]>
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, |
9e029689 | 129 | ROUND_UP(new_l1_size2, 512)); |
de82815d KW |
130 | if (new_l1_table == NULL) { |
131 | return -ENOMEM; | |
132 | } | |
9e029689 | 133 | memset(new_l1_table, 0, ROUND_UP(new_l1_size2, 512)); |
de82815d | 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 | * | |
e2b5713e AG |
198 | * @bs: The BlockDriverState |
199 | * @offset: A guest offset, used to calculate what slice of the L2 | |
200 | * table to load. | |
201 | * @l2_offset: Offset to the L2 table in the image file. | |
202 | * @l2_slice: Location to store the pointer to the L2 slice. | |
45aba42f | 203 | * |
e2b5713e AG |
204 | * Loads a L2 slice into memory (L2 slices are the parts of L2 tables |
205 | * that are loaded by the qcow2 cache). If the slice is in the cache, | |
206 | * the cache is used; otherwise the L2 slice is loaded from the image | |
207 | * file. | |
45aba42f | 208 | */ |
e2b5713e AG |
209 | static int l2_load(BlockDriverState *bs, uint64_t offset, |
210 | uint64_t l2_offset, uint64_t **l2_slice) | |
45aba42f | 211 | { |
ff99129a | 212 | BDRVQcow2State *s = bs->opaque; |
e2b5713e AG |
213 | int start_of_slice = sizeof(uint64_t) * |
214 | (offset_to_l2_index(s, offset) - offset_to_l2_slice_index(s, offset)); | |
45aba42f | 215 | |
e2b5713e AG |
216 | return qcow2_cache_get(bs, s->l2_table_cache, l2_offset + start_of_slice, |
217 | (void **)l2_slice); | |
45aba42f KW |
218 | } |
219 | ||
6583e3c7 KW |
220 | /* |
221 | * Writes one sector of the L1 table to the disk (can't update single entries | |
222 | * and we really don't want bdrv_pread to perform a read-modify-write) | |
223 | */ | |
224 | #define L1_ENTRIES_PER_SECTOR (512 / 8) | |
e23e400e | 225 | int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index) |
6583e3c7 | 226 | { |
ff99129a | 227 | BDRVQcow2State *s = bs->opaque; |
a1391444 | 228 | uint64_t buf[L1_ENTRIES_PER_SECTOR] = { 0 }; |
6583e3c7 | 229 | int l1_start_index; |
f7defcb6 | 230 | int i, ret; |
6583e3c7 KW |
231 | |
232 | l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1); | |
a1391444 HR |
233 | for (i = 0; i < L1_ENTRIES_PER_SECTOR && l1_start_index + i < s->l1_size; |
234 | i++) | |
235 | { | |
6583e3c7 KW |
236 | buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]); |
237 | } | |
238 | ||
231bb267 | 239 | ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1, |
cf93980e HR |
240 | s->l1_table_offset + 8 * l1_start_index, sizeof(buf)); |
241 | if (ret < 0) { | |
242 | return ret; | |
243 | } | |
244 | ||
66f82cee | 245 | BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); |
d9ca2ea2 | 246 | ret = bdrv_pwrite_sync(bs->file, |
9a4f4c31 KW |
247 | s->l1_table_offset + 8 * l1_start_index, |
248 | buf, sizeof(buf)); | |
f7defcb6 KW |
249 | if (ret < 0) { |
250 | return ret; | |
6583e3c7 KW |
251 | } |
252 | ||
253 | return 0; | |
254 | } | |
255 | ||
45aba42f KW |
256 | /* |
257 | * l2_allocate | |
258 | * | |
259 | * Allocate a new l2 entry in the file. If l1_index points to an already | |
260 | * used entry in the L2 table (i.e. we are doing a copy on write for the L2 | |
261 | * table) copy the contents of the old L2 table into the newly allocated one. | |
262 | * Otherwise the new table is initialized with zeros. | |
263 | * | |
264 | */ | |
265 | ||
3861946a | 266 | static int l2_allocate(BlockDriverState *bs, int l1_index) |
45aba42f | 267 | { |
ff99129a | 268 | BDRVQcow2State *s = bs->opaque; |
6583e3c7 | 269 | uint64_t old_l2_offset; |
3861946a AG |
270 | uint64_t *l2_slice = NULL; |
271 | unsigned slice, slice_size2, n_slices; | |
f4f0d391 | 272 | int64_t l2_offset; |
c46e1167 | 273 | int ret; |
45aba42f KW |
274 | |
275 | old_l2_offset = s->l1_table[l1_index]; | |
276 | ||
3cce16f4 KW |
277 | trace_qcow2_l2_allocate(bs, l1_index); |
278 | ||
45aba42f KW |
279 | /* allocate a new l2 entry */ |
280 | ||
ed6ccf0f | 281 | l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t)); |
5d757b56 | 282 | if (l2_offset < 0) { |
be0b742e HR |
283 | ret = l2_offset; |
284 | goto fail; | |
5d757b56 | 285 | } |
29c1a730 | 286 | |
98839750 AG |
287 | /* If we're allocating the table at offset 0 then something is wrong */ |
288 | if (l2_offset == 0) { | |
289 | qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid " | |
290 | "allocation of L2 table at offset 0"); | |
291 | ret = -EIO; | |
292 | goto fail; | |
293 | } | |
294 | ||
29c1a730 KW |
295 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
296 | if (ret < 0) { | |
297 | goto fail; | |
298 | } | |
45aba42f | 299 | |
45aba42f KW |
300 | /* allocate a new entry in the l2 cache */ |
301 | ||
3861946a AG |
302 | slice_size2 = s->l2_slice_size * sizeof(uint64_t); |
303 | n_slices = s->cluster_size / slice_size2; | |
304 | ||
3cce16f4 | 305 | trace_qcow2_l2_allocate_get_empty(bs, l1_index); |
3861946a | 306 | for (slice = 0; slice < n_slices; slice++) { |
6580bb09 | 307 | ret = qcow2_cache_get_empty(bs, s->l2_table_cache, |
3861946a AG |
308 | l2_offset + slice * slice_size2, |
309 | (void **) &l2_slice); | |
6580bb09 AG |
310 | if (ret < 0) { |
311 | goto fail; | |
312 | } | |
29c1a730 | 313 | |
6580bb09 | 314 | if ((old_l2_offset & L1E_OFFSET_MASK) == 0) { |
3861946a AG |
315 | /* if there was no old l2 table, clear the new slice */ |
316 | memset(l2_slice, 0, slice_size2); | |
6580bb09 | 317 | } else { |
3861946a AG |
318 | uint64_t *old_slice; |
319 | uint64_t old_l2_slice_offset = | |
320 | (old_l2_offset & L1E_OFFSET_MASK) + slice * slice_size2; | |
29c1a730 | 321 | |
3861946a | 322 | /* if there was an old l2 table, read a slice from the disk */ |
6580bb09 | 323 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ); |
3861946a AG |
324 | ret = qcow2_cache_get(bs, s->l2_table_cache, old_l2_slice_offset, |
325 | (void **) &old_slice); | |
6580bb09 AG |
326 | if (ret < 0) { |
327 | goto fail; | |
328 | } | |
329 | ||
3861946a | 330 | memcpy(l2_slice, old_slice, slice_size2); |
6580bb09 | 331 | |
3861946a | 332 | qcow2_cache_put(s->l2_table_cache, (void **) &old_slice); |
29c1a730 KW |
333 | } |
334 | ||
3861946a | 335 | /* write the l2 slice to the file */ |
6580bb09 | 336 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE); |
29c1a730 | 337 | |
6580bb09 | 338 | trace_qcow2_l2_allocate_write_l2(bs, l1_index); |
3861946a AG |
339 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
340 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
45aba42f | 341 | } |
29c1a730 | 342 | |
29c1a730 | 343 | ret = qcow2_cache_flush(bs, s->l2_table_cache); |
c46e1167 | 344 | if (ret < 0) { |
175e1152 KW |
345 | goto fail; |
346 | } | |
347 | ||
348 | /* update the L1 entry */ | |
3cce16f4 | 349 | trace_qcow2_l2_allocate_write_l1(bs, l1_index); |
175e1152 | 350 | s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; |
e23e400e | 351 | ret = qcow2_write_l1_entry(bs, l1_index); |
175e1152 KW |
352 | if (ret < 0) { |
353 | goto fail; | |
c46e1167 | 354 | } |
45aba42f | 355 | |
3cce16f4 | 356 | trace_qcow2_l2_allocate_done(bs, l1_index, 0); |
c46e1167 | 357 | return 0; |
175e1152 KW |
358 | |
359 | fail: | |
3cce16f4 | 360 | trace_qcow2_l2_allocate_done(bs, l1_index, ret); |
3861946a AG |
361 | if (l2_slice != NULL) { |
362 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
8585afd8 | 363 | } |
68dba0bf | 364 | s->l1_table[l1_index] = old_l2_offset; |
e3b21ef9 HR |
365 | if (l2_offset > 0) { |
366 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), | |
367 | QCOW2_DISCARD_ALWAYS); | |
368 | } | |
175e1152 | 369 | return ret; |
45aba42f KW |
370 | } |
371 | ||
2bfcc4a0 | 372 | /* |
13f893c4 | 373 | * Checks how many clusters in a given L2 slice are contiguous in the image |
2bfcc4a0 KW |
374 | * file. As soon as one of the flags in the bitmask stop_flags changes compared |
375 | * to the first cluster, the search is stopped and the cluster is not counted | |
376 | * as contiguous. (This allows it, for example, to stop at the first compressed | |
377 | * cluster which may require a different handling) | |
378 | */ | |
b6d36def | 379 | static int count_contiguous_clusters(int nb_clusters, int cluster_size, |
13f893c4 | 380 | uint64_t *l2_slice, uint64_t stop_flags) |
45aba42f KW |
381 | { |
382 | int i; | |
3ef95218 | 383 | QCow2ClusterType first_cluster_type; |
78a52ad5 | 384 | uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED; |
13f893c4 | 385 | uint64_t first_entry = be64_to_cpu(l2_slice[0]); |
15684a47 | 386 | uint64_t offset = first_entry & mask; |
45aba42f | 387 | |
564a6b69 | 388 | if (!offset) { |
45aba42f | 389 | return 0; |
564a6b69 | 390 | } |
45aba42f | 391 | |
564a6b69 HR |
392 | /* must be allocated */ |
393 | first_cluster_type = qcow2_get_cluster_type(first_entry); | |
394 | assert(first_cluster_type == QCOW2_CLUSTER_NORMAL || | |
fdfab37d | 395 | first_cluster_type == QCOW2_CLUSTER_ZERO_ALLOC); |
15684a47 | 396 | |
61653008 | 397 | for (i = 0; i < nb_clusters; i++) { |
13f893c4 | 398 | uint64_t l2_entry = be64_to_cpu(l2_slice[i]) & mask; |
2bfcc4a0 | 399 | if (offset + (uint64_t) i * cluster_size != l2_entry) { |
45aba42f | 400 | break; |
2bfcc4a0 KW |
401 | } |
402 | } | |
45aba42f | 403 | |
61653008 | 404 | return i; |
45aba42f KW |
405 | } |
406 | ||
4341df8a EB |
407 | /* |
408 | * Checks how many consecutive unallocated clusters in a given L2 | |
c26f10ba | 409 | * slice have the same cluster type. |
