_size * 512);
552 header_size = sizeof(header);
553 backing_filename_len = 0;
554 if (backing_file) {
555 if (strcmp(backing_file, "fat:")) {
556 const char *p;
557 /* XXX: this is a hack: we do not attempt to check for URL
558 like syntax */
559 p = strchr(backing_file, ':');
560 if (p && (p - backing_file) >= 2) {
561 /* URL like but exclude "c:" like filenames */
562 pstrcpy(backing_filename, sizeof(backing_filename),
563 backing_file);
564 } else {
565 realpath(backing_file, backing_filename);
566 if (stat(backing_filename, &st) != 0) {
567 return -1;
568 }
569 }
570 header.backing_file_offset = cpu_to_be64(header_size);
571 backing_filename_len = strlen(backing_filename);
572 header.backing_file_size = cpu_to_be32(backing_filename_len);
573 header_size += backing_filename_len;
574 } else
575 backing_file = NULL;
576 header.mtime = cpu_to_be32(st.st_mtime);
577 header.cluster_bits = 9; /* 512 byte cluster to avoid copying
578 unmodifyed sectors */
579 header.l2_bits = 12; /* 32 KB L2 tables */
580 } else {
581 header.cluster_bits = 12; /* 4 KB clusters */
582 header.l2_bits = 9; /* 4 KB L2 tables */
583 }
584 header_size = (header_size + 7) & ~7;
585 shift = header.cluster_bits + header.l2_bits;
586 l1_size = ((total_size * 512) + (1LL << shift) - 1) >> shift;
588 header.l1_table_offset = cpu_to_be64(header_size);
589 if (flags) {
590 header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
591 } else {
592 header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
593 }
595 /* write all the data */
596 write(fd, &header, sizeof(header));
597 if (backing_file) {
598 write(fd, backing_filename, backing_filename_len);
599 }
600 lseek(fd, header_size, SEEK_SET);
601 tmp = 0;
602 for(i = 0;i < l1_size; i++) {
603 write(fd, &tmp, sizeof(tmp));
604 }
605 close(fd);
606 return 0;
607 }
609 int qcow_make_empty(BlockDriverState *bs)
610 {
611 BDRVQcowState *s = bs->opaque;
612 uint32_t l1_length = s->l1_size * sizeof(uint64_t);
614 memset(s->l1_table, 0, l1_length);
615 lseek(s->fd, s->l1_table_offset, SEEK_SET);
616 if (write(s->fd, s->l1_table, l1_length) < 0)
617 return -1;
618 ftruncate(s->fd, s->l1_table_offset + l1_length);
620 memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
621 memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
622 memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
624 return 0;
625 }
627 int qcow_get_cluster_size(BlockDriverState *bs)
628 {
629 BDRVQcowState *s = bs->opaque;
630 if (bs->drv != &bdrv_qcow)
631 return -1;
632 return s->cluster_size;
633 }
635 /* XXX: put compressed sectors first, then all the cluster aligned
636 tables to avoid losing bytes in alignment */
637 int qcow_compress_cluster(BlockDriverState *bs, int64_t sector_num,
638 const uint8_t *buf)
639 {
640 BDRVQcowState *s = bs->opaque;
641 z_stream strm;
642 int ret, out_len;
643 uint8_t *out_buf;
644 uint64_t cluster_offset;
646 if (bs->drv != &bdrv_qcow)
647 return -1;
649 out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
650 if (!out_buf)
651 return -1;
653 /* best compression, small window, no zlib header */
654 memset(&strm, 0, sizeof(strm));
655 ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
656 Z_DEFLATED, -12,
657 9, Z_DEFAULT_STRATEGY);
658 if (ret != 0) {
659 qemu_free(out_buf);
660 return -1;
661 }
663 strm.avail_in = s->cluster_size;
664 strm.next_in = (uint8_t *)buf;
665 strm.avail_out = s->cluster_size;
666 strm.next_out = out_buf;
668 ret = deflate(&strm, Z_FINISH);
669 if (ret != Z_STREAM_END && ret != Z_OK) {
670 qemu_free(out_buf);
671 deflateEnd(&strm);
672 return -1;
673 }
674 out_len = strm.next_out - out_buf;
676 deflateEnd(&strm);
678 if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
679 /* could not compress: write normal cluster */
680 qcow_write(bs, sector_num, buf, s->cluster_sectors);
681 } else {
682 cluster_offset = get_cluster_offset(bs, sector_num << 9, 2,
683 out_len, 0, 0);
684 cluster_offset &= s->cluster_offset_mask;
685 lseek(s->fd, cluster_offset, SEEK_SET);
686 if (write(s->fd, out_buf, out_len) != out_len) {
687 qemu_free(out_buf);
688 return -1;
689 }
690 }
692 qemu_free(out_buf);
693 return 0;
694 }
696 BlockDriver bdrv_qcow = {
697 "qcow",
698 sizeof(BDRVQcowState),
699 qcow_probe,
700 qcow_open,
701 qcow_read,
702 qcow_write,
703 qcow_close,
704 qcow_create,
705 qcow_is_allocated,
706 qcow_set_key,
707 qcow_make_empty
708 };