]>
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 | ||
80c71a24 | 25 | #include "qemu/osdep.h" |
45aba42f KW |
26 | #include <zlib.h> |
27 | ||
28 | #include "qemu-common.h" | |
737e150e | 29 | #include "block/block_int.h" |
45aba42f | 30 | #include "block/qcow2.h" |
58369e22 | 31 | #include "qemu/bswap.h" |
3cce16f4 | 32 | #include "trace.h" |
45aba42f | 33 | |
46b732cd PB |
34 | int qcow2_shrink_l1_table(BlockDriverState *bs, uint64_t exact_size) |
35 | { | |
36 | BDRVQcow2State *s = bs->opaque; | |
37 | int new_l1_size, i, ret; | |
38 | ||
39 | if (exact_size >= s->l1_size) { | |
40 | return 0; | |
41 | } | |
42 | ||
43 | new_l1_size = exact_size; | |
44 | ||
45 | #ifdef DEBUG_ALLOC2 | |
46 | fprintf(stderr, "shrink l1_table from %d to %d\n", s->l1_size, new_l1_size); | |
47 | #endif | |
48 | ||
49 | BLKDBG_EVENT(bs->file, BLKDBG_L1_SHRINK_WRITE_TABLE); | |
50 | ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset + | |
51 | new_l1_size * sizeof(uint64_t), | |
52 | (s->l1_size - new_l1_size) * sizeof(uint64_t), 0); | |
53 | if (ret < 0) { | |
54 | goto fail; | |
55 | } | |
56 | ||
57 | ret = bdrv_flush(bs->file->bs); | |
58 | if (ret < 0) { | |
59 | goto fail; | |
60 | } | |
61 | ||
62 | BLKDBG_EVENT(bs->file, BLKDBG_L1_SHRINK_FREE_L2_CLUSTERS); | |
63 | for (i = s->l1_size - 1; i > new_l1_size - 1; i--) { | |
64 | if ((s->l1_table[i] & L1E_OFFSET_MASK) == 0) { | |
65 | continue; | |
66 | } | |
67 | qcow2_free_clusters(bs, s->l1_table[i] & L1E_OFFSET_MASK, | |
68 | s->cluster_size, QCOW2_DISCARD_ALWAYS); | |
69 | s->l1_table[i] = 0; | |
70 | } | |
71 | return 0; | |
72 | ||
73 | fail: | |
74 | /* | |
75 | * If the write in the l1_table failed the image may contain a partially | |
76 | * overwritten l1_table. In this case it would be better to clear the | |
77 | * l1_table in memory to avoid possible image corruption. | |
78 | */ | |
79 | memset(s->l1_table + new_l1_size, 0, | |
80 | (s->l1_size - new_l1_size) * sizeof(uint64_t)); | |
81 | return ret; | |
82 | } | |
83 | ||
2cf7cfa1 KW |
84 | int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size, |
85 | bool exact_size) | |
45aba42f | 86 | { |
ff99129a | 87 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 | 88 | int new_l1_size2, ret, i; |
45aba42f | 89 | uint64_t *new_l1_table; |
fda74f82 | 90 | int64_t old_l1_table_offset, old_l1_size; |
2cf7cfa1 | 91 | int64_t new_l1_table_offset, new_l1_size; |
45aba42f KW |
92 | uint8_t data[12]; |
93 | ||
72893756 | 94 | if (min_size <= s->l1_size) |
45aba42f | 95 | return 0; |
72893756 | 96 | |
b93f9950 HR |
97 | /* Do a sanity check on min_size before trying to calculate new_l1_size |
98 | * (this prevents overflows during the while loop for the calculation of | |
99 | * new_l1_size) */ | |
100 | if (min_size > INT_MAX / sizeof(uint64_t)) { | |
101 | return -EFBIG; | |
102 | } | |
103 | ||
72893756 SH |
104 | if (exact_size) { |
105 | new_l1_size = min_size; | |
106 | } else { | |
107 | /* Bump size up to reduce the number of times we have to grow */ | |
108 | new_l1_size = s->l1_size; | |
109 | if (new_l1_size == 0) { | |
110 | new_l1_size = 1; | |
111 | } | |
112 | while (min_size > new_l1_size) { | |
21cf3e12 | 113 | new_l1_size = DIV_ROUND_UP(new_l1_size * 3, 2); |
72893756 | 114 | } |
45aba42f | 115 | } |
72893756 | 116 | |
84c26520 HR |
117 | QEMU_BUILD_BUG_ON(QCOW_MAX_L1_SIZE > INT_MAX); |
118 | if (new_l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) { | |
2cf7cfa1 KW |
119 | return -EFBIG; |
120 | } | |
121 | ||
45aba42f | 122 | #ifdef DEBUG_ALLOC2 |
2cf7cfa1 KW |
123 | fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n", |
124 | s->l1_size, new_l1_size); | |
45aba42f KW |
125 | #endif |
126 | ||
127 | new_l1_size2 = sizeof(uint64_t) * new_l1_size; | |
9a4f4c31 | 128 | new_l1_table = qemu_try_blockalign(bs->file->bs, |
de82815d KW |
129 | align_offset(new_l1_size2, 512)); |
130 | if (new_l1_table == NULL) { | |
131 | return -ENOMEM; | |
132 | } | |
133 | memset(new_l1_table, 0, align_offset(new_l1_size2, 512)); | |
134 | ||
0647d47c SH |
135 | if (s->l1_size) { |
136 | memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t)); | |
137 | } | |
45aba42f KW |
138 | |
139 | /* write new table (align to cluster) */ | |
66f82cee | 140 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE); |
ed6ccf0f | 141 | new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2); |
5d757b56 | 142 | if (new_l1_table_offset < 0) { |
de82815d | 143 | qemu_vfree(new_l1_table); |
5d757b56 KW |
144 | return new_l1_table_offset; |
145 | } | |
29c1a730 KW |
146 | |
147 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
148 | if (ret < 0) { | |
80fa3341 | 149 | goto fail; |
29c1a730 | 150 | } |
45aba42f | 151 | |
cf93980e HR |
152 | /* the L1 position has not yet been updated, so these clusters must |
153 | * indeed be completely free */ | |
231bb267 HR |
154 | ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset, |
155 | new_l1_size2); | |
cf93980e HR |
156 | if (ret < 0) { |
157 | goto fail; | |
158 | } | |
159 | ||
66f82cee | 160 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE); |
45aba42f KW |
161 | for(i = 0; i < s->l1_size; i++) |
162 | new_l1_table[i] = cpu_to_be64(new_l1_table[i]); | |
d9ca2ea2 | 163 | ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, |
9a4f4c31 | 164 | new_l1_table, new_l1_size2); |
8b3b7206 | 165 | if (ret < 0) |
45aba42f KW |
166 | goto fail; |
167 | for(i = 0; i < s->l1_size; i++) | |
168 | new_l1_table[i] = be64_to_cpu(new_l1_table[i]); | |
169 | ||
170 | /* set new table */ | |
66f82cee | 171 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE); |
f1f7a1dd | 172 | stl_be_p(data, new_l1_size); |
e4ef9f46 | 173 | stq_be_p(data + 4, new_l1_table_offset); |
d9ca2ea2 | 174 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), |
9a4f4c31 | 175 | data, sizeof(data)); |
8b3b7206 | 176 | if (ret < 0) { |
45aba42f | 177 | goto fail; |
fb8fa77c | 178 | } |
de82815d | 179 | qemu_vfree(s->l1_table); |
fda74f82 | 180 | old_l1_table_offset = s->l1_table_offset; |
45aba42f KW |
181 | s->l1_table_offset = new_l1_table_offset; |
182 | s->l1_table = new_l1_table; | |
fda74f82 | 183 | old_l1_size = s->l1_size; |
45aba42f | 184 | s->l1_size = new_l1_size; |
fda74f82 HR |
185 | qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * sizeof(uint64_t), |
186 | QCOW2_DISCARD_OTHER); | |
45aba42f KW |
187 | return 0; |
188 | fail: | |
de82815d | 189 | qemu_vfree(new_l1_table); |
6cfcb9b8 KW |
190 | qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2, |
191 | QCOW2_DISCARD_OTHER); | |
8b3b7206 | 192 | return ret; |
45aba42f KW |
193 | } |
194 | ||
45aba42f KW |
195 | /* |
196 | * l2_load | |
197 | * | |
198 | * Loads a L2 table into memory. If the table is in the cache, the cache | |
199 | * is used; otherwise the L2 table is loaded from the image file. | |
200 | * | |
201 | * Returns a pointer to the L2 table on success, or NULL if the read from | |
202 | * the image file failed. | |
203 | */ | |
204 | ||
55c17e98 KW |
205 | static int l2_load(BlockDriverState *bs, uint64_t l2_offset, |
206 | uint64_t **l2_table) | |
45aba42f | 207 | { |
ff99129a | 208 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 209 | |
9be38598 EH |
210 | return qcow2_cache_get(bs, s->l2_table_cache, l2_offset, |
211 | (void **)l2_table); | |
45aba42f KW |
212 | } |
213 | ||
6583e3c7 KW |
214 | /* |
215 | * Writes one sector of the L1 table to the disk (can't update single entries | |
216 | * and we really don't want bdrv_pread to perform a read-modify-write) | |
217 | */ | |
218 | #define L1_ENTRIES_PER_SECTOR (512 / 8) | |
e23e400e | 219 | int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index) |
6583e3c7 | 220 | { |
ff99129a | 221 | BDRVQcow2State *s = bs->opaque; |
a1391444 | 222 | uint64_t buf[L1_ENTRIES_PER_SECTOR] = { 0 }; |
6583e3c7 | 223 | int l1_start_index; |
f7defcb6 | 224 | int i, ret; |
6583e3c7 KW |
225 | |
226 | l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1); | |
a1391444 HR |
227 | for (i = 0; i < L1_ENTRIES_PER_SECTOR && l1_start_index + i < s->l1_size; |
228 | i++) | |
229 | { | |
6583e3c7 KW |
230 | buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]); |
231 | } | |
232 | ||
231bb267 | 233 | ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1, |
cf93980e HR |
234 | s->l1_table_offset + 8 * l1_start_index, sizeof(buf)); |
235 | if (ret < 0) { | |
236 | return ret; | |
237 | } | |
238 | ||
66f82cee | 239 | BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); |
d9ca2ea2 | 240 | ret = bdrv_pwrite_sync(bs->file, |
9a4f4c31 KW |
241 | s->l1_table_offset + 8 * l1_start_index, |
242 | buf, sizeof(buf)); | |
f7defcb6 KW |
243 | if (ret < 0) { |
244 | return ret; | |
6583e3c7 KW |
245 | } |
246 | ||
247 | return 0; | |
248 | } | |
249 | ||
45aba42f KW |
250 | /* |
251 | * l2_allocate | |
252 | * | |
253 | * Allocate a new l2 entry in the file. If l1_index points to an already | |
254 | * used entry in the L2 table (i.e. we are doing a copy on write for the L2 | |
255 | * table) copy the contents of the old L2 table into the newly allocated one. | |
256 | * Otherwise the new table is initialized with zeros. | |
257 | * | |
258 | */ | |
259 | ||
c46e1167 | 260 | static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table) |
45aba42f | 261 | { |
ff99129a | 262 | BDRVQcow2State *s = bs->opaque; |
6583e3c7 | 263 | uint64_t old_l2_offset; |
8585afd8 | 264 | uint64_t *l2_table = NULL; |
f4f0d391 | 265 | int64_t l2_offset; |
c46e1167 | 266 | int ret; |
45aba42f KW |
267 | |
268 | old_l2_offset = s->l1_table[l1_index]; | |
269 | ||
3cce16f4 KW |
270 | trace_qcow2_l2_allocate(bs, l1_index); |
271 | ||
45aba42f KW |
272 | /* allocate a new l2 entry */ |
273 | ||
ed6ccf0f | 274 | l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t)); |
5d757b56 | 275 | if (l2_offset < 0) { |
be0b742e HR |
276 | ret = l2_offset; |
277 | goto fail; | |
5d757b56 | 278 | } |
29c1a730 | 279 | |
98839750 AG |
280 | /* If we're allocating the table at offset 0 then something is wrong */ |
281 | if (l2_offset == 0) { | |
282 | qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid " | |
283 | "allocation of L2 table at offset 0"); | |
284 | ret = -EIO; | |
285 | goto fail; | |
286 | } | |
287 | ||
29c1a730 KW |
288 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
289 | if (ret < 0) { | |
290 | goto fail; | |
291 | } | |
45aba42f | 292 | |
45aba42f KW |
293 | /* allocate a new entry in the l2 cache */ |
