]>
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 | ||
2013c3d4 | 320 | qcow2_cache_put(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); |
2d135ee9 | 327 | qcow2_cache_entry_mark_dirty(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 | 347 | if (l2_table != NULL) { |
2013c3d4 | 348 | qcow2_cache_put(s->l2_table_cache, (void **) table); |
8585afd8 | 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 | 542 | |
05b5b6ee | 543 | l1_index = offset_to_l1_index(s, offset); |
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 | ||
2013c3d4 | 622 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_table); |
29c1a730 | 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: | |
2013c3d4 | 640 | qcow2_cache_put(s->l2_table_cache, (void **)&l2_table); |
a97c67ee | 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 | * | |
d4c373b8 | 650 | * the cluster index in the l2 table is given to the caller. |
45aba42f | 651 | * |
1e3e8f1a | 652 | * Returns 0 on success, -errno in failure case |
45aba42f | 653 | */ |
45aba42f KW |
654 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
655 | uint64_t **new_l2_table, | |
45aba42f KW |
656 | int *new_l2_index) |
657 | { | |
ff99129a | 658 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
659 | unsigned int l2_index; |
660 | uint64_t l1_index, l2_offset; | |
c46e1167 | 661 | uint64_t *l2_table = NULL; |
80ee15a6 | 662 | int ret; |
45aba42f | 663 | |
b6af0975 | 664 | /* seek to the l2 offset in the l1 table */ |
45aba42f | 665 | |
05b5b6ee | 666 | l1_index = offset_to_l1_index(s, offset); |
45aba42f | 667 | if (l1_index >= s->l1_size) { |
72893756 | 668 | ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
1e3e8f1a KW |
669 | if (ret < 0) { |
670 | return ret; | |
671 | } | |
45aba42f | 672 | } |
8e37f681 | 673 | |
2cf7cfa1 | 674 | assert(l1_index < s->l1_size); |
8e37f681 | 675 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
a97c67ee HR |
676 | if (offset_into_cluster(s, l2_offset)) { |
677 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
678 | " unaligned (L1 index: %#" PRIx64 ")", | |
679 | l2_offset, l1_index); | |
680 | return -EIO; | |
681 | } | |
45aba42f KW |
682 | |
683 | /* seek the l2 table of the given l2 offset */ | |
684 | ||
8e37f681 | 685 | if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) { |
45aba42f | 686 | /* load the l2 table in memory */ |
55c17e98 KW |
687 | ret = l2_load(bs, l2_offset, &l2_table); |
688 | if (ret < 0) { | |
689 | return ret; | |
1e3e8f1a | 690 | } |
45aba42f | 691 | } else { |
16fde5f2 | 692 | /* First allocate a new L2 table (and do COW if needed) */ |
c46e1167 KW |
693 | ret = l2_allocate(bs, l1_index, &l2_table); |
694 | if (ret < 0) { | |
695 | return ret; | |
1e3e8f1a | 696 | } |
16fde5f2 KW |
697 | |
698 | /* Then decrease the refcount of the old table */ | |
699 | if (l2_offset) { | |
6cfcb9b8 KW |
700 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), |
701 | QCOW2_DISCARD_OTHER); | |
16fde5f2 | 702 | } |
45aba42f KW |
703 | } |
704 | ||
705 | /* find the cluster offset for the given disk offset */ | |
706 | ||
24990c5b | 707 | l2_index = offset_to_l2_index(s, offset); |
45aba42f KW |
708 | |
709 | *new_l2_table = l2_table; | |
45aba42f KW |
710 | *new_l2_index = l2_index; |
711 | ||
1e3e8f1a | 712 | return 0; |
45aba42f KW |
713 | } |
714 | ||
715 | /* | |
716 | * alloc_compressed_cluster_offset | |
717 | * | |
718 | * For a given offset of the disk image, return cluster offset in | |
719 | * qcow2 file. | |
720 | * | |
721 | * If the offset is not found, allocate a new compressed cluster. | |
722 | * | |
723 | * Return the cluster offset if successful, | |
724 | * Return 0, otherwise. | |
725 | * | |
726 | */ | |
727 | ||
ed6ccf0f KW |
728 | uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
729 | uint64_t offset, | |
730 | int compressed_size) | |
45aba42f | 731 | { |
ff99129a | 732 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 733 | int l2_index, ret; |
3948d1d4 | 734 | uint64_t *l2_table; |
f4f0d391 | 735 | int64_t cluster_offset; |
45aba42f KW |
736 | int nb_csectors; |
737 | ||
3948d1d4 | 738 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
1e3e8f1a | 739 | if (ret < 0) { |
45aba42f | 740 | return 0; |
1e3e8f1a | 741 | } |
45aba42f | 742 | |
b0b6862e KW |
743 | /* Compression can't overwrite anything. Fail if the cluster was already |
744 | * allocated. */ | |
45aba42f | 745 | cluster_offset = be64_to_cpu(l2_table[l2_index]); |
b0b6862e | 746 | if (cluster_offset & L2E_OFFSET_MASK) { |
2013c3d4 | 747 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_table); |
8f1efd00 KW |
748 | return 0; |
749 | } | |
45aba42f | 750 | |
ed6ccf0f | 751 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
5d757b56 | 752 | if (cluster_offset < 0) { |
2013c3d4 | 753 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_table); |
5d757b56 KW |
754 | return 0; |
755 | } | |
756 | ||
45aba42f KW |
757 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
758 | (cluster_offset >> 9); | |
759 | ||
760 | cluster_offset |= QCOW_OFLAG_COMPRESSED | | |
761 | ((uint64_t)nb_csectors << s->csize_shift); | |
762 | ||
763 | /* update L2 table */ | |
764 | ||
765 | /* compressed clusters never have the copied flag */ | |
766 | ||
66f82cee | 767 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
2d135ee9 | 768 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
45aba42f | 769 | l2_table[l2_index] = cpu_to_be64(cluster_offset); |
2013c3d4 | 770 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_table); |
4c1612d9 | 771 | |
29c1a730 | 772 | return cluster_offset; |
4c1612d9 KW |
773 | } |
774 | ||
99450c6f | 775 | static int perform_cow(BlockDriverState *bs, QCowL2Meta *m) |
593fb83c | 776 | { |
ff99129a | 777 | BDRVQcow2State *s = bs->opaque; |
99450c6f AG |
778 | Qcow2COWRegion *start = &m->cow_start; |
779 | Qcow2COWRegion *end = &m->cow_end; | |
672f0f2c | 780 | unsigned buffer_size; |
b3cf1c7c AG |
781 | unsigned data_bytes = end->offset - (start->offset + start->nb_bytes); |
782 | bool merge_reads; | |
672f0f2c | 783 | uint8_t *start_buffer, *end_buffer; |
86b862c4 | 784 | QEMUIOVector qiov; |
593fb83c KW |
785 | int ret; |
786 | ||
672f0f2c | 787 | assert(start->nb_bytes <= UINT_MAX - end->nb_bytes); |
b3cf1c7c AG |
788 | assert(start->nb_bytes + end->nb_bytes <= UINT_MAX - data_bytes); |
789 | assert(start->offset + start->nb_bytes <= end->offset); | |
ee22a9d8 | 790 | assert(!m->data_qiov || m->data_qiov->size == data_bytes); |
672f0f2c | 791 | |
99450c6f | 792 | if (start->nb_bytes == 0 && end->nb_bytes == 0) { |
593fb83c KW |
793 | return 0; |
794 | } | |
795 | ||
b3cf1c7c AG |
796 | /* If we have to read both the start and end COW regions and the |
797 | * middle region is not too large then perform just one read | |
798 | * operation */ | |
799 | merge_reads = start->nb_bytes && end->nb_bytes && data_bytes <= 16384; | |
800 | if (merge_reads) { | |
801 | buffer_size = start->nb_bytes + data_bytes + end->nb_bytes; | |
802 | } else { | |
803 | /* If we have to do two reads, add some padding in the middle | |
804 | * if necessary to make sure that the end region is optimally | |
805 | * aligned. */ | |
806 | size_t align = bdrv_opt_mem_align(bs); | |
807 | assert(align > 0 && align <= UINT_MAX); | |
808 | assert(QEMU_ALIGN_UP(start->nb_bytes, align) <= | |
809 | UINT_MAX - end->nb_bytes); | |
810 | buffer_size = QEMU_ALIGN_UP(start->nb_bytes, align) + end->nb_bytes; | |
811 | } | |
812 | ||
