]>
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" | |
3cce16f4 | 30 | #include "trace.h" |
45aba42f | 31 | |
72893756 | 32 | int qcow2_grow_l1_table(BlockDriverState *bs, int min_size, bool exact_size) |
45aba42f KW |
33 | { |
34 | BDRVQcowState *s = bs->opaque; | |
35 | int new_l1_size, new_l1_size2, ret, i; | |
36 | uint64_t *new_l1_table; | |
5d757b56 | 37 | int64_t new_l1_table_offset; |
45aba42f KW |
38 | uint8_t data[12]; |
39 | ||
72893756 | 40 | if (min_size <= s->l1_size) |
45aba42f | 41 | return 0; |
72893756 SH |
42 | |
43 | if (exact_size) { | |
44 | new_l1_size = min_size; | |
45 | } else { | |
46 | /* Bump size up to reduce the number of times we have to grow */ | |
47 | new_l1_size = s->l1_size; | |
48 | if (new_l1_size == 0) { | |
49 | new_l1_size = 1; | |
50 | } | |
51 | while (min_size > new_l1_size) { | |
52 | new_l1_size = (new_l1_size * 3 + 1) / 2; | |
53 | } | |
45aba42f | 54 | } |
72893756 | 55 | |
45aba42f | 56 | #ifdef DEBUG_ALLOC2 |
35ee5e39 | 57 | fprintf(stderr, "grow l1_table from %d to %d\n", s->l1_size, new_l1_size); |
45aba42f KW |
58 | #endif |
59 | ||
60 | new_l1_size2 = sizeof(uint64_t) * new_l1_size; | |
7267c094 | 61 | new_l1_table = g_malloc0(align_offset(new_l1_size2, 512)); |
45aba42f KW |
62 | memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t)); |
63 | ||
64 | /* write new table (align to cluster) */ | |
66f82cee | 65 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE); |
ed6ccf0f | 66 | new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2); |
5d757b56 | 67 | if (new_l1_table_offset < 0) { |
7267c094 | 68 | g_free(new_l1_table); |
5d757b56 KW |
69 | return new_l1_table_offset; |
70 | } | |
29c1a730 KW |
71 | |
72 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
73 | if (ret < 0) { | |
80fa3341 | 74 | goto fail; |
29c1a730 | 75 | } |
45aba42f | 76 | |
66f82cee | 77 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE); |
45aba42f KW |
78 | for(i = 0; i < s->l1_size; i++) |
79 | new_l1_table[i] = cpu_to_be64(new_l1_table[i]); | |
8b3b7206 KW |
80 | ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, new_l1_table, new_l1_size2); |
81 | if (ret < 0) | |
45aba42f KW |
82 | goto fail; |
83 | for(i = 0; i < s->l1_size; i++) | |
84 | new_l1_table[i] = be64_to_cpu(new_l1_table[i]); | |
85 | ||
86 | /* set new table */ | |
66f82cee | 87 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE); |
45aba42f | 88 | cpu_to_be32w((uint32_t*)data, new_l1_size); |
653df36b | 89 | cpu_to_be64wu((uint64_t*)(data + 4), new_l1_table_offset); |
8b3b7206 KW |
90 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), data,sizeof(data)); |
91 | if (ret < 0) { | |
45aba42f | 92 | goto fail; |
fb8fa77c | 93 | } |
7267c094 | 94 | g_free(s->l1_table); |
ed6ccf0f | 95 | qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t)); |
45aba42f KW |
96 | s->l1_table_offset = new_l1_table_offset; |
97 | s->l1_table = new_l1_table; | |
98 | s->l1_size = new_l1_size; | |
99 | return 0; | |
100 | fail: | |
7267c094 | 101 | g_free(new_l1_table); |
fb8fa77c | 102 | qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2); |
8b3b7206 | 103 | return ret; |
45aba42f KW |
104 | } |
105 | ||
45aba42f KW |
106 | /* |
107 | * l2_load | |
108 | * | |
109 | * Loads a L2 table into memory. If the table is in the cache, the cache | |
110 | * is used; otherwise the L2 table is loaded from the image file. | |
111 | * | |
112 | * Returns a pointer to the L2 table on success, or NULL if the read from | |
113 | * the image file failed. | |
114 | */ | |
115 | ||
55c17e98 KW |
116 | static int l2_load(BlockDriverState *bs, uint64_t l2_offset, |
117 | uint64_t **l2_table) | |
45aba42f KW |
118 | { |
119 | BDRVQcowState *s = bs->opaque; | |
55c17e98 | 120 | int ret; |
45aba42f | 121 | |
29c1a730 | 122 | ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, (void**) l2_table); |
45aba42f | 123 | |
29c1a730 | 124 | return ret; |
45aba42f KW |
125 | } |
126 | ||
6583e3c7 KW |
127 | /* |
128 | * Writes one sector of the L1 table to the disk (can't update single entries | |
129 | * and we really don't want bdrv_pread to perform a read-modify-write) | |
130 | */ | |
131 | #define L1_ENTRIES_PER_SECTOR (512 / 8) | |
66f82cee | 132 | static int write_l1_entry(BlockDriverState *bs, int l1_index) |
6583e3c7 | 133 | { |
66f82cee | 134 | BDRVQcowState *s = bs->opaque; |