4341df8a EB |
410 | */ |
411 | static int count_contiguous_clusters_unallocated(int nb_clusters, | |
c26f10ba | 412 | uint64_t *l2_slice, |
3ef95218 | 413 | QCow2ClusterType wanted_type) |
45aba42f | 414 | { |
2bfcc4a0 KW |
415 | int i; |
416 | ||
fdfab37d | 417 | assert(wanted_type == QCOW2_CLUSTER_ZERO_PLAIN || |
4341df8a | 418 | wanted_type == QCOW2_CLUSTER_UNALLOCATED); |
2bfcc4a0 | 419 | for (i = 0; i < nb_clusters; i++) { |
c26f10ba | 420 | uint64_t entry = be64_to_cpu(l2_slice[i]); |
3ef95218 | 421 | QCow2ClusterType type = qcow2_get_cluster_type(entry); |
45aba42f | 422 | |
fdfab37d | 423 | if (type != wanted_type) { |
2bfcc4a0 KW |
424 | break; |
425 | } | |
426 | } | |
45aba42f KW |
427 | |
428 | return i; | |
429 | } | |
430 | ||
672f0f2c AG |
431 | static int coroutine_fn do_perform_cow_read(BlockDriverState *bs, |
432 | uint64_t src_cluster_offset, | |
433 | unsigned offset_in_cluster, | |
86b862c4 | 434 | QEMUIOVector *qiov) |
45aba42f | 435 | { |
aaa4d20b | 436 | int ret; |
1b9f1491 | 437 | |
86b862c4 | 438 | if (qiov->size == 0) { |
99450c6f AG |
439 | return 0; |
440 | } | |
441 | ||
66f82cee | 442 | BLKDBG_EVENT(bs->file, BLKDBG_COW_READ); |
aef4acb6 | 443 | |
dba28555 | 444 | if (!bs->drv) { |
672f0f2c | 445 | return -ENOMEDIUM; |
dba28555 HR |
446 | } |
447 | ||
aef4acb6 SH |
448 | /* Call .bdrv_co_readv() directly instead of using the public block-layer |
449 | * interface. This avoids double I/O throttling and request tracking, | |
450 | * which can lead to deadlock when block layer copy-on-read is enabled. | |
451 | */ | |
aaa4d20b | 452 | ret = bs->drv->bdrv_co_preadv(bs, src_cluster_offset + offset_in_cluster, |
86b862c4 | 453 | qiov->size, qiov, 0); |
1b9f1491 | 454 | if (ret < 0) { |
672f0f2c | 455 | return ret; |
1b9f1491 KW |
456 | } |
457 | ||
672f0f2c AG |
458 | return 0; |
459 | } | |
460 | ||
461 | static bool coroutine_fn do_perform_cow_encrypt(BlockDriverState *bs, | |
462 | uint64_t src_cluster_offset, | |
4652b8f3 | 463 | uint64_t cluster_offset, |
672f0f2c AG |
464 | unsigned offset_in_cluster, |
465 | uint8_t *buffer, | |
466 | unsigned bytes) | |
467 | { | |
468 | if (bytes && bs->encrypted) { | |
469 | BDRVQcow2State *s = bs->opaque; | |
4609742a | 470 | int64_t offset = (s->crypt_physical_offset ? |
4652b8f3 | 471 | (cluster_offset + offset_in_cluster) : |
4609742a | 472 | (src_cluster_offset + offset_in_cluster)); |
aaa4d20b KW |
473 | assert((offset_in_cluster & ~BDRV_SECTOR_MASK) == 0); |
474 | assert((bytes & ~BDRV_SECTOR_MASK) == 0); | |
b25b387f | 475 | assert(s->crypto); |
4609742a | 476 | if (qcrypto_block_encrypt(s->crypto, offset, buffer, bytes, NULL) < 0) { |
672f0f2c | 477 | return false; |
f6fa64f6 | 478 | } |
45aba42f | 479 | } |
672f0f2c AG |
480 | return true; |
481 | } | |
482 | ||
483 | static int coroutine_fn do_perform_cow_write(BlockDriverState *bs, | |
484 | uint64_t cluster_offset, | |
485 | unsigned offset_in_cluster, | |
86b862c4 | 486 | QEMUIOVector *qiov) |
672f0f2c | 487 | { |
672f0f2c AG |
488 | int ret; |
489 | ||
86b862c4 | 490 | if (qiov->size == 0) { |
672f0f2c AG |
491 | return 0; |
492 | } | |
493 | ||
231bb267 | 494 | ret = qcow2_pre_write_overlap_check(bs, 0, |
86b862c4 | 495 | cluster_offset + offset_in_cluster, qiov->size); |
cf93980e | 496 | if (ret < 0) { |
672f0f2c | 497 | return ret; |
cf93980e HR |
498 | } |
499 | ||
66f82cee | 500 | BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE); |
a03ef88f | 501 | ret = bdrv_co_pwritev(bs->file, cluster_offset + offset_in_cluster, |
86b862c4 | 502 | qiov->size, qiov, 0); |
1b9f1491 | 503 | if (ret < 0) { |
672f0f2c | 504 | return ret; |
1b9f1491 KW |
505 | } |
506 | ||
672f0f2c | 507 | return 0; |
45aba42f KW |
508 | } |
509 | ||
510 | ||
511 | /* | |
512 | * get_cluster_offset | |
513 | * | |
ecfe1863 KW |
514 | * For a given offset of the virtual disk, find the cluster type and offset in |
515 | * the qcow2 file. The offset is stored in *cluster_offset. | |
45aba42f | 516 | * |
ecfe1863 KW |
517 | * On entry, *bytes is the maximum number of contiguous bytes starting at |
518 | * offset that we are interested in. | |
45aba42f | 519 | * |
ecfe1863 KW |
520 | * On exit, *bytes is the number of bytes starting at offset that have the same |
521 | * cluster type and (if applicable) are stored contiguously in the image file. | |
522 | * Compressed clusters are always returned one by one. | |
45aba42f | 523 | * |
68d000a3 KW |
524 | * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error |
525 | * cases. | |
45aba42f | 526 | */ |
1c46efaa | 527 | int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, |
ecfe1863 | 528 | unsigned int *bytes, uint64_t *cluster_offset) |
45aba42f | 529 | { |
ff99129a | 530 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 | 531 | unsigned int l2_index; |
fd630039 AG |
532 | uint64_t l1_index, l2_offset, *l2_slice; |
533 | int c; | |
c834cba9 HR |
534 | unsigned int offset_in_cluster; |
535 | uint64_t bytes_available, bytes_needed, nb_clusters; | |
3ef95218 | 536 | QCow2ClusterType type; |
55c17e98 | 537 | int ret; |
45aba42f | 538 | |
b2f65d6b | 539 | offset_in_cluster = offset_into_cluster(s, offset); |
ecfe1863 | 540 | bytes_needed = (uint64_t) *bytes + offset_in_cluster; |
45aba42f | 541 | |
b2f65d6b | 542 | /* compute how many bytes there are between the start of the cluster |
fd630039 AG |
543 | * containing offset and the end of the l2 slice that contains |
544 | * the entry pointing to it */ | |
545 | bytes_available = | |
546 | ((uint64_t) (s->l2_slice_size - offset_to_l2_slice_index(s, offset))) | |
547 | << s->cluster_bits; | |
45aba42f | 548 | |
b2f65d6b KW |
549 | if (bytes_needed > bytes_available) { |
550 | bytes_needed = bytes_available; | |
45aba42f KW |
551 | } |
552 | ||
1c46efaa | 553 | *cluster_offset = 0; |
45aba42f | 554 | |
b6af0975 | 555 | /* seek to the l2 offset in the l1 table */ |
45aba42f | 556 | |
05b5b6ee | 557 | l1_index = offset_to_l1_index(s, offset); |
68d000a3 | 558 | if (l1_index >= s->l1_size) { |
3ef95218 | 559 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 560 | goto out; |
68d000a3 | 561 | } |
45aba42f | 562 | |
68d000a3 KW |
563 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
564 | if (!l2_offset) { | |
3ef95218 | 565 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 566 | goto out; |
68d000a3 | 567 | } |
45aba42f | 568 | |
a97c67ee HR |
569 | if (offset_into_cluster(s, l2_offset)) { |
570 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
571 | " unaligned (L1 index: %#" PRIx64 ")", | |
572 | l2_offset, l1_index); | |
573 | return -EIO; | |
574 | } | |
575 | ||
fd630039 | 576 | /* load the l2 slice in memory */ |
45aba42f | 577 | |
fd630039 | 578 | ret = l2_load(bs, offset, l2_offset, &l2_slice); |
55c17e98 KW |
579 | if (ret < 0) { |
580 | return ret; | |
1c46efaa | 581 | } |
45aba42f KW |
582 | |
583 | /* find the cluster offset for the given disk offset */ | |
584 | ||
fd630039 AG |
585 | l2_index = offset_to_l2_slice_index(s, offset); |
586 | *cluster_offset = be64_to_cpu(l2_slice[l2_index]); | |
b6d36def | 587 | |
b2f65d6b | 588 | nb_clusters = size_to_clusters(s, bytes_needed); |
c834cba9 HR |
589 | /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned |
590 | * integers; the minimum cluster size is 512, so this assertion is always | |
591 | * true */ | |
592 | assert(nb_clusters <= INT_MAX); | |
45aba42f | 593 | |
3ef95218 | 594 | type = qcow2_get_cluster_type(*cluster_offset); |
fdfab37d EB |
595 | if (s->qcow_version < 3 && (type == QCOW2_CLUSTER_ZERO_PLAIN || |
596 | type == QCOW2_CLUSTER_ZERO_ALLOC)) { | |
597 | qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found" | |
598 | " in pre-v3 image (L2 offset: %#" PRIx64 | |
599 | ", L2 index: %#x)", l2_offset, l2_index); | |
600 | ret = -EIO; | |
601 | goto fail; | |
602 | } | |
3ef95218 | 603 | switch (type) { |
68d000a3 KW |
604 | case QCOW2_CLUSTER_COMPRESSED: |
605 | /* Compressed clusters can only be processed one by one */ | |
606 | c = 1; | |
607 | *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK; | |
608 | break; | |
fdfab37d | 609 | case QCOW2_CLUSTER_ZERO_PLAIN: |
68d000a3 | 610 | case QCOW2_CLUSTER_UNALLOCATED: |
45aba42f | 611 | /* how many empty clusters ? */ |
4341df8a | 612 | c = count_contiguous_clusters_unallocated(nb_clusters, |
fd630039 | 613 | &l2_slice[l2_index], type); |
68d000a3 KW |
614 | *cluster_offset = 0; |
615 | break; | |
fdfab37d | 616 | case QCOW2_CLUSTER_ZERO_ALLOC: |
68d000a3 | 617 | case QCOW2_CLUSTER_NORMAL: |
45aba42f KW |
618 | /* how many allocated clusters ? */ |
619 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
fd630039 | 620 | &l2_slice[l2_index], QCOW_OFLAG_ZERO); |
68d000a3 | 621 | *cluster_offset &= L2E_OFFSET_MASK; |
a97c67ee | 622 | if (offset_into_cluster(s, *cluster_offset)) { |
fdfab37d EB |
623 | qcow2_signal_corruption(bs, true, -1, -1, |
624 | "Cluster allocation offset %#" | |
a97c67ee HR |
625 | PRIx64 " unaligned (L2 offset: %#" PRIx64 |
626 | ", L2 index: %#x)", *cluster_offset, | |
627 | l2_offset, l2_index); | |
628 | ret = -EIO; | |
629 | goto fail; | |
630 | } | |
68d000a3 | 631 | break; |
1417d7e4 KW |
632 | default: |
633 | abort(); | |
45aba42f KW |
634 | } |
635 | ||
fd630039 | 636 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
29c1a730 | 637 | |
