]>
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" | |
737e150e | 28 | #include "block/block_int.h" |
45aba42f | 29 | #include "block/qcow2.h" |
3cce16f4 | 30 | #include "trace.h" |
45aba42f | 31 | |
2cf7cfa1 KW |
32 | int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size, |
33 | bool exact_size) | |
45aba42f KW |
34 | { |
35 | BDRVQcowState *s = bs->opaque; | |
2cf7cfa1 | 36 | int new_l1_size2, ret, i; |
45aba42f | 37 | uint64_t *new_l1_table; |
fda74f82 | 38 | int64_t old_l1_table_offset, old_l1_size; |
2cf7cfa1 | 39 | int64_t new_l1_table_offset, new_l1_size; |
45aba42f KW |
40 | uint8_t data[12]; |
41 | ||
72893756 | 42 | if (min_size <= s->l1_size) |
45aba42f | 43 | return 0; |
72893756 SH |
44 | |
45 | if (exact_size) { | |
46 | new_l1_size = min_size; | |
47 | } else { | |
48 | /* Bump size up to reduce the number of times we have to grow */ | |
49 | new_l1_size = s->l1_size; | |
50 | if (new_l1_size == 0) { | |
51 | new_l1_size = 1; | |
52 | } | |
53 | while (min_size > new_l1_size) { | |
54 | new_l1_size = (new_l1_size * 3 + 1) / 2; | |
55 | } | |
45aba42f | 56 | } |
72893756 | 57 | |
2cf7cfa1 KW |
58 | if (new_l1_size > INT_MAX) { |
59 | return -EFBIG; | |
60 | } | |
61 | ||
45aba42f | 62 | #ifdef DEBUG_ALLOC2 |
2cf7cfa1 KW |
63 | fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n", |
64 | s->l1_size, new_l1_size); | |
45aba42f KW |
65 | #endif |
66 | ||
67 | new_l1_size2 = sizeof(uint64_t) * new_l1_size; | |
7267c094 | 68 | new_l1_table = g_malloc0(align_offset(new_l1_size2, 512)); |
45aba42f KW |
69 | memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t)); |
70 | ||
71 | /* write new table (align to cluster) */ | |
66f82cee | 72 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE); |
ed6ccf0f | 73 | new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2); |
5d757b56 | 74 | if (new_l1_table_offset < 0) { |
7267c094 | 75 | g_free(new_l1_table); |
5d757b56 KW |
76 | return new_l1_table_offset; |
77 | } | |
29c1a730 KW |
78 | |
79 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
80 | if (ret < 0) { | |
80fa3341 | 81 | goto fail; |
29c1a730 | 82 | } |
45aba42f | 83 | |
cf93980e MR |
84 | /* the L1 position has not yet been updated, so these clusters must |
85 | * indeed be completely free */ | |
86 | ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_DEFAULT, | |
87 | new_l1_table_offset, new_l1_size2); | |
88 | if (ret < 0) { | |
89 | goto fail; | |
90 | } | |
91 | ||
66f82cee | 92 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE); |
45aba42f KW |
93 | for(i = 0; i < s->l1_size; i++) |
94 | new_l1_table[i] = cpu_to_be64(new_l1_table[i]); | |
8b3b7206 KW |
95 | ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, new_l1_table, new_l1_size2); |
96 | if (ret < 0) | |
45aba42f KW |
97 | goto fail; |
98 | for(i = 0; i < s->l1_size; i++) | |
99 | new_l1_table[i] = be64_to_cpu(new_l1_table[i]); | |
100 | ||
101 | /* set new table */ | |
66f82cee | 102 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE); |
45aba42f | 103 | cpu_to_be32w((uint32_t*)data, new_l1_size); |
653df36b | 104 | cpu_to_be64wu((uint64_t*)(data + 4), new_l1_table_offset); |
8b3b7206 KW |
105 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), data,sizeof(data)); |
106 | if (ret < 0) { | |
45aba42f | 107 | goto fail; |
fb8fa77c | 108 | } |
7267c094 | 109 | g_free(s->l1_table); |
fda74f82 | 110 | old_l1_table_offset = s->l1_table_offset; |
45aba42f KW |
111 | s->l1_table_offset = new_l1_table_offset; |
112 | s->l1_table = new_l1_table; | |
fda74f82 | 113 | old_l1_size = s->l1_size; |
45aba42f | 114 | s->l1_size = new_l1_size; |
fda74f82 MR |
115 | qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * sizeof(uint64_t), |
116 | QCOW2_DISCARD_OTHER); | |
45aba42f KW |
117 | return 0; |
118 | fail: | |
7267c094 | 119 | g_free(new_l1_table); |
6cfcb9b8 KW |
120 | qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2, |
121 | QCOW2_DISCARD_OTHER); | |
8b3b7206 | 122 | return ret; |
45aba42f KW |
123 | } |
124 | ||
45aba42f KW |
125 | /* |
126 | * l2_load | |
127 | * | |
128 | * Loads a L2 table into memory. If the table is in the cache, the cache | |
129 | * is used; otherwise the L2 table is loaded from the image file. | |
130 | * | |
131 | * Returns a pointer to the L2 table on success, or NULL if the read from | |
132 | * the image file failed. | |
133 | */ | |
134 | ||
55c17e98 KW |
135 | static int l2_load(BlockDriverState *bs, uint64_t l2_offset, |
136 | uint64_t **l2_table) | |
45aba42f KW |
137 | { |
138 | BDRVQcowState *s = bs->opaque; | |
55c17e98 | 139 | int ret; |
45aba42f | 140 | |
29c1a730 | 141 | ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, (void**) l2_table); |
45aba42f | 142 | |
29c1a730 | 143 | return ret; |
45aba42f KW |
144 | } |
145 | ||
6583e3c7 KW |
146 | /* |
147 | * Writes one sector of the L1 table to the disk (can't update single entries | |
148 | * and we really don't want bdrv_pread to perform a read-modify-write) | |
149 | */ | |
150 | #define L1_ENTRIES_PER_SECTOR (512 / 8) | |
e23e400e | 151 | int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index) |
6583e3c7 | 152 | { |
66f82cee | 153 | BDRVQcowState *s = bs->opaque; |
6583e3c7 KW |
154 | uint64_t buf[L1_ENTRIES_PER_SECTOR]; |
155 | int l1_start_index; | |
f7defcb6 | 156 | int i, ret; |
6583e3c7 KW |
157 | |
158 | l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1); | |
159 | for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) { | |
160 | buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]); | |
161 | } | |
162 | ||
cf93980e MR |
163 | ret = qcow2_pre_write_overlap_check(bs, |
164 | QCOW2_OL_DEFAULT & ~QCOW2_OL_ACTIVE_L1, | |
165 | s->l1_table_offset + 8 * l1_start_index, sizeof(buf)); | |
166 | if (ret < 0) { | |
167 | return ret; | |
168 | } | |
169 | ||
66f82cee | 170 | BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); |
8b3b7206 | 171 | ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset + 8 * l1_start_index, |
f7defcb6 KW |
172 | buf, sizeof(buf)); |
173 | if (ret < 0) { | |
174 | return ret; | |
6583e3c7 KW |
175 | } |
176 | ||
177 | return 0; | |
178 | } | |
179 | ||
45aba42f KW |
180 | /* |
181 | * l2_allocate | |
182 | * | |
183 | * Allocate a new l2 entry in the file. If l1_index points to an already | |
184 | * used entry in the L2 table (i.e. we are doing a copy on write for the L2 | |
185 | * table) copy the contents of the old L2 table into the newly allocated one. | |
186 | * Otherwise the new table is initialized with zeros. | |
187 | * | |
188 | */ | |
189 | ||
c46e1167 | 190 | static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table) |
45aba42f KW |
191 | { |
192 | BDRVQcowState *s = bs->opaque; | |
6583e3c7 | 193 | uint64_t old_l2_offset; |
8585afd8 | 194 | uint64_t *l2_table = NULL; |
f4f0d391 | 195 | int64_t l2_offset; |
c46e1167 | 196 | int ret; |
45aba42f KW |
197 | |
198 | old_l2_offset = s->l1_table[l1_index]; | |
199 | ||
3cce16f4 KW |
200 | trace_qcow2_l2_allocate(bs, l1_index); |
201 | ||
45aba42f KW |
202 | /* allocate a new l2 entry */ |
203 | ||
ed6ccf0f | 204 | l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t)); |
5d757b56 | 205 | if (l2_offset < 0) { |
be0b742e MR |
206 | ret = l2_offset; |
207 | goto fail; | |
5d757b56 | 208 | } |
29c1a730 KW |
209 | |
210 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
211 | if (ret < 0) { | |
212 | goto fail; | |
213 | } | |
45aba42f | 214 | |
45aba42f KW |
215 | /* allocate a new entry in the l2 cache */ |
216 | ||
3cce16f4 | 217 | trace_qcow2_l2_allocate_get_empty(bs, l1_index); |
29c1a730 KW |
218 | ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table); |
219 | if (ret < 0) { | |
be0b742e | 220 | goto fail; |
29c1a730 KW |
221 | } |
222 | ||
223 | l2_table = *table; | |
45aba42f | 224 | |
8e37f681 | 225 | if ((old_l2_offset & L1E_OFFSET_MASK) == 0) { |
45aba42f KW |
226 | /* if there was no old l2 table, clear the new table */ |
227 | memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); | |
228 | } else { | |
29c1a730 KW |
229 | uint64_t* old_table; |
230 | ||
45aba42f | 231 | /* if there was an old l2 table, read it from the disk */ |
66f82cee | 232 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ); |
8e37f681 KW |
233 | ret = qcow2_cache_get(bs, s->l2_table_cache, |
234 | old_l2_offset & L1E_OFFSET_MASK, | |
29c1a730 KW |
235 | (void**) &old_table); |
