]>
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 | 197 | |
8e37f681 | 198 | if ((old_l2_offset & L1E_OFFSET_MASK) == 0) { |
45aba42f KW |
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); |
8e37f681 KW |
206 | ret = qcow2_cache_get(bs, s->l2_table_cache, |
207 | old_l2_offset & L1E_OFFSET_MASK, | |
29c1a730 KW |
208 | (void**) &old_table); |
209 | if (ret < 0) { | |
210 | goto fail; | |
211 | } | |
212 | ||
213 | memcpy(l2_table, old_table, s->cluster_size); | |
214 | ||
215 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &old_table); | |
c46e1167 | 216 | if (ret < 0) { |
175e1152 | 217 | goto fail; |
c46e1167 | 218 | } |
45aba42f | 219 | } |
29c1a730 | 220 | |
45aba42f | 221 | /* write the l2 table to the file */ |
66f82cee | 222 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE); |
29c1a730 | 223 | |
3cce16f4 | 224 | trace_qcow2_l2_allocate_write_l2(bs, l1_index); |
29c1a730 KW |
225 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
226 | ret = qcow2_cache_flush(bs, s->l2_table_cache); | |
c46e1167 | 227 | if (ret < 0) { |
175e1152 KW |
228 | goto fail; |
229 | } | |
230 | ||
231 | /* update the L1 entry */ | |
3cce16f4 | 232 | trace_qcow2_l2_allocate_write_l1(bs, l1_index); |
175e1152 KW |
233 | s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; |
234 | ret = write_l1_entry(bs, l1_index); | |
235 | if (ret < 0) { | |
236 | goto fail; | |
c46e1167 | 237 | } |
45aba42f | 238 | |
c46e1167 | 239 | *table = l2_table; |
3cce16f4 | 240 | trace_qcow2_l2_allocate_done(bs, l1_index, 0); |
c46e1167 | 241 | return 0; |
175e1152 KW |
242 | |
243 | fail: | |
3cce16f4 | 244 | trace_qcow2_l2_allocate_done(bs, l1_index, ret); |
29c1a730 | 245 | qcow2_cache_put(bs, s->l2_table_cache, (void**) table); |
68dba0bf | 246 | s->l1_table[l1_index] = old_l2_offset; |
175e1152 | 247 | return ret; |
45aba42f KW |
248 | } |
249 | ||
2bfcc4a0 KW |
250 | /* |
251 | * Checks how many clusters in a given L2 table are contiguous in the image | |
252 | * file. As soon as one of the flags in the bitmask stop_flags changes compared | |
253 | * to the first cluster, the search is stopped and the cluster is not counted | |
254 | * as contiguous. (This allows it, for example, to stop at the first compressed | |
255 | * cluster which may require a different handling) | |
256 | */ | |
45aba42f | 257 | static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size, |
2bfcc4a0 | 258 | uint64_t *l2_table, uint64_t start, uint64_t stop_flags) |
45aba42f KW |
259 | { |
260 | int i; | |
2bfcc4a0 KW |
261 | uint64_t mask = stop_flags | L2E_OFFSET_MASK; |
262 | uint64_t offset = be64_to_cpu(l2_table[0]) & mask; | |
45aba42f KW |
263 | |
264 | if (!offset) | |
265 | return 0; | |
266 | ||
2bfcc4a0 KW |
267 | for (i = start; i < start + nb_clusters; i++) { |
268 | uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask; | |
269 | if (offset + (uint64_t) i * cluster_size != l2_entry) { | |
45aba42f | 270 | break; |
2bfcc4a0 KW |
271 | } |
272 | } | |
45aba42f KW |
273 | |
274 | return (i - start); | |
275 | } | |
276 | ||
277 | static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table) | |
278 | { | |
2bfcc4a0 KW |
279 | int i; |
280 | ||
281 | for (i = 0; i < nb_clusters; i++) { | |
282 | int type = qcow2_get_cluster_type(be64_to_cpu(l2_table[i])); | |
45aba42f | 283 | |
2bfcc4a0 KW |
284 | if (type != QCOW2_CLUSTER_UNALLOCATED) { |
285 | break; | |
286 | } | |
287 | } | |
45aba42f KW |
288 | |
289 | return i; | |
290 | } | |
291 | ||
292 | /* The crypt function is compatible with the linux cryptoloop | |
293 | algorithm for < 4 GB images. NOTE: out_buf == in_buf is | |
294 | supported */ | |
ed6ccf0f KW |
295 | void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num, |
296 | uint8_t *out_buf, const uint8_t *in_buf, | |
297 | int nb_sectors, int enc, | |
298 | const AES_KEY *key) | |
45aba42f KW |
299 | { |
300 | union { | |
301 | uint64_t ll[2]; | |
302 | uint8_t b[16]; | |
303 | } ivec; | |
304 | int i; | |
305 | ||
306 | for(i = 0; i < nb_sectors; i++) { | |
307 | ivec.ll[0] = cpu_to_le64(sector_num); | |
308 | ivec.ll[1] = 0; | |
309 | AES_cbc_encrypt(in_buf, out_buf, 512, key, | |
310 | ivec.b, enc); | |
311 | sector_num++; | |
312 | in_buf += 512; | |
313 | out_buf += 512; | |
314 | } | |
315 | } | |
316 | ||
aef4acb6 SH |
317 | static int coroutine_fn copy_sectors(BlockDriverState *bs, |
318 | uint64_t start_sect, | |
319 | uint64_t cluster_offset, | |
320 | int n_start, int n_end) | |
45aba42f KW |
321 | { |
322 | BDRVQcowState *s = bs->opaque; | |
aef4acb6 SH |
323 | QEMUIOVector qiov; |
324 | struct iovec iov; | |
45aba42f | 325 | int n, ret; |
1b9f1491 KW |
326 | |
327 | /* | |