c834cba9 | 638 | bytes_available = (int64_t)c * s->cluster_size; |
68d000a3 | 639 | |
45aba42f | 640 | out: |
b2f65d6b KW |
641 | if (bytes_available > bytes_needed) { |
642 | bytes_available = bytes_needed; | |
643 | } | |
45aba42f | 644 | |
c834cba9 HR |
645 | /* bytes_available <= bytes_needed <= *bytes + offset_in_cluster; |
646 | * subtracting offset_in_cluster will therefore definitely yield something | |
647 | * not exceeding UINT_MAX */ | |
648 | assert(bytes_available - offset_in_cluster <= UINT_MAX); | |
ecfe1863 | 649 | *bytes = bytes_available - offset_in_cluster; |
45aba42f | 650 | |
3ef95218 | 651 | return type; |
a97c67ee HR |
652 | |
653 | fail: | |
fd630039 | 654 | qcow2_cache_put(s->l2_table_cache, (void **)&l2_slice); |
a97c67ee | 655 | return ret; |
45aba42f KW |
656 | } |
657 | ||
658 | /* | |
659 | * get_cluster_table | |
660 | * | |
661 | * for a given disk offset, load (and allocate if needed) | |
c03bfc5b | 662 | * the appropriate slice of its l2 table. |
45aba42f | 663 | * |
c03bfc5b | 664 | * the cluster index in the l2 slice is given to the caller. |
45aba42f | 665 | * |
1e3e8f1a | 666 | * Returns 0 on success, -errno in failure case |
45aba42f | 667 | */ |
45aba42f | 668 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
c03bfc5b | 669 | uint64_t **new_l2_slice, |
45aba42f KW |
670 | int *new_l2_index) |
671 | { | |
ff99129a | 672 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
673 | unsigned int l2_index; |
674 | uint64_t l1_index, l2_offset; | |
c03bfc5b | 675 | uint64_t *l2_slice = NULL; |
80ee15a6 | 676 | int ret; |
45aba42f | 677 | |
b6af0975 | 678 | /* seek to the l2 offset in the l1 table */ |
45aba42f | 679 | |
05b5b6ee | 680 | l1_index = offset_to_l1_index(s, offset); |
45aba42f | 681 | if (l1_index >= s->l1_size) { |
72893756 | 682 | ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
1e3e8f1a KW |
683 | if (ret < 0) { |
684 | return ret; | |
685 | } | |
45aba42f | 686 | } |
8e37f681 | 687 | |
2cf7cfa1 | 688 | assert(l1_index < s->l1_size); |
8e37f681 | 689 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
a97c67ee HR |
690 | if (offset_into_cluster(s, l2_offset)) { |
691 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
692 | " unaligned (L1 index: %#" PRIx64 ")", | |
693 | l2_offset, l1_index); | |
694 | return -EIO; | |
695 | } | |
45aba42f | 696 | |
05f9ee46 | 697 | if (!(s->l1_table[l1_index] & QCOW_OFLAG_COPIED)) { |
16fde5f2 | 698 | /* First allocate a new L2 table (and do COW if needed) */ |
3861946a | 699 | ret = l2_allocate(bs, l1_index); |
c46e1167 KW |
700 | if (ret < 0) { |
701 | return ret; | |
1e3e8f1a | 702 | } |
16fde5f2 KW |
703 | |
704 | /* Then decrease the refcount of the old table */ | |
705 | if (l2_offset) { | |
6cfcb9b8 KW |
706 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), |
707 | QCOW2_DISCARD_OTHER); | |
16fde5f2 | 708 | } |
3861946a AG |
709 | |
710 | /* Get the offset of the newly-allocated l2 table */ | |
711 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; | |
712 | assert(offset_into_cluster(s, l2_offset) == 0); | |
05f9ee46 AG |
713 | } |
714 | ||
c03bfc5b AG |
715 | /* load the l2 slice in memory */ |
716 | ret = l2_load(bs, offset, l2_offset, &l2_slice); | |
05f9ee46 AG |
717 | if (ret < 0) { |
718 | return ret; | |
45aba42f KW |
719 | } |
720 | ||
721 | /* find the cluster offset for the given disk offset */ | |
722 | ||
c03bfc5b | 723 | l2_index = offset_to_l2_slice_index(s, offset); |
45aba42f | 724 | |
c03bfc5b | 725 | *new_l2_slice = l2_slice; |
45aba42f KW |
726 | *new_l2_index = l2_index; |
727 | ||
1e3e8f1a | 728 | return 0; |
45aba42f KW |
729 | } |
730 | ||
731 | /* | |
732 | * alloc_compressed_cluster_offset | |
733 | * | |
734 | * For a given offset of the disk image, return cluster offset in | |
735 | * qcow2 file. | |
736 | * | |
737 | * If the offset is not found, allocate a new compressed cluster. | |
738 | * | |
739 | * Return the cluster offset if successful, | |
740 | * Return 0, otherwise. | |
741 | * | |
742 | */ | |
743 | ||
ed6ccf0f KW |
744 | uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
745 | uint64_t offset, | |
746 | int compressed_size) | |
45aba42f | 747 | { |
ff99129a | 748 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 749 | int l2_index, ret; |
e4e72548 | 750 | uint64_t *l2_slice; |
f4f0d391 | 751 | int64_t cluster_offset; |
45aba42f KW |
752 | int nb_csectors; |
753 | ||
e4e72548 | 754 | ret = get_cluster_table(bs, offset, &l2_slice, &l2_index); |
1e3e8f1a | 755 | if (ret < 0) { |
45aba42f | 756 | return 0; |
1e3e8f1a | 757 | } |
45aba42f | 758 | |
b0b6862e KW |
759 | /* Compression can't overwrite anything. Fail if the cluster was already |
760 | * allocated. */ | |
e4e72548 | 761 | cluster_offset = be64_to_cpu(l2_slice[l2_index]); |
b0b6862e | 762 | if (cluster_offset & L2E_OFFSET_MASK) { |
e4e72548 | 763 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
8f1efd00 KW |
764 | return 0; |
765 | } | |
45aba42f | 766 | |
ed6ccf0f | 767 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
5d757b56 | 768 | if (cluster_offset < 0) { |
e4e72548 | 769 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
5d757b56 KW |
770 | return 0; |
771 | } | |
772 | ||
45aba42f KW |
773 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
774 | (cluster_offset >> 9); | |
775 | ||
776 | cluster_offset |= QCOW_OFLAG_COMPRESSED | | |
777 | ((uint64_t)nb_csectors << s->csize_shift); | |
778 | ||
779 | /* update L2 table */ | |
780 | ||
781 | /* compressed clusters never have the copied flag */ | |
782 | ||
66f82cee | 783 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
e4e72548 AG |
784 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
785 | l2_slice[l2_index] = cpu_to_be64(cluster_offset); | |
786 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
4c1612d9 | 787 | |
29c1a730 | 788 | return cluster_offset; |
4c1612d9 KW |
789 | } |
790 | ||
99450c6f | 791 | static int perform_cow(BlockDriverState *bs, QCowL2Meta *m) |
593fb83c | 792 | { |
ff99129a | 793 | BDRVQcow2State *s = bs->opaque; |
99450c6f AG |
794 | Qcow2COWRegion *start = &m->cow_start; |
795 | Qcow2COWRegion *end = &m->cow_end; | |
672f0f2c | 796 | unsigned buffer_size; |
b3cf1c7c AG |
797 | unsigned data_bytes = end->offset - (start->offset + start->nb_bytes); |
798 | bool merge_reads; | |
672f0f2c | 799 | uint8_t *start_buffer, *end_buffer; |
86b862c4 | 800 | QEMUIOVector qiov; |
593fb83c KW |
801 | int ret; |
802 | ||
672f0f2c | 803 | assert(start->nb_bytes <= UINT_MAX - end->nb_bytes); |
b3cf1c7c AG |
804 | assert(start->nb_bytes + end->nb_bytes <= UINT_MAX - data_bytes); |
805 | assert(start->offset + start->nb_bytes <= end->offset); | |
ee22a9d8 | 806 | assert(!m->data_qiov || m->data_qiov->size == data_bytes); |
672f0f2c | 807 | |
99450c6f | 808 | if (start->nb_bytes == 0 && end->nb_bytes == 0) { |
593fb83c KW |
809 | return 0; |
810 | } | |
811 | ||
b3cf1c7c AG |
812 | /* If we have to read both the start and end COW regions and the |
813 | * middle region is not too large then perform just one read | |
814 | * operation */ | |
815 | merge_reads = start->nb_bytes && end->nb_bytes && data_bytes <= 16384; | |
816 | if (merge_reads) { | |
817 | buffer_size = start->nb_bytes + data_bytes + end->nb_bytes; | |
818 | } else { | |
819 | /* If we have to do two reads, add some padding in the middle | |
820 | * if necessary to make sure that the end region is optimally | |
821 | * aligned. */ | |
822 | size_t align = bdrv_opt_mem_align(bs); | |
823 | assert(align > 0 && align <= UINT_MAX); | |
824 | assert(QEMU_ALIGN_UP(start->nb_bytes, align) <= | |
825 | UINT_MAX - end->nb_bytes); | |
826 | buffer_size = QEMU_ALIGN_UP(start->nb_bytes, align) + end->nb_bytes; | |
827 | } | |
828 | ||
829 | /* Reserve a buffer large enough to store all the data that we're | |
830 | * going to read */ | |
672f0f2c AG |
831 | start_buffer = qemu_try_blockalign(bs, buffer_size); |
832 | if (start_buffer == NULL) { | |
833 | return -ENOMEM; | |
834 | } | |
835 | /* The part of the buffer where the end region is located */ | |
836 | end_buffer = start_buffer + buffer_size - end->nb_bytes; | |
837 | ||
ee22a9d8 | 838 | qemu_iovec_init(&qiov, 2 + (m->data_qiov ? m->data_qiov->niov : 0)); |
86b862c4 | 839 | |
593fb83c | 840 | qemu_co_mutex_unlock(&s->lock); |
b3cf1c7c AG |
841 | /* First we read the existing data from both COW regions. We |
842 | * either read the whole region in one go, or the start and end | |
843 | * regions separately. */ | |
844 | if (merge_reads) { | |
86b862c4 AG |
845 | qemu_iovec_add(&qiov, start_buffer, buffer_size); |
846 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c | 847 | } else { |
86b862c4 AG |
848 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); |
849 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c AG |
850 | if (ret < 0) { |
851 | goto fail; | |
852 | } | |
672f0f2c | 853 | |
86b862c4 AG |
854 | qemu_iovec_reset(&qiov); |
855 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
856 | ret = do_perform_cow_read(bs, m->offset, end->offset, &qiov); | |
b3cf1c7c | 857 | } |
593fb83c | 858 | if (ret < 0) { |
99450c6f | 859 | goto fail; |
593fb83c KW |
860 | } |
861 | ||
672f0f2c AG |
862 | /* Encrypt the data if necessary before writing it */ |
863 | if (bs->encrypted) { | |
4652b8f3 DB |
864 | if (!do_perform_cow_encrypt(bs, m->offset, m->alloc_offset, |