294 | ||
3cce16f4 | 295 | trace_qcow2_l2_allocate_get_empty(bs, l1_index); |
29c1a730 KW |
296 | ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table); |
297 | if (ret < 0) { | |
be0b742e | 298 | goto fail; |
29c1a730 KW |
299 | } |
300 | ||
301 | l2_table = *table; | |
45aba42f | 302 | |
8e37f681 | 303 | if ((old_l2_offset & L1E_OFFSET_MASK) == 0) { |
45aba42f KW |
304 | /* if there was no old l2 table, clear the new table */ |
305 | memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); | |
306 | } else { | |
29c1a730 KW |
307 | uint64_t* old_table; |
308 | ||
45aba42f | 309 | /* if there was an old l2 table, read it from the disk */ |
66f82cee | 310 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ); |
8e37f681 KW |
311 | ret = qcow2_cache_get(bs, s->l2_table_cache, |
312 | old_l2_offset & L1E_OFFSET_MASK, | |
29c1a730 KW |
313 | (void**) &old_table); |
314 | if (ret < 0) { | |
315 | goto fail; | |
316 | } | |
317 | ||
318 | memcpy(l2_table, old_table, s->cluster_size); | |
319 | ||
a3f1afb4 | 320 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &old_table); |
45aba42f | 321 | } |
29c1a730 | 322 | |
45aba42f | 323 | /* write the l2 table to the file */ |
66f82cee | 324 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE); |
29c1a730 | 325 | |
3cce16f4 | 326 | trace_qcow2_l2_allocate_write_l2(bs, l1_index); |
72e80b89 | 327 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
29c1a730 | 328 | ret = qcow2_cache_flush(bs, s->l2_table_cache); |
c46e1167 | 329 | if (ret < 0) { |
175e1152 KW |
330 | goto fail; |
331 | } | |
332 | ||
333 | /* update the L1 entry */ | |
3cce16f4 | 334 | trace_qcow2_l2_allocate_write_l1(bs, l1_index); |
175e1152 | 335 | s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; |
e23e400e | 336 | ret = qcow2_write_l1_entry(bs, l1_index); |
175e1152 KW |
337 | if (ret < 0) { |
338 | goto fail; | |
c46e1167 | 339 | } |
45aba42f | 340 | |
c46e1167 | 341 | *table = l2_table; |
3cce16f4 | 342 | trace_qcow2_l2_allocate_done(bs, l1_index, 0); |
c46e1167 | 343 | return 0; |
175e1152 KW |
344 | |
345 | fail: | |
3cce16f4 | 346 | trace_qcow2_l2_allocate_done(bs, l1_index, ret); |
8585afd8 HR |
347 | if (l2_table != NULL) { |
348 | qcow2_cache_put(bs, s->l2_table_cache, (void**) table); | |
349 | } | |
68dba0bf | 350 | s->l1_table[l1_index] = old_l2_offset; |
e3b21ef9 HR |
351 | if (l2_offset > 0) { |
352 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), | |
353 | QCOW2_DISCARD_ALWAYS); | |
354 | } | |
175e1152 | 355 | return ret; |
45aba42f KW |
356 | } |
357 | ||
2bfcc4a0 KW |
358 | /* |
359 | * Checks how many clusters in a given L2 table are contiguous in the image | |
360 | * file. As soon as one of the flags in the bitmask stop_flags changes compared | |
361 | * to the first cluster, the search is stopped and the cluster is not counted | |
362 | * as contiguous. (This allows it, for example, to stop at the first compressed | |
363 | * cluster which may require a different handling) | |
364 | */ | |
b6d36def | 365 | static int count_contiguous_clusters(int nb_clusters, int cluster_size, |
61653008 | 366 | uint64_t *l2_table, uint64_t stop_flags) |
45aba42f KW |
367 | { |
368 | int i; | |
3ef95218 | 369 | QCow2ClusterType first_cluster_type; |
78a52ad5 | 370 | uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED; |
15684a47 HR |
371 | uint64_t first_entry = be64_to_cpu(l2_table[0]); |
372 | uint64_t offset = first_entry & mask; | |
45aba42f | 373 | |
564a6b69 | 374 | if (!offset) { |
45aba42f | 375 | return 0; |
564a6b69 | 376 | } |
45aba42f | 377 | |
564a6b69 HR |
378 | /* must be allocated */ |
379 | first_cluster_type = qcow2_get_cluster_type(first_entry); | |
380 | assert(first_cluster_type == QCOW2_CLUSTER_NORMAL || | |
fdfab37d | 381 | first_cluster_type == QCOW2_CLUSTER_ZERO_ALLOC); |
15684a47 | 382 | |
61653008 | 383 | for (i = 0; i < nb_clusters; i++) { |
2bfcc4a0 KW |
384 | uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask; |
385 | if (offset + (uint64_t) i * cluster_size != l2_entry) { | |
45aba42f | 386 | break; |
2bfcc4a0 KW |
387 | } |
388 | } | |
45aba42f | 389 | |
61653008 | 390 | return i; |
45aba42f KW |
391 | } |
392 | ||
4341df8a EB |
393 | /* |
394 | * Checks how many consecutive unallocated clusters in a given L2 | |
395 | * table have the same cluster type. | |
396 | */ | |
397 | static int count_contiguous_clusters_unallocated(int nb_clusters, | |
398 | uint64_t *l2_table, | |
3ef95218 | 399 | QCow2ClusterType wanted_type) |
45aba42f | 400 | { |
2bfcc4a0 KW |
401 | int i; |
402 | ||
fdfab37d | 403 | assert(wanted_type == QCOW2_CLUSTER_ZERO_PLAIN || |
4341df8a | 404 | wanted_type == QCOW2_CLUSTER_UNALLOCATED); |
2bfcc4a0 | 405 | for (i = 0; i < nb_clusters; i++) { |
4341df8a | 406 | uint64_t entry = be64_to_cpu(l2_table[i]); |
3ef95218 | 407 | QCow2ClusterType type = qcow2_get_cluster_type(entry); |
45aba42f | 408 | |
fdfab37d | 409 | if (type != wanted_type) { |
2bfcc4a0 KW |
410 | break; |
411 | } | |
412 | } | |
45aba42f KW |
413 | |
414 | return i; | |
415 | } | |
416 | ||
672f0f2c AG |
417 | static int coroutine_fn do_perform_cow_read(BlockDriverState *bs, |
418 | uint64_t src_cluster_offset, | |
419 | unsigned offset_in_cluster, | |
86b862c4 | 420 | QEMUIOVector *qiov) |
45aba42f | 421 | { |
aaa4d20b | 422 | int ret; |
1b9f1491 | 423 | |
86b862c4 | 424 | if (qiov->size == 0) { |
99450c6f AG |
425 | return 0; |
426 | } | |
427 | ||
66f82cee | 428 | BLKDBG_EVENT(bs->file, BLKDBG_COW_READ); |
aef4acb6 | 429 | |
dba28555 | 430 | if (!bs->drv) { |
672f0f2c | 431 | return -ENOMEDIUM; |
dba28555 HR |
432 | } |
433 | ||
aef4acb6 SH |
434 | /* Call .bdrv_co_readv() directly instead of using the public block-layer |
435 | * interface. This avoids double I/O throttling and request tracking, | |
436 | * which can lead to deadlock when block layer copy-on-read is enabled. | |
437 | */ | |
aaa4d20b | 438 | ret = bs->drv->bdrv_co_preadv(bs, src_cluster_offset + offset_in_cluster, |
86b862c4 | 439 | qiov->size, qiov, 0); |
1b9f1491 | 440 | if (ret < 0) { |
672f0f2c | 441 | return ret; |
1b9f1491 KW |
442 | } |
443 | ||
672f0f2c AG |
444 | return 0; |
445 | } | |
446 | ||
447 | static bool coroutine_fn do_perform_cow_encrypt(BlockDriverState *bs, | |
448 | uint64_t src_cluster_offset, | |
4652b8f3 | 449 | uint64_t cluster_offset, |
672f0f2c AG |
450 | unsigned offset_in_cluster, |
451 | uint8_t *buffer, | |
452 | unsigned bytes) | |
453 | { | |
454 | if (bytes && bs->encrypted) { | |
455 | BDRVQcow2State *s = bs->opaque; | |
4609742a | 456 | int64_t offset = (s->crypt_physical_offset ? |
4652b8f3 | 457 | (cluster_offset + offset_in_cluster) : |
4609742a | 458 | (src_cluster_offset + offset_in_cluster)); |
aaa4d20b KW |
459 | assert((offset_in_cluster & ~BDRV_SECTOR_MASK) == 0); |
460 | assert((bytes & ~BDRV_SECTOR_MASK) == 0); | |
b25b387f | 461 | assert(s->crypto); |
4609742a | 462 | if (qcrypto_block_encrypt(s->crypto, offset, buffer, bytes, NULL) < 0) { |
672f0f2c | 463 | return false; |
f6fa64f6 | 464 | } |
45aba42f | 465 | } |
672f0f2c AG |
466 | return true; |
467 | } | |
468 | ||
469 | static int coroutine_fn do_perform_cow_write(BlockDriverState *bs, | |
470 | uint64_t cluster_offset, | |
471 | unsigned offset_in_cluster, | |
86b862c4 | 472 | QEMUIOVector *qiov) |
672f0f2c | 473 | { |
672f0f2c AG |
474 | int ret; |
475 | ||
86b862c4 | 476 | if (qiov->size == 0) { |
672f0f2c AG |
477 | return 0; |
478 | } | |
479 | ||
231bb267 | 480 | ret = qcow2_pre_write_overlap_check(bs, 0, |
86b862c4 | 481 | cluster_offset + offset_in_cluster, qiov->size); |
cf93980e | 482 | if (ret < 0) { |
672f0f2c | 483 | return ret; |
cf93980e HR |
484 | } |
485 | ||
66f82cee | 486 | BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE); |
a03ef88f | 487 | ret = bdrv_co_pwritev(bs->file, cluster_offset + offset_in_cluster, |
86b862c4 | 488 | qiov->size, qiov, 0); |
1b9f1491 | 489 | if (ret < 0) { |
672f0f2c | 490 | return ret; |
1b9f1491 KW |
491 | } |
492 | ||
672f0f2c | 493 | return 0; |
45aba42f KW |
494 | } |
495 | ||
496 | ||
497 | /* | |
498 | * get_cluster_offset | |
499 | * | |
ecfe1863 KW |
500 | * For a given offset of the virtual disk, find the cluster type and offset in |
501 | * the qcow2 file. The offset is stored in *cluster_offset. | |
45aba42f | 502 | * |
ecfe1863 KW |
503 | * On entry, *bytes is the maximum number of contiguous bytes starting at |
504 | * offset that we are interested in. | |
45aba42f | 505 | * |
ecfe1863 KW |
506 | * On exit, *bytes is the number of bytes starting at offset that have the same |
507 | * cluster type and (if applicable) are stored contiguously in the image file. | |
508 | * Compressed clusters are always returned one by one. | |
45aba42f | 509 | * |
68d000a3 KW |
510 | * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error |
511 | * cases. | |
45aba42f | 512 | */ |
1c46efaa | 513 | int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, |
ecfe1863 | 514 | unsigned int *bytes, uint64_t *cluster_offset) |
45aba42f | 515 | { |
ff99129a | 516 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
517 | unsigned int l2_index; |
518 | uint64_t l1_index, l2_offset, *l2_table; | |
45aba42f | 519 | int l1_bits, c; |
c834cba9 HR |
520 | unsigned int offset_in_cluster; |
521 | uint64_t bytes_available, bytes_needed, nb_clusters; | |
3ef95218 | 522 | QCow2ClusterType type; |
55c17e98 | 523 | int ret; |
45aba42f | 524 | |
b2f65d6b | 525 | offset_in_cluster = offset_into_cluster(s, offset); |
ecfe1863 | 526 | bytes_needed = (uint64_t) *bytes + offset_in_cluster; |
45aba42f | 527 | |
b2f65d6b | 528 | l1_bits = s->l2_bits + s->cluster_bits; |
45aba42f | 529 | |
b2f65d6b KW |
530 | /* compute how many bytes there are between the start of the cluster |
531 | * containing offset and the end of the l1 entry */ | |
532 | bytes_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1)) | |
533 | + offset_in_cluster; | |
45aba42f | 534 | |
b2f65d6b KW |
535 | if (bytes_needed > bytes_available) { |
536 | bytes_needed = bytes_available; | |
45aba42f KW |
537 | } |
538 | ||
1c46efaa | 539 | *cluster_offset = 0; |
45aba42f | 540 | |
b6af0975 | 541 | /* seek to the l2 offset in the l1 table */ |
45aba42f KW |
542 | |
543 | l1_index = offset >> l1_bits; | |
68d000a3 | 544 | if (l1_index >= s->l1_size) { |
3ef95218 | 545 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 546 | goto out; |
68d000a3 | 547 | } |
45aba42f | 548 | |
68d000a3 KW |