236 | if (ret < 0) { | |
237 | goto fail; | |
238 | } | |
239 | ||
240 | memcpy(l2_table, old_table, s->cluster_size); | |
241 | ||
242 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &old_table); | |
c46e1167 | 243 | if (ret < 0) { |
175e1152 | 244 | goto fail; |
c46e1167 | 245 | } |
45aba42f | 246 | } |
29c1a730 | 247 | |
45aba42f | 248 | /* write the l2 table to the file */ |
66f82cee | 249 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE); |
29c1a730 | 250 | |
3cce16f4 | 251 | trace_qcow2_l2_allocate_write_l2(bs, l1_index); |
29c1a730 KW |
252 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
253 | ret = qcow2_cache_flush(bs, s->l2_table_cache); | |
c46e1167 | 254 | if (ret < 0) { |
175e1152 KW |
255 | goto fail; |
256 | } | |
257 | ||
258 | /* update the L1 entry */ | |
3cce16f4 | 259 | trace_qcow2_l2_allocate_write_l1(bs, l1_index); |
175e1152 | 260 | s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; |
e23e400e | 261 | ret = qcow2_write_l1_entry(bs, l1_index); |
175e1152 KW |
262 | if (ret < 0) { |
263 | goto fail; | |
c46e1167 | 264 | } |
45aba42f | 265 | |
c46e1167 | 266 | *table = l2_table; |
3cce16f4 | 267 | trace_qcow2_l2_allocate_done(bs, l1_index, 0); |
c46e1167 | 268 | return 0; |
175e1152 KW |
269 | |
270 | fail: | |
3cce16f4 | 271 | trace_qcow2_l2_allocate_done(bs, l1_index, ret); |
8585afd8 MR |
272 | if (l2_table != NULL) { |
273 | qcow2_cache_put(bs, s->l2_table_cache, (void**) table); | |
274 | } | |
68dba0bf | 275 | s->l1_table[l1_index] = old_l2_offset; |
e3b21ef9 MR |
276 | if (l2_offset > 0) { |
277 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), | |
278 | QCOW2_DISCARD_ALWAYS); | |
279 | } | |
175e1152 | 280 | return ret; |
45aba42f KW |
281 | } |
282 | ||
2bfcc4a0 KW |
283 | /* |
284 | * Checks how many clusters in a given L2 table are contiguous in the image | |
285 | * file. As soon as one of the flags in the bitmask stop_flags changes compared | |
286 | * to the first cluster, the search is stopped and the cluster is not counted | |
287 | * as contiguous. (This allows it, for example, to stop at the first compressed | |
288 | * cluster which may require a different handling) | |
289 | */ | |
45aba42f | 290 | static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size, |
61653008 | 291 | uint64_t *l2_table, uint64_t stop_flags) |
45aba42f KW |
292 | { |
293 | int i; | |
15684a47 MR |
294 | uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW2_CLUSTER_COMPRESSED; |
295 | uint64_t first_entry = be64_to_cpu(l2_table[0]); | |
296 | uint64_t offset = first_entry & mask; | |
45aba42f KW |
297 | |
298 | if (!offset) | |
299 | return 0; | |
300 | ||
15684a47 MR |
301 | assert(qcow2_get_cluster_type(first_entry) != QCOW2_CLUSTER_COMPRESSED); |
302 | ||
61653008 | 303 | for (i = 0; i < nb_clusters; i++) { |
2bfcc4a0 KW |
304 | uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask; |
305 | if (offset + (uint64_t) i * cluster_size != l2_entry) { | |
45aba42f | 306 | break; |
2bfcc4a0 KW |
307 | } |
308 | } | |
45aba42f | 309 | |
61653008 | 310 | return i; |
45aba42f KW |
311 | } |
312 | ||
313 | static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table) | |
314 | { | |
2bfcc4a0 KW |
315 | int i; |
316 | ||
317 | for (i = 0; i < nb_clusters; i++) { | |
318 | int type = qcow2_get_cluster_type(be64_to_cpu(l2_table[i])); | |
45aba42f | 319 | |
2bfcc4a0 KW |
320 | if (type != QCOW2_CLUSTER_UNALLOCATED) { |
321 | break; | |
322 | } | |
323 | } | |
45aba42f KW |
324 | |
325 | return i; | |
326 | } | |
327 | ||
328 | /* The crypt function is compatible with the linux cryptoloop | |
329 | algorithm for < 4 GB images. NOTE: out_buf == in_buf is | |
330 | supported */ | |
ed6ccf0f KW |
331 | void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num, |
332 | uint8_t *out_buf, const uint8_t *in_buf, | |
333 | int nb_sectors, int enc, | |
334 | const AES_KEY *key) | |
45aba42f KW |
335 | { |
336 | union { | |
337 | uint64_t ll[2]; | |
338 | uint8_t b[16]; | |
339 | } ivec; | |
340 | int i; | |
341 | ||
342 | for(i = 0; i < nb_sectors; i++) { | |
343 | ivec.ll[0] = cpu_to_le64(sector_num); | |
344 | ivec.ll[1] = 0; | |
345 | AES_cbc_encrypt(in_buf, out_buf, 512, key, | |
346 | ivec.b, enc); | |
347 | sector_num++; | |
348 | in_buf += 512; | |
349 | out_buf += 512; | |
350 | } | |
351 | } | |
352 | ||
aef4acb6 SH |
353 | static int coroutine_fn copy_sectors(BlockDriverState *bs, |
354 | uint64_t start_sect, | |
355 | uint64_t cluster_offset, | |
356 | int n_start, int n_end) | |
45aba42f KW |
357 | { |
358 | BDRVQcowState *s = bs->opaque; | |
aef4acb6 SH |
359 | QEMUIOVector qiov; |
360 | struct iovec iov; | |
45aba42f | 361 | int n, ret; |
1b9f1491 KW |
362 | |
363 | /* | |
364 | * If this is the last cluster and it is only partially used, we must only | |
365 | * copy until the end of the image, or bdrv_check_request will fail for the | |
366 | * bdrv_read/write calls below. | |
367 | */ | |
368 | if (start_sect + n_end > bs->total_sectors) { | |
369 | n_end = bs->total_sectors - start_sect; | |
370 | } | |
45aba42f KW |
371 | |
372 | n = n_end - n_start; | |
1b9f1491 | 373 | if (n <= 0) { |
45aba42f | 374 | return 0; |
1b9f1491 KW |
375 | } |
376 | ||
aef4acb6 SH |
377 | iov.iov_len = n * BDRV_SECTOR_SIZE; |
378 | iov.iov_base = qemu_blockalign(bs, iov.iov_len); | |
379 | ||
380 | qemu_iovec_init_external(&qiov, &iov, 1); | |
1b9f1491 | 381 | |
66f82cee | 382 | BLKDBG_EVENT(bs->file, BLKDBG_COW_READ); |
aef4acb6 SH |
383 | |
384 | /* Call .bdrv_co_readv() directly instead of using the public block-layer | |
385 | * interface. This avoids double I/O throttling and request tracking, | |
386 | * which can lead to deadlock when block layer copy-on-read is enabled. | |
387 | */ | |
388 | ret = bs->drv->bdrv_co_readv(bs, start_sect + n_start, n, &qiov); | |
1b9f1491 KW |
389 | if (ret < 0) { |
390 | goto out; | |
391 | } | |
392 | ||
45aba42f | 393 | if (s->crypt_method) { |
ed6ccf0f | 394 | qcow2_encrypt_sectors(s, start_sect + n_start, |
aef4acb6 | 395 | iov.iov_base, iov.iov_base, n, 1, |
45aba42f KW |
396 | &s->aes_encrypt_key); |
397 | } | |
1b9f1491 | 398 | |
cf93980e MR |
399 | ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_DEFAULT, |
400 | cluster_offset + n_start * BDRV_SECTOR_SIZE, n * BDRV_SECTOR_SIZE); | |
401 | if (ret < 0) { | |
402 | goto out; | |
403 | } | |
404 | ||
66f82cee | 405 | BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE); |
aef4acb6 | 406 | ret = bdrv_co_writev(bs->file, (cluster_offset >> 9) + n_start, n, &qiov); |
1b9f1491 KW |
407 | if (ret < 0) { |
408 | goto out; | |
409 | } | |
410 | ||
411 | ret = 0; | |
412 | out: | |
aef4acb6 | 413 | qemu_vfree(iov.iov_base); |
1b9f1491 | 414 | return ret; |
45aba42f KW |
415 | } |
416 | ||
417 | ||
418 | /* | |
419 | * get_cluster_offset | |
420 | * | |
1c46efaa KW |
421 | * For a given offset of the disk image, find the cluster offset in |
422 | * qcow2 file. The offset is stored in *cluster_offset. | |
45aba42f | 423 | * |
d57237f2 | 424 | * on entry, *num is the number of contiguous sectors we'd like to |
45aba42f KW |
425 | * access following offset. |
426 | * | |
d57237f2 | 427 | * on exit, *num is the number of contiguous sectors we can read. |
45aba42f | 428 | * |
68d000a3 KW |
429 | * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error |
430 | * cases. | |
45aba42f | 431 | */ |
1c46efaa KW |
432 | int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, |
433 | int *num, uint64_t *cluster_offset) | |
45aba42f KW |
434 | { |
435 | BDRVQcowState *s = bs->opaque; | |
2cf7cfa1 KW |
436 | unsigned int l2_index; |
437 | uint64_t l1_index, l2_offset, *l2_table; | |
45aba42f | 438 | int l1_bits, c; |
80ee15a6 KW |
439 | unsigned int index_in_cluster, nb_clusters; |
440 | uint64_t nb_available, nb_needed; | |
55c17e98 | 441 | int ret; |
45aba42f KW |
442 | |
443 | index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1); | |
444 | nb_needed = *num + index_in_cluster; | |
445 | ||
446 | l1_bits = s->l2_bits + s->cluster_bits; | |
447 | ||
448 | /* compute how many bytes there are between the offset and | |
449 | * the end of the l1 entry | |
450 | */ | |
451 | ||
80ee15a6 | 452 | nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1)); |
45aba42f KW |
453 | |
454 | /* compute the number of available sectors */ | |
455 | ||
456 | nb_available = (nb_available >> 9) + index_in_cluster; | |