865 | start->offset, start_buffer, | |
866 | start->nb_bytes) || | |
867 | !do_perform_cow_encrypt(bs, m->offset, m->alloc_offset, | |
868 | end->offset, end_buffer, end->nb_bytes)) { | |
672f0f2c AG |
869 | ret = -EIO; |
870 | goto fail; | |
871 | } | |
872 | } | |
873 | ||
ee22a9d8 AG |
874 | /* And now we can write everything. If we have the guest data we |
875 | * can write everything in one single operation */ | |
876 | if (m->data_qiov) { | |
877 | qemu_iovec_reset(&qiov); | |
878 | if (start->nb_bytes) { | |
879 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
880 | } | |
881 | qemu_iovec_concat(&qiov, m->data_qiov, 0, data_bytes); | |
882 | if (end->nb_bytes) { | |
883 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
884 | } | |
885 | /* NOTE: we have a write_aio blkdebug event here followed by | |
886 | * a cow_write one in do_perform_cow_write(), but there's only | |
887 | * one single I/O operation */ | |
888 | BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); | |
889 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); | |
890 | } else { | |
891 | /* If there's no guest data then write both COW regions separately */ | |
892 | qemu_iovec_reset(&qiov); | |
893 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
894 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); | |
895 | if (ret < 0) { | |
896 | goto fail; | |
897 | } | |
898 | ||
899 | qemu_iovec_reset(&qiov); | |
900 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
901 | ret = do_perform_cow_write(bs, m->alloc_offset, end->offset, &qiov); | |
672f0f2c | 902 | } |
99450c6f AG |
903 | |
904 | fail: | |
905 | qemu_co_mutex_lock(&s->lock); | |
906 | ||
593fb83c KW |
907 | /* |
908 | * Before we update the L2 table to actually point to the new cluster, we | |
909 | * need to be sure that the refcounts have been increased and COW was | |
910 | * handled. | |
911 | */ | |
99450c6f AG |
912 | if (ret == 0) { |
913 | qcow2_cache_depends_on_flush(s->l2_table_cache); | |
914 | } | |
593fb83c | 915 | |
672f0f2c | 916 | qemu_vfree(start_buffer); |
86b862c4 | 917 | qemu_iovec_destroy(&qiov); |
99450c6f | 918 | return ret; |
593fb83c KW |
919 | } |
920 | ||
148da7ea | 921 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) |
45aba42f | 922 | { |
ff99129a | 923 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 924 | int i, j = 0, l2_index, ret; |
a002c0b0 | 925 | uint64_t *old_cluster, *l2_slice; |
250196f1 | 926 | uint64_t cluster_offset = m->alloc_offset; |
45aba42f | 927 | |
3cce16f4 | 928 | trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
f50f88b9 | 929 | assert(m->nb_clusters > 0); |
45aba42f | 930 | |
5839e53b | 931 | old_cluster = g_try_new(uint64_t, m->nb_clusters); |
de82815d KW |
932 | if (old_cluster == NULL) { |
933 | ret = -ENOMEM; | |
934 | goto err; | |
935 | } | |
45aba42f KW |
936 | |
937 | /* copy content of unmodified sectors */ | |
99450c6f | 938 | ret = perform_cow(bs, m); |
593fb83c KW |
939 | if (ret < 0) { |
940 | goto err; | |
29c1a730 KW |
941 | } |
942 | ||
593fb83c | 943 | /* Update L2 table. */ |
74c4510a | 944 | if (s->use_lazy_refcounts) { |
280d3735 KW |
945 | qcow2_mark_dirty(bs); |
946 | } | |
bfe8043e SH |
947 | if (qcow2_need_accurate_refcounts(s)) { |
948 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
949 | s->refcount_block_cache); | |
950 | } | |
280d3735 | 951 | |
a002c0b0 | 952 | ret = get_cluster_table(bs, m->offset, &l2_slice, &l2_index); |
1e3e8f1a | 953 | if (ret < 0) { |
45aba42f | 954 | goto err; |
1e3e8f1a | 955 | } |
a002c0b0 | 956 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
45aba42f | 957 | |
a002c0b0 | 958 | assert(l2_index + m->nb_clusters <= s->l2_slice_size); |
45aba42f KW |
959 | for (i = 0; i < m->nb_clusters; i++) { |
960 | /* if two concurrent writes happen to the same unallocated cluster | |
aaa4d20b KW |
961 | * each write allocates separate cluster and writes data concurrently. |
962 | * The first one to complete updates l2 table with pointer to its | |
963 | * cluster the second one has to do RMW (which is done above by | |
964 | * perform_cow()), update l2 table with its cluster pointer and free | |
965 | * old cluster. This is what this loop does */ | |
a002c0b0 AG |
966 | if (l2_slice[l2_index + i] != 0) { |
967 | old_cluster[j++] = l2_slice[l2_index + i]; | |
aaa4d20b | 968 | } |
45aba42f | 969 | |
a002c0b0 | 970 | l2_slice[l2_index + i] = cpu_to_be64((cluster_offset + |
45aba42f KW |
971 | (i << s->cluster_bits)) | QCOW_OFLAG_COPIED); |
972 | } | |
973 | ||
9f8e668e | 974 | |
a002c0b0 | 975 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
45aba42f | 976 | |
7ec5e6a4 KW |
977 | /* |
978 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
6cfcb9b8 KW |
979 | * |
980 | * Don't discard clusters that reach a refcount of 0 (e.g. compressed | |
981 | * clusters), the next write will reuse them anyway. | |
7ec5e6a4 | 982 | */ |
564a6b69 | 983 | if (!m->keep_old_clusters && j != 0) { |
7ec5e6a4 | 984 | for (i = 0; i < j; i++) { |
6cfcb9b8 KW |
985 | qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1, |
986 | QCOW2_DISCARD_NEVER); | |
7ec5e6a4 KW |
987 | } |
988 | } | |
45aba42f KW |
989 | |
990 | ret = 0; | |
991 | err: | |
7267c094 | 992 | g_free(old_cluster); |
45aba42f KW |
993 | return ret; |
994 | } | |
995 | ||
bf319ece KW |
996 | /* |
997 | * Returns the number of contiguous clusters that can be used for an allocating | |
998 | * write, but require COW to be performed (this includes yet unallocated space, | |
999 | * which must copy from the backing file) | |
1000 | */ | |
ff99129a | 1001 | static int count_cow_clusters(BDRVQcow2State *s, int nb_clusters, |
dd32c881 | 1002 | uint64_t *l2_slice, int l2_index) |
bf319ece | 1003 | { |
143550a8 | 1004 | int i; |
bf319ece | 1005 | |
143550a8 | 1006 | for (i = 0; i < nb_clusters; i++) { |
dd32c881 | 1007 | uint64_t l2_entry = be64_to_cpu(l2_slice[l2_index + i]); |
3ef95218 | 1008 | QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry); |
143550a8 KW |
1009 | |
1010 | switch(cluster_type) { | |
1011 | case QCOW2_CLUSTER_NORMAL: | |
1012 | if (l2_entry & QCOW_OFLAG_COPIED) { | |
1013 | goto out; | |
1014 | } | |
bf319ece | 1015 | break; |
143550a8 KW |
1016 | case QCOW2_CLUSTER_UNALLOCATED: |
1017 | case QCOW2_CLUSTER_COMPRESSED: | |
fdfab37d EB |
1018 | case QCOW2_CLUSTER_ZERO_PLAIN: |
1019 | case QCOW2_CLUSTER_ZERO_ALLOC: | |
bf319ece | 1020 | break; |
143550a8 KW |
1021 | default: |
1022 | abort(); | |
1023 | } | |
bf319ece KW |
1024 | } |
1025 | ||
143550a8 | 1026 | out: |
bf319ece KW |
1027 | assert(i <= nb_clusters); |
1028 | return i; | |
1029 | } | |
1030 | ||
250196f1 | 1031 | /* |
226c3c26 KW |
1032 | * Check if there already is an AIO write request in flight which allocates |
1033 | * the same cluster. In this case we need to wait until the previous | |
1034 | * request has completed and updated the L2 table accordingly. | |
65eb2e35 KW |
1035 | * |
1036 | * Returns: | |
1037 | * 0 if there was no dependency. *cur_bytes indicates the number of | |
1038 | * bytes from guest_offset that can be read before the next | |
1039 | * dependency must be processed (or the request is complete) | |
1040 | * | |
1041 | * -EAGAIN if we had to wait for another request, previously gathered | |
1042 | * information on cluster allocation may be invalid now. The caller | |
1043 | * must start over anyway, so consider *cur_bytes undefined. | |
250196f1 | 1044 | */ |
226c3c26 | 1045 | static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, |
ecdd5333 | 1046 | uint64_t *cur_bytes, QCowL2Meta **m) |
250196f1 | 1047 | { |
ff99129a | 1048 | BDRVQcow2State *s = bs->opaque; |
250196f1 | 1049 | QCowL2Meta *old_alloc; |
65eb2e35 | 1050 | uint64_t bytes = *cur_bytes; |
250196f1 | 1051 | |
250196f1 KW |
1052 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
1053 | ||
65eb2e35 KW |
1054 | uint64_t start = guest_offset; |
1055 | uint64_t end = start + bytes; | |
1056 | uint64_t old_start = l2meta_cow_start(old_alloc); | |
1057 | uint64_t old_end = l2meta_cow_end(old_alloc); | |
250196f1 | 1058 | |
d9d74f41 | 1059 | if (end <= old_start || start >= old_end) { |
250196f1 KW |
1060 | /* No intersection */ |
1061 | } else { | |
1062 | if (start < old_start) { | |
1063 | /* Stop at the start of a running allocation */ | |
65eb2e35 | 1064 | bytes = old_start - start; |
250196f1 | 1065 | } else { |
65eb2e35 | 1066 | bytes = 0; |
250196f1 KW |
1067 | } |
1068 | ||
ecdd5333 KW |
1069 | /* Stop if already an l2meta exists. After yielding, it wouldn't |
1070 | * be valid any more, so we'd have to clean up the old L2Metas | |
1071 | * and deal with requests depending on them before starting to | |
1072 | * gather new ones. Not worth the trouble. */ | |
1073 | if (bytes == 0 && *m) { | |
1074 | *cur_bytes = 0; | |
1075 | return 0; | |
1076 | } | |
1077 | ||
65eb2e35 | 1078 | if (bytes == 0) { |
250196f1 KW |
1079 | /* Wait for the dependency to complete. We need to recheck |
1080 | * the free/allocated clusters when we continue. */ | |
1ace7cea | 1081 | qemu_co_queue_wait(&old_alloc->dependent_requests, &s->lock); |
250196f1 KW |
1082 | return -EAGAIN; |
1083 | } | |
1084 | } | |
1085 | } | |
1086 | ||
65eb2e35 KW |
1087 | /* Make sure that existing clusters and new allocations are only used up to |
1088 | * the next dependency if we shortened the request above */ | |