549 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
550 | if (!l2_offset) { | |
3ef95218 | 551 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 552 | goto out; |
68d000a3 | 553 | } |
45aba42f | 554 | |
a97c67ee HR |
555 | if (offset_into_cluster(s, l2_offset)) { |
556 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
557 | " unaligned (L1 index: %#" PRIx64 ")", | |
558 | l2_offset, l1_index); | |
559 | return -EIO; | |
560 | } | |
561 | ||
45aba42f KW |
562 | /* load the l2 table in memory */ |
563 | ||
55c17e98 KW |
564 | ret = l2_load(bs, l2_offset, &l2_table); |
565 | if (ret < 0) { | |
566 | return ret; | |
1c46efaa | 567 | } |
45aba42f KW |
568 | |
569 | /* find the cluster offset for the given disk offset */ | |
570 | ||
24990c5b | 571 | l2_index = offset_to_l2_index(s, offset); |
1c46efaa | 572 | *cluster_offset = be64_to_cpu(l2_table[l2_index]); |
b6d36def | 573 | |
b2f65d6b | 574 | nb_clusters = size_to_clusters(s, bytes_needed); |
c834cba9 HR |
575 | /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned |
576 | * integers; the minimum cluster size is 512, so this assertion is always | |
577 | * true */ | |
578 | assert(nb_clusters <= INT_MAX); | |
45aba42f | 579 | |
3ef95218 | 580 | type = qcow2_get_cluster_type(*cluster_offset); |
fdfab37d EB |
581 | if (s->qcow_version < 3 && (type == QCOW2_CLUSTER_ZERO_PLAIN || |
582 | type == QCOW2_CLUSTER_ZERO_ALLOC)) { | |
583 | qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found" | |
584 | " in pre-v3 image (L2 offset: %#" PRIx64 | |
585 | ", L2 index: %#x)", l2_offset, l2_index); | |
586 | ret = -EIO; | |
587 | goto fail; | |
588 | } | |
3ef95218 | 589 | switch (type) { |
68d000a3 KW |
590 | case QCOW2_CLUSTER_COMPRESSED: |
591 | /* Compressed clusters can only be processed one by one */ | |
592 | c = 1; | |
593 | *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK; | |
594 | break; | |
fdfab37d | 595 | case QCOW2_CLUSTER_ZERO_PLAIN: |
68d000a3 | 596 | case QCOW2_CLUSTER_UNALLOCATED: |
45aba42f | 597 | /* how many empty clusters ? */ |
4341df8a | 598 | c = count_contiguous_clusters_unallocated(nb_clusters, |
fdfab37d | 599 | &l2_table[l2_index], type); |
68d000a3 KW |
600 | *cluster_offset = 0; |
601 | break; | |
fdfab37d | 602 | case QCOW2_CLUSTER_ZERO_ALLOC: |
68d000a3 | 603 | case QCOW2_CLUSTER_NORMAL: |
45aba42f KW |
604 | /* how many allocated clusters ? */ |
605 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
fdfab37d | 606 | &l2_table[l2_index], QCOW_OFLAG_ZERO); |
68d000a3 | 607 | *cluster_offset &= L2E_OFFSET_MASK; |
a97c67ee | 608 | if (offset_into_cluster(s, *cluster_offset)) { |
fdfab37d EB |
609 | qcow2_signal_corruption(bs, true, -1, -1, |
610 | "Cluster allocation offset %#" | |
a97c67ee HR |
611 | PRIx64 " unaligned (L2 offset: %#" PRIx64 |
612 | ", L2 index: %#x)", *cluster_offset, | |
613 | l2_offset, l2_index); | |
614 | ret = -EIO; | |
615 | goto fail; | |
616 | } | |
68d000a3 | 617 | break; |
1417d7e4 KW |
618 | default: |
619 | abort(); | |
45aba42f KW |
620 | } |
621 | ||
29c1a730 KW |
622 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
623 | ||
c834cba9 | 624 | bytes_available = (int64_t)c * s->cluster_size; |
68d000a3 | 625 | |
45aba42f | 626 | out: |
b2f65d6b KW |
627 | if (bytes_available > bytes_needed) { |
628 | bytes_available = bytes_needed; | |
629 | } | |
45aba42f | 630 | |
c834cba9 HR |
631 | /* bytes_available <= bytes_needed <= *bytes + offset_in_cluster; |
632 | * subtracting offset_in_cluster will therefore definitely yield something | |
633 | * not exceeding UINT_MAX */ | |
634 | assert(bytes_available - offset_in_cluster <= UINT_MAX); | |
ecfe1863 | 635 | *bytes = bytes_available - offset_in_cluster; |
45aba42f | 636 | |
3ef95218 | 637 | return type; |
a97c67ee HR |
638 | |
639 | fail: | |
640 | qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table); | |
641 | return ret; | |
45aba42f KW |
642 | } |
643 | ||
644 | /* | |
645 | * get_cluster_table | |
646 | * | |
647 | * for a given disk offset, load (and allocate if needed) | |
648 | * the l2 table. | |
649 | * | |
650 | * the l2 table offset in the qcow2 file and the cluster index | |
651 | * in the l2 table are given to the caller. | |
652 | * | |
1e3e8f1a | 653 | * Returns 0 on success, -errno in failure case |
45aba42f | 654 | */ |
45aba42f KW |
655 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
656 | uint64_t **new_l2_table, | |
45aba42f KW |
657 | int *new_l2_index) |
658 | { | |
ff99129a | 659 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
660 | unsigned int l2_index; |
661 | uint64_t l1_index, l2_offset; | |
c46e1167 | 662 | uint64_t *l2_table = NULL; |
80ee15a6 | 663 | int ret; |
45aba42f | 664 | |
b6af0975 | 665 | /* seek to the l2 offset in the l1 table */ |
45aba42f KW |
666 | |
667 | l1_index = offset >> (s->l2_bits + s->cluster_bits); | |
668 | if (l1_index >= s->l1_size) { | |
72893756 | 669 | ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
1e3e8f1a KW |
670 | if (ret < 0) { |
671 | return ret; | |
672 | } | |
45aba42f | 673 | } |
8e37f681 | 674 | |
2cf7cfa1 | 675 | assert(l1_index < s->l1_size); |
8e37f681 | 676 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
a97c67ee HR |
677 | if (offset_into_cluster(s, l2_offset)) { |
678 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
679 | " unaligned (L1 index: %#" PRIx64 ")", | |
680 | l2_offset, l1_index); | |
681 | return -EIO; | |
682 | } | |
45aba42f KW |
683 | |
684 | /* seek the l2 table of the given l2 offset */ | |
685 | ||
8e37f681 | 686 | if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) { |
45aba42f | 687 | /* load the l2 table in memory */ |
55c17e98 KW |
688 | ret = l2_load(bs, l2_offset, &l2_table); |
689 | if (ret < 0) { | |
690 | return ret; | |
1e3e8f1a | 691 | } |
45aba42f | 692 | } else { |
16fde5f2 | 693 | /* First allocate a new L2 table (and do COW if needed) */ |
c46e1167 KW |
694 | ret = l2_allocate(bs, l1_index, &l2_table); |
695 | if (ret < 0) { | |
696 | return ret; | |
1e3e8f1a | 697 | } |
16fde5f2 KW |
698 | |
699 | /* Then decrease the refcount of the old table */ | |
700 | if (l2_offset) { | |
6cfcb9b8 KW |
701 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), |
702 | QCOW2_DISCARD_OTHER); | |
16fde5f2 | 703 | } |
45aba42f KW |
704 | } |
705 | ||
706 | /* find the cluster offset for the given disk offset */ | |
707 | ||
24990c5b | 708 | l2_index = offset_to_l2_index(s, offset); |
45aba42f KW |
709 | |
710 | *new_l2_table = l2_table; | |
45aba42f KW |
711 | *new_l2_index = l2_index; |
712 | ||
1e3e8f1a | 713 | return 0; |
45aba42f KW |
714 | } |
715 | ||
716 | /* | |
717 | * alloc_compressed_cluster_offset | |
718 | * | |
719 | * For a given offset of the disk image, return cluster offset in | |
720 | * qcow2 file. | |
721 | * | |
722 | * If the offset is not found, allocate a new compressed cluster. | |
723 | * | |
724 | * Return the cluster offset if successful, | |
725 | * Return 0, otherwise. | |
726 | * | |
727 | */ | |
728 | ||
ed6ccf0f KW |
729 | uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
730 | uint64_t offset, | |
731 | int compressed_size) | |
45aba42f | 732 | { |
ff99129a | 733 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 734 | int l2_index, ret; |
3948d1d4 | 735 | uint64_t *l2_table; |
f4f0d391 | 736 | int64_t cluster_offset; |
45aba42f KW |
737 | int nb_csectors; |
738 | ||
3948d1d4 | 739 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
1e3e8f1a | 740 | if (ret < 0) { |
45aba42f | 741 | return 0; |
1e3e8f1a | 742 | } |
45aba42f | 743 | |
b0b6862e KW |
744 | /* Compression can't overwrite anything. Fail if the cluster was already |
745 | * allocated. */ | |
45aba42f | 746 | cluster_offset = be64_to_cpu(l2_table[l2_index]); |
b0b6862e | 747 | if (cluster_offset & L2E_OFFSET_MASK) { |
8f1efd00 KW |
748 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
749 | return 0; | |
750 | } | |
45aba42f | 751 | |
ed6ccf0f | 752 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
5d757b56 | 753 | if (cluster_offset < 0) { |
29c1a730 | 754 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
5d757b56 KW |
755 | return 0; |
756 | } | |
757 | ||
45aba42f KW |
758 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
759 | (cluster_offset >> 9); | |
760 | ||
761 | cluster_offset |= QCOW_OFLAG_COMPRESSED | | |
762 | ((uint64_t)nb_csectors << s->csize_shift); | |
763 | ||
764 | /* update L2 table */ | |
765 | ||
766 | /* compressed clusters never have the copied flag */ | |
767 | ||
66f82cee | 768 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
72e80b89 | 769 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
45aba42f | 770 | l2_table[l2_index] = cpu_to_be64(cluster_offset); |
a3f1afb4 | 771 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
4c1612d9 | 772 | |
29c1a730 | 773 | return cluster_offset; |
4c1612d9 KW |
774 | } |
775 | ||
99450c6f | 776 | static int perform_cow(BlockDriverState *bs, QCowL2Meta *m) |
593fb83c | 777 | { |
ff99129a | 778 | BDRVQcow2State *s = bs->opaque; |
99450c6f AG |
779 | Qcow2COWRegion *start = &m->cow_start; |
780 | Qcow2COWRegion *end = &m->cow_end; | |
672f0f2c | 781 | unsigned buffer_size; |
b3cf1c7c AG |
782 | unsigned data_bytes = end->offset - (start->offset + start->nb_bytes); |
783 | bool merge_reads; | |
672f0f2c | 784 | uint8_t *start_buffer, *end_buffer; |
86b862c4 | 785 | QEMUIOVector qiov; |
593fb83c KW |
786 | int ret; |
787 | ||
672f0f2c | 788 | assert(start->nb_bytes <= UINT_MAX - end->nb_bytes); |
b3cf1c7c AG |
789 | assert(start->nb_bytes + end->nb_bytes <= UINT_MAX - data_bytes); |
790 | assert(start->offset + start->nb_bytes <= end->offset); | |
ee22a9d8 | 791 | assert(!m->data_qiov || m->data_qiov->size == data_bytes); |
672f0f2c | 792 | |
99450c6f | 793 | if (start->nb_bytes == 0 && end->nb_bytes == 0) { |
593fb83c KW |
794 | return 0; |
795 | } | |
796 | ||
b3cf1c7c AG |
797 | /* If we have to read both the start and end COW regions and the |
798 | * middle region is not too large then perform just one read | |
799 | * operation */ | |
800 | merge_reads = start->nb_bytes && end->nb_bytes && data_bytes <= 16384; | |
801 | if (merge_reads) { | |
802 | buffer_size = start->nb_bytes + data_bytes + end->nb_bytes; | |
803 | } else { | |
804 | /* If we have to do two reads, add some padding in the middle | |
805 | * if necessary to make sure that the end region is optimally | |
806 | * aligned. */ | |
807 | size_t align = bdrv_opt_mem_align(bs); | |
808 | assert(align > 0 && align <= UINT_MAX); | |