6583e3c7 KW |
135 | uint64_t buf[L1_ENTRIES_PER_SECTOR]; |
136 | int l1_start_index; | |
f7defcb6 | 137 | int i, ret; |
6583e3c7 KW |
138 | |
139 | l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1); | |
140 | for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) { | |
141 | buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]); | |
142 | } | |
143 | ||
66f82cee | 144 | BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); |
8b3b7206 | 145 | ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset + 8 * l1_start_index, |
f7defcb6 KW |
146 | buf, sizeof(buf)); |
147 | if (ret < 0) { | |
148 | return ret; | |
6583e3c7 KW |
149 | } |
150 | ||
151 | return 0; | |
152 | } | |
153 | ||
45aba42f KW |
154 | /* |
155 | * l2_allocate | |
156 | * | |
157 | * Allocate a new l2 entry in the file. If l1_index points to an already | |
158 | * used entry in the L2 table (i.e. we are doing a copy on write for the L2 | |
159 | * table) copy the contents of the old L2 table into the newly allocated one. | |
160 | * Otherwise the new table is initialized with zeros. | |
161 | * | |
162 | */ | |
163 | ||
c46e1167 | 164 | static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table) |
45aba42f KW |
165 | { |
166 | BDRVQcowState *s = bs->opaque; | |
6583e3c7 | 167 | uint64_t old_l2_offset; |
f4f0d391 KW |
168 | uint64_t *l2_table; |
169 | int64_t l2_offset; | |
c46e1167 | 170 | int ret; |
45aba42f KW |
171 | |
172 | old_l2_offset = s->l1_table[l1_index]; | |
173 | ||
3cce16f4 KW |
174 | trace_qcow2_l2_allocate(bs, l1_index); |
175 | ||
45aba42f KW |
176 | /* allocate a new l2 entry */ |
177 | ||
ed6ccf0f | 178 | l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t)); |
5d757b56 | 179 | if (l2_offset < 0) { |
c46e1167 | 180 | return l2_offset; |
5d757b56 | 181 | } |
29c1a730 KW |
182 | |
183 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
184 | if (ret < 0) { | |
185 | goto fail; | |
186 | } | |
45aba42f | 187 | |
45aba42f KW |
188 | /* allocate a new entry in the l2 cache */ |
189 | ||
3cce16f4 | 190 | trace_qcow2_l2_allocate_get_empty(bs, l1_index); |
29c1a730 KW |
191 | ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table); |
192 | if (ret < 0) { | |
193 | return ret; | |
194 | } | |
195 | ||
196 | l2_table = *table; | |
45aba42f KW |
197 | |
198 | if (old_l2_offset == 0) { | |
199 | /* if there was no old l2 table, clear the new table */ | |
200 | memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); | |
201 | } else { | |
29c1a730 KW |
202 | uint64_t* old_table; |
203 | ||
45aba42f | 204 | /* if there was an old l2 table, read it from the disk */ |
66f82cee | 205 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ); |
29c1a730 KW |
206 | ret = qcow2_cache_get(bs, s->l2_table_cache, old_l2_offset, |
207 | (void**) &old_table); | |
208 | if (ret < 0) { | |
209 | goto fail; | |
210 | } | |
211 | ||
212 | memcpy(l2_table, old_table, s->cluster_size); | |
213 | ||
214 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &old_table); | |
c46e1167 | 215 | if (ret < 0) { |
175e1152 | 216 | goto fail; |
c46e1167 | 217 | } |
45aba42f | 218 | } |
29c1a730 | 219 | |
45aba42f | 220 | /* write the l2 table to the file */ |
66f82cee | 221 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE); |
29c1a730 | 222 | |
3cce16f4 | 223 | trace_qcow2_l2_allocate_write_l2(bs, l1_index); |
29c1a730 KW |
224 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
225 | ret = qcow2_cache_flush(bs, s->l2_table_cache); | |
c46e1167 | 226 | if (ret < 0) { |
175e1152 KW |
227 | goto fail; |
228 | } | |
229 | ||
230 | /* update the L1 entry */ | |
3cce16f4 | 231 | trace_qcow2_l2_allocate_write_l1(bs, l1_index); |
175e1152 KW |
232 | s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; |
233 | ret = write_l1_entry(bs, l1_index); | |
234 | if (ret < 0) { | |
235 | goto fail; | |
c46e1167 | 236 | } |
45aba42f | 237 | |
c46e1167 | 238 | *table = l2_table; |
3cce16f4 | 239 | trace_qcow2_l2_allocate_done(bs, l1_index, 0); |
c46e1167 | 240 | return 0; |
175e1152 KW |
241 | |
242 | fail: | |
3cce16f4 | 243 | trace_qcow2_l2_allocate_done(bs, l1_index, ret); |
29c1a730 | 244 | qcow2_cache_put(bs, s->l2_table_cache, (void**) table); |
68dba0bf | 245 | s->l1_table[l1_index] = old_l2_offset; |
175e1152 | 246 | return ret; |
45aba42f KW |