1089 | *cur_bytes = bytes; | |
250196f1 | 1090 | |
226c3c26 KW |
1091 | return 0; |
1092 | } | |
1093 | ||
0af729ec KW |
1094 | /* |
1095 | * Checks how many already allocated clusters that don't require a copy on | |
1096 | * write there are at the given guest_offset (up to *bytes). If | |
1097 | * *host_offset is not zero, only physically contiguous clusters beginning at | |
1098 | * this host offset are counted. | |
1099 | * | |
411d62b0 KW |
1100 | * Note that guest_offset may not be cluster aligned. In this case, the |
1101 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1102 | * therefore isn't cluster aligned as well. | |
0af729ec KW |
1103 | * |
1104 | * Returns: | |
1105 | * 0: if no allocated clusters are available at the given offset. | |
1106 | * *bytes is normally unchanged. It is set to 0 if the cluster | |
1107 | * is allocated and doesn't need COW, but doesn't have the right | |
1108 | * physical offset. | |
1109 | * | |
1110 | * 1: if allocated clusters that don't require a COW are available at | |
1111 | * the requested offset. *bytes may have decreased and describes | |
1112 | * the length of the area that can be written to. | |
1113 | * | |
1114 | * -errno: in error cases | |
0af729ec KW |
1115 | */ |
1116 | static int handle_copied(BlockDriverState *bs, uint64_t guest_offset, | |
c53ede9f | 1117 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
0af729ec | 1118 | { |
ff99129a | 1119 | BDRVQcow2State *s = bs->opaque; |
0af729ec KW |
1120 | int l2_index; |
1121 | uint64_t cluster_offset; | |
cde91766 | 1122 | uint64_t *l2_slice; |
b6d36def | 1123 | uint64_t nb_clusters; |
c53ede9f | 1124 | unsigned int keep_clusters; |
a3f1afb4 | 1125 | int ret; |
0af729ec KW |
1126 | |
1127 | trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset, | |
1128 | *bytes); | |
0af729ec | 1129 | |
411d62b0 KW |
1130 | assert(*host_offset == 0 || offset_into_cluster(s, guest_offset) |
1131 | == offset_into_cluster(s, *host_offset)); | |
1132 | ||
acb0467f | 1133 | /* |
cde91766 | 1134 | * Calculate the number of clusters to look for. We stop at L2 slice |
acb0467f KW |
1135 | * boundaries to keep things simple. |
1136 | */ | |
1137 | nb_clusters = | |
1138 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1139 | ||
cde91766 AG |
1140 | l2_index = offset_to_l2_slice_index(s, guest_offset); |
1141 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
b6d36def | 1142 | assert(nb_clusters <= INT_MAX); |
acb0467f | 1143 | |
0af729ec | 1144 | /* Find L2 entry for the first involved cluster */ |
cde91766 | 1145 | ret = get_cluster_table(bs, guest_offset, &l2_slice, &l2_index); |
0af729ec KW |
1146 | if (ret < 0) { |
1147 | return ret; | |
1148 | } | |
1149 | ||
cde91766 | 1150 | cluster_offset = be64_to_cpu(l2_slice[l2_index]); |
0af729ec KW |
1151 | |
1152 | /* Check how many clusters are already allocated and don't need COW */ | |
1153 | if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL | |
1154 | && (cluster_offset & QCOW_OFLAG_COPIED)) | |
1155 | { | |
e62daaf6 KW |
1156 | /* If a specific host_offset is required, check it */ |
1157 | bool offset_matches = | |
1158 | (cluster_offset & L2E_OFFSET_MASK) == *host_offset; | |
1159 | ||
a97c67ee HR |
1160 | if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) { |
1161 | qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset " | |
1162 | "%#llx unaligned (guest offset: %#" PRIx64 | |
1163 | ")", cluster_offset & L2E_OFFSET_MASK, | |
1164 | guest_offset); | |
1165 | ret = -EIO; | |
1166 | goto out; | |
1167 | } | |
1168 | ||
e62daaf6 KW |
1169 | if (*host_offset != 0 && !offset_matches) { |
1170 | *bytes = 0; | |
1171 | ret = 0; | |
1172 | goto out; | |
1173 | } | |
1174 | ||
0af729ec | 1175 | /* We keep all QCOW_OFLAG_COPIED clusters */ |
c53ede9f | 1176 | keep_clusters = |
acb0467f | 1177 | count_contiguous_clusters(nb_clusters, s->cluster_size, |
cde91766 | 1178 | &l2_slice[l2_index], |
0af729ec | 1179 | QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); |
c53ede9f KW |
1180 | assert(keep_clusters <= nb_clusters); |
1181 | ||
1182 | *bytes = MIN(*bytes, | |
1183 | keep_clusters * s->cluster_size | |
1184 | - offset_into_cluster(s, guest_offset)); | |
0af729ec KW |
1185 | |
1186 | ret = 1; | |
1187 | } else { | |
0af729ec KW |
1188 | ret = 0; |
1189 | } | |
1190 | ||
0af729ec | 1191 | /* Cleanup */ |
e62daaf6 | 1192 | out: |
cde91766 | 1193 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
0af729ec | 1194 | |
e62daaf6 KW |
1195 | /* Only return a host offset if we actually made progress. Otherwise we |
1196 | * would make requirements for handle_alloc() that it can't fulfill */ | |
a97c67ee | 1197 | if (ret > 0) { |
411d62b0 KW |
1198 | *host_offset = (cluster_offset & L2E_OFFSET_MASK) |
1199 | + offset_into_cluster(s, guest_offset); | |
e62daaf6 KW |
1200 | } |
1201 | ||
0af729ec KW |
1202 | return ret; |
1203 | } | |
1204 | ||
226c3c26 KW |
1205 | /* |
1206 | * Allocates new clusters for the given guest_offset. | |
1207 | * | |
1208 | * At most *nb_clusters are allocated, and on return *nb_clusters is updated to | |
1209 | * contain the number of clusters that have been allocated and are contiguous | |
1210 | * in the image file. | |
1211 | * | |
1212 | * If *host_offset is non-zero, it specifies the offset in the image file at | |
1213 | * which the new clusters must start. *nb_clusters can be 0 on return in this | |
1214 | * case if the cluster at host_offset is already in use. If *host_offset is | |
1215 | * zero, the clusters can be allocated anywhere in the image file. | |
1216 | * | |
1217 | * *host_offset is updated to contain the offset into the image file at which | |
1218 | * the first allocated cluster starts. | |
1219 | * | |
1220 | * Return 0 on success and -errno in error cases. -EAGAIN means that the | |
1221 | * function has been waiting for another request and the allocation must be | |
1222 | * restarted, but the whole request should not be failed. | |
1223 | */ | |
1224 | static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, | |
b6d36def | 1225 | uint64_t *host_offset, uint64_t *nb_clusters) |
226c3c26 | 1226 | { |
ff99129a | 1227 | BDRVQcow2State *s = bs->opaque; |
226c3c26 KW |
1228 | |
1229 | trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, | |
1230 | *host_offset, *nb_clusters); | |
1231 | ||
250196f1 KW |
1232 | /* Allocate new clusters */ |
1233 | trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); | |
1234 | if (*host_offset == 0) { | |
df021791 KW |
1235 | int64_t cluster_offset = |
1236 | qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); | |
1237 | if (cluster_offset < 0) { | |
1238 | return cluster_offset; | |
1239 | } | |
1240 | *host_offset = cluster_offset; | |
1241 | return 0; | |
250196f1 | 1242 | } else { |
b6d36def | 1243 | int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); |
df021791 KW |
1244 | if (ret < 0) { |
1245 | return ret; | |
1246 | } | |
1247 | *nb_clusters = ret; | |
1248 | return 0; | |
250196f1 | 1249 | } |
250196f1 KW |
1250 | } |
1251 | ||
10f0ed8b KW |
1252 | /* |
1253 | * Allocates new clusters for an area that either is yet unallocated or needs a | |
1254 | * copy on write. If *host_offset is non-zero, clusters are only allocated if | |
1255 | * the new allocation can match the specified host offset. | |
1256 | * | |
411d62b0 KW |
1257 | * Note that guest_offset may not be cluster aligned. In this case, the |
1258 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1259 | * therefore isn't cluster aligned as well. | |
10f0ed8b KW |
1260 | * |
1261 | * Returns: | |
1262 | * 0: if no clusters could be allocated. *bytes is set to 0, | |
1263 | * *host_offset is left unchanged. | |
1264 | * | |
1265 | * 1: if new clusters were allocated. *bytes may be decreased if the | |
1266 | * new allocation doesn't cover all of the requested area. | |
1267 | * *host_offset is updated to contain the host offset of the first | |
1268 | * newly allocated cluster. | |
1269 | * | |
1270 | * -errno: in error cases | |
10f0ed8b KW |
1271 | */ |
1272 | static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, | |
c37f4cd7 | 1273 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
10f0ed8b | 1274 | { |
ff99129a | 1275 | BDRVQcow2State *s = bs->opaque; |
10f0ed8b | 1276 | int l2_index; |
6d99a344 | 1277 | uint64_t *l2_slice; |
10f0ed8b | 1278 | uint64_t entry; |
b6d36def | 1279 | uint64_t nb_clusters; |
10f0ed8b | 1280 | int ret; |
564a6b69 | 1281 | bool keep_old_clusters = false; |
10f0ed8b | 1282 | |
564a6b69 | 1283 | uint64_t alloc_cluster_offset = 0; |
10f0ed8b KW |
1284 | |
1285 | trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset, | |
1286 | *bytes); | |
1287 | assert(*bytes > 0); | |
1288 | ||
f5bc6350 | 1289 | /* |
6d99a344 | 1290 | * Calculate the number of clusters to look for. We stop at L2 slice |
f5bc6350 KW |
1291 | * boundaries to keep things simple. |
1292 | */ | |
c37f4cd7 KW |
1293 | nb_clusters = |
1294 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1295 | ||
6d99a344 AG |
1296 | l2_index = offset_to_l2_slice_index(s, guest_offset); |
1297 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
b6d36def | 1298 | assert(nb_clusters <= INT_MAX); |
f5bc6350 | 1299 | |
10f0ed8b | 1300 | /* Find L2 entry for the first involved cluster */ |
