]>
Commit | Line | Data |
---|---|---|
585f8587 FB |
1 | /* |
2 | * Block driver for the QCOW version 2 format | |
5fafdf24 | 3 | * |
585f8587 | 4 | * Copyright (c) 2004-2006 Fabrice Bellard |
5fafdf24 | 5 | * |
585f8587 FB |
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 | */ | |
faf07963 | 24 | #include "qemu-common.h" |
585f8587 | 25 | #include "block_int.h" |
5efa9d5a | 26 | #include "module.h" |
585f8587 FB |
27 | #include <zlib.h> |
28 | #include "aes.h" | |
f7d0fe02 | 29 | #include "block/qcow2.h" |
585f8587 FB |
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. | |
5fafdf24 | 38 | - Size of compressed clusters is stored in sectors to reduce bit usage |
585f8587 FB |
39 | in the cluster offsets. |
40 | - Support for storing additional data (such as the VM state) in the | |
3b46e624 | 41 | snapshots. |
585f8587 FB |
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 | //#define DEBUG_ALLOC | |
48 | //#define DEBUG_ALLOC2 | |
9b80ddf3 | 49 | //#define DEBUG_EXT |
5fafdf24 | 50 | |
9b80ddf3 AL |
51 | |
52 | typedef struct { | |
53 | uint32_t magic; | |
54 | uint32_t len; | |
55 | } QCowExtension; | |
56 | #define QCOW_EXT_MAGIC_END 0 | |
f965509c | 57 | #define QCOW_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA |
9b80ddf3 AL |
58 | |
59 | ||
585f8587 FB |
60 | typedef struct __attribute__((packed)) QCowSnapshotHeader { |
61 | /* header is 8 byte aligned */ | |
62 | uint64_t l1_table_offset; | |
63 | ||
64 | uint32_t l1_size; | |
65 | uint16_t id_str_size; | |
66 | uint16_t name_size; | |
67 | ||
68 | uint32_t date_sec; | |
69 | uint32_t date_nsec; | |
70 | ||
71 | uint64_t vm_clock_nsec; | |
72 | ||
73 | uint32_t vm_state_size; | |
74 | uint32_t extra_data_size; /* for extension */ | |
75 | /* extra data follows */ | |
76 | /* id_str follows */ | |
77 | /* name follows */ | |
78 | } QCowSnapshotHeader; | |
79 | ||
585f8587 FB |
80 | |
81 | static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset); | |
5fafdf24 | 82 | static int qcow_read(BlockDriverState *bs, int64_t sector_num, |
585f8587 FB |
83 | uint8_t *buf, int nb_sectors); |
84 | static int qcow_read_snapshots(BlockDriverState *bs); | |
85 | static void qcow_free_snapshots(BlockDriverState *bs); | |
585f8587 FB |
86 | |
87 | static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename) | |
88 | { | |
89 | const QCowHeader *cow_header = (const void *)buf; | |
3b46e624 | 90 | |
585f8587 FB |
91 | if (buf_size >= sizeof(QCowHeader) && |
92 | be32_to_cpu(cow_header->magic) == QCOW_MAGIC && | |
5fafdf24 | 93 | be32_to_cpu(cow_header->version) == QCOW_VERSION) |
585f8587 FB |
94 | return 100; |
95 | else | |
96 | return 0; | |
97 | } | |
98 | ||
9b80ddf3 AL |
99 | |
100 | /* | |
101 | * read qcow2 extension and fill bs | |
102 | * start reading from start_offset | |
103 | * finish reading upon magic of value 0 or when end_offset reached | |
104 | * unknown magic is skipped (future extension this version knows nothing about) | |
105 | * return 0 upon success, non-0 otherwise | |
106 | */ | |
107 | static int qcow_read_extensions(BlockDriverState *bs, uint64_t start_offset, | |
108 | uint64_t end_offset) | |
109 | { | |
110 | BDRVQcowState *s = bs->opaque; | |
111 | QCowExtension ext; | |
112 | uint64_t offset; | |
113 | ||
114 | #ifdef DEBUG_EXT | |
115 | printf("qcow_read_extensions: start=%ld end=%ld\n", start_offset, end_offset); | |
116 | #endif | |
117 | offset = start_offset; | |
118 | while (offset < end_offset) { | |
119 | ||
120 | #ifdef DEBUG_EXT | |
121 | /* Sanity check */ | |
122 | if (offset > s->cluster_size) | |
123 | printf("qcow_handle_extension: suspicious offset %lu\n", offset); | |
124 | ||
125 | printf("attemting to read extended header in offset %lu\n", offset); | |
126 | #endif | |
127 | ||
128 | if (bdrv_pread(s->hd, offset, &ext, sizeof(ext)) != sizeof(ext)) { | |
4c978075 AL |
129 | fprintf(stderr, "qcow_handle_extension: ERROR: pread fail from offset %llu\n", |
130 | (unsigned long long)offset); | |
9b80ddf3 AL |
131 | return 1; |
132 | } | |
133 | be32_to_cpus(&ext.magic); | |
134 | be32_to_cpus(&ext.len); | |
135 | offset += sizeof(ext); | |
136 | #ifdef DEBUG_EXT | |
137 | printf("ext.magic = 0x%x\n", ext.magic); | |
138 | #endif | |
139 | switch (ext.magic) { | |
140 | case QCOW_EXT_MAGIC_END: | |
141 | return 0; | |
f965509c AL |
142 | |
143 | case QCOW_EXT_MAGIC_BACKING_FORMAT: | |
144 | if (ext.len >= sizeof(bs->backing_format)) { | |
145 | fprintf(stderr, "ERROR: ext_backing_format: len=%u too large" | |
4c978075 | 146 | " (>=%zu)\n", |
f965509c AL |
147 | ext.len, sizeof(bs->backing_format)); |
148 | return 2; | |
149 | } | |
150 | if (bdrv_pread(s->hd, offset , bs->backing_format, | |
151 | ext.len) != ext.len) | |
152 | return 3; | |
153 | bs->backing_format[ext.len] = '\0'; | |
154 | #ifdef DEBUG_EXT | |
155 | printf("Qcow2: Got format extension %s\n", bs->backing_format); | |
156 | #endif | |
157 | offset += ((ext.len + 7) & ~7); | |
158 | break; | |
159 | ||
9b80ddf3 AL |
160 | default: |
161 | /* unknown magic -- just skip it */ | |
162 | offset += ((ext.len + 7) & ~7); | |
163 | break; | |
164 | } | |
165 | } | |
166 | ||
167 | return 0; | |
168 | } | |
169 | ||
170 | ||
585f8587 FB |
171 | static int qcow_open(BlockDriverState *bs, const char *filename, int flags) |
172 | { | |
173 | BDRVQcowState *s = bs->opaque; | |
174 | int len, i, shift, ret; | |
175 | QCowHeader header; | |
9b80ddf3 | 176 | uint64_t ext_end; |
585f8587 | 177 | |
4dc822d7 AL |
178 | /* Performance is terrible right now with cache=writethrough due mainly |
179 | * to reference count updates. If the user does not explicitly specify | |
180 | * a caching type, force to writeback caching. | |
181 | */ | |
182 | if ((flags & BDRV_O_CACHE_DEF)) { | |
183 | flags |= BDRV_O_CACHE_WB; | |
184 | flags &= ~BDRV_O_CACHE_DEF; | |
185 | } | |
b5eff355 | 186 | ret = bdrv_file_open(&s->hd, filename, flags); |
585f8587 FB |
187 | if (ret < 0) |
188 | return ret; | |
189 | if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header)) | |
190 | goto fail; | |
191 | be32_to_cpus(&header.magic); | |
192 | be32_to_cpus(&header.version); | |
193 | be64_to_cpus(&header.backing_file_offset); | |
194 | be32_to_cpus(&header.backing_file_size); | |
195 | be64_to_cpus(&header.size); | |
196 | be32_to_cpus(&header.cluster_bits); | |
197 | be32_to_cpus(&header.crypt_method); | |
198 | be64_to_cpus(&header.l1_table_offset); | |
199 | be32_to_cpus(&header.l1_size); | |
200 | be64_to_cpus(&header.refcount_table_offset); | |
201 | be32_to_cpus(&header.refcount_table_clusters); | |
202 | be64_to_cpus(&header.snapshots_offset); | |
203 | be32_to_cpus(&header.nb_snapshots); | |
3b46e624 | 204 | |
585f8587 FB |
205 | if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION) |
206 | goto fail; | |
5fafdf24 | 207 | if (header.size <= 1 || |
73c632ed KW |
208 | header.cluster_bits < MIN_CLUSTER_BITS || |
209 | header.cluster_bits > MAX_CLUSTER_BITS) | |
585f8587 FB |
210 | goto fail; |
211 | if (header.crypt_method > QCOW_CRYPT_AES) | |
212 | goto fail; | |
213 | s->crypt_method_header = header.crypt_method; | |
214 | if (s->crypt_method_header) | |
215 | bs->encrypted = 1; | |
216 | s->cluster_bits = header.cluster_bits; | |
217 | s->cluster_size = 1 << s->cluster_bits; | |
218 | s->cluster_sectors = 1 << (s->cluster_bits - 9); | |
219 | s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ | |
220 | s->l2_size = 1 << s->l2_bits; | |
221 | bs->total_sectors = header.size / 512; | |
222 | s->csize_shift = (62 - (s->cluster_bits - 8)); | |
223 | s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; | |
224 | s->cluster_offset_mask = (1LL << s->csize_shift) - 1; | |
225 | s->refcount_table_offset = header.refcount_table_offset; | |
5fafdf24 | 226 | s->refcount_table_size = |
585f8587 FB |
227 | header.refcount_table_clusters << (s->cluster_bits - 3); |
228 | ||
229 | s->snapshots_offset = header.snapshots_offset; | |
230 | s->nb_snapshots = header.nb_snapshots; | |
231 | ||
232 | /* read the level 1 table */ | |
233 | s->l1_size = header.l1_size; | |
234 | shift = s->cluster_bits + s->l2_bits; | |
235 | s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift; | |
236 | /* the L1 table must contain at least enough entries to put | |
237 | header.size bytes */ | |
238 | if (s->l1_size < s->l1_vm_state_index) | |
239 | goto fail; | |
240 | s->l1_table_offset = header.l1_table_offset; | |
241 | s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t)); | |
5fafdf24 | 242 | if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) != |
585f8587 FB |
243 | s->l1_size * sizeof(uint64_t)) |
244 | goto fail; | |
245 | for(i = 0;i < s->l1_size; i++) { | |
246 | be64_to_cpus(&s->l1_table[i]); | |
247 | } | |
248 | /* alloc L2 cache */ | |
249 | s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); | |
585f8587 | 250 | s->cluster_cache = qemu_malloc(s->cluster_size); |
585f8587 | 251 | /* one more sector for decompressed data alignment */ |
095a9c58 AL |
252 | s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size |
253 | + 512); | |
585f8587 | 254 | s->cluster_cache_offset = -1; |
3b46e624 | 255 | |
585f8587 FB |
256 | if (refcount_init(bs) < 0) |
257 | goto fail; | |
258 | ||
9b80ddf3 AL |
259 | /* read qcow2 extensions */ |
260 | if (header.backing_file_offset) | |
261 | ext_end = header.backing_file_offset; | |
262 | else | |
263 | ext_end = s->cluster_size; | |
264 | if (qcow_read_extensions(bs, sizeof(header), ext_end)) | |
265 | goto fail; | |
266 | ||
585f8587 FB |
267 | /* read the backing file name */ |
268 | if (header.backing_file_offset != 0) { | |
269 | len = header.backing_file_size; | |
270 | if (len > 1023) | |
271 | len = 1023; | |
272 | if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len) | |
273 | goto fail; | |
274 | bs->backing_file[len] = '\0'; | |
275 | } | |
276 | if (qcow_read_snapshots(bs) < 0) | |
277 | goto fail; | |
278 | ||
