]>
Commit | Line | Data |
---|---|---|
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 | /* Sanity check */ | |
91 | if (offset > s->cluster_size) | |
92 | printf("qcow2_read_extension: suspicious offset %lu\n", offset); | |
93 | ||
94 | printf("attemting to read extended header in offset %lu\n", offset); | |
95 | #endif | |
96 | ||
97 | if (bdrv_pread(bs->file, offset, &ext, sizeof(ext)) != sizeof(ext)) { | |
98 | fprintf(stderr, "qcow2_read_extension: ERROR: " | |
99 | "pread fail from offset %" PRIu64 "\n", | |
100 | offset); | |
101 | return 1; | |
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 QCOW2_EXT_MAGIC_END: | |
111 | return 0; | |
112 | ||
113 | case QCOW2_EXT_MAGIC_BACKING_FORMAT: | |
114 | if (ext.len >= sizeof(bs->backing_format)) { | |
115 | fprintf(stderr, "ERROR: ext_backing_format: len=%u too large" | |
116 | " (>=%zu)\n", | |
117 | ext.len, sizeof(bs->backing_format)); | |
118 | return 2; | |
119 | } | |
120 | if (bdrv_pread(bs->file, offset , bs->backing_format, | |
121 | ext.len) != ext.len) | |
122 | return 3; | |
123 | bs->backing_format[ext.len] = '\0'; | |
124 | #ifdef DEBUG_EXT | |
125 | printf("Qcow2: Got format extension %s\n", bs->backing_format); | |
126 | #endif | |
127 | offset = ((offset + ext.len + 7) & ~7); | |
128 | break; | |
129 | ||
130 | default: | |
131 | /* unknown magic -- just skip it */ | |
132 | offset = ((offset + ext.len + 7) & ~7); | |
133 | break; | |
134 | } | |
135 | } | |
136 | ||
137 | return 0; | |
138 | } | |
139 | ||
140 | ||
141 | static int qcow2_open(BlockDriverState *bs, int flags) | |
142 | { | |
143 | BDRVQcowState *s = bs->opaque; | |
144 | int len, i, ret = 0; | |
145 | QCowHeader header; | |
146 | uint64_t ext_end; | |
147 | bool writethrough; | |
148 | ||
149 | ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); | |
150 | if (ret < 0) { | |
151 | goto fail; | |
152 | } | |
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) { | |
168 | ret = -EINVAL; | |
169 | goto fail; | |
170 | } | |
171 | if (header.version != QCOW_VERSION) { | |
172 | char version[64]; | |
173 | snprintf(version, sizeof(version), "QCOW version %d", header.version); | |
174 | qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, | |
175 | bs->device_name, "qcow2", version); | |
176 | ret = -ENOTSUP; | |
177 | goto fail; | |
178 | } | |
179 | if (header.cluster_bits < MIN_CLUSTER_BITS || | |
180 | header.cluster_bits > MAX_CLUSTER_BITS) { | |
181 | ret = -EINVAL; | |
182 | goto fail; | |
183 | } | |
184 | if (header.crypt_method > QCOW_CRYPT_AES) { | |
185 | ret = -EINVAL; | |
186 | goto fail; | |
187 | } | |
188 | s->crypt_method_header = header.crypt_method; | |
189 | if (s->crypt_method_header) { | |
190 | bs->encrypted = 1; | |
191 | } | |
192 | s->cluster_bits = header.cluster_bits; | |
193 | s->cluster_size = 1 << s->cluster_bits; | |
194 | s->cluster_sectors = 1 << (s->cluster_bits - 9); | |
195 | s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ | |
196 | s->l2_size = 1 << s->l2_bits; | |
197 | bs->total_sectors = header.size / 512; | |
198 | s->csize_shift = (62 - (s->cluster_bits - 8)); | |
199 | s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; | |
200 | s->cluster_offset_mask = (1LL << s->csize_shift) - 1; | |
201 | s->refcount_table_offset = header.refcount_table_offset; | |
202 | s->refcount_table_size = | |
203 | header.refcount_table_clusters << (s->cluster_bits - 3); | |
204 | ||
205 | s->snapshots_offset = header.snapshots_offset; | |
206 | s->nb_snapshots = header.nb_snapshots; | |
207 | ||
208 | /* read the level 1 table */ | |
209 | s->l1_size = header.l1_size; | |
210 | s->l1_vm_state_index = size_to_l1(s, header.size); | |
211 | /* the L1 table must contain at least enough entries to put | |
212 | header.size bytes */ | |
213 | if (s->l1_size < s->l1_vm_state_index) { | |
214 | ret = -EINVAL; | |
215 | goto fail; | |
216 | } | |
217 | s->l1_table_offset = header.l1_table_offset; | |
218 | if (s->l1_size > 0) { | |
219 | s->l1_table = qemu_mallocz( | |
220 | align_offset(s->l1_size * sizeof(uint64_t), 512)); | |
221 | ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, | |
222 | s->l1_size * sizeof(uint64_t)); | |
223 | if (ret < 0) { | |
224 | goto fail; | |
225 | } | |
226 | for(i = 0;i < s->l1_size; i++) { | |
227 | be64_to_cpus(&s->l1_table[i]); | |
228 | } | |
229 | } | |
230 | ||
231 | /* alloc L2 table/refcount block cache */ | |
232 | writethrough = ((flags & BDRV_O_CACHE_WB) == 0); | |
233 | s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE, writethrough); | |
234 | s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE, | |
235 | writethrough); | |
236 | ||
237 | s->cluster_cache = qemu_malloc(s->cluster_size); | |
238 | /* one more sector for decompressed data alignment */ | |
239 | s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size | |
240 | + 512); | |
241 | s->cluster_cache_offset = -1; | |
242 | ||
243 | ret = qcow2_refcount_init(bs); | |
244 | if (ret != 0) { | |
245 | goto fail; | |
246 | } | |
247 | ||
248 | QLIST_INIT(&s->cluster_allocs); | |
249 | ||
250 | /* read qcow2 extensions */ | |
251 | if (header.backing_file_offset) { | |
252 | ext_end = header.backing_file_offset; | |
253 | } else { | |
254 | ext_end = s->cluster_size; | |
255 | } | |
256 | if (qcow2_read_extensions(bs, sizeof(header), ext_end)) { | |
257 | ret = -EINVAL; | |
258 | goto fail; | |
259 | } | |
260 | ||
261 | /* read the backing file name */ | |
262 | if (header.backing_file_offset != 0) { | |
