]>
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; | |
5d757b56 | 36 | int64_t new_l1_table_offset; |
45aba42f KW |
37 | uint8_t data[12]; |
38 | ||
39 | new_l1_size = s->l1_size; | |
40 | if (min_size <= new_l1_size) | |
41 | return 0; | |
d191d12d SW |
42 | if (new_l1_size == 0) { |
43 | new_l1_size = 1; | |
44 | } | |
45aba42f KW |
45 | while (min_size > new_l1_size) { |
46 | new_l1_size = (new_l1_size * 3 + 1) / 2; | |
47 | } | |
48 | #ifdef DEBUG_ALLOC2 | |
49 | printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size); | |
50 | #endif | |
51 | ||
52 | new_l1_size2 = sizeof(uint64_t) * new_l1_size; | |
3f6a3ee5 | 53 | new_l1_table = qemu_mallocz(align_offset(new_l1_size2, 512)); |
45aba42f KW |
54 | memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t)); |
55 | ||
56 | /* write new table (align to cluster) */ | |
66f82cee | 57 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE); |
ed6ccf0f | 58 | new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2); |
5d757b56 KW |
59 | if (new_l1_table_offset < 0) { |
60 | qemu_free(new_l1_table); | |
61 | return new_l1_table_offset; | |
62 | } | |
29216ed1 | 63 | bdrv_flush(bs->file); |
45aba42f | 64 | |
66f82cee | 65 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE); |
45aba42f KW |
66 | for(i = 0; i < s->l1_size; i++) |
67 | new_l1_table[i] = cpu_to_be64(new_l1_table[i]); | |
8b3b7206 KW |
68 | ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, new_l1_table, new_l1_size2); |
69 | if (ret < 0) | |
45aba42f KW |
70 | goto fail; |
71 | for(i = 0; i < s->l1_size; i++) | |
72 | new_l1_table[i] = be64_to_cpu(new_l1_table[i]); | |
73 | ||
74 | /* set new table */ | |
66f82cee | 75 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE); |
45aba42f KW |
76 | cpu_to_be32w((uint32_t*)data, new_l1_size); |
77 | cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset); | |
8b3b7206 KW |
78 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), data,sizeof(data)); |
79 | if (ret < 0) { | |
45aba42f | 80 | goto fail; |
fb8fa77c | 81 | } |
45aba42f | 82 | qemu_free(s->l1_table); |
ed6ccf0f | 83 | qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t)); |
45aba42f KW |
84 | s->l1_table_offset = new_l1_table_offset; |
85 | s->l1_table = new_l1_table; | |
86 | s->l1_size = new_l1_size; | |
87 | return 0; | |
88 | fail: | |
fb8fa77c KW |
89 | qemu_free(new_l1_table); |
90 | qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2); | |
8b3b7206 | 91 | return ret; |
45aba42f KW |
92 | } |
93 | ||
ed6ccf0f | 94 | void qcow2_l2_cache_reset(BlockDriverState *bs) |
45aba42f KW |
95 | { |
96 | BDRVQcowState *s = bs->opaque; | |
97 | ||
98 | memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t)); | |
99 | memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t)); | |
100 | memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t)); | |
101 | } | |
102 | ||
103 | static inline int l2_cache_new_entry(BlockDriverState *bs) | |
104 | { | |
105 | BDRVQcowState *s = bs->opaque; | |
106 | uint32_t min_count; | |
107 | int min_index, i; | |
108 | ||
109 | /* find a new entry in the least used one */ | |
110 | min_index = 0; | |
111 | min_count = 0xffffffff; | |
112 | for(i = 0; i < L2_CACHE_SIZE; i++) { | |
113 | if (s->l2_cache_counts[i] < min_count) { | |
114 | min_count = s->l2_cache_counts[i]; | |
115 | min_index = i; | |
116 | } | |
117 | } | |
118 | return min_index; | |
119 | } | |
120 | ||
121 | /* | |
122 | * seek_l2_table | |
123 | * | |
124 | * seek l2_offset in the l2_cache table | |
125 | * if not found, return NULL, | |
126 | * if found, | |
127 | * increments the l2 cache hit count of the entry, | |
128 | * if counter overflow, divide by two all counters | |
129 | * return the pointer to the l2 cache entry | |
130 | * | |
131 | */ | |
132 | ||
133 | static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset) | |
134 | { | |
135 | int i, j; | |
136 | ||
137 | for(i = 0; i < L2_CACHE_SIZE; i++) { | |
138 | if (l2_offset == s->l2_cache_offsets[i]) { | |
139 | /* increment the hit count */ | |
140 | if (++s->l2_cache_counts[i] == 0xffffffff) { | |