279 | #ifdef DEBUG_ALLOC | |
280 | check_refcounts(bs); | |
281 | #endif | |
282 | return 0; | |
283 | ||
284 | fail: | |
285 | qcow_free_snapshots(bs); | |
286 | refcount_close(bs); | |
287 | qemu_free(s->l1_table); | |
288 | qemu_free(s->l2_cache); | |
289 | qemu_free(s->cluster_cache); | |
290 | qemu_free(s->cluster_data); | |
291 | bdrv_delete(s->hd); | |
292 | return -1; | |
293 | } | |
294 | ||
295 | static int qcow_set_key(BlockDriverState *bs, const char *key) | |
296 | { | |
297 | BDRVQcowState *s = bs->opaque; | |
298 | uint8_t keybuf[16]; | |
299 | int len, i; | |
3b46e624 | 300 | |
585f8587 FB |
301 | memset(keybuf, 0, 16); |
302 | len = strlen(key); | |
303 | if (len > 16) | |
304 | len = 16; | |
305 | /* XXX: we could compress the chars to 7 bits to increase | |
306 | entropy */ | |
307 | for(i = 0;i < len;i++) { | |
308 | keybuf[i] = key[i]; | |
309 | } | |
310 | s->crypt_method = s->crypt_method_header; | |
311 | ||
312 | if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0) | |
313 | return -1; | |
314 | if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0) | |
315 | return -1; | |
316 | #if 0 | |
317 | /* test */ | |
318 | { | |
319 | uint8_t in[16]; | |
320 | uint8_t out[16]; | |
321 | uint8_t tmp[16]; | |
322 | for(i=0;i<16;i++) | |
323 | in[i] = i; | |
324 | AES_encrypt(in, tmp, &s->aes_encrypt_key); | |
325 | AES_decrypt(tmp, out, &s->aes_decrypt_key); | |
326 | for(i = 0; i < 16; i++) | |
327 | printf(" %02x", tmp[i]); | |
328 | printf("\n"); | |
329 | for(i = 0; i < 16; i++) | |
330 | printf(" %02x", out[i]); | |
331 | printf("\n"); | |
332 | } | |
333 | #endif | |
334 | return 0; | |
335 | } | |
336 | ||
337 | /* The crypt function is compatible with the linux cryptoloop | |
338 | algorithm for < 4 GB images. NOTE: out_buf == in_buf is | |
339 | supported */ | |
340 | static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num, | |
341 | uint8_t *out_buf, const uint8_t *in_buf, | |
342 | int nb_sectors, int enc, | |
343 | const AES_KEY *key) | |
344 | { | |
345 | union { | |
346 | uint64_t ll[2]; | |
347 | uint8_t b[16]; | |
348 | } ivec; | |
349 | int i; | |
350 | ||
351 | for(i = 0; i < nb_sectors; i++) { | |
352 | ivec.ll[0] = cpu_to_le64(sector_num); | |
353 | ivec.ll[1] = 0; | |
5fafdf24 | 354 | AES_cbc_encrypt(in_buf, out_buf, 512, key, |
585f8587 FB |
355 | ivec.b, enc); |
356 | sector_num++; | |
357 | in_buf += 512; | |
358 | out_buf += 512; | |
359 | } | |
360 | } | |
361 | ||
362 | static int copy_sectors(BlockDriverState *bs, uint64_t start_sect, | |
363 | uint64_t cluster_offset, int n_start, int n_end) | |
364 | { | |
365 | BDRVQcowState *s = bs->opaque; | |
366 | int n, ret; | |
367 | ||
368 | n = n_end - n_start; | |
369 | if (n <= 0) | |
370 | return 0; | |
371 | ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n); | |
372 | if (ret < 0) | |
373 | return ret; | |
374 | if (s->crypt_method) { | |
5fafdf24 TS |
375 | encrypt_sectors(s, start_sect + n_start, |
376 | s->cluster_data, | |
585f8587 FB |
377 | s->cluster_data, n, 1, |
378 | &s->aes_encrypt_key); | |
379 | } | |
5fafdf24 | 380 | ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start, |
585f8587 FB |
381 | s->cluster_data, n); |
382 | if (ret < 0) | |
383 | return ret; | |
384 | return 0; | |
385 | } | |
386 | ||
f7d0fe02 | 387 | void l2_cache_reset(BlockDriverState *bs) |
585f8587 FB |
388 | { |
389 | BDRVQcowState *s = bs->opaque; | |
390 | ||
391 | memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); | |
392 | memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t)); | |
393 | memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t)); | |
394 | } | |
395 | ||
396 | static inline int l2_cache_new_entry(BlockDriverState *bs) | |
397 | { | |
398 | BDRVQcowState *s = bs->opaque; | |
399 | uint32_t min_count; | |
400 | int min_index, i; | |
401 | ||
402 | /* find a new entry in the least used one */ | |
403 | min_index = 0; | |
404 | min_count = 0xffffffff; | |
405 | for(i = 0; i < L2_CACHE_SIZE; i++) { | |
406 | if (s->l2_cache_counts[i] < min_count) { | |
407 | min_count = s->l2_cache_counts[i]; | |
408 | min_index = i; | |
409 | } | |
410 | } | |
411 | return min_index; | |
412 | } | |
413 | ||
414 | static int64_t align_offset(int64_t offset, int n) | |
415 | { | |
416 | offset = (offset + n - 1) & ~(n - 1); | |
417 | return offset; | |
418 | } | |
419 | ||
420 | static int grow_l1_table(BlockDriverState *bs, int min_size) | |
421 | { | |
422 | BDRVQcowState *s = bs->opaque; | |
423 | int new_l1_size, new_l1_size2, ret, i; | |
424 | uint64_t *new_l1_table; | |
425 | uint64_t new_l1_table_offset; | |
643e5399 | 426 | uint8_t data[12]; |
585f8587 FB |
427 | |
428 | new_l1_size = s->l1_size; | |
429 | if (min_size <= new_l1_size) | |
430 | return 0; | |
431 | while (min_size > new_l1_size) { | |
432 | new_l1_size = (new_l1_size * 3 + 1) / 2; | |
433 | } | |
434 | #ifdef DEBUG_ALLOC2 | |
435 | printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size); | |
436 | #endif | |
437 | ||
438 | new_l1_size2 = sizeof(uint64_t) * new_l1_size; | |
439 | new_l1_table = qemu_mallocz(new_l1_size2); | |
585f8587 FB |
440 | memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t)); |
441 | ||
442 | /* write new table (align to cluster) */ | |
443 | new_l1_table_offset = alloc_clusters(bs, new_l1_size2); | |
3b46e624 | 444 | |
585f8587 FB |
445 | for(i = 0; i < s->l1_size; i++) |
446 | new_l1_table[i] = cpu_to_be64(new_l1_table[i]); | |
447 | ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2); | |
448 | if (ret != new_l1_size2) | |
449 | goto fail; | |
450 | for(i = 0; i < s->l1_size; i++) | |
451 | new_l1_table[i] = be64_to_cpu(new_l1_table[i]); | |
3b46e624 | 452 | |
585f8587 | 453 | /* set new table */ |
643e5399 AL |
454 | cpu_to_be32w((uint32_t*)data, new_l1_size); |
455 | cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset); | |
456 | if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data, | |
457 | sizeof(data)) != sizeof(data)) | |
585f8587 FB |
458 | goto fail; |
459 | qemu_free(s->l1_table); | |
460 | free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t)); | |
461 | s->l1_table_offset = new_l1_table_offset; | |
462 | s->l1_table = new_l1_table; | |
463 | s->l1_size = new_l1_size; | |
464 | return 0; | |
465 | fail: | |
466 | qemu_free(s->l1_table); | |
467 | return -EIO; | |
468 | } | |
469 | ||
108534b9 AL |
470 | /* |
471 | * seek_l2_table | |
585f8587 | 472 | * |
108534b9 AL |
473 | * seek l2_offset in the l2_cache table |
474 | * if not found, return NULL, | |
475 | * if found, | |
476 | * increments the l2 cache hit count of the entry, | |
477 | * if counter overflow, divide by two all counters | |
478 | * return the pointer to the l2 cache entry | |
585f8587 | 479 | * |
108534b9 AL |
480 | */ |
481 | ||
482 | static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset) | |
483 | { | |
484 | int i, j; | |
485 | ||
486 | for(i = 0; i < L2_CACHE_SIZE; i++) { | |
487 | if (l2_offset == s->l2_cache_offsets[i]) { | |
488 | /* increment the hit count */ | |
489 | if (++s->l2_cache_counts[i] == 0xffffffff) { | |
490 | for(j = 0; j < L2_CACHE_SIZE; j++) { | |
491 | s->l2_cache_counts[j] >>= 1; | |
492 | } | |
493 | } | |
494 | return s->l2_cache + (i << s->l2_bits); | |
495 | } | |
496 | } | |
497 | return NULL; | |
498 | } | |
499 | ||
500 | /* | |
501 | * l2_load | |
502 | * | |
503 | * Loads a L2 table into memory. If the table is in the cache, the cache | |
504 | * is used; otherwise the L2 table is loaded from the image file. | |
505 | * | |
506 | * Returns a pointer to the L2 table on success, or NULL if the read from | |
507 | * the image file failed. | |
508 | */ | |
509 | ||
510 | static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset) | |
511 | { | |
512 | BDRVQcowState *s = bs->opaque; | |
513 | int min_index; | |
514 | uint64_t *l2_table; | |
515 | ||
516 | /* seek if the table for the given offset is in the cache */ | |
517 | ||
518 | l2_table = seek_l2_table(s, l2_offset); | |
519 | if (l2_table != NULL) | |
520 | return l2_table; | |
521 | ||
522 | /* not found: load a new entry in the least used one */ | |
523 | ||
524 | min_index = l2_cache_new_entry(bs); | |
525 | l2_table = s->l2_cache + (min_index << s->l2_bits); | |
526 | if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) != | |
527 | s->l2_size * sizeof(uint64_t)) | |
528 | return NULL; | |
529 | s->l2_cache_offsets[min_index] = l2_offset; | |
530 | s->l2_cache_counts[min_index] = 1; | |
531 | ||
532 | return l2_table; | |
533 | } | |
534 | ||
535 | /* | |
536 | * l2_allocate | |
585f8587 | 537 | * |
108534b9 AL |
538 | * Allocate a new l2 entry in the file. If l1_index points to an already |
539 | * used entry in the L2 table (i.e. we are doing a copy on write for the L2 | |
540 | * table) copy the contents of the old L2 table into the newly allocated one. | |
541 | * Otherwise the new table is initialized with zeros. | |
585f8587 | 542 | * |
585f8587 | 543 | */ |
108534b9 AL |
544 | |
545 | static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index) | |
546 | { | |
547 | BDRVQcowState *s = bs->opaque; | |
548 | int min_index; | |
549 | uint64_t old_l2_offset, tmp; | |
550 | uint64_t *l2_table, l2_offset; | |
551 | ||
552 | old_l2_offset = s->l1_table[l1_index]; | |
553 | ||
554 | /* allocate a new l2 entry */ | |
555 | ||
556 | l2_offset = alloc_clusters(bs, s->l2_size * sizeof(uint64_t)); | |
557 | ||
558 | /* update the L1 entry */ | |
559 | ||
560 | s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; | |
561 | ||
562 | tmp = cpu_to_be64(l2_offset | QCOW_OFLAG_COPIED); | |
563 | if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp), | |
564 | &tmp, sizeof(tmp)) != sizeof(tmp)) | |
565 | return NULL; | |
566 | ||
567 | /* allocate a new entry in the l2 cache */ | |
568 | ||
569 | min_index = l2_cache_new_entry(bs); | |
570 | l2_table = s->l2_cache + (min_index << s->l2_bits); | |
571 | ||