247 | } |
248 | ||
249 | static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size, | |
250 | uint64_t *l2_table, uint64_t start, uint64_t mask) | |
251 | { | |
252 | int i; | |
253 | uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask; | |
254 | ||
255 | if (!offset) | |
256 | return 0; | |
257 | ||
258 | for (i = start; i < start + nb_clusters; i++) | |
80ee15a6 | 259 | if (offset + (uint64_t) i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask)) |
45aba42f KW |
260 | break; |
261 | ||
262 | return (i - start); | |
263 | } | |
264 | ||
265 | static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table) | |
266 | { | |
267 | int i = 0; | |
268 | ||
269 | while(nb_clusters-- && l2_table[i] == 0) | |
270 | i++; | |
271 | ||
272 | return i; | |
273 | } | |
274 | ||
275 | /* The crypt function is compatible with the linux cryptoloop | |
276 | algorithm for < 4 GB images. NOTE: out_buf == in_buf is | |
277 | supported */ | |
ed6ccf0f KW |
278 | void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num, |
279 | uint8_t *out_buf, const uint8_t *in_buf, | |
280 | int nb_sectors, int enc, | |
281 | const AES_KEY *key) | |
45aba42f KW |
282 | { |
283 | union { | |
284 | uint64_t ll[2]; | |
285 | uint8_t b[16]; | |
286 | } ivec; | |
287 | int i; | |
288 | ||
289 | for(i = 0; i < nb_sectors; i++) { | |
290 | ivec.ll[0] = cpu_to_le64(sector_num); | |
291 | ivec.ll[1] = 0; | |
292 | AES_cbc_encrypt(in_buf, out_buf, 512, key, | |
293 | ivec.b, enc); | |
294 | sector_num++; | |
295 | in_buf += 512; | |
296 | out_buf += 512; | |
297 | } | |
298 | } | |
299 | ||
aef4acb6 SH |
300 | static int coroutine_fn copy_sectors(BlockDriverState *bs, |
301 | uint64_t start_sect, | |
302 | uint64_t cluster_offset, | |
303 | int n_start, int n_end) | |
45aba42f KW |
304 | { |
305 | BDRVQcowState *s = bs->opaque; | |
aef4acb6 SH |
306 | QEMUIOVector qiov; |
307 | struct iovec iov; | |
45aba42f | 308 | int n, ret; |
1b9f1491 KW |
309 | |
310 | /* | |
311 | * If this is the last cluster and it is only partially used, we must only | |
312 | * copy until the end of the image, or bdrv_check_request will fail for the | |
313 | * bdrv_read/write calls below. | |
314 | */ | |
315 | if (start_sect + n_end > bs->total_sectors) { | |
316 | n_end = bs->total_sectors - start_sect; | |
317 | } | |
45aba42f KW |
318 | |
319 | n = n_end - n_start; | |
1b9f1491 | 320 | if (n <= 0) { |
45aba42f | 321 | return 0; |
1b9f1491 KW |
322 | } |
323 | ||
aef4acb6 SH |
324 | iov.iov_len = n * BDRV_SECTOR_SIZE; |
325 | iov.iov_base = qemu_blockalign(bs, iov.iov_len); | |
326 | ||
327 | qemu_iovec_init_external(&qiov, &iov, 1); | |
1b9f1491 | 328 | |
66f82cee | 329 | BLKDBG_EVENT(bs->file, BLKDBG_COW_READ); |
aef4acb6 SH |
330 | |
331 | /* Call .bdrv_co_readv() directly instead of using the public block-layer | |
332 | * interface. This avoids double I/O throttling and request tracking, | |
333 | * which can lead to deadlock when block layer copy-on-read is enabled. | |
334 | */ | |
335 | ret = bs->drv->bdrv_co_readv(bs, start_sect + n_start, n, &qiov); | |
1b9f1491 KW |
336 | if (ret < 0) { |
337 | goto out; | |
338 | } | |
339 | ||
45aba42f | 340 | if (s->crypt_method) { |
ed6ccf0f | 341 | qcow2_encrypt_sectors(s, start_sect + n_start, |
aef4acb6 | 342 | iov.iov_base, iov.iov_base, n, 1, |
45aba42f KW |
343 | &s->aes_encrypt_key); |
344 | } | |
1b9f1491 | 345 | |
66f82cee | 346 | BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE); |
aef4acb6 | 347 | ret = bdrv_co_writev(bs->file, (cluster_offset >> 9) + n_start, n, &qiov); |
1b9f1491 KW |
348 | if (ret < 0) { |
349 | goto out; | |
350 | } | |
351 | ||
352 | ret = 0; | |
353 | out: | |
aef4acb6 | 354 | qemu_vfree(iov.iov_base); |
1b9f1491 | 355 | return ret; |
45aba42f KW |
356 | } |
357 | ||
358 | ||
359 | /* | |
360 | * get_cluster_offset | |
361 | * | |
1c46efaa KW |
362 | * For a given offset of the disk image, find the cluster offset in |
363 | * qcow2 file. The offset is stored in *cluster_offset. | |
45aba42f | 364 | * |
d57237f2 | 365 | * on entry, *num is the number of contiguous sectors we'd like to |
45aba42f KW |
366 | * access following offset. |
367 | * | |
d57237f2 | 368 | * on exit, *num is the number of contiguous sectors we can read. |
45aba42f | 369 | * |
1c46efaa KW |
370 | * Return 0, if the offset is found |