263 | len = header.backing_file_size; | |
264 | if (len > 1023) { | |
265 | len = 1023; | |
266 | } | |
267 | ret = bdrv_pread(bs->file, header.backing_file_offset, | |
268 | bs->backing_file, len); | |
269 | if (ret < 0) { | |
270 | goto fail; | |
271 | } | |
272 | bs->backing_file[len] = '\0'; | |
273 | } | |
274 | if (qcow2_read_snapshots(bs) < 0) { | |
275 | ret = -EINVAL; | |
276 | goto fail; | |
277 | } | |
278 | ||
279 | #ifdef DEBUG_ALLOC | |
280 | qcow2_check_refcounts(bs); | |
281 | #endif | |
282 | return ret; | |
283 | ||
284 | fail: | |
285 | qcow2_free_snapshots(bs); | |
286 | qcow2_refcount_close(bs); | |
287 | qemu_free(s->l1_table); | |
288 | if (s->l2_table_cache) { | |
289 | qcow2_cache_destroy(bs, s->l2_table_cache); | |
290 | } | |
291 | qemu_free(s->cluster_cache); | |
292 | qemu_free(s->cluster_data); | |
293 | return ret; | |
294 | } | |
295 | ||
296 | static int qcow2_set_key(BlockDriverState *bs, const char *key) | |
297 | { | |
298 | BDRVQcowState *s = bs->opaque; | |
299 | uint8_t keybuf[16]; | |
300 | int len, i; | |
301 | ||
302 | memset(keybuf, 0, 16); | |
303 | len = strlen(key); | |
304 | if (len > 16) | |
305 | len = 16; | |
306 | /* XXX: we could compress the chars to 7 bits to increase | |
307 | entropy */ | |
308 | for(i = 0;i < len;i++) { | |
309 | keybuf[i] = key[i]; | |
310 | } | |
311 | s->crypt_method = s->crypt_method_header; | |
312 | ||
313 | if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0) | |
314 | return -1; | |
315 | if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0) | |
316 | return -1; | |
317 | #if 0 | |
318 | /* test */ | |
319 | { | |
320 | uint8_t in[16]; | |
321 | uint8_t out[16]; | |
322 | uint8_t tmp[16]; | |
323 | for(i=0;i<16;i++) | |
324 | in[i] = i; | |
325 | AES_encrypt(in, tmp, &s->aes_encrypt_key); | |
326 | AES_decrypt(tmp, out, &s->aes_decrypt_key); | |
327 | for(i = 0; i < 16; i++) | |
328 | printf(" %02x", tmp[i]); | |
329 | printf("\n"); | |
330 | for(i = 0; i < 16; i++) | |
331 | printf(" %02x", out[i]); | |
332 | printf("\n"); | |
333 | } | |
334 | #endif | |
335 | return 0; | |
336 | } | |
337 | ||
338 | static int qcow2_is_allocated(BlockDriverState *bs, int64_t sector_num, | |
339 | int nb_sectors, int *pnum) | |
340 | { | |
341 | uint64_t cluster_offset; | |
342 | int ret; | |
343 | ||
344 | *pnum = nb_sectors; | |
345 | /* FIXME We can get errors here, but the bdrv_is_allocated interface can't | |
346 | * pass them on today */ | |
347 | ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset); | |
348 | if (ret < 0) { | |
349 | *pnum = 0; | |
350 | } | |
351 | ||
352 | return (cluster_offset != 0); | |
353 | } | |
354 | ||
355 | /* handle reading after the end of the backing file */ | |
356 | int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov, | |
357 | int64_t sector_num, int nb_sectors) | |
358 | { | |
359 | int n1; | |
360 | if ((sector_num + nb_sectors) <= bs->total_sectors) | |
361 | return nb_sectors; | |
362 | if (sector_num >= bs->total_sectors) | |
363 | n1 = 0; | |
364 | else | |
365 | n1 = bs->total_sectors - sector_num; | |
366 | ||
367 | qemu_iovec_memset_skip(qiov, 0, 512 * (nb_sectors - n1), 512 * n1); | |
368 | ||
369 | return n1; | |
370 | } | |
371 | ||
372 | typedef struct QCowAIOCB { | |
373 | BlockDriverAIOCB common; | |
374 | int64_t sector_num; | |
375 | QEMUIOVector *qiov; | |
376 | int remaining_sectors; | |
377 | int cur_nr_sectors; /* number of sectors in current iteration */ | |
378 | uint64_t bytes_done; | |
379 | uint64_t cluster_offset; | |
380 | uint8_t *cluster_data; | |
381 | bool is_write; | |
382 | BlockDriverAIOCB *hd_aiocb; | |
383 | QEMUIOVector hd_qiov; | |
384 | QEMUBH *bh; | |
385 | QCowL2Meta l2meta; | |
386 | QLIST_ENTRY(QCowAIOCB) next_depend; | |
387 | } QCowAIOCB; | |
388 | ||
389 | static void qcow2_aio_cancel(BlockDriverAIOCB *blockacb) | |
390 | { | |
391 | QCowAIOCB *acb = container_of(blockacb, QCowAIOCB, common); | |
392 | if (acb->hd_aiocb) | |
393 | bdrv_aio_cancel(acb->hd_aiocb); | |
394 | qemu_aio_release(acb); | |
395 | } | |
396 | ||
397 | static AIOPool qcow2_aio_pool = { | |
398 | .aiocb_size = sizeof(QCowAIOCB), | |
399 | .cancel = qcow2_aio_cancel, | |
400 | }; | |
401 | ||
402 | static void qcow2_aio_read_cb(void *opaque, int ret); | |
403 | static void qcow2_aio_write_cb(void *opaque, int ret); | |
404 | ||
405 | static void qcow2_aio_rw_bh(void *opaque) | |
406 | { | |
407 | QCowAIOCB *acb = opaque; | |
408 | qemu_bh_delete(acb->bh); | |
409 | acb->bh = NULL; | |
410 | ||
411 | if (acb->is_write) { | |
412 | qcow2_aio_write_cb(opaque, 0); | |
413 | } else { | |
414 | qcow2_aio_read_cb(opaque, 0); | |
415 | } | |
416 | } | |
417 | ||
418 | static int qcow2_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb) | |
419 | { | |
420 | if (acb->bh) | |
421 | return -EIO; | |
422 | ||
423 | acb->bh = qemu_bh_new(cb, acb); | |
424 | if (!acb->bh) | |
425 | return -EIO; | |
426 | ||
427 | qemu_bh_schedule(acb->bh); | |
428 | ||
429 | return 0; | |
430 | } | |
431 | ||
432 | static void qcow2_aio_read_cb(void *opaque, int ret) | |
433 | { | |
434 | QCowAIOCB *acb = opaque; | |
435 | BlockDriverState *bs = acb->common.bs; | |
436 | BDRVQcowState *s = bs->opaque; | |
437 | int index_in_cluster, n1; | |
438 | ||
439 | acb->hd_aiocb = NULL; | |
440 | if (ret < 0) | |
441 | goto done; | |
442 | ||
443 | /* post process the read buffer */ | |
444 | if (!acb->cluster_offset) { | |
445 | /* nothing to do */ | |