328 | * If this is the last cluster and it is only partially used, we must only | |
329 | * copy until the end of the image, or bdrv_check_request will fail for the | |
330 | * bdrv_read/write calls below. | |
331 | */ | |
332 | if (start_sect + n_end > bs->total_sectors) { | |
333 | n_end = bs->total_sectors - start_sect; | |
334 | } | |
45aba42f KW |
335 | |
336 | n = n_end - n_start; | |
1b9f1491 | 337 | if (n <= 0) { |
45aba42f | 338 | return 0; |
1b9f1491 KW |
339 | } |
340 | ||
aef4acb6 SH |
341 | iov.iov_len = n * BDRV_SECTOR_SIZE; |
342 | iov.iov_base = qemu_blockalign(bs, iov.iov_len); | |
343 | ||
344 | qemu_iovec_init_external(&qiov, &iov, 1); | |
1b9f1491 | 345 | |
66f82cee | 346 | BLKDBG_EVENT(bs->file, BLKDBG_COW_READ); |
aef4acb6 SH |
347 | |
348 | /* Call .bdrv_co_readv() directly instead of using the public block-layer | |
349 | * interface. This avoids double I/O throttling and request tracking, | |
350 | * which can lead to deadlock when block layer copy-on-read is enabled. | |
351 | */ | |
352 | ret = bs->drv->bdrv_co_readv(bs, start_sect + n_start, n, &qiov); | |
1b9f1491 KW |
353 | if (ret < 0) { |
354 | goto out; | |
355 | } | |
356 | ||
45aba42f | 357 | if (s->crypt_method) { |
ed6ccf0f | 358 | qcow2_encrypt_sectors(s, start_sect + n_start, |
aef4acb6 | 359 | iov.iov_base, iov.iov_base, n, 1, |
45aba42f KW |
360 | &s->aes_encrypt_key); |
361 | } | |
1b9f1491 | 362 | |
66f82cee | 363 | BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE); |
aef4acb6 | 364 | ret = bdrv_co_writev(bs->file, (cluster_offset >> 9) + n_start, n, &qiov); |
1b9f1491 KW |
365 | if (ret < 0) { |
366 | goto out; | |
367 | } | |
368 | ||
369 | ret = 0; | |
370 | out: | |
aef4acb6 | 371 | qemu_vfree(iov.iov_base); |
1b9f1491 | 372 | return ret; |
45aba42f KW |
373 | } |
374 | ||
375 | ||
376 | /* | |
377 | * get_cluster_offset | |
378 | * | |
1c46efaa KW |
379 | * For a given offset of the disk image, find the cluster offset in |
380 | * qcow2 file. The offset is stored in *cluster_offset. | |
45aba42f | 381 | * |
d57237f2 | 382 | * on entry, *num is the number of contiguous sectors we'd like to |
45aba42f KW |
383 | * access following offset. |
384 | * | |
d57237f2 | 385 | * on exit, *num is the number of contiguous sectors we can read. |
45aba42f | 386 | * |
68d000a3 KW |
387 | * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error |
388 | * cases. | |
45aba42f | 389 | */ |
1c46efaa KW |
390 | int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, |
391 | int *num, uint64_t *cluster_offset) | |
45aba42f KW |
392 | { |
393 | BDRVQcowState *s = bs->opaque; | |
80ee15a6 | 394 | unsigned int l1_index, l2_index; |
1c46efaa | 395 | uint64_t l2_offset, *l2_table; |
45aba42f | 396 | int l1_bits, c; |
80ee15a6 KW |
397 | unsigned int index_in_cluster, nb_clusters; |
398 | uint64_t nb_available, nb_needed; | |
55c17e98 | 399 | int ret; |
45aba42f KW |
400 | |
401 | index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1); | |
402 | nb_needed = *num + index_in_cluster; | |
403 | ||
404 | l1_bits = s->l2_bits + s->cluster_bits; | |
405 | ||
406 | /* compute how many bytes there are between the offset and | |
407 | * the end of the l1 entry | |
408 | */ | |
409 | ||
80ee15a6 | 410 | nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1)); |
45aba42f KW |
411 | |
412 | /* compute the number of available sectors */ | |
413 | ||
414 | nb_available = (nb_available >> 9) + index_in_cluster; | |
415 | ||
416 | if (nb_needed > nb_available) { | |
417 | nb_needed = nb_available; | |
418 | } | |
419 | ||
1c46efaa | 420 | *cluster_offset = 0; |
45aba42f KW |
421 | |
422 | /* seek the the l2 offset in the l1 table */ | |
423 | ||
424 | l1_index = offset >> l1_bits; | |
68d000a3 KW |
425 | if (l1_index >= s->l1_size) { |
426 | ret = QCOW2_CLUSTER_UNALLOCATED; | |
45aba42f | 427 | goto out; |
68d000a3 | 428 | } |
45aba42f | 429 | |
68d000a3 KW |
430 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
431 | if (!l2_offset) { | |
432 | ret = QCOW2_CLUSTER_UNALLOCATED; | |
45aba42f | 433 | goto out; |
68d000a3 | 434 | } |
45aba42f KW |
435 | |
436 | /* load the l2 table in memory */ | |
437 | ||
55c17e98 KW |
438 | ret = l2_load(bs, l2_offset, &l2_table); |
439 | if (ret < 0) { | |
440 | return ret; | |
1c46efaa | 441 | } |
45aba42f KW |
442 | |
443 | /* find the cluster offset for the given disk offset */ | |
444 | ||
445 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
1c46efaa | 446 | *cluster_offset = be64_to_cpu(l2_table[l2_index]); |
45aba42f KW |
447 | nb_clusters = size_to_clusters(s, nb_needed << 9); |
448 | ||
68d000a3 KW |
449 | ret = qcow2_get_cluster_type(*cluster_offset); |
450 | switch (ret) { | |
451 | case QCOW2_CLUSTER_COMPRESSED: | |