371 | * Return -errno, otherwise. | |
45aba42f KW |
372 | * |
373 | */ | |
374 | ||
1c46efaa KW |
375 | int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, |
376 | int *num, uint64_t *cluster_offset) | |
45aba42f KW |
377 | { |
378 | BDRVQcowState *s = bs->opaque; | |
80ee15a6 | 379 | unsigned int l1_index, l2_index; |
1c46efaa | 380 | uint64_t l2_offset, *l2_table; |
45aba42f | 381 | int l1_bits, c; |
80ee15a6 KW |
382 | unsigned int index_in_cluster, nb_clusters; |
383 | uint64_t nb_available, nb_needed; | |
55c17e98 | 384 | int ret; |
45aba42f KW |
385 | |
386 | index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1); | |
387 | nb_needed = *num + index_in_cluster; | |
388 | ||
389 | l1_bits = s->l2_bits + s->cluster_bits; | |
390 | ||
391 | /* compute how many bytes there are between the offset and | |
392 | * the end of the l1 entry | |
393 | */ | |
394 | ||
80ee15a6 | 395 | nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1)); |
45aba42f KW |
396 | |
397 | /* compute the number of available sectors */ | |
398 | ||
399 | nb_available = (nb_available >> 9) + index_in_cluster; | |
400 | ||
401 | if (nb_needed > nb_available) { | |
402 | nb_needed = nb_available; | |
403 | } | |
404 | ||
1c46efaa | 405 | *cluster_offset = 0; |
45aba42f KW |
406 | |
407 | /* seek the the l2 offset in the l1 table */ | |
408 | ||
409 | l1_index = offset >> l1_bits; | |
410 | if (l1_index >= s->l1_size) | |
411 | goto out; | |
412 | ||
413 | l2_offset = s->l1_table[l1_index]; | |
414 | ||
415 | /* seek the l2 table of the given l2 offset */ | |
416 | ||
417 | if (!l2_offset) | |
418 | goto out; | |
419 | ||
420 | /* load the l2 table in memory */ | |
421 | ||
422 | l2_offset &= ~QCOW_OFLAG_COPIED; | |
55c17e98 KW |
423 | ret = l2_load(bs, l2_offset, &l2_table); |
424 | if (ret < 0) { | |
425 | return ret; | |
1c46efaa | 426 | } |
45aba42f KW |
427 | |
428 | /* find the cluster offset for the given disk offset */ | |
429 | ||
430 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
1c46efaa | 431 | *cluster_offset = be64_to_cpu(l2_table[l2_index]); |
45aba42f KW |
432 | nb_clusters = size_to_clusters(s, nb_needed << 9); |
433 | ||
1c46efaa | 434 | if (!*cluster_offset) { |
45aba42f KW |
435 | /* how many empty clusters ? */ |
436 | c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]); | |
437 | } else { | |
438 | /* how many allocated clusters ? */ | |
439 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
440 | &l2_table[l2_index], 0, QCOW_OFLAG_COPIED); | |
441 | } | |
442 | ||
29c1a730 KW |
443 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
444 | ||
45aba42f KW |
445 | nb_available = (c * s->cluster_sectors); |
446 | out: | |
447 | if (nb_available > nb_needed) | |
448 | nb_available = nb_needed; | |
449 | ||
450 | *num = nb_available - index_in_cluster; | |
451 | ||
1c46efaa KW |
452 | *cluster_offset &=~QCOW_OFLAG_COPIED; |
453 | return 0; | |
45aba42f KW |
454 | } |
455 | ||
456 | /* | |
457 | * get_cluster_table | |
458 | * | |
459 | * for a given disk offset, load (and allocate if needed) | |
460 | * the l2 table. | |
461 | * | |
462 | * the l2 table offset in the qcow2 file and the cluster index | |
463 | * in the l2 table are given to the caller. | |
464 | * | |
1e3e8f1a | 465 | * Returns 0 on success, -errno in failure case |
45aba42f | 466 | */ |
45aba42f KW |
467 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
468 | uint64_t **new_l2_table, | |
469 | uint64_t *new_l2_offset, | |
470 | int *new_l2_index) | |
471 | { | |
472 | BDRVQcowState *s = bs->opaque; | |
80ee15a6 | 473 | unsigned int l1_index, l2_index; |
c46e1167 KW |
474 | uint64_t l2_offset; |
475 | uint64_t *l2_table = NULL; | |
80ee15a6 | 476 | int ret; |
45aba42f KW |
477 | |
478 | /* seek the the l2 offset in the l1 table */ | |
479 | ||
480 | l1_index = offset >> (s->l2_bits + s->cluster_bits); | |
481 | if (l1_index >= s->l1_size) { | |
72893756 | 482 | ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
1e3e8f1a KW |
483 | if (ret < 0) { |
484 | return ret; | |
485 | } | |
45aba42f KW |
486 | } |
487 | l2_offset = s->l1_table[l1_index]; | |
488 | ||
489 | /* seek the l2 table of the given l2 offset */ | |
490 | ||
491 | if (l2_offset & QCOW_OFLAG_COPIED) { | |
492 | /* load the l2 table in memory */ | |
493 | l2_offset &= ~QCOW_OFLAG_COPIED; | |