457 | ||
458 | if (nb_needed > nb_available) { | |
459 | nb_needed = nb_available; | |
460 | } | |
461 | ||
1c46efaa | 462 | *cluster_offset = 0; |
45aba42f KW |
463 | |
464 | /* seek the the l2 offset in the l1 table */ | |
465 | ||
466 | l1_index = offset >> l1_bits; | |
68d000a3 KW |
467 | if (l1_index >= s->l1_size) { |
468 | ret = QCOW2_CLUSTER_UNALLOCATED; | |
45aba42f | 469 | goto out; |
68d000a3 | 470 | } |
45aba42f | 471 | |
68d000a3 KW |
472 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
473 | if (!l2_offset) { | |
474 | ret = QCOW2_CLUSTER_UNALLOCATED; | |
45aba42f | 475 | goto out; |
68d000a3 | 476 | } |
45aba42f KW |
477 | |
478 | /* load the l2 table in memory */ | |
479 | ||
55c17e98 KW |
480 | ret = l2_load(bs, l2_offset, &l2_table); |
481 | if (ret < 0) { | |
482 | return ret; | |
1c46efaa | 483 | } |
45aba42f KW |
484 | |
485 | /* find the cluster offset for the given disk offset */ | |
486 | ||
487 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
1c46efaa | 488 | *cluster_offset = be64_to_cpu(l2_table[l2_index]); |
45aba42f KW |
489 | nb_clusters = size_to_clusters(s, nb_needed << 9); |
490 | ||
68d000a3 KW |
491 | ret = qcow2_get_cluster_type(*cluster_offset); |
492 | switch (ret) { | |
493 | case QCOW2_CLUSTER_COMPRESSED: | |
494 | /* Compressed clusters can only be processed one by one */ | |
495 | c = 1; | |
496 | *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK; | |
497 | break; | |
6377af48 | 498 | case QCOW2_CLUSTER_ZERO: |
381b487d PB |
499 | if (s->qcow_version < 3) { |
500 | return -EIO; | |
501 | } | |
6377af48 | 502 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, |
61653008 | 503 | &l2_table[l2_index], QCOW_OFLAG_ZERO); |
6377af48 KW |
504 | *cluster_offset = 0; |
505 | break; | |
68d000a3 | 506 | case QCOW2_CLUSTER_UNALLOCATED: |
45aba42f KW |
507 | /* how many empty clusters ? */ |
508 | c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]); | |
68d000a3 KW |
509 | *cluster_offset = 0; |
510 | break; | |
511 | case QCOW2_CLUSTER_NORMAL: | |
45aba42f KW |
512 | /* how many allocated clusters ? */ |
513 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
61653008 | 514 | &l2_table[l2_index], QCOW_OFLAG_ZERO); |
68d000a3 KW |
515 | *cluster_offset &= L2E_OFFSET_MASK; |
516 | break; | |
1417d7e4 KW |
517 | default: |
518 | abort(); | |
45aba42f KW |
519 | } |
520 | ||
29c1a730 KW |
521 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
522 | ||
68d000a3 KW |
523 | nb_available = (c * s->cluster_sectors); |
524 | ||
45aba42f KW |
525 | out: |
526 | if (nb_available > nb_needed) | |
527 | nb_available = nb_needed; | |
528 | ||
529 | *num = nb_available - index_in_cluster; | |
530 | ||
68d000a3 | 531 | return ret; |
45aba42f KW |
532 | } |
533 | ||
534 | /* | |
535 | * get_cluster_table | |
536 | * | |
537 | * for a given disk offset, load (and allocate if needed) | |
538 | * the l2 table. | |
539 | * | |
540 | * the l2 table offset in the qcow2 file and the cluster index | |
541 | * in the l2 table are given to the caller. | |
542 | * | |
1e3e8f1a | 543 | * Returns 0 on success, -errno in failure case |
45aba42f | 544 | */ |
45aba42f KW |
545 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
546 | uint64_t **new_l2_table, | |
45aba42f KW |
547 | int *new_l2_index) |
548 | { | |
549 | BDRVQcowState *s = bs->opaque; | |
2cf7cfa1 KW |
550 | unsigned int l2_index; |
551 | uint64_t l1_index, l2_offset; | |
c46e1167 | 552 | uint64_t *l2_table = NULL; |
80ee15a6 | 553 | int ret; |
45aba42f KW |
554 | |
555 | /* seek the the l2 offset in the l1 table */ | |
556 | ||
557 | l1_index = offset >> (s->l2_bits + s->cluster_bits); | |
558 | if (l1_index >= s->l1_size) { | |
72893756 | 559 | ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
1e3e8f1a KW |
560 | if (ret < 0) { |
561 | return ret; | |
562 | } | |
45aba42f | 563 | } |
8e37f681 | 564 | |
2cf7cfa1 | 565 | assert(l1_index < s->l1_size); |
8e37f681 | 566 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
45aba42f KW |
567 | |
568 | /* seek the l2 table of the given l2 offset */ | |
569 | ||
8e37f681 | 570 | if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) { |
45aba42f | 571 | /* load the l2 table in memory */ |
55c17e98 KW |
572 | ret = l2_load(bs, l2_offset, &l2_table); |
573 | if (ret < 0) { | |
574 | return ret; | |
1e3e8f1a | 575 | } |
45aba42f | 576 | } else { |
16fde5f2 | 577 | /* First allocate a new L2 table (and do COW if needed) */ |
c46e1167 KW |
578 | ret = l2_allocate(bs, l1_index, &l2_table); |
579 | if (ret < 0) { | |
580 | return ret; | |
1e3e8f1a | 581 | } |
16fde5f2 KW |
582 | |
583 | /* Then decrease the refcount of the old table */ | |
584 | if (l2_offset) { | |
6cfcb9b8 KW |
585 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), |
586 | QCOW2_DISCARD_OTHER); | |
16fde5f2 | 587 | } |
45aba42f KW |
588 | } |
589 | ||
590 | /* find the cluster offset for the given disk offset */ | |
591 | ||
592 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
593 | ||
594 | *new_l2_table = l2_table; | |
45aba42f KW |
595 | *new_l2_index = l2_index; |
596 | ||
1e3e8f1a | 597 | return 0; |
45aba42f KW |
598 | } |
599 | ||
600 | /* | |
601 | * alloc_compressed_cluster_offset | |
602 | * | |
603 | * For a given offset of the disk image, return cluster offset in | |
604 | * qcow2 file. | |
605 | * | |
606 | * If the offset is not found, allocate a new compressed cluster. | |
607 | * | |
608 | * Return the cluster offset if successful, | |
609 | * Return 0, otherwise. | |
610 | * | |
611 | */ | |
612 | ||
ed6ccf0f KW |
613 | uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
614 | uint64_t offset, | |
615 | int compressed_size) | |
45aba42f KW |
616 | { |
617 | BDRVQcowState *s = bs->opaque; | |
618 | int l2_index, ret; | |
3948d1d4 | 619 | uint64_t *l2_table; |
f4f0d391 | 620 | int64_t cluster_offset; |
45aba42f KW |
621 | int nb_csectors; |
622 | ||
3948d1d4 | 623 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
1e3e8f1a | 624 | if (ret < 0) { |
45aba42f | 625 | return 0; |
1e3e8f1a | 626 | } |
45aba42f | 627 | |
b0b6862e KW |
628 | /* Compression can't overwrite anything. Fail if the cluster was already |
629 | * allocated. */ | |
45aba42f | 630 | cluster_offset = be64_to_cpu(l2_table[l2_index]); |
b0b6862e | 631 | if (cluster_offset & L2E_OFFSET_MASK) { |
8f1efd00 KW |
632 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
633 | return 0; | |
634 | } | |
45aba42f | 635 | |
ed6ccf0f | 636 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
5d757b56 | 637 | if (cluster_offset < 0) { |
29c1a730 | 638 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
5d757b56 KW |
639 | return 0; |
640 | } | |
641 | ||
45aba42f KW |
642 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
643 | (cluster_offset >> 9); | |
644 | ||
645 | cluster_offset |= QCOW_OFLAG_COMPRESSED | | |
646 | ((uint64_t)nb_csectors << s->csize_shift); | |
647 | ||
648 | /* update L2 table */ | |
649 | ||
650 | /* compressed clusters never have the copied flag */ | |
651 | ||
66f82cee | 652 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
29c1a730 | 653 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
45aba42f | 654 | l2_table[l2_index] = cpu_to_be64(cluster_offset); |
29c1a730 | 655 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
79a31189 | 656 | if (ret < 0) { |
29c1a730 | 657 | return 0; |
4c1612d9 KW |
658 | } |
659 | ||
29c1a730 | 660 | return cluster_offset; |
4c1612d9 KW |
661 | } |
662 | ||
593fb83c KW |
663 | static int perform_cow(BlockDriverState *bs, QCowL2Meta *m, Qcow2COWRegion *r) |
664 | { | |
665 | BDRVQcowState *s = bs->opaque; | |
666 | int ret; | |
667 | ||
668 | if (r->nb_sectors == 0) { | |
669 | return 0; | |
670 | } | |
671 | ||
672 | qemu_co_mutex_unlock(&s->lock); | |
673 | ret = copy_sectors(bs, m->offset / BDRV_SECTOR_SIZE, m->alloc_offset, | |
674 | r->offset / BDRV_SECTOR_SIZE, | |
675 | r->offset / BDRV_SECTOR_SIZE + r->nb_sectors); | |
676 | qemu_co_mutex_lock(&s->lock); | |
677 | ||
678 | if (ret < 0) { | |
679 | return ret; | |
680 | } | |
681 | ||
682 | /* | |
683 | * Before we update the L2 table to actually point to the new cluster, we | |
684 | * need to be sure that the refcounts have been increased and COW was | |
685 | * handled. | |
686 | */ | |
687 | qcow2_cache_depends_on_flush(s->l2_table_cache); | |
688 | ||