572 | if (old_l2_offset == 0) { | |
573 | /* if there was no old l2 table, clear the new table */ | |
574 | memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); | |
575 | } else { | |
576 | /* if there was an old l2 table, read it from the disk */ | |
577 | if (bdrv_pread(s->hd, old_l2_offset, | |
578 | l2_table, s->l2_size * sizeof(uint64_t)) != | |
579 | s->l2_size * sizeof(uint64_t)) | |
580 | return NULL; | |
581 | } | |
582 | /* write the l2 table to the file */ | |
583 | if (bdrv_pwrite(s->hd, l2_offset, | |
584 | l2_table, s->l2_size * sizeof(uint64_t)) != | |
585 | s->l2_size * sizeof(uint64_t)) | |
586 | return NULL; | |
587 | ||
588 | /* update the l2 cache entry */ | |
589 | ||
590 | s->l2_cache_offsets[min_index] = l2_offset; | |
591 | s->l2_cache_counts[min_index] = 1; | |
592 | ||
593 | return l2_table; | |
594 | } | |
595 | ||
6db6c638 | 596 | static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size, |
ff4b91c2 | 597 | uint64_t *l2_table, uint64_t start, uint64_t mask) |
6db6c638 AL |
598 | { |
599 | int i; | |
600 | uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask; | |
601 | ||
ab5ccbd6 AL |
602 | if (!offset) |
603 | return 0; | |
604 | ||
ff4b91c2 | 605 | for (i = start; i < start + nb_clusters; i++) |
6db6c638 AL |
606 | if (offset + i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask)) |
607 | break; | |
608 | ||
ff4b91c2 | 609 | return (i - start); |
6db6c638 AL |
610 | } |
611 | ||
612 | static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table) | |
613 | { | |
614 | int i = 0; | |
615 | ||
616 | while(nb_clusters-- && l2_table[i] == 0) | |
617 | i++; | |
618 | ||
619 | return i; | |
620 | } | |
621 | ||
05203524 AL |
622 | /* |
623 | * get_cluster_offset | |
624 | * | |
625 | * For a given offset of the disk image, return cluster offset in | |
626 | * qcow2 file. | |
627 | * | |
095a9c58 AL |
628 | * on entry, *num is the number of contiguous clusters we'd like to |
629 | * access following offset. | |
630 | * | |
631 | * on exit, *num is the number of contiguous clusters we can read. | |
632 | * | |
05203524 AL |
633 | * Return 1, if the offset is found |
634 | * Return 0, otherwise. | |
635 | * | |
636 | */ | |
637 | ||
095a9c58 AL |
638 | static uint64_t get_cluster_offset(BlockDriverState *bs, |
639 | uint64_t offset, int *num) | |
05203524 AL |
640 | { |
641 | BDRVQcowState *s = bs->opaque; | |
642 | int l1_index, l2_index; | |
6db6c638 AL |
643 | uint64_t l2_offset, *l2_table, cluster_offset; |
644 | int l1_bits, c; | |
645 | int index_in_cluster, nb_available, nb_needed, nb_clusters; | |
095a9c58 AL |
646 | |
647 | index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1); | |
648 | nb_needed = *num + index_in_cluster; | |
649 | ||
650 | l1_bits = s->l2_bits + s->cluster_bits; | |
651 | ||
652 | /* compute how many bytes there are between the offset and | |
6db6c638 | 653 | * the end of the l1 entry |
095a9c58 AL |
654 | */ |
655 | ||
656 | nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1)); | |
657 | ||
658 | /* compute the number of available sectors */ | |
659 | ||
660 | nb_available = (nb_available >> 9) + index_in_cluster; | |
661 | ||
f8de1660 AL |
662 | if (nb_needed > nb_available) { |
663 | nb_needed = nb_available; | |
664 | } | |
665 | ||
095a9c58 | 666 | cluster_offset = 0; |
05203524 AL |
667 | |
668 | /* seek the the l2 offset in the l1 table */ | |
669 | ||
095a9c58 | 670 | l1_index = offset >> l1_bits; |
05203524 | 671 | if (l1_index >= s->l1_size) |
095a9c58 | 672 | goto out; |
05203524 AL |
673 | |
674 | l2_offset = s->l1_table[l1_index]; | |
675 | ||
676 | /* seek the l2 table of the given l2 offset */ | |
677 | ||
678 | if (!l2_offset) | |
095a9c58 | 679 | goto out; |
05203524 AL |
680 | |
681 | /* load the l2 table in memory */ | |
682 | ||
683 | l2_offset &= ~QCOW_OFLAG_COPIED; | |
684 | l2_table = l2_load(bs, l2_offset); | |
685 | if (l2_table == NULL) | |
768706a5 | 686 | return 0; |
05203524 AL |
687 | |
688 | /* find the cluster offset for the given disk offset */ | |
689 | ||
690 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
691 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
6db6c638 | 692 | nb_clusters = size_to_clusters(s, nb_needed << 9); |
095a9c58 AL |
693 | |
694 | if (!cluster_offset) { | |
6db6c638 AL |
695 | /* how many empty clusters ? */ |
696 | c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]); | |
095a9c58 | 697 | } else { |
6db6c638 AL |
698 | /* how many allocated clusters ? */ |
699 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
ff4b91c2 | 700 | &l2_table[l2_index], 0, QCOW_OFLAG_COPIED); |
6db6c638 | 701 | } |
095a9c58 | 702 | |
6db6c638 | 703 | nb_available = (c * s->cluster_sectors); |
095a9c58 AL |
704 | out: |
705 | if (nb_available > nb_needed) | |
706 | nb_available = nb_needed; | |
707 | ||
708 | *num = nb_available - index_in_cluster; | |
709 | ||
6db6c638 | 710 | return cluster_offset & ~QCOW_OFLAG_COPIED; |
05203524 AL |
711 | } |
712 | ||
713 | /* | |
52d893ec | 714 | * free_any_clusters |
05203524 | 715 | * |
52d893ec | 716 | * free clusters according to its type: compressed or not |
05203524 | 717 | * |
52d893ec AL |
718 | */ |
719 | ||
720 | static void free_any_clusters(BlockDriverState *bs, | |
095a9c58 | 721 | uint64_t cluster_offset, int nb_clusters) |
52d893ec AL |
722 | { |
723 | BDRVQcowState *s = bs->opaque; | |
724 | ||
52d893ec AL |
725 | /* free the cluster */ |
726 | ||
727 | if (cluster_offset & QCOW_OFLAG_COMPRESSED) { | |
728 | int nb_csectors; | |
729 | nb_csectors = ((cluster_offset >> s->csize_shift) & | |
730 | s->csize_mask) + 1; | |
731 | free_clusters(bs, (cluster_offset & s->cluster_offset_mask) & ~511, | |
732 | nb_csectors * 512); | |
733 | return; | |
734 | } | |
735 | ||
095a9c58 AL |
736 | free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits); |
737 | ||
738 | return; | |
52d893ec AL |
739 | } |
740 | ||
741 | /* | |
742 | * get_cluster_table | |
05203524 | 743 | * |
52d893ec AL |
744 | * for a given disk offset, load (and allocate if needed) |
745 | * the l2 table. | |
746 | * | |
747 | * the l2 table offset in the qcow2 file and the cluster index | |
748 | * in the l2 table are given to the caller. | |
05203524 AL |
749 | * |
750 | */ | |
751 | ||
52d893ec AL |
752 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
753 | uint64_t **new_l2_table, | |
754 | uint64_t *new_l2_offset, | |
755 | int *new_l2_index) | |
585f8587 FB |
756 | { |
757 | BDRVQcowState *s = bs->opaque; | |
108534b9 | 758 | int l1_index, l2_index, ret; |
52d893ec | 759 | uint64_t l2_offset, *l2_table; |
108534b9 AL |
760 | |
761 | /* seek the the l2 offset in the l1 table */ | |
3b46e624 | 762 | |
585f8587 FB |
763 | l1_index = offset >> (s->l2_bits + s->cluster_bits); |
764 | if (l1_index >= s->l1_size) { | |
108534b9 AL |
765 | ret = grow_l1_table(bs, l1_index + 1); |
766 | if (ret < 0) | |
585f8587 FB |
767 | return 0; |
768 | } | |
769 | l2_offset = s->l1_table[l1_index]; | |
108534b9 AL |
770 | |
771 | /* seek the l2 table of the given l2 offset */ | |
772 | ||
05203524 AL |
773 | if (l2_offset & QCOW_OFLAG_COPIED) { |
774 | /* load the l2 table in memory */ | |
775 | l2_offset &= ~QCOW_OFLAG_COPIED; | |
776 | l2_table = l2_load(bs, l2_offset); | |
777 | if (l2_table == NULL) | |
585f8587 | 778 | return 0; |
05203524 AL |
779 | } else { |
780 | if (l2_offset) | |
781 | free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t)); | |
108534b9 AL |
782 | l2_table = l2_allocate(bs, l1_index); |
783 | if (l2_table == NULL) | |
585f8587 | 784 | return 0; |
108534b9 | 785 | l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED; |
585f8587 | 786 | } |
108534b9 AL |
787 | |
788 | /* find the cluster offset for the given disk offset */ | |
789 | ||
585f8587 | 790 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); |
05203524 | 791 | |
52d893ec AL |
792 | *new_l2_table = l2_table; |
793 | *new_l2_offset = l2_offset; | |
794 | *new_l2_index = l2_index; | |
795 | ||
796 | return 1; | |
797 | } | |
798 | ||
799 | /* | |
800 | * alloc_compressed_cluster_offset | |
801 | * | |
802 | * For a given offset of the disk image, return cluster offset in | |
803 | * qcow2 file. | |
804 | * | |
805 | * If the offset is not found, allocate a new compressed cluster. | |
806 | * | |
807 | * Return the cluster offset if successful, | |
808 | * Return 0, otherwise. | |
809 | * | |
810 | */ | |
811 | ||
812 | static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs, | |
813 | uint64_t offset, | |
814 | int compressed_size) | |
815 | { | |
816 | BDRVQcowState *s = bs->opaque; | |
817 | int l2_index, ret; | |
818 | uint64_t l2_offset, *l2_table, cluster_offset; | |
819 | int nb_csectors; | |
820 | ||
821 | ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index); | |
822 | if (ret == 0) | |
823 | return 0; | |
824 | ||
825 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
05203524 AL |
826 | if (cluster_offset & QCOW_OFLAG_COPIED) |
827 | return cluster_offset & ~QCOW_OFLAG_COPIED; | |
828 | ||
095a9c58 AL |
829 | if (cluster_offset) |
830 | free_any_clusters(bs, cluster_offset, 1); | |
108534b9 | 831 | |
52d893ec AL |
832 | cluster_offset = alloc_bytes(bs, compressed_size); |
833 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - | |
834 | (cluster_offset >> 9); | |
05203524 | 835 | |
52d893ec AL |
836 | cluster_offset |= QCOW_OFLAG_COMPRESSED | |
837 | ((uint64_t)nb_csectors << s->csize_shift); | |
05203524 | 838 | |
52d893ec | 839 | /* update L2 table */ |
05203524 | 840 | |
52d893ec | 841 | /* compressed clusters never have the copied flag */ |
05203524 | 842 | |
52d893ec AL |
843 | l2_table[l2_index] = cpu_to_be64(cluster_offset); |
844 | if (bdrv_pwrite(s->hd, | |
845 | l2_offset + l2_index * sizeof(uint64_t), | |
846 | l2_table + l2_index, | |
847 | sizeof(uint64_t)) != sizeof(uint64_t)) | |
848 | return 0; | |
05203524 | 849 | |
52d893ec AL |
850 | return cluster_offset; |