446 | } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) { | |
447 | /* nothing to do */ | |
448 | } else { | |
449 | if (s->crypt_method) { | |
450 | qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data, | |
451 | acb->cluster_data, acb->cur_nr_sectors, 0, &s->aes_decrypt_key); | |
452 | qemu_iovec_reset(&acb->hd_qiov); | |
453 | qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done, | |
454 | acb->cur_nr_sectors * 512); | |
455 | qemu_iovec_from_buffer(&acb->hd_qiov, acb->cluster_data, | |
456 | 512 * acb->cur_nr_sectors); | |
457 | } | |
458 | } | |
459 | ||
460 | acb->remaining_sectors -= acb->cur_nr_sectors; | |
461 | acb->sector_num += acb->cur_nr_sectors; | |
462 | acb->bytes_done += acb->cur_nr_sectors * 512; | |
463 | ||
464 | if (acb->remaining_sectors == 0) { | |
465 | /* request completed */ | |
466 | ret = 0; | |
467 | goto done; | |
468 | } | |
469 | ||
470 | /* prepare next AIO request */ | |
471 | acb->cur_nr_sectors = acb->remaining_sectors; | |
472 | if (s->crypt_method) { | |
473 | acb->cur_nr_sectors = MIN(acb->cur_nr_sectors, | |
474 | QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); | |
475 | } | |
476 | ||
477 | ret = qcow2_get_cluster_offset(bs, acb->sector_num << 9, | |
478 | &acb->cur_nr_sectors, &acb->cluster_offset); | |
479 | if (ret < 0) { | |
480 | goto done; | |
481 | } | |
482 | ||
483 | index_in_cluster = acb->sector_num & (s->cluster_sectors - 1); | |
484 | ||
485 | qemu_iovec_reset(&acb->hd_qiov); | |
486 | qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done, | |
487 | acb->cur_nr_sectors * 512); | |
488 | ||
489 | if (!acb->cluster_offset) { | |
490 | ||
491 | if (bs->backing_hd) { | |
492 | /* read from the base image */ | |
493 | n1 = qcow2_backing_read1(bs->backing_hd, &acb->hd_qiov, | |
494 | acb->sector_num, acb->cur_nr_sectors); | |
495 | if (n1 > 0) { | |
496 | BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); | |
497 | acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num, | |
498 | &acb->hd_qiov, n1, qcow2_aio_read_cb, acb); | |
499 | if (acb->hd_aiocb == NULL) { | |
500 | ret = -EIO; | |
501 | goto done; | |
502 | } | |
503 | } else { | |
504 | ret = qcow2_schedule_bh(qcow2_aio_rw_bh, acb); | |
505 | if (ret < 0) | |
506 | goto done; | |
507 | } | |
508 | } else { | |
509 | /* Note: in this case, no need to wait */ | |
510 | qemu_iovec_memset(&acb->hd_qiov, 0, 512 * acb->cur_nr_sectors); | |
511 | ret = qcow2_schedule_bh(qcow2_aio_rw_bh, acb); | |
512 | if (ret < 0) | |
513 | goto done; | |
514 | } | |
515 | } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) { | |
516 | /* add AIO support for compressed blocks ? */ | |
517 | ret = qcow2_decompress_cluster(bs, acb->cluster_offset); | |
518 | if (ret < 0) { | |
519 | goto done; | |
520 | } | |
521 | ||
522 | qemu_iovec_from_buffer(&acb->hd_qiov, | |
523 | s->cluster_cache + index_in_cluster * 512, | |
524 | 512 * acb->cur_nr_sectors); | |
525 | ||
526 | ret = qcow2_schedule_bh(qcow2_aio_rw_bh, acb); | |
527 | if (ret < 0) | |
528 | goto done; | |
529 | } else { | |
530 | if ((acb->cluster_offset & 511) != 0) { | |
531 | ret = -EIO; | |
532 | goto done; | |
533 | } | |
534 | ||
535 | if (s->crypt_method) { | |
536 | /* | |
537 | * For encrypted images, read everything into a temporary | |
538 | * contiguous buffer on which the AES functions can work. | |
539 | */ | |
540 | if (!acb->cluster_data) { | |
541 | acb->cluster_data = | |
542 | qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); | |
543 | } | |
544 | ||
545 | assert(acb->cur_nr_sectors <= | |
546 | QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); | |
547 | qemu_iovec_reset(&acb->hd_qiov); | |
548 | qemu_iovec_add(&acb->hd_qiov, acb->cluster_data, | |
549 | 512 * acb->cur_nr_sectors); | |
550 | } | |
551 | ||
552 | BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); | |
553 | acb->hd_aiocb = bdrv_aio_readv(bs->file, | |
554 | (acb->cluster_offset >> 9) + index_in_cluster, | |
555 | &acb->hd_qiov, acb->cur_nr_sectors, | |
556 | qcow2_aio_read_cb, acb); | |
557 | if (acb->hd_aiocb == NULL) { | |
558 | ret = -EIO; | |
559 | goto done; | |
560 | } | |
561 | } | |
562 | ||
563 | return; | |
564 | done: | |
565 | acb->common.cb(acb->common.opaque, ret); | |
566 | qemu_iovec_destroy(&acb->hd_qiov); | |
567 | qemu_aio_release(acb); | |
568 | } | |
569 | ||
570 | static QCowAIOCB *qcow2_aio_setup(BlockDriverState *bs, int64_t sector_num, | |
571 | QEMUIOVector *qiov, int nb_sectors, | |
572 | BlockDriverCompletionFunc *cb, | |
573 | void *opaque, int is_write) | |
574 | { | |
575 | QCowAIOCB *acb; | |
576 | ||
577 | acb = qemu_aio_get(&qcow2_aio_pool, bs, cb, opaque); | |
578 | if (!acb) | |
579 | return NULL; | |
580 | acb->hd_aiocb = NULL; | |
581 | acb->sector_num = sector_num; | |
582 | acb->qiov = qiov; | |
583 | acb->is_write = is_write; | |
584 | ||
585 | qemu_iovec_init(&acb->hd_qiov, qiov->niov); | |
586 | ||
587 | acb->bytes_done = 0; | |
588 | acb->remaining_sectors = nb_sectors; | |
589 | acb->cur_nr_sectors = 0; | |
590 | acb->cluster_offset = 0; | |
591 | acb->l2meta.nb_clusters = 0; | |
592 | QLIST_INIT(&acb->l2meta.dependent_requests); | |
593 | return acb; | |
594 | } | |
595 | ||
596 | static BlockDriverAIOCB *qcow2_aio_readv(BlockDriverState *bs, | |
597 | int64_t sector_num, | |
598 | QEMUIOVector *qiov, int nb_sectors, | |
599 | BlockDriverCompletionFunc *cb, | |
600 | void *opaque) | |
601 | { | |
602 | QCowAIOCB *acb; | |
603 | int ret; | |
604 | ||