141 | for(j = 0; j < L2_CACHE_SIZE; j++) { | |
142 | s->l2_cache_counts[j] >>= 1; | |
143 | } | |
144 | } | |
145 | return s->l2_cache + (i << s->l2_bits); | |
146 | } | |
147 | } | |
148 | return NULL; | |
149 | } | |
150 | ||
151 | /* | |
152 | * l2_load | |
153 | * | |
154 | * Loads a L2 table into memory. If the table is in the cache, the cache | |
155 | * is used; otherwise the L2 table is loaded from the image file. | |
156 | * | |
157 | * Returns a pointer to the L2 table on success, or NULL if the read from | |
158 | * the image file failed. | |
159 | */ | |
160 | ||
55c17e98 KW |
161 | static int l2_load(BlockDriverState *bs, uint64_t l2_offset, |
162 | uint64_t **l2_table) | |
45aba42f KW |
163 | { |
164 | BDRVQcowState *s = bs->opaque; | |
165 | int min_index; | |
55c17e98 | 166 | int ret; |
45aba42f KW |
167 | |
168 | /* seek if the table for the given offset is in the cache */ | |
169 | ||
55c17e98 KW |
170 | *l2_table = seek_l2_table(s, l2_offset); |
171 | if (*l2_table != NULL) { | |
172 | return 0; | |
173 | } | |
45aba42f KW |
174 | |
175 | /* not found: load a new entry in the least used one */ | |
176 | ||
177 | min_index = l2_cache_new_entry(bs); | |
55c17e98 | 178 | *l2_table = s->l2_cache + (min_index << s->l2_bits); |
8252278a | 179 | |
66f82cee | 180 | BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD); |
55c17e98 KW |
181 | ret = bdrv_pread(bs->file, l2_offset, *l2_table, |
182 | s->l2_size * sizeof(uint64_t)); | |
183 | if (ret < 0) { | |
184 | return ret; | |
185 | } | |
186 | ||
45aba42f KW |
187 | s->l2_cache_offsets[min_index] = l2_offset; |
188 | s->l2_cache_counts[min_index] = 1; | |
189 | ||
55c17e98 | 190 | return 0; |
45aba42f KW |
191 | } |
192 | ||
6583e3c7 KW |
193 | /* |
194 | * Writes one sector of the L1 table to the disk (can't update single entries | |
195 | * and we really don't want bdrv_pread to perform a read-modify-write) | |
196 | */ | |
197 | #define L1_ENTRIES_PER_SECTOR (512 / 8) | |
66f82cee | 198 | static int write_l1_entry(BlockDriverState *bs, int l1_index) |
6583e3c7 | 199 | { |
66f82cee | 200 | BDRVQcowState *s = bs->opaque; |
6583e3c7 KW |
201 | uint64_t buf[L1_ENTRIES_PER_SECTOR]; |
202 | int l1_start_index; | |
f7defcb6 | 203 | int i, ret; |
6583e3c7 KW |
204 | |
205 | l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1); | |
206 | for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) { | |
207 | buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]); | |
208 | } | |
209 | ||
66f82cee | 210 | BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); |
8b3b7206 | 211 | ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset + 8 * l1_start_index, |
f7defcb6 KW |
212 | buf, sizeof(buf)); |
213 | if (ret < 0) { | |
214 | return ret; | |
6583e3c7 KW |
215 | } |
216 | ||
217 | return 0; | |
218 | } | |
219 | ||
45aba42f KW |
220 | /* |
221 | * l2_allocate | |
222 | * | |
223 | * Allocate a new l2 entry in the file. If l1_index points to an already | |
224 | * used entry in the L2 table (i.e. we are doing a copy on write for the L2 | |
225 | * table) copy the contents of the old L2 table into the newly allocated one. | |
226 | * Otherwise the new table is initialized with zeros. | |
227 | * | |
228 | */ | |
229 | ||
c46e1167 | 230 | static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table) |
45aba42f KW |
231 | { |
232 | BDRVQcowState *s = bs->opaque; | |
233 | int min_index; | |
6583e3c7 | 234 | uint64_t old_l2_offset; |
f4f0d391 KW |
235 | uint64_t *l2_table; |
236 | int64_t l2_offset; | |
c46e1167 | 237 | int ret; |
45aba42f KW |
238 | |
239 | old_l2_offset = s->l1_table[l1_index]; | |
240 | ||
241 | /* allocate a new l2 entry */ | |
242 | ||
ed6ccf0f | 243 | l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t)); |
5d757b56 | 244 | if (l2_offset < 0) { |
c46e1167 | 245 | return l2_offset; |
5d757b56 | 246 | } |
29216ed1 | 247 | bdrv_flush(bs->file); |
45aba42f | 248 | |
45aba42f KW |
249 | /* allocate a new entry in the l2 cache */ |
250 | ||
251 | min_index = l2_cache_new_entry(bs); | |
252 | l2_table = s->l2_cache + (min_index << s->l2_bits); | |
253 | ||
254 | if (old_l2_offset == 0) { | |