813 | /* Reserve a buffer large enough to store all the data that we're | |
814 | * going to read */ | |
672f0f2c AG |
815 | start_buffer = qemu_try_blockalign(bs, buffer_size); |
816 | if (start_buffer == NULL) { | |
817 | return -ENOMEM; | |
818 | } | |
819 | /* The part of the buffer where the end region is located */ | |
820 | end_buffer = start_buffer + buffer_size - end->nb_bytes; | |
821 | ||
ee22a9d8 | 822 | qemu_iovec_init(&qiov, 2 + (m->data_qiov ? m->data_qiov->niov : 0)); |
86b862c4 | 823 | |
593fb83c | 824 | qemu_co_mutex_unlock(&s->lock); |
b3cf1c7c AG |
825 | /* First we read the existing data from both COW regions. We |
826 | * either read the whole region in one go, or the start and end | |
827 | * regions separately. */ | |
828 | if (merge_reads) { | |
86b862c4 AG |
829 | qemu_iovec_add(&qiov, start_buffer, buffer_size); |
830 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c | 831 | } else { |
86b862c4 AG |
832 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); |
833 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c AG |
834 | if (ret < 0) { |
835 | goto fail; | |
836 | } | |
672f0f2c | 837 | |
86b862c4 AG |
838 | qemu_iovec_reset(&qiov); |
839 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
840 | ret = do_perform_cow_read(bs, m->offset, end->offset, &qiov); | |
b3cf1c7c | 841 | } |
593fb83c | 842 | if (ret < 0) { |
99450c6f | 843 | goto fail; |
593fb83c KW |
844 | } |
845 | ||
672f0f2c AG |
846 | /* Encrypt the data if necessary before writing it */ |
847 | if (bs->encrypted) { | |
4652b8f3 DB |
848 | if (!do_perform_cow_encrypt(bs, m->offset, m->alloc_offset, |
849 | start->offset, start_buffer, | |
850 | start->nb_bytes) || | |
851 | !do_perform_cow_encrypt(bs, m->offset, m->alloc_offset, | |
852 | end->offset, end_buffer, end->nb_bytes)) { | |
672f0f2c AG |
853 | ret = -EIO; |
854 | goto fail; | |
855 | } | |
856 | } | |
857 | ||
ee22a9d8 AG |
858 | /* And now we can write everything. If we have the guest data we |
859 | * can write everything in one single operation */ | |
860 | if (m->data_qiov) { | |
861 | qemu_iovec_reset(&qiov); | |
862 | if (start->nb_bytes) { | |
863 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
864 | } | |
865 | qemu_iovec_concat(&qiov, m->data_qiov, 0, data_bytes); | |
866 | if (end->nb_bytes) { | |
867 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
868 | } | |
869 | /* NOTE: we have a write_aio blkdebug event here followed by | |
870 | * a cow_write one in do_perform_cow_write(), but there's only | |
871 | * one single I/O operation */ | |
872 | BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); | |
873 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); | |
874 | } else { | |
875 | /* If there's no guest data then write both COW regions separately */ | |
876 | qemu_iovec_reset(&qiov); | |
877 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
878 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); | |
879 | if (ret < 0) { | |
880 | goto fail; | |
881 | } | |
882 | ||
883 | qemu_iovec_reset(&qiov); | |
884 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
885 | ret = do_perform_cow_write(bs, m->alloc_offset, end->offset, &qiov); | |
672f0f2c | 886 | } |
99450c6f AG |
887 | |
888 | fail: | |
889 | qemu_co_mutex_lock(&s->lock); | |
890 | ||
593fb83c KW |
891 | /* |
892 | * Before we update the L2 table to actually point to the new cluster, we | |
893 | * need to be sure that the refcounts have been increased and COW was | |
894 | * handled. | |
895 | */ | |
99450c6f AG |
896 | if (ret == 0) { |
897 | qcow2_cache_depends_on_flush(s->l2_table_cache); | |
898 | } | |
593fb83c | 899 | |
672f0f2c | 900 | qemu_vfree(start_buffer); |
86b862c4 | 901 | qemu_iovec_destroy(&qiov); |
99450c6f | 902 | return ret; |
593fb83c KW |
903 | } |
904 | ||
148da7ea | 905 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) |
45aba42f | 906 | { |
ff99129a | 907 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 908 | int i, j = 0, l2_index, ret; |
593fb83c | 909 | uint64_t *old_cluster, *l2_table; |
250196f1 | 910 | uint64_t cluster_offset = m->alloc_offset; |
45aba42f | 911 | |
3cce16f4 | 912 | trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
f50f88b9 | 913 | assert(m->nb_clusters > 0); |
45aba42f | 914 | |
5839e53b | 915 | old_cluster = g_try_new(uint64_t, m->nb_clusters); |
de82815d KW |
916 | if (old_cluster == NULL) { |
917 | ret = -ENOMEM; | |
918 | goto err; | |
919 | } | |
45aba42f KW |
920 | |
921 | /* copy content of unmodified sectors */ | |
99450c6f | 922 | ret = perform_cow(bs, m); |
593fb83c KW |
923 | if (ret < 0) { |
924 | goto err; | |
29c1a730 KW |
925 | } |
926 | ||
593fb83c | 927 | /* Update L2 table. */ |
74c4510a | 928 | if (s->use_lazy_refcounts) { |
280d3735 KW |
929 | qcow2_mark_dirty(bs); |
930 | } | |
bfe8043e SH |
931 | if (qcow2_need_accurate_refcounts(s)) { |
932 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
933 | s->refcount_block_cache); | |
934 | } | |
280d3735 | 935 | |
3948d1d4 | 936 | ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index); |
1e3e8f1a | 937 | if (ret < 0) { |
45aba42f | 938 | goto err; |
1e3e8f1a | 939 | } |
2d135ee9 | 940 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
45aba42f | 941 | |
c01dbccb | 942 | assert(l2_index + m->nb_clusters <= s->l2_size); |
45aba42f KW |
943 | for (i = 0; i < m->nb_clusters; i++) { |
944 | /* if two concurrent writes happen to the same unallocated cluster | |
aaa4d20b KW |
945 | * each write allocates separate cluster and writes data concurrently. |
946 | * The first one to complete updates l2 table with pointer to its | |
947 | * cluster the second one has to do RMW (which is done above by | |
948 | * perform_cow()), update l2 table with its cluster pointer and free | |
949 | * old cluster. This is what this loop does */ | |
950 | if (l2_table[l2_index + i] != 0) { | |
45aba42f | 951 | old_cluster[j++] = l2_table[l2_index + i]; |
aaa4d20b | 952 | } |
45aba42f KW |
953 | |
954 | l2_table[l2_index + i] = cpu_to_be64((cluster_offset + | |
955 | (i << s->cluster_bits)) | QCOW_OFLAG_COPIED); | |
956 | } | |
957 | ||
9f8e668e | 958 | |
2013c3d4 | 959 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_table); |
45aba42f | 960 | |
7ec5e6a4 KW |
961 | /* |
962 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
6cfcb9b8 KW |
963 | * |
964 | * Don't discard clusters that reach a refcount of 0 (e.g. compressed | |
965 | * clusters), the next write will reuse them anyway. | |
7ec5e6a4 | 966 | */ |
564a6b69 | 967 | if (!m->keep_old_clusters && j != 0) { |
7ec5e6a4 | 968 | for (i = 0; i < j; i++) { |
6cfcb9b8 KW |
969 | qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1, |
970 | QCOW2_DISCARD_NEVER); | |
7ec5e6a4 KW |
971 | } |
972 | } | |
45aba42f KW |
973 | |
974 | ret = 0; | |
975 | err: | |
7267c094 | 976 | g_free(old_cluster); |
45aba42f KW |
977 | return ret; |
978 | } | |
979 | ||
bf319ece KW |
980 | /* |
981 | * Returns the number of contiguous clusters that can be used for an allocating | |
982 | * write, but require COW to be performed (this includes yet unallocated space, | |
983 | * which must copy from the backing file) | |
984 | */ | |
ff99129a | 985 | static int count_cow_clusters(BDRVQcow2State *s, int nb_clusters, |
bf319ece KW |
986 | uint64_t *l2_table, int l2_index) |
987 | { | |
143550a8 | 988 | int i; |
bf319ece | 989 | |
143550a8 KW |
990 | for (i = 0; i < nb_clusters; i++) { |
991 | uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]); | |