605 | acb = qcow2_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0); | |
606 | if (!acb) | |
607 | return NULL; | |
608 | ||
609 | ret = qcow2_schedule_bh(qcow2_aio_rw_bh, acb); | |
610 | if (ret < 0) { | |
611 | qemu_iovec_destroy(&acb->hd_qiov); | |
612 | qemu_aio_release(acb); | |
613 | return NULL; | |
614 | } | |
615 | ||
616 | return &acb->common; | |
617 | } | |
618 | ||
619 | static void run_dependent_requests(QCowL2Meta *m) | |
620 | { | |
621 | QCowAIOCB *req; | |
622 | QCowAIOCB *next; | |
623 | ||
624 | /* Take the request off the list of running requests */ | |
625 | if (m->nb_clusters != 0) { | |
626 | QLIST_REMOVE(m, next_in_flight); | |
627 | } | |
628 | ||
629 | /* Restart all dependent requests */ | |
630 | QLIST_FOREACH_SAFE(req, &m->dependent_requests, next_depend, next) { | |
631 | qcow2_aio_write_cb(req, 0); | |
632 | } | |
633 | ||
634 | /* Empty the list for the next part of the request */ | |
635 | QLIST_INIT(&m->dependent_requests); | |
636 | } | |
637 | ||
638 | static void qcow2_aio_write_cb(void *opaque, int ret) | |
639 | { | |
640 | QCowAIOCB *acb = opaque; | |
641 | BlockDriverState *bs = acb->common.bs; | |
642 | BDRVQcowState *s = bs->opaque; | |
643 | int index_in_cluster; | |
644 | int n_end; | |
645 | ||
646 | acb->hd_aiocb = NULL; | |
647 | ||
648 | if (ret >= 0) { | |
649 | ret = qcow2_alloc_cluster_link_l2(bs, &acb->l2meta); | |
650 | } | |
651 | ||
652 | run_dependent_requests(&acb->l2meta); | |
653 | ||
654 | if (ret < 0) | |
655 | goto done; | |
656 | ||
657 | acb->remaining_sectors -= acb->cur_nr_sectors; | |
658 | acb->sector_num += acb->cur_nr_sectors; | |
659 | acb->bytes_done += acb->cur_nr_sectors * 512; | |
660 | ||
661 | if (acb->remaining_sectors == 0) { | |
662 | /* request completed */ | |
663 | ret = 0; | |
664 | goto done; | |
665 | } | |
666 | ||
667 | index_in_cluster = acb->sector_num & (s->cluster_sectors - 1); | |
668 | n_end = index_in_cluster + acb->remaining_sectors; | |
669 | if (s->crypt_method && | |
670 | n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) | |
671 | n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors; | |
672 | ||
673 | ret = qcow2_alloc_cluster_offset(bs, acb->sector_num << 9, | |
674 | index_in_cluster, n_end, &acb->cur_nr_sectors, &acb->l2meta); | |
675 | if (ret < 0) { | |
676 | goto done; | |
677 | } | |
678 | ||
679 | acb->cluster_offset = acb->l2meta.cluster_offset; | |
680 | ||
681 | /* Need to wait for another request? If so, we are done for now. */ | |
682 | if (acb->l2meta.nb_clusters == 0 && acb->l2meta.depends_on != NULL) { | |
683 | QLIST_INSERT_HEAD(&acb->l2meta.depends_on->dependent_requests, | |
684 | acb, next_depend); | |
685 | return; | |
686 | } | |
687 | ||
688 | assert((acb->cluster_offset & 511) == 0); | |
689 | ||
690 | qemu_iovec_reset(&acb->hd_qiov); | |
691 | qemu_iovec_copy(&acb->hd_qiov, acb->qiov, acb->bytes_done, | |
692 | acb->cur_nr_sectors * 512); | |
693 | ||
694 | if (s->crypt_method) { | |
695 | if (!acb->cluster_data) { | |
696 | acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS * | |
697 | s->cluster_size); | |
698 | } | |
699 | ||
700 | assert(acb->hd_qiov.size <= QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); | |
701 | qemu_iovec_to_buffer(&acb->hd_qiov, acb->cluster_data); | |
702 | ||
703 | qcow2_encrypt_sectors(s, acb->sector_num, acb->cluster_data, | |
704 | acb->cluster_data, acb->cur_nr_sectors, 1, &s->aes_encrypt_key); | |
705 | ||
706 | qemu_iovec_reset(&acb->hd_qiov); | |
707 | qemu_iovec_add(&acb->hd_qiov, acb->cluster_data, | |
708 | acb->cur_nr_sectors * 512); | |
709 | } | |
710 | ||
711 | BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); | |
712 | acb->hd_aiocb = bdrv_aio_writev(bs->file, | |
713 | (acb->cluster_offset >> 9) + index_in_cluster, | |
714 | &acb->hd_qiov, acb->cur_nr_sectors, | |
715 | qcow2_aio_write_cb, acb); | |
716 | if (acb->hd_aiocb == NULL) { | |
717 | ret = -EIO; | |
718 | goto fail; | |
719 | } | |
720 | ||
721 | return; | |
722 | ||
723 | fail: | |
724 | if (acb->l2meta.nb_clusters != 0) { | |
725 | QLIST_REMOVE(&acb->l2meta, next_in_flight); | |
726 | } | |
727 | done: | |
728 | acb->common.cb(acb->common.opaque, ret); | |
729 | qemu_iovec_destroy(&acb->hd_qiov); | |
730 | qemu_aio_release(acb); | |
731 | } | |
732 | ||
733 | static BlockDriverAIOCB *qcow2_aio_writev(BlockDriverState *bs, | |
734 | int64_t sector_num, | |
735 | QEMUIOVector *qiov, int nb_sectors, | |
736 | BlockDriverCompletionFunc *cb, | |
737 | void *opaque) | |
738 | { | |
739 | BDRVQcowState *s = bs->opaque; | |
740 | QCowAIOCB *acb; | |
741 | int ret; | |
742 | ||
743 | s->cluster_cache_offset = -1; /* disable compressed cache */ | |
744 | ||
745 | acb = qcow2_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1); | |
746 | if (!acb) | |
747 | return NULL; | |
748 | ||
749 | ret = qcow2_schedule_bh(qcow2_aio_rw_bh, acb); | |
750 | if (ret < 0) { | |
751 | qemu_iovec_destroy(&acb->hd_qiov); | |
752 | qemu_aio_release(acb); | |
753 | return NULL; | |
754 | } | |
755 | ||
756 | return &acb->common; | |
757 | } | |
758 | ||
759 | static void qcow2_close(BlockDriverState *bs) | |
760 | { | |
761 | BDRVQcowState *s = bs->opaque; | |
762 | qemu_free(s->l1_table); | |
763 | ||
764 | qcow2_cache_flush(bs, s->l2_table_cache); | |
765 | qcow2_cache_flush(bs, s->refcount_block_cache); | |
766 | ||
767 | qcow2_cache_destroy(bs, s->l2_table_cache); | |
768 | qcow2_cache_destroy(bs, s->refcount_block_cache); | |
769 | ||
770 | qemu_free(s->cluster_cache); | |