689 | return 0; | |
690 | } | |
691 | ||
148da7ea | 692 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) |
45aba42f KW |
693 | { |
694 | BDRVQcowState *s = bs->opaque; | |
695 | int i, j = 0, l2_index, ret; | |
593fb83c | 696 | uint64_t *old_cluster, *l2_table; |
250196f1 | 697 | uint64_t cluster_offset = m->alloc_offset; |
45aba42f | 698 | |
3cce16f4 | 699 | trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
f50f88b9 | 700 | assert(m->nb_clusters > 0); |
45aba42f | 701 | |
7267c094 | 702 | old_cluster = g_malloc(m->nb_clusters * sizeof(uint64_t)); |
45aba42f KW |
703 | |
704 | /* copy content of unmodified sectors */ | |
593fb83c KW |
705 | ret = perform_cow(bs, m, &m->cow_start); |
706 | if (ret < 0) { | |
707 | goto err; | |
45aba42f KW |
708 | } |
709 | ||
593fb83c KW |
710 | ret = perform_cow(bs, m, &m->cow_end); |
711 | if (ret < 0) { | |
712 | goto err; | |
29c1a730 KW |
713 | } |
714 | ||
593fb83c | 715 | /* Update L2 table. */ |
74c4510a | 716 | if (s->use_lazy_refcounts) { |
280d3735 KW |
717 | qcow2_mark_dirty(bs); |
718 | } | |
bfe8043e SH |
719 | if (qcow2_need_accurate_refcounts(s)) { |
720 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
721 | s->refcount_block_cache); | |
722 | } | |
280d3735 | 723 | |
3948d1d4 | 724 | ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index); |
1e3e8f1a | 725 | if (ret < 0) { |
45aba42f | 726 | goto err; |
1e3e8f1a | 727 | } |
29c1a730 | 728 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
45aba42f | 729 | |
c01dbccb | 730 | assert(l2_index + m->nb_clusters <= s->l2_size); |
45aba42f KW |
731 | for (i = 0; i < m->nb_clusters; i++) { |
732 | /* if two concurrent writes happen to the same unallocated cluster | |
733 | * each write allocates separate cluster and writes data concurrently. | |
734 | * The first one to complete updates l2 table with pointer to its | |
735 | * cluster the second one has to do RMW (which is done above by | |
736 | * copy_sectors()), update l2 table with its cluster pointer and free | |
737 | * old cluster. This is what this loop does */ | |
738 | if(l2_table[l2_index + i] != 0) | |
739 | old_cluster[j++] = l2_table[l2_index + i]; | |
740 | ||
741 | l2_table[l2_index + i] = cpu_to_be64((cluster_offset + | |
742 | (i << s->cluster_bits)) | QCOW_OFLAG_COPIED); | |
743 | } | |
744 | ||
9f8e668e | 745 | |
29c1a730 | 746 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
c835d00f | 747 | if (ret < 0) { |
45aba42f | 748 | goto err; |
4c1612d9 | 749 | } |
45aba42f | 750 | |
7ec5e6a4 KW |
751 | /* |
752 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
753 | * Also flush bs->file to get the right order for L2 and refcount update. | |
6cfcb9b8 KW |
754 | * |
755 | * Don't discard clusters that reach a refcount of 0 (e.g. compressed | |
756 | * clusters), the next write will reuse them anyway. | |
7ec5e6a4 KW |
757 | */ |
758 | if (j != 0) { | |
7ec5e6a4 | 759 | for (i = 0; i < j; i++) { |
6cfcb9b8 KW |
760 | qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1, |
761 | QCOW2_DISCARD_NEVER); | |
7ec5e6a4 KW |
762 | } |
763 | } | |
45aba42f KW |
764 | |
765 | ret = 0; | |
766 | err: | |
7267c094 | 767 | g_free(old_cluster); |
45aba42f KW |
768 | return ret; |
769 | } | |
770 | ||
bf319ece KW |
771 | /* |
772 | * Returns the number of contiguous clusters that can be used for an allocating | |
773 | * write, but require COW to be performed (this includes yet unallocated space, | |
774 | * which must copy from the backing file) | |
775 | */ | |
776 | static int count_cow_clusters(BDRVQcowState *s, int nb_clusters, | |
777 | uint64_t *l2_table, int l2_index) | |
778 | { | |
143550a8 | 779 | int i; |
bf319ece | 780 | |
143550a8 KW |
781 | for (i = 0; i < nb_clusters; i++) { |
782 | uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]); | |
783 | int cluster_type = qcow2_get_cluster_type(l2_entry); | |
784 | ||
785 | switch(cluster_type) { | |
786 | case QCOW2_CLUSTER_NORMAL: | |
787 | if (l2_entry & QCOW_OFLAG_COPIED) { | |
788 | goto out; | |
789 | } | |
bf319ece | 790 | break; |
143550a8 KW |
791 | case QCOW2_CLUSTER_UNALLOCATED: |
792 | case QCOW2_CLUSTER_COMPRESSED: | |
6377af48 | 793 | case QCOW2_CLUSTER_ZERO: |
bf319ece | 794 | break; |
143550a8 KW |
795 | default: |
796 | abort(); | |
797 | } | |
bf319ece KW |
798 | } |
799 | ||
143550a8 | 800 | out: |
bf319ece KW |
801 | assert(i <= nb_clusters); |
802 | return i; | |
803 | } | |
804 | ||
250196f1 | 805 | /* |
226c3c26 KW |
806 | * Check if there already is an AIO write request in flight which allocates |
807 | * the same cluster. In this case we need to wait until the previous | |
808 | * request has completed and updated the L2 table accordingly. | |
65eb2e35 KW |
809 | * |
810 | * Returns: | |
811 | * 0 if there was no dependency. *cur_bytes indicates the number of | |
812 | * bytes from guest_offset that can be read before the next | |
813 | * dependency must be processed (or the request is complete) | |
814 | * | |
815 | * -EAGAIN if we had to wait for another request, previously gathered | |
816 | * information on cluster allocation may be invalid now. The caller | |
817 | * must start over anyway, so consider *cur_bytes undefined. | |
250196f1 | 818 | */ |
226c3c26 | 819 | static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, |
ecdd5333 | 820 | uint64_t *cur_bytes, QCowL2Meta **m) |
250196f1 KW |
821 | { |
822 | BDRVQcowState *s = bs->opaque; | |
250196f1 | 823 | QCowL2Meta *old_alloc; |
65eb2e35 | 824 | uint64_t bytes = *cur_bytes; |
250196f1 | 825 | |
250196f1 KW |
826 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
827 | ||
65eb2e35 KW |
828 | uint64_t start = guest_offset; |
829 | uint64_t end = start + bytes; | |
830 | uint64_t old_start = l2meta_cow_start(old_alloc); | |
831 | uint64_t old_end = l2meta_cow_end(old_alloc); | |
250196f1 | 832 | |
d9d74f41 | 833 | if (end <= old_start || start >= old_end) { |
250196f1 KW |
834 | /* No intersection */ |
835 | } else { | |
836 | if (start < old_start) { | |
837 | /* Stop at the start of a running allocation */ | |
65eb2e35 | 838 | bytes = old_start - start; |
250196f1 | 839 | } else { |
65eb2e35 | 840 | bytes = 0; |
250196f1 KW |
841 | } |
842 | ||
ecdd5333 KW |
843 | /* Stop if already an l2meta exists. After yielding, it wouldn't |
844 | * be valid any more, so we'd have to clean up the old L2Metas | |
845 | * and deal with requests depending on them before starting to | |
846 | * gather new ones. Not worth the trouble. */ | |
847 | if (bytes == 0 && *m) { | |
848 | *cur_bytes = 0; | |
849 | return 0; | |
850 | } | |
851 | ||
65eb2e35 | 852 | if (bytes == 0) { |
250196f1 KW |
853 | /* Wait for the dependency to complete. We need to recheck |
854 | * the free/allocated clusters when we continue. */ | |
855 | qemu_co_mutex_unlock(&s->lock); | |
856 | qemu_co_queue_wait(&old_alloc->dependent_requests); | |
857 | qemu_co_mutex_lock(&s->lock); | |
858 | return -EAGAIN; | |
859 | } | |
860 | } | |
861 | } | |
862 | ||
65eb2e35 KW |
863 | /* Make sure that existing clusters and new allocations are only used up to |
864 | * the next dependency if we shortened the request above */ | |
865 | *cur_bytes = bytes; | |
250196f1 | 866 | |
226c3c26 KW |
867 | return 0; |
868 | } | |
869 | ||
0af729ec KW |
870 | /* |
871 | * Checks how many already allocated clusters that don't require a copy on | |
872 | * write there are at the given guest_offset (up to *bytes). If | |
873 | * *host_offset is not zero, only physically contiguous clusters beginning at | |
874 | * this host offset are counted. | |
875 | * | |
411d62b0 KW |
876 | * Note that guest_offset may not be cluster aligned. In this case, the |
877 | * returned *host_offset points to exact byte referenced by guest_offset and | |
878 | * therefore isn't cluster aligned as well. | |
0af729ec KW |
879 | * |
880 | * Returns: | |
881 | * 0: if no allocated clusters are available at the given offset. | |
882 | * *bytes is normally unchanged. It is set to 0 if the cluster | |
883 | * is allocated and doesn't need COW, but doesn't have the right | |
884 | * physical offset. | |
885 | * | |
886 | * 1: if allocated clusters that don't require a COW are available at | |
887 | * the requested offset. *bytes may have decreased and describes | |
888 | * the length of the area that can be written to. | |
889 | * | |
890 | * -errno: in error cases | |