851 | } | |
05203524 | 852 | |
e976c6a1 AL |
853 | typedef struct QCowL2Meta |
854 | { | |
855 | uint64_t offset; | |
856 | int n_start; | |
857 | int nb_available; | |
858 | int nb_clusters; | |
859 | } QCowL2Meta; | |
860 | ||
861 | static int alloc_cluster_link_l2(BlockDriverState *bs, uint64_t cluster_offset, | |
862 | QCowL2Meta *m) | |
863 | { | |
864 | BDRVQcowState *s = bs->opaque; | |
865 | int i, j = 0, l2_index, ret; | |
866 | uint64_t *old_cluster, start_sect, l2_offset, *l2_table; | |
867 | ||
868 | if (m->nb_clusters == 0) | |
869 | return 0; | |
870 | ||
3ec88e80 | 871 | old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t)); |
e976c6a1 AL |
872 | |
873 | /* copy content of unmodified sectors */ | |
874 | start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9; | |
875 | if (m->n_start) { | |
876 | ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start); | |
877 | if (ret < 0) | |
878 | goto err; | |
879 | } | |
880 | ||
881 | if (m->nb_available & (s->cluster_sectors - 1)) { | |
882 | uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1); | |
883 | ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9), | |
884 | m->nb_available - end, s->cluster_sectors); | |
885 | if (ret < 0) | |
886 | goto err; | |
887 | } | |
888 | ||
889 | ret = -EIO; | |
890 | /* update L2 table */ | |
891 | if (!get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index)) | |
892 | goto err; | |
893 | ||
894 | for (i = 0; i < m->nb_clusters; i++) { | |
d4d698f0 GN |
895 | /* if two concurrent writes happen to the same unallocated cluster |
896 | * each write allocates separate cluster and writes data concurrently. | |
897 | * The first one to complete updates l2 table with pointer to its | |
898 | * cluster the second one has to do RMW (which is done above by | |
899 | * copy_sectors()), update l2 table with its cluster pointer and free | |
900 | * old cluster. This is what this loop does */ | |
e976c6a1 AL |
901 | if(l2_table[l2_index + i] != 0) |
902 | old_cluster[j++] = l2_table[l2_index + i]; | |
903 | ||
904 | l2_table[l2_index + i] = cpu_to_be64((cluster_offset + | |
905 | (i << s->cluster_bits)) | QCOW_OFLAG_COPIED); | |
906 | } | |
907 | ||
908 | if (bdrv_pwrite(s->hd, l2_offset + l2_index * sizeof(uint64_t), | |
909 | l2_table + l2_index, m->nb_clusters * sizeof(uint64_t)) != | |
910 | m->nb_clusters * sizeof(uint64_t)) | |
911 | goto err; | |
912 | ||
913 | for (i = 0; i < j; i++) | |
d4d698f0 GN |
914 | free_any_clusters(bs, be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, |
915 | 1); | |
e976c6a1 AL |
916 | |
917 | ret = 0; | |
918 | err: | |
919 | qemu_free(old_cluster); | |
920 | return ret; | |
921 | } | |
922 | ||
52d893ec AL |
923 | /* |
924 | * alloc_cluster_offset | |
925 | * | |
926 | * For a given offset of the disk image, return cluster offset in | |
927 | * qcow2 file. | |
928 | * | |
929 | * If the offset is not found, allocate a new cluster. | |
930 | * | |
931 | * Return the cluster offset if successful, | |
932 | * Return 0, otherwise. | |
933 | * | |
934 | */ | |
935 | ||
936 | static uint64_t alloc_cluster_offset(BlockDriverState *bs, | |
937 | uint64_t offset, | |
095a9c58 | 938 | int n_start, int n_end, |
e976c6a1 | 939 | int *num, QCowL2Meta *m) |
52d893ec AL |
940 | { |
941 | BDRVQcowState *s = bs->opaque; | |
942 | int l2_index, ret; | |
943 | uint64_t l2_offset, *l2_table, cluster_offset; | |
e976c6a1 | 944 | int nb_clusters, i = 0; |
52d893ec AL |
945 | |
946 | ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index); | |
947 | if (ret == 0) | |
948 | return 0; | |
949 | ||
6db6c638 AL |
950 | nb_clusters = size_to_clusters(s, n_end << 9); |
951 | ||
e976c6a1 | 952 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
095a9c58 | 953 | |
52d893ec | 954 | cluster_offset = be64_to_cpu(l2_table[l2_index]); |
52d893ec | 955 | |
095a9c58 AL |
956 | /* We keep all QCOW_OFLAG_COPIED clusters */ |
957 | ||
958 | if (cluster_offset & QCOW_OFLAG_COPIED) { | |
6db6c638 | 959 | nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size, |
ff4b91c2 | 960 | &l2_table[l2_index], 0, 0); |
095a9c58 | 961 | |
095a9c58 | 962 | cluster_offset &= ~QCOW_OFLAG_COPIED; |
e976c6a1 | 963 | m->nb_clusters = 0; |
095a9c58 AL |
964 | |
965 | goto out; | |
966 | } | |
967 | ||
968 | /* for the moment, multiple compressed clusters are not managed */ | |
969 | ||
970 | if (cluster_offset & QCOW_OFLAG_COMPRESSED) | |
971 | nb_clusters = 1; | |
972 | ||
bc352085 | 973 | /* how many available clusters ? */ |
095a9c58 | 974 | |
bc352085 | 975 | while (i < nb_clusters) { |
ab5ccbd6 | 976 | i += count_contiguous_clusters(nb_clusters - i, s->cluster_size, |
ff4b91c2 | 977 | &l2_table[l2_index], i, 0); |
ab5ccbd6 AL |
978 | |
979 | if(be64_to_cpu(l2_table[l2_index + i])) | |
980 | break; | |
981 | ||
6db6c638 AL |
982 | i += count_contiguous_free_clusters(nb_clusters - i, |
983 | &l2_table[l2_index + i]); | |
095a9c58 | 984 | |
6db6c638 | 985 | cluster_offset = be64_to_cpu(l2_table[l2_index + i]); |
095a9c58 | 986 | |
6db6c638 | 987 | if ((cluster_offset & QCOW_OFLAG_COPIED) || |
bc352085 | 988 | (cluster_offset & QCOW_OFLAG_COMPRESSED)) |
6db6c638 | 989 | break; |
095a9c58 | 990 | } |
bc352085 | 991 | nb_clusters = i; |
05203524 AL |
992 | |
993 | /* allocate a new cluster */ | |
994 | ||
095a9c58 | 995 | cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size); |
05203524 | 996 | |
e976c6a1 AL |
997 | /* save info needed for meta data update */ |
998 | m->offset = offset; | |
999 | m->n_start = n_start; | |
1000 | m->nb_clusters = nb_clusters; | |
05203524 | 1001 | |
095a9c58 | 1002 | out: |
e976c6a1 AL |
1003 | m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end); |
1004 | ||
1005 | *num = m->nb_available - n_start; | |
095a9c58 | 1006 | |
585f8587 FB |
1007 | return cluster_offset; |
1008 | } | |
1009 | ||
5fafdf24 | 1010 | static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num, |
585f8587 FB |
1011 | int nb_sectors, int *pnum) |
1012 | { | |
585f8587 FB |
1013 | uint64_t cluster_offset; |
1014 | ||
095a9c58 AL |
1015 | *pnum = nb_sectors; |
1016 | cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum); | |
1017 | ||
585f8587 FB |
1018 | return (cluster_offset != 0); |
1019 | } | |
1020 | ||
1021 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
1022 | const uint8_t *buf, int buf_size) | |
1023 | { | |
1024 | z_stream strm1, *strm = &strm1; | |
1025 | int ret, out_len; | |
1026 | ||
1027 | memset(strm, 0, sizeof(*strm)); | |
1028 | ||
1029 | strm->next_in = (uint8_t *)buf; | |
1030 | strm->avail_in = buf_size; | |
1031 | strm->next_out = out_buf; | |
1032 | strm->avail_out = out_buf_size; | |
1033 | ||
1034 | ret = inflateInit2(strm, -12); | |
1035 | if (ret != Z_OK) | |
1036 | return -1; | |
1037 | ret = inflate(strm, Z_FINISH); | |
1038 | out_len = strm->next_out - out_buf; | |
1039 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
1040 | out_len != out_buf_size) { | |
1041 | inflateEnd(strm); | |
1042 | return -1; | |
1043 | } | |
1044 | inflateEnd(strm); | |
1045 | return 0; | |
1046 | } | |
3b46e624 | 1047 | |
585f8587 FB |
1048 | static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset) |
1049 | { | |
1050 | int ret, csize, nb_csectors, sector_offset; | |
1051 | uint64_t coffset; | |
1052 | ||
1053 | coffset = cluster_offset & s->cluster_offset_mask; | |
1054 | if (s->cluster_cache_offset != coffset) { | |
1055 | nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1; | |
1056 | sector_offset = coffset & 511; | |
1057 | csize = nb_csectors * 512 - sector_offset; | |
1058 | ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors); | |
1059 | if (ret < 0) { | |
1060 | return -1; | |
1061 | } | |
1062 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
1063 | s->cluster_data + sector_offset, csize) < 0) { | |
1064 | return -1; | |
1065 | } | |
1066 | s->cluster_cache_offset = coffset; | |
1067 | } | |
1068 | return 0; | |
1069 | } | |
1070 | ||
a9465922 | 1071 | /* handle reading after the end of the backing file */ |
5fafdf24 | 1072 | static int backing_read1(BlockDriverState *bs, |
a9465922 FB |
1073 | int64_t sector_num, uint8_t *buf, int nb_sectors) |
1074 | { | |
1075 | int n1; | |
1076 | if ((sector_num + nb_sectors) <= bs->total_sectors) | |
1077 | return nb_sectors; | |
1078 | if (sector_num >= bs->total_sectors) | |
1079 | n1 = 0; | |
1080 | else | |
1081 | n1 = bs->total_sectors - sector_num; | |
1082 | memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1)); | |
1083 | return n1; | |
1084 | } | |
1085 | ||
5fafdf24 | 1086 | static int qcow_read(BlockDriverState *bs, int64_t sector_num, |
585f8587 FB |
1087 | uint8_t *buf, int nb_sectors) |
1088 | { | |
1089 | BDRVQcowState *s = bs->opaque; | |
a9465922 | 1090 | int ret, index_in_cluster, n, n1; |
585f8587 | 1091 | uint64_t cluster_offset; |
3b46e624 | 1092 | |
585f8587 | 1093 | while (nb_sectors > 0) { |
095a9c58 AL |
1094 | n = nb_sectors; |
1095 | cluster_offset = get_cluster_offset(bs, sector_num << 9, &n); | |
585f8587 | 1096 | index_in_cluster = sector_num & (s->cluster_sectors - 1); |
585f8587 FB |
1097 | if (!cluster_offset) { |
1098 | if (bs->backing_hd) { | |
1099 | /* read from the base image */ | |
a9465922 FB |
1100 | n1 = backing_read1(bs->backing_hd, sector_num, buf, n); |
1101 | if (n1 > 0) { | |
1102 | ret = bdrv_read(bs->backing_hd, sector_num, buf, n1); | |
1103 | if (ret < 0) | |
1104 | return -1; | |
1105 | } | |
585f8587 FB |
1106 | } else { |
1107 | memset(buf, 0, 512 * n); | |
1108 | } | |
1109 | } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) { | |
1110 | if (decompress_cluster(s, cluster_offset) < 0) | |
1111 | return -1; | |
1112 | memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n); | |
1113 | } else { | |
1114 | ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512); | |
5fafdf24 | 1115 | if (ret != n * 512) |
585f8587 FB |
1116 | return -1; |