771 | qemu_free(s->cluster_data); | |
772 | qcow2_refcount_close(bs); | |
773 | } | |
774 | ||
775 | /* | |
776 | * Updates the variable length parts of the qcow2 header, i.e. the backing file | |
777 | * name and all extensions. qcow2 was not designed to allow such changes, so if | |
778 | * we run out of space (we can only use the first cluster) this function may | |
779 | * fail. | |
780 | * | |
781 | * Returns 0 on success, -errno in error cases. | |
782 | */ | |
783 | static int qcow2_update_ext_header(BlockDriverState *bs, | |
784 | const char *backing_file, const char *backing_fmt) | |
785 | { | |
786 | size_t backing_file_len = 0; | |
787 | size_t backing_fmt_len = 0; | |
788 | BDRVQcowState *s = bs->opaque; | |
789 | QCowExtension ext_backing_fmt = {0, 0}; | |
790 | int ret; | |
791 | ||
792 | /* Backing file format doesn't make sense without a backing file */ | |
793 | if (backing_fmt && !backing_file) { | |
794 | return -EINVAL; | |
795 | } | |
796 | ||
797 | /* Prepare the backing file format extension if needed */ | |
798 | if (backing_fmt) { | |
799 | ext_backing_fmt.len = cpu_to_be32(strlen(backing_fmt)); | |
800 | ext_backing_fmt.magic = cpu_to_be32(QCOW2_EXT_MAGIC_BACKING_FORMAT); | |
801 | backing_fmt_len = ((sizeof(ext_backing_fmt) | |
802 | + strlen(backing_fmt) + 7) & ~7); | |
803 | } | |
804 | ||
805 | /* Check if we can fit the new header into the first cluster */ | |
806 | if (backing_file) { | |
807 | backing_file_len = strlen(backing_file); | |
808 | } | |
809 | ||
810 | size_t header_size = sizeof(QCowHeader) + backing_file_len | |
811 | + backing_fmt_len; | |
812 | ||
813 | if (header_size > s->cluster_size) { | |
814 | return -ENOSPC; | |
815 | } | |
816 | ||
817 | /* Rewrite backing file name and qcow2 extensions */ | |
818 | size_t ext_size = header_size - sizeof(QCowHeader); | |
819 | uint8_t buf[ext_size]; | |
820 | size_t offset = 0; | |
821 | size_t backing_file_offset = 0; | |
822 | ||
823 | if (backing_file) { | |
824 | if (backing_fmt) { | |
825 | int padding = backing_fmt_len - | |
826 | (sizeof(ext_backing_fmt) + strlen(backing_fmt)); | |
827 | ||
828 | memcpy(buf + offset, &ext_backing_fmt, sizeof(ext_backing_fmt)); | |
829 | offset += sizeof(ext_backing_fmt); | |
830 | ||
831 | memcpy(buf + offset, backing_fmt, strlen(backing_fmt)); | |
832 | offset += strlen(backing_fmt); | |
833 | ||
834 | memset(buf + offset, 0, padding); | |
835 | offset += padding; | |
836 | } | |
837 | ||
838 | memcpy(buf + offset, backing_file, backing_file_len); | |
839 | backing_file_offset = sizeof(QCowHeader) + offset; | |
840 | } | |
841 | ||
842 | ret = bdrv_pwrite_sync(bs->file, sizeof(QCowHeader), buf, ext_size); | |
843 | if (ret < 0) { | |
844 | goto fail; | |
845 | } | |
846 | ||
847 | /* Update header fields */ | |
848 | uint64_t be_backing_file_offset = cpu_to_be64(backing_file_offset); | |
849 | uint32_t be_backing_file_size = cpu_to_be32(backing_file_len); | |
850 | ||
851 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_offset), | |
852 | &be_backing_file_offset, sizeof(uint64_t)); | |
853 | if (ret < 0) { | |
854 | goto fail; | |
855 | } | |
856 | ||
857 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, backing_file_size), | |
858 | &be_backing_file_size, sizeof(uint32_t)); | |
859 | if (ret < 0) { | |
860 | goto fail; | |
861 | } | |
862 | ||
863 | ret = 0; | |
864 | fail: | |
865 | return ret; | |
866 | } | |
867 | ||
868 | static int qcow2_change_backing_file(BlockDriverState *bs, | |
869 | const char *backing_file, const char *backing_fmt) | |
870 | { | |
871 | return qcow2_update_ext_header(bs, backing_file, backing_fmt); | |
872 | } | |
873 | ||
874 | static int preallocate(BlockDriverState *bs) | |
875 | { | |
876 | uint64_t nb_sectors; | |
877 | uint64_t offset; | |
878 | int num; | |
879 | int ret; | |
880 | QCowL2Meta meta; | |
881 | ||
882 | nb_sectors = bdrv_getlength(bs) >> 9; | |
883 | offset = 0; | |
884 | QLIST_INIT(&meta.dependent_requests); | |
885 | meta.cluster_offset = 0; | |
886 | ||
887 | while (nb_sectors) { | |
888 | num = MIN(nb_sectors, INT_MAX >> 9); | |
889 | ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta); | |
890 | if (ret < 0) { | |
891 | return ret; | |
892 | } | |
893 | ||
894 | ret = qcow2_alloc_cluster_link_l2(bs, &meta); | |
895 | if (ret < 0) { | |
896 | qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters); | |
897 | return ret; | |
898 | } | |
899 | ||
900 | /* There are no dependent requests, but we need to remove our request | |
901 | * from the list of in-flight requests */ | |
902 | run_dependent_requests(&meta); | |
903 | ||
904 | /* TODO Preallocate data if requested */ | |
905 | ||
906 | nb_sectors -= num; | |
907 | offset += num << 9; | |
908 | } | |
909 | ||
910 | /* | |
911 | * It is expected that the image file is large enough to actually contain | |
912 | * all of the allocated clusters (otherwise we get failing reads after | |
913 | * EOF). Extend the image to the last allocated sector. | |
914 | */ | |
915 | if (meta.cluster_offset != 0) { | |
916 | uint8_t buf[512]; | |
917 | memset(buf, 0, 512); | |
918 | ret = bdrv_write(bs->file, (meta.cluster_offset >> 9) + num - 1, buf, 1); | |
919 | if (ret < 0) { | |
920 | return ret; | |
921 | } | |
922 | } | |
923 | ||
924 | return 0; | |
925 | } | |
926 | ||
927 | static int qcow2_create2(const char *filename, int64_t total_size, | |
928 | const char *backing_file, const char *backing_format, | |
929 | int flags, size_t cluster_size, int prealloc, | |
