]>
Commit | Line | Data |
---|---|---|
45aba42f KW |
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 | ||
25 | #include <zlib.h> | |
26 | ||
27 | #include "qemu-common.h" | |
28 | #include "block_int.h" | |
29 | #include "block/qcow2.h" | |
30 | ||
ed6ccf0f | 31 | int qcow2_grow_l1_table(BlockDriverState *bs, int min_size) |
45aba42f KW |
32 | { |
33 | BDRVQcowState *s = bs->opaque; | |
34 | int new_l1_size, new_l1_size2, ret, i; | |
35 | uint64_t *new_l1_table; | |
36 | uint64_t new_l1_table_offset; | |
37 | uint8_t data[12]; | |
38 | ||
39 | new_l1_size = s->l1_size; | |
40 | if (min_size <= new_l1_size) | |
41 | return 0; | |
42 | while (min_size > new_l1_size) { | |
43 | new_l1_size = (new_l1_size * 3 + 1) / 2; | |
44 | } | |
45 | #ifdef DEBUG_ALLOC2 | |
46 | printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size); | |
47 | #endif | |
48 | ||
49 | new_l1_size2 = sizeof(uint64_t) * new_l1_size; | |
3f6a3ee5 | 50 | new_l1_table = qemu_mallocz(align_offset(new_l1_size2, 512)); |
45aba42f KW |
51 | memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t)); |
52 | ||
53 | /* write new table (align to cluster) */ | |
ed6ccf0f | 54 | new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2); |
45aba42f KW |
55 | |
56 | for(i = 0; i < s->l1_size; i++) | |
57 | new_l1_table[i] = cpu_to_be64(new_l1_table[i]); | |
58 | ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2); | |
59 | if (ret != new_l1_size2) | |
60 | goto fail; | |
61 | for(i = 0; i < s->l1_size; i++) | |
62 | new_l1_table[i] = be64_to_cpu(new_l1_table[i]); | |
63 | ||
64 | /* set new table */ | |
65 | cpu_to_be32w((uint32_t*)data, new_l1_size); | |
66 | cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset); | |
67 | if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data, | |
68 | sizeof(data)) != sizeof(data)) | |
69 | goto fail; | |
70 | qemu_free(s->l1_table); | |
ed6ccf0f | 71 | qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t)); |
45aba42f KW |
72 | s->l1_table_offset = new_l1_table_offset; |
73 | s->l1_table = new_l1_table; | |
74 | s->l1_size = new_l1_size; | |
75 | return 0; | |
76 | fail: | |
77 | qemu_free(s->l1_table); | |
78 | return -EIO; | |
79 | } | |
80 | ||
ed6ccf0f | 81 | void qcow2_l2_cache_reset(BlockDriverState *bs) |
45aba42f KW |
82 | { |
83 | BDRVQcowState *s = bs->opaque; | |
84 | ||
85 | memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); | |
86 | memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t)); | |
87 | memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t)); | |
88 | } | |
89 | ||
90 | static inline int l2_cache_new_entry(BlockDriverState *bs) | |
91 | { | |
92 | BDRVQcowState *s = bs->opaque; | |
93 | uint32_t min_count; | |
94 | int min_index, i; | |
95 | ||
96 | /* find a new entry in the least used one */ | |
97 | min_index = 0; | |
98 | min_count = 0xffffffff; | |
99 | for(i = 0; i < L2_CACHE_SIZE; i++) { | |
100 | if (s->l2_cache_counts[i] < min_count) { | |
101 | min_count = s->l2_cache_counts[i]; | |
102 | min_index = i; | |
103 | } | |
104 | } | |
105 | return min_index; | |
106 | } | |
107 | ||
108 | /* | |
109 | * seek_l2_table | |
110 | * | |
111 | * seek l2_offset in the l2_cache table | |
112 | * if not found, return NULL, | |
113 | * if found, | |
114 | * increments the l2 cache hit count of the entry, | |
115 | * if counter overflow, divide by two all counters | |
116 | * return the pointer to the l2 cache entry | |
117 | * | |
118 | */ | |
119 | ||