0af729ec KW |
891 | */ |
892 | static int handle_copied(BlockDriverState *bs, uint64_t guest_offset, | |
c53ede9f | 893 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
0af729ec KW |
894 | { |
895 | BDRVQcowState *s = bs->opaque; | |
896 | int l2_index; | |
897 | uint64_t cluster_offset; | |
898 | uint64_t *l2_table; | |
acb0467f | 899 | unsigned int nb_clusters; |
c53ede9f | 900 | unsigned int keep_clusters; |
0af729ec KW |
901 | int ret, pret; |
902 | ||
903 | trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset, | |
904 | *bytes); | |
0af729ec | 905 | |
411d62b0 KW |
906 | assert(*host_offset == 0 || offset_into_cluster(s, guest_offset) |
907 | == offset_into_cluster(s, *host_offset)); | |
908 | ||
acb0467f KW |
909 | /* |
910 | * Calculate the number of clusters to look for. We stop at L2 table | |
911 | * boundaries to keep things simple. | |
912 | */ | |
913 | nb_clusters = | |
914 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
915 | ||
916 | l2_index = offset_to_l2_index(s, guest_offset); | |
917 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
918 | ||
0af729ec KW |
919 | /* Find L2 entry for the first involved cluster */ |
920 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
921 | if (ret < 0) { | |
922 | return ret; | |
923 | } | |
924 | ||
925 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
926 | ||
927 | /* Check how many clusters are already allocated and don't need COW */ | |
928 | if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL | |
929 | && (cluster_offset & QCOW_OFLAG_COPIED)) | |
930 | { | |
e62daaf6 KW |
931 | /* If a specific host_offset is required, check it */ |
932 | bool offset_matches = | |
933 | (cluster_offset & L2E_OFFSET_MASK) == *host_offset; | |
934 | ||
935 | if (*host_offset != 0 && !offset_matches) { | |
936 | *bytes = 0; | |
937 | ret = 0; | |
938 | goto out; | |
939 | } | |
940 | ||
0af729ec | 941 | /* We keep all QCOW_OFLAG_COPIED clusters */ |
c53ede9f | 942 | keep_clusters = |
acb0467f | 943 | count_contiguous_clusters(nb_clusters, s->cluster_size, |
61653008 | 944 | &l2_table[l2_index], |
0af729ec | 945 | QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); |
c53ede9f KW |
946 | assert(keep_clusters <= nb_clusters); |
947 | ||
948 | *bytes = MIN(*bytes, | |
949 | keep_clusters * s->cluster_size | |
950 | - offset_into_cluster(s, guest_offset)); | |
0af729ec KW |
951 | |
952 | ret = 1; | |
953 | } else { | |
0af729ec KW |
954 | ret = 0; |
955 | } | |
956 | ||
0af729ec | 957 | /* Cleanup */ |
e62daaf6 | 958 | out: |
0af729ec KW |
959 | pret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
960 | if (pret < 0) { | |
961 | return pret; | |
962 | } | |
963 | ||
e62daaf6 KW |
964 | /* Only return a host offset if we actually made progress. Otherwise we |
965 | * would make requirements for handle_alloc() that it can't fulfill */ | |
966 | if (ret) { | |
411d62b0 KW |
967 | *host_offset = (cluster_offset & L2E_OFFSET_MASK) |
968 | + offset_into_cluster(s, guest_offset); | |
e62daaf6 KW |
969 | } |
970 | ||
0af729ec KW |
971 | return ret; |
972 | } | |
973 | ||
226c3c26 KW |
974 | /* |
975 | * Allocates new clusters for the given guest_offset. | |
976 | * | |
977 | * At most *nb_clusters are allocated, and on return *nb_clusters is updated to | |
978 | * contain the number of clusters that have been allocated and are contiguous | |
979 | * in the image file. | |
980 | * | |
981 | * If *host_offset is non-zero, it specifies the offset in the image file at | |
982 | * which the new clusters must start. *nb_clusters can be 0 on return in this | |
983 | * case if the cluster at host_offset is already in use. If *host_offset is | |
984 | * zero, the clusters can be allocated anywhere in the image file. | |
985 | * | |
986 | * *host_offset is updated to contain the offset into the image file at which | |
987 | * the first allocated cluster starts. | |
988 | * | |
989 | * Return 0 on success and -errno in error cases. -EAGAIN means that the | |
990 | * function has been waiting for another request and the allocation must be | |
991 | * restarted, but the whole request should not be failed. | |
992 | */ | |
993 | static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, | |
994 | uint64_t *host_offset, unsigned int *nb_clusters) | |
995 | { | |
996 | BDRVQcowState *s = bs->opaque; | |
226c3c26 KW |
997 | |
998 | trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, | |
999 | *host_offset, *nb_clusters); | |
1000 | ||
250196f1 KW |
1001 | /* Allocate new clusters */ |
1002 | trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); | |
1003 | if (*host_offset == 0) { | |
df021791 KW |
1004 | int64_t cluster_offset = |
1005 | qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); | |
1006 | if (cluster_offset < 0) { | |
1007 | return cluster_offset; | |
1008 | } | |
1009 | *host_offset = cluster_offset; | |
1010 | return 0; | |
250196f1 | 1011 | } else { |
17a71e58 | 1012 | int ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); |
df021791 KW |
1013 | if (ret < 0) { |
1014 | return ret; | |
1015 | } | |
1016 | *nb_clusters = ret; | |
1017 | return 0; | |
250196f1 | 1018 | } |
250196f1 KW |
1019 | } |
1020 | ||
10f0ed8b KW |
1021 | /* |
1022 | * Allocates new clusters for an area that either is yet unallocated or needs a | |
1023 | * copy on write. If *host_offset is non-zero, clusters are only allocated if | |
1024 | * the new allocation can match the specified host offset. | |
1025 | * | |
411d62b0 KW |
1026 | * Note that guest_offset may not be cluster aligned. In this case, the |
1027 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1028 | * therefore isn't cluster aligned as well. | |
10f0ed8b KW |
1029 | * |
1030 | * Returns: | |
1031 | * 0: if no clusters could be allocated. *bytes is set to 0, | |
1032 | * *host_offset is left unchanged. | |
1033 | * | |
1034 | * 1: if new clusters were allocated. *bytes may be decreased if the | |
1035 | * new allocation doesn't cover all of the requested area. | |
1036 | * *host_offset is updated to contain the host offset of the first | |
1037 | * newly allocated cluster. | |
1038 | * | |
1039 | * -errno: in error cases | |
10f0ed8b KW |
1040 | */ |
1041 | static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, | |
c37f4cd7 | 1042 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
10f0ed8b KW |
1043 | { |
1044 | BDRVQcowState *s = bs->opaque; | |
1045 | int l2_index; | |
1046 | uint64_t *l2_table; | |
1047 | uint64_t entry; | |
f5bc6350 | 1048 | unsigned int nb_clusters; |
10f0ed8b KW |
1049 | int ret; |
1050 | ||
10f0ed8b | 1051 | uint64_t alloc_cluster_offset; |
10f0ed8b KW |
1052 | |
1053 | trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset, | |
1054 | *bytes); | |
1055 | assert(*bytes > 0); | |
1056 | ||
f5bc6350 KW |
1057 | /* |
1058 | * Calculate the number of clusters to look for. We stop at L2 table | |
1059 | * boundaries to keep things simple. | |
1060 | */ | |
c37f4cd7 KW |
1061 | nb_clusters = |
1062 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1063 | ||
f5bc6350 | 1064 | l2_index = offset_to_l2_index(s, guest_offset); |
c37f4cd7 | 1065 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
f5bc6350 | 1066 | |
10f0ed8b KW |
1067 | /* Find L2 entry for the first involved cluster */ |
1068 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
1069 | if (ret < 0) { | |
1070 | return ret; | |
1071 | } | |
1072 | ||
3b8e2e26 | 1073 | entry = be64_to_cpu(l2_table[l2_index]); |
10f0ed8b KW |
1074 | |
1075 | /* For the moment, overwrite compressed clusters one by one */ | |
1076 | if (entry & QCOW_OFLAG_COMPRESSED) { | |
1077 | nb_clusters = 1; | |
1078 | } else { | |
3b8e2e26 | 1079 | nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index); |
10f0ed8b KW |
1080 | } |
1081 | ||
ecdd5333 KW |
1082 | /* This function is only called when there were no non-COW clusters, so if |
1083 | * we can't find any unallocated or COW clusters either, something is | |
1084 | * wrong with our code. */ | |
1085 | assert(nb_clusters > 0); | |
1086 | ||
10f0ed8b KW |
1087 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
1088 | if (ret < 0) { | |
1089 | return ret; | |
1090 | } | |
1091 | ||
10f0ed8b | 1092 | /* Allocate, if necessary at a given offset in the image file */ |
411d62b0 | 1093 | alloc_cluster_offset = start_of_cluster(s, *host_offset); |