255 | /* if there was no old l2 table, clear the new table */ | |
256 | memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); | |
257 | } else { | |
258 | /* if there was an old l2 table, read it from the disk */ | |
66f82cee KW |
259 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ); |
260 | ret = bdrv_pread(bs->file, old_l2_offset, l2_table, | |
c46e1167 KW |
261 | s->l2_size * sizeof(uint64_t)); |
262 | if (ret < 0) { | |
175e1152 | 263 | goto fail; |
c46e1167 | 264 | } |
45aba42f KW |
265 | } |
266 | /* write the l2 table to the file */ | |
66f82cee | 267 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE); |
8b3b7206 | 268 | ret = bdrv_pwrite_sync(bs->file, l2_offset, l2_table, |
c46e1167 KW |
269 | s->l2_size * sizeof(uint64_t)); |
270 | if (ret < 0) { | |
175e1152 KW |
271 | goto fail; |
272 | } | |
273 | ||
274 | /* update the L1 entry */ | |
275 | s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; | |
276 | ret = write_l1_entry(bs, l1_index); | |
277 | if (ret < 0) { | |
278 | goto fail; | |
c46e1167 | 279 | } |
45aba42f KW |
280 | |
281 | /* update the l2 cache entry */ | |
282 | ||
283 | s->l2_cache_offsets[min_index] = l2_offset; | |
284 | s->l2_cache_counts[min_index] = 1; | |
285 | ||
c46e1167 KW |
286 | *table = l2_table; |
287 | return 0; | |
175e1152 KW |
288 | |
289 | fail: | |
68dba0bf | 290 | s->l1_table[l1_index] = old_l2_offset; |
175e1152 KW |
291 | qcow2_l2_cache_reset(bs); |
292 | return ret; | |
45aba42f KW |
293 | } |
294 | ||
295 | static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size, | |
296 | uint64_t *l2_table, uint64_t start, uint64_t mask) | |
297 | { | |
298 | int i; | |
299 | uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask; | |
300 | ||
301 | if (!offset) | |
302 | return 0; | |
303 | ||
304 | for (i = start; i < start + nb_clusters; i++) | |
80ee15a6 | 305 | if (offset + (uint64_t) i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask)) |
45aba42f KW |
306 | break; |
307 | ||
308 | return (i - start); | |
309 | } | |
310 | ||
311 | static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table) | |
312 | { | |
313 | int i = 0; | |
314 | ||
315 | while(nb_clusters-- && l2_table[i] == 0) | |
316 | i++; | |
317 | ||
318 | return i; | |
319 | } | |
320 | ||
321 | /* The crypt function is compatible with the linux cryptoloop | |
322 | algorithm for < 4 GB images. NOTE: out_buf == in_buf is | |
323 | supported */ | |
ed6ccf0f KW |
324 | void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num, |
325 | uint8_t *out_buf, const uint8_t *in_buf, | |
326 | int nb_sectors, int enc, | |
327 | const AES_KEY *key) | |
45aba42f KW |
328 | { |
329 | union { | |
330 | uint64_t ll[2]; | |
331 | uint8_t b[16]; | |
332 | } ivec; | |
333 | int i; | |
334 | ||
335 | for(i = 0; i < nb_sectors; i++) { | |
336 | ivec.ll[0] = cpu_to_le64(sector_num); | |
337 | ivec.ll[1] = 0; | |
338 | AES_cbc_encrypt(in_buf, out_buf, 512, key, | |
339 | ivec.b, enc); | |
340 | sector_num++; | |
341 | in_buf += 512; | |
342 | out_buf += 512; | |
343 | } | |
344 | } | |
345 | ||
346 | ||
72ecf02d KW |
347 | static int qcow_read(BlockDriverState *bs, int64_t sector_num, |
348 | uint8_t *buf, int nb_sectors) | |
45aba42f KW |
349 | { |
350 | BDRVQcowState *s = bs->opaque; | |
351 | int ret, index_in_cluster, n, n1; | |
352 | uint64_t cluster_offset; | |
353 | ||
354 | while (nb_sectors > 0) { | |
355 | n = nb_sectors; | |
1c46efaa KW |
356 | |
357 | ret = qcow2_get_cluster_offset(bs, sector_num << 9, &n, | |
358 | &cluster_offset); | |
359 | if (ret < 0) { | |
360 | return ret; | |
361 | } | |
362 | ||
45aba42f KW |
363 | index_in_cluster = sector_num & (s->cluster_sectors - 1); |
364 | if (!cluster_offset) { | |
365 | if (bs->backing_hd) { | |
366 | /* read from the base image */ | |
ed6ccf0f | 367 | n1 = qcow2_backing_read1(bs->backing_hd, sector_num, buf, n); |
45aba42f | 368 | if (n1 > 0) { |
66f82cee | 369 | BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING); |
45aba42f KW |
370 | ret = bdrv_read(bs->backing_hd, sector_num, buf, n1); |
371 | if (ret < 0) | |
372 | return -1; | |