452 | /* Compressed clusters can only be processed one by one */ | |
453 | c = 1; | |
454 | *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK; | |
455 | break; | |
6377af48 KW |
456 | case QCOW2_CLUSTER_ZERO: |
457 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
458 | &l2_table[l2_index], 0, | |
459 | QCOW_OFLAG_COMPRESSED | QCOW_OFLAG_ZERO); | |
460 | *cluster_offset = 0; | |
461 | break; | |
68d000a3 | 462 | case QCOW2_CLUSTER_UNALLOCATED: |
45aba42f KW |
463 | /* how many empty clusters ? */ |
464 | c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]); | |
68d000a3 KW |
465 | *cluster_offset = 0; |
466 | break; | |
467 | case QCOW2_CLUSTER_NORMAL: | |
45aba42f KW |
468 | /* how many allocated clusters ? */ |
469 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
6377af48 KW |
470 | &l2_table[l2_index], 0, |
471 | QCOW_OFLAG_COMPRESSED | QCOW_OFLAG_ZERO); | |
68d000a3 KW |
472 | *cluster_offset &= L2E_OFFSET_MASK; |
473 | break; | |
1417d7e4 KW |
474 | default: |
475 | abort(); | |
45aba42f KW |
476 | } |
477 | ||
29c1a730 KW |
478 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
479 | ||
68d000a3 KW |
480 | nb_available = (c * s->cluster_sectors); |
481 | ||
45aba42f KW |
482 | out: |
483 | if (nb_available > nb_needed) | |
484 | nb_available = nb_needed; | |
485 | ||
486 | *num = nb_available - index_in_cluster; | |
487 | ||
68d000a3 | 488 | return ret; |
45aba42f KW |
489 | } |
490 | ||
491 | /* | |
492 | * get_cluster_table | |
493 | * | |
494 | * for a given disk offset, load (and allocate if needed) | |
495 | * the l2 table. | |
496 | * | |
497 | * the l2 table offset in the qcow2 file and the cluster index | |
498 | * in the l2 table are given to the caller. | |
499 | * | |
1e3e8f1a | 500 | * Returns 0 on success, -errno in failure case |
45aba42f | 501 | */ |
45aba42f KW |
502 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
503 | uint64_t **new_l2_table, | |
45aba42f KW |
504 | int *new_l2_index) |
505 | { | |
506 | BDRVQcowState *s = bs->opaque; | |
80ee15a6 | 507 | unsigned int l1_index, l2_index; |
c46e1167 KW |
508 | uint64_t l2_offset; |
509 | uint64_t *l2_table = NULL; | |
80ee15a6 | 510 | int ret; |
45aba42f KW |
511 | |
512 | /* seek the the l2 offset in the l1 table */ | |
513 | ||
514 | l1_index = offset >> (s->l2_bits + s->cluster_bits); | |
515 | if (l1_index >= s->l1_size) { | |
72893756 | 516 | ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
1e3e8f1a KW |
517 | if (ret < 0) { |
518 | return ret; | |
519 | } | |
45aba42f | 520 | } |
8e37f681 KW |
521 | |
522 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; | |
45aba42f KW |
523 | |
524 | /* seek the l2 table of the given l2 offset */ | |
525 | ||
8e37f681 | 526 | if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) { |
45aba42f | 527 | /* load the l2 table in memory */ |
55c17e98 KW |
528 | ret = l2_load(bs, l2_offset, &l2_table); |
529 | if (ret < 0) { | |
530 | return ret; | |
1e3e8f1a | 531 | } |
45aba42f | 532 | } else { |
16fde5f2 | 533 | /* First allocate a new L2 table (and do COW if needed) */ |
c46e1167 KW |
534 | ret = l2_allocate(bs, l1_index, &l2_table); |
535 | if (ret < 0) { | |
536 | return ret; | |
1e3e8f1a | 537 | } |
16fde5f2 KW |
538 | |
539 | /* Then decrease the refcount of the old table */ | |
540 | if (l2_offset) { | |
541 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t)); | |
542 | } | |
45aba42f KW |
543 | } |
544 | ||
545 | /* find the cluster offset for the given disk offset */ | |
546 | ||
547 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
548 | ||
549 | *new_l2_table = l2_table; | |
45aba42f KW |
550 | *new_l2_index = l2_index; |
551 | ||
1e3e8f1a | 552 | return 0; |
45aba42f KW |
553 | } |
554 | ||
555 | /* | |
556 | * alloc_compressed_cluster_offset | |
557 | * | |
558 | * For a given offset of the disk image, return cluster offset in | |
559 | * qcow2 file. | |
560 | * | |
561 | * If the offset is not found, allocate a new compressed cluster. | |
562 | * | |
563 | * Return the cluster offset if successful, | |
564 | * Return 0, otherwise. | |
565 | * | |
566 | */ | |
567 | ||
ed6ccf0f KW |
568 | uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
569 | uint64_t offset, | |
570 | int compressed_size) | |
45aba42f KW |
571 | { |
572 | BDRVQcowState *s = bs->opaque; | |
573 | int l2_index, ret; | |
3948d1d4 | 574 | uint64_t *l2_table; |
f4f0d391 | 575 | int64_t cluster_offset; |
45aba42f KW |
576 | int nb_csectors; |
577 | ||
3948d1d4 | 578 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
1e3e8f1a | 579 | if (ret < 0) { |
45aba42f | 580 | return 0; |
1e3e8f1a | 581 | } |
45aba42f | 582 | |
b0b6862e KW |
583 | /* Compression can't overwrite anything. Fail if the cluster was already |