6d99a344 | 1301 | ret = get_cluster_table(bs, guest_offset, &l2_slice, &l2_index); |
10f0ed8b KW |
1302 | if (ret < 0) { |
1303 | return ret; | |
1304 | } | |
1305 | ||
6d99a344 | 1306 | entry = be64_to_cpu(l2_slice[l2_index]); |
10f0ed8b KW |
1307 | |
1308 | /* For the moment, overwrite compressed clusters one by one */ | |
1309 | if (entry & QCOW_OFLAG_COMPRESSED) { | |
1310 | nb_clusters = 1; | |
1311 | } else { | |
6d99a344 | 1312 | nb_clusters = count_cow_clusters(s, nb_clusters, l2_slice, l2_index); |
10f0ed8b KW |
1313 | } |
1314 | ||
ecdd5333 KW |
1315 | /* This function is only called when there were no non-COW clusters, so if |
1316 | * we can't find any unallocated or COW clusters either, something is | |
1317 | * wrong with our code. */ | |
1318 | assert(nb_clusters > 0); | |
1319 | ||
fdfab37d EB |
1320 | if (qcow2_get_cluster_type(entry) == QCOW2_CLUSTER_ZERO_ALLOC && |
1321 | (entry & QCOW_OFLAG_COPIED) && | |
564a6b69 HR |
1322 | (!*host_offset || |
1323 | start_of_cluster(s, *host_offset) == (entry & L2E_OFFSET_MASK))) | |
1324 | { | |
93bbaf03 HR |
1325 | int preallocated_nb_clusters; |
1326 | ||
1327 | if (offset_into_cluster(s, entry & L2E_OFFSET_MASK)) { | |
1328 | qcow2_signal_corruption(bs, true, -1, -1, "Preallocated zero " | |
1329 | "cluster offset %#llx unaligned (guest " | |
1330 | "offset: %#" PRIx64 ")", | |
1331 | entry & L2E_OFFSET_MASK, guest_offset); | |
1332 | ret = -EIO; | |
1333 | goto fail; | |
1334 | } | |
1335 | ||
564a6b69 HR |
1336 | /* Try to reuse preallocated zero clusters; contiguous normal clusters |
1337 | * would be fine, too, but count_cow_clusters() above has limited | |
1338 | * nb_clusters already to a range of COW clusters */ | |
93bbaf03 | 1339 | preallocated_nb_clusters = |
564a6b69 | 1340 | count_contiguous_clusters(nb_clusters, s->cluster_size, |
6d99a344 | 1341 | &l2_slice[l2_index], QCOW_OFLAG_COPIED); |
564a6b69 | 1342 | assert(preallocated_nb_clusters > 0); |
10f0ed8b | 1343 | |
564a6b69 HR |
1344 | nb_clusters = preallocated_nb_clusters; |
1345 | alloc_cluster_offset = entry & L2E_OFFSET_MASK; | |
10f0ed8b | 1346 | |
564a6b69 HR |
1347 | /* We want to reuse these clusters, so qcow2_alloc_cluster_link_l2() |
1348 | * should not free them. */ | |
1349 | keep_old_clusters = true; | |
10f0ed8b KW |
1350 | } |
1351 | ||
6d99a344 | 1352 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
564a6b69 | 1353 | |
ff52aab2 | 1354 | if (!alloc_cluster_offset) { |
564a6b69 HR |
1355 | /* Allocate, if necessary at a given offset in the image file */ |
1356 | alloc_cluster_offset = start_of_cluster(s, *host_offset); | |
1357 | ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset, | |
1358 | &nb_clusters); | |
1359 | if (ret < 0) { | |
1360 | goto fail; | |
1361 | } | |
1362 | ||
1363 | /* Can't extend contiguous allocation */ | |
1364 | if (nb_clusters == 0) { | |
1365 | *bytes = 0; | |
1366 | return 0; | |
1367 | } | |
1368 | ||
1369 | /* !*host_offset would overwrite the image header and is reserved for | |
1370 | * "no host offset preferred". If 0 was a valid host offset, it'd | |
1371 | * trigger the following overlap check; do that now to avoid having an | |
1372 | * invalid value in *host_offset. */ | |
1373 | if (!alloc_cluster_offset) { | |
1374 | ret = qcow2_pre_write_overlap_check(bs, 0, alloc_cluster_offset, | |
1375 | nb_clusters * s->cluster_size); | |
1376 | assert(ret < 0); | |
1377 | goto fail; | |
1378 | } | |
ff52aab2 HR |
1379 | } |
1380 | ||
83baa9a4 KW |
1381 | /* |
1382 | * Save info needed for meta data update. | |
1383 | * | |
85567393 | 1384 | * requested_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1385 | * newly allocated cluster to the end of the (possibly shortened |
1386 | * before) write request. | |
1387 | * | |
85567393 | 1388 | * avail_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1389 | * newly allocated to the end of the last newly allocated cluster. |
1390 | * | |
85567393 | 1391 | * nb_bytes: The number of bytes from the start of the first |
83baa9a4 KW |
1392 | * newly allocated cluster to the end of the area that the write |
1393 | * request actually writes to (excluding COW at the end) | |
1394 | */ | |
85567393 KW |
1395 | uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset); |
1396 | int avail_bytes = MIN(INT_MAX, nb_clusters << s->cluster_bits); | |
1397 | int nb_bytes = MIN(requested_bytes, avail_bytes); | |
88c6588c | 1398 | QCowL2Meta *old_m = *m; |
83baa9a4 | 1399 | |
83baa9a4 KW |
1400 | *m = g_malloc0(sizeof(**m)); |
1401 | ||
1402 | **m = (QCowL2Meta) { | |
88c6588c KW |
1403 | .next = old_m, |
1404 | ||
411d62b0 | 1405 | .alloc_offset = alloc_cluster_offset, |
83baa9a4 KW |
1406 | .offset = start_of_cluster(s, guest_offset), |
1407 | .nb_clusters = nb_clusters, | |
83baa9a4 | 1408 | |
564a6b69 HR |
1409 | .keep_old_clusters = keep_old_clusters, |
1410 | ||
83baa9a4 KW |
1411 | .cow_start = { |
1412 | .offset = 0, | |
85567393 | 1413 | .nb_bytes = offset_into_cluster(s, guest_offset), |
83baa9a4 KW |
1414 | }, |
1415 | .cow_end = { | |
85567393 KW |
1416 | .offset = nb_bytes, |
1417 | .nb_bytes = avail_bytes - nb_bytes, | |
83baa9a4 KW |
1418 | }, |
1419 | }; | |
1420 | qemu_co_queue_init(&(*m)->dependent_requests); | |
1421 | QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); | |
1422 | ||
411d62b0 | 1423 | *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset); |
85567393 | 1424 | *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset)); |
83baa9a4 KW |
1425 | assert(*bytes != 0); |
1426 | ||
10f0ed8b KW |
1427 | return 1; |
1428 | ||
1429 | fail: | |
1430 | if (*m && (*m)->nb_clusters > 0) { | |
1431 | QLIST_REMOVE(*m, next_in_flight); | |
1432 | } | |
1433 | return ret; | |
1434 | } | |
1435 | ||
45aba42f KW |
1436 | /* |
1437 | * alloc_cluster_offset | |
1438 | * | |
250196f1 KW |
1439 | * For a given offset on the virtual disk, find the cluster offset in qcow2 |
1440 | * file. If the offset is not found, allocate a new cluster. | |
45aba42f | 1441 | * |
250196f1 | 1442 | * If the cluster was already allocated, m->nb_clusters is set to 0 and |
a7912369 | 1443 | * other fields in m are meaningless. |
148da7ea KW |
1444 | * |
1445 | * If the cluster is newly allocated, m->nb_clusters is set to the number of | |
68d100e9 KW |
1446 | * contiguous clusters that have been allocated. In this case, the other |
1447 | * fields of m are valid and contain information about the first allocated | |
1448 | * cluster. | |
45aba42f | 1449 | * |
68d100e9 KW |
1450 | * If the request conflicts with another write request in flight, the coroutine |
1451 | * is queued and will be reentered when the dependency has completed. | |
148da7ea KW |
1452 | * |
1453 | * Return 0 on success and -errno in error cases | |
45aba42f | 1454 | */ |
f4f0d391 | 1455 | int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, |
d46a0bb2 KW |
1456 | unsigned int *bytes, uint64_t *host_offset, |
1457 | QCowL2Meta **m) | |
45aba42f | 1458 | { |
ff99129a | 1459 | BDRVQcow2State *s = bs->opaque; |
710c2496 | 1460 | uint64_t start, remaining; |
250196f1 | 1461 | uint64_t cluster_offset; |
65eb2e35 | 1462 | uint64_t cur_bytes; |
710c2496 | 1463 | int ret; |
45aba42f | 1464 | |
d46a0bb2 | 1465 | trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *bytes); |
710c2496 | 1466 | |
72424114 | 1467 | again: |
16f0587e | 1468 | start = offset; |
d46a0bb2 | 1469 | remaining = *bytes; |
0af729ec KW |
1470 | cluster_offset = 0; |
1471 | *host_offset = 0; | |
ecdd5333 KW |
1472 | cur_bytes = 0; |
1473 | *m = NULL; | |
0af729ec | 1474 | |
2c3b32d2 | 1475 | while (true) { |
ecdd5333 KW |
1476 | |
1477 | if (!*host_offset) { | |
1478 | *host_offset = start_of_cluster(s, cluster_offset); | |
1479 | } | |
1480 | ||
1481 | assert(remaining >= cur_bytes); | |
1482 | ||
1483 | start += cur_bytes; | |
1484 | remaining -= cur_bytes; | |
1485 | cluster_offset += cur_bytes; | |
1486 | ||
1487 | if (remaining == 0) { | |
1488 | break; | |
1489 | } | |
1490 | ||
1491 | cur_bytes = remaining; | |
1492 | ||
2c3b32d2 KW |
1493 | /* |
1494 | * Now start gathering as many contiguous clusters as possible: | |
1495 | * | |
1496 | * 1. Check for overlaps with in-flight allocations | |
1497 | * | |
1498 | * a) Overlap not in the first cluster -> shorten this request and | |
1499 | * let the caller handle the rest in its next loop iteration. | |
1500 | * | |
1501 | * b) Real overlaps of two requests. Yield and restart the search | |
1502 | * for contiguous clusters (the situation could have changed | |
1503 | * while we were sleeping) | |
1504 | * | |
1505 | * c) TODO: Request starts in the same cluster as the in-flight | |
1506 | * allocation ends. Shorten the COW of the in-fight allocation, | |
1507 | * set cluster_offset to write to the same cluster and set up | |
1508 | * the right synchronisation between the in-flight request and | |
1509 | * the new one. | |
1510 | */ | |
ecdd5333 | 1511 | ret = handle_dependencies(bs, start, &cur_bytes, m); |
2c3b32d2 | 1512 | if (ret == -EAGAIN) { |
ecdd5333 KW |
1513 | /* Currently handle_dependencies() doesn't yield if we already had |
1514 | * an allocation. If it did, we would have to clean up the L2Meta | |
1515 | * structs before starting over. */ | |
1516 | assert(*m == NULL); | |
2c3b32d2 KW |
1517 | goto again; |
1518 | } else if (ret < 0) { | |
1519 | return ret; | |