809 | assert(QEMU_ALIGN_UP(start->nb_bytes, align) <= | |
810 | UINT_MAX - end->nb_bytes); | |
811 | buffer_size = QEMU_ALIGN_UP(start->nb_bytes, align) + end->nb_bytes; | |
812 | } | |
813 | ||
814 | /* Reserve a buffer large enough to store all the data that we're | |
815 | * going to read */ | |
672f0f2c AG |
816 | start_buffer = qemu_try_blockalign(bs, buffer_size); |
817 | if (start_buffer == NULL) { | |
818 | return -ENOMEM; | |
819 | } | |
820 | /* The part of the buffer where the end region is located */ | |
821 | end_buffer = start_buffer + buffer_size - end->nb_bytes; | |
822 | ||
ee22a9d8 | 823 | qemu_iovec_init(&qiov, 2 + (m->data_qiov ? m->data_qiov->niov : 0)); |
86b862c4 | 824 | |
593fb83c | 825 | qemu_co_mutex_unlock(&s->lock); |
b3cf1c7c AG |
826 | /* First we read the existing data from both COW regions. We |
827 | * either read the whole region in one go, or the start and end | |
828 | * regions separately. */ | |
829 | if (merge_reads) { | |
86b862c4 AG |
830 | qemu_iovec_add(&qiov, start_buffer, buffer_size); |
831 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c | 832 | } else { |
86b862c4 AG |
833 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); |
834 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c AG |
835 | if (ret < 0) { |
836 | goto fail; | |
837 | } | |
672f0f2c | 838 | |
86b862c4 AG |
839 | qemu_iovec_reset(&qiov); |
840 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
841 | ret = do_perform_cow_read(bs, m->offset, end->offset, &qiov); | |
b3cf1c7c | 842 | } |
593fb83c | 843 | if (ret < 0) { |
99450c6f | 844 | goto fail; |
593fb83c KW |
845 | } |
846 | ||
672f0f2c AG |
847 | /* Encrypt the data if necessary before writing it */ |
848 | if (bs->encrypted) { | |
4652b8f3 DB |
849 | if (!do_perform_cow_encrypt(bs, m->offset, m->alloc_offset, |
850 | start->offset, start_buffer, | |
851 | start->nb_bytes) || | |
852 | !do_perform_cow_encrypt(bs, m->offset, m->alloc_offset, | |
853 | end->offset, end_buffer, end->nb_bytes)) { | |
672f0f2c AG |
854 | ret = -EIO; |
855 | goto fail; | |
856 | } | |
857 | } | |
858 | ||
ee22a9d8 AG |
859 | /* And now we can write everything. If we have the guest data we |
860 | * can write everything in one single operation */ | |
861 | if (m->data_qiov) { | |
862 | qemu_iovec_reset(&qiov); | |
863 | if (start->nb_bytes) { | |
864 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
865 | } | |
866 | qemu_iovec_concat(&qiov, m->data_qiov, 0, data_bytes); | |
867 | if (end->nb_bytes) { | |
868 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
869 | } | |
870 | /* NOTE: we have a write_aio blkdebug event here followed by | |
871 | * a cow_write one in do_perform_cow_write(), but there's only | |
872 | * one single I/O operation */ | |
873 | BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); | |
874 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); | |
875 | } else { | |
876 | /* If there's no guest data then write both COW regions separately */ | |
877 | qemu_iovec_reset(&qiov); | |
878 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
879 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); | |
880 | if (ret < 0) { | |
881 | goto fail; | |
882 | } | |
883 | ||
884 | qemu_iovec_reset(&qiov); | |
885 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
886 | ret = do_perform_cow_write(bs, m->alloc_offset, end->offset, &qiov); | |
672f0f2c | 887 | } |
99450c6f AG |
888 | |
889 | fail: | |
890 | qemu_co_mutex_lock(&s->lock); | |
891 | ||
593fb83c KW |
892 | /* |
893 | * Before we update the L2 table to actually point to the new cluster, we | |
894 | * need to be sure that the refcounts have been increased and COW was | |
895 | * handled. | |
896 | */ | |
99450c6f AG |
897 | if (ret == 0) { |
898 | qcow2_cache_depends_on_flush(s->l2_table_cache); | |
899 | } | |
593fb83c | 900 | |
672f0f2c | 901 | qemu_vfree(start_buffer); |
86b862c4 | 902 | qemu_iovec_destroy(&qiov); |
99450c6f | 903 | return ret; |
593fb83c KW |
904 | } |
905 | ||
148da7ea | 906 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) |
45aba42f | 907 | { |
ff99129a | 908 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 909 | int i, j = 0, l2_index, ret; |
593fb83c | 910 | uint64_t *old_cluster, *l2_table; |
250196f1 | 911 | uint64_t cluster_offset = m->alloc_offset; |
45aba42f | 912 | |
3cce16f4 | 913 | trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
f50f88b9 | 914 | assert(m->nb_clusters > 0); |
45aba42f | 915 | |
5839e53b | 916 | old_cluster = g_try_new(uint64_t, m->nb_clusters); |
de82815d KW |
917 | if (old_cluster == NULL) { |
918 | ret = -ENOMEM; | |
919 | goto err; | |
920 | } | |
45aba42f KW |
921 | |
922 | /* copy content of unmodified sectors */ | |
99450c6f | 923 | ret = perform_cow(bs, m); |
593fb83c KW |
924 | if (ret < 0) { |
925 | goto err; | |
29c1a730 KW |
926 | } |
927 | ||
593fb83c | 928 | /* Update L2 table. */ |
74c4510a | 929 | if (s->use_lazy_refcounts) { |
280d3735 KW |
930 | qcow2_mark_dirty(bs); |
931 | } | |
bfe8043e SH |
932 | if (qcow2_need_accurate_refcounts(s)) { |
933 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
934 | s->refcount_block_cache); | |
935 | } | |
280d3735 | 936 | |
3948d1d4 | 937 | ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index); |
1e3e8f1a | 938 | if (ret < 0) { |
45aba42f | 939 | goto err; |
1e3e8f1a | 940 | } |
72e80b89 | 941 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
45aba42f | 942 | |
c01dbccb | 943 | assert(l2_index + m->nb_clusters <= s->l2_size); |
45aba42f KW |
944 | for (i = 0; i < m->nb_clusters; i++) { |
945 | /* if two concurrent writes happen to the same unallocated cluster | |
aaa4d20b KW |
946 | * each write allocates separate cluster and writes data concurrently. |
947 | * The first one to complete updates l2 table with pointer to its | |
948 | * cluster the second one has to do RMW (which is done above by | |
949 | * perform_cow()), update l2 table with its cluster pointer and free | |
950 | * old cluster. This is what this loop does */ | |
951 | if (l2_table[l2_index + i] != 0) { | |
45aba42f | 952 | old_cluster[j++] = l2_table[l2_index + i]; |
aaa4d20b | 953 | } |
45aba42f KW |
954 | |
955 | l2_table[l2_index + i] = cpu_to_be64((cluster_offset + | |
956 | (i << s->cluster_bits)) | QCOW_OFLAG_COPIED); | |
957 | } | |
958 | ||
9f8e668e | 959 | |
a3f1afb4 | 960 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
45aba42f | 961 | |
7ec5e6a4 KW |
962 | /* |
963 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
6cfcb9b8 KW |
964 | * |
965 | * Don't discard clusters that reach a refcount of 0 (e.g. compressed | |
966 | * clusters), the next write will reuse them anyway. | |
7ec5e6a4 | 967 | */ |
564a6b69 | 968 | if (!m->keep_old_clusters && j != 0) { |
7ec5e6a4 | 969 | for (i = 0; i < j; i++) { |
6cfcb9b8 KW |
970 | qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1, |
971 | QCOW2_DISCARD_NEVER); | |
7ec5e6a4 KW |
972 | } |
973 | } | |
45aba42f KW |
974 | |
975 | ret = 0; | |
976 | err: | |
7267c094 | 977 | g_free(old_cluster); |
45aba42f KW |
978 | return ret; |
979 | } | |
980 | ||
bf319ece KW |
981 | /* |
982 | * Returns the number of contiguous clusters that can be used for an allocating | |
983 | * write, but require COW to be performed (this includes yet unallocated space, | |
984 | * which must copy from the backing file) | |
985 | */ | |
ff99129a | 986 | static int count_cow_clusters(BDRVQcow2State *s, int nb_clusters, |
bf319ece KW |
987 | uint64_t *l2_table, int l2_index) |
988 | { | |
143550a8 | 989 | int i; |
bf319ece | 990 | |
143550a8 KW |
991 | for (i = 0; i < nb_clusters; i++) { |
992 | uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]); | |
3ef95218 | 993 | QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry); |
143550a8 KW |
994 | |
995 | switch(cluster_type) { | |
996 | case QCOW2_CLUSTER_NORMAL: | |
997 | if (l2_entry & QCOW_OFLAG_COPIED) { | |
998 | goto out; | |
999 | } | |
bf319ece | 1000 | break; |
143550a8 KW |
1001 | case QCOW2_CLUSTER_UNALLOCATED: |
1002 | case QCOW2_CLUSTER_COMPRESSED: | |
fdfab37d EB |
1003 | case QCOW2_CLUSTER_ZERO_PLAIN: |
1004 | case QCOW2_CLUSTER_ZERO_ALLOC: | |
bf319ece | 1005 | break; |
143550a8 KW |
1006 | default: |
1007 | abort(); | |
1008 | } | |
bf319ece KW |
1009 | } |
1010 | ||
143550a8 | 1011 | out: |
bf319ece KW |
1012 | assert(i <= nb_clusters); |
1013 | return i; | |
1014 | } | |
1015 | ||
250196f1 | 1016 | /* |
226c3c26 KW |
1017 | * Check if there already is an AIO write request in flight which allocates |
1018 | * the same cluster. In this case we need to wait until the previous | |
1019 | * request has completed and updated the L2 table accordingly. | |
65eb2e35 KW |
1020 | * |
1021 | * Returns: | |
1022 | * 0 if there was no dependency. *cur_bytes indicates the number of | |
1023 | * bytes from guest_offset that can be read before the next | |
1024 | * dependency must be processed (or the request is complete) | |
1025 | * | |
1026 | * -EAGAIN if we had to wait for another request, previously gathered | |
1027 | * information on cluster allocation may be invalid now. The caller | |
1028 | * must start over anyway, so consider *cur_bytes undefined. | |
250196f1 | 1029 | */ |
226c3c26 | 1030 | static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, |
ecdd5333 | 1031 | uint64_t *cur_bytes, QCowL2Meta **m) |
250196f1 | 1032 | { |
ff99129a | 1033 | BDRVQcow2State *s = bs->opaque; |
250196f1 | 1034 | QCowL2Meta *old_alloc; |
65eb2e35 | 1035 | uint64_t bytes = *cur_bytes; |
250196f1 | 1036 | |
250196f1 KW |
1037 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
1038 | ||
65eb2e35 KW |
1039 | uint64_t start = guest_offset; |
1040 | uint64_t end = start + bytes; | |
1041 | uint64_t old_start = l2meta_cow_start(old_alloc); | |
1042 | uint64_t old_end = l2meta_cow_end(old_alloc); | |
250196f1 | 1043 | |
d9d74f41 | 1044 | if (end <= old_start || start >= old_end) { |
250196f1 KW |
1045 | /* No intersection */ |
1046 | } else { | |
1047 | if (start < old_start) { | |
1048 | /* Stop at the start of a running allocation */ | |
65eb2e35 | 1049 | bytes = old_start - start; |
250196f1 | 1050 | } else { |
65eb2e35 | 1051 | bytes = 0; |
250196f1 KW |
1052 | } |
1053 | ||
ecdd5333 KW |
1054 | /* Stop if already an l2meta exists. After yielding, it wouldn't |
1055 | * be valid any more, so we'd have to clean up the old L2Metas | |
1056 | * and deal with requests depending on them before starting to | |