1117 | if (s->crypt_method) { | |
5fafdf24 | 1118 | encrypt_sectors(s, sector_num, buf, buf, n, 0, |
585f8587 FB |
1119 | &s->aes_decrypt_key); |
1120 | } | |
1121 | } | |
1122 | nb_sectors -= n; | |
1123 | sector_num += n; | |
1124 | buf += n * 512; | |
1125 | } | |
1126 | return 0; | |
1127 | } | |
1128 | ||
ce1a14dc PB |
1129 | typedef struct QCowAIOCB { |
1130 | BlockDriverAIOCB common; | |
585f8587 | 1131 | int64_t sector_num; |
f141eafe | 1132 | QEMUIOVector *qiov; |
585f8587 | 1133 | uint8_t *buf; |
f141eafe | 1134 | void *orig_buf; |
585f8587 FB |
1135 | int nb_sectors; |
1136 | int n; | |
1137 | uint64_t cluster_offset; | |
5fafdf24 | 1138 | uint8_t *cluster_data; |
585f8587 | 1139 | BlockDriverAIOCB *hd_aiocb; |
c87c0672 AL |
1140 | struct iovec hd_iov; |
1141 | QEMUIOVector hd_qiov; | |
1490791f | 1142 | QEMUBH *bh; |
e976c6a1 | 1143 | QCowL2Meta l2meta; |
585f8587 FB |
1144 | } QCowAIOCB; |
1145 | ||
c16b5a2c CH |
1146 | static void qcow_aio_cancel(BlockDriverAIOCB *blockacb) |
1147 | { | |
1148 | QCowAIOCB *acb = (QCowAIOCB *)blockacb; | |
1149 | if (acb->hd_aiocb) | |
1150 | bdrv_aio_cancel(acb->hd_aiocb); | |
1151 | qemu_aio_release(acb); | |
1152 | } | |
1153 | ||
1154 | static AIOPool qcow_aio_pool = { | |
1155 | .aiocb_size = sizeof(QCowAIOCB), | |
1156 | .cancel = qcow_aio_cancel, | |
1157 | }; | |
1158 | ||
1490791f AL |
1159 | static void qcow_aio_read_cb(void *opaque, int ret); |
1160 | static void qcow_aio_read_bh(void *opaque) | |
1161 | { | |
1162 | QCowAIOCB *acb = opaque; | |
1163 | qemu_bh_delete(acb->bh); | |
1164 | acb->bh = NULL; | |
1165 | qcow_aio_read_cb(opaque, 0); | |
1166 | } | |
1167 | ||
a32ef786 AL |
1168 | static int qcow_schedule_bh(QEMUBHFunc *cb, QCowAIOCB *acb) |
1169 | { | |
1170 | if (acb->bh) | |
1171 | return -EIO; | |
1172 | ||
1173 | acb->bh = qemu_bh_new(cb, acb); | |
1174 | if (!acb->bh) | |
1175 | return -EIO; | |
1176 | ||
1177 | qemu_bh_schedule(acb->bh); | |
1178 | ||
1179 | return 0; | |
1180 | } | |
1181 | ||
585f8587 FB |
1182 | static void qcow_aio_read_cb(void *opaque, int ret) |
1183 | { | |
ce1a14dc PB |
1184 | QCowAIOCB *acb = opaque; |
1185 | BlockDriverState *bs = acb->common.bs; | |
585f8587 | 1186 | BDRVQcowState *s = bs->opaque; |
a9465922 | 1187 | int index_in_cluster, n1; |
585f8587 | 1188 | |
ce1a14dc | 1189 | acb->hd_aiocb = NULL; |
f141eafe AL |
1190 | if (ret < 0) |
1191 | goto done; | |
585f8587 | 1192 | |
585f8587 | 1193 | /* post process the read buffer */ |
ce1a14dc | 1194 | if (!acb->cluster_offset) { |
585f8587 | 1195 | /* nothing to do */ |
ce1a14dc | 1196 | } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) { |
585f8587 FB |
1197 | /* nothing to do */ |
1198 | } else { | |
1199 | if (s->crypt_method) { | |
5fafdf24 TS |
1200 | encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf, |
1201 | acb->n, 0, | |
585f8587 FB |
1202 | &s->aes_decrypt_key); |
1203 | } | |
1204 | } | |
1205 | ||
ce1a14dc PB |
1206 | acb->nb_sectors -= acb->n; |
1207 | acb->sector_num += acb->n; | |
1208 | acb->buf += acb->n * 512; | |
585f8587 | 1209 | |
ce1a14dc | 1210 | if (acb->nb_sectors == 0) { |
585f8587 | 1211 | /* request completed */ |
f141eafe AL |
1212 | ret = 0; |
1213 | goto done; | |
585f8587 | 1214 | } |
3b46e624 | 1215 | |
585f8587 | 1216 | /* prepare next AIO request */ |
095a9c58 AL |
1217 | acb->n = acb->nb_sectors; |
1218 | acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n); | |
ce1a14dc | 1219 | index_in_cluster = acb->sector_num & (s->cluster_sectors - 1); |
ce1a14dc PB |
1220 | |
1221 | if (!acb->cluster_offset) { | |
585f8587 FB |
1222 | if (bs->backing_hd) { |
1223 | /* read from the base image */ | |
5fafdf24 | 1224 | n1 = backing_read1(bs->backing_hd, acb->sector_num, |
ce1a14dc | 1225 | acb->buf, acb->n); |
a9465922 | 1226 | if (n1 > 0) { |
3f4cb3d3 | 1227 | acb->hd_iov.iov_base = (void *)acb->buf; |
c87c0672 AL |
1228 | acb->hd_iov.iov_len = acb->n * 512; |
1229 | qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1); | |
1230 | acb->hd_aiocb = bdrv_aio_readv(bs->backing_hd, acb->sector_num, | |
1231 | &acb->hd_qiov, acb->n, | |
1232 | qcow_aio_read_cb, acb); | |
ce1a14dc | 1233 | if (acb->hd_aiocb == NULL) |
f141eafe | 1234 | goto done; |
a9465922 | 1235 | } else { |
a32ef786 AL |
1236 | ret = qcow_schedule_bh(qcow_aio_read_bh, acb); |
1237 | if (ret < 0) | |
f141eafe | 1238 | goto done; |
a9465922 | 1239 | } |
585f8587 FB |
1240 | } else { |
1241 | /* Note: in this case, no need to wait */ | |
ce1a14dc | 1242 | memset(acb->buf, 0, 512 * acb->n); |
a32ef786 AL |
1243 | ret = qcow_schedule_bh(qcow_aio_read_bh, acb); |
1244 | if (ret < 0) | |
f141eafe | 1245 | goto done; |
585f8587 | 1246 | } |
ce1a14dc | 1247 | } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) { |
585f8587 | 1248 | /* add AIO support for compressed blocks ? */ |
ce1a14dc | 1249 | if (decompress_cluster(s, acb->cluster_offset) < 0) |
f141eafe | 1250 | goto done; |
5fafdf24 | 1251 | memcpy(acb->buf, |
ce1a14dc | 1252 | s->cluster_cache + index_in_cluster * 512, 512 * acb->n); |
a32ef786 AL |
1253 | ret = qcow_schedule_bh(qcow_aio_read_bh, acb); |
1254 | if (ret < 0) | |
f141eafe | 1255 | goto done; |
585f8587 | 1256 | } else { |
ce1a14dc | 1257 | if ((acb->cluster_offset & 511) != 0) { |
585f8587 | 1258 | ret = -EIO; |
f141eafe | 1259 | goto done; |
585f8587 | 1260 | } |
c87c0672 | 1261 | |
3f4cb3d3 | 1262 | acb->hd_iov.iov_base = (void *)acb->buf; |
c87c0672 AL |
1263 | acb->hd_iov.iov_len = acb->n * 512; |
1264 | qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1); | |
1265 | acb->hd_aiocb = bdrv_aio_readv(s->hd, | |
5fafdf24 | 1266 | (acb->cluster_offset >> 9) + index_in_cluster, |
c87c0672 | 1267 | &acb->hd_qiov, acb->n, qcow_aio_read_cb, acb); |
ce1a14dc | 1268 | if (acb->hd_aiocb == NULL) |
f141eafe AL |
1269 | goto done; |
1270 | } | |
1271 | ||
1272 | return; | |
1273 | done: | |
1274 | if (acb->qiov->niov > 1) { | |
1275 | qemu_iovec_from_buffer(acb->qiov, acb->orig_buf, acb->qiov->size); | |
1276 | qemu_vfree(acb->orig_buf); | |
585f8587 | 1277 | } |
f141eafe AL |
1278 | acb->common.cb(acb->common.opaque, ret); |
1279 | qemu_aio_release(acb); | |
585f8587 FB |
1280 | } |
1281 | ||
ce1a14dc | 1282 | static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs, |
f141eafe AL |
1283 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, |
1284 | BlockDriverCompletionFunc *cb, void *opaque, int is_write) | |
585f8587 | 1285 | { |
ce1a14dc PB |
1286 | QCowAIOCB *acb; |
1287 | ||
c16b5a2c | 1288 | acb = qemu_aio_get(&qcow_aio_pool, bs, cb, opaque); |
ce1a14dc PB |
1289 | if (!acb) |
1290 | return NULL; | |
1291 | acb->hd_aiocb = NULL; | |
1292 | acb->sector_num = sector_num; | |
f141eafe AL |
1293 | acb->qiov = qiov; |
1294 | if (qiov->niov > 1) { | |
e268ca52 | 1295 | acb->buf = acb->orig_buf = qemu_blockalign(bs, qiov->size); |
f141eafe AL |
1296 | if (is_write) |
1297 | qemu_iovec_to_buffer(qiov, acb->buf); | |
3f4cb3d3 BS |
1298 | } else { |
1299 | acb->buf = (uint8_t *)qiov->iov->iov_base; | |
1300 | } | |
ce1a14dc PB |
1301 | acb->nb_sectors = nb_sectors; |
1302 | acb->n = 0; | |
1303 | acb->cluster_offset = 0; | |
e976c6a1 | 1304 | acb->l2meta.nb_clusters = 0; |
ce1a14dc PB |
1305 | return acb; |
1306 | } | |
1307 | ||
f141eafe AL |
1308 | static BlockDriverAIOCB *qcow_aio_readv(BlockDriverState *bs, |
1309 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
ce1a14dc PB |
1310 | BlockDriverCompletionFunc *cb, void *opaque) |
1311 | { | |
1312 | QCowAIOCB *acb; | |
1313 | ||
f141eafe | 1314 | acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0); |
ce1a14dc PB |
1315 | if (!acb) |
1316 | return NULL; | |
585f8587 FB |
1317 | |
1318 | qcow_aio_read_cb(acb, 0); | |
ce1a14dc | 1319 | return &acb->common; |
585f8587 FB |
1320 | } |
1321 | ||
1322 | static void qcow_aio_write_cb(void *opaque, int ret) | |
1323 | { | |
ce1a14dc PB |
1324 | QCowAIOCB *acb = opaque; |
1325 | BlockDriverState *bs = acb->common.bs; | |
585f8587 | 1326 | BDRVQcowState *s = bs->opaque; |
585f8587 | 1327 | int index_in_cluster; |
585f8587 | 1328 | const uint8_t *src_buf; |
095a9c58 | 1329 | int n_end; |
ce1a14dc PB |
1330 | |
1331 | acb->hd_aiocb = NULL; | |
1332 | ||
f141eafe AL |
1333 | if (ret < 0) |
1334 | goto done; | |
585f8587 | 1335 | |
e976c6a1 AL |
1336 | if (alloc_cluster_link_l2(bs, acb->cluster_offset, &acb->l2meta) < 0) { |
1337 | free_any_clusters(bs, acb->cluster_offset, acb->l2meta.nb_clusters); | |
f141eafe | 1338 | goto done; |
e976c6a1 AL |
1339 | } |
1340 | ||
ce1a14dc PB |
1341 | acb->nb_sectors -= acb->n; |
1342 | acb->sector_num += acb->n; | |
1343 | acb->buf += acb->n * 512; | |
585f8587 | 1344 | |
ce1a14dc | 1345 | if (acb->nb_sectors == 0) { |
585f8587 | 1346 | /* request completed */ |
f141eafe AL |
1347 | ret = 0; |
1348 | goto done; | |
585f8587 | 1349 | } |
3b46e624 | 1350 | |
ce1a14dc | 1351 | index_in_cluster = acb->sector_num & (s->cluster_sectors - 1); |
095a9c58 AL |
1352 | n_end = index_in_cluster + acb->nb_sectors; |
1353 | if (s->crypt_method && | |
1354 | n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) | |
1355 | n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors; | |
1356 | ||
e976c6a1 | 1357 | acb->cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9, |
05203524 | 1358 | index_in_cluster, |
e976c6a1 AL |
1359 | n_end, &acb->n, &acb->l2meta); |
1360 | if (!acb->cluster_offset || (acb->cluster_offset & 511) != 0) { | |
585f8587 | 1361 | ret = -EIO; |
f141eafe | 1362 | goto done; |
585f8587 FB |
1363 | } |
1364 | if (s->crypt_method) { | |
ce1a14dc | 1365 | if (!acb->cluster_data) { |
095a9c58 AL |
1366 | acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS * |
1367 | s->cluster_size); | |
585f8587 | 1368 | } |
5fafdf24 | 1369 | encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf, |
ce1a14dc PB |
1370 | acb->n, 1, &s->aes_encrypt_key); |
1371 | src_buf = acb->cluster_data; | |
585f8587 | 1372 | } else { |