55c17e98 KW |
494 | ret = l2_load(bs, l2_offset, &l2_table); |
495 | if (ret < 0) { | |
496 | return ret; | |
1e3e8f1a | 497 | } |
45aba42f | 498 | } else { |
16fde5f2 | 499 | /* First allocate a new L2 table (and do COW if needed) */ |
c46e1167 KW |
500 | ret = l2_allocate(bs, l1_index, &l2_table); |
501 | if (ret < 0) { | |
502 | return ret; | |
1e3e8f1a | 503 | } |
16fde5f2 KW |
504 | |
505 | /* Then decrease the refcount of the old table */ | |
506 | if (l2_offset) { | |
507 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t)); | |
508 | } | |
45aba42f KW |
509 | l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED; |
510 | } | |
511 | ||
512 | /* find the cluster offset for the given disk offset */ | |
513 | ||
514 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
515 | ||
516 | *new_l2_table = l2_table; | |
517 | *new_l2_offset = l2_offset; | |
518 | *new_l2_index = l2_index; | |
519 | ||
1e3e8f1a | 520 | return 0; |
45aba42f KW |
521 | } |
522 | ||
523 | /* | |
524 | * alloc_compressed_cluster_offset | |
525 | * | |
526 | * For a given offset of the disk image, return cluster offset in | |
527 | * qcow2 file. | |
528 | * | |
529 | * If the offset is not found, allocate a new compressed cluster. | |
530 | * | |
531 | * Return the cluster offset if successful, | |
532 | * Return 0, otherwise. | |
533 | * | |
534 | */ | |
535 | ||
ed6ccf0f KW |
536 | uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
537 | uint64_t offset, | |
538 | int compressed_size) | |
45aba42f KW |
539 | { |
540 | BDRVQcowState *s = bs->opaque; | |
541 | int l2_index, ret; | |
f4f0d391 KW |
542 | uint64_t l2_offset, *l2_table; |
543 | int64_t cluster_offset; | |
45aba42f KW |
544 | int nb_csectors; |
545 | ||
546 | ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index); | |
1e3e8f1a | 547 | if (ret < 0) { |
45aba42f | 548 | return 0; |
1e3e8f1a | 549 | } |
45aba42f KW |
550 | |
551 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
8f1efd00 KW |
552 | if (cluster_offset & QCOW_OFLAG_COPIED) { |
553 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); | |
554 | return 0; | |
555 | } | |
45aba42f KW |
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); |
5d757b56 | 561 | if (cluster_offset < 0) { |
29c1a730 | 562 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
5d757b56 KW |
563 | return 0; |
564 | } | |
565 | ||
45aba42f KW |
566 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
567 | (cluster_offset >> 9); | |
568 | ||
569 | cluster_offset |= QCOW_OFLAG_COMPRESSED | | |
570 | ((uint64_t)nb_csectors << s->csize_shift); | |
571 | ||
572 | /* update L2 table */ | |
573 | ||
574 | /* compressed clusters never have the copied flag */ | |
575 | ||
66f82cee | 576 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
29c1a730 | 577 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
45aba42f | 578 | l2_table[l2_index] = cpu_to_be64(cluster_offset); |
29c1a730 | 579 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
79a31189 | 580 | if (ret < 0) { |
29c1a730 | 581 | return 0; |
4c1612d9 KW |
582 | } |
583 | ||
29c1a730 | 584 | return cluster_offset; |
4c1612d9 KW |
585 | } |
586 | ||
148da7ea | 587 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) |
45aba42f KW |
588 | { |
589 | BDRVQcowState *s = bs->opaque; | |
590 | int i, j = 0, l2_index, ret; | |
591 | uint64_t *old_cluster, start_sect, l2_offset, *l2_table; | |
148da7ea | 592 | uint64_t cluster_offset = m->cluster_offset; |
29c1a730 | 593 | bool cow = false; |
45aba42f | 594 | |
3cce16f4 KW |
595 | trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
596 | ||
45aba42f KW |
597 | if (m->nb_clusters == 0) |
598 | return 0; | |
599 | ||
7267c094 | 600 | old_cluster = g_malloc(m->nb_clusters * sizeof(uint64_t)); |
45aba42f KW |
601 | |
602 | /* copy content of unmodified sectors */ | |
603 | start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9; | |
604 | if (m->n_start) { | |
29c1a730 | 605 | cow = true; |
1b9f1491 | 606 | qemu_co_mutex_unlock(&s->lock); |
45aba42f | 607 | ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start); |
1b9f1491 | 608 | qemu_co_mutex_lock(&s->lock); |
45aba42f KW |
609 | if (ret < 0) |
610 | goto err; | |
611 | } | |
612 | ||
613 | if (m->nb_available & (s->cluster_sectors - 1)) { | |
614 | uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1); | |