3ef95218 | 992 | QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry); |
143550a8 KW |
993 | |
994 | switch(cluster_type) { | |
995 | case QCOW2_CLUSTER_NORMAL: | |
996 | if (l2_entry & QCOW_OFLAG_COPIED) { | |
997 | goto out; | |
998 | } | |
bf319ece | 999 | break; |
143550a8 KW |
1000 | case QCOW2_CLUSTER_UNALLOCATED: |
1001 | case QCOW2_CLUSTER_COMPRESSED: | |
fdfab37d EB |
1002 | case QCOW2_CLUSTER_ZERO_PLAIN: |
1003 | case QCOW2_CLUSTER_ZERO_ALLOC: | |
bf319ece | 1004 | break; |
143550a8 KW |
1005 | default: |
1006 | abort(); | |
1007 | } | |
bf319ece KW |
1008 | } |
1009 | ||
143550a8 | 1010 | out: |
bf319ece KW |
1011 | assert(i <= nb_clusters); |
1012 | return i; | |
1013 | } | |
1014 | ||
250196f1 | 1015 | /* |
226c3c26 KW |
1016 | * Check if there already is an AIO write request in flight which allocates |
1017 | * the same cluster. In this case we need to wait until the previous | |
1018 | * request has completed and updated the L2 table accordingly. | |
65eb2e35 KW |
1019 | * |
1020 | * Returns: | |
1021 | * 0 if there was no dependency. *cur_bytes indicates the number of | |
1022 | * bytes from guest_offset that can be read before the next | |
1023 | * dependency must be processed (or the request is complete) | |
1024 | * | |
1025 | * -EAGAIN if we had to wait for another request, previously gathered | |
1026 | * information on cluster allocation may be invalid now. The caller | |
1027 | * must start over anyway, so consider *cur_bytes undefined. | |
250196f1 | 1028 | */ |
226c3c26 | 1029 | static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, |
ecdd5333 | 1030 | uint64_t *cur_bytes, QCowL2Meta **m) |
250196f1 | 1031 | { |
ff99129a | 1032 | BDRVQcow2State *s = bs->opaque; |
250196f1 | 1033 | QCowL2Meta *old_alloc; |
65eb2e35 | 1034 | uint64_t bytes = *cur_bytes; |
250196f1 | 1035 | |
250196f1 KW |
1036 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
1037 | ||
65eb2e35 KW |
1038 | uint64_t start = guest_offset; |
1039 | uint64_t end = start + bytes; | |
1040 | uint64_t old_start = l2meta_cow_start(old_alloc); | |
1041 | uint64_t old_end = l2meta_cow_end(old_alloc); | |
250196f1 | 1042 | |
d9d74f41 | 1043 | if (end <= old_start || start >= old_end) { |
250196f1 KW |
1044 | /* No intersection */ |
1045 | } else { | |
1046 | if (start < old_start) { | |
1047 | /* Stop at the start of a running allocation */ | |
65eb2e35 | 1048 | bytes = old_start - start; |
250196f1 | 1049 | } else { |
65eb2e35 | 1050 | bytes = 0; |
250196f1 KW |
1051 | } |
1052 | ||
ecdd5333 KW |
1053 | /* Stop if already an l2meta exists. After yielding, it wouldn't |
1054 | * be valid any more, so we'd have to clean up the old L2Metas | |
1055 | * and deal with requests depending on them before starting to | |
1056 | * gather new ones. Not worth the trouble. */ | |
1057 | if (bytes == 0 && *m) { | |
1058 | *cur_bytes = 0; | |
1059 | return 0; | |
1060 | } | |
1061 | ||
65eb2e35 | 1062 | if (bytes == 0) { |
250196f1 KW |
1063 | /* Wait for the dependency to complete. We need to recheck |
1064 | * the free/allocated clusters when we continue. */ | |
1ace7cea | 1065 | qemu_co_queue_wait(&old_alloc->dependent_requests, &s->lock); |
250196f1 KW |
1066 | return -EAGAIN; |
1067 | } | |
1068 | } | |
1069 | } | |
1070 | ||
65eb2e35 KW |
1071 | /* Make sure that existing clusters and new allocations are only used up to |
1072 | * the next dependency if we shortened the request above */ | |
1073 | *cur_bytes = bytes; | |
250196f1 | 1074 | |
226c3c26 KW |
1075 | return 0; |
1076 | } | |
1077 | ||
0af729ec KW |
1078 | /* |
1079 | * Checks how many already allocated clusters that don't require a copy on | |
1080 | * write there are at the given guest_offset (up to *bytes). If | |
1081 | * *host_offset is not zero, only physically contiguous clusters beginning at | |
1082 | * this host offset are counted. | |
1083 | * | |
411d62b0 KW |
1084 | * Note that guest_offset may not be cluster aligned. In this case, the |
1085 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1086 | * therefore isn't cluster aligned as well. | |
0af729ec KW |
1087 | * |
1088 | * Returns: | |
1089 | * 0: if no allocated clusters are available at the given offset. | |
1090 | * *bytes is normally unchanged. It is set to 0 if the cluster | |
1091 | * is allocated and doesn't need COW, but doesn't have the right | |
1092 | * physical offset. | |
1093 | * | |
1094 | * 1: if allocated clusters that don't require a COW are available at | |
1095 | * the requested offset. *bytes may have decreased and describes | |
1096 | * the length of the area that can be written to. | |
1097 | * | |
1098 | * -errno: in error cases | |
0af729ec KW |
1099 | */ |
1100 | static int handle_copied(BlockDriverState *bs, uint64_t guest_offset, | |
c53ede9f | 1101 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
0af729ec | 1102 | { |
ff99129a | 1103 | BDRVQcow2State *s = bs->opaque; |
0af729ec KW |
1104 | int l2_index; |
1105 | uint64_t cluster_offset; | |
1106 | uint64_t *l2_table; | |
b6d36def | 1107 | uint64_t nb_clusters; |
c53ede9f | 1108 | unsigned int keep_clusters; |
a3f1afb4 | 1109 | int ret; |
0af729ec KW |
1110 | |
1111 | trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset, | |
1112 | *bytes); | |
0af729ec | 1113 | |
411d62b0 KW |
1114 | assert(*host_offset == 0 || offset_into_cluster(s, guest_offset) |
1115 | == offset_into_cluster(s, *host_offset)); | |
1116 | ||
acb0467f KW |
1117 | /* |
1118 | * Calculate the number of clusters to look for. We stop at L2 table | |
1119 | * boundaries to keep things simple. | |
1120 | */ | |
1121 | nb_clusters = | |
1122 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1123 | ||
1124 | l2_index = offset_to_l2_index(s, guest_offset); | |
1125 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1126 | assert(nb_clusters <= INT_MAX); |
acb0467f | 1127 | |
0af729ec KW |
1128 | /* Find L2 entry for the first involved cluster */ |
1129 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
1130 | if (ret < 0) { | |
1131 | return ret; | |
1132 | } | |
1133 | ||
1134 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
1135 | ||
1136 | /* Check how many clusters are already allocated and don't need COW */ | |
1137 | if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL | |
1138 | && (cluster_offset & QCOW_OFLAG_COPIED)) | |
1139 | { | |
e62daaf6 KW |
1140 | /* If a specific host_offset is required, check it */ |
1141 | bool offset_matches = | |
1142 | (cluster_offset & L2E_OFFSET_MASK) == *host_offset; | |
1143 | ||
a97c67ee HR |
1144 | if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) { |
1145 | qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset " | |
1146 | "%#llx unaligned (guest offset: %#" PRIx64 | |
1147 | ")", cluster_offset & L2E_OFFSET_MASK, | |
1148 | guest_offset); | |
1149 | ret = -EIO; | |
1150 | goto out; | |
1151 | } | |
1152 | ||
e62daaf6 KW |
1153 | if (*host_offset != 0 && !offset_matches) { |
1154 | *bytes = 0; | |
1155 | ret = 0; | |
1156 | goto out; | |
1157 | } | |
1158 | ||
0af729ec | 1159 | /* We keep all QCOW_OFLAG_COPIED clusters */ |
c53ede9f | 1160 | keep_clusters = |
acb0467f | 1161 | count_contiguous_clusters(nb_clusters, s->cluster_size, |
61653008 | 1162 | &l2_table[l2_index], |
0af729ec | 1163 | QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); |
c53ede9f KW |
1164 | assert(keep_clusters <= nb_clusters); |
1165 | ||
1166 | *bytes = MIN(*bytes, | |