1057 | * gather new ones. Not worth the trouble. */ | |
1058 | if (bytes == 0 && *m) { | |
1059 | *cur_bytes = 0; | |
1060 | return 0; | |
1061 | } | |
1062 | ||
65eb2e35 | 1063 | if (bytes == 0) { |
250196f1 KW |
1064 | /* Wait for the dependency to complete. We need to recheck |
1065 | * the free/allocated clusters when we continue. */ | |
1ace7cea | 1066 | qemu_co_queue_wait(&old_alloc->dependent_requests, &s->lock); |
250196f1 KW |
1067 | return -EAGAIN; |
1068 | } | |
1069 | } | |
1070 | } | |
1071 | ||
65eb2e35 KW |
1072 | /* Make sure that existing clusters and new allocations are only used up to |
1073 | * the next dependency if we shortened the request above */ | |
1074 | *cur_bytes = bytes; | |
250196f1 | 1075 | |
226c3c26 KW |
1076 | return 0; |
1077 | } | |
1078 | ||
0af729ec KW |
1079 | /* |
1080 | * Checks how many already allocated clusters that don't require a copy on | |
1081 | * write there are at the given guest_offset (up to *bytes). If | |
1082 | * *host_offset is not zero, only physically contiguous clusters beginning at | |
1083 | * this host offset are counted. | |
1084 | * | |
411d62b0 KW |
1085 | * Note that guest_offset may not be cluster aligned. In this case, the |
1086 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1087 | * therefore isn't cluster aligned as well. | |
0af729ec KW |
1088 | * |
1089 | * Returns: | |
1090 | * 0: if no allocated clusters are available at the given offset. | |
1091 | * *bytes is normally unchanged. It is set to 0 if the cluster | |
1092 | * is allocated and doesn't need COW, but doesn't have the right | |
1093 | * physical offset. | |
1094 | * | |
1095 | * 1: if allocated clusters that don't require a COW are available at | |
1096 | * the requested offset. *bytes may have decreased and describes | |
1097 | * the length of the area that can be written to. | |
1098 | * | |
1099 | * -errno: in error cases | |
0af729ec KW |
1100 | */ |
1101 | static int handle_copied(BlockDriverState *bs, uint64_t guest_offset, | |
c53ede9f | 1102 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
0af729ec | 1103 | { |
ff99129a | 1104 | BDRVQcow2State *s = bs->opaque; |
0af729ec KW |
1105 | int l2_index; |
1106 | uint64_t cluster_offset; | |
1107 | uint64_t *l2_table; | |
b6d36def | 1108 | uint64_t nb_clusters; |
c53ede9f | 1109 | unsigned int keep_clusters; |
a3f1afb4 | 1110 | int ret; |
0af729ec KW |
1111 | |
1112 | trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset, | |
1113 | *bytes); | |
0af729ec | 1114 | |
411d62b0 KW |
1115 | assert(*host_offset == 0 || offset_into_cluster(s, guest_offset) |
1116 | == offset_into_cluster(s, *host_offset)); | |
1117 | ||
acb0467f KW |
1118 | /* |
1119 | * Calculate the number of clusters to look for. We stop at L2 table | |
1120 | * boundaries to keep things simple. | |
1121 | */ | |
1122 | nb_clusters = | |
1123 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1124 | ||
1125 | l2_index = offset_to_l2_index(s, guest_offset); | |
1126 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1127 | assert(nb_clusters <= INT_MAX); |
acb0467f | 1128 | |
0af729ec KW |
1129 | /* Find L2 entry for the first involved cluster */ |
1130 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
1131 | if (ret < 0) { | |
1132 | return ret; | |
1133 | } | |
1134 | ||
1135 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
1136 | ||
1137 | /* Check how many clusters are already allocated and don't need COW */ | |
1138 | if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL | |
1139 | && (cluster_offset & QCOW_OFLAG_COPIED)) | |
1140 | { | |
e62daaf6 KW |
1141 | /* If a specific host_offset is required, check it */ |
1142 | bool offset_matches = | |
1143 | (cluster_offset & L2E_OFFSET_MASK) == *host_offset; | |
1144 | ||
a97c67ee HR |
1145 | if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) { |
1146 | qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset " | |
1147 | "%#llx unaligned (guest offset: %#" PRIx64 | |
1148 | ")", cluster_offset & L2E_OFFSET_MASK, | |
1149 | guest_offset); | |
1150 | ret = -EIO; | |
1151 | goto out; | |
1152 | } | |
1153 | ||
e62daaf6 KW |
1154 | if (*host_offset != 0 && !offset_matches) { |
1155 | *bytes = 0; | |
1156 | ret = 0; | |
1157 | goto out; | |
1158 | } | |
1159 | ||
0af729ec | 1160 | /* We keep all QCOW_OFLAG_COPIED clusters */ |
c53ede9f | 1161 | keep_clusters = |
acb0467f | 1162 | count_contiguous_clusters(nb_clusters, s->cluster_size, |
61653008 | 1163 | &l2_table[l2_index], |
0af729ec | 1164 | QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); |
c53ede9f KW |
1165 | assert(keep_clusters <= nb_clusters); |
1166 | ||
1167 | *bytes = MIN(*bytes, | |
1168 | keep_clusters * s->cluster_size | |
1169 | - offset_into_cluster(s, guest_offset)); | |
0af729ec KW |
1170 | |
1171 | ret = 1; | |
1172 | } else { | |
0af729ec KW |
1173 | ret = 0; |
1174 | } | |
1175 | ||
0af729ec | 1176 | /* Cleanup */ |
e62daaf6 | 1177 | out: |
a3f1afb4 | 1178 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
0af729ec | 1179 | |
e62daaf6 KW |
1180 | /* Only return a host offset if we actually made progress. Otherwise we |
1181 | * would make requirements for handle_alloc() that it can't fulfill */ | |
a97c67ee | 1182 | if (ret > 0) { |
411d62b0 KW |
1183 | *host_offset = (cluster_offset & L2E_OFFSET_MASK) |
1184 | + offset_into_cluster(s, guest_offset); | |
e62daaf6 KW |
1185 | } |
1186 | ||
0af729ec KW |
1187 | return ret; |
1188 | } | |
1189 | ||
226c3c26 KW |
1190 | /* |
1191 | * Allocates new clusters for the given guest_offset. | |
1192 | * | |
1193 | * At most *nb_clusters are allocated, and on return *nb_clusters is updated to | |
1194 | * contain the number of clusters that have been allocated and are contiguous | |
1195 | * in the image file. | |
1196 | * | |
1197 | * If *host_offset is non-zero, it specifies the offset in the image file at | |
1198 | * which the new clusters must start. *nb_clusters can be 0 on return in this | |
1199 | * case if the cluster at host_offset is already in use. If *host_offset is | |
1200 | * zero, the clusters can be allocated anywhere in the image file. | |
1201 | * | |
1202 | * *host_offset is updated to contain the offset into the image file at which | |
1203 | * the first allocated cluster starts. | |
1204 | * | |
1205 | * Return 0 on success and -errno in error cases. -EAGAIN means that the | |
1206 | * function has been waiting for another request and the allocation must be | |
1207 | * restarted, but the whole request should not be failed. | |
1208 | */ | |
1209 | static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, | |
b6d36def | 1210 | uint64_t *host_offset, uint64_t *nb_clusters) |
226c3c26 | 1211 | { |
ff99129a | 1212 | BDRVQcow2State *s = bs->opaque; |
226c3c26 KW |
1213 | |
1214 | trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, | |
1215 | *host_offset, *nb_clusters); | |
1216 | ||
250196f1 KW |
1217 | /* Allocate new clusters */ |
1218 | trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); | |
1219 | if (*host_offset == 0) { | |
df021791 KW |
1220 | int64_t cluster_offset = |
1221 | qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); | |
1222 | if (cluster_offset < 0) { | |
1223 | return cluster_offset; | |
1224 | } | |
1225 | *host_offset = cluster_offset; | |
1226 | return 0; | |
250196f1 | 1227 | } else { |
b6d36def | 1228 | int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); |
df021791 KW |
1229 | if (ret < 0) { |
1230 | return ret; | |
1231 | } | |
1232 | *nb_clusters = ret; | |
1233 | return 0; | |
250196f1 | 1234 | } |
250196f1 KW |
1235 | } |
1236 | ||
10f0ed8b KW |
1237 | /* |
1238 | * Allocates new clusters for an area that either is yet unallocated or needs a | |
1239 | * copy on write. If *host_offset is non-zero, clusters are only allocated if | |
1240 | * the new allocation can match the specified host offset. | |
1241 | * | |
411d62b0 KW |
1242 | * Note that guest_offset may not be cluster aligned. In this case, the |
1243 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1244 | * therefore isn't cluster aligned as well. | |
10f0ed8b KW |
1245 | * |
1246 | * Returns: | |
1247 | * 0: if no clusters could be allocated. *bytes is set to 0, | |
1248 | * *host_offset is left unchanged. | |
1249 | * | |
1250 | * 1: if new clusters were allocated. *bytes may be decreased if the | |
1251 | * new allocation doesn't cover all of the requested area. | |
1252 | * *host_offset is updated to contain the host offset of the first | |
1253 | * newly allocated cluster. | |
1254 | * | |
1255 | * -errno: in error cases | |
10f0ed8b KW |
1256 | */ |
1257 | static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, | |
c37f4cd7 | 1258 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
10f0ed8b | 1259 | { |
ff99129a | 1260 | BDRVQcow2State *s = bs->opaque; |
10f0ed8b KW |
1261 | int l2_index; |
1262 | uint64_t *l2_table; | |
1263 | uint64_t entry; | |
b6d36def | 1264 | uint64_t nb_clusters; |
10f0ed8b | 1265 | int ret; |
564a6b69 | 1266 | bool keep_old_clusters = false; |
10f0ed8b | 1267 | |
564a6b69 | 1268 | uint64_t alloc_cluster_offset = 0; |
10f0ed8b KW |
1269 | |
1270 | trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset, | |
1271 | *bytes); | |
1272 | assert(*bytes > 0); | |
1273 | ||
f5bc6350 KW |
1274 | /* |
1275 | * Calculate the number of clusters to look for. We stop at L2 table | |
1276 | * boundaries to keep things simple. | |
1277 | */ | |
c37f4cd7 KW |
1278 | nb_clusters = |
1279 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1280 | ||
f5bc6350 | 1281 | l2_index = offset_to_l2_index(s, guest_offset); |
c37f4cd7 | 1282 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
b6d36def | 1283 | assert(nb_clusters <= INT_MAX); |
f5bc6350 | 1284 | |
10f0ed8b KW |
1285 | /* Find L2 entry for the first involved cluster */ |
1286 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
1287 | if (ret < 0) { | |
1288 | return ret; | |
1289 | } | |
1290 | ||
3b8e2e26 | 1291 | entry = be64_to_cpu(l2_table[l2_index]); |
10f0ed8b KW |
1292 | |
1293 | /* For the moment, overwrite compressed clusters one by one */ | |
1294 | if (entry & QCOW_OFLAG_COMPRESSED) { | |
1295 | nb_clusters = 1; | |
1296 | } else { | |
3b8e2e26 | 1297 | nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index); |
10f0ed8b KW |
1298 | } |
1299 | ||
ecdd5333 KW |
1300 | /* This function is only called when there were no non-COW clusters, so if |