584 | * allocated. */ | |
45aba42f | 585 | cluster_offset = be64_to_cpu(l2_table[l2_index]); |
b0b6862e | 586 | if (cluster_offset & L2E_OFFSET_MASK) { |
8f1efd00 KW |
587 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
588 | return 0; | |
589 | } | |
45aba42f | 590 | |
ed6ccf0f | 591 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
5d757b56 | 592 | if (cluster_offset < 0) { |
29c1a730 | 593 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
5d757b56 KW |
594 | return 0; |
595 | } | |
596 | ||
45aba42f KW |
597 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
598 | (cluster_offset >> 9); | |
599 | ||
600 | cluster_offset |= QCOW_OFLAG_COMPRESSED | | |
601 | ((uint64_t)nb_csectors << s->csize_shift); | |
602 | ||
603 | /* update L2 table */ | |
604 | ||
605 | /* compressed clusters never have the copied flag */ | |
606 | ||
66f82cee | 607 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
29c1a730 | 608 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
45aba42f | 609 | l2_table[l2_index] = cpu_to_be64(cluster_offset); |
29c1a730 | 610 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
79a31189 | 611 | if (ret < 0) { |
29c1a730 | 612 | return 0; |
4c1612d9 KW |
613 | } |
614 | ||
29c1a730 | 615 | return cluster_offset; |
4c1612d9 KW |
616 | } |
617 | ||
148da7ea | 618 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) |
45aba42f KW |
619 | { |
620 | BDRVQcowState *s = bs->opaque; | |
621 | int i, j = 0, l2_index, ret; | |
3948d1d4 | 622 | uint64_t *old_cluster, start_sect, *l2_table; |
250196f1 | 623 | uint64_t cluster_offset = m->alloc_offset; |
29c1a730 | 624 | bool cow = false; |
45aba42f | 625 | |
3cce16f4 KW |
626 | trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
627 | ||
45aba42f KW |
628 | if (m->nb_clusters == 0) |
629 | return 0; | |
630 | ||
7267c094 | 631 | old_cluster = g_malloc(m->nb_clusters * sizeof(uint64_t)); |
45aba42f KW |
632 | |
633 | /* copy content of unmodified sectors */ | |
634 | start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9; | |
635 | if (m->n_start) { | |
29c1a730 | 636 | cow = true; |
1b9f1491 | 637 | qemu_co_mutex_unlock(&s->lock); |
45aba42f | 638 | ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start); |
1b9f1491 | 639 | qemu_co_mutex_lock(&s->lock); |
45aba42f KW |
640 | if (ret < 0) |
641 | goto err; | |
642 | } | |
643 | ||
644 | if (m->nb_available & (s->cluster_sectors - 1)) { | |
29c1a730 | 645 | cow = true; |
1b9f1491 | 646 | qemu_co_mutex_unlock(&s->lock); |
cdba7fee KW |
647 | ret = copy_sectors(bs, start_sect, cluster_offset, m->nb_available, |
648 | align_offset(m->nb_available, s->cluster_sectors)); | |
1b9f1491 | 649 | qemu_co_mutex_lock(&s->lock); |
45aba42f KW |
650 | if (ret < 0) |
651 | goto err; | |
652 | } | |
653 | ||
29c1a730 KW |
654 | /* |
655 | * Update L2 table. | |
656 | * | |
657 | * Before we update the L2 table to actually point to the new cluster, we | |
658 | * need to be sure that the refcounts have been increased and COW was | |
659 | * handled. | |
660 | */ | |
661 | if (cow) { | |
3de0a294 | 662 | qcow2_cache_depends_on_flush(s->l2_table_cache); |
29c1a730 KW |
663 | } |
664 | ||
bfe8043e SH |
665 | if (qcow2_need_accurate_refcounts(s)) { |
666 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
667 | s->refcount_block_cache); | |
668 | } | |
3948d1d4 | 669 | ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index); |
1e3e8f1a | 670 | if (ret < 0) { |
45aba42f | 671 | goto err; |
1e3e8f1a | 672 | } |
29c1a730 | 673 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
45aba42f KW |
674 | |
675 | for (i = 0; i < m->nb_clusters; i++) { | |
676 | /* if two concurrent writes happen to the same unallocated cluster | |
677 | * each write allocates separate cluster and writes data concurrently. | |
678 | * The first one to complete updates l2 table with pointer to its | |
679 | * cluster the second one has to do RMW (which is done above by | |
680 | * copy_sectors()), update l2 table with its cluster pointer and free | |
681 | * old cluster. This is what this loop does */ | |
682 | if(l2_table[l2_index + i] != 0) | |
683 | old_cluster[j++] = l2_table[l2_index + i]; | |
684 | ||
685 | l2_table[l2_index + i] = cpu_to_be64((cluster_offset + | |
686 | (i << s->cluster_bits)) | QCOW_OFLAG_COPIED); | |
687 | } | |
688 | ||
9f8e668e | 689 | |
29c1a730 | 690 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
c835d00f | 691 | if (ret < 0) { |
45aba42f | 692 | goto err; |
4c1612d9 | 693 | } |
45aba42f | 694 | |
7ec5e6a4 KW |
695 | /* |
696 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
697 | * Also flush bs->file to get the right order for L2 and refcount update. | |
698 | */ | |