1167 | keep_clusters * s->cluster_size | |
1168 | - offset_into_cluster(s, guest_offset)); | |
0af729ec KW |
1169 | |
1170 | ret = 1; | |
1171 | } else { | |
0af729ec KW |
1172 | ret = 0; |
1173 | } | |
1174 | ||
0af729ec | 1175 | /* Cleanup */ |
e62daaf6 | 1176 | out: |
2013c3d4 | 1177 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_table); |
0af729ec | 1178 | |
e62daaf6 KW |
1179 | /* Only return a host offset if we actually made progress. Otherwise we |
1180 | * would make requirements for handle_alloc() that it can't fulfill */ | |
a97c67ee | 1181 | if (ret > 0) { |
411d62b0 KW |
1182 | *host_offset = (cluster_offset & L2E_OFFSET_MASK) |
1183 | + offset_into_cluster(s, guest_offset); | |
e62daaf6 KW |
1184 | } |
1185 | ||
0af729ec KW |
1186 | return ret; |
1187 | } | |
1188 | ||
226c3c26 KW |
1189 | /* |
1190 | * Allocates new clusters for the given guest_offset. | |
1191 | * | |
1192 | * At most *nb_clusters are allocated, and on return *nb_clusters is updated to | |
1193 | * contain the number of clusters that have been allocated and are contiguous | |
1194 | * in the image file. | |
1195 | * | |
1196 | * If *host_offset is non-zero, it specifies the offset in the image file at | |
1197 | * which the new clusters must start. *nb_clusters can be 0 on return in this | |
1198 | * case if the cluster at host_offset is already in use. If *host_offset is | |
1199 | * zero, the clusters can be allocated anywhere in the image file. | |
1200 | * | |
1201 | * *host_offset is updated to contain the offset into the image file at which | |
1202 | * the first allocated cluster starts. | |
1203 | * | |
1204 | * Return 0 on success and -errno in error cases. -EAGAIN means that the | |
1205 | * function has been waiting for another request and the allocation must be | |
1206 | * restarted, but the whole request should not be failed. | |
1207 | */ | |
1208 | static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, | |
b6d36def | 1209 | uint64_t *host_offset, uint64_t *nb_clusters) |
226c3c26 | 1210 | { |
ff99129a | 1211 | BDRVQcow2State *s = bs->opaque; |
226c3c26 KW |
1212 | |
1213 | trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, | |
1214 | *host_offset, *nb_clusters); | |
1215 | ||
250196f1 KW |
1216 | /* Allocate new clusters */ |
1217 | trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); | |
1218 | if (*host_offset == 0) { | |
df021791 KW |
1219 | int64_t cluster_offset = |
1220 | qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); | |
1221 | if (cluster_offset < 0) { | |
1222 | return cluster_offset; | |
1223 | } | |
1224 | *host_offset = cluster_offset; | |
1225 | return 0; | |
250196f1 | 1226 | } else { |
b6d36def | 1227 | int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); |
df021791 KW |
1228 | if (ret < 0) { |
1229 | return ret; | |
1230 | } | |
1231 | *nb_clusters = ret; | |
1232 | return 0; | |
250196f1 | 1233 | } |
250196f1 KW |
1234 | } |
1235 | ||
10f0ed8b KW |
1236 | /* |
1237 | * Allocates new clusters for an area that either is yet unallocated or needs a | |
1238 | * copy on write. If *host_offset is non-zero, clusters are only allocated if | |
1239 | * the new allocation can match the specified host offset. | |
1240 | * | |
411d62b0 KW |
1241 | * Note that guest_offset may not be cluster aligned. In this case, the |
1242 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1243 | * therefore isn't cluster aligned as well. | |
10f0ed8b KW |
1244 | * |
1245 | * Returns: | |
1246 | * 0: if no clusters could be allocated. *bytes is set to 0, | |
1247 | * *host_offset is left unchanged. | |
1248 | * | |
1249 | * 1: if new clusters were allocated. *bytes may be decreased if the | |
1250 | * new allocation doesn't cover all of the requested area. | |
1251 | * *host_offset is updated to contain the host offset of the first | |
1252 | * newly allocated cluster. | |
1253 | * | |
1254 | * -errno: in error cases | |
10f0ed8b KW |
1255 | */ |
1256 | static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, | |
c37f4cd7 | 1257 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
10f0ed8b | 1258 | { |
ff99129a | 1259 | BDRVQcow2State *s = bs->opaque; |
10f0ed8b KW |
1260 | int l2_index; |
1261 | uint64_t *l2_table; | |
1262 | uint64_t entry; | |
b6d36def | 1263 | uint64_t nb_clusters; |
10f0ed8b | 1264 | int ret; |
564a6b69 | 1265 | bool keep_old_clusters = false; |
10f0ed8b | 1266 | |
564a6b69 | 1267 | uint64_t alloc_cluster_offset = 0; |
10f0ed8b KW |
1268 | |
1269 | trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset, | |
1270 | *bytes); | |
1271 | assert(*bytes > 0); | |
1272 | ||
f5bc6350 KW |
1273 | /* |
1274 | * Calculate the number of clusters to look for. We stop at L2 table | |
1275 | * boundaries to keep things simple. | |
1276 | */ | |
c37f4cd7 KW |
1277 | nb_clusters = |
1278 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1279 | ||
f5bc6350 | 1280 | l2_index = offset_to_l2_index(s, guest_offset); |
c37f4cd7 | 1281 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
b6d36def | 1282 | assert(nb_clusters <= INT_MAX); |
f5bc6350 | 1283 | |
10f0ed8b KW |
1284 | /* Find L2 entry for the first involved cluster */ |
1285 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
1286 | if (ret < 0) { | |
1287 | return ret; | |
1288 | } | |
1289 | ||
3b8e2e26 | 1290 | entry = be64_to_cpu(l2_table[l2_index]); |
10f0ed8b KW |
1291 | |
1292 | /* For the moment, overwrite compressed clusters one by one */ | |
1293 | if (entry & QCOW_OFLAG_COMPRESSED) { | |
1294 | nb_clusters = 1; | |
1295 | } else { | |
3b8e2e26 | 1296 | nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index); |
10f0ed8b KW |
1297 | } |
1298 | ||
ecdd5333 KW |
1299 | /* This function is only called when there were no non-COW clusters, so if |
1300 | * we can't find any unallocated or COW clusters either, something is | |
1301 | * wrong with our code. */ | |
1302 | assert(nb_clusters > 0); | |
1303 | ||
fdfab37d EB |
1304 | if (qcow2_get_cluster_type(entry) == QCOW2_CLUSTER_ZERO_ALLOC && |
1305 | (entry & QCOW_OFLAG_COPIED) && | |
564a6b69 HR |
1306 | (!*host_offset || |
1307 | start_of_cluster(s, *host_offset) == (entry & L2E_OFFSET_MASK))) | |
1308 | { | |
93bbaf03 HR |
1309 | int preallocated_nb_clusters; |
1310 | ||
1311 | if (offset_into_cluster(s, entry & L2E_OFFSET_MASK)) { | |
1312 | qcow2_signal_corruption(bs, true, -1, -1, "Preallocated zero " | |
1313 | "cluster offset %#llx unaligned (guest " | |
1314 | "offset: %#" PRIx64 ")", | |
1315 | entry & L2E_OFFSET_MASK, guest_offset); | |
1316 | ret = -EIO; | |
1317 | goto fail; | |
1318 | } | |
1319 | ||
564a6b69 HR |
1320 | /* Try to reuse preallocated zero clusters; contiguous normal clusters |
1321 | * would be fine, too, but count_cow_clusters() above has limited | |
1322 | * nb_clusters already to a range of COW clusters */ | |
93bbaf03 | 1323 | preallocated_nb_clusters = |
564a6b69 HR |
1324 | count_contiguous_clusters(nb_clusters, s->cluster_size, |
1325 | &l2_table[l2_index], QCOW_OFLAG_COPIED); | |
1326 | assert(preallocated_nb_clusters > 0); | |
10f0ed8b | 1327 | |
564a6b69 HR |
1328 | nb_clusters = preallocated_nb_clusters; |
1329 | alloc_cluster_offset = entry & L2E_OFFSET_MASK; | |
10f0ed8b | 1330 | |
564a6b69 HR |
1331 | /* We want to reuse these clusters, so qcow2_alloc_cluster_link_l2() |
1332 | * should not free them. */ | |
1333 | keep_old_clusters = true; | |
10f0ed8b KW |
1334 | } |
1335 | ||
2013c3d4 | 1336 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_table); |