373 | } | |
374 | } else { | |
375 | memset(buf, 0, 512 * n); | |
376 | } | |
377 | } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) { | |
66f82cee | 378 | if (qcow2_decompress_cluster(bs, cluster_offset) < 0) |
45aba42f KW |
379 | return -1; |
380 | memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n); | |
381 | } else { | |
66f82cee KW |
382 | BLKDBG_EVENT(bs->file, BLKDBG_READ); |
383 | ret = bdrv_pread(bs->file, cluster_offset + index_in_cluster * 512, buf, n * 512); | |
45aba42f KW |
384 | if (ret != n * 512) |
385 | return -1; | |
386 | if (s->crypt_method) { | |
ed6ccf0f | 387 | qcow2_encrypt_sectors(s, sector_num, buf, buf, n, 0, |
45aba42f KW |
388 | &s->aes_decrypt_key); |
389 | } | |
390 | } | |
391 | nb_sectors -= n; | |
392 | sector_num += n; | |
393 | buf += n * 512; | |
394 | } | |
395 | return 0; | |
396 | } | |
397 | ||
398 | static int copy_sectors(BlockDriverState *bs, uint64_t start_sect, | |
399 | uint64_t cluster_offset, int n_start, int n_end) | |
400 | { | |
401 | BDRVQcowState *s = bs->opaque; | |
402 | int n, ret; | |
403 | ||
404 | n = n_end - n_start; | |
405 | if (n <= 0) | |
406 | return 0; | |
66f82cee | 407 | BLKDBG_EVENT(bs->file, BLKDBG_COW_READ); |
72ecf02d | 408 | ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n); |
45aba42f KW |
409 | if (ret < 0) |
410 | return ret; | |
411 | if (s->crypt_method) { | |
ed6ccf0f | 412 | qcow2_encrypt_sectors(s, start_sect + n_start, |
45aba42f KW |
413 | s->cluster_data, |
414 | s->cluster_data, n, 1, | |
415 | &s->aes_encrypt_key); | |
416 | } | |
66f82cee | 417 | BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE); |
8b3b7206 KW |
418 | ret = bdrv_write_sync(bs->file, (cluster_offset >> 9) + n_start, |
419 | s->cluster_data, n); | |
45aba42f KW |
420 | if (ret < 0) |
421 | return ret; | |
422 | return 0; | |
423 | } | |
424 | ||
425 | ||
426 | /* | |
427 | * get_cluster_offset | |
428 | * | |
1c46efaa KW |
429 | * For a given offset of the disk image, find the cluster offset in |
430 | * qcow2 file. The offset is stored in *cluster_offset. | |
45aba42f KW |
431 | * |
432 | * on entry, *num is the number of contiguous clusters we'd like to | |
433 | * access following offset. | |
434 | * | |
435 | * on exit, *num is the number of contiguous clusters we can read. | |
436 | * | |
1c46efaa KW |
437 | * Return 0, if the offset is found |
438 | * Return -errno, otherwise. | |
45aba42f KW |
439 | * |
440 | */ | |
441 | ||
1c46efaa KW |
442 | int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, |
443 | int *num, uint64_t *cluster_offset) | |
45aba42f KW |
444 | { |
445 | BDRVQcowState *s = bs->opaque; | |
80ee15a6 | 446 | unsigned int l1_index, l2_index; |
1c46efaa | 447 | uint64_t l2_offset, *l2_table; |
45aba42f | 448 | int l1_bits, c; |
80ee15a6 KW |
449 | unsigned int index_in_cluster, nb_clusters; |
450 | uint64_t nb_available, nb_needed; | |
55c17e98 | 451 | int ret; |
45aba42f KW |
452 | |
453 | index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1); | |
454 | nb_needed = *num + index_in_cluster; | |
455 | ||
456 | l1_bits = s->l2_bits + s->cluster_bits; | |
457 | ||
458 | /* compute how many bytes there are between the offset and | |
459 | * the end of the l1 entry | |
460 | */ | |
461 | ||
80ee15a6 | 462 | nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1)); |
45aba42f KW |
463 | |
464 | /* compute the number of available sectors */ | |
465 | ||
466 | nb_available = (nb_available >> 9) + index_in_cluster; | |
467 | ||
468 | if (nb_needed > nb_available) { | |
469 | nb_needed = nb_available; | |
470 | } | |
471 | ||
1c46efaa | 472 | *cluster_offset = 0; |
45aba42f KW |
473 | |
474 | /* seek the the l2 offset in the l1 table */ | |
475 | ||
476 | l1_index = offset >> l1_bits; | |
477 | if (l1_index >= s->l1_size) | |
478 | goto out; | |
479 | ||
480 | l2_offset = s->l1_table[l1_index]; | |
481 | ||
482 | /* seek the l2 table of the given l2 offset */ | |
483 | ||
484 | if (!l2_offset) | |
485 | goto out; | |
486 | ||
487 | /* load the l2 table in memory */ | |
488 | ||
489 | l2_offset &= ~QCOW_OFLAG_COPIED; | |
55c17e98 KW |