930 | QEMUOptionParameter *options) | |
931 | { | |
932 | /* Calulate cluster_bits */ | |
933 | int cluster_bits; | |
934 | cluster_bits = ffs(cluster_size) - 1; | |
935 | if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS || | |
936 | (1 << cluster_bits) != cluster_size) | |
937 | { | |
938 | error_report( | |
939 | "Cluster size must be a power of two between %d and %dk", | |
940 | 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); | |
941 | return -EINVAL; | |
942 | } | |
943 | ||
944 | /* | |
945 | * Open the image file and write a minimal qcow2 header. | |
946 | * | |
947 | * We keep things simple and start with a zero-sized image. We also | |
948 | * do without refcount blocks or a L1 table for now. We'll fix the | |
949 | * inconsistency later. | |
950 | * | |
951 | * We do need a refcount table because growing the refcount table means | |
952 | * allocating two new refcount blocks - the seconds of which would be at | |
953 | * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file | |
954 | * size for any qcow2 image. | |
955 | */ | |
956 | BlockDriverState* bs; | |
957 | QCowHeader header; | |
958 | uint8_t* refcount_table; | |
959 | int ret; | |
960 | ||
961 | ret = bdrv_create_file(filename, options); | |
962 | if (ret < 0) { | |
963 | return ret; | |
964 | } | |
965 | ||
966 | ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR); | |
967 | if (ret < 0) { | |
968 | return ret; | |
969 | } | |
970 | ||
971 | /* Write the header */ | |
972 | memset(&header, 0, sizeof(header)); | |
973 | header.magic = cpu_to_be32(QCOW_MAGIC); | |
974 | header.version = cpu_to_be32(QCOW_VERSION); | |
975 | header.cluster_bits = cpu_to_be32(cluster_bits); | |
976 | header.size = cpu_to_be64(0); | |
977 | header.l1_table_offset = cpu_to_be64(0); | |
978 | header.l1_size = cpu_to_be32(0); | |
979 | header.refcount_table_offset = cpu_to_be64(cluster_size); | |
980 | header.refcount_table_clusters = cpu_to_be32(1); | |
981 | ||
982 | if (flags & BLOCK_FLAG_ENCRYPT) { | |
983 | header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES); | |
984 | } else { | |
985 | header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); | |
986 | } | |
987 | ||
988 | ret = bdrv_pwrite(bs, 0, &header, sizeof(header)); | |
989 | if (ret < 0) { | |
990 | goto out; | |
991 | } | |
992 | ||
993 | /* Write an empty refcount table */ | |
994 | refcount_table = qemu_mallocz(cluster_size); | |
995 | ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size); | |
996 | qemu_free(refcount_table); | |
997 | ||
998 | if (ret < 0) { | |
999 | goto out; | |
1000 | } | |
1001 | ||
1002 | bdrv_close(bs); | |
1003 | ||
1004 | /* | |
1005 | * And now open the image and make it consistent first (i.e. increase the | |
1006 | * refcount of the cluster that is occupied by the header and the refcount | |
1007 | * table) | |
1008 | */ | |
1009 | BlockDriver* drv = bdrv_find_format("qcow2"); | |
1010 | assert(drv != NULL); | |
1011 | ret = bdrv_open(bs, filename, | |
1012 | BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv); | |
1013 | if (ret < 0) { | |
1014 | goto out; | |
1015 | } | |
1016 | ||
1017 | ret = qcow2_alloc_clusters(bs, 2 * cluster_size); | |
1018 | if (ret < 0) { | |
1019 | goto out; | |
1020 | ||
1021 | } else if (ret != 0) { | |
1022 | error_report("Huh, first cluster in empty image is already in use?"); | |
1023 | abort(); | |
1024 | } | |
1025 | ||
1026 | /* Okay, now that we have a valid image, let's give it the right size */ | |
1027 | ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE); | |
1028 | if (ret < 0) { | |
1029 | goto out; | |
1030 | } | |
1031 | ||
1032 | /* Want a backing file? There you go.*/ | |
1033 | if (backing_file) { | |
1034 | ret = bdrv_change_backing_file(bs, backing_file, backing_format); | |
1035 | if (ret < 0) { | |
1036 | goto out; | |
1037 | } | |
1038 | } | |
1039 | ||
1040 | /* And if we're supposed to preallocate metadata, do that now */ | |
1041 | if (prealloc) { | |
1042 | ret = preallocate(bs); | |
1043 | if (ret < 0) { | |
1044 | goto out; | |
1045 | } | |
1046 | } | |
1047 | ||
1048 | ret = 0; | |
1049 | out: | |
1050 | bdrv_delete(bs); | |
1051 | return ret; | |
1052 | } | |
1053 | ||
1054 | static int qcow2_create(const char *filename, QEMUOptionParameter *options) | |
1055 | { | |
1056 | const char *backing_file = NULL; | |
1057 | const char *backing_fmt = NULL; | |
1058 | uint64_t sectors = 0; | |
1059 | int flags = 0; | |
1060 | size_t cluster_size = DEFAULT_CLUSTER_SIZE; | |
1061 | int prealloc = 0; | |
1062 | ||
1063 | /* Read out options */ | |
1064 | while (options && options->name) { | |
1065 | if (!strcmp(options->name, BLOCK_OPT_SIZE)) { | |
1066 | sectors = options->value.n / 512; | |
1067 | } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { | |
1068 | backing_file = options->value.s; | |
1069 | } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) { | |
1070 | backing_fmt = options->value.s; | |
1071 | } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) { | |
1072 | flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0; | |
1073 | } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) { | |
1074 | if (options->value.n) { | |
1075 | cluster_size = options->value.n; | |
1076 | } | |
1077 | } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) { | |
1078 | if (!options->value.s || !strcmp(options->value.s, "off")) { | |
1079 | prealloc = 0; | |
1080 | } else if (!strcmp(options->value.s, "metadata")) { | |