699 | if (j != 0) { | |
7ec5e6a4 | 700 | for (i = 0; i < j; i++) { |
8e37f681 | 701 | qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1); |
7ec5e6a4 KW |
702 | } |
703 | } | |
45aba42f KW |
704 | |
705 | ret = 0; | |
706 | err: | |
7267c094 | 707 | g_free(old_cluster); |
45aba42f KW |
708 | return ret; |
709 | } | |
710 | ||
bf319ece KW |
711 | /* |
712 | * Returns the number of contiguous clusters that can be used for an allocating | |
713 | * write, but require COW to be performed (this includes yet unallocated space, | |
714 | * which must copy from the backing file) | |
715 | */ | |
716 | static int count_cow_clusters(BDRVQcowState *s, int nb_clusters, | |
717 | uint64_t *l2_table, int l2_index) | |
718 | { | |
143550a8 | 719 | int i; |
bf319ece | 720 | |
143550a8 KW |
721 | for (i = 0; i < nb_clusters; i++) { |
722 | uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]); | |
723 | int cluster_type = qcow2_get_cluster_type(l2_entry); | |
724 | ||
725 | switch(cluster_type) { | |
726 | case QCOW2_CLUSTER_NORMAL: | |
727 | if (l2_entry & QCOW_OFLAG_COPIED) { | |
728 | goto out; | |
729 | } | |
bf319ece | 730 | break; |
143550a8 KW |
731 | case QCOW2_CLUSTER_UNALLOCATED: |
732 | case QCOW2_CLUSTER_COMPRESSED: | |
6377af48 | 733 | case QCOW2_CLUSTER_ZERO: |
bf319ece | 734 | break; |
143550a8 KW |
735 | default: |
736 | abort(); | |
737 | } | |
bf319ece KW |
738 | } |
739 | ||
143550a8 | 740 | out: |
bf319ece KW |
741 | assert(i <= nb_clusters); |
742 | return i; | |
743 | } | |
744 | ||
250196f1 KW |
745 | /* |
746 | * Allocates new clusters for the given guest_offset. | |
747 | * | |
748 | * At most *nb_clusters are allocated, and on return *nb_clusters is updated to | |
749 | * contain the number of clusters that have been allocated and are contiguous | |
750 | * in the image file. | |
751 | * | |
752 | * If *host_offset is non-zero, it specifies the offset in the image file at | |
753 | * which the new clusters must start. *nb_clusters can be 0 on return in this | |
754 | * case if the cluster at host_offset is already in use. If *host_offset is | |
755 | * zero, the clusters can be allocated anywhere in the image file. | |
756 | * | |
757 | * *host_offset is updated to contain the offset into the image file at which | |
758 | * the first allocated cluster starts. | |
759 | * | |
760 | * Return 0 on success and -errno in error cases. -EAGAIN means that the | |
761 | * function has been waiting for another request and the allocation must be | |
762 | * restarted, but the whole request should not be failed. | |
763 | */ | |
764 | static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, | |
60651f90 | 765 | uint64_t *host_offset, unsigned int *nb_clusters) |
250196f1 KW |
766 | { |
767 | BDRVQcowState *s = bs->opaque; | |
250196f1 KW |
768 | QCowL2Meta *old_alloc; |
769 | ||
770 | trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, | |
771 | *host_offset, *nb_clusters); | |
772 | ||
773 | /* | |
774 | * Check if there already is an AIO write request in flight which allocates | |
775 | * the same cluster. In this case we need to wait until the previous | |
776 | * request has completed and updated the L2 table accordingly. | |
777 | */ | |
778 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { | |
779 | ||
780 | uint64_t start = guest_offset >> s->cluster_bits; | |
781 | uint64_t end = start + *nb_clusters; | |
782 | uint64_t old_start = old_alloc->offset >> s->cluster_bits; | |
783 | uint64_t old_end = old_start + old_alloc->nb_clusters; | |
784 | ||
785 | if (end < old_start || start > old_end) { | |
786 | /* No intersection */ | |
787 | } else { | |
788 | if (start < old_start) { | |
789 | /* Stop at the start of a running allocation */ | |
790 | *nb_clusters = old_start - start; | |
791 | } else { | |
792 | *nb_clusters = 0; | |
793 | } | |
794 | ||
795 | if (*nb_clusters == 0) { | |
796 | /* Wait for the dependency to complete. We need to recheck | |
797 | * the free/allocated clusters when we continue. */ | |
798 | qemu_co_mutex_unlock(&s->lock); | |
799 | qemu_co_queue_wait(&old_alloc->dependent_requests); | |
800 | qemu_co_mutex_lock(&s->lock); | |
801 | return -EAGAIN; | |
802 | } | |
803 | } | |
804 | } | |
805 | ||
806 | if (!*nb_clusters) { | |
807 | abort(); | |
808 | } | |
809 | ||
810 | /* Allocate new clusters */ | |
811 | trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); | |
812 | if (*host_offset == 0) { | |
df021791 KW |
813 | int64_t cluster_offset = |
814 | qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); | |
815 | if (cluster_offset < 0) { | |
816 | return cluster_offset; | |
817 | } | |
818 | *host_offset = cluster_offset; | |
819 | return 0; | |
250196f1 | 820 | } else { |