490 | ret = l2_load(bs, l2_offset, &l2_table); |
491 | if (ret < 0) { | |
492 | return ret; | |
1c46efaa | 493 | } |
45aba42f KW |
494 | |
495 | /* find the cluster offset for the given disk offset */ | |
496 | ||
497 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
1c46efaa | 498 | *cluster_offset = be64_to_cpu(l2_table[l2_index]); |
45aba42f KW |
499 | nb_clusters = size_to_clusters(s, nb_needed << 9); |
500 | ||
1c46efaa | 501 | if (!*cluster_offset) { |
45aba42f KW |
502 | /* how many empty clusters ? */ |
503 | c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]); | |
504 | } else { | |
505 | /* how many allocated clusters ? */ | |
506 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
507 | &l2_table[l2_index], 0, QCOW_OFLAG_COPIED); | |
508 | } | |
509 | ||
510 | nb_available = (c * s->cluster_sectors); | |
511 | out: | |
512 | if (nb_available > nb_needed) | |
513 | nb_available = nb_needed; | |
514 | ||
515 | *num = nb_available - index_in_cluster; | |
516 | ||
1c46efaa KW |
517 | *cluster_offset &=~QCOW_OFLAG_COPIED; |
518 | return 0; | |
45aba42f KW |
519 | } |
520 | ||
521 | /* | |
522 | * get_cluster_table | |
523 | * | |
524 | * for a given disk offset, load (and allocate if needed) | |
525 | * the l2 table. | |
526 | * | |
527 | * the l2 table offset in the qcow2 file and the cluster index | |
528 | * in the l2 table are given to the caller. | |
529 | * | |
1e3e8f1a | 530 | * Returns 0 on success, -errno in failure case |
45aba42f | 531 | */ |
45aba42f KW |
532 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
533 | uint64_t **new_l2_table, | |
534 | uint64_t *new_l2_offset, | |
535 | int *new_l2_index) | |
536 | { | |
537 | BDRVQcowState *s = bs->opaque; | |
80ee15a6 | 538 | unsigned int l1_index, l2_index; |
c46e1167 KW |
539 | uint64_t l2_offset; |
540 | uint64_t *l2_table = NULL; | |
80ee15a6 | 541 | int ret; |
45aba42f KW |
542 | |
543 | /* seek the the l2 offset in the l1 table */ | |
544 | ||
545 | l1_index = offset >> (s->l2_bits + s->cluster_bits); | |
546 | if (l1_index >= s->l1_size) { | |
ed6ccf0f | 547 | ret = qcow2_grow_l1_table(bs, l1_index + 1); |
1e3e8f1a KW |
548 | if (ret < 0) { |
549 | return ret; | |
550 | } | |
45aba42f KW |
551 | } |
552 | l2_offset = s->l1_table[l1_index]; | |
553 | ||
554 | /* seek the l2 table of the given l2 offset */ | |
555 | ||
556 | if (l2_offset & QCOW_OFLAG_COPIED) { | |
557 | /* load the l2 table in memory */ | |
558 | l2_offset &= ~QCOW_OFLAG_COPIED; | |
55c17e98 KW |
559 | ret = l2_load(bs, l2_offset, &l2_table); |
560 | if (ret < 0) { | |
561 | return ret; | |
1e3e8f1a | 562 | } |
45aba42f KW |
563 | } else { |
564 | if (l2_offset) | |
ed6ccf0f | 565 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t)); |
c46e1167 KW |
566 | ret = l2_allocate(bs, l1_index, &l2_table); |
567 | if (ret < 0) { | |
568 | return ret; | |
1e3e8f1a | 569 | } |
45aba42f KW |
570 | l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED; |
571 | } | |
572 | ||
573 | /* find the cluster offset for the given disk offset */ | |
574 | ||
575 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
576 | ||
577 | *new_l2_table = l2_table; | |
578 | *new_l2_offset = l2_offset; | |
579 | *new_l2_index = l2_index; | |
580 | ||
1e3e8f1a | 581 | return 0; |
45aba42f KW |
582 | } |
583 | ||
584 | /* | |
585 | * alloc_compressed_cluster_offset | |
586 | * | |
587 | * For a given offset of the disk image, return cluster offset in | |
588 | * qcow2 file. | |
589 | * | |
590 | * If the offset is not found, allocate a new compressed cluster. | |
591 | * | |
592 | * Return the cluster offset if successful, | |
593 | * Return 0, otherwise. | |
594 | * | |
595 | */ | |
596 | ||
ed6ccf0f KW |
597 | uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
598 | uint64_t offset, | |
599 | int compressed_size) | |
45aba42f KW |
600 | { |
601 | BDRVQcowState *s = bs->opaque; | |
602 | int l2_index, ret; | |
f4f0d391 KW |
603 | uint64_t l2_offset, *l2_table; |
604 | int64_t cluster_offset; | |
45aba42f KW |
605 | int nb_csectors; |
606 | ||
607 | ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index); | |