29c1a730 | 615 | cow = true; |
1b9f1491 | 616 | qemu_co_mutex_unlock(&s->lock); |
45aba42f KW |
617 | ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9), |
618 | m->nb_available - end, s->cluster_sectors); | |
1b9f1491 | 619 | qemu_co_mutex_lock(&s->lock); |
45aba42f KW |
620 | if (ret < 0) |
621 | goto err; | |
622 | } | |
623 | ||
29c1a730 KW |
624 | /* |
625 | * Update L2 table. | |
626 | * | |
627 | * Before we update the L2 table to actually point to the new cluster, we | |
628 | * need to be sure that the refcounts have been increased and COW was | |
629 | * handled. | |
630 | */ | |
631 | if (cow) { | |
3de0a294 | 632 | qcow2_cache_depends_on_flush(s->l2_table_cache); |
29c1a730 KW |
633 | } |
634 | ||
635 | qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache); | |
1e3e8f1a KW |
636 | ret = get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index); |
637 | if (ret < 0) { | |
45aba42f | 638 | goto err; |
1e3e8f1a | 639 | } |
29c1a730 | 640 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
45aba42f KW |
641 | |
642 | for (i = 0; i < m->nb_clusters; i++) { | |
643 | /* if two concurrent writes happen to the same unallocated cluster | |
644 | * each write allocates separate cluster and writes data concurrently. | |
645 | * The first one to complete updates l2 table with pointer to its | |
646 | * cluster the second one has to do RMW (which is done above by | |
647 | * copy_sectors()), update l2 table with its cluster pointer and free | |
648 | * old cluster. This is what this loop does */ | |
649 | if(l2_table[l2_index + i] != 0) | |
650 | old_cluster[j++] = l2_table[l2_index + i]; | |
651 | ||
652 | l2_table[l2_index + i] = cpu_to_be64((cluster_offset + | |
653 | (i << s->cluster_bits)) | QCOW_OFLAG_COPIED); | |
654 | } | |
655 | ||
9f8e668e | 656 | |
29c1a730 | 657 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
c835d00f | 658 | if (ret < 0) { |
45aba42f | 659 | goto err; |
4c1612d9 | 660 | } |
45aba42f | 661 | |
7ec5e6a4 KW |
662 | /* |
663 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
664 | * Also flush bs->file to get the right order for L2 and refcount update. | |
665 | */ | |
666 | if (j != 0) { | |
7ec5e6a4 KW |
667 | for (i = 0; i < j; i++) { |
668 | qcow2_free_any_clusters(bs, | |
669 | be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, 1); | |
670 | } | |
671 | } | |
45aba42f KW |
672 | |
673 | ret = 0; | |
674 | err: | |
7267c094 | 675 | g_free(old_cluster); |
45aba42f KW |
676 | return ret; |
677 | } | |
678 | ||
679 | /* | |
680 | * alloc_cluster_offset | |
681 | * | |
148da7ea | 682 | * For a given offset of the disk image, return cluster offset in qcow2 file. |
45aba42f KW |
683 | * If the offset is not found, allocate a new cluster. |
684 | * | |
148da7ea | 685 | * If the cluster was already allocated, m->nb_clusters is set to 0, |
a7912369 | 686 | * other fields in m are meaningless. |
148da7ea KW |
687 | * |
688 | * If the cluster is newly allocated, m->nb_clusters is set to the number of | |
68d100e9 KW |
689 | * contiguous clusters that have been allocated. In this case, the other |
690 | * fields of m are valid and contain information about the first allocated | |
691 | * cluster. | |
45aba42f | 692 | * |
68d100e9 KW |
693 | * If the request conflicts with another write request in flight, the coroutine |
694 | * is queued and will be reentered when the dependency has completed. | |
148da7ea KW |
695 | * |
696 | * Return 0 on success and -errno in error cases | |
45aba42f | 697 | */ |
f4f0d391 KW |
698 | int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, |
699 | int n_start, int n_end, int *num, QCowL2Meta *m) | |
45aba42f KW |
700 | { |
701 | BDRVQcowState *s = bs->opaque; | |
702 | int l2_index, ret; | |
5d757b56 KW |
703 | uint64_t l2_offset, *l2_table; |
704 | int64_t cluster_offset; | |
80ee15a6 | 705 | unsigned int nb_clusters, i = 0; |
f214978a | 706 | QCowL2Meta *old_alloc; |
45aba42f | 707 | |
3cce16f4 KW |
708 | trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, |
709 | n_start, n_end); | |
710 | ||
45aba42f | 711 | ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index); |
1e3e8f1a | 712 | if (ret < 0) { |
148da7ea | 713 | return ret; |
1e3e8f1a | 714 | } |
45aba42f | 715 | |
68d100e9 | 716 | again: |
45aba42f KW |
717 | nb_clusters = size_to_clusters(s, n_end << 9); |