df021791 KW |
821 | int ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); |
822 | if (ret < 0) { | |
823 | return ret; | |
824 | } | |
825 | *nb_clusters = ret; | |
826 | return 0; | |
250196f1 | 827 | } |
250196f1 KW |
828 | } |
829 | ||
45aba42f KW |
830 | /* |
831 | * alloc_cluster_offset | |
832 | * | |
250196f1 KW |
833 | * For a given offset on the virtual disk, find the cluster offset in qcow2 |
834 | * file. If the offset is not found, allocate a new cluster. | |
45aba42f | 835 | * |
250196f1 | 836 | * If the cluster was already allocated, m->nb_clusters is set to 0 and |
a7912369 | 837 | * other fields in m are meaningless. |
148da7ea KW |
838 | * |
839 | * If the cluster is newly allocated, m->nb_clusters is set to the number of | |
68d100e9 KW |
840 | * contiguous clusters that have been allocated. In this case, the other |
841 | * fields of m are valid and contain information about the first allocated | |
842 | * cluster. | |
45aba42f | 843 | * |
68d100e9 KW |
844 | * If the request conflicts with another write request in flight, the coroutine |
845 | * is queued and will be reentered when the dependency has completed. | |
148da7ea KW |
846 | * |
847 | * Return 0 on success and -errno in error cases | |
45aba42f | 848 | */ |
f4f0d391 KW |
849 | int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, |
850 | int n_start, int n_end, int *num, QCowL2Meta *m) | |
45aba42f KW |
851 | { |
852 | BDRVQcowState *s = bs->opaque; | |
250196f1 | 853 | int l2_index, ret, sectors; |
3948d1d4 | 854 | uint64_t *l2_table; |
250196f1 KW |
855 | unsigned int nb_clusters, keep_clusters; |
856 | uint64_t cluster_offset; | |
45aba42f | 857 | |
3cce16f4 KW |
858 | trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, |
859 | n_start, n_end); | |
860 | ||
250196f1 | 861 | /* Find L2 entry for the first involved cluster */ |
72424114 | 862 | again: |
3948d1d4 | 863 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
1e3e8f1a | 864 | if (ret < 0) { |
148da7ea | 865 | return ret; |
1e3e8f1a | 866 | } |
45aba42f | 867 | |
250196f1 KW |
868 | /* |
869 | * Calculate the number of clusters to look for. We stop at L2 table | |
870 | * boundaries to keep things simple. | |
871 | */ | |
250196f1 KW |
872 | nb_clusters = MIN(size_to_clusters(s, n_end << BDRV_SECTOR_BITS), |
873 | s->l2_size - l2_index); | |
45aba42f KW |
874 | |
875 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
876 | ||
250196f1 KW |
877 | /* |
878 | * Check how many clusters are already allocated and don't need COW, and how | |
879 | * many need a new allocation. | |
880 | */ | |
8e37f681 KW |
881 | if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL |
882 | && (cluster_offset & QCOW_OFLAG_COPIED)) | |
883 | { | |
250196f1 | 884 | /* We keep all QCOW_OFLAG_COPIED clusters */ |
6377af48 KW |
885 | keep_clusters = |
886 | count_contiguous_clusters(nb_clusters, s->cluster_size, | |
887 | &l2_table[l2_index], 0, | |
888 | QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); | |
250196f1 KW |
889 | assert(keep_clusters <= nb_clusters); |
890 | nb_clusters -= keep_clusters; | |
891 | } else { | |
54e68143 KW |
892 | keep_clusters = 0; |
893 | cluster_offset = 0; | |
894 | } | |
895 | ||
896 | if (nb_clusters > 0) { | |
250196f1 | 897 | /* For the moment, overwrite compressed clusters one by one */ |
54e68143 KW |
898 | uint64_t entry = be64_to_cpu(l2_table[l2_index + keep_clusters]); |
899 | if (entry & QCOW_OFLAG_COMPRESSED) { | |
250196f1 KW |
900 | nb_clusters = 1; |
901 | } else { | |
54e68143 KW |
902 | nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, |
903 | l2_index + keep_clusters); | |
250196f1 | 904 | } |
45aba42f KW |
905 | } |
906 | ||
8e37f681 | 907 | cluster_offset &= L2E_OFFSET_MASK; |
45aba42f | 908 | |
72424114 KW |
909 | /* |
910 | * The L2 table isn't used any more after this. As long as the cache works | |
911 | * synchronously, it's important to release it before calling | |
912 | * do_alloc_cluster_offset, which may yield if we need to wait for another | |
913 | * request to complete. If we still had the reference, we could use up the | |
914 | * whole cache with sleeping requests. | |
915 | */ | |
916 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); | |
917 | if (ret < 0) { | |
918 | return ret; | |
919 | } | |
920 | ||
250196f1 KW |
921 | /* If there is something left to allocate, do that now */ |
922 | *m = (QCowL2Meta) { | |
923 | .cluster_offset = cluster_offset, | |
924 | .nb_clusters = 0, | |
925 | }; | |
926 | qemu_co_queue_init(&m->dependent_requests); | |
45aba42f | 927 | |
250196f1 KW |
928 | if (nb_clusters > 0) { |
929 | uint64_t alloc_offset; | |
930 | uint64_t alloc_cluster_offset; | |