1081 | prealloc = 1; | |
1082 | } else { | |
1083 | fprintf(stderr, "Invalid preallocation mode: '%s'\n", | |
1084 | options->value.s); | |
1085 | return -EINVAL; | |
1086 | } | |
1087 | } | |
1088 | options++; | |
1089 | } | |
1090 | ||
1091 | if (backing_file && prealloc) { | |
1092 | fprintf(stderr, "Backing file and preallocation cannot be used at " | |
1093 | "the same time\n"); | |
1094 | return -EINVAL; | |
1095 | } | |
1096 | ||
1097 | return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags, | |
1098 | cluster_size, prealloc, options); | |
1099 | } | |
1100 | ||
1101 | static int qcow2_make_empty(BlockDriverState *bs) | |
1102 | { | |
1103 | #if 0 | |
1104 | /* XXX: not correct */ | |
1105 | BDRVQcowState *s = bs->opaque; | |
1106 | uint32_t l1_length = s->l1_size * sizeof(uint64_t); | |
1107 | int ret; | |
1108 | ||
1109 | memset(s->l1_table, 0, l1_length); | |
1110 | if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0) | |
1111 | return -1; | |
1112 | ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length); | |
1113 | if (ret < 0) | |
1114 | return ret; | |
1115 | ||
1116 | l2_cache_reset(bs); | |
1117 | #endif | |
1118 | return 0; | |
1119 | } | |
1120 | ||
1121 | static int qcow2_discard(BlockDriverState *bs, int64_t sector_num, | |
1122 | int nb_sectors) | |
1123 | { | |
1124 | return qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS, | |
1125 | nb_sectors); | |
1126 | } | |
1127 | ||
1128 | static int qcow2_truncate(BlockDriverState *bs, int64_t offset) | |
1129 | { | |
1130 | BDRVQcowState *s = bs->opaque; | |
1131 | int ret, new_l1_size; | |
1132 | ||
1133 | if (offset & 511) { | |
1134 | return -EINVAL; | |
1135 | } | |
1136 | ||
1137 | /* cannot proceed if image has snapshots */ | |
1138 | if (s->nb_snapshots) { | |
1139 | return -ENOTSUP; | |
1140 | } | |
1141 | ||
1142 | /* shrinking is currently not supported */ | |
1143 | if (offset < bs->total_sectors * 512) { | |
1144 | return -ENOTSUP; | |
1145 | } | |
1146 | ||
1147 | new_l1_size = size_to_l1(s, offset); | |
1148 | ret = qcow2_grow_l1_table(bs, new_l1_size, true); | |
1149 | if (ret < 0) { | |
1150 | return ret; | |
1151 | } | |
1152 | ||
1153 | /* write updated header.size */ | |
1154 | offset = cpu_to_be64(offset); | |
1155 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), | |
1156 | &offset, sizeof(uint64_t)); | |
1157 | if (ret < 0) { | |
1158 | return ret; | |
1159 | } | |
1160 | ||
1161 | s->l1_vm_state_index = new_l1_size; | |
1162 | return 0; | |
1163 | } | |
1164 | ||
1165 | /* XXX: put compressed sectors first, then all the cluster aligned | |
1166 | tables to avoid losing bytes in alignment */ | |
1167 | static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num, | |
1168 | const uint8_t *buf, int nb_sectors) | |
1169 | { | |
1170 | BDRVQcowState *s = bs->opaque; | |
1171 | z_stream strm; | |
1172 | int ret, out_len; | |
1173 | uint8_t *out_buf; | |
1174 | uint64_t cluster_offset; | |
1175 | ||
1176 | if (nb_sectors == 0) { | |
1177 | /* align end of file to a sector boundary to ease reading with | |
1178 | sector based I/Os */ | |
1179 | cluster_offset = bdrv_getlength(bs->file); | |
1180 | cluster_offset = (cluster_offset + 511) & ~511; | |
1181 | bdrv_truncate(bs->file, cluster_offset); | |
1182 | return 0; | |
1183 | } | |
1184 | ||
1185 | if (nb_sectors != s->cluster_sectors) | |
1186 | return -EINVAL; | |
1187 | ||
1188 | out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128); | |
1189 | ||
1190 | /* best compression, small window, no zlib header */ | |
1191 | memset(&strm, 0, sizeof(strm)); | |
1192 | ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, | |
1193 | Z_DEFLATED, -12, | |
1194 | 9, Z_DEFAULT_STRATEGY); | |
1195 | if (ret != 0) { | |
1196 | qemu_free(out_buf); | |
1197 | return -1; | |
1198 | } | |
1199 | ||
1200 | strm.avail_in = s->cluster_size; | |
1201 | strm.next_in = (uint8_t *)buf; | |
1202 | strm.avail_out = s->cluster_size; | |
1203 | strm.next_out = out_buf; | |
1204 | ||
1205 | ret = deflate(&strm, Z_FINISH); | |
1206 | if (ret != Z_STREAM_END && ret != Z_OK) { | |
1207 | qemu_free(out_buf); | |
1208 | deflateEnd(&strm); | |
1209 | return -1; | |
1210 | } | |
1211 | out_len = strm.next_out - out_buf; | |
1212 | ||
1213 | deflateEnd(&strm); | |
1214 | ||
1215 | if (ret != Z_STREAM_END || out_len >= s->cluster_size) { | |
1216 | /* could not compress: write normal cluster */ | |
1217 | bdrv_write(bs, sector_num, buf, s->cluster_sectors); | |
1218 | } else { | |
1219 | cluster_offset = qcow2_alloc_compressed_cluster_offset(bs, | |
1220 | sector_num << 9, out_len); | |
1221 | if (!cluster_offset) | |
1222 | return -1; | |
1223 | cluster_offset &= s->cluster_offset_mask; | |
1224 | BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); | |
1225 | if (bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len) != out_len) { | |
1226 | qemu_free(out_buf); | |
1227 | return -1; | |
1228 | } | |
1229 | } | |
1230 | ||
1231 | qemu_free(out_buf); | |
1232 | return 0; | |
1233 | } | |
1234 | ||
1235 | static int qcow2_flush(BlockDriverState *bs) | |
1236 | { | |
1237 | BDRVQcowState *s = bs->opaque; | |
1238 | int ret; | |
1239 | ||
1240 | ret = qcow2_cache_flush(bs, s->l2_table_cache); | |
1241 | if (ret < 0) { | |
1242 | return ret; | |
1243 | } | |
1244 | ||
1245 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
1246 | if (ret < 0) { | |
1247 | return ret; | |
1248 | } | |
1249 | ||
1250 | return bdrv_flush(bs->file); | |
1251 | } | |