83baa9a4 | 1094 | ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset, |
10f0ed8b KW |
1095 | &nb_clusters); |
1096 | if (ret < 0) { | |
1097 | goto fail; | |
1098 | } | |
1099 | ||
83baa9a4 KW |
1100 | /* Can't extend contiguous allocation */ |
1101 | if (nb_clusters == 0) { | |
10f0ed8b KW |
1102 | *bytes = 0; |
1103 | return 0; | |
1104 | } | |
1105 | ||
83baa9a4 KW |
1106 | /* |
1107 | * Save info needed for meta data update. | |
1108 | * | |
1109 | * requested_sectors: Number of sectors from the start of the first | |
1110 | * newly allocated cluster to the end of the (possibly shortened | |
1111 | * before) write request. | |
1112 | * | |
1113 | * avail_sectors: Number of sectors from the start of the first | |
1114 | * newly allocated to the end of the last newly allocated cluster. | |
1115 | * | |
1116 | * nb_sectors: The number of sectors from the start of the first | |
1117 | * newly allocated cluster to the end of the area that the write | |
1118 | * request actually writes to (excluding COW at the end) | |
1119 | */ | |
1120 | int requested_sectors = | |
1121 | (*bytes + offset_into_cluster(s, guest_offset)) | |
1122 | >> BDRV_SECTOR_BITS; | |
1123 | int avail_sectors = nb_clusters | |
1124 | << (s->cluster_bits - BDRV_SECTOR_BITS); | |
1125 | int alloc_n_start = offset_into_cluster(s, guest_offset) | |
1126 | >> BDRV_SECTOR_BITS; | |
1127 | int nb_sectors = MIN(requested_sectors, avail_sectors); | |
88c6588c | 1128 | QCowL2Meta *old_m = *m; |
83baa9a4 | 1129 | |
83baa9a4 KW |
1130 | *m = g_malloc0(sizeof(**m)); |
1131 | ||
1132 | **m = (QCowL2Meta) { | |
88c6588c KW |
1133 | .next = old_m, |
1134 | ||
411d62b0 | 1135 | .alloc_offset = alloc_cluster_offset, |
83baa9a4 KW |
1136 | .offset = start_of_cluster(s, guest_offset), |
1137 | .nb_clusters = nb_clusters, | |
1138 | .nb_available = nb_sectors, | |
1139 | ||
1140 | .cow_start = { | |
1141 | .offset = 0, | |
1142 | .nb_sectors = alloc_n_start, | |
1143 | }, | |
1144 | .cow_end = { | |
1145 | .offset = nb_sectors * BDRV_SECTOR_SIZE, | |
1146 | .nb_sectors = avail_sectors - nb_sectors, | |
1147 | }, | |
1148 | }; | |
1149 | qemu_co_queue_init(&(*m)->dependent_requests); | |
1150 | QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); | |
1151 | ||
411d62b0 | 1152 | *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset); |
83baa9a4 KW |
1153 | *bytes = MIN(*bytes, (nb_sectors * BDRV_SECTOR_SIZE) |
1154 | - offset_into_cluster(s, guest_offset)); | |
1155 | assert(*bytes != 0); | |
1156 | ||
10f0ed8b KW |
1157 | return 1; |
1158 | ||
1159 | fail: | |
1160 | if (*m && (*m)->nb_clusters > 0) { | |
1161 | QLIST_REMOVE(*m, next_in_flight); | |
1162 | } | |
1163 | return ret; | |
1164 | } | |
1165 | ||
45aba42f KW |
1166 | /* |
1167 | * alloc_cluster_offset | |
1168 | * | |
250196f1 KW |
1169 | * For a given offset on the virtual disk, find the cluster offset in qcow2 |
1170 | * file. If the offset is not found, allocate a new cluster. | |
45aba42f | 1171 | * |
250196f1 | 1172 | * If the cluster was already allocated, m->nb_clusters is set to 0 and |
a7912369 | 1173 | * other fields in m are meaningless. |
148da7ea KW |
1174 | * |
1175 | * If the cluster is newly allocated, m->nb_clusters is set to the number of | |
68d100e9 KW |
1176 | * contiguous clusters that have been allocated. In this case, the other |
1177 | * fields of m are valid and contain information about the first allocated | |
1178 | * cluster. | |
45aba42f | 1179 | * |
68d100e9 KW |
1180 | * If the request conflicts with another write request in flight, the coroutine |
1181 | * is queued and will be reentered when the dependency has completed. | |
148da7ea KW |
1182 | * |
1183 | * Return 0 on success and -errno in error cases | |
45aba42f | 1184 | */ |
f4f0d391 | 1185 | int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, |
f50f88b9 | 1186 | int n_start, int n_end, int *num, uint64_t *host_offset, QCowL2Meta **m) |
45aba42f KW |
1187 | { |
1188 | BDRVQcowState *s = bs->opaque; | |
710c2496 | 1189 | uint64_t start, remaining; |
250196f1 | 1190 | uint64_t cluster_offset; |
65eb2e35 | 1191 | uint64_t cur_bytes; |
710c2496 | 1192 | int ret; |
45aba42f | 1193 | |
3cce16f4 KW |
1194 | trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, |
1195 | n_start, n_end); | |
1196 | ||
710c2496 KW |
1197 | assert(n_start * BDRV_SECTOR_SIZE == offset_into_cluster(s, offset)); |
1198 | offset = start_of_cluster(s, offset); | |
1199 | ||
72424114 | 1200 | again: |
710c2496 KW |
1201 | start = offset + (n_start << BDRV_SECTOR_BITS); |
1202 | remaining = (n_end - n_start) << BDRV_SECTOR_BITS; | |
0af729ec KW |
1203 | cluster_offset = 0; |
1204 | *host_offset = 0; | |
ecdd5333 KW |
1205 | cur_bytes = 0; |
1206 | *m = NULL; | |
0af729ec | 1207 | |
2c3b32d2 | 1208 | while (true) { |
ecdd5333 KW |
1209 | |
1210 | if (!*host_offset) { | |
1211 | *host_offset = start_of_cluster(s, cluster_offset); | |
1212 | } | |
1213 | ||
1214 | assert(remaining >= cur_bytes); | |
1215 | ||
1216 | start += cur_bytes; | |
1217 | remaining -= cur_bytes; | |
1218 | cluster_offset += cur_bytes; | |
1219 | ||
1220 | if (remaining == 0) { | |
1221 | break; | |
1222 | } | |
1223 | ||
1224 | cur_bytes = remaining; | |
1225 | ||
2c3b32d2 KW |
1226 | /* |
1227 | * Now start gathering as many contiguous clusters as possible: | |
1228 | * | |
1229 | * 1. Check for overlaps with in-flight allocations | |
1230 | * | |
1231 | * a) Overlap not in the first cluster -> shorten this request and | |
1232 | * let the caller handle the rest in its next loop iteration. | |
1233 | * | |
1234 | * b) Real overlaps of two requests. Yield and restart the search | |
1235 | * for contiguous clusters (the situation could have changed | |
1236 | * while we were sleeping) | |
1237 | * | |
1238 | * c) TODO: Request starts in the same cluster as the in-flight | |
1239 | * allocation ends. Shorten the COW of the in-fight allocation, | |
1240 | * set cluster_offset to write to the same cluster and set up | |
1241 | * the right synchronisation between the in-flight request and | |
1242 | * the new one. | |
1243 | */ | |
ecdd5333 | 1244 | ret = handle_dependencies(bs, start, &cur_bytes, m); |
2c3b32d2 | 1245 | if (ret == -EAGAIN) { |
ecdd5333 KW |
1246 | /* Currently handle_dependencies() doesn't yield if we already had |
1247 | * an allocation. If it did, we would have to clean up the L2Meta | |
1248 | * structs before starting over. */ | |
1249 | assert(*m == NULL); | |
2c3b32d2 KW |
1250 | goto again; |
1251 | } else if (ret < 0) { | |
1252 | return ret; | |
ecdd5333 KW |
1253 | } else if (cur_bytes == 0) { |
1254 | break; | |
2c3b32d2 KW |
1255 | } else { |
1256 | /* handle_dependencies() may have decreased cur_bytes (shortened | |
1257 | * the allocations below) so that the next dependency is processed | |
1258 | * correctly during the next loop iteration. */ | |
0af729ec | 1259 | } |
710c2496 | 1260 | |
2c3b32d2 KW |
1261 | /* |
1262 | * 2. Count contiguous COPIED clusters. | |
1263 | */ | |
1264 | ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m); | |
1265 | if (ret < 0) { | |
1266 | return ret; | |
1267 | } else if (ret) { | |
ecdd5333 | 1268 | continue; |
2c3b32d2 KW |
1269 | } else if (cur_bytes == 0) { |
1270 | break; | |
1271 | } | |
060bee89 | 1272 | |
2c3b32d2 KW |
1273 | /* |
1274 | * 3. If the request still hasn't completed, allocate new clusters, | |
1275 | * considering any cluster_offset of steps 1c or 2. | |
1276 | */ | |
1277 | ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m); | |
1278 | if (ret < 0) { | |
1279 | return ret; | |
1280 | } else if (ret) { | |
ecdd5333 | 1281 | continue; |
2c3b32d2 KW |
1282 | } else { |
1283 | assert(cur_bytes == 0); | |
1284 | break; | |
1285 | } | |
f5bc6350 | 1286 | } |
10f0ed8b | 1287 | |
710c2496 KW |
1288 | *num = (n_end - n_start) - (remaining >> BDRV_SECTOR_BITS); |
1289 | assert(*num > 0); | |
1290 | assert(*host_offset != 0); | |
45aba42f | 1291 | |
148da7ea | 1292 | return 0; |
45aba42f KW |
1293 | } |
1294 | ||
1295 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
1296 | const uint8_t *buf, int buf_size) | |
1297 | { | |
1298 | z_stream strm1, *strm = &strm1; | |
1299 | int ret, out_len; | |
1300 | ||
1301 | memset(strm, 0, sizeof(*strm)); | |
1302 | ||
1303 | strm->next_in = (uint8_t *)buf; | |
1304 | strm->avail_in = buf_size; | |
1305 | strm->next_out = out_buf; | |