931 | uint64_t keep_bytes = keep_clusters * s->cluster_size; | |
45aba42f | 932 | |
250196f1 KW |
933 | /* Calculate start and size of allocation */ |
934 | alloc_offset = offset + keep_bytes; | |
45aba42f | 935 | |
250196f1 KW |
936 | if (keep_clusters == 0) { |
937 | alloc_cluster_offset = 0; | |
f214978a | 938 | } else { |
250196f1 | 939 | alloc_cluster_offset = cluster_offset + keep_bytes; |
f214978a | 940 | } |
f214978a | 941 | |
250196f1 KW |
942 | /* Allocate, if necessary at a given offset in the image file */ |
943 | ret = do_alloc_cluster_offset(bs, alloc_offset, &alloc_cluster_offset, | |
60651f90 | 944 | &nb_clusters); |
250196f1 KW |
945 | if (ret == -EAGAIN) { |
946 | goto again; | |
947 | } else if (ret < 0) { | |
948 | goto fail; | |
949 | } | |
f214978a | 950 | |
250196f1 KW |
951 | /* save info needed for meta data update */ |
952 | if (nb_clusters > 0) { | |
b7ab0fea KW |
953 | /* |
954 | * requested_sectors: Number of sectors from the start of the first | |
955 | * newly allocated cluster to the end of the (possibly shortened | |
956 | * before) write request. | |
957 | * | |
958 | * avail_sectors: Number of sectors from the start of the first | |
959 | * newly allocated to the end of the last newly allocated cluster. | |
960 | */ | |
250196f1 | 961 | int requested_sectors = n_end - keep_clusters * s->cluster_sectors; |
b7ab0fea | 962 | int avail_sectors = nb_clusters |
250196f1 KW |
963 | << (s->cluster_bits - BDRV_SECTOR_BITS); |
964 | ||
965 | *m = (QCowL2Meta) { | |
966 | .cluster_offset = keep_clusters == 0 ? | |
967 | alloc_cluster_offset : cluster_offset, | |
968 | .alloc_offset = alloc_cluster_offset, | |
969 | .offset = alloc_offset, | |
970 | .n_start = keep_clusters == 0 ? n_start : 0, | |
971 | .nb_clusters = nb_clusters, | |
972 | .nb_available = MIN(requested_sectors, avail_sectors), | |
973 | }; | |
974 | qemu_co_queue_init(&m->dependent_requests); | |
975 | QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight); | |
976 | } | |
5d757b56 | 977 | } |
45aba42f | 978 | |
250196f1 | 979 | /* Some cleanup work */ |
250196f1 KW |
980 | sectors = (keep_clusters + nb_clusters) << (s->cluster_bits - 9); |
981 | if (sectors > n_end) { | |
982 | sectors = n_end; | |
983 | } | |
45aba42f | 984 | |
250196f1 KW |
985 | assert(sectors > n_start); |
986 | *num = sectors - n_start; | |
45aba42f | 987 | |
148da7ea | 988 | return 0; |
29c1a730 KW |
989 | |
990 | fail: | |
8dc0a5e7 | 991 | if (m->nb_clusters > 0) { |
250196f1 KW |
992 | QLIST_REMOVE(m, next_in_flight); |
993 | } | |
29c1a730 | 994 | return ret; |
45aba42f KW |
995 | } |
996 | ||
997 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
998 | const uint8_t *buf, int buf_size) | |
999 | { | |
1000 | z_stream strm1, *strm = &strm1; | |
1001 | int ret, out_len; | |
1002 | ||
1003 | memset(strm, 0, sizeof(*strm)); | |
1004 | ||
1005 | strm->next_in = (uint8_t *)buf; | |
1006 | strm->avail_in = buf_size; | |
1007 | strm->next_out = out_buf; | |
1008 | strm->avail_out = out_buf_size; | |
1009 | ||
1010 | ret = inflateInit2(strm, -12); | |
1011 | if (ret != Z_OK) | |
1012 | return -1; | |
1013 | ret = inflate(strm, Z_FINISH); | |
1014 | out_len = strm->next_out - out_buf; | |
1015 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
1016 | out_len != out_buf_size) { | |
1017 | inflateEnd(strm); | |
1018 | return -1; | |
1019 | } | |
1020 | inflateEnd(strm); | |
1021 | return 0; | |
1022 | } | |
1023 | ||
66f82cee | 1024 | int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) |
45aba42f | 1025 | { |
66f82cee | 1026 | BDRVQcowState *s = bs->opaque; |
45aba42f KW |
1027 | int ret, csize, nb_csectors, sector_offset; |
1028 | uint64_t coffset; | |
1029 | ||
1030 | coffset = cluster_offset & s->cluster_offset_mask; | |
1031 | if (s->cluster_cache_offset != coffset) { | |
1032 | nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1; | |
1033 | sector_offset = coffset & 511; | |
1034 | csize = nb_csectors * 512 - sector_offset; | |
66f82cee KW |
1035 | BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
1036 | ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, nb_csectors); | |
45aba42f | 1037 | if (ret < 0) { |
8af36488 | 1038 | return ret; |
45aba42f KW |
1039 | } |
1040 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
1041 | s->cluster_data + sector_offset, csize) < 0) { | |
8af36488 | 1042 | return -EIO; |
45aba42f KW |
1043 | } |
1044 | s->cluster_cache_offset = coffset; | |
1045 | } | |
1046 | return 0; | |
1047 | } | |
5ea929e3 KW |
1048 | |
1049 | /* | |
1050 | * This discards as many clusters of nb_clusters as possible at once (i.e. | |
1051 | * all clusters in the same L2 table) and returns the number of discarded | |
1052 | * clusters. | |