120 | static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset) | |
121 | { | |
122 | int i, j; | |
123 | ||
124 | for(i = 0; i < L2_CACHE_SIZE; i++) { | |
125 | if (l2_offset == s->l2_cache_offsets[i]) { | |
126 | /* increment the hit count */ | |
127 | if (++s->l2_cache_counts[i] == 0xffffffff) { | |
128 | for(j = 0; j < L2_CACHE_SIZE; j++) { | |
129 | s->l2_cache_counts[j] >>= 1; | |
130 | } | |
131 | } | |
132 | return s->l2_cache + (i << s->l2_bits); | |
133 | } | |
134 | } | |
135 | return NULL; | |
136 | } | |
137 | ||
138 | /* | |
139 | * l2_load | |
140 | * | |
141 | * Loads a L2 table into memory. If the table is in the cache, the cache | |
142 | * is used; otherwise the L2 table is loaded from the image file. | |
143 | * | |
144 | * Returns a pointer to the L2 table on success, or NULL if the read from | |
145 | * the image file failed. | |
146 | */ | |
147 | ||
148 | static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset) | |
149 | { | |
150 | BDRVQcowState *s = bs->opaque; | |
151 | int min_index; | |
152 | uint64_t *l2_table; | |
153 | ||
154 | /* seek if the table for the given offset is in the cache */ | |
155 | ||
156 | l2_table = seek_l2_table(s, l2_offset); | |
157 | if (l2_table != NULL) | |
158 | return l2_table; | |
159 | ||
160 | /* not found: load a new entry in the least used one */ | |
161 | ||
162 | min_index = l2_cache_new_entry(bs); | |
163 | l2_table = s->l2_cache + (min_index << s->l2_bits); | |
164 | if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) != | |
165 | s->l2_size * sizeof(uint64_t)) | |
166 | return NULL; | |
167 | s->l2_cache_offsets[min_index] = l2_offset; | |
168 | s->l2_cache_counts[min_index] = 1; | |
169 | ||
170 | return l2_table; | |
171 | } | |
172 | ||
6583e3c7 KW |
173 | /* |
174 | * Writes one sector of the L1 table to the disk (can't update single entries | |
175 | * and we really don't want bdrv_pread to perform a read-modify-write) | |
176 | */ | |
177 | #define L1_ENTRIES_PER_SECTOR (512 / 8) | |
178 | static int write_l1_entry(BDRVQcowState *s, int l1_index) | |
179 | { | |
180 | uint64_t buf[L1_ENTRIES_PER_SECTOR]; | |
181 | int l1_start_index; | |
182 | int i; | |
183 | ||
184 | l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1); | |
185 | for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) { | |
186 | buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]); | |
187 | } | |
188 | ||
189 | if (bdrv_pwrite(s->hd, s->l1_table_offset + 8 * l1_start_index, | |
190 | buf, sizeof(buf)) != sizeof(buf)) | |
191 | { | |
192 | return -1; | |
193 | } | |
194 | ||
195 | return 0; | |
196 | } | |
197 | ||
45aba42f KW |
198 | /* |
199 | * l2_allocate | |
200 | * | |
201 | * Allocate a new l2 entry in the file. If l1_index points to an already | |
202 | * used entry in the L2 table (i.e. we are doing a copy on write for the L2 | |
203 | * table) copy the contents of the old L2 table into the newly allocated one. | |
204 | * Otherwise the new table is initialized with zeros. | |
205 | * | |
206 | */ | |
207 | ||
208 | static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index) | |
209 | { | |
210 | BDRVQcowState *s = bs->opaque; | |
211 | int min_index; | |
6583e3c7 | 212 | uint64_t old_l2_offset; |
45aba42f KW |
213 | uint64_t *l2_table, l2_offset; |
214 | ||
215 | old_l2_offset = s->l1_table[l1_index]; | |
216 | ||
217 | /* allocate a new l2 entry */ | |
218 | ||
ed6ccf0f | 219 | l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t)); |
45aba42f KW |
220 | |