1306 | strm->avail_out = out_buf_size; | |
1307 | ||
1308 | ret = inflateInit2(strm, -12); | |
1309 | if (ret != Z_OK) | |
1310 | return -1; | |
1311 | ret = inflate(strm, Z_FINISH); | |
1312 | out_len = strm->next_out - out_buf; | |
1313 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
1314 | out_len != out_buf_size) { | |
1315 | inflateEnd(strm); | |
1316 | return -1; | |
1317 | } | |
1318 | inflateEnd(strm); | |
1319 | return 0; | |
1320 | } | |
1321 | ||
66f82cee | 1322 | int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) |
45aba42f | 1323 | { |
66f82cee | 1324 | BDRVQcowState *s = bs->opaque; |
45aba42f KW |
1325 | int ret, csize, nb_csectors, sector_offset; |
1326 | uint64_t coffset; | |
1327 | ||
1328 | coffset = cluster_offset & s->cluster_offset_mask; | |
1329 | if (s->cluster_cache_offset != coffset) { | |
1330 | nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1; | |
1331 | sector_offset = coffset & 511; | |
1332 | csize = nb_csectors * 512 - sector_offset; | |
66f82cee KW |
1333 | BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
1334 | ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, nb_csectors); | |
45aba42f | 1335 | if (ret < 0) { |
8af36488 | 1336 | return ret; |
45aba42f KW |
1337 | } |
1338 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
1339 | s->cluster_data + sector_offset, csize) < 0) { | |
8af36488 | 1340 | return -EIO; |
45aba42f KW |
1341 | } |
1342 | s->cluster_cache_offset = coffset; | |
1343 | } | |
1344 | return 0; | |
1345 | } | |
5ea929e3 KW |
1346 | |
1347 | /* | |
1348 | * This discards as many clusters of nb_clusters as possible at once (i.e. | |
1349 | * all clusters in the same L2 table) and returns the number of discarded | |
1350 | * clusters. | |
1351 | */ | |
1352 | static int discard_single_l2(BlockDriverState *bs, uint64_t offset, | |
670df5e3 | 1353 | unsigned int nb_clusters, enum qcow2_discard_type type) |
5ea929e3 KW |
1354 | { |
1355 | BDRVQcowState *s = bs->opaque; | |
3948d1d4 | 1356 | uint64_t *l2_table; |
5ea929e3 KW |
1357 | int l2_index; |
1358 | int ret; | |
1359 | int i; | |
1360 | ||
3948d1d4 | 1361 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
5ea929e3 KW |
1362 | if (ret < 0) { |
1363 | return ret; | |
1364 | } | |
1365 | ||
1366 | /* Limit nb_clusters to one L2 table */ | |
1367 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
1368 | ||
1369 | for (i = 0; i < nb_clusters; i++) { | |
1370 | uint64_t old_offset; | |
1371 | ||
1372 | old_offset = be64_to_cpu(l2_table[l2_index + i]); | |
8e37f681 | 1373 | if ((old_offset & L2E_OFFSET_MASK) == 0) { |
5ea929e3 KW |
1374 | continue; |
1375 | } | |
1376 | ||
1377 | /* First remove L2 entries */ | |
1378 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); | |
1379 | l2_table[l2_index + i] = cpu_to_be64(0); | |
1380 | ||
1381 | /* Then decrease the refcount */ | |
670df5e3 | 1382 | qcow2_free_any_clusters(bs, old_offset, 1, type); |
5ea929e3 KW |
1383 | } |
1384 | ||
1385 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); | |
1386 | if (ret < 0) { | |
1387 | return ret; | |
1388 | } | |
1389 | ||
1390 | return nb_clusters; | |
1391 | } | |
1392 | ||
1393 | int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset, | |
670df5e3 | 1394 | int nb_sectors, enum qcow2_discard_type type) |
5ea929e3 KW |
1395 | { |
1396 | BDRVQcowState *s = bs->opaque; | |
1397 | uint64_t end_offset; | |
1398 | unsigned int nb_clusters; | |
1399 | int ret; | |
1400 | ||
1401 | end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS); | |
1402 | ||
1403 | /* Round start up and end down */ | |
1404 | offset = align_offset(offset, s->cluster_size); | |
1405 | end_offset &= ~(s->cluster_size - 1); | |
1406 | ||
1407 | if (offset > end_offset) { | |
1408 | return 0; | |
1409 | } | |
1410 | ||
1411 | nb_clusters = size_to_clusters(s, end_offset - offset); | |
1412 | ||
0b919fae KW |
1413 | s->cache_discards = true; |
1414 | ||
5ea929e3 KW |
1415 | /* Each L2 table is handled by its own loop iteration */ |
1416 | while (nb_clusters > 0) { | |
670df5e3 | 1417 | ret = discard_single_l2(bs, offset, nb_clusters, type); |
5ea929e3 | 1418 | if (ret < 0) { |
0b919fae | 1419 | goto fail; |
5ea929e3 KW |
1420 | } |
1421 | ||
1422 | nb_clusters -= ret; | |
1423 | offset += (ret * s->cluster_size); | |
1424 | } | |
1425 | ||
0b919fae KW |
1426 | ret = 0; |
1427 | fail: | |
1428 | s->cache_discards = false; | |
1429 | qcow2_process_discards(bs, ret); | |
1430 | ||
1431 | return ret; | |
5ea929e3 | 1432 | } |
621f0589 KW |
1433 | |
1434 | /* | |
1435 | * This zeroes as many clusters of nb_clusters as possible at once (i.e. | |
1436 | * all clusters in the same L2 table) and returns the number of zeroed | |
1437 | * clusters. | |
1438 | */ | |
1439 | static int zero_single_l2(BlockDriverState *bs, uint64_t offset, | |
1440 | unsigned int nb_clusters) | |
1441 | { | |
1442 | BDRVQcowState *s = bs->opaque; | |
1443 | uint64_t *l2_table; | |
1444 | int l2_index; | |
1445 | int ret; | |
1446 | int i; | |
1447 | ||
1448 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); | |
1449 | if (ret < 0) { | |
1450 | return ret; | |
1451 | } | |
1452 | ||
1453 | /* Limit nb_clusters to one L2 table */ | |
1454 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
1455 | ||
1456 | for (i = 0; i < nb_clusters; i++) { | |
1457 | uint64_t old_offset; | |
1458 | ||
1459 | old_offset = be64_to_cpu(l2_table[l2_index + i]); | |
1460 | ||
1461 | /* Update L2 entries */ | |
1462 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); | |
1463 | if (old_offset & QCOW_OFLAG_COMPRESSED) { | |
1464 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); | |
6cfcb9b8 | 1465 | qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST); |
621f0589 KW |
1466 | } else { |
1467 | l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO); | |
1468 | } | |
1469 | } | |
1470 | ||
1471 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); | |
1472 | if (ret < 0) { | |
1473 | return ret; | |
1474 | } | |
1475 | ||
1476 | return nb_clusters; | |
1477 | } | |
1478 | ||
1479 | int qcow2_zero_clusters(BlockDriverState *bs, uint64_t offset, int nb_sectors) | |
1480 | { | |
1481 | BDRVQcowState *s = bs->opaque; | |
1482 | unsigned int nb_clusters; | |
1483 | int ret; | |
1484 | ||
1485 | /* The zero flag is only supported by version 3 and newer */ | |
1486 | if (s->qcow_version < 3) { | |
1487 | return -ENOTSUP; | |
1488 | } | |
1489 | ||
1490 | /* Each L2 table is handled by its own loop iteration */ | |
1491 | nb_clusters = size_to_clusters(s, nb_sectors << BDRV_SECTOR_BITS); | |
1492 | ||
0b919fae KW |
1493 | s->cache_discards = true; |
1494 | ||
621f0589 KW |
1495 | while (nb_clusters > 0) { |
1496 | ret = zero_single_l2(bs, offset, nb_clusters); | |
1497 | if (ret < 0) { | |
0b919fae | 1498 | goto fail; |
621f0589 KW |
1499 | } |
1500 | ||
1501 | nb_clusters -= ret; | |
1502 | offset += (ret * s->cluster_size); | |
1503 | } | |
1504 | ||
0b919fae KW |
1505 | ret = 0; |
1506 | fail: | |
1507 | s->cache_discards = false; | |
1508 | qcow2_process_discards(bs, ret); | |
1509 | ||
1510 | return ret; | |
621f0589 | 1511 | } |
32b6444d MR |
1512 | |
1513 | /* | |
1514 | * Expands all zero clusters in a specific L1 table (or deallocates them, for | |
1515 | * non-backed non-pre-allocated zero clusters). | |
1516 | * | |
1517 | * expanded_clusters is a bitmap where every bit corresponds to one cluster in | |
1518 | * the image file; a bit gets set if the corresponding cluster has been used for | |
1519 | * zero expansion (i.e., has been filled with zeroes and is referenced from an | |
1520 | * L2 table). nb_clusters contains the total cluster count of the image file, | |
1521 | * i.e., the number of bits in expanded_clusters. | |
1522 | */ | |
1523 | static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table, | |
e390cf5a MR |
1524 | int l1_size, uint8_t **expanded_clusters, |
1525 | uint64_t *nb_clusters) | |
32b6444d MR |
1526 | { |
1527 | BDRVQcowState *s = bs->opaque; | |
1528 | bool is_active_l1 = (l1_table == s->l1_table); | |
1529 | uint64_t *l2_table = NULL; | |
1530 | int ret; | |
1531 | int i, j; | |
1532 | ||
1533 | if (!is_active_l1) { | |
1534 | /* inactive L2 tables require a buffer to be stored in when loading | |
1535 | * them from disk */ | |
1536 | l2_table = qemu_blockalign(bs, s->cluster_size); | |
1537 | } | |