564a6b69 | 1337 | |
ff52aab2 | 1338 | if (!alloc_cluster_offset) { |
564a6b69 HR |
1339 | /* Allocate, if necessary at a given offset in the image file */ |
1340 | alloc_cluster_offset = start_of_cluster(s, *host_offset); | |
1341 | ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset, | |
1342 | &nb_clusters); | |
1343 | if (ret < 0) { | |
1344 | goto fail; | |
1345 | } | |
1346 | ||
1347 | /* Can't extend contiguous allocation */ | |
1348 | if (nb_clusters == 0) { | |
1349 | *bytes = 0; | |
1350 | return 0; | |
1351 | } | |
1352 | ||
1353 | /* !*host_offset would overwrite the image header and is reserved for | |
1354 | * "no host offset preferred". If 0 was a valid host offset, it'd | |
1355 | * trigger the following overlap check; do that now to avoid having an | |
1356 | * invalid value in *host_offset. */ | |
1357 | if (!alloc_cluster_offset) { | |
1358 | ret = qcow2_pre_write_overlap_check(bs, 0, alloc_cluster_offset, | |
1359 | nb_clusters * s->cluster_size); | |
1360 | assert(ret < 0); | |
1361 | goto fail; | |
1362 | } | |
ff52aab2 HR |
1363 | } |
1364 | ||
83baa9a4 KW |
1365 | /* |
1366 | * Save info needed for meta data update. | |
1367 | * | |
85567393 | 1368 | * requested_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1369 | * newly allocated cluster to the end of the (possibly shortened |
1370 | * before) write request. | |
1371 | * | |
85567393 | 1372 | * avail_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1373 | * newly allocated to the end of the last newly allocated cluster. |
1374 | * | |
85567393 | 1375 | * nb_bytes: The number of bytes from the start of the first |
83baa9a4 KW |
1376 | * newly allocated cluster to the end of the area that the write |
1377 | * request actually writes to (excluding COW at the end) | |
1378 | */ | |
85567393 KW |
1379 | uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset); |
1380 | int avail_bytes = MIN(INT_MAX, nb_clusters << s->cluster_bits); | |
1381 | int nb_bytes = MIN(requested_bytes, avail_bytes); | |
88c6588c | 1382 | QCowL2Meta *old_m = *m; |
83baa9a4 | 1383 | |
83baa9a4 KW |
1384 | *m = g_malloc0(sizeof(**m)); |
1385 | ||
1386 | **m = (QCowL2Meta) { | |
88c6588c KW |
1387 | .next = old_m, |
1388 | ||
411d62b0 | 1389 | .alloc_offset = alloc_cluster_offset, |
83baa9a4 KW |
1390 | .offset = start_of_cluster(s, guest_offset), |
1391 | .nb_clusters = nb_clusters, | |
83baa9a4 | 1392 | |
564a6b69 HR |
1393 | .keep_old_clusters = keep_old_clusters, |
1394 | ||
83baa9a4 KW |
1395 | .cow_start = { |
1396 | .offset = 0, | |
85567393 | 1397 | .nb_bytes = offset_into_cluster(s, guest_offset), |
83baa9a4 KW |
1398 | }, |
1399 | .cow_end = { | |
85567393 KW |
1400 | .offset = nb_bytes, |
1401 | .nb_bytes = avail_bytes - nb_bytes, | |
83baa9a4 KW |
1402 | }, |
1403 | }; | |
1404 | qemu_co_queue_init(&(*m)->dependent_requests); | |
1405 | QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); | |
1406 | ||
411d62b0 | 1407 | *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset); |
85567393 | 1408 | *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset)); |
83baa9a4 KW |
1409 | assert(*bytes != 0); |
1410 | ||
10f0ed8b KW |
1411 | return 1; |
1412 | ||
1413 | fail: | |
1414 | if (*m && (*m)->nb_clusters > 0) { | |
1415 | QLIST_REMOVE(*m, next_in_flight); | |
1416 | } | |
1417 | return ret; | |
1418 | } | |
1419 | ||
45aba42f KW |
1420 | /* |
1421 | * alloc_cluster_offset | |
1422 | * | |
250196f1 KW |
1423 | * For a given offset on the virtual disk, find the cluster offset in qcow2 |
1424 | * file. If the offset is not found, allocate a new cluster. | |
45aba42f | 1425 | * |
250196f1 | 1426 | * If the cluster was already allocated, m->nb_clusters is set to 0 and |
a7912369 | 1427 | * other fields in m are meaningless. |
148da7ea KW |
1428 | * |
1429 | * If the cluster is newly allocated, m->nb_clusters is set to the number of | |
68d100e9 KW |
1430 | * contiguous clusters that have been allocated. In this case, the other |
1431 | * fields of m are valid and contain information about the first allocated | |
1432 | * cluster. | |
45aba42f | 1433 | * |
68d100e9 KW |
1434 | * If the request conflicts with another write request in flight, the coroutine |
1435 | * is queued and will be reentered when the dependency has completed. | |
148da7ea KW |
1436 | * |
1437 | * Return 0 on success and -errno in error cases | |
45aba42f | 1438 | */ |
f4f0d391 | 1439 | int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, |
d46a0bb2 KW |
1440 | unsigned int *bytes, uint64_t *host_offset, |
1441 | QCowL2Meta **m) | |
45aba42f | 1442 | { |
ff99129a | 1443 | BDRVQcow2State *s = bs->opaque; |
710c2496 | 1444 | uint64_t start, remaining; |
250196f1 | 1445 | uint64_t cluster_offset; |
65eb2e35 | 1446 | uint64_t cur_bytes; |
710c2496 | 1447 | int ret; |
45aba42f | 1448 | |
d46a0bb2 | 1449 | trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *bytes); |
710c2496 | 1450 | |
72424114 | 1451 | again: |
16f0587e | 1452 | start = offset; |
d46a0bb2 | 1453 | remaining = *bytes; |
0af729ec KW |
1454 | cluster_offset = 0; |
1455 | *host_offset = 0; | |
ecdd5333 KW |
1456 | cur_bytes = 0; |
1457 | *m = NULL; | |
0af729ec | 1458 | |
2c3b32d2 | 1459 | while (true) { |
ecdd5333 KW |
1460 | |
1461 | if (!*host_offset) { | |
1462 | *host_offset = start_of_cluster(s, cluster_offset); | |
1463 | } | |
1464 | ||
1465 | assert(remaining >= cur_bytes); | |
1466 | ||
1467 | start += cur_bytes; | |
1468 | remaining -= cur_bytes; | |
1469 | cluster_offset += cur_bytes; | |
1470 | ||
1471 | if (remaining == 0) { | |
1472 | break; | |
1473 | } | |
1474 | ||
1475 | cur_bytes = remaining; | |
1476 | ||
2c3b32d2 KW |
1477 | /* |
1478 | * Now start gathering as many contiguous clusters as possible: | |
1479 | * | |
1480 | * 1. Check for overlaps with in-flight allocations | |
1481 | * | |
1482 | * a) Overlap not in the first cluster -> shorten this request and | |
1483 | * let the caller handle the rest in its next loop iteration. | |
1484 | * | |
1485 | * b) Real overlaps of two requests. Yield and restart the search | |
1486 | * for contiguous clusters (the situation could have changed | |
1487 | * while we were sleeping) | |
1488 | * | |
1489 | * c) TODO: Request starts in the same cluster as the in-flight | |
1490 | * allocation ends. Shorten the COW of the in-fight allocation, | |
1491 | * set cluster_offset to write to the same cluster and set up | |
1492 | * the right synchronisation between the in-flight request and | |
1493 | * the new one. | |
1494 | */ | |
ecdd5333 | 1495 | ret = handle_dependencies(bs, start, &cur_bytes, m); |
2c3b32d2 | 1496 | if (ret == -EAGAIN) { |
ecdd5333 KW |
1497 | /* Currently handle_dependencies() doesn't yield if we already had |
1498 | * an allocation. If it did, we would have to clean up the L2Meta | |
1499 | * structs before starting over. */ | |
1500 | assert(*m == NULL); | |
2c3b32d2 KW |
1501 | goto again; |
1502 | } else if (ret < 0) { | |
1503 | return ret; | |
ecdd5333 KW |
1504 | } else if (cur_bytes == 0) { |
1505 | break; | |
2c3b32d2 KW |
1506 | } else { |
1507 | /* handle_dependencies() may have decreased cur_bytes (shortened | |
1508 | * the allocations below) so that the next dependency is processed | |
1509 | * correctly during the next loop iteration. */ | |
0af729ec | 1510 | } |
710c2496 | 1511 | |
2c3b32d2 KW |
1512 | /* |
1513 | * 2. Count contiguous COPIED clusters. | |
1514 | */ | |
1515 | ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m); | |