221 | /* update the L1 entry */ | |
222 | ||
223 | s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; | |
6583e3c7 | 224 | if (write_l1_entry(s, l1_index) < 0) { |
45aba42f | 225 | return NULL; |
6583e3c7 | 226 | } |
45aba42f KW |
227 | |
228 | /* allocate a new entry in the l2 cache */ | |
229 | ||
230 | min_index = l2_cache_new_entry(bs); | |
231 | l2_table = s->l2_cache + (min_index << s->l2_bits); | |
232 | ||
233 | if (old_l2_offset == 0) { | |
234 | /* if there was no old l2 table, clear the new table */ | |
235 | memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); | |
236 | } else { | |
237 | /* if there was an old l2 table, read it from the disk */ | |
238 | if (bdrv_pread(s->hd, old_l2_offset, | |
239 | l2_table, s->l2_size * sizeof(uint64_t)) != | |
240 | s->l2_size * sizeof(uint64_t)) | |
241 | return NULL; | |
242 | } | |
243 | /* write the l2 table to the file */ | |
244 | if (bdrv_pwrite(s->hd, l2_offset, | |
245 | l2_table, s->l2_size * sizeof(uint64_t)) != | |
246 | s->l2_size * sizeof(uint64_t)) | |
247 | return NULL; | |
248 | ||
249 | /* update the l2 cache entry */ | |
250 | ||
251 | s->l2_cache_offsets[min_index] = l2_offset; | |
252 | s->l2_cache_counts[min_index] = 1; | |
253 | ||
254 | return l2_table; | |
255 | } | |
256 | ||
257 | static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size, | |
258 | uint64_t *l2_table, uint64_t start, uint64_t mask) | |
259 | { | |
260 | int i; | |
261 | uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask; | |
262 | ||
263 | if (!offset) | |
264 | return 0; | |
265 | ||
266 | for (i = start; i < start + nb_clusters; i++) | |
267 | if (offset + i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask)) | |
268 | break; | |
269 | ||
270 | return (i - start); | |
271 | } | |
272 | ||
273 | static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table) | |
274 | { | |
275 | int i = 0; | |
276 | ||
277 | while(nb_clusters-- && l2_table[i] == 0) | |
278 | i++; | |
279 | ||
280 | return i; | |
281 | } | |
282 | ||
283 | /* The crypt function is compatible with the linux cryptoloop | |
284 | algorithm for < 4 GB images. NOTE: out_buf == in_buf is | |
285 | supported */ | |
ed6ccf0f KW |
286 | void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num, |
287 | uint8_t *out_buf, const uint8_t *in_buf, | |
288 | int nb_sectors, int enc, | |
289 | const AES_KEY *key) | |
45aba42f KW |
290 | { |
291 | union { | |
292 | uint64_t ll[2]; | |
293 | uint8_t b[16]; | |
294 | } ivec; | |
295 | int i; | |
296 | ||
297 | for(i = 0; i < nb_sectors; i++) { | |
298 | ivec.ll[0] = cpu_to_le64(sector_num); | |
299 | ivec.ll[1] = 0; | |
300 | AES_cbc_encrypt(in_buf, out_buf, 512, key, | |
301 | ivec.b, enc); | |
302 | sector_num++; | |
303 | in_buf += 512; | |
304 | out_buf += 512; | |
305 | } | |
306 | } | |
307 | ||
308 | ||
309 | static int qcow_read(BlockDriverState *bs, int64_t sector_num, | |
310 | uint8_t *buf, int nb_sectors) | |
311 | { | |
312 | BDRVQcowState *s = bs->opaque; | |
313 | int ret, index_in_cluster, n, n1; | |
314 | uint64_t cluster_offset; | |
315 | ||
316 | while (nb_sectors > 0) { | |
317 | n = nb_sectors; | |
ed6ccf0f | 318 | cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, &n); |
45aba42f KW |
319 | index_in_cluster = sector_num & (s->cluster_sectors - 1); |
320 | if (!cluster_offset) { | |
321 | if (bs->backing_hd) { | |
322 | /* read from the base image */ | |
ed6ccf0f | 323 | n1 = qcow2_backing_read1(bs->backing_hd, sector_num, buf, n); |