ce1a14dc | 1373 | src_buf = acb->buf; |
585f8587 | 1374 | } |
c87c0672 AL |
1375 | acb->hd_iov.iov_base = (void *)src_buf; |
1376 | acb->hd_iov.iov_len = acb->n * 512; | |
1377 | qemu_iovec_init_external(&acb->hd_qiov, &acb->hd_iov, 1); | |
1378 | acb->hd_aiocb = bdrv_aio_writev(s->hd, | |
1379 | (acb->cluster_offset >> 9) + index_in_cluster, | |
1380 | &acb->hd_qiov, acb->n, | |
1381 | qcow_aio_write_cb, acb); | |
ce1a14dc | 1382 | if (acb->hd_aiocb == NULL) |
f141eafe AL |
1383 | goto done; |
1384 | ||
1385 | return; | |
1386 | ||
1387 | done: | |
1388 | if (acb->qiov->niov > 1) | |
1389 | qemu_vfree(acb->orig_buf); | |
1390 | acb->common.cb(acb->common.opaque, ret); | |
1391 | qemu_aio_release(acb); | |
585f8587 FB |
1392 | } |
1393 | ||
f141eafe AL |
1394 | static BlockDriverAIOCB *qcow_aio_writev(BlockDriverState *bs, |
1395 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
ce1a14dc | 1396 | BlockDriverCompletionFunc *cb, void *opaque) |
585f8587 | 1397 | { |
585f8587 | 1398 | BDRVQcowState *s = bs->opaque; |
ce1a14dc | 1399 | QCowAIOCB *acb; |
3b46e624 | 1400 | |
585f8587 FB |
1401 | s->cluster_cache_offset = -1; /* disable compressed cache */ |
1402 | ||
f141eafe | 1403 | acb = qcow_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 1); |
ce1a14dc PB |
1404 | if (!acb) |
1405 | return NULL; | |
3b46e624 | 1406 | |
585f8587 | 1407 | qcow_aio_write_cb(acb, 0); |
ce1a14dc | 1408 | return &acb->common; |
585f8587 FB |
1409 | } |
1410 | ||
585f8587 FB |
1411 | static void qcow_close(BlockDriverState *bs) |
1412 | { | |
1413 | BDRVQcowState *s = bs->opaque; | |
1414 | qemu_free(s->l1_table); | |
1415 | qemu_free(s->l2_cache); | |
1416 | qemu_free(s->cluster_cache); | |
1417 | qemu_free(s->cluster_data); | |
1418 | refcount_close(bs); | |
1419 | bdrv_delete(s->hd); | |
1420 | } | |
1421 | ||
73c632ed KW |
1422 | static int get_bits_from_size(size_t size) |
1423 | { | |
1424 | int res = 0; | |
1425 | ||
1426 | if (size == 0) { | |
1427 | return -1; | |
1428 | } | |
1429 | ||
1430 | while (size != 1) { | |
1431 | /* Not a power of two */ | |
1432 | if (size & 1) { | |
1433 | return -1; | |
1434 | } | |
1435 | ||
1436 | size >>= 1; | |
1437 | res++; | |
1438 | } | |
1439 | ||
1440 | return res; | |
1441 | } | |
1442 | ||
f965509c AL |
1443 | static int qcow_create2(const char *filename, int64_t total_size, |
1444 | const char *backing_file, const char *backing_format, | |
73c632ed | 1445 | int flags, size_t cluster_size) |
585f8587 | 1446 | { |
f965509c | 1447 | |
585f8587 | 1448 | int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits; |
2d2431f0 | 1449 | int ref_clusters, backing_format_len = 0; |
585f8587 FB |
1450 | QCowHeader header; |
1451 | uint64_t tmp, offset; | |
1452 | QCowCreateState s1, *s = &s1; | |
f965509c AL |
1453 | QCowExtension ext_bf = {0, 0}; |
1454 | ||
3b46e624 | 1455 | |
585f8587 FB |
1456 | memset(s, 0, sizeof(*s)); |
1457 | ||
1458 | fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644); | |
1459 | if (fd < 0) | |
1460 | return -1; | |
1461 | memset(&header, 0, sizeof(header)); | |
1462 | header.magic = cpu_to_be32(QCOW_MAGIC); | |
1463 | header.version = cpu_to_be32(QCOW_VERSION); | |
1464 | header.size = cpu_to_be64(total_size * 512); | |
1465 | header_size = sizeof(header); | |
1466 | backing_filename_len = 0; | |
1467 | if (backing_file) { | |
f965509c AL |
1468 | if (backing_format) { |
1469 | ext_bf.magic = QCOW_EXT_MAGIC_BACKING_FORMAT; | |
1470 | backing_format_len = strlen(backing_format); | |
1471 | ext_bf.len = (backing_format_len + 7) & ~7; | |
1472 | header_size += ((sizeof(ext_bf) + ext_bf.len + 7) & ~7); | |
1473 | } | |
585f8587 FB |
1474 | header.backing_file_offset = cpu_to_be64(header_size); |
1475 | backing_filename_len = strlen(backing_file); | |
1476 | header.backing_file_size = cpu_to_be32(backing_filename_len); | |
1477 | header_size += backing_filename_len; | |
1478 | } | |
73c632ed KW |
1479 | |
1480 | /* Cluster size */ | |
1481 | s->cluster_bits = get_bits_from_size(cluster_size); | |
1482 | if (s->cluster_bits < MIN_CLUSTER_BITS || | |
1483 | s->cluster_bits > MAX_CLUSTER_BITS) | |
1484 | { | |
1485 | fprintf(stderr, "Cluster size must be a power of two between " | |
1486 | "%d and %dk\n", | |
1487 | 1 << MIN_CLUSTER_BITS, | |
1488 | 1 << (MAX_CLUSTER_BITS - 10)); | |
1489 | return -EINVAL; | |
1490 | } | |
585f8587 | 1491 | s->cluster_size = 1 << s->cluster_bits; |
73c632ed | 1492 | |
585f8587 FB |
1493 | header.cluster_bits = cpu_to_be32(s->cluster_bits); |
1494 | header_size = (header_size + 7) & ~7; | |
ec36ba14 | 1495 | if (flags & BLOCK_FLAG_ENCRYPT) { |
585f8587 FB |
1496 | header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES); |
1497 | } else { | |
1498 | header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); | |
1499 | } | |
1500 | l2_bits = s->cluster_bits - 3; | |
1501 | shift = s->cluster_bits + l2_bits; | |
1502 | l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift); | |
1503 | offset = align_offset(header_size, s->cluster_size); | |
1504 | s->l1_table_offset = offset; | |
1505 | header.l1_table_offset = cpu_to_be64(s->l1_table_offset); | |
1506 | header.l1_size = cpu_to_be32(l1_size); | |
15e6690a | 1507 | offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size); |
585f8587 FB |
1508 | |
1509 | s->refcount_table = qemu_mallocz(s->cluster_size); | |
3b46e624 | 1510 | |
585f8587 FB |
1511 | s->refcount_table_offset = offset; |
1512 | header.refcount_table_offset = cpu_to_be64(offset); | |
1513 | header.refcount_table_clusters = cpu_to_be32(1); | |
1514 | offset += s->cluster_size; | |
585f8587 | 1515 | s->refcount_block_offset = offset; |
2d2431f0 AL |
1516 | |
1517 | /* count how many refcount blocks needed */ | |
1518 | tmp = offset >> s->cluster_bits; | |
1519 | ref_clusters = (tmp >> (s->cluster_bits - REFCOUNT_SHIFT)) + 1; | |
1520 | for (i=0; i < ref_clusters; i++) { | |
1521 | s->refcount_table[i] = cpu_to_be64(offset); | |
1522 | offset += s->cluster_size; | |
1523 | } | |
1524 | ||
1525 | s->refcount_block = qemu_mallocz(ref_clusters * s->cluster_size); | |
585f8587 FB |
1526 | |
1527 | /* update refcounts */ | |
1528 | create_refcount_update(s, 0, header_size); | |
15e6690a | 1529 | create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t)); |
585f8587 | 1530 | create_refcount_update(s, s->refcount_table_offset, s->cluster_size); |
2d2431f0 | 1531 | create_refcount_update(s, s->refcount_block_offset, ref_clusters * s->cluster_size); |
3b46e624 | 1532 | |
585f8587 FB |
1533 | /* write all the data */ |
1534 | write(fd, &header, sizeof(header)); | |
1535 | if (backing_file) { | |
f965509c AL |
1536 | if (backing_format_len) { |
1537 | char zero[16]; | |
1538 | int d = ext_bf.len - backing_format_len; | |
1539 | ||
1540 | memset(zero, 0, sizeof(zero)); | |
1541 | cpu_to_be32s(&ext_bf.magic); | |
1542 | cpu_to_be32s(&ext_bf.len); | |
1543 | write(fd, &ext_bf, sizeof(ext_bf)); | |
1544 | write(fd, backing_format, backing_format_len); | |
1545 | if (d>0) { | |
1546 | write(fd, zero, d); | |
1547 | } | |
1548 | } | |
585f8587 FB |
1549 | write(fd, backing_file, backing_filename_len); |
1550 | } | |
1551 | lseek(fd, s->l1_table_offset, SEEK_SET); | |
1552 | tmp = 0; | |
1553 | for(i = 0;i < l1_size; i++) { | |
1554 | write(fd, &tmp, sizeof(tmp)); | |
1555 | } | |
1556 | lseek(fd, s->refcount_table_offset, SEEK_SET); | |
1557 | write(fd, s->refcount_table, s->cluster_size); | |
3b46e624 | 1558 | |
585f8587 | 1559 | lseek(fd, s->refcount_block_offset, SEEK_SET); |
2d2431f0 | 1560 | write(fd, s->refcount_block, ref_clusters * s->cluster_size); |
585f8587 FB |
1561 | |
1562 | qemu_free(s->refcount_table); | |
1563 | qemu_free(s->refcount_block); | |
1564 | close(fd); | |
1565 | return 0; | |
585f8587 FB |
1566 | } |
1567 | ||
0e7e1989 KW |
1568 | static int qcow_create(const char *filename, QEMUOptionParameter *options) |
1569 | { | |
1570 | const char *backing_file = NULL; | |
1571 | const char *backing_fmt = NULL; | |
1572 | uint64_t sectors = 0; | |
1573 | int flags = 0; | |
9ccb258e | 1574 | size_t cluster_size = 65536; |
0e7e1989 KW |
1575 | |
1576 | /* Read out options */ | |
1577 | while (options && options->name) { | |
1578 | if (!strcmp(options->name, BLOCK_OPT_SIZE)) { | |
1579 | sectors = options->value.n / 512; | |
1580 | } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { | |
1581 | backing_file = options->value.s; | |
1582 | } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) { | |
1583 | backing_fmt = options->value.s; | |
1584 | } else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) { | |
1585 | flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0; | |
73c632ed KW |
1586 | } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) { |
1587 | if (options->value.n) { | |
1588 | cluster_size = options->value.n; | |
1589 | } | |
0e7e1989 KW |
1590 | } |
1591 | options++; | |
1592 | } | |
1593 | ||
73c632ed KW |
1594 | return qcow_create2(filename, sectors, backing_file, backing_fmt, flags, |
1595 | cluster_size); | |
f965509c AL |
1596 | } |
1597 | ||
585f8587 FB |
1598 | static int qcow_make_empty(BlockDriverState *bs) |
1599 | { | |
1600 | #if 0 | |
1601 | /* XXX: not correct */ | |
1602 | BDRVQcowState *s = bs->opaque; | |
1603 | uint32_t l1_length = s->l1_size * sizeof(uint64_t); | |
1604 | int ret; | |
1605 | ||
1606 | memset(s->l1_table, 0, l1_length); | |
1607 | if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0) | |
ac674887 | 1608 | return -1; |
585f8587 FB |
1609 | ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length); |
1610 | if (ret < 0) | |
1611 | return ret; | |
3b46e624 | 1612 | |
585f8587 FB |
1613 | l2_cache_reset(bs); |
1614 | #endif | |
1615 | return 0; | |
1616 | } | |
1617 | ||
1618 | /* XXX: put compressed sectors first, then all the cluster aligned | |
1619 | tables to avoid losing bytes in alignment */ | |
5fafdf24 | 1620 | static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num, |