1e3e8f1a | 608 | if (ret < 0) { |
45aba42f | 609 | return 0; |
1e3e8f1a | 610 | } |
45aba42f KW |
611 | |
612 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
613 | if (cluster_offset & QCOW_OFLAG_COPIED) | |
614 | return cluster_offset & ~QCOW_OFLAG_COPIED; | |
615 | ||
616 | if (cluster_offset) | |
ed6ccf0f | 617 | qcow2_free_any_clusters(bs, cluster_offset, 1); |
45aba42f | 618 | |
ed6ccf0f | 619 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
5d757b56 KW |
620 | if (cluster_offset < 0) { |
621 | return 0; | |
622 | } | |
623 | ||
45aba42f KW |
624 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
625 | (cluster_offset >> 9); | |
626 | ||
627 | cluster_offset |= QCOW_OFLAG_COMPRESSED | | |
628 | ((uint64_t)nb_csectors << s->csize_shift); | |
629 | ||
630 | /* update L2 table */ | |
631 | ||
632 | /* compressed clusters never have the copied flag */ | |
633 | ||
66f82cee | 634 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
45aba42f | 635 | l2_table[l2_index] = cpu_to_be64(cluster_offset); |
8b3b7206 | 636 | if (bdrv_pwrite_sync(bs->file, |
45aba42f KW |
637 | l2_offset + l2_index * sizeof(uint64_t), |
638 | l2_table + l2_index, | |
8b3b7206 | 639 | sizeof(uint64_t)) < 0) |
45aba42f KW |
640 | return 0; |
641 | ||
642 | return cluster_offset; | |
643 | } | |
644 | ||
4c1612d9 KW |
645 | /* |
646 | * Write L2 table updates to disk, writing whole sectors to avoid a | |
647 | * read-modify-write in bdrv_pwrite | |
648 | */ | |
649 | #define L2_ENTRIES_PER_SECTOR (512 / 8) | |
66f82cee | 650 | static int write_l2_entries(BlockDriverState *bs, uint64_t *l2_table, |
4c1612d9 KW |
651 | uint64_t l2_offset, int l2_index, int num) |
652 | { | |
653 | int l2_start_index = l2_index & ~(L1_ENTRIES_PER_SECTOR - 1); | |
654 | int start_offset = (8 * l2_index) & ~511; | |
655 | int end_offset = (8 * (l2_index + num) + 511) & ~511; | |
656 | size_t len = end_offset - start_offset; | |
79a31189 | 657 | int ret; |
4c1612d9 | 658 | |
66f82cee | 659 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE); |
7ec5e6a4 | 660 | ret = bdrv_pwrite(bs->file, l2_offset + start_offset, |
79a31189 KW |
661 | &l2_table[l2_start_index], len); |
662 | if (ret < 0) { | |
663 | return ret; | |
4c1612d9 KW |
664 | } |
665 | ||
666 | return 0; | |
667 | } | |
668 | ||
148da7ea | 669 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) |
45aba42f KW |
670 | { |
671 | BDRVQcowState *s = bs->opaque; | |
672 | int i, j = 0, l2_index, ret; | |
673 | uint64_t *old_cluster, start_sect, l2_offset, *l2_table; | |
148da7ea | 674 | uint64_t cluster_offset = m->cluster_offset; |
45aba42f KW |
675 | |
676 | if (m->nb_clusters == 0) | |
677 | return 0; | |
678 | ||
679 | old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t)); | |
680 | ||
681 | /* copy content of unmodified sectors */ | |
682 | start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9; | |
683 | if (m->n_start) { | |
684 | ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start); | |
685 | if (ret < 0) | |
686 | goto err; | |
687 | } | |
688 | ||
689 | if (m->nb_available & (s->cluster_sectors - 1)) { | |
690 | uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1); | |
691 | ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9), | |
692 | m->nb_available - end, s->cluster_sectors); | |
693 | if (ret < 0) | |
694 | goto err; | |
695 | } | |
696 | ||
45aba42f | 697 | /* update L2 table */ |
1e3e8f1a KW |
698 | ret = get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index); |
699 | if (ret < 0) { | |
45aba42f | 700 | goto err; |
1e3e8f1a | 701 | } |
45aba42f KW |
702 | |
703 | for (i = 0; i < m->nb_clusters; i++) { | |
704 | /* if two concurrent writes happen to the same unallocated cluster | |
705 | * each write allocates separate cluster and writes data concurrently. | |
706 | * The first one to complete updates l2 table with pointer to its | |
707 | * cluster the second one has to do RMW (which is done above by | |
708 | * copy_sectors()), update l2 table with its cluster pointer and free | |
709 | * old cluster. This is what this loop does */ | |
710 | if(l2_table[l2_index + i] != 0) | |