45aba42f KW |
324 | if (n1 > 0) { |
325 | ret = bdrv_read(bs->backing_hd, sector_num, buf, n1); | |
326 | if (ret < 0) | |
327 | return -1; | |
328 | } | |
329 | } else { | |
330 | memset(buf, 0, 512 * n); | |
331 | } | |
332 | } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) { | |
ed6ccf0f | 333 | if (qcow2_decompress_cluster(s, cluster_offset) < 0) |
45aba42f KW |
334 | return -1; |
335 | memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n); | |
336 | } else { | |
337 | ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512); | |
338 | if (ret != n * 512) | |
339 | return -1; | |
340 | if (s->crypt_method) { | |
ed6ccf0f | 341 | qcow2_encrypt_sectors(s, sector_num, buf, buf, n, 0, |
45aba42f KW |
342 | &s->aes_decrypt_key); |
343 | } | |
344 | } | |
345 | nb_sectors -= n; | |
346 | sector_num += n; | |
347 | buf += n * 512; | |
348 | } | |
349 | return 0; | |
350 | } | |
351 | ||
352 | static int copy_sectors(BlockDriverState *bs, uint64_t start_sect, | |
353 | uint64_t cluster_offset, int n_start, int n_end) | |
354 | { | |
355 | BDRVQcowState *s = bs->opaque; | |
356 | int n, ret; | |
357 | ||
358 | n = n_end - n_start; | |
359 | if (n <= 0) | |
360 | return 0; | |
361 | ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n); | |
362 | if (ret < 0) | |
363 | return ret; | |
364 | if (s->crypt_method) { | |
ed6ccf0f | 365 | qcow2_encrypt_sectors(s, start_sect + n_start, |
45aba42f KW |
366 | s->cluster_data, |
367 | s->cluster_data, n, 1, | |
368 | &s->aes_encrypt_key); | |
369 | } | |
370 | ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start, | |
371 | s->cluster_data, n); | |
372 | if (ret < 0) | |
373 | return ret; | |
374 | return 0; | |
375 | } | |
376 | ||
377 | ||
378 | /* | |
379 | * get_cluster_offset | |
380 | * | |
381 | * For a given offset of the disk image, return cluster offset in | |
382 | * qcow2 file. | |
383 | * | |
384 | * on entry, *num is the number of contiguous clusters we'd like to | |
385 | * access following offset. | |
386 | * | |
387 | * on exit, *num is the number of contiguous clusters we can read. | |
388 | * | |
389 | * Return 1, if the offset is found | |
390 | * Return 0, otherwise. | |
391 | * | |
392 | */ | |
393 | ||
ed6ccf0f KW |
394 | uint64_t qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, |
395 | int *num) | |
45aba42f KW |
396 | { |
397 | BDRVQcowState *s = bs->opaque; | |
398 | int l1_index, l2_index; | |
399 | uint64_t l2_offset, *l2_table, cluster_offset; | |
400 | int l1_bits, c; | |
401 | int index_in_cluster, nb_available, nb_needed, nb_clusters; | |
402 | ||
403 | index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1); | |
404 | nb_needed = *num + index_in_cluster; | |
405 | ||
406 | l1_bits = s->l2_bits + s->cluster_bits; | |
407 | ||
408 | /* compute how many bytes there are between the offset and | |
409 | * the end of the l1 entry | |
410 | */ | |
411 | ||
412 | nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1)); | |
413 | ||
414 | /* compute the number of available sectors */ | |
415 | ||
416 | nb_available = (nb_available >> 9) + index_in_cluster; | |
417 | ||
418 | if (nb_needed > nb_available) { | |
419 | nb_needed = nb_available; | |
420 | } | |
421 | ||
422 | cluster_offset = 0; | |
423 | ||
424 | /* seek the the l2 offset in the l1 table */ | |
425 | ||
426 | l1_index = offset >> l1_bits; | |
427 | if (l1_index >= s->l1_size) | |
428 | goto out; | |
429 | ||
430 | l2_offset = s->l1_table[l1_index]; | |
431 | ||