585f8587 FB |
1621 | const uint8_t *buf, int nb_sectors) |
1622 | { | |
1623 | BDRVQcowState *s = bs->opaque; | |
1624 | z_stream strm; | |
1625 | int ret, out_len; | |
1626 | uint8_t *out_buf; | |
1627 | uint64_t cluster_offset; | |
1628 | ||
1629 | if (nb_sectors == 0) { | |
1630 | /* align end of file to a sector boundary to ease reading with | |
1631 | sector based I/Os */ | |
1632 | cluster_offset = bdrv_getlength(s->hd); | |
1633 | cluster_offset = (cluster_offset + 511) & ~511; | |
1634 | bdrv_truncate(s->hd, cluster_offset); | |
1635 | return 0; | |
1636 | } | |
1637 | ||
1638 | if (nb_sectors != s->cluster_sectors) | |
1639 | return -EINVAL; | |
1640 | ||
1641 | out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128); | |
585f8587 FB |
1642 | |
1643 | /* best compression, small window, no zlib header */ | |
1644 | memset(&strm, 0, sizeof(strm)); | |
1645 | ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, | |
5fafdf24 | 1646 | Z_DEFLATED, -12, |
585f8587 FB |
1647 | 9, Z_DEFAULT_STRATEGY); |
1648 | if (ret != 0) { | |
1649 | qemu_free(out_buf); | |
1650 | return -1; | |
1651 | } | |
1652 | ||
1653 | strm.avail_in = s->cluster_size; | |
1654 | strm.next_in = (uint8_t *)buf; | |
1655 | strm.avail_out = s->cluster_size; | |
1656 | strm.next_out = out_buf; | |
1657 | ||
1658 | ret = deflate(&strm, Z_FINISH); | |
1659 | if (ret != Z_STREAM_END && ret != Z_OK) { | |
1660 | qemu_free(out_buf); | |
1661 | deflateEnd(&strm); | |
1662 | return -1; | |
1663 | } | |
1664 | out_len = strm.next_out - out_buf; | |
1665 | ||
1666 | deflateEnd(&strm); | |
1667 | ||
1668 | if (ret != Z_STREAM_END || out_len >= s->cluster_size) { | |
1669 | /* could not compress: write normal cluster */ | |
ade40677 | 1670 | bdrv_write(bs, sector_num, buf, s->cluster_sectors); |
585f8587 | 1671 | } else { |
52d893ec AL |
1672 | cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9, |
1673 | out_len); | |
1674 | if (!cluster_offset) | |
1675 | return -1; | |
585f8587 FB |
1676 | cluster_offset &= s->cluster_offset_mask; |
1677 | if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) { | |
1678 | qemu_free(out_buf); | |
1679 | return -1; | |
1680 | } | |
1681 | } | |
3b46e624 | 1682 | |
585f8587 FB |
1683 | qemu_free(out_buf); |
1684 | return 0; | |
1685 | } | |
1686 | ||
1687 | static void qcow_flush(BlockDriverState *bs) | |
1688 | { | |
1689 | BDRVQcowState *s = bs->opaque; | |
1690 | bdrv_flush(s->hd); | |
1691 | } | |
1692 | ||
1693 | static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
1694 | { | |
1695 | BDRVQcowState *s = bs->opaque; | |
1696 | bdi->cluster_size = s->cluster_size; | |
5fafdf24 | 1697 | bdi->vm_state_offset = (int64_t)s->l1_vm_state_index << |
585f8587 FB |
1698 | (s->cluster_bits + s->l2_bits); |
1699 | return 0; | |
1700 | } | |
1701 | ||
1702 | /*********************************************************/ | |
1703 | /* snapshot support */ | |
1704 | ||
585f8587 FB |
1705 | |
1706 | static void qcow_free_snapshots(BlockDriverState *bs) | |
1707 | { | |
1708 | BDRVQcowState *s = bs->opaque; | |
1709 | int i; | |
1710 | ||
1711 | for(i = 0; i < s->nb_snapshots; i++) { | |
1712 | qemu_free(s->snapshots[i].name); | |
1713 | qemu_free(s->snapshots[i].id_str); | |
1714 | } | |
1715 | qemu_free(s->snapshots); | |
1716 | s->snapshots = NULL; | |
1717 | s->nb_snapshots = 0; | |
1718 | } | |
1719 | ||
1720 | static int qcow_read_snapshots(BlockDriverState *bs) | |
1721 | { | |
1722 | BDRVQcowState *s = bs->opaque; | |
1723 | QCowSnapshotHeader h; | |
1724 | QCowSnapshot *sn; | |
1725 | int i, id_str_size, name_size; | |
1726 | int64_t offset; | |
1727 | uint32_t extra_data_size; | |
1728 | ||
63c75dcd | 1729 | if (!s->nb_snapshots) { |
1730 | s->snapshots = NULL; | |
1731 | s->snapshots_size = 0; | |
1732 | return 0; | |
1733 | } | |
1734 | ||
585f8587 FB |
1735 | offset = s->snapshots_offset; |
1736 | s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot)); | |
585f8587 FB |
1737 | for(i = 0; i < s->nb_snapshots; i++) { |
1738 | offset = align_offset(offset, 8); | |
1739 | if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h)) | |
1740 | goto fail; | |
1741 | offset += sizeof(h); | |
1742 | sn = s->snapshots + i; | |
1743 | sn->l1_table_offset = be64_to_cpu(h.l1_table_offset); | |
1744 | sn->l1_size = be32_to_cpu(h.l1_size); | |
1745 | sn->vm_state_size = be32_to_cpu(h.vm_state_size); | |
1746 | sn->date_sec = be32_to_cpu(h.date_sec); | |
1747 | sn->date_nsec = be32_to_cpu(h.date_nsec); | |
1748 | sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec); | |
1749 | extra_data_size = be32_to_cpu(h.extra_data_size); | |
1750 | ||
1751 | id_str_size = be16_to_cpu(h.id_str_size); | |
1752 | name_size = be16_to_cpu(h.name_size); | |
1753 | ||
1754 | offset += extra_data_size; | |
1755 | ||
1756 | sn->id_str = qemu_malloc(id_str_size + 1); | |
585f8587 FB |
1757 | if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size) |
1758 | goto fail; | |
1759 | offset += id_str_size; | |
1760 | sn->id_str[id_str_size] = '\0'; | |
1761 | ||
1762 | sn->name = qemu_malloc(name_size + 1); | |
585f8587 FB |
1763 | if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size) |
1764 | goto fail; | |
1765 | offset += name_size; | |
1766 | sn->name[name_size] = '\0'; | |
1767 | } | |
1768 | s->snapshots_size = offset - s->snapshots_offset; | |
1769 | return 0; | |
1770 | fail: | |
1771 | qcow_free_snapshots(bs); | |
1772 | return -1; | |
1773 | } | |
1774 | ||
1775 | /* add at the end of the file a new list of snapshots */ | |
1776 | static int qcow_write_snapshots(BlockDriverState *bs) | |
1777 | { | |
1778 | BDRVQcowState *s = bs->opaque; | |
1779 | QCowSnapshot *sn; | |
1780 | QCowSnapshotHeader h; | |
1781 | int i, name_size, id_str_size, snapshots_size; | |
1782 | uint64_t data64; | |
1783 | uint32_t data32; | |
1784 | int64_t offset, snapshots_offset; | |
1785 | ||
1786 | /* compute the size of the snapshots */ | |
1787 | offset = 0; | |
1788 | for(i = 0; i < s->nb_snapshots; i++) { | |
1789 | sn = s->snapshots + i; | |
1790 | offset = align_offset(offset, 8); | |
1791 | offset += sizeof(h); | |
1792 | offset += strlen(sn->id_str); | |
1793 | offset += strlen(sn->name); | |
1794 | } | |
1795 | snapshots_size = offset; | |
1796 | ||
1797 | snapshots_offset = alloc_clusters(bs, snapshots_size); | |
1798 | offset = snapshots_offset; | |
3b46e624 | 1799 | |
585f8587 FB |
1800 | for(i = 0; i < s->nb_snapshots; i++) { |
1801 | sn = s->snapshots + i; | |
1802 | memset(&h, 0, sizeof(h)); | |
1803 | h.l1_table_offset = cpu_to_be64(sn->l1_table_offset); | |
1804 | h.l1_size = cpu_to_be32(sn->l1_size); | |
1805 | h.vm_state_size = cpu_to_be32(sn->vm_state_size); | |
1806 | h.date_sec = cpu_to_be32(sn->date_sec); | |
1807 | h.date_nsec = cpu_to_be32(sn->date_nsec); | |
1808 | h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec); | |
3b46e624 | 1809 | |
585f8587 FB |
1810 | id_str_size = strlen(sn->id_str); |
1811 | name_size = strlen(sn->name); | |
1812 | h.id_str_size = cpu_to_be16(id_str_size); | |
1813 | h.name_size = cpu_to_be16(name_size); | |
1814 | offset = align_offset(offset, 8); | |
1815 | if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h)) | |
1816 | goto fail; | |
1817 | offset += sizeof(h); | |
1818 | if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size) | |
1819 | goto fail; | |
1820 | offset += id_str_size; | |
1821 | if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size) | |
1822 | goto fail; | |
1823 | offset += name_size; | |
1824 | } | |
1825 | ||
1826 | /* update the various header fields */ | |
1827 | data64 = cpu_to_be64(snapshots_offset); | |
1828 | if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset), | |
1829 | &data64, sizeof(data64)) != sizeof(data64)) | |
1830 | goto fail; | |
1831 | data32 = cpu_to_be32(s->nb_snapshots); | |
1832 | if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots), | |
1833 | &data32, sizeof(data32)) != sizeof(data32)) | |
1834 | goto fail; | |
1835 | ||
1836 | /* free the old snapshot table */ | |
1837 | free_clusters(bs, s->snapshots_offset, s->snapshots_size); | |
1838 | s->snapshots_offset = snapshots_offset; | |
1839 | s->snapshots_size = snapshots_size; | |
1840 | return 0; | |
1841 | fail: | |
1842 | return -1; | |
1843 | } | |
1844 | ||
1845 | static void find_new_snapshot_id(BlockDriverState *bs, | |
1846 | char *id_str, int id_str_size) | |
1847 | { | |
1848 | BDRVQcowState *s = bs->opaque; | |
1849 | QCowSnapshot *sn; | |
1850 | int i, id, id_max = 0; | |
1851 | ||
1852 | for(i = 0; i < s->nb_snapshots; i++) { | |
1853 | sn = s->snapshots + i; | |
1854 | id = strtoul(sn->id_str, NULL, 10); | |
1855 | if (id > id_max) | |
1856 | id_max = id; | |
1857 | } | |
1858 | snprintf(id_str, id_str_size, "%d", id_max + 1); | |
1859 | } | |
1860 | ||
1861 | static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str) | |
1862 | { | |
1863 | BDRVQcowState *s = bs->opaque; | |
1864 | int i; | |
1865 | ||
1866 | for(i = 0; i < s->nb_snapshots; i++) { | |
1867 | if (!strcmp(s->snapshots[i].id_str, id_str)) | |
1868 | return i; | |
1869 | } | |
1870 | return -1; | |
1871 | } | |
1872 | ||
1873 | static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name) | |
1874 | { | |
1875 | BDRVQcowState *s = bs->opaque; | |
1876 | int i, ret; | |
3b46e624 | 1877 | |
585f8587 FB |
1878 | ret = find_snapshot_by_id(bs, name); |
1879 | if (ret >= 0) | |
1880 | return ret; | |
1881 | for(i = 0; i < s->nb_snapshots; i++) { | |
1882 | if (!strcmp(s->snapshots[i].name, name)) | |
1883 | return i; | |
1884 | } | |
1885 | return -1; | |
1886 | } | |
1887 | ||
1888 | /* if no id is provided, a new one is constructed */ | |
5fafdf24 | 1889 | static int qcow_snapshot_create(BlockDriverState *bs, |
585f8587 FB |
1890 | QEMUSnapshotInfo *sn_info) |
1891 | { | |
1892 | BDRVQcowState *s = bs->opaque; | |
1893 | QCowSnapshot *snapshots1, sn1, *sn = &sn1; | |