ecdd5333 KW |
1520 | } else if (cur_bytes == 0) { |
1521 | break; | |
2c3b32d2 KW |
1522 | } else { |
1523 | /* handle_dependencies() may have decreased cur_bytes (shortened | |
1524 | * the allocations below) so that the next dependency is processed | |
1525 | * correctly during the next loop iteration. */ | |
0af729ec | 1526 | } |
710c2496 | 1527 | |
2c3b32d2 KW |
1528 | /* |
1529 | * 2. Count contiguous COPIED clusters. | |
1530 | */ | |
1531 | ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m); | |
1532 | if (ret < 0) { | |
1533 | return ret; | |
1534 | } else if (ret) { | |
ecdd5333 | 1535 | continue; |
2c3b32d2 KW |
1536 | } else if (cur_bytes == 0) { |
1537 | break; | |
1538 | } | |
060bee89 | 1539 | |
2c3b32d2 KW |
1540 | /* |
1541 | * 3. If the request still hasn't completed, allocate new clusters, | |
1542 | * considering any cluster_offset of steps 1c or 2. | |
1543 | */ | |
1544 | ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m); | |
1545 | if (ret < 0) { | |
1546 | return ret; | |
1547 | } else if (ret) { | |
ecdd5333 | 1548 | continue; |
2c3b32d2 KW |
1549 | } else { |
1550 | assert(cur_bytes == 0); | |
1551 | break; | |
1552 | } | |
f5bc6350 | 1553 | } |
10f0ed8b | 1554 | |
d46a0bb2 KW |
1555 | *bytes -= remaining; |
1556 | assert(*bytes > 0); | |
710c2496 | 1557 | assert(*host_offset != 0); |
45aba42f | 1558 | |
148da7ea | 1559 | return 0; |
45aba42f KW |
1560 | } |
1561 | ||
1562 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
1563 | const uint8_t *buf, int buf_size) | |
1564 | { | |
1565 | z_stream strm1, *strm = &strm1; | |
1566 | int ret, out_len; | |
1567 | ||
1568 | memset(strm, 0, sizeof(*strm)); | |
1569 | ||
1570 | strm->next_in = (uint8_t *)buf; | |
1571 | strm->avail_in = buf_size; | |
1572 | strm->next_out = out_buf; | |
1573 | strm->avail_out = out_buf_size; | |
1574 | ||
1575 | ret = inflateInit2(strm, -12); | |
1576 | if (ret != Z_OK) | |
1577 | return -1; | |
1578 | ret = inflate(strm, Z_FINISH); | |
1579 | out_len = strm->next_out - out_buf; | |
1580 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
1581 | out_len != out_buf_size) { | |
1582 | inflateEnd(strm); | |
1583 | return -1; | |
1584 | } | |
1585 | inflateEnd(strm); | |
1586 | return 0; | |
1587 | } | |
1588 | ||
66f82cee | 1589 | int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) |
45aba42f | 1590 | { |
ff99129a | 1591 | BDRVQcow2State *s = bs->opaque; |
45aba42f KW |
1592 | int ret, csize, nb_csectors, sector_offset; |
1593 | uint64_t coffset; | |
1594 | ||
1595 | coffset = cluster_offset & s->cluster_offset_mask; | |
1596 | if (s->cluster_cache_offset != coffset) { | |
1597 | nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1; | |
1598 | sector_offset = coffset & 511; | |
1599 | csize = nb_csectors * 512 - sector_offset; | |
3e4c7052 SH |
1600 | |
1601 | /* Allocate buffers on first decompress operation, most images are | |
1602 | * uncompressed and the memory overhead can be avoided. The buffers | |
1603 | * are freed in .bdrv_close(). | |
1604 | */ | |
1605 | if (!s->cluster_data) { | |
1606 | /* one more sector for decompressed data alignment */ | |
1607 | s->cluster_data = qemu_try_blockalign(bs->file->bs, | |
1608 | QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size + 512); | |
1609 | if (!s->cluster_data) { | |
1610 | return -ENOMEM; | |
1611 | } | |
1612 | } | |
1613 | if (!s->cluster_cache) { | |
1614 | s->cluster_cache = g_malloc(s->cluster_size); | |
1615 | } | |
1616 | ||
66f82cee | 1617 | BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
fbcbbf4e | 1618 | ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, |
9a4f4c31 | 1619 | nb_csectors); |
45aba42f | 1620 | if (ret < 0) { |
8af36488 | 1621 | return ret; |
45aba42f KW |
1622 | } |
1623 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
1624 | s->cluster_data + sector_offset, csize) < 0) { | |
8af36488 | 1625 | return -EIO; |
45aba42f KW |
1626 | } |
1627 | s->cluster_cache_offset = coffset; | |
1628 | } | |
1629 | return 0; | |
1630 | } | |
5ea929e3 KW |
1631 | |
1632 | /* | |
1633 | * This discards as many clusters of nb_clusters as possible at once (i.e. | |
21ab3add | 1634 | * all clusters in the same L2 slice) and returns the number of discarded |
5ea929e3 KW |
1635 | * clusters. |
1636 | */ | |
21ab3add AG |
1637 | static int discard_in_l2_slice(BlockDriverState *bs, uint64_t offset, |
1638 | uint64_t nb_clusters, | |
1639 | enum qcow2_discard_type type, bool full_discard) | |
5ea929e3 | 1640 | { |
ff99129a | 1641 | BDRVQcow2State *s = bs->opaque; |
21ab3add | 1642 | uint64_t *l2_slice; |
5ea929e3 KW |
1643 | int l2_index; |
1644 | int ret; | |
1645 | int i; | |
1646 | ||
21ab3add | 1647 | ret = get_cluster_table(bs, offset, &l2_slice, &l2_index); |
5ea929e3 KW |
1648 | if (ret < 0) { |
1649 | return ret; | |
1650 | } | |
1651 | ||
21ab3add AG |
1652 | /* Limit nb_clusters to one L2 slice */ |
1653 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
b6d36def | 1654 | assert(nb_clusters <= INT_MAX); |
5ea929e3 KW |
1655 | |
1656 | for (i = 0; i < nb_clusters; i++) { | |
c883db0d | 1657 | uint64_t old_l2_entry; |
5ea929e3 | 1658 | |
21ab3add | 1659 | old_l2_entry = be64_to_cpu(l2_slice[l2_index + i]); |
a71835a0 KW |
1660 | |
1661 | /* | |
808c4b6f HR |
1662 | * If full_discard is false, make sure that a discarded area reads back |
1663 | * as zeroes for v3 images (we cannot do it for v2 without actually | |
1664 | * writing a zero-filled buffer). We can skip the operation if the | |
1665 | * cluster is already marked as zero, or if it's unallocated and we | |
1666 | * don't have a backing file. | |
a71835a0 | 1667 | * |
237d78f8 | 1668 | * TODO We might want to use bdrv_block_status(bs) here, but we're |
a71835a0 | 1669 | * holding s->lock, so that doesn't work today. |
808c4b6f HR |
1670 | * |
1671 | * If full_discard is true, the sector should not read back as zeroes, | |
1672 | * but rather fall through to the backing file. | |
a71835a0 | 1673 | */ |
c883db0d | 1674 | switch (qcow2_get_cluster_type(old_l2_entry)) { |
bbd995d8 EB |
1675 | case QCOW2_CLUSTER_UNALLOCATED: |
1676 | if (full_discard || !bs->backing) { | |
1677 | continue; | |
1678 | } | |
1679 | break; | |
1680 | ||
fdfab37d EB |
1681 | case QCOW2_CLUSTER_ZERO_PLAIN: |
1682 | if (!full_discard) { | |
bbd995d8 EB |
1683 | continue; |
1684 | } | |
1685 | break; | |
1686 | ||
fdfab37d | 1687 | case QCOW2_CLUSTER_ZERO_ALLOC: |
bbd995d8 EB |
1688 | case QCOW2_CLUSTER_NORMAL: |
1689 | case QCOW2_CLUSTER_COMPRESSED: | |
1690 | break; | |
1691 | ||
1692 | default: | |
1693 | abort(); | |
5ea929e3 KW |
1694 | } |
1695 | ||
1696 | /* First remove L2 entries */ | |
21ab3add | 1697 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
808c4b6f | 1698 | if (!full_discard && s->qcow_version >= 3) { |
21ab3add | 1699 | l2_slice[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
a71835a0 | 1700 | } else { |
21ab3add | 1701 | l2_slice[l2_index + i] = cpu_to_be64(0); |
a71835a0 | 1702 | } |
5ea929e3 KW |
1703 | |
1704 | /* Then decrease the refcount */ | |
c883db0d | 1705 | qcow2_free_any_clusters(bs, old_l2_entry, 1, type); |
5ea929e3 KW |
1706 | } |
1707 | ||
21ab3add | 1708 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
5ea929e3 KW |
1709 | |
1710 | return nb_clusters; | |
1711 | } | |
1712 | ||
d2cb36af EB |
1713 | int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset, |
1714 | uint64_t bytes, enum qcow2_discard_type type, | |
1715 | bool full_discard) | |
5ea929e3 | 1716 | { |
ff99129a | 1717 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1718 | uint64_t end_offset = offset + bytes; |
b6d36def | 1719 | uint64_t nb_clusters; |
d2cb36af | 1720 | int64_t cleared; |
5ea929e3 KW |
1721 | int ret; |
1722 | ||
f10ee139 | 1723 | /* Caller must pass aligned values, except at image end */ |
0c1bd469 | 1724 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); |
f10ee139 EB |
1725 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || |
1726 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
5ea929e3 | 1727 | |
d2cb36af | 1728 | nb_clusters = size_to_clusters(s, bytes); |
5ea929e3 | 1729 | |
0b919fae KW |
1730 | s->cache_discards = true; |
1731 | ||
21ab3add | 1732 | /* Each L2 slice is handled by its own loop iteration */ |
5ea929e3 | 1733 | while (nb_clusters > 0) { |
21ab3add AG |
1734 | cleared = discard_in_l2_slice(bs, offset, nb_clusters, type, |
1735 | full_discard); | |
d2cb36af EB |
1736 | if (cleared < 0) { |
1737 | ret = cleared; | |
0b919fae | 1738 | goto fail; |
5ea929e3 KW |
1739 | } |
1740 | ||
d2cb36af EB |
1741 | nb_clusters -= cleared; |
1742 | offset += (cleared * s->cluster_size); | |
5ea929e3 KW |
1743 | } |
1744 | ||
0b919fae KW |
1745 | ret = 0; |
1746 | fail: | |
1747 | s->cache_discards = false; | |
1748 | qcow2_process_discards(bs, ret); | |
1749 | ||
1750 | return ret; | |
5ea929e3 | 1751 | } |
621f0589 KW |
1752 | |
1753 | /* | |
1754 | * This zeroes as many clusters of nb_clusters as possible at once (i.e. | |
a9a9f8f0 | 1755 | * all clusters in the same L2 slice) and returns the number of zeroed |
621f0589 KW |
1756 | * clusters. |
1757 | */ | |
a9a9f8f0 AG |
1758 | static int zero_in_l2_slice(BlockDriverState *bs, uint64_t offset, |
1759 | uint64_t nb_clusters, int flags) | |
621f0589 | 1760 | { |
ff99129a | 1761 | BDRVQcow2State *s = bs->opaque; |