432 | /* seek the l2 table of the given l2 offset */ | |
433 | ||
434 | if (!l2_offset) | |
435 | goto out; | |
436 | ||
437 | /* load the l2 table in memory */ | |
438 | ||
439 | l2_offset &= ~QCOW_OFLAG_COPIED; | |
440 | l2_table = l2_load(bs, l2_offset); | |
441 | if (l2_table == NULL) | |
442 | return 0; | |
443 | ||
444 | /* find the cluster offset for the given disk offset */ | |
445 | ||
446 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
447 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
448 | nb_clusters = size_to_clusters(s, nb_needed << 9); | |
449 | ||
450 | if (!cluster_offset) { | |
451 | /* how many empty clusters ? */ | |
452 | c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]); | |
453 | } else { | |
454 | /* how many allocated clusters ? */ | |
455 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
456 | &l2_table[l2_index], 0, QCOW_OFLAG_COPIED); | |
457 | } | |
458 | ||
459 | nb_available = (c * s->cluster_sectors); | |
460 | out: | |
461 | if (nb_available > nb_needed) | |
462 | nb_available = nb_needed; | |
463 | ||
464 | *num = nb_available - index_in_cluster; | |
465 | ||
466 | return cluster_offset & ~QCOW_OFLAG_COPIED; | |
467 | } | |
468 | ||
469 | /* | |
470 | * get_cluster_table | |
471 | * | |
472 | * for a given disk offset, load (and allocate if needed) | |
473 | * the l2 table. | |
474 | * | |
475 | * the l2 table offset in the qcow2 file and the cluster index | |
476 | * in the l2 table are given to the caller. | |
477 | * | |
478 | */ | |
479 | ||
480 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, | |
481 | uint64_t **new_l2_table, | |
482 | uint64_t *new_l2_offset, | |
483 | int *new_l2_index) | |
484 | { | |
485 | BDRVQcowState *s = bs->opaque; | |
486 | int l1_index, l2_index, ret; | |
487 | uint64_t l2_offset, *l2_table; | |
488 | ||
489 | /* seek the the l2 offset in the l1 table */ | |
490 | ||
491 | l1_index = offset >> (s->l2_bits + s->cluster_bits); | |
492 | if (l1_index >= s->l1_size) { | |
ed6ccf0f | 493 | ret = qcow2_grow_l1_table(bs, l1_index + 1); |
45aba42f KW |
494 | if (ret < 0) |
495 | return 0; | |
496 | } | |
497 | l2_offset = s->l1_table[l1_index]; | |
498 | ||
499 | /* seek the l2 table of the given l2 offset */ | |
500 | ||
501 | if (l2_offset & QCOW_OFLAG_COPIED) { | |
502 | /* load the l2 table in memory */ | |
503 | l2_offset &= ~QCOW_OFLAG_COPIED; | |
504 | l2_table = l2_load(bs, l2_offset); | |
505 | if (l2_table == NULL) | |
506 | return 0; | |
507 | } else { | |
508 | if (l2_offset) | |
ed6ccf0f | 509 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t)); |
45aba42f KW |
510 | l2_table = l2_allocate(bs, l1_index); |
511 | if (l2_table == NULL) | |
512 | return 0; | |
513 | l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED; | |
514 | } | |
515 | ||
516 | /* find the cluster offset for the given disk offset */ | |
517 | ||
518 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
519 | ||
520 | *new_l2_table = l2_table; | |
521 | *new_l2_offset = l2_offset; | |
522 | *new_l2_index = l2_index; | |
523 | ||
524 | return 1; | |
525 | } | |
526 | ||
527 | /* | |
528 | * alloc_compressed_cluster_offset | |
529 | * | |
530 | * For a given offset of the disk image, return cluster offset in | |
531 | * qcow2 file. | |
532 | * | |
533 | * If the offset is not found, allocate a new compressed cluster. | |
534 | * | |
535 | * Return the cluster offset if successful, | |
536 | * Return 0, otherwise. | |
537 | * | |
538 | */ | |
539 | ||
ed6ccf0f KW |