1516 | if (ret < 0) { | |
1517 | return ret; | |
1518 | } else if (ret) { | |
ecdd5333 | 1519 | continue; |
2c3b32d2 KW |
1520 | } else if (cur_bytes == 0) { |
1521 | break; | |
1522 | } | |
060bee89 | 1523 | |
2c3b32d2 KW |
1524 | /* |
1525 | * 3. If the request still hasn't completed, allocate new clusters, | |
1526 | * considering any cluster_offset of steps 1c or 2. | |
1527 | */ | |
1528 | ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m); | |
1529 | if (ret < 0) { | |
1530 | return ret; | |
1531 | } else if (ret) { | |
ecdd5333 | 1532 | continue; |
2c3b32d2 KW |
1533 | } else { |
1534 | assert(cur_bytes == 0); | |
1535 | break; | |
1536 | } | |
f5bc6350 | 1537 | } |
10f0ed8b | 1538 | |
d46a0bb2 KW |
1539 | *bytes -= remaining; |
1540 | assert(*bytes > 0); | |
710c2496 | 1541 | assert(*host_offset != 0); |
45aba42f | 1542 | |
148da7ea | 1543 | return 0; |
45aba42f KW |
1544 | } |
1545 | ||
1546 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
1547 | const uint8_t *buf, int buf_size) | |
1548 | { | |
1549 | z_stream strm1, *strm = &strm1; | |
1550 | int ret, out_len; | |
1551 | ||
1552 | memset(strm, 0, sizeof(*strm)); | |
1553 | ||
1554 | strm->next_in = (uint8_t *)buf; | |
1555 | strm->avail_in = buf_size; | |
1556 | strm->next_out = out_buf; | |
1557 | strm->avail_out = out_buf_size; | |
1558 | ||
1559 | ret = inflateInit2(strm, -12); | |
1560 | if (ret != Z_OK) | |
1561 | return -1; | |
1562 | ret = inflate(strm, Z_FINISH); | |
1563 | out_len = strm->next_out - out_buf; | |
1564 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
1565 | out_len != out_buf_size) { | |
1566 | inflateEnd(strm); | |
1567 | return -1; | |
1568 | } | |
1569 | inflateEnd(strm); | |
1570 | return 0; | |
1571 | } | |
1572 | ||
66f82cee | 1573 | int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) |
45aba42f | 1574 | { |
ff99129a | 1575 | BDRVQcow2State *s = bs->opaque; |
45aba42f KW |
1576 | int ret, csize, nb_csectors, sector_offset; |
1577 | uint64_t coffset; | |
1578 | ||
1579 | coffset = cluster_offset & s->cluster_offset_mask; | |
1580 | if (s->cluster_cache_offset != coffset) { | |
1581 | nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1; | |
1582 | sector_offset = coffset & 511; | |
1583 | csize = nb_csectors * 512 - sector_offset; | |
3e4c7052 SH |
1584 | |
1585 | /* Allocate buffers on first decompress operation, most images are | |
1586 | * uncompressed and the memory overhead can be avoided. The buffers | |
1587 | * are freed in .bdrv_close(). | |
1588 | */ | |
1589 | if (!s->cluster_data) { | |
1590 | /* one more sector for decompressed data alignment */ | |
1591 | s->cluster_data = qemu_try_blockalign(bs->file->bs, | |
1592 | QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size + 512); | |
1593 | if (!s->cluster_data) { | |
1594 | return -ENOMEM; | |
1595 | } | |
1596 | } | |
1597 | if (!s->cluster_cache) { | |
1598 | s->cluster_cache = g_malloc(s->cluster_size); | |
1599 | } | |
1600 | ||
66f82cee | 1601 | BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
fbcbbf4e | 1602 | ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, |
9a4f4c31 | 1603 | nb_csectors); |
45aba42f | 1604 | if (ret < 0) { |
8af36488 | 1605 | return ret; |
45aba42f KW |
1606 | } |
1607 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
1608 | s->cluster_data + sector_offset, csize) < 0) { | |
8af36488 | 1609 | return -EIO; |
45aba42f KW |
1610 | } |
1611 | s->cluster_cache_offset = coffset; | |
1612 | } | |
1613 | return 0; | |
1614 | } | |
5ea929e3 KW |
1615 | |
1616 | /* | |
1617 | * This discards as many clusters of nb_clusters as possible at once (i.e. | |
1618 | * all clusters in the same L2 table) and returns the number of discarded | |
1619 | * clusters. | |
1620 | */ | |
1621 | static int discard_single_l2(BlockDriverState *bs, uint64_t offset, | |
b6d36def HR |
1622 | uint64_t nb_clusters, enum qcow2_discard_type type, |
1623 | bool full_discard) | |
5ea929e3 | 1624 | { |
ff99129a | 1625 | BDRVQcow2State *s = bs->opaque; |
3948d1d4 | 1626 | uint64_t *l2_table; |
5ea929e3 KW |
1627 | int l2_index; |
1628 | int ret; | |
1629 | int i; | |
1630 | ||
3948d1d4 | 1631 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
5ea929e3 KW |
1632 | if (ret < 0) { |
1633 | return ret; | |
1634 | } | |
1635 | ||
1636 | /* Limit nb_clusters to one L2 table */ | |
1637 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1638 | assert(nb_clusters <= INT_MAX); |
5ea929e3 KW |
1639 | |
1640 | for (i = 0; i < nb_clusters; i++) { | |
c883db0d | 1641 | uint64_t old_l2_entry; |
5ea929e3 | 1642 | |
c883db0d | 1643 | old_l2_entry = be64_to_cpu(l2_table[l2_index + i]); |
a71835a0 KW |
1644 | |
1645 | /* | |
808c4b6f HR |
1646 | * If full_discard is false, make sure that a discarded area reads back |
1647 | * as zeroes for v3 images (we cannot do it for v2 without actually | |
1648 | * writing a zero-filled buffer). We can skip the operation if the | |
1649 | * cluster is already marked as zero, or if it's unallocated and we | |
1650 | * don't have a backing file. | |
a71835a0 | 1651 | * |
237d78f8 | 1652 | * TODO We might want to use bdrv_block_status(bs) here, but we're |
a71835a0 | 1653 | * holding s->lock, so that doesn't work today. |
808c4b6f HR |
1654 | * |
1655 | * If full_discard is true, the sector should not read back as zeroes, | |
1656 | * but rather fall through to the backing file. | |
a71835a0 | 1657 | */ |
c883db0d | 1658 | switch (qcow2_get_cluster_type(old_l2_entry)) { |
bbd995d8 EB |
1659 | case QCOW2_CLUSTER_UNALLOCATED: |
1660 | if (full_discard || !bs->backing) { | |
1661 | continue; | |
1662 | } | |
1663 | break; | |
1664 | ||
fdfab37d EB |
1665 | case QCOW2_CLUSTER_ZERO_PLAIN: |
1666 | if (!full_discard) { | |
bbd995d8 EB |
1667 | continue; |
1668 | } | |
1669 | break; | |
1670 | ||
fdfab37d | 1671 | case QCOW2_CLUSTER_ZERO_ALLOC: |
bbd995d8 EB |
1672 | case QCOW2_CLUSTER_NORMAL: |
1673 | case QCOW2_CLUSTER_COMPRESSED: | |
1674 | break; | |
1675 | ||
1676 | default: | |
1677 | abort(); | |
5ea929e3 KW |
1678 | } |
1679 | ||
1680 | /* First remove L2 entries */ | |
2d135ee9 | 1681 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
808c4b6f | 1682 | if (!full_discard && s->qcow_version >= 3) { |
a71835a0 KW |
1683 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
1684 | } else { | |
1685 | l2_table[l2_index + i] = cpu_to_be64(0); | |
1686 | } | |
5ea929e3 KW |
1687 | |
1688 | /* Then decrease the refcount */ | |
c883db0d | 1689 | qcow2_free_any_clusters(bs, old_l2_entry, 1, type); |
5ea929e3 KW |
1690 | } |
1691 | ||
2013c3d4 | 1692 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_table); |
5ea929e3 KW |
1693 | |
1694 | return nb_clusters; | |
1695 | } | |
1696 | ||
d2cb36af EB |
1697 | int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset, |
1698 | uint64_t bytes, enum qcow2_discard_type type, | |
1699 | bool full_discard) | |
5ea929e3 | 1700 | { |
ff99129a | 1701 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1702 | uint64_t end_offset = offset + bytes; |
b6d36def | 1703 | uint64_t nb_clusters; |
d2cb36af | 1704 | int64_t cleared; |
5ea929e3 KW |
1705 | int ret; |
1706 | ||
f10ee139 | 1707 | /* Caller must pass aligned values, except at image end */ |
0c1bd469 | 1708 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); |
f10ee139 EB |
1709 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || |
1710 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