1301 | * we can't find any unallocated or COW clusters either, something is | |
1302 | * wrong with our code. */ | |
1303 | assert(nb_clusters > 0); | |
1304 | ||
fdfab37d EB |
1305 | if (qcow2_get_cluster_type(entry) == QCOW2_CLUSTER_ZERO_ALLOC && |
1306 | (entry & QCOW_OFLAG_COPIED) && | |
564a6b69 HR |
1307 | (!*host_offset || |
1308 | start_of_cluster(s, *host_offset) == (entry & L2E_OFFSET_MASK))) | |
1309 | { | |
93bbaf03 HR |
1310 | int preallocated_nb_clusters; |
1311 | ||
1312 | if (offset_into_cluster(s, entry & L2E_OFFSET_MASK)) { | |
1313 | qcow2_signal_corruption(bs, true, -1, -1, "Preallocated zero " | |
1314 | "cluster offset %#llx unaligned (guest " | |
1315 | "offset: %#" PRIx64 ")", | |
1316 | entry & L2E_OFFSET_MASK, guest_offset); | |
1317 | ret = -EIO; | |
1318 | goto fail; | |
1319 | } | |
1320 | ||
564a6b69 HR |
1321 | /* Try to reuse preallocated zero clusters; contiguous normal clusters |
1322 | * would be fine, too, but count_cow_clusters() above has limited | |
1323 | * nb_clusters already to a range of COW clusters */ | |
93bbaf03 | 1324 | preallocated_nb_clusters = |
564a6b69 HR |
1325 | count_contiguous_clusters(nb_clusters, s->cluster_size, |
1326 | &l2_table[l2_index], QCOW_OFLAG_COPIED); | |
1327 | assert(preallocated_nb_clusters > 0); | |
10f0ed8b | 1328 | |
564a6b69 HR |
1329 | nb_clusters = preallocated_nb_clusters; |
1330 | alloc_cluster_offset = entry & L2E_OFFSET_MASK; | |
10f0ed8b | 1331 | |
564a6b69 HR |
1332 | /* We want to reuse these clusters, so qcow2_alloc_cluster_link_l2() |
1333 | * should not free them. */ | |
1334 | keep_old_clusters = true; | |
10f0ed8b KW |
1335 | } |
1336 | ||
564a6b69 HR |
1337 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
1338 | ||
ff52aab2 | 1339 | if (!alloc_cluster_offset) { |
564a6b69 HR |
1340 | /* Allocate, if necessary at a given offset in the image file */ |
1341 | alloc_cluster_offset = start_of_cluster(s, *host_offset); | |
1342 | ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset, | |
1343 | &nb_clusters); | |
1344 | if (ret < 0) { | |
1345 | goto fail; | |
1346 | } | |
1347 | ||
1348 | /* Can't extend contiguous allocation */ | |
1349 | if (nb_clusters == 0) { | |
1350 | *bytes = 0; | |
1351 | return 0; | |
1352 | } | |
1353 | ||
1354 | /* !*host_offset would overwrite the image header and is reserved for | |
1355 | * "no host offset preferred". If 0 was a valid host offset, it'd | |
1356 | * trigger the following overlap check; do that now to avoid having an | |
1357 | * invalid value in *host_offset. */ | |
1358 | if (!alloc_cluster_offset) { | |
1359 | ret = qcow2_pre_write_overlap_check(bs, 0, alloc_cluster_offset, | |
1360 | nb_clusters * s->cluster_size); | |
1361 | assert(ret < 0); | |
1362 | goto fail; | |
1363 | } | |
ff52aab2 HR |
1364 | } |
1365 | ||
83baa9a4 KW |
1366 | /* |
1367 | * Save info needed for meta data update. | |
1368 | * | |
85567393 | 1369 | * requested_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1370 | * newly allocated cluster to the end of the (possibly shortened |
1371 | * before) write request. | |
1372 | * | |
85567393 | 1373 | * avail_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1374 | * newly allocated to the end of the last newly allocated cluster. |
1375 | * | |
85567393 | 1376 | * nb_bytes: The number of bytes from the start of the first |
83baa9a4 KW |
1377 | * newly allocated cluster to the end of the area that the write |
1378 | * request actually writes to (excluding COW at the end) | |
1379 | */ | |
85567393 KW |
1380 | uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset); |
1381 | int avail_bytes = MIN(INT_MAX, nb_clusters << s->cluster_bits); | |
1382 | int nb_bytes = MIN(requested_bytes, avail_bytes); | |
88c6588c | 1383 | QCowL2Meta *old_m = *m; |
83baa9a4 | 1384 | |
83baa9a4 KW |
1385 | *m = g_malloc0(sizeof(**m)); |
1386 | ||
1387 | **m = (QCowL2Meta) { | |
88c6588c KW |
1388 | .next = old_m, |
1389 | ||
411d62b0 | 1390 | .alloc_offset = alloc_cluster_offset, |
83baa9a4 KW |
1391 | .offset = start_of_cluster(s, guest_offset), |
1392 | .nb_clusters = nb_clusters, | |
83baa9a4 | 1393 | |
564a6b69 HR |
1394 | .keep_old_clusters = keep_old_clusters, |
1395 | ||
83baa9a4 KW |
1396 | .cow_start = { |
1397 | .offset = 0, | |
85567393 | 1398 | .nb_bytes = offset_into_cluster(s, guest_offset), |
83baa9a4 KW |
1399 | }, |
1400 | .cow_end = { | |
85567393 KW |
1401 | .offset = nb_bytes, |
1402 | .nb_bytes = avail_bytes - nb_bytes, | |
83baa9a4 KW |
1403 | }, |
1404 | }; | |
1405 | qemu_co_queue_init(&(*m)->dependent_requests); | |
1406 | QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); | |
1407 | ||
411d62b0 | 1408 | *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset); |
85567393 | 1409 | *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset)); |
83baa9a4 KW |
1410 | assert(*bytes != 0); |
1411 | ||
10f0ed8b KW |
1412 | return 1; |
1413 | ||
1414 | fail: | |
1415 | if (*m && (*m)->nb_clusters > 0) { | |
1416 | QLIST_REMOVE(*m, next_in_flight); | |
1417 | } | |
1418 | return ret; | |
1419 | } | |
1420 | ||
45aba42f KW |
1421 | /* |
1422 | * alloc_cluster_offset | |
1423 | * | |
250196f1 KW |
1424 | * For a given offset on the virtual disk, find the cluster offset in qcow2 |
1425 | * file. If the offset is not found, allocate a new cluster. | |
45aba42f | 1426 | * |
250196f1 | 1427 | * If the cluster was already allocated, m->nb_clusters is set to 0 and |
a7912369 | 1428 | * other fields in m are meaningless. |
148da7ea KW |
1429 | * |
1430 | * If the cluster is newly allocated, m->nb_clusters is set to the number of | |
68d100e9 KW |
1431 | * contiguous clusters that have been allocated. In this case, the other |
1432 | * fields of m are valid and contain information about the first allocated | |
1433 | * cluster. | |
45aba42f | 1434 | * |
68d100e9 KW |
1435 | * If the request conflicts with another write request in flight, the coroutine |
1436 | * is queued and will be reentered when the dependency has completed. | |
148da7ea KW |
1437 | * |
1438 | * Return 0 on success and -errno in error cases | |
45aba42f | 1439 | */ |
f4f0d391 | 1440 | int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, |
d46a0bb2 KW |
1441 | unsigned int *bytes, uint64_t *host_offset, |
1442 | QCowL2Meta **m) | |
45aba42f | 1443 | { |
ff99129a | 1444 | BDRVQcow2State *s = bs->opaque; |
710c2496 | 1445 | uint64_t start, remaining; |
250196f1 | 1446 | uint64_t cluster_offset; |
65eb2e35 | 1447 | uint64_t cur_bytes; |
710c2496 | 1448 | int ret; |
45aba42f | 1449 | |
d46a0bb2 | 1450 | trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *bytes); |
710c2496 | 1451 | |
72424114 | 1452 | again: |
16f0587e | 1453 | start = offset; |
d46a0bb2 | 1454 | remaining = *bytes; |
0af729ec KW |
1455 | cluster_offset = 0; |
1456 | *host_offset = 0; | |
ecdd5333 KW |
1457 | cur_bytes = 0; |
1458 | *m = NULL; | |
0af729ec | 1459 | |
2c3b32d2 | 1460 | while (true) { |
ecdd5333 KW |
1461 | |
1462 | if (!*host_offset) { | |
1463 | *host_offset = start_of_cluster(s, cluster_offset); | |
1464 | } | |
1465 | ||
1466 | assert(remaining >= cur_bytes); | |
1467 | ||
1468 | start += cur_bytes; | |
1469 | remaining -= cur_bytes; | |
1470 | cluster_offset += cur_bytes; | |
1471 | ||
1472 | if (remaining == 0) { | |
1473 | break; | |
1474 | } | |
1475 | ||
1476 | cur_bytes = remaining; | |
1477 | ||
2c3b32d2 KW |
1478 | /* |
1479 | * Now start gathering as many contiguous clusters as possible: | |
1480 | * | |
1481 | * 1. Check for overlaps with in-flight allocations | |
1482 | * | |
1483 | * a) Overlap not in the first cluster -> shorten this request and | |
1484 | * let the caller handle the rest in its next loop iteration. | |
1485 | * | |
1486 | * b) Real overlaps of two requests. Yield and restart the search | |
1487 | * for contiguous clusters (the situation could have changed | |
1488 | * while we were sleeping) | |
1489 | * | |
1490 | * c) TODO: Request starts in the same cluster as the in-flight | |
1491 | * allocation ends. Shorten the COW of the in-fight allocation, | |
1492 | * set cluster_offset to write to the same cluster and set up | |
1493 | * the right synchronisation between the in-flight request and | |
1494 | * the new one. | |
1495 | */ | |
ecdd5333 | 1496 | ret = handle_dependencies(bs, start, &cur_bytes, m); |
2c3b32d2 | 1497 | if (ret == -EAGAIN) { |
ecdd5333 KW |
1498 | /* Currently handle_dependencies() doesn't yield if we already had |
1499 | * an allocation. If it did, we would have to clean up the L2Meta | |
1500 | * structs before starting over. */ | |
1501 | assert(*m == NULL); | |
2c3b32d2 KW |
1502 | goto again; |
1503 | } else if (ret < 0) { | |
1504 | return ret; | |
ecdd5333 KW |
1505 | } else if (cur_bytes == 0) { |
1506 | break; | |
2c3b32d2 KW |
1507 | } else { |
1508 | /* handle_dependencies() may have decreased cur_bytes (shortened | |
1509 | * the allocations below) so that the next dependency is processed | |
1510 | * correctly during the next loop iteration. */ | |
0af729ec | 1511 | } |
710c2496 | 1512 | |
2c3b32d2 KW |
1513 | /* |
1514 | * 2. Count contiguous COPIED clusters. | |
1515 | */ | |
1516 | ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m); | |
1517 | if (ret < 0) { | |
1518 | return ret; | |
1519 | } else if (ret) { | |
ecdd5333 | 1520 | continue; |
2c3b32d2 KW |
1521 | } else if (cur_bytes == 0) { |
1522 | break; | |
1523 | } | |
060bee89 | 1524 | |
2c3b32d2 KW |
1525 | /* |
1526 | * 3. If the request still hasn't completed, allocate new clusters, | |
1527 | * considering any cluster_offset of steps 1c or 2. | |
1528 | */ | |
1529 | ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m); | |
1530 | if (ret < 0) { | |
1531 | return ret; | |
1532 | } else if (ret) { | |
ecdd5333 | 1533 | continue; |
2c3b32d2 KW |
1534 | } else { |
1535 | assert(cur_bytes == 0); | |
1536 | break; | |
1537 | } | |
f5bc6350 | 1538 | } |
10f0ed8b | 1539 | |
d46a0bb2 KW |
1540 | *bytes -= remaining; |
1541 | assert(*bytes > 0); | |
710c2496 | 1542 | assert(*host_offset != 0); |
45aba42f | 1543 | |
148da7ea | 1544 | return 0; |
45aba42f KW |
1545 | } |
1546 | ||
1547 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
1548 | const uint8_t *buf, int buf_size) | |
1549 | { | |
1550 | z_stream strm1, *strm = &strm1; | |
1551 | int ret, out_len; | |
1552 | ||
1553 | memset(strm, 0, sizeof(*strm)); | |
1554 | ||
1555 | strm->next_in = (uint8_t *)buf; | |
1556 | strm->avail_in = buf_size; | |