1894 | int i, ret; | |
1895 | uint64_t *l1_table = NULL; | |
3b46e624 | 1896 | |
585f8587 FB |
1897 | memset(sn, 0, sizeof(*sn)); |
1898 | ||
1899 | if (sn_info->id_str[0] == '\0') { | |
1900 | /* compute a new id */ | |
1901 | find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str)); | |
1902 | } | |
1903 | ||
1904 | /* check that the ID is unique */ | |
1905 | if (find_snapshot_by_id(bs, sn_info->id_str) >= 0) | |
1906 | return -ENOENT; | |
1907 | ||
1908 | sn->id_str = qemu_strdup(sn_info->id_str); | |
1909 | if (!sn->id_str) | |
1910 | goto fail; | |
1911 | sn->name = qemu_strdup(sn_info->name); | |
1912 | if (!sn->name) | |
1913 | goto fail; | |
1914 | sn->vm_state_size = sn_info->vm_state_size; | |
1915 | sn->date_sec = sn_info->date_sec; | |
1916 | sn->date_nsec = sn_info->date_nsec; | |
1917 | sn->vm_clock_nsec = sn_info->vm_clock_nsec; | |
1918 | ||
1919 | ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1); | |
1920 | if (ret < 0) | |
1921 | goto fail; | |
1922 | ||
1923 | /* create the L1 table of the snapshot */ | |
1924 | sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t)); | |
1925 | sn->l1_size = s->l1_size; | |
1926 | ||
1927 | l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t)); | |
585f8587 FB |
1928 | for(i = 0; i < s->l1_size; i++) { |
1929 | l1_table[i] = cpu_to_be64(s->l1_table[i]); | |
1930 | } | |
1931 | if (bdrv_pwrite(s->hd, sn->l1_table_offset, | |
5fafdf24 | 1932 | l1_table, s->l1_size * sizeof(uint64_t)) != |
585f8587 FB |
1933 | (s->l1_size * sizeof(uint64_t))) |
1934 | goto fail; | |
1935 | qemu_free(l1_table); | |
1936 | l1_table = NULL; | |
1937 | ||
1938 | snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot)); | |
63c75dcd | 1939 | if (s->snapshots) { |
1940 | memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot)); | |
1941 | qemu_free(s->snapshots); | |
1942 | } | |
585f8587 FB |
1943 | s->snapshots = snapshots1; |
1944 | s->snapshots[s->nb_snapshots++] = *sn; | |
1945 | ||
1946 | if (qcow_write_snapshots(bs) < 0) | |
1947 | goto fail; | |
1948 | #ifdef DEBUG_ALLOC | |
1949 | check_refcounts(bs); | |
1950 | #endif | |
1951 | return 0; | |
1952 | fail: | |
1953 | qemu_free(sn->name); | |
1954 | qemu_free(l1_table); | |
1955 | return -1; | |
1956 | } | |
1957 | ||
1958 | /* copy the snapshot 'snapshot_name' into the current disk image */ | |
5fafdf24 | 1959 | static int qcow_snapshot_goto(BlockDriverState *bs, |
585f8587 FB |
1960 | const char *snapshot_id) |
1961 | { | |
1962 | BDRVQcowState *s = bs->opaque; | |
1963 | QCowSnapshot *sn; | |
1964 | int i, snapshot_index, l1_size2; | |
1965 | ||
1966 | snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id); | |
1967 | if (snapshot_index < 0) | |
1968 | return -ENOENT; | |
1969 | sn = &s->snapshots[snapshot_index]; | |
1970 | ||
1971 | if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0) | |
1972 | goto fail; | |
1973 | ||
1974 | if (grow_l1_table(bs, sn->l1_size) < 0) | |
1975 | goto fail; | |
1976 | ||
1977 | s->l1_size = sn->l1_size; | |
1978 | l1_size2 = s->l1_size * sizeof(uint64_t); | |
1979 | /* copy the snapshot l1 table to the current l1 table */ | |
5fafdf24 | 1980 | if (bdrv_pread(s->hd, sn->l1_table_offset, |
585f8587 FB |
1981 | s->l1_table, l1_size2) != l1_size2) |
1982 | goto fail; | |
1983 | if (bdrv_pwrite(s->hd, s->l1_table_offset, | |
1984 | s->l1_table, l1_size2) != l1_size2) | |
1985 | goto fail; | |
1986 | for(i = 0;i < s->l1_size; i++) { | |
1987 | be64_to_cpus(&s->l1_table[i]); | |
1988 | } | |
1989 | ||
1990 | if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0) | |
1991 | goto fail; | |
1992 | ||
1993 | #ifdef DEBUG_ALLOC | |
1994 | check_refcounts(bs); | |
1995 | #endif | |
1996 | return 0; | |
1997 | fail: | |
1998 | return -EIO; | |
1999 | } | |
2000 | ||
2001 | static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id) | |
2002 | { | |
2003 | BDRVQcowState *s = bs->opaque; | |
2004 | QCowSnapshot *sn; | |
2005 | int snapshot_index, ret; | |
3b46e624 | 2006 | |
585f8587 FB |
2007 | snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id); |
2008 | if (snapshot_index < 0) | |
2009 | return -ENOENT; | |
2010 | sn = &s->snapshots[snapshot_index]; | |
2011 | ||
2012 | ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1); | |
2013 | if (ret < 0) | |
2014 | return ret; | |
2015 | /* must update the copied flag on the current cluster offsets */ | |
2016 | ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0); | |
2017 | if (ret < 0) | |
2018 | return ret; | |
2019 | free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t)); | |
2020 | ||
2021 | qemu_free(sn->id_str); | |
2022 | qemu_free(sn->name); | |
2023 | memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn)); | |
2024 | s->nb_snapshots--; | |
2025 | ret = qcow_write_snapshots(bs); | |
2026 | if (ret < 0) { | |
2027 | /* XXX: restore snapshot if error ? */ | |
2028 | return ret; | |
2029 | } | |
2030 | #ifdef DEBUG_ALLOC | |
2031 | check_refcounts(bs); | |
2032 | #endif | |
2033 | return 0; | |
2034 | } | |
2035 | ||
5fafdf24 | 2036 | static int qcow_snapshot_list(BlockDriverState *bs, |
585f8587 FB |
2037 | QEMUSnapshotInfo **psn_tab) |
2038 | { | |
2039 | BDRVQcowState *s = bs->opaque; | |
2040 | QEMUSnapshotInfo *sn_tab, *sn_info; | |
2041 | QCowSnapshot *sn; | |
2042 | int i; | |
2043 | ||
eb0b64f7 | 2044 | if (!s->nb_snapshots) { |
2045 | *psn_tab = NULL; | |
2046 | return s->nb_snapshots; | |
2047 | } | |
2048 | ||
585f8587 | 2049 | sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo)); |
585f8587 FB |
2050 | for(i = 0; i < s->nb_snapshots; i++) { |
2051 | sn_info = sn_tab + i; | |
2052 | sn = s->snapshots + i; | |
2053 | pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), | |
2054 | sn->id_str); | |
2055 | pstrcpy(sn_info->name, sizeof(sn_info->name), | |
2056 | sn->name); | |
2057 | sn_info->vm_state_size = sn->vm_state_size; | |
2058 | sn_info->date_sec = sn->date_sec; | |
2059 | sn_info->date_nsec = sn->date_nsec; | |
2060 | sn_info->vm_clock_nsec = sn->vm_clock_nsec; | |
2061 | } | |
2062 | *psn_tab = sn_tab; | |
2063 | return s->nb_snapshots; | |
585f8587 FB |
2064 | } |
2065 | ||
e97fc193 AL |
2066 | static int qcow_check(BlockDriverState *bs) |
2067 | { | |
2068 | return check_refcounts(bs); | |
585f8587 FB |
2069 | } |
2070 | ||
2071 | #if 0 | |
2072 | static void dump_refcounts(BlockDriverState *bs) | |
2073 | { | |
2074 | BDRVQcowState *s = bs->opaque; | |
2075 | int64_t nb_clusters, k, k1, size; | |
2076 | int refcount; | |
2077 | ||
2078 | size = bdrv_getlength(s->hd); | |
6db6c638 | 2079 | nb_clusters = size_to_clusters(s, size); |
585f8587 FB |
2080 | for(k = 0; k < nb_clusters;) { |
2081 | k1 = k; | |
2082 | refcount = get_refcount(bs, k); | |
2083 | k++; | |
2084 | while (k < nb_clusters && get_refcount(bs, k) == refcount) | |
2085 | k++; | |
2086 | printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1); | |
2087 | } | |
2088 | } | |
2089 | #endif | |
585f8587 | 2090 | |
178e08a5 AL |
2091 | static int qcow_put_buffer(BlockDriverState *bs, const uint8_t *buf, |
2092 | int64_t pos, int size) | |
2093 | { | |
2094 | int growable = bs->growable; | |
2095 | ||
2096 | bs->growable = 1; | |
2097 | bdrv_pwrite(bs, pos, buf, size); | |
2098 | bs->growable = growable; | |
2099 | ||
2100 | return size; | |
2101 | } | |
2102 | ||
2103 | static int qcow_get_buffer(BlockDriverState *bs, uint8_t *buf, | |
2104 | int64_t pos, int size) | |
2105 | { | |
2106 | int growable = bs->growable; | |
2107 | int ret; | |
2108 | ||
2109 | bs->growable = 1; | |
2110 | ret = bdrv_pread(bs, pos, buf, size); | |
2111 | bs->growable = growable; | |
2112 | ||
2113 | return ret; | |
2114 | } | |
2115 | ||
0e7e1989 | 2116 | static QEMUOptionParameter qcow_create_options[] = { |
db08adf5 KW |
2117 | { |
2118 | .name = BLOCK_OPT_SIZE, | |
2119 | .type = OPT_SIZE, | |
2120 | .help = "Virtual disk size" | |
2121 | }, | |
2122 | { | |
2123 | .name = BLOCK_OPT_BACKING_FILE, | |
2124 | .type = OPT_STRING, | |
2125 | .help = "File name of a base image" | |
2126 | }, | |
2127 | { | |
2128 | .name = BLOCK_OPT_BACKING_FMT, | |
2129 | .type = OPT_STRING, | |
2130 | .help = "Image format of the base image" | |
2131 | }, | |
2132 | { | |
2133 | .name = BLOCK_OPT_ENCRYPT, | |
2134 | .type = OPT_FLAG, | |
2135 | .help = "Encrypt the image" | |
2136 | }, | |
2137 | { | |
2138 | .name = BLOCK_OPT_CLUSTER_SIZE, | |
2139 | .type = OPT_SIZE, | |
2140 | .help = "qcow2 cluster size" | |
2141 | }, | |
0e7e1989 KW |
2142 | { NULL } |
2143 | }; | |
2144 | ||
5efa9d5a | 2145 | static BlockDriver bdrv_qcow2 = { |
e60f469c AJ |
2146 | .format_name = "qcow2", |
2147 | .instance_size = sizeof(BDRVQcowState), | |
2148 | .bdrv_probe = qcow_probe, | |
2149 | .bdrv_open = qcow_open, | |
2150 | .bdrv_close = qcow_close, | |
2151 | .bdrv_create = qcow_create, | |
2152 | .bdrv_flush = qcow_flush, | |
2153 | .bdrv_is_allocated = qcow_is_allocated, | |
2154 | .bdrv_set_key = qcow_set_key, | |
2155 | .bdrv_make_empty = qcow_make_empty, | |
2156 | ||
f141eafe AL |
2157 | .bdrv_aio_readv = qcow_aio_readv, |
2158 | .bdrv_aio_writev = qcow_aio_writev, | |
585f8587 FB |
2159 | .bdrv_write_compressed = qcow_write_compressed, |
2160 | ||
2161 | .bdrv_snapshot_create = qcow_snapshot_create, | |
e60f469c | 2162 | .bdrv_snapshot_goto = qcow_snapshot_goto, |
585f8587 | 2163 | .bdrv_snapshot_delete = qcow_snapshot_delete, |
e60f469c AJ |
2164 | .bdrv_snapshot_list = qcow_snapshot_list, |
2165 | .bdrv_get_info = qcow_get_info, | |
f965509c | 2166 | |
178e08a5 AL |
2167 | .bdrv_put_buffer = qcow_put_buffer, |
2168 | .bdrv_get_buffer = qcow_get_buffer, | |
2169 | ||
0e7e1989 | 2170 | .create_options = qcow_create_options, |
e97fc193 | 2171 | .bdrv_check = qcow_check, |
585f8587 | 2172 | }; |
5efa9d5a AL |
2173 | |
2174 | static void bdrv_qcow2_init(void) | |
2175 | { | |
2176 | bdrv_register(&bdrv_qcow2); | |
2177 | } | |
2178 | ||
2179 | block_init(bdrv_qcow2_init); |