5ea929e3 | 1711 | |
d2cb36af | 1712 | nb_clusters = size_to_clusters(s, bytes); |
5ea929e3 | 1713 | |
0b919fae KW |
1714 | s->cache_discards = true; |
1715 | ||
5ea929e3 KW |
1716 | /* Each L2 table is handled by its own loop iteration */ |
1717 | while (nb_clusters > 0) { | |
d2cb36af EB |
1718 | cleared = discard_single_l2(bs, offset, nb_clusters, type, |
1719 | full_discard); | |
1720 | if (cleared < 0) { | |
1721 | ret = cleared; | |
0b919fae | 1722 | goto fail; |
5ea929e3 KW |
1723 | } |
1724 | ||
d2cb36af EB |
1725 | nb_clusters -= cleared; |
1726 | offset += (cleared * s->cluster_size); | |
5ea929e3 KW |
1727 | } |
1728 | ||
0b919fae KW |
1729 | ret = 0; |
1730 | fail: | |
1731 | s->cache_discards = false; | |
1732 | qcow2_process_discards(bs, ret); | |
1733 | ||
1734 | return ret; | |
5ea929e3 | 1735 | } |
621f0589 KW |
1736 | |
1737 | /* | |
1738 | * This zeroes as many clusters of nb_clusters as possible at once (i.e. | |
1739 | * all clusters in the same L2 table) and returns the number of zeroed | |
1740 | * clusters. | |
1741 | */ | |
1742 | static int zero_single_l2(BlockDriverState *bs, uint64_t offset, | |
170f4b2e | 1743 | uint64_t nb_clusters, int flags) |
621f0589 | 1744 | { |
ff99129a | 1745 | BDRVQcow2State *s = bs->opaque; |
621f0589 KW |
1746 | uint64_t *l2_table; |
1747 | int l2_index; | |
1748 | int ret; | |
1749 | int i; | |
06cc5e2b | 1750 | bool unmap = !!(flags & BDRV_REQ_MAY_UNMAP); |
621f0589 KW |
1751 | |
1752 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); | |
1753 | if (ret < 0) { | |
1754 | return ret; | |
1755 | } | |
1756 | ||
1757 | /* Limit nb_clusters to one L2 table */ | |
1758 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1759 | assert(nb_clusters <= INT_MAX); |
621f0589 KW |
1760 | |
1761 | for (i = 0; i < nb_clusters; i++) { | |
1762 | uint64_t old_offset; | |
06cc5e2b | 1763 | QCow2ClusterType cluster_type; |
621f0589 KW |
1764 | |
1765 | old_offset = be64_to_cpu(l2_table[l2_index + i]); | |
1766 | ||
06cc5e2b EB |
1767 | /* |
1768 | * Minimize L2 changes if the cluster already reads back as | |
1769 | * zeroes with correct allocation. | |
1770 | */ | |
1771 | cluster_type = qcow2_get_cluster_type(old_offset); | |
1772 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN || | |
1773 | (cluster_type == QCOW2_CLUSTER_ZERO_ALLOC && !unmap)) { | |
1774 | continue; | |
1775 | } | |
1776 | ||
2d135ee9 | 1777 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
06cc5e2b | 1778 | if (cluster_type == QCOW2_CLUSTER_COMPRESSED || unmap) { |
621f0589 | 1779 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
6cfcb9b8 | 1780 | qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST); |
621f0589 KW |
1781 | } else { |
1782 | l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO); | |
1783 | } | |
1784 | } | |
1785 | ||
2013c3d4 | 1786 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_table); |
621f0589 KW |
1787 | |
1788 | return nb_clusters; | |
1789 | } | |
1790 | ||
d2cb36af EB |
1791 | int qcow2_cluster_zeroize(BlockDriverState *bs, uint64_t offset, |
1792 | uint64_t bytes, int flags) | |
621f0589 | 1793 | { |
ff99129a | 1794 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1795 | uint64_t end_offset = offset + bytes; |
b6d36def | 1796 | uint64_t nb_clusters; |
d2cb36af | 1797 | int64_t cleared; |
621f0589 KW |
1798 | int ret; |
1799 | ||
f10ee139 EB |
1800 | /* Caller must pass aligned values, except at image end */ |
1801 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); | |
1802 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || | |
1803 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
1804 | ||
621f0589 KW |
1805 | /* The zero flag is only supported by version 3 and newer */ |
1806 | if (s->qcow_version < 3) { | |
1807 | return -ENOTSUP; | |
1808 | } | |
1809 | ||
1810 | /* Each L2 table is handled by its own loop iteration */ | |
d2cb36af | 1811 | nb_clusters = size_to_clusters(s, bytes); |
621f0589 | 1812 | |
0b919fae KW |
1813 | s->cache_discards = true; |
1814 | ||
621f0589 | 1815 | while (nb_clusters > 0) { |
d2cb36af EB |
1816 | cleared = zero_single_l2(bs, offset, nb_clusters, flags); |
1817 | if (cleared < 0) { | |
1818 | ret = cleared; | |
0b919fae | 1819 | goto fail; |
621f0589 KW |
1820 | } |
1821 | ||
d2cb36af EB |
1822 | nb_clusters -= cleared; |
1823 | offset += (cleared * s->cluster_size); | |
621f0589 KW |
1824 | } |
1825 | ||
0b919fae KW |
1826 | ret = 0; |
1827 | fail: | |
1828 | s->cache_discards = false; | |
1829 | qcow2_process_discards(bs, ret); | |
1830 | ||
1831 | return ret; | |
621f0589 | 1832 | } |
32b6444d HR |
1833 | |
1834 | /* | |
1835 | * Expands all zero clusters in a specific L1 table (or deallocates them, for | |
1836 | * non-backed non-pre-allocated zero clusters). | |
1837 | * | |
4057a2b2 HR |
1838 | * l1_entries and *visited_l1_entries are used to keep track of progress for |
1839 | * status_cb(). l1_entries contains the total number of L1 entries and | |
1840 | * *visited_l1_entries counts all visited L1 entries. | |
32b6444d HR |
1841 | */ |
1842 | static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table, | |
ecf58777 | 1843 | int l1_size, int64_t *visited_l1_entries, |
4057a2b2 | 1844 | int64_t l1_entries, |
8b13976d HR |
1845 | BlockDriverAmendStatusCB *status_cb, |
1846 | void *cb_opaque) | |
32b6444d | 1847 | { |
ff99129a | 1848 | BDRVQcow2State *s = bs->opaque; |
32b6444d HR |
1849 | bool is_active_l1 = (l1_table == s->l1_table); |
1850 | uint64_t *l2_table = NULL; | |
1851 | int ret; | |
1852 | int i, j; | |
1853 | ||
1854 | if (!is_active_l1) { | |
1855 | /* inactive L2 tables require a buffer to be stored in when loading | |
1856 | * them from disk */ | |
9a4f4c31 | 1857 | l2_table = qemu_try_blockalign(bs->file->bs, s->cluster_size); |
de82815d KW |
1858 | if (l2_table == NULL) { |
1859 | return -ENOMEM; | |
1860 | } | |
32b6444d HR |
1861 | } |
1862 | ||
1863 | for (i = 0; i < l1_size; i++) { | |
1864 | uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK; | |
1865 | bool l2_dirty = false; | |
0e06528e | 1866 | uint64_t l2_refcount; |
32b6444d HR |
1867 | |
1868 | if (!l2_offset) { | |
1869 | /* unallocated */ | |
4057a2b2 HR |
1870 | (*visited_l1_entries)++; |
1871 | if (status_cb) { | |
8b13976d | 1872 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 1873 | } |
32b6444d HR |
1874 | continue; |
1875 | } | |
1876 | ||
8dd93d93 HR |
1877 | if (offset_into_cluster(s, l2_offset)) { |
1878 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" | |
1879 | PRIx64 " unaligned (L1 index: %#x)", | |
1880 | l2_offset, i); | |
1881 | ret = -EIO; | |
1882 | goto fail; | |
1883 | } | |
1884 | ||
32b6444d HR |
1885 | if (is_active_l1) { |
1886 | /* get active L2 tables from cache */ | |
1887 | ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, | |
1888 | (void **)&l2_table); | |
1889 | } else { | |
1890 | /* load inactive L2 tables from disk */ | |
fbcbbf4e | 1891 | ret = bdrv_read(bs->file, l2_offset / BDRV_SECTOR_SIZE, |
9a4f4c31 | 1892 | (void *)l2_table, s->cluster_sectors); |
32b6444d HR |
1893 | } |
1894 | if (ret < 0) { | |
1895 | goto fail; | |
1896 | } | |
1897 | ||
7324c10f HR |
1898 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
1899 | &l2_refcount); | |
1900 | if (ret < 0) { | |
ecf58777 HR |
1901 | goto fail; |
1902 | } | |
1903 | ||
32b6444d HR |
1904 | for (j = 0; j < s->l2_size; j++) { |
1905 | uint64_t l2_entry = be64_to_cpu(l2_table[j]); | |