1557 | strm->next_out = out_buf; | |
1558 | strm->avail_out = out_buf_size; | |
1559 | ||
1560 | ret = inflateInit2(strm, -12); | |
1561 | if (ret != Z_OK) | |
1562 | return -1; | |
1563 | ret = inflate(strm, Z_FINISH); | |
1564 | out_len = strm->next_out - out_buf; | |
1565 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
1566 | out_len != out_buf_size) { | |
1567 | inflateEnd(strm); | |
1568 | return -1; | |
1569 | } | |
1570 | inflateEnd(strm); | |
1571 | return 0; | |
1572 | } | |
1573 | ||
66f82cee | 1574 | int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) |
45aba42f | 1575 | { |
ff99129a | 1576 | BDRVQcow2State *s = bs->opaque; |
45aba42f KW |
1577 | int ret, csize, nb_csectors, sector_offset; |
1578 | uint64_t coffset; | |
1579 | ||
1580 | coffset = cluster_offset & s->cluster_offset_mask; | |
1581 | if (s->cluster_cache_offset != coffset) { | |
1582 | nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1; | |
1583 | sector_offset = coffset & 511; | |
1584 | csize = nb_csectors * 512 - sector_offset; | |
3e4c7052 SH |
1585 | |
1586 | /* Allocate buffers on first decompress operation, most images are | |
1587 | * uncompressed and the memory overhead can be avoided. The buffers | |
1588 | * are freed in .bdrv_close(). | |
1589 | */ | |
1590 | if (!s->cluster_data) { | |
1591 | /* one more sector for decompressed data alignment */ | |
1592 | s->cluster_data = qemu_try_blockalign(bs->file->bs, | |
1593 | QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size + 512); | |
1594 | if (!s->cluster_data) { | |
1595 | return -ENOMEM; | |
1596 | } | |
1597 | } | |
1598 | if (!s->cluster_cache) { | |
1599 | s->cluster_cache = g_malloc(s->cluster_size); | |
1600 | } | |
1601 | ||
66f82cee | 1602 | BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
fbcbbf4e | 1603 | ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, |
9a4f4c31 | 1604 | nb_csectors); |
45aba42f | 1605 | if (ret < 0) { |
8af36488 | 1606 | return ret; |
45aba42f KW |
1607 | } |
1608 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
1609 | s->cluster_data + sector_offset, csize) < 0) { | |
8af36488 | 1610 | return -EIO; |
45aba42f KW |
1611 | } |
1612 | s->cluster_cache_offset = coffset; | |
1613 | } | |
1614 | return 0; | |
1615 | } | |
5ea929e3 KW |
1616 | |
1617 | /* | |
1618 | * This discards as many clusters of nb_clusters as possible at once (i.e. | |
1619 | * all clusters in the same L2 table) and returns the number of discarded | |
1620 | * clusters. | |
1621 | */ | |
1622 | static int discard_single_l2(BlockDriverState *bs, uint64_t offset, | |
b6d36def HR |
1623 | uint64_t nb_clusters, enum qcow2_discard_type type, |
1624 | bool full_discard) | |
5ea929e3 | 1625 | { |
ff99129a | 1626 | BDRVQcow2State *s = bs->opaque; |
3948d1d4 | 1627 | uint64_t *l2_table; |
5ea929e3 KW |
1628 | int l2_index; |
1629 | int ret; | |
1630 | int i; | |
1631 | ||
3948d1d4 | 1632 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
5ea929e3 KW |
1633 | if (ret < 0) { |
1634 | return ret; | |
1635 | } | |
1636 | ||
1637 | /* Limit nb_clusters to one L2 table */ | |
1638 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1639 | assert(nb_clusters <= INT_MAX); |
5ea929e3 KW |
1640 | |
1641 | for (i = 0; i < nb_clusters; i++) { | |
c883db0d | 1642 | uint64_t old_l2_entry; |
5ea929e3 | 1643 | |
c883db0d | 1644 | old_l2_entry = be64_to_cpu(l2_table[l2_index + i]); |
a71835a0 KW |
1645 | |
1646 | /* | |
808c4b6f HR |
1647 | * If full_discard is false, make sure that a discarded area reads back |
1648 | * as zeroes for v3 images (we cannot do it for v2 without actually | |
1649 | * writing a zero-filled buffer). We can skip the operation if the | |
1650 | * cluster is already marked as zero, or if it's unallocated and we | |
1651 | * don't have a backing file. | |
a71835a0 | 1652 | * |
237d78f8 | 1653 | * TODO We might want to use bdrv_block_status(bs) here, but we're |
a71835a0 | 1654 | * holding s->lock, so that doesn't work today. |
808c4b6f HR |
1655 | * |
1656 | * If full_discard is true, the sector should not read back as zeroes, | |
1657 | * but rather fall through to the backing file. | |
a71835a0 | 1658 | */ |
c883db0d | 1659 | switch (qcow2_get_cluster_type(old_l2_entry)) { |
bbd995d8 EB |
1660 | case QCOW2_CLUSTER_UNALLOCATED: |
1661 | if (full_discard || !bs->backing) { | |
1662 | continue; | |
1663 | } | |
1664 | break; | |
1665 | ||
fdfab37d EB |
1666 | case QCOW2_CLUSTER_ZERO_PLAIN: |
1667 | if (!full_discard) { | |
bbd995d8 EB |
1668 | continue; |
1669 | } | |
1670 | break; | |
1671 | ||
fdfab37d | 1672 | case QCOW2_CLUSTER_ZERO_ALLOC: |
bbd995d8 EB |
1673 | case QCOW2_CLUSTER_NORMAL: |
1674 | case QCOW2_CLUSTER_COMPRESSED: | |
1675 | break; | |
1676 | ||
1677 | default: | |
1678 | abort(); | |
5ea929e3 KW |
1679 | } |
1680 | ||
1681 | /* First remove L2 entries */ | |
72e80b89 | 1682 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
808c4b6f | 1683 | if (!full_discard && s->qcow_version >= 3) { |
a71835a0 KW |
1684 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
1685 | } else { | |
1686 | l2_table[l2_index + i] = cpu_to_be64(0); | |
1687 | } | |
5ea929e3 KW |
1688 | |
1689 | /* Then decrease the refcount */ | |
c883db0d | 1690 | qcow2_free_any_clusters(bs, old_l2_entry, 1, type); |
5ea929e3 KW |
1691 | } |
1692 | ||
a3f1afb4 | 1693 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
5ea929e3 KW |
1694 | |
1695 | return nb_clusters; | |
1696 | } | |
1697 | ||
d2cb36af EB |
1698 | int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset, |
1699 | uint64_t bytes, enum qcow2_discard_type type, | |
1700 | bool full_discard) | |
5ea929e3 | 1701 | { |
ff99129a | 1702 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1703 | uint64_t end_offset = offset + bytes; |
b6d36def | 1704 | uint64_t nb_clusters; |
d2cb36af | 1705 | int64_t cleared; |
5ea929e3 KW |
1706 | int ret; |
1707 | ||
f10ee139 | 1708 | /* Caller must pass aligned values, except at image end */ |
0c1bd469 | 1709 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); |
f10ee139 EB |
1710 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || |
1711 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
5ea929e3 | 1712 | |
d2cb36af | 1713 | nb_clusters = size_to_clusters(s, bytes); |
5ea929e3 | 1714 | |
0b919fae KW |
1715 | s->cache_discards = true; |
1716 | ||
5ea929e3 KW |
1717 | /* Each L2 table is handled by its own loop iteration */ |
1718 | while (nb_clusters > 0) { | |
d2cb36af EB |
1719 | cleared = discard_single_l2(bs, offset, nb_clusters, type, |
1720 | full_discard); | |
1721 | if (cleared < 0) { | |
1722 | ret = cleared; | |
0b919fae | 1723 | goto fail; |
5ea929e3 KW |
1724 | } |
1725 | ||
d2cb36af EB |
1726 | nb_clusters -= cleared; |
1727 | offset += (cleared * s->cluster_size); | |
5ea929e3 KW |
1728 | } |
1729 | ||
0b919fae KW |
1730 | ret = 0; |
1731 | fail: | |
1732 | s->cache_discards = false; | |
1733 | qcow2_process_discards(bs, ret); | |
1734 | ||
1735 | return ret; | |
5ea929e3 | 1736 | } |
621f0589 KW |
1737 | |
1738 | /* | |
1739 | * This zeroes as many clusters of nb_clusters as possible at once (i.e. | |
1740 | * all clusters in the same L2 table) and returns the number of zeroed | |
1741 | * clusters. | |
1742 | */ | |
1743 | static int zero_single_l2(BlockDriverState *bs, uint64_t offset, | |
170f4b2e | 1744 | uint64_t nb_clusters, int flags) |
621f0589 | 1745 | { |
ff99129a | 1746 | BDRVQcow2State *s = bs->opaque; |
621f0589 KW |
1747 | uint64_t *l2_table; |
1748 | int l2_index; | |
1749 | int ret; | |
1750 | int i; | |
06cc5e2b | 1751 | bool unmap = !!(flags & BDRV_REQ_MAY_UNMAP); |
621f0589 KW |
1752 | |
1753 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); | |
1754 | if (ret < 0) { | |
1755 | return ret; | |
1756 | } | |
1757 | ||
1758 | /* Limit nb_clusters to one L2 table */ | |
1759 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1760 | assert(nb_clusters <= INT_MAX); |
621f0589 KW |
1761 | |
1762 | for (i = 0; i < nb_clusters; i++) { | |
1763 | uint64_t old_offset; | |
06cc5e2b | 1764 | QCow2ClusterType cluster_type; |
621f0589 KW |
1765 | |
1766 | old_offset = be64_to_cpu(l2_table[l2_index + i]); | |
1767 | ||
06cc5e2b EB |
1768 | /* |
1769 | * Minimize L2 changes if the cluster already reads back as | |
1770 | * zeroes with correct allocation. | |
1771 | */ | |
1772 | cluster_type = qcow2_get_cluster_type(old_offset); | |
1773 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN || | |
1774 | (cluster_type == QCOW2_CLUSTER_ZERO_ALLOC && !unmap)) { | |
1775 | continue; | |
1776 | } | |
1777 | ||
72e80b89 | 1778 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
06cc5e2b | 1779 | if (cluster_type == QCOW2_CLUSTER_COMPRESSED || unmap) { |
621f0589 | 1780 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
6cfcb9b8 | 1781 | qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST); |
621f0589 KW |
1782 | } else { |
1783 | l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO); | |
1784 | } | |
1785 | } | |
1786 | ||
a3f1afb4 | 1787 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
621f0589 KW |
1788 | |
1789 | return nb_clusters; | |
1790 | } | |
1791 | ||
d2cb36af EB |
1792 | int qcow2_cluster_zeroize(BlockDriverState *bs, uint64_t offset, |
1793 | uint64_t bytes, int flags) | |
621f0589 | 1794 | { |
ff99129a | 1795 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1796 | uint64_t end_offset = offset + bytes; |
b6d36def | 1797 | uint64_t nb_clusters; |
d2cb36af | 1798 | int64_t cleared; |
621f0589 KW |
1799 | int ret; |
1800 | ||
f10ee139 EB |
1801 | /* Caller must pass aligned values, except at image end */ |
1802 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); | |
1803 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || | |
1804 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
1805 | ||
621f0589 KW |
1806 | /* The zero flag is only supported by version 3 and newer */ |
1807 | if (s->qcow_version < 3) { | |
1808 | return -ENOTSUP; | |
1809 | } | |
1810 | ||
1811 | /* Each L2 table is handled by its own loop iteration */ | |
d2cb36af | 1812 | nb_clusters = size_to_clusters(s, bytes); |
621f0589 | 1813 | |
0b919fae KW |
1814 | s->cache_discards = true; |
1815 | ||
621f0589 | 1816 | while (nb_clusters > 0) { |
d2cb36af EB |
1817 | cleared = zero_single_l2(bs, offset, nb_clusters, flags); |
1818 | if (cleared < 0) { | |
1819 | ret = cleared; | |
0b919fae | 1820 | goto fail; |
621f0589 KW |
1821 | } |
1822 | ||