1538 | ||
1539 | for (i = 0; i < l1_size; i++) { | |
1540 | uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK; | |
1541 | bool l2_dirty = false; | |
1542 | ||
1543 | if (!l2_offset) { | |
1544 | /* unallocated */ | |
1545 | continue; | |
1546 | } | |
1547 | ||
1548 | if (is_active_l1) { | |
1549 | /* get active L2 tables from cache */ | |
1550 | ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, | |
1551 | (void **)&l2_table); | |
1552 | } else { | |
1553 | /* load inactive L2 tables from disk */ | |
1554 | ret = bdrv_read(bs->file, l2_offset / BDRV_SECTOR_SIZE, | |
1555 | (void *)l2_table, s->cluster_sectors); | |
1556 | } | |
1557 | if (ret < 0) { | |
1558 | goto fail; | |
1559 | } | |
1560 | ||
1561 | for (j = 0; j < s->l2_size; j++) { | |
1562 | uint64_t l2_entry = be64_to_cpu(l2_table[j]); | |
1563 | int64_t offset = l2_entry & L2E_OFFSET_MASK, cluster_index; | |
1564 | int cluster_type = qcow2_get_cluster_type(l2_entry); | |
320c7066 | 1565 | bool preallocated = offset != 0; |
32b6444d MR |
1566 | |
1567 | if (cluster_type == QCOW2_CLUSTER_NORMAL) { | |
1568 | cluster_index = offset >> s->cluster_bits; | |
e390cf5a MR |
1569 | assert((cluster_index >= 0) && (cluster_index < *nb_clusters)); |
1570 | if ((*expanded_clusters)[cluster_index / 8] & | |
32b6444d MR |
1571 | (1 << (cluster_index % 8))) { |
1572 | /* Probably a shared L2 table; this cluster was a zero | |
1573 | * cluster which has been expanded, its refcount | |
1574 | * therefore most likely requires an update. */ | |
1575 | ret = qcow2_update_cluster_refcount(bs, cluster_index, 1, | |
1576 | QCOW2_DISCARD_NEVER); | |
1577 | if (ret < 0) { | |
1578 | goto fail; | |
1579 | } | |
1580 | /* Since we just increased the refcount, the COPIED flag may | |
1581 | * no longer be set. */ | |
1582 | l2_table[j] = cpu_to_be64(l2_entry & ~QCOW_OFLAG_COPIED); | |
1583 | l2_dirty = true; | |
1584 | } | |
1585 | continue; | |
1586 | } | |
1587 | else if (qcow2_get_cluster_type(l2_entry) != QCOW2_CLUSTER_ZERO) { | |
1588 | continue; | |
1589 | } | |
1590 | ||
320c7066 | 1591 | if (!preallocated) { |
32b6444d MR |
1592 | if (!bs->backing_hd) { |
1593 | /* not backed; therefore we can simply deallocate the | |
1594 | * cluster */ | |
1595 | l2_table[j] = 0; | |
1596 | l2_dirty = true; | |
1597 | continue; | |
1598 | } | |
1599 | ||
1600 | offset = qcow2_alloc_clusters(bs, s->cluster_size); | |
1601 | if (offset < 0) { | |
1602 | ret = offset; | |
1603 | goto fail; | |
1604 | } | |
1605 | } | |
1606 | ||
1607 | ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_DEFAULT, | |
1608 | offset, s->cluster_size); | |
1609 | if (ret < 0) { | |
320c7066 MR |
1610 | if (!preallocated) { |
1611 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1612 | QCOW2_DISCARD_ALWAYS); | |
1613 | } | |
32b6444d MR |
1614 | goto fail; |
1615 | } | |
1616 | ||
1617 | ret = bdrv_write_zeroes(bs->file, offset / BDRV_SECTOR_SIZE, | |
1618 | s->cluster_sectors); | |
1619 | if (ret < 0) { | |
320c7066 MR |
1620 | if (!preallocated) { |
1621 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1622 | QCOW2_DISCARD_ALWAYS); | |
1623 | } | |
32b6444d MR |
1624 | goto fail; |
1625 | } | |
1626 | ||
1627 | l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED); | |
1628 | l2_dirty = true; | |
1629 | ||
1630 | cluster_index = offset >> s->cluster_bits; | |
e390cf5a MR |
1631 | |
1632 | if (cluster_index >= *nb_clusters) { | |
1633 | uint64_t old_bitmap_size = (*nb_clusters + 7) / 8; | |
1634 | uint64_t new_bitmap_size; | |
1635 | /* The offset may lie beyond the old end of the underlying image | |
1636 | * file for growable files only */ | |
1637 | assert(bs->file->growable); | |
1638 | *nb_clusters = size_to_clusters(s, bs->file->total_sectors * | |
1639 | BDRV_SECTOR_SIZE); | |
1640 | new_bitmap_size = (*nb_clusters + 7) / 8; | |
1641 | *expanded_clusters = g_realloc(*expanded_clusters, | |
1642 | new_bitmap_size); | |
1643 | /* clear the newly allocated space */ | |
1644 | memset(&(*expanded_clusters)[old_bitmap_size], 0, | |
1645 | new_bitmap_size - old_bitmap_size); | |
1646 | } | |
1647 | ||
1648 | assert((cluster_index >= 0) && (cluster_index < *nb_clusters)); | |
1649 | (*expanded_clusters)[cluster_index / 8] |= 1 << (cluster_index % 8); | |
32b6444d MR |
1650 | } |
1651 | ||
1652 | if (is_active_l1) { | |
1653 | if (l2_dirty) { | |
1654 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); | |
1655 | qcow2_cache_depends_on_flush(s->l2_table_cache); | |
1656 | } | |
1657 | ret = qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table); | |
1658 | if (ret < 0) { | |
1659 | l2_table = NULL; | |
1660 | goto fail; | |
1661 | } | |
1662 | } else { | |
1663 | if (l2_dirty) { | |
1664 | ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_DEFAULT & | |
1665 | ~(QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2), l2_offset, | |
1666 | s->cluster_size); | |
1667 | if (ret < 0) { | |
1668 | goto fail; | |
1669 | } | |
1670 | ||
1671 | ret = bdrv_write(bs->file, l2_offset / BDRV_SECTOR_SIZE, | |
1672 | (void *)l2_table, s->cluster_sectors); | |
1673 | if (ret < 0) { | |
1674 | goto fail; | |
1675 | } | |
1676 | } | |
1677 | } | |
1678 | } | |
1679 | ||
1680 | ret = 0; | |
1681 | ||
1682 | fail: | |
1683 | if (l2_table) { | |
1684 | if (!is_active_l1) { | |
1685 | qemu_vfree(l2_table); | |
1686 | } else { | |
1687 | if (ret < 0) { | |
1688 | qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table); | |
1689 | } else { | |
1690 | ret = qcow2_cache_put(bs, s->l2_table_cache, | |
1691 | (void **)&l2_table); | |
1692 | } | |
1693 | } | |
1694 | } | |
1695 | return ret; | |
1696 | } | |
1697 | ||
1698 | /* | |
1699 | * For backed images, expands all zero clusters on the image. For non-backed | |
1700 | * images, deallocates all non-pre-allocated zero clusters (and claims the | |
1701 | * allocation for pre-allocated ones). This is important for downgrading to a | |
1702 | * qcow2 version which doesn't yet support metadata zero clusters. | |
1703 | */ | |
1704 | int qcow2_expand_zero_clusters(BlockDriverState *bs) | |
1705 | { | |
1706 | BDRVQcowState *s = bs->opaque; | |
1707 | uint64_t *l1_table = NULL; | |
32b6444d MR |
1708 | uint64_t nb_clusters; |
1709 | uint8_t *expanded_clusters; | |
1710 | int ret; | |
1711 | int i, j; | |
1712 | ||
e390cf5a MR |
1713 | nb_clusters = size_to_clusters(s, bs->file->total_sectors * |
1714 | BDRV_SECTOR_SIZE); | |
32b6444d MR |
1715 | expanded_clusters = g_malloc0((nb_clusters + 7) / 8); |
1716 | ||
1717 | ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size, | |
e390cf5a | 1718 | &expanded_clusters, &nb_clusters); |
32b6444d MR |
1719 | if (ret < 0) { |
1720 | goto fail; | |
1721 | } | |
1722 | ||
1723 | /* Inactive L1 tables may point to active L2 tables - therefore it is | |
1724 | * necessary to flush the L2 table cache before trying to access the L2 | |
1725 | * tables pointed to by inactive L1 entries (else we might try to expand | |
1726 | * zero clusters that have already been expanded); furthermore, it is also | |
1727 | * necessary to empty the L2 table cache, since it may contain tables which | |
1728 | * are now going to be modified directly on disk, bypassing the cache. | |
1729 | * qcow2_cache_empty() does both for us. */ | |
1730 | ret = qcow2_cache_empty(bs, s->l2_table_cache); | |
1731 | if (ret < 0) { | |
1732 | goto fail; | |
1733 | } | |
1734 | ||
1735 | for (i = 0; i < s->nb_snapshots; i++) { | |
1736 | int l1_sectors = (s->snapshots[i].l1_size * sizeof(uint64_t) + | |
1737 | BDRV_SECTOR_SIZE - 1) / BDRV_SECTOR_SIZE; | |
1738 | ||
1739 | l1_table = g_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE); | |
1740 | ||
1741 | ret = bdrv_read(bs->file, s->snapshots[i].l1_table_offset / | |
1742 | BDRV_SECTOR_SIZE, (void *)l1_table, l1_sectors); | |
1743 | if (ret < 0) { | |
1744 | goto fail; | |
1745 | } | |
1746 | ||
1747 | for (j = 0; j < s->snapshots[i].l1_size; j++) { | |
1748 | be64_to_cpus(&l1_table[j]); | |
1749 | } | |
1750 | ||
1751 | ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size, | |
e390cf5a | 1752 | &expanded_clusters, &nb_clusters); |
32b6444d MR |
1753 | if (ret < 0) { |
1754 | goto fail; | |
1755 | } | |
1756 | } | |
1757 | ||
1758 | ret = 0; | |
1759 | ||
1760 | fail: | |
1761 | g_free(expanded_clusters); | |
1762 | g_free(l1_table); | |
1763 | return ret; | |
1764 | } |