718 | ||
719 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
720 | ||
721 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
722 | ||
723 | /* We keep all QCOW_OFLAG_COPIED clusters */ | |
724 | ||
725 | if (cluster_offset & QCOW_OFLAG_COPIED) { | |
726 | nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
727 | &l2_table[l2_index], 0, 0); | |
728 | ||
729 | cluster_offset &= ~QCOW_OFLAG_COPIED; | |
730 | m->nb_clusters = 0; | |
731 | ||
732 | goto out; | |
733 | } | |
734 | ||
735 | /* for the moment, multiple compressed clusters are not managed */ | |
736 | ||
737 | if (cluster_offset & QCOW_OFLAG_COMPRESSED) | |
738 | nb_clusters = 1; | |
739 | ||
740 | /* how many available clusters ? */ | |
741 | ||
742 | while (i < nb_clusters) { | |
743 | i += count_contiguous_clusters(nb_clusters - i, s->cluster_size, | |
744 | &l2_table[l2_index], i, 0); | |
4805bb66 | 745 | if ((i >= nb_clusters) || be64_to_cpu(l2_table[l2_index + i])) { |
45aba42f | 746 | break; |
4805bb66 | 747 | } |
45aba42f KW |
748 | |
749 | i += count_contiguous_free_clusters(nb_clusters - i, | |
750 | &l2_table[l2_index + i]); | |
4805bb66 KW |
751 | if (i >= nb_clusters) { |
752 | break; | |
753 | } | |
45aba42f KW |
754 | |
755 | cluster_offset = be64_to_cpu(l2_table[l2_index + i]); | |
756 | ||
757 | if ((cluster_offset & QCOW_OFLAG_COPIED) || | |
758 | (cluster_offset & QCOW_OFLAG_COMPRESSED)) | |
759 | break; | |
760 | } | |
4805bb66 | 761 | assert(i <= nb_clusters); |
45aba42f KW |
762 | nb_clusters = i; |
763 | ||
f214978a KW |
764 | /* |
765 | * Check if there already is an AIO write request in flight which allocates | |
766 | * the same cluster. In this case we need to wait until the previous | |
767 | * request has completed and updated the L2 table accordingly. | |
768 | */ | |
72cf2d4f | 769 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
f214978a | 770 | |
ee18e730 FZ |
771 | uint64_t start = offset >> s->cluster_bits; |
772 | uint64_t end = start + nb_clusters; | |
773 | uint64_t old_start = old_alloc->offset >> s->cluster_bits; | |
774 | uint64_t old_end = old_start + old_alloc->nb_clusters; | |
f214978a | 775 | |
ee18e730 | 776 | if (end < old_start || start > old_end) { |
f214978a KW |
777 | /* No intersection */ |
778 | } else { | |
ee18e730 | 779 | if (start < old_start) { |
f214978a | 780 | /* Stop at the start of a running allocation */ |
ee18e730 | 781 | nb_clusters = old_start - start; |
f214978a KW |
782 | } else { |
783 | nb_clusters = 0; | |
784 | } | |
785 | ||
786 | if (nb_clusters == 0) { | |
68d100e9 KW |
787 | /* Wait for the dependency to complete. We need to recheck |
788 | * the free/allocated clusters when we continue. */ | |
789 | qemu_co_mutex_unlock(&s->lock); | |
790 | qemu_co_queue_wait(&old_alloc->dependent_requests); | |
791 | qemu_co_mutex_lock(&s->lock); | |
792 | goto again; | |
f214978a KW |
793 | } |
794 | } | |
795 | } | |
796 | ||
797 | if (!nb_clusters) { | |
798 | abort(); | |
799 | } | |
800 | ||
05140499 FZ |
801 | /* save info needed for meta data update */ |
802 | m->offset = offset; | |
803 | m->n_start = n_start; | |
804 | m->nb_clusters = nb_clusters; | |
805 | ||
72cf2d4f | 806 | QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight); |
f214978a | 807 | |
45aba42f | 808 | /* allocate a new cluster */ |
3cce16f4 | 809 | trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); |
ed6ccf0f | 810 | cluster_offset = qcow2_alloc_clusters(bs, nb_clusters * s->cluster_size); |
5d757b56 | 811 | if (cluster_offset < 0) { |
29c1a730 KW |
812 | ret = cluster_offset; |
813 | goto fail; | |
5d757b56 | 814 | } |
45aba42f | 815 | |
45aba42f | 816 | out: |
29c1a730 KW |
817 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
818 | if (ret < 0) { | |
9e2a3701 | 819 | goto fail_put; |
29c1a730 KW |
820 | } |
821 | ||
45aba42f | 822 | m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end); |
148da7ea | 823 | m->cluster_offset = cluster_offset; |
45aba42f KW |
824 | |
825 | *num = m->nb_available - n_start; | |
826 | ||
148da7ea | 827 | return 0; |
29c1a730 KW |
828 | |
829 | fail: | |
830 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); | |
9e2a3701 KW |
831 | fail_put: |
832 | QLIST_REMOVE(m, next_in_flight); | |
29c1a730 | 833 | return ret; |
45aba42f KW |
834 | } |
835 | ||