540 | uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
541 | uint64_t offset, | |
542 | int compressed_size) | |
45aba42f KW |
543 | { |
544 | BDRVQcowState *s = bs->opaque; | |
545 | int l2_index, ret; | |
546 | uint64_t l2_offset, *l2_table, cluster_offset; | |
547 | int nb_csectors; | |
548 | ||
549 | ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index); | |
550 | if (ret == 0) | |
551 | return 0; | |
552 | ||
553 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
554 | if (cluster_offset & QCOW_OFLAG_COPIED) | |
555 | return cluster_offset & ~QCOW_OFLAG_COPIED; | |
556 | ||
557 | if (cluster_offset) | |
ed6ccf0f | 558 | qcow2_free_any_clusters(bs, cluster_offset, 1); |
45aba42f | 559 | |
ed6ccf0f | 560 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
45aba42f KW |
561 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
562 | (cluster_offset >> 9); | |
563 | ||
564 | cluster_offset |= QCOW_OFLAG_COMPRESSED | | |
565 | ((uint64_t)nb_csectors << s->csize_shift); | |
566 | ||
567 | /* update L2 table */ | |
568 | ||
569 | /* compressed clusters never have the copied flag */ | |
570 | ||
571 | l2_table[l2_index] = cpu_to_be64(cluster_offset); | |
572 | if (bdrv_pwrite(s->hd, | |
573 | l2_offset + l2_index * sizeof(uint64_t), | |
574 | l2_table + l2_index, | |
575 | sizeof(uint64_t)) != sizeof(uint64_t)) | |
576 | return 0; | |
577 | ||
578 | return cluster_offset; | |
579 | } | |
580 | ||
4c1612d9 KW |
581 | /* |
582 | * Write L2 table updates to disk, writing whole sectors to avoid a | |
583 | * read-modify-write in bdrv_pwrite | |
584 | */ | |
585 | #define L2_ENTRIES_PER_SECTOR (512 / 8) | |
586 | static int write_l2_entries(BDRVQcowState *s, uint64_t *l2_table, | |
587 | uint64_t l2_offset, int l2_index, int num) | |
588 | { | |
589 | int l2_start_index = l2_index & ~(L1_ENTRIES_PER_SECTOR - 1); | |
590 | int start_offset = (8 * l2_index) & ~511; | |
591 | int end_offset = (8 * (l2_index + num) + 511) & ~511; | |
592 | size_t len = end_offset - start_offset; | |
593 | ||
594 | if (bdrv_pwrite(s->hd, l2_offset + start_offset, &l2_table[l2_start_index], | |
595 | len) != len) | |
596 | { | |
597 | return -1; | |
598 | } | |
599 | ||
600 | return 0; | |
601 | } | |
602 | ||
ed6ccf0f | 603 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, uint64_t cluster_offset, |
45aba42f KW |
604 | QCowL2Meta *m) |
605 | { | |
606 | BDRVQcowState *s = bs->opaque; | |
607 | int i, j = 0, l2_index, ret; | |
608 | uint64_t *old_cluster, start_sect, l2_offset, *l2_table; | |
609 | ||
610 | if (m->nb_clusters == 0) | |
611 | return 0; | |
612 | ||
613 | old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t)); | |
614 | ||
615 | /* copy content of unmodified sectors */ | |
616 | start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9; | |
617 | if (m->n_start) { | |
618 | ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start); | |
619 | if (ret < 0) | |
620 | goto err; | |
621 | } | |
622 | ||
623 | if (m->nb_available & (s->cluster_sectors - 1)) { | |
624 | uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1); | |
625 | ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9), | |
626 | m->nb_available - end, s->cluster_sectors); | |
627 | if (ret < 0) | |
628 | goto err; | |
629 | } | |
630 | ||
631 | ret = -EIO; | |
632 | /* update L2 table */ | |
633 | if (!get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index)) | |
634 | goto err; | |
635 | ||
636 | for (i = 0; i < m->nb_clusters; i++) { | |