ecf58777 | 1906 | int64_t offset = l2_entry & L2E_OFFSET_MASK; |
3ef95218 | 1907 | QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry); |
32b6444d | 1908 | |
fdfab37d EB |
1909 | if (cluster_type != QCOW2_CLUSTER_ZERO_PLAIN && |
1910 | cluster_type != QCOW2_CLUSTER_ZERO_ALLOC) { | |
32b6444d HR |
1911 | continue; |
1912 | } | |
1913 | ||
fdfab37d | 1914 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
760e0063 | 1915 | if (!bs->backing) { |
32b6444d HR |
1916 | /* not backed; therefore we can simply deallocate the |
1917 | * cluster */ | |
1918 | l2_table[j] = 0; | |
1919 | l2_dirty = true; | |
1920 | continue; | |
1921 | } | |
1922 | ||
1923 | offset = qcow2_alloc_clusters(bs, s->cluster_size); | |
1924 | if (offset < 0) { | |
1925 | ret = offset; | |
1926 | goto fail; | |
1927 | } | |
ecf58777 HR |
1928 | |
1929 | if (l2_refcount > 1) { | |
1930 | /* For shared L2 tables, set the refcount accordingly (it is | |
1931 | * already 1 and needs to be l2_refcount) */ | |
1932 | ret = qcow2_update_cluster_refcount(bs, | |
2aabe7c7 HR |
1933 | offset >> s->cluster_bits, |
1934 | refcount_diff(1, l2_refcount), false, | |
ecf58777 HR |
1935 | QCOW2_DISCARD_OTHER); |
1936 | if (ret < 0) { | |
1937 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1938 | QCOW2_DISCARD_OTHER); | |
1939 | goto fail; | |
1940 | } | |
1941 | } | |
32b6444d HR |
1942 | } |
1943 | ||
8dd93d93 | 1944 | if (offset_into_cluster(s, offset)) { |
bcb07dba EB |
1945 | qcow2_signal_corruption(bs, true, -1, -1, |
1946 | "Cluster allocation offset " | |
8dd93d93 HR |
1947 | "%#" PRIx64 " unaligned (L2 offset: %#" |
1948 | PRIx64 ", L2 index: %#x)", offset, | |
1949 | l2_offset, j); | |
fdfab37d | 1950 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
8dd93d93 HR |
1951 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1952 | QCOW2_DISCARD_ALWAYS); | |
1953 | } | |
1954 | ret = -EIO; | |
1955 | goto fail; | |
1956 | } | |
1957 | ||
231bb267 | 1958 | ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size); |
32b6444d | 1959 | if (ret < 0) { |
fdfab37d | 1960 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
320c7066 HR |
1961 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1962 | QCOW2_DISCARD_ALWAYS); | |
1963 | } | |
32b6444d HR |
1964 | goto fail; |
1965 | } | |
1966 | ||
720ff280 | 1967 | ret = bdrv_pwrite_zeroes(bs->file, offset, s->cluster_size, 0); |
32b6444d | 1968 | if (ret < 0) { |
fdfab37d | 1969 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
320c7066 HR |
1970 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1971 | QCOW2_DISCARD_ALWAYS); | |
1972 | } | |
32b6444d HR |
1973 | goto fail; |
1974 | } | |
1975 | ||
ecf58777 HR |
1976 | if (l2_refcount == 1) { |
1977 | l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED); | |
1978 | } else { | |
1979 | l2_table[j] = cpu_to_be64(offset); | |
e390cf5a | 1980 | } |
ecf58777 | 1981 | l2_dirty = true; |
32b6444d HR |
1982 | } |
1983 | ||
1984 | if (is_active_l1) { | |
1985 | if (l2_dirty) { | |
2d135ee9 | 1986 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table); |
32b6444d HR |
1987 | qcow2_cache_depends_on_flush(s->l2_table_cache); |
1988 | } | |
2013c3d4 | 1989 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_table); |
32b6444d HR |
1990 | } else { |
1991 | if (l2_dirty) { | |
231bb267 HR |
1992 | ret = qcow2_pre_write_overlap_check(bs, |
1993 | QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, l2_offset, | |
32b6444d HR |
1994 | s->cluster_size); |
1995 | if (ret < 0) { | |
1996 | goto fail; | |
1997 | } | |
1998 | ||
18d51c4b | 1999 | ret = bdrv_write(bs->file, l2_offset / BDRV_SECTOR_SIZE, |
9a4f4c31 | 2000 | (void *)l2_table, s->cluster_sectors); |
32b6444d HR |
2001 | if (ret < 0) { |
2002 | goto fail; | |
2003 | } | |
2004 | } | |
2005 | } | |
4057a2b2 HR |
2006 | |
2007 | (*visited_l1_entries)++; | |
2008 | if (status_cb) { | |
8b13976d | 2009 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 2010 | } |
32b6444d HR |
2011 | } |
2012 | ||
2013 | ret = 0; | |
2014 | ||
2015 | fail: | |
2016 | if (l2_table) { | |
2017 | if (!is_active_l1) { | |
2018 | qemu_vfree(l2_table); | |
2019 | } else { | |
2013c3d4 | 2020 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_table); |
32b6444d HR |
2021 | } |
2022 | } | |
2023 | return ret; | |
2024 | } | |
2025 | ||
2026 | /* | |
2027 | * For backed images, expands all zero clusters on the image. For non-backed | |
2028 | * images, deallocates all non-pre-allocated zero clusters (and claims the | |
2029 | * allocation for pre-allocated ones). This is important for downgrading to a | |
2030 | * qcow2 version which doesn't yet support metadata zero clusters. | |
2031 | */ | |
4057a2b2 | 2032 | int qcow2_expand_zero_clusters(BlockDriverState *bs, |
8b13976d HR |
2033 | BlockDriverAmendStatusCB *status_cb, |
2034 | void *cb_opaque) | |
32b6444d | 2035 | { |
ff99129a | 2036 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 2037 | uint64_t *l1_table = NULL; |
4057a2b2 | 2038 | int64_t l1_entries = 0, visited_l1_entries = 0; |
32b6444d HR |
2039 | int ret; |
2040 | int i, j; | |
2041 | ||
4057a2b2 HR |
2042 | if (status_cb) { |
2043 | l1_entries = s->l1_size; | |
2044 | for (i = 0; i < s->nb_snapshots; i++) { | |
2045 | l1_entries += s->snapshots[i].l1_size; | |
2046 | } | |
2047 | } | |
2048 | ||
32b6444d | 2049 | ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size, |
4057a2b2 | 2050 | &visited_l1_entries, l1_entries, |
8b13976d | 2051 | status_cb, cb_opaque); |
32b6444d HR |
2052 | if (ret < 0) { |
2053 | goto fail; | |
2054 | } | |
2055 | ||
2056 | /* Inactive L1 tables may point to active L2 tables - therefore it is | |
2057 | * necessary to flush the L2 table cache before trying to access the L2 | |
2058 | * tables pointed to by inactive L1 entries (else we might try to expand | |
2059 | * zero clusters that have already been expanded); furthermore, it is also | |
2060 | * necessary to empty the L2 table cache, since it may contain tables which | |
2061 | * are now going to be modified directly on disk, bypassing the cache. | |
2062 | * qcow2_cache_empty() does both for us. */ | |
2063 | ret = qcow2_cache_empty(bs, s->l2_table_cache); | |
2064 | if (ret < 0) { | |
2065 | goto fail; | |
2066 | } | |
2067 | ||
2068 | for (i = 0; i < s->nb_snapshots; i++) { | |
d737b78c LV |
2069 | int l1_sectors = DIV_ROUND_UP(s->snapshots[i].l1_size * |
2070 | sizeof(uint64_t), BDRV_SECTOR_SIZE); | |
32b6444d | 2071 | |
de7269d2 AG |
2072 | uint64_t *new_l1_table = |
2073 | g_try_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE); | |
2074 | ||
2075 | if (!new_l1_table) { | |
2076 | ret = -ENOMEM; | |
2077 | goto fail; | |
2078 | } | |
2079 | ||
2080 | l1_table = new_l1_table; | |
32b6444d | 2081 | |
fbcbbf4e | 2082 | ret = bdrv_read(bs->file, |
9a4f4c31 KW |
2083 | s->snapshots[i].l1_table_offset / BDRV_SECTOR_SIZE, |
2084 | (void *)l1_table, l1_sectors); | |
32b6444d HR |
2085 | if (ret < 0) { |
2086 | goto fail; | |
2087 | } | |
2088 | ||
2089 | for (j = 0; j < s->snapshots[i].l1_size; j++) { | |
2090 | be64_to_cpus(&l1_table[j]); | |
2091 | } | |
2092 | ||
2093 | ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size, | |
4057a2b2 | 2094 | &visited_l1_entries, l1_entries, |
8b13976d | 2095 | status_cb, cb_opaque); |
32b6444d HR |
2096 | if (ret < 0) { |
2097 | goto fail; | |
2098 | } | |
2099 | } | |
2100 | ||
2101 | ret = 0; | |
2102 | ||
2103 | fail: | |
32b6444d HR |
2104 | g_free(l1_table); |
2105 | return ret; | |
2106 | } |