d2cb36af EB |
1823 | nb_clusters -= cleared; |
1824 | offset += (cleared * s->cluster_size); | |
621f0589 KW |
1825 | } |
1826 | ||
0b919fae KW |
1827 | ret = 0; |
1828 | fail: | |
1829 | s->cache_discards = false; | |
1830 | qcow2_process_discards(bs, ret); | |
1831 | ||
1832 | return ret; | |
621f0589 | 1833 | } |
32b6444d HR |
1834 | |
1835 | /* | |
1836 | * Expands all zero clusters in a specific L1 table (or deallocates them, for | |
1837 | * non-backed non-pre-allocated zero clusters). | |
1838 | * | |
4057a2b2 HR |
1839 | * l1_entries and *visited_l1_entries are used to keep track of progress for |
1840 | * status_cb(). l1_entries contains the total number of L1 entries and | |
1841 | * *visited_l1_entries counts all visited L1 entries. | |
32b6444d HR |
1842 | */ |
1843 | static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table, | |
ecf58777 | 1844 | int l1_size, int64_t *visited_l1_entries, |
4057a2b2 | 1845 | int64_t l1_entries, |
8b13976d HR |
1846 | BlockDriverAmendStatusCB *status_cb, |
1847 | void *cb_opaque) | |
32b6444d | 1848 | { |
ff99129a | 1849 | BDRVQcow2State *s = bs->opaque; |
32b6444d HR |
1850 | bool is_active_l1 = (l1_table == s->l1_table); |
1851 | uint64_t *l2_table = NULL; | |
1852 | int ret; | |
1853 | int i, j; | |
1854 | ||
1855 | if (!is_active_l1) { | |
1856 | /* inactive L2 tables require a buffer to be stored in when loading | |
1857 | * them from disk */ | |
9a4f4c31 | 1858 | l2_table = qemu_try_blockalign(bs->file->bs, s->cluster_size); |
de82815d KW |
1859 | if (l2_table == NULL) { |
1860 | return -ENOMEM; | |
1861 | } | |
32b6444d HR |
1862 | } |
1863 | ||
1864 | for (i = 0; i < l1_size; i++) { | |
1865 | uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK; | |
1866 | bool l2_dirty = false; | |
0e06528e | 1867 | uint64_t l2_refcount; |
32b6444d HR |
1868 | |
1869 | if (!l2_offset) { | |
1870 | /* unallocated */ | |
4057a2b2 HR |
1871 | (*visited_l1_entries)++; |
1872 | if (status_cb) { | |
8b13976d | 1873 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 1874 | } |
32b6444d HR |
1875 | continue; |
1876 | } | |
1877 | ||
8dd93d93 HR |
1878 | if (offset_into_cluster(s, l2_offset)) { |
1879 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" | |
1880 | PRIx64 " unaligned (L1 index: %#x)", | |
1881 | l2_offset, i); | |
1882 | ret = -EIO; | |
1883 | goto fail; | |
1884 | } | |
1885 | ||
32b6444d HR |
1886 | if (is_active_l1) { |
1887 | /* get active L2 tables from cache */ | |
1888 | ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, | |
1889 | (void **)&l2_table); | |
1890 | } else { | |
1891 | /* load inactive L2 tables from disk */ | |
fbcbbf4e | 1892 | ret = bdrv_read(bs->file, l2_offset / BDRV_SECTOR_SIZE, |
9a4f4c31 | 1893 | (void *)l2_table, s->cluster_sectors); |
32b6444d HR |
1894 | } |
1895 | if (ret < 0) { | |
1896 | goto fail; | |
1897 | } | |
1898 | ||
7324c10f HR |
1899 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
1900 | &l2_refcount); | |
1901 | if (ret < 0) { | |
ecf58777 HR |
1902 | goto fail; |
1903 | } | |
1904 | ||
32b6444d HR |
1905 | for (j = 0; j < s->l2_size; j++) { |
1906 | uint64_t l2_entry = be64_to_cpu(l2_table[j]); | |
ecf58777 | 1907 | int64_t offset = l2_entry & L2E_OFFSET_MASK; |
3ef95218 | 1908 | QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry); |
32b6444d | 1909 | |
fdfab37d EB |
1910 | if (cluster_type != QCOW2_CLUSTER_ZERO_PLAIN && |
1911 | cluster_type != QCOW2_CLUSTER_ZERO_ALLOC) { | |
32b6444d HR |
1912 | continue; |
1913 | } | |
1914 | ||
fdfab37d | 1915 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
760e0063 | 1916 | if (!bs->backing) { |
32b6444d HR |
1917 | /* not backed; therefore we can simply deallocate the |
1918 | * cluster */ | |
1919 | l2_table[j] = 0; | |
1920 | l2_dirty = true; | |
1921 | continue; | |
1922 | } | |
1923 | ||
1924 | offset = qcow2_alloc_clusters(bs, s->cluster_size); | |
1925 | if (offset < 0) { | |
1926 | ret = offset; | |
1927 | goto fail; | |
1928 | } | |
ecf58777 HR |
1929 | |
1930 | if (l2_refcount > 1) { | |
1931 | /* For shared L2 tables, set the refcount accordingly (it is | |
1932 | * already 1 and needs to be l2_refcount) */ | |
1933 | ret = qcow2_update_cluster_refcount(bs, | |
2aabe7c7 HR |
1934 | offset >> s->cluster_bits, |
1935 | refcount_diff(1, l2_refcount), false, | |
ecf58777 HR |
1936 | QCOW2_DISCARD_OTHER); |
1937 | if (ret < 0) { | |
1938 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1939 | QCOW2_DISCARD_OTHER); | |
1940 | goto fail; | |
1941 | } | |
1942 | } | |
32b6444d HR |
1943 | } |
1944 | ||
8dd93d93 | 1945 | if (offset_into_cluster(s, offset)) { |
bcb07dba EB |
1946 | qcow2_signal_corruption(bs, true, -1, -1, |
1947 | "Cluster allocation offset " | |
8dd93d93 HR |
1948 | "%#" PRIx64 " unaligned (L2 offset: %#" |
1949 | PRIx64 ", L2 index: %#x)", offset, | |
1950 | l2_offset, j); | |
fdfab37d | 1951 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
8dd93d93 HR |
1952 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1953 | QCOW2_DISCARD_ALWAYS); | |
1954 | } | |
1955 | ret = -EIO; | |
1956 | goto fail; | |
1957 | } | |
1958 | ||
231bb267 | 1959 | ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size); |
32b6444d | 1960 | if (ret < 0) { |
fdfab37d | 1961 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
320c7066 HR |
1962 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1963 | QCOW2_DISCARD_ALWAYS); | |
1964 | } | |
32b6444d HR |
1965 | goto fail; |
1966 | } | |
1967 | ||
720ff280 | 1968 | ret = bdrv_pwrite_zeroes(bs->file, offset, s->cluster_size, 0); |
32b6444d | 1969 | if (ret < 0) { |
fdfab37d | 1970 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
320c7066 HR |
1971 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1972 | QCOW2_DISCARD_ALWAYS); | |
1973 | } | |
32b6444d HR |
1974 | goto fail; |
1975 | } | |
1976 | ||
ecf58777 HR |
1977 | if (l2_refcount == 1) { |
1978 | l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED); | |
1979 | } else { | |
1980 | l2_table[j] = cpu_to_be64(offset); | |
e390cf5a | 1981 | } |
ecf58777 | 1982 | l2_dirty = true; |
32b6444d HR |
1983 | } |
1984 | ||
1985 | if (is_active_l1) { | |
1986 | if (l2_dirty) { | |
72e80b89 | 1987 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
32b6444d HR |
1988 | qcow2_cache_depends_on_flush(s->l2_table_cache); |
1989 | } | |
a3f1afb4 | 1990 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
32b6444d HR |
1991 | } else { |
1992 | if (l2_dirty) { | |
231bb267 HR |
1993 | ret = qcow2_pre_write_overlap_check(bs, |
1994 | QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, l2_offset, | |
32b6444d HR |
1995 | s->cluster_size); |
1996 | if (ret < 0) { | |
1997 | goto fail; | |
1998 | } | |
1999 | ||
18d51c4b | 2000 | ret = bdrv_write(bs->file, l2_offset / BDRV_SECTOR_SIZE, |
9a4f4c31 | 2001 | (void *)l2_table, s->cluster_sectors); |
32b6444d HR |
2002 | if (ret < 0) { |
2003 | goto fail; | |
2004 | } | |
2005 | } | |
2006 | } | |
4057a2b2 HR |
2007 | |
2008 | (*visited_l1_entries)++; | |
2009 | if (status_cb) { | |
8b13976d | 2010 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 2011 | } |
32b6444d HR |
2012 | } |
2013 | ||
2014 | ret = 0; | |
2015 | ||
2016 | fail: | |
2017 | if (l2_table) { | |
2018 | if (!is_active_l1) { | |
2019 | qemu_vfree(l2_table); | |
2020 | } else { | |
a3f1afb4 | 2021 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
32b6444d HR |
2022 | } |
2023 | } | |
2024 | return ret; | |
2025 | } | |
2026 | ||
2027 | /* | |
2028 | * For backed images, expands all zero clusters on the image. For non-backed | |
2029 | * images, deallocates all non-pre-allocated zero clusters (and claims the | |
2030 | * allocation for pre-allocated ones). This is important for downgrading to a | |
2031 | * qcow2 version which doesn't yet support metadata zero clusters. | |
2032 | */ | |
4057a2b2 | 2033 | int qcow2_expand_zero_clusters(BlockDriverState *bs, |
8b13976d HR |
2034 | BlockDriverAmendStatusCB *status_cb, |
2035 | void *cb_opaque) | |
32b6444d | 2036 | { |
ff99129a | 2037 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 2038 | uint64_t *l1_table = NULL; |
4057a2b2 | 2039 | int64_t l1_entries = 0, visited_l1_entries = 0; |
32b6444d HR |
2040 | int ret; |
2041 | int i, j; | |
2042 | ||
4057a2b2 HR |
2043 | if (status_cb) { |
2044 | l1_entries = s->l1_size; | |
2045 | for (i = 0; i < s->nb_snapshots; i++) { | |
2046 | l1_entries += s->snapshots[i].l1_size; | |
2047 | } | |
2048 | } | |
2049 | ||
32b6444d | 2050 | ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size, |
4057a2b2 | 2051 | &visited_l1_entries, l1_entries, |
8b13976d | 2052 | status_cb, cb_opaque); |
32b6444d HR |
2053 | if (ret < 0) { |
2054 | goto fail; | |
2055 | } | |
2056 | ||
2057 | /* Inactive L1 tables may point to active L2 tables - therefore it is | |
2058 | * necessary to flush the L2 table cache before trying to access the L2 | |
2059 | * tables pointed to by inactive L1 entries (else we might try to expand | |
2060 | * zero clusters that have already been expanded); furthermore, it is also | |
2061 | * necessary to empty the L2 table cache, since it may contain tables which | |
2062 | * are now going to be modified directly on disk, bypassing the cache. | |
2063 | * qcow2_cache_empty() does both for us. */ | |
2064 | ret = qcow2_cache_empty(bs, s->l2_table_cache); | |
2065 | if (ret < 0) { | |
2066 | goto fail; | |
2067 | } | |
2068 | ||
2069 | for (i = 0; i < s->nb_snapshots; i++) { | |
d737b78c LV |
2070 | int l1_sectors = DIV_ROUND_UP(s->snapshots[i].l1_size * |
2071 | sizeof(uint64_t), BDRV_SECTOR_SIZE); | |
32b6444d HR |
2072 | |
2073 | l1_table = g_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE); | |
2074 | ||
fbcbbf4e | 2075 | ret = bdrv_read(bs->file, |
9a4f4c31 KW |
2076 | s->snapshots[i].l1_table_offset / BDRV_SECTOR_SIZE, |
2077 | (void *)l1_table, l1_sectors); | |
32b6444d HR |
2078 | if (ret < 0) { |
2079 | goto fail; | |
2080 | } | |
2081 | ||
2082 | for (j = 0; j < s->snapshots[i].l1_size; j++) { | |
2083 | be64_to_cpus(&l1_table[j]); | |
2084 | } | |
2085 | ||
2086 | ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size, | |
4057a2b2 | 2087 | &visited_l1_entries, l1_entries, |
8b13976d | 2088 | status_cb, cb_opaque); |
32b6444d HR |
2089 | if (ret < 0) { |
2090 | goto fail; | |
2091 | } | |
2092 | } | |
2093 | ||
2094 | ret = 0; | |
2095 | ||
2096 | fail: | |
32b6444d HR |
2097 | g_free(l1_table); |
2098 | return ret; | |
2099 | } |