637 | /* if two concurrent writes happen to the same unallocated cluster | |
638 | * each write allocates separate cluster and writes data concurrently. | |
639 | * The first one to complete updates l2 table with pointer to its | |
640 | * cluster the second one has to do RMW (which is done above by | |
641 | * copy_sectors()), update l2 table with its cluster pointer and free | |
642 | * old cluster. This is what this loop does */ | |
643 | if(l2_table[l2_index + i] != 0) | |
644 | old_cluster[j++] = l2_table[l2_index + i]; | |
645 | ||
646 | l2_table[l2_index + i] = cpu_to_be64((cluster_offset + | |
647 | (i << s->cluster_bits)) | QCOW_OFLAG_COPIED); | |
648 | } | |
649 | ||
4c1612d9 KW |
650 | if (write_l2_entries(s, l2_table, l2_offset, l2_index, m->nb_clusters) < 0) { |
651 | ret = -1; | |
45aba42f | 652 | goto err; |
4c1612d9 | 653 | } |
45aba42f KW |
654 | |
655 | for (i = 0; i < j; i++) | |
ed6ccf0f KW |
656 | qcow2_free_any_clusters(bs, |
657 | be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, 1); | |
45aba42f KW |
658 | |
659 | ret = 0; | |
660 | err: | |
661 | qemu_free(old_cluster); | |
662 | return ret; | |
663 | } | |
664 | ||
665 | /* | |
666 | * alloc_cluster_offset | |
667 | * | |
668 | * For a given offset of the disk image, return cluster offset in | |
669 | * qcow2 file. | |
670 | * | |
671 | * If the offset is not found, allocate a new cluster. | |
672 | * | |
673 | * Return the cluster offset if successful, | |
674 | * Return 0, otherwise. | |
675 | * | |
676 | */ | |
677 | ||
ed6ccf0f KW |
678 | uint64_t qcow2_alloc_cluster_offset(BlockDriverState *bs, |
679 | uint64_t offset, | |
680 | int n_start, int n_end, | |
681 | int *num, QCowL2Meta *m) | |
45aba42f KW |
682 | { |
683 | BDRVQcowState *s = bs->opaque; | |
684 | int l2_index, ret; | |
685 | uint64_t l2_offset, *l2_table, cluster_offset; | |
686 | int nb_clusters, i = 0; | |
f214978a | 687 | QCowL2Meta *old_alloc; |
45aba42f KW |
688 | |
689 | ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index); | |
690 | if (ret == 0) | |
691 | return 0; | |
692 | ||
693 | nb_clusters = size_to_clusters(s, n_end << 9); | |
694 | ||
695 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
696 | ||
697 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
698 | ||
699 | /* We keep all QCOW_OFLAG_COPIED clusters */ | |
700 | ||
701 | if (cluster_offset & QCOW_OFLAG_COPIED) { | |
702 | nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
703 | &l2_table[l2_index], 0, 0); | |
704 | ||
705 | cluster_offset &= ~QCOW_OFLAG_COPIED; | |
706 | m->nb_clusters = 0; | |
707 | ||
708 | goto out; | |
709 | } | |
710 | ||
711 | /* for the moment, multiple compressed clusters are not managed */ | |
712 | ||
713 | if (cluster_offset & QCOW_OFLAG_COMPRESSED) | |
714 | nb_clusters = 1; | |
715 | ||
716 | /* how many available clusters ? */ | |
717 | ||
718 | while (i < nb_clusters) { | |
719 | i += count_contiguous_clusters(nb_clusters - i, s->cluster_size, | |
720 | &l2_table[l2_index], i, 0); | |
721 | ||
722 | if(be64_to_cpu(l2_table[l2_index + i])) | |
723 | break; | |
724 | ||
725 | i += count_contiguous_free_clusters(nb_clusters - i, | |
726 | &l2_table[l2_index + i]); | |
727 | ||
728 | cluster_offset = be64_to_cpu(l2_table[l2_index + i]); | |
729 | ||
730 | if ((cluster_offset & QCOW_OFLAG_COPIED) || | |
731 | (cluster_offset & QCOW_OFLAG_COMPRESSED)) | |
732 | break; | |
733 | } | |
734 | nb_clusters = i; | |
735 | ||
f214978a KW |
736 | /* |