1252 | ||
1253 | static BlockDriverAIOCB *qcow2_aio_flush(BlockDriverState *bs, | |
1254 | BlockDriverCompletionFunc *cb, | |
1255 | void *opaque) | |
1256 | { | |
1257 | BDRVQcowState *s = bs->opaque; | |
1258 | int ret; | |
1259 | ||
1260 | ret = qcow2_cache_flush(bs, s->l2_table_cache); | |
1261 | if (ret < 0) { | |
1262 | return NULL; | |
1263 | } | |
1264 | ||
1265 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
1266 | if (ret < 0) { | |
1267 | return NULL; | |
1268 | } | |
1269 | ||
1270 | return bdrv_aio_flush(bs->file, cb, opaque); | |
1271 | } | |
1272 | ||
1273 | static int64_t qcow2_vm_state_offset(BDRVQcowState *s) | |
1274 | { | |
1275 | return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits); | |
1276 | } | |
1277 | ||
1278 | static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
1279 | { | |
1280 | BDRVQcowState *s = bs->opaque; | |
1281 | bdi->cluster_size = s->cluster_size; | |
1282 | bdi->vm_state_offset = qcow2_vm_state_offset(s); | |
1283 | return 0; | |
1284 | } | |
1285 | ||
1286 | ||
1287 | static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result) | |
1288 | { | |
1289 | return qcow2_check_refcounts(bs, result); | |
1290 | } | |
1291 | ||
1292 | #if 0 | |
1293 | static void dump_refcounts(BlockDriverState *bs) | |
1294 | { | |
1295 | BDRVQcowState *s = bs->opaque; | |
1296 | int64_t nb_clusters, k, k1, size; | |
1297 | int refcount; | |
1298 | ||
1299 | size = bdrv_getlength(bs->file); | |
1300 | nb_clusters = size_to_clusters(s, size); | |
1301 | for(k = 0; k < nb_clusters;) { | |
1302 | k1 = k; | |
1303 | refcount = get_refcount(bs, k); | |
1304 | k++; | |
1305 | while (k < nb_clusters && get_refcount(bs, k) == refcount) | |
1306 | k++; | |
1307 | printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount, | |
1308 | k - k1); | |
1309 | } | |
1310 | } | |
1311 | #endif | |
1312 | ||
1313 | static int qcow2_save_vmstate(BlockDriverState *bs, const uint8_t *buf, | |
1314 | int64_t pos, int size) | |
1315 | { | |
1316 | BDRVQcowState *s = bs->opaque; | |
1317 | int growable = bs->growable; | |
1318 | int ret; | |
1319 | ||
1320 | BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); | |
1321 | bs->growable = 1; | |
1322 | ret = bdrv_pwrite(bs, qcow2_vm_state_offset(s) + pos, buf, size); | |
1323 | bs->growable = growable; | |
1324 | ||
1325 | return ret; | |
1326 | } | |
1327 | ||
1328 | static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf, | |
1329 | int64_t pos, int size) | |
1330 | { | |
1331 | BDRVQcowState *s = bs->opaque; | |
1332 | int growable = bs->growable; | |
1333 | int ret; | |
1334 | ||
1335 | BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); | |
1336 | bs->growable = 1; | |
1337 | ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size); | |
1338 | bs->growable = growable; | |
1339 | ||
1340 | return ret; | |
1341 | } | |
1342 | ||
1343 | static QEMUOptionParameter qcow2_create_options[] = { | |
1344 | { | |
1345 | .name = BLOCK_OPT_SIZE, | |
1346 | .type = OPT_SIZE, | |
1347 | .help = "Virtual disk size" | |
1348 | }, | |
1349 | { | |
1350 | .name = BLOCK_OPT_BACKING_FILE, | |
1351 | .type = OPT_STRING, | |
1352 | .help = "File name of a base image" | |
1353 | }, | |
1354 | { | |
1355 | .name = BLOCK_OPT_BACKING_FMT, | |
1356 | .type = OPT_STRING, | |
1357 | .help = "Image format of the base image" | |
1358 | }, | |
1359 | { | |
1360 | .name = BLOCK_OPT_ENCRYPT, | |
1361 | .type = OPT_FLAG, | |
1362 | .help = "Encrypt the image" | |
1363 | }, | |
1364 | { | |
1365 | .name = BLOCK_OPT_CLUSTER_SIZE, | |
1366 | .type = OPT_SIZE, | |
1367 | .help = "qcow2 cluster size", | |
1368 | .value = { .n = DEFAULT_CLUSTER_SIZE }, | |
1369 | }, | |
1370 | { | |
1371 | .name = BLOCK_OPT_PREALLOC, | |
1372 | .type = OPT_STRING, | |
1373 | .help = "Preallocation mode (allowed values: off, metadata)" | |
1374 | }, | |
1375 | { NULL } | |
1376 | }; | |
1377 | ||
1378 | static BlockDriver bdrv_qcow2 = { | |
1379 | .format_name = "qcow2", | |
1380 | .instance_size = sizeof(BDRVQcowState), | |
1381 | .bdrv_probe = qcow2_probe, | |
1382 | .bdrv_open = qcow2_open, | |
1383 | .bdrv_close = qcow2_close, | |
1384 | .bdrv_create = qcow2_create, | |
1385 | .bdrv_flush = qcow2_flush, | |
1386 | .bdrv_is_allocated = qcow2_is_allocated, | |
1387 | .bdrv_set_key = qcow2_set_key, | |
1388 | .bdrv_make_empty = qcow2_make_empty, | |
1389 | ||
1390 | .bdrv_aio_readv = qcow2_aio_readv, | |
1391 | .bdrv_aio_writev = qcow2_aio_writev, | |
1392 | .bdrv_aio_flush = qcow2_aio_flush, | |
1393 | ||
1394 | .bdrv_discard = qcow2_discard, | |
1395 | .bdrv_truncate = qcow2_truncate, | |
1396 | .bdrv_write_compressed = qcow2_write_compressed, | |
1397 | ||
1398 | .bdrv_snapshot_create = qcow2_snapshot_create, | |
1399 | .bdrv_snapshot_goto = qcow2_snapshot_goto, | |
1400 | .bdrv_snapshot_delete = qcow2_snapshot_delete, | |
1401 | .bdrv_snapshot_list = qcow2_snapshot_list, | |
1402 | .bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, | |
1403 | .bdrv_get_info = qcow2_get_info, | |
1404 | ||
1405 | .bdrv_save_vmstate = qcow2_save_vmstate, | |
1406 | .bdrv_load_vmstate = qcow2_load_vmstate, | |
1407 | ||
1408 | .bdrv_change_backing_file = qcow2_change_backing_file, | |
1409 | ||
1410 | .create_options = qcow2_create_options, | |
1411 | .bdrv_check = qcow2_check, | |
1412 | }; | |
1413 | ||
1414 | static void bdrv_qcow2_init(void) | |
1415 | { | |
1416 | bdrv_register(&bdrv_qcow2); | |
1417 | } | |
1418 | ||
1419 | block_init(bdrv_qcow2_init); |