1053 | */ | |
1054 | static int discard_single_l2(BlockDriverState *bs, uint64_t offset, | |
1055 | unsigned int nb_clusters) | |
1056 | { | |
1057 | BDRVQcowState *s = bs->opaque; | |
3948d1d4 | 1058 | uint64_t *l2_table; |
5ea929e3 KW |
1059 | int l2_index; |
1060 | int ret; | |
1061 | int i; | |
1062 | ||
3948d1d4 | 1063 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
5ea929e3 KW |
1064 | if (ret < 0) { |
1065 | return ret; | |
1066 | } | |
1067 | ||
1068 | /* Limit nb_clusters to one L2 table */ | |
1069 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
1070 | ||
1071 | for (i = 0; i < nb_clusters; i++) { | |
1072 | uint64_t old_offset; | |
1073 | ||
1074 | old_offset = be64_to_cpu(l2_table[l2_index + i]); | |
8e37f681 | 1075 | if ((old_offset & L2E_OFFSET_MASK) == 0) { |
5ea929e3 KW |
1076 | continue; |
1077 | } | |
1078 | ||
1079 | /* First remove L2 entries */ | |
1080 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); | |
1081 | l2_table[l2_index + i] = cpu_to_be64(0); | |
1082 | ||
1083 | /* Then decrease the refcount */ | |
1084 | qcow2_free_any_clusters(bs, old_offset, 1); | |
1085 | } | |
1086 | ||
1087 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); | |
1088 | if (ret < 0) { | |
1089 | return ret; | |
1090 | } | |
1091 | ||
1092 | return nb_clusters; | |
1093 | } | |
1094 | ||
1095 | int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset, | |
1096 | int nb_sectors) | |
1097 | { | |
1098 | BDRVQcowState *s = bs->opaque; | |
1099 | uint64_t end_offset; | |
1100 | unsigned int nb_clusters; | |
1101 | int ret; | |
1102 | ||
1103 | end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS); | |
1104 | ||
1105 | /* Round start up and end down */ | |
1106 | offset = align_offset(offset, s->cluster_size); | |
1107 | end_offset &= ~(s->cluster_size - 1); | |
1108 | ||
1109 | if (offset > end_offset) { | |
1110 | return 0; | |
1111 | } | |
1112 | ||
1113 | nb_clusters = size_to_clusters(s, end_offset - offset); | |
1114 | ||
1115 | /* Each L2 table is handled by its own loop iteration */ | |
1116 | while (nb_clusters > 0) { | |
1117 | ret = discard_single_l2(bs, offset, nb_clusters); | |
1118 | if (ret < 0) { | |
1119 | return ret; | |
1120 | } | |
1121 | ||
1122 | nb_clusters -= ret; | |
1123 | offset += (ret * s->cluster_size); | |
1124 | } | |
1125 | ||
1126 | return 0; | |
1127 | } | |
621f0589 KW |
1128 | |
1129 | /* | |
1130 | * This zeroes as many clusters of nb_clusters as possible at once (i.e. | |
1131 | * all clusters in the same L2 table) and returns the number of zeroed | |
1132 | * clusters. | |
1133 | */ | |
1134 | static int zero_single_l2(BlockDriverState *bs, uint64_t offset, | |
1135 | unsigned int nb_clusters) | |
1136 | { | |
1137 | BDRVQcowState *s = bs->opaque; | |
1138 | uint64_t *l2_table; | |
1139 | int l2_index; | |
1140 | int ret; | |
1141 | int i; | |
1142 | ||
1143 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); | |
1144 | if (ret < 0) { | |
1145 | return ret; | |
1146 | } | |
1147 | ||
1148 | /* Limit nb_clusters to one L2 table */ | |
1149 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
1150 | ||
1151 | for (i = 0; i < nb_clusters; i++) { | |
1152 | uint64_t old_offset; | |
1153 | ||
1154 | old_offset = be64_to_cpu(l2_table[l2_index + i]); | |
1155 | ||
1156 | /* Update L2 entries */ | |
1157 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); | |
1158 | if (old_offset & QCOW_OFLAG_COMPRESSED) { | |
1159 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); | |
1160 | qcow2_free_any_clusters(bs, old_offset, 1); | |
1161 | } else { | |
1162 | l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO); | |
1163 | } | |
1164 | } | |
1165 | ||
1166 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); | |
1167 | if (ret < 0) { | |
1168 | return ret; | |
1169 | } | |
1170 | ||
1171 | return nb_clusters; | |
1172 | } | |
1173 | ||
1174 | int qcow2_zero_clusters(BlockDriverState *bs, uint64_t offset, int nb_sectors) | |
1175 | { | |
1176 | BDRVQcowState *s = bs->opaque; | |
1177 | unsigned int nb_clusters; | |
1178 | int ret; | |
1179 | ||
1180 | /* The zero flag is only supported by version 3 and newer */ | |
1181 | if (s->qcow_version < 3) { | |
1182 | return -ENOTSUP; | |
1183 | } | |
1184 | ||
1185 | /* Each L2 table is handled by its own loop iteration */ | |
1186 | nb_clusters = size_to_clusters(s, nb_sectors << BDRV_SECTOR_BITS); | |
1187 | ||
1188 | while (nb_clusters > 0) { | |
1189 | ret = zero_single_l2(bs, offset, nb_clusters); | |
1190 | if (ret < 0) { | |
1191 | return ret; | |
1192 | } | |
1193 | ||
1194 | nb_clusters -= ret; | |
1195 | offset += (ret * s->cluster_size); | |
1196 | } | |
1197 | ||
1198 | return 0; | |
1199 | } |