737 | * Check if there already is an AIO write request in flight which allocates | |
738 | * the same cluster. In this case we need to wait until the previous | |
739 | * request has completed and updated the L2 table accordingly. | |
740 | */ | |
72cf2d4f | 741 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
f214978a KW |
742 | |
743 | uint64_t end_offset = offset + nb_clusters * s->cluster_size; | |
744 | uint64_t old_offset = old_alloc->offset; | |
745 | uint64_t old_end_offset = old_alloc->offset + | |
746 | old_alloc->nb_clusters * s->cluster_size; | |
747 | ||
748 | if (end_offset < old_offset || offset > old_end_offset) { | |
749 | /* No intersection */ | |
750 | } else { | |
751 | if (offset < old_offset) { | |
752 | /* Stop at the start of a running allocation */ | |
753 | nb_clusters = (old_offset - offset) >> s->cluster_bits; | |
754 | } else { | |
755 | nb_clusters = 0; | |
756 | } | |
757 | ||
758 | if (nb_clusters == 0) { | |
759 | /* Set dependency and wait for a callback */ | |
760 | m->depends_on = old_alloc; | |
761 | m->nb_clusters = 0; | |
762 | *num = 0; | |
763 | return 0; | |
764 | } | |
765 | } | |
766 | } | |
767 | ||
768 | if (!nb_clusters) { | |
769 | abort(); | |
770 | } | |
771 | ||
72cf2d4f | 772 | QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight); |
f214978a | 773 | |
45aba42f KW |
774 | /* allocate a new cluster */ |
775 | ||
ed6ccf0f | 776 | cluster_offset = qcow2_alloc_clusters(bs, nb_clusters * s->cluster_size); |
45aba42f KW |
777 | |
778 | /* save info needed for meta data update */ | |
779 | m->offset = offset; | |
780 | m->n_start = n_start; | |
781 | m->nb_clusters = nb_clusters; | |
782 | ||
783 | out: | |
784 | m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end); | |
785 | ||
786 | *num = m->nb_available - n_start; | |
787 | ||
788 | return cluster_offset; | |
789 | } | |
790 | ||
791 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
792 | const uint8_t *buf, int buf_size) | |
793 | { | |
794 | z_stream strm1, *strm = &strm1; | |
795 | int ret, out_len; | |
796 | ||
797 | memset(strm, 0, sizeof(*strm)); | |
798 | ||
799 | strm->next_in = (uint8_t *)buf; | |
800 | strm->avail_in = buf_size; | |
801 | strm->next_out = out_buf; | |
802 | strm->avail_out = out_buf_size; | |
803 | ||
804 | ret = inflateInit2(strm, -12); | |
805 | if (ret != Z_OK) | |
806 | return -1; | |
807 | ret = inflate(strm, Z_FINISH); | |
808 | out_len = strm->next_out - out_buf; | |
809 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
810 | out_len != out_buf_size) { | |
811 | inflateEnd(strm); | |
812 | return -1; | |
813 | } | |
814 | inflateEnd(strm); | |
815 | return 0; | |
816 | } | |
817 | ||
ed6ccf0f | 818 | int qcow2_decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset) |
45aba42f KW |
819 | { |
820 | int ret, csize, nb_csectors, sector_offset; | |
821 | uint64_t coffset; | |
822 | ||
823 | coffset = cluster_offset & s->cluster_offset_mask; | |
824 | if (s->cluster_cache_offset != coffset) { | |
825 | nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1; | |
826 | sector_offset = coffset & 511; | |
827 | csize = nb_csectors * 512 - sector_offset; | |
828 | ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors); | |
829 | if (ret < 0) { | |
830 | return -1; | |
831 | } | |
832 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
833 | s->cluster_data + sector_offset, csize) < 0) { | |
834 | return -1; | |
835 | } | |
836 | s->cluster_cache_offset = coffset; | |
837 | } | |
838 | return 0; | |
839 | } |