711 | old_cluster[j++] = l2_table[l2_index + i]; | |
712 | ||
713 | l2_table[l2_index + i] = cpu_to_be64((cluster_offset + | |
714 | (i << s->cluster_bits)) | QCOW_OFLAG_COPIED); | |
715 | } | |
716 | ||
66f82cee | 717 | ret = write_l2_entries(bs, l2_table, l2_offset, l2_index, m->nb_clusters); |
c835d00f | 718 | if (ret < 0) { |
1b7c801b | 719 | qcow2_l2_cache_reset(bs); |
45aba42f | 720 | goto err; |
4c1612d9 | 721 | } |
45aba42f | 722 | |
7ec5e6a4 KW |
723 | /* |
724 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
725 | * Also flush bs->file to get the right order for L2 and refcount update. | |
726 | */ | |
727 | if (j != 0) { | |
728 | bdrv_flush(bs->file); | |
729 | for (i = 0; i < j; i++) { | |
730 | qcow2_free_any_clusters(bs, | |
731 | be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, 1); | |
732 | } | |
733 | } | |
45aba42f KW |
734 | |
735 | ret = 0; | |
736 | err: | |
737 | qemu_free(old_cluster); | |
738 | return ret; | |
739 | } | |
740 | ||
741 | /* | |
742 | * alloc_cluster_offset | |
743 | * | |
148da7ea | 744 | * For a given offset of the disk image, return cluster offset in qcow2 file. |
45aba42f KW |
745 | * If the offset is not found, allocate a new cluster. |
746 | * | |
148da7ea KW |
747 | * If the cluster was already allocated, m->nb_clusters is set to 0, |
748 | * m->depends_on is set to NULL and the other fields in m are meaningless. | |
749 | * | |
750 | * If the cluster is newly allocated, m->nb_clusters is set to the number of | |
751 | * contiguous clusters that have been allocated. This may be 0 if the request | |
752 | * conflict with another write request in flight; in this case, m->depends_on | |
753 | * is set and the remaining fields of m are meaningless. | |
45aba42f | 754 | * |
148da7ea KW |
755 | * If m->nb_clusters is non-zero, the other fields of m are valid and contain |
756 | * information about the first allocated cluster. | |
757 | * | |
758 | * Return 0 on success and -errno in error cases | |
45aba42f | 759 | */ |
f4f0d391 KW |
760 | int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, |
761 | int n_start, int n_end, int *num, QCowL2Meta *m) | |
45aba42f KW |
762 | { |
763 | BDRVQcowState *s = bs->opaque; | |
764 | int l2_index, ret; | |
5d757b56 KW |
765 | uint64_t l2_offset, *l2_table; |
766 | int64_t cluster_offset; | |
80ee15a6 | 767 | unsigned int nb_clusters, i = 0; |
f214978a | 768 | QCowL2Meta *old_alloc; |
45aba42f KW |
769 | |
770 | ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index); | |
1e3e8f1a | 771 | if (ret < 0) { |
148da7ea | 772 | return ret; |
1e3e8f1a | 773 | } |
45aba42f KW |
774 | |
775 | nb_clusters = size_to_clusters(s, n_end << 9); | |
776 | ||
777 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
778 | ||
779 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
780 | ||
781 | /* We keep all QCOW_OFLAG_COPIED clusters */ | |
782 | ||
783 | if (cluster_offset & QCOW_OFLAG_COPIED) { | |
784 | nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
785 | &l2_table[l2_index], 0, 0); | |
786 | ||
787 | cluster_offset &= ~QCOW_OFLAG_COPIED; | |
788 | m->nb_clusters = 0; | |
148da7ea | 789 | m->depends_on = NULL; |
45aba42f KW |
790 | |
791 | goto out; | |
792 | } | |
793 | ||
794 | /* for the moment, multiple compressed clusters are not managed */ | |
795 | ||
796 | if (cluster_offset & QCOW_OFLAG_COMPRESSED) | |
797 | nb_clusters = 1; | |
798 | ||
799 | /* how many available clusters ? */ | |
800 | ||
801 | while (i < nb_clusters) { | |
802 | i += count_contiguous_clusters(nb_clusters - i, s->cluster_size, | |
803 | &l2_table[l2_index], i, 0); | |
4805bb66 | 804 | if ((i >= nb_clusters) || be64_to_cpu(l2_table[l2_index + i])) { |
45aba42f | 805 | break; |
4805bb66 | 806 | } |
45aba42f KW |
807 | |
808 | i += count_contiguous_free_clusters(nb_clusters - i, | |
809 | &l2_table[l2_index + i]); | |
4805bb66 KW |
810 | if (i >= nb_clusters) { |
811 | break; | |
812 | } | |
45aba42f KW |
813 | |
814 | cluster_offset = be64_to_cpu(l2_table[l2_index + i]); | |
815 | ||
816 | if ((cluster_offset & QCOW_OFLAG_COPIED) || | |