836 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
837 | const uint8_t *buf, int buf_size) | |
838 | { | |
839 | z_stream strm1, *strm = &strm1; | |
840 | int ret, out_len; | |
841 | ||
842 | memset(strm, 0, sizeof(*strm)); | |
843 | ||
844 | strm->next_in = (uint8_t *)buf; | |
845 | strm->avail_in = buf_size; | |
846 | strm->next_out = out_buf; | |
847 | strm->avail_out = out_buf_size; | |
848 | ||
849 | ret = inflateInit2(strm, -12); | |
850 | if (ret != Z_OK) | |
851 | return -1; | |
852 | ret = inflate(strm, Z_FINISH); | |
853 | out_len = strm->next_out - out_buf; | |
854 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
855 | out_len != out_buf_size) { | |
856 | inflateEnd(strm); | |
857 | return -1; | |
858 | } | |
859 | inflateEnd(strm); | |
860 | return 0; | |
861 | } | |
862 | ||
66f82cee | 863 | int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) |
45aba42f | 864 | { |
66f82cee | 865 | BDRVQcowState *s = bs->opaque; |
45aba42f KW |
866 | int ret, csize, nb_csectors, sector_offset; |
867 | uint64_t coffset; | |
868 | ||
869 | coffset = cluster_offset & s->cluster_offset_mask; | |
870 | if (s->cluster_cache_offset != coffset) { | |
871 | nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1; | |
872 | sector_offset = coffset & 511; | |
873 | csize = nb_csectors * 512 - sector_offset; | |
66f82cee KW |
874 | BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
875 | ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, nb_csectors); | |
45aba42f | 876 | if (ret < 0) { |
8af36488 | 877 | return ret; |
45aba42f KW |
878 | } |
879 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
880 | s->cluster_data + sector_offset, csize) < 0) { | |
8af36488 | 881 | return -EIO; |
45aba42f KW |
882 | } |
883 | s->cluster_cache_offset = coffset; | |
884 | } | |
885 | return 0; | |
886 | } | |
5ea929e3 KW |
887 | |
888 | /* | |
889 | * This discards as many clusters of nb_clusters as possible at once (i.e. | |
890 | * all clusters in the same L2 table) and returns the number of discarded | |
891 | * clusters. | |
892 | */ | |
893 | static int discard_single_l2(BlockDriverState *bs, uint64_t offset, | |
894 | unsigned int nb_clusters) | |
895 | { | |
896 | BDRVQcowState *s = bs->opaque; | |
897 | uint64_t l2_offset, *l2_table; | |
898 | int l2_index; | |
899 | int ret; | |
900 | int i; | |
901 | ||
902 | ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index); | |
903 | if (ret < 0) { | |
904 | return ret; | |
905 | } | |
906 | ||
907 | /* Limit nb_clusters to one L2 table */ | |
908 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
909 | ||
910 | for (i = 0; i < nb_clusters; i++) { | |
911 | uint64_t old_offset; | |
912 | ||
913 | old_offset = be64_to_cpu(l2_table[l2_index + i]); | |
914 | old_offset &= ~QCOW_OFLAG_COPIED; | |
915 | ||
916 | if (old_offset == 0) { | |
917 | continue; | |
918 | } | |
919 | ||
920 | /* First remove L2 entries */ | |
921 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); | |
922 | l2_table[l2_index + i] = cpu_to_be64(0); | |
923 | ||
924 | /* Then decrease the refcount */ | |
925 | qcow2_free_any_clusters(bs, old_offset, 1); | |
926 | } | |
927 | ||
928 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); | |
929 | if (ret < 0) { | |
930 | return ret; | |
931 | } | |
932 | ||
933 | return nb_clusters; | |
934 | } | |
935 | ||
936 | int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset, | |
937 | int nb_sectors) | |
938 | { | |
939 | BDRVQcowState *s = bs->opaque; | |
940 | uint64_t end_offset; | |
941 | unsigned int nb_clusters; | |
942 | int ret; | |
943 | ||
944 | end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS); | |
945 | ||
946 | /* Round start up and end down */ | |
947 | offset = align_offset(offset, s->cluster_size); | |
948 | end_offset &= ~(s->cluster_size - 1); | |
949 | ||
950 | if (offset > end_offset) { | |
951 | return 0; | |
952 | } | |
953 | ||
954 | nb_clusters = size_to_clusters(s, end_offset - offset); | |
955 | ||
956 | /* Each L2 table is handled by its own loop iteration */ | |
957 | while (nb_clusters > 0) { | |
958 | ret = discard_single_l2(bs, offset, nb_clusters); | |
959 | if (ret < 0) { | |
960 | return ret; | |
961 | } | |
962 | ||
963 | nb_clusters -= ret; | |
964 | offset += (ret * s->cluster_size); | |
965 | } | |
966 | ||
967 | return 0; | |
968 | } |