817 | (cluster_offset & QCOW_OFLAG_COMPRESSED)) | |
818 | break; | |
819 | } | |
4805bb66 | 820 | assert(i <= nb_clusters); |
45aba42f KW |
821 | nb_clusters = i; |
822 | ||
f214978a KW |
823 | /* |
824 | * Check if there already is an AIO write request in flight which allocates | |
825 | * the same cluster. In this case we need to wait until the previous | |
826 | * request has completed and updated the L2 table accordingly. | |
827 | */ | |
72cf2d4f | 828 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
f214978a KW |
829 | |
830 | uint64_t end_offset = offset + nb_clusters * s->cluster_size; | |
831 | uint64_t old_offset = old_alloc->offset; | |
832 | uint64_t old_end_offset = old_alloc->offset + | |
833 | old_alloc->nb_clusters * s->cluster_size; | |
834 | ||
835 | if (end_offset < old_offset || offset > old_end_offset) { | |
836 | /* No intersection */ | |
837 | } else { | |
838 | if (offset < old_offset) { | |
839 | /* Stop at the start of a running allocation */ | |
840 | nb_clusters = (old_offset - offset) >> s->cluster_bits; | |
841 | } else { | |
842 | nb_clusters = 0; | |
843 | } | |
844 | ||
845 | if (nb_clusters == 0) { | |
846 | /* Set dependency and wait for a callback */ | |
847 | m->depends_on = old_alloc; | |
848 | m->nb_clusters = 0; | |
849 | *num = 0; | |
850 | return 0; | |
851 | } | |
852 | } | |
853 | } | |
854 | ||
855 | if (!nb_clusters) { | |
856 | abort(); | |
857 | } | |
858 | ||
72cf2d4f | 859 | QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight); |
f214978a | 860 | |
45aba42f KW |
861 | /* allocate a new cluster */ |
862 | ||
ed6ccf0f | 863 | cluster_offset = qcow2_alloc_clusters(bs, nb_clusters * s->cluster_size); |
5d757b56 | 864 | if (cluster_offset < 0) { |
c644db3d | 865 | QLIST_REMOVE(m, next_in_flight); |
5d757b56 KW |
866 | return cluster_offset; |
867 | } | |
29216ed1 | 868 | bdrv_flush(bs->file); |
45aba42f KW |
869 | |
870 | /* save info needed for meta data update */ | |
871 | m->offset = offset; | |
872 | m->n_start = n_start; | |
873 | m->nb_clusters = nb_clusters; | |
874 | ||
875 | out: | |
876 | m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end); | |
148da7ea | 877 | m->cluster_offset = cluster_offset; |
45aba42f KW |
878 | |
879 | *num = m->nb_available - n_start; | |
880 | ||
148da7ea | 881 | return 0; |
45aba42f KW |
882 | } |
883 | ||
884 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
885 | const uint8_t *buf, int buf_size) | |
886 | { | |
887 | z_stream strm1, *strm = &strm1; | |
888 | int ret, out_len; | |
889 | ||
890 | memset(strm, 0, sizeof(*strm)); | |
891 | ||
892 | strm->next_in = (uint8_t *)buf; | |
893 | strm->avail_in = buf_size; | |
894 | strm->next_out = out_buf; | |
895 | strm->avail_out = out_buf_size; | |
896 | ||
897 | ret = inflateInit2(strm, -12); | |
898 | if (ret != Z_OK) | |
899 | return -1; | |
900 | ret = inflate(strm, Z_FINISH); | |
901 | out_len = strm->next_out - out_buf; | |
902 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
903 | out_len != out_buf_size) { | |
904 | inflateEnd(strm); | |
905 | return -1; | |
906 | } | |
907 | inflateEnd(strm); | |
908 | return 0; | |
909 | } | |
910 | ||
66f82cee | 911 | int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) |
45aba42f | 912 | { |
66f82cee | 913 | BDRVQcowState *s = bs->opaque; |
45aba42f KW |
914 | int ret, csize, nb_csectors, sector_offset; |
915 | uint64_t coffset; | |
916 | ||
917 | coffset = cluster_offset & s->cluster_offset_mask; | |
918 | if (s->cluster_cache_offset != coffset) { | |
919 | nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1; | |
920 | sector_offset = coffset & 511; | |
921 | csize = nb_csectors * 512 - sector_offset; | |
66f82cee KW |
922 | BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
923 | ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, nb_csectors); | |
45aba42f KW |
924 | if (ret < 0) { |
925 | return -1; | |
926 | } | |
927 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
928 | s->cluster_data + sector_offset, csize) < 0) { | |
929 | return -1; | |
930 | } | |
931 | s->cluster_cache_offset = coffset; | |
932 | } | |
933 | return 0; | |
934 | } |