a9a9f8f0 | 1762 | uint64_t *l2_slice; |
621f0589 KW |
1763 | int l2_index; |
1764 | int ret; | |
1765 | int i; | |
06cc5e2b | 1766 | bool unmap = !!(flags & BDRV_REQ_MAY_UNMAP); |
621f0589 | 1767 | |
a9a9f8f0 | 1768 | ret = get_cluster_table(bs, offset, &l2_slice, &l2_index); |
621f0589 KW |
1769 | if (ret < 0) { |
1770 | return ret; | |
1771 | } | |
1772 | ||
a9a9f8f0 AG |
1773 | /* Limit nb_clusters to one L2 slice */ |
1774 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
b6d36def | 1775 | assert(nb_clusters <= INT_MAX); |
621f0589 KW |
1776 | |
1777 | for (i = 0; i < nb_clusters; i++) { | |
1778 | uint64_t old_offset; | |
06cc5e2b | 1779 | QCow2ClusterType cluster_type; |
621f0589 | 1780 | |
a9a9f8f0 | 1781 | old_offset = be64_to_cpu(l2_slice[l2_index + i]); |
621f0589 | 1782 | |
06cc5e2b EB |
1783 | /* |
1784 | * Minimize L2 changes if the cluster already reads back as | |
1785 | * zeroes with correct allocation. | |
1786 | */ | |
1787 | cluster_type = qcow2_get_cluster_type(old_offset); | |
1788 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN || | |
1789 | (cluster_type == QCOW2_CLUSTER_ZERO_ALLOC && !unmap)) { | |
1790 | continue; | |
1791 | } | |
1792 | ||
a9a9f8f0 | 1793 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
06cc5e2b | 1794 | if (cluster_type == QCOW2_CLUSTER_COMPRESSED || unmap) { |
a9a9f8f0 | 1795 | l2_slice[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
6cfcb9b8 | 1796 | qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST); |
621f0589 | 1797 | } else { |
a9a9f8f0 | 1798 | l2_slice[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO); |
621f0589 KW |
1799 | } |
1800 | } | |
1801 | ||
a9a9f8f0 | 1802 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
621f0589 KW |
1803 | |
1804 | return nb_clusters; | |
1805 | } | |
1806 | ||
d2cb36af EB |
1807 | int qcow2_cluster_zeroize(BlockDriverState *bs, uint64_t offset, |
1808 | uint64_t bytes, int flags) | |
621f0589 | 1809 | { |
ff99129a | 1810 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1811 | uint64_t end_offset = offset + bytes; |
b6d36def | 1812 | uint64_t nb_clusters; |
d2cb36af | 1813 | int64_t cleared; |
621f0589 KW |
1814 | int ret; |
1815 | ||
f10ee139 EB |
1816 | /* Caller must pass aligned values, except at image end */ |
1817 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); | |
1818 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || | |
1819 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
1820 | ||
621f0589 KW |
1821 | /* The zero flag is only supported by version 3 and newer */ |
1822 | if (s->qcow_version < 3) { | |
1823 | return -ENOTSUP; | |
1824 | } | |
1825 | ||
a9a9f8f0 | 1826 | /* Each L2 slice is handled by its own loop iteration */ |
d2cb36af | 1827 | nb_clusters = size_to_clusters(s, bytes); |
621f0589 | 1828 | |
0b919fae KW |
1829 | s->cache_discards = true; |
1830 | ||
621f0589 | 1831 | while (nb_clusters > 0) { |
a9a9f8f0 | 1832 | cleared = zero_in_l2_slice(bs, offset, nb_clusters, flags); |
d2cb36af EB |
1833 | if (cleared < 0) { |
1834 | ret = cleared; | |
0b919fae | 1835 | goto fail; |
621f0589 KW |
1836 | } |
1837 | ||
d2cb36af EB |
1838 | nb_clusters -= cleared; |
1839 | offset += (cleared * s->cluster_size); | |
621f0589 KW |
1840 | } |
1841 | ||
0b919fae KW |
1842 | ret = 0; |
1843 | fail: | |
1844 | s->cache_discards = false; | |
1845 | qcow2_process_discards(bs, ret); | |
1846 | ||
1847 | return ret; | |
621f0589 | 1848 | } |
32b6444d HR |
1849 | |
1850 | /* | |
1851 | * Expands all zero clusters in a specific L1 table (or deallocates them, for | |
1852 | * non-backed non-pre-allocated zero clusters). | |
1853 | * | |
4057a2b2 HR |
1854 | * l1_entries and *visited_l1_entries are used to keep track of progress for |
1855 | * status_cb(). l1_entries contains the total number of L1 entries and | |
1856 | * *visited_l1_entries counts all visited L1 entries. | |
32b6444d HR |
1857 | */ |
1858 | static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table, | |
ecf58777 | 1859 | int l1_size, int64_t *visited_l1_entries, |
4057a2b2 | 1860 | int64_t l1_entries, |
8b13976d HR |
1861 | BlockDriverAmendStatusCB *status_cb, |
1862 | void *cb_opaque) | |
32b6444d | 1863 | { |
ff99129a | 1864 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 1865 | bool is_active_l1 = (l1_table == s->l1_table); |
415184f5 AG |
1866 | uint64_t *l2_slice = NULL; |
1867 | unsigned slice, slice_size2, n_slices; | |
32b6444d HR |
1868 | int ret; |
1869 | int i, j; | |
1870 | ||
415184f5 AG |
1871 | slice_size2 = s->l2_slice_size * sizeof(uint64_t); |
1872 | n_slices = s->cluster_size / slice_size2; | |
1873 | ||
32b6444d HR |
1874 | if (!is_active_l1) { |
1875 | /* inactive L2 tables require a buffer to be stored in when loading | |
1876 | * them from disk */ | |
415184f5 AG |
1877 | l2_slice = qemu_try_blockalign(bs->file->bs, slice_size2); |
1878 | if (l2_slice == NULL) { | |
de82815d KW |
1879 | return -ENOMEM; |
1880 | } | |
32b6444d HR |
1881 | } |
1882 | ||
1883 | for (i = 0; i < l1_size; i++) { | |
1884 | uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK; | |
0e06528e | 1885 | uint64_t l2_refcount; |
32b6444d HR |
1886 | |
1887 | if (!l2_offset) { | |
1888 | /* unallocated */ | |
4057a2b2 HR |
1889 | (*visited_l1_entries)++; |
1890 | if (status_cb) { | |
8b13976d | 1891 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 1892 | } |
32b6444d HR |
1893 | continue; |
1894 | } | |
1895 | ||
8dd93d93 HR |
1896 | if (offset_into_cluster(s, l2_offset)) { |
1897 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" | |
1898 | PRIx64 " unaligned (L1 index: %#x)", | |
1899 | l2_offset, i); | |
1900 | ret = -EIO; | |
1901 | goto fail; | |
1902 | } | |
1903 | ||
9b765486 AG |
1904 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
1905 | &l2_refcount); | |
1906 | if (ret < 0) { | |
1907 | goto fail; | |
1908 | } | |
1909 | ||
415184f5 AG |
1910 | for (slice = 0; slice < n_slices; slice++) { |
1911 | uint64_t slice_offset = l2_offset + slice * slice_size2; | |
1912 | bool l2_dirty = false; | |
226494ff AG |
1913 | if (is_active_l1) { |
1914 | /* get active L2 tables from cache */ | |
415184f5 AG |
1915 | ret = qcow2_cache_get(bs, s->l2_table_cache, slice_offset, |
1916 | (void **)&l2_slice); | |
226494ff AG |
1917 | } else { |
1918 | /* load inactive L2 tables from disk */ | |
415184f5 | 1919 | ret = bdrv_pread(bs->file, slice_offset, l2_slice, slice_size2); |
226494ff AG |
1920 | } |
1921 | if (ret < 0) { | |
1922 | goto fail; | |
32b6444d HR |
1923 | } |
1924 | ||
415184f5 AG |
1925 | for (j = 0; j < s->l2_slice_size; j++) { |
1926 | uint64_t l2_entry = be64_to_cpu(l2_slice[j]); | |
226494ff AG |
1927 | int64_t offset = l2_entry & L2E_OFFSET_MASK; |
1928 | QCow2ClusterType cluster_type = | |
1929 | qcow2_get_cluster_type(l2_entry); | |
1930 | ||
1931 | if (cluster_type != QCOW2_CLUSTER_ZERO_PLAIN && | |
1932 | cluster_type != QCOW2_CLUSTER_ZERO_ALLOC) { | |
32b6444d HR |
1933 | continue; |
1934 | } | |
1935 | ||
226494ff AG |
1936 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
1937 | if (!bs->backing) { | |
1938 | /* not backed; therefore we can simply deallocate the | |
1939 | * cluster */ | |
415184f5 | 1940 | l2_slice[j] = 0; |
226494ff AG |
1941 | l2_dirty = true; |
1942 | continue; | |
1943 | } | |
1944 | ||
1945 | offset = qcow2_alloc_clusters(bs, s->cluster_size); | |
1946 | if (offset < 0) { | |
1947 | ret = offset; | |
1948 | goto fail; | |
1949 | } | |
ecf58777 | 1950 | |
226494ff AG |
1951 | if (l2_refcount > 1) { |
1952 | /* For shared L2 tables, set the refcount accordingly | |
1953 | * (it is already 1 and needs to be l2_refcount) */ | |
1954 | ret = qcow2_update_cluster_refcount( | |
1955 | bs, offset >> s->cluster_bits, | |
2aabe7c7 | 1956 | refcount_diff(1, l2_refcount), false, |
ecf58777 | 1957 | QCOW2_DISCARD_OTHER); |
226494ff AG |
1958 | if (ret < 0) { |
1959 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1960 | QCOW2_DISCARD_OTHER); | |
1961 | goto fail; | |
1962 | } | |
ecf58777 HR |
1963 | } |
1964 | } | |
32b6444d | 1965 | |
226494ff | 1966 | if (offset_into_cluster(s, offset)) { |
415184f5 | 1967 | int l2_index = slice * s->l2_slice_size + j; |
226494ff AG |
1968 | qcow2_signal_corruption( |
1969 | bs, true, -1, -1, | |
1970 | "Cluster allocation offset " | |
1971 | "%#" PRIx64 " unaligned (L2 offset: %#" | |
1972 | PRIx64 ", L2 index: %#x)", offset, | |
415184f5 | 1973 | l2_offset, l2_index); |
226494ff AG |
1974 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
1975 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1976 | QCOW2_DISCARD_ALWAYS); | |
1977 | } | |
1978 | ret = -EIO; | |
1979 | goto fail; | |
8dd93d93 | 1980 | } |
8dd93d93 | 1981 | |
226494ff AG |
1982 | ret = qcow2_pre_write_overlap_check(bs, 0, offset, |
1983 | s->cluster_size); | |
1984 | if (ret < 0) { | |
1985 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { | |
1986 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1987 | QCOW2_DISCARD_ALWAYS); | |
1988 | } | |
1989 | goto fail; | |
320c7066 | 1990 | } |
32b6444d | 1991 | |
226494ff AG |
1992 | ret = bdrv_pwrite_zeroes(bs->file, offset, s->cluster_size, 0); |
1993 | if (ret < 0) { | |
1994 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { | |
1995 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1996 | QCOW2_DISCARD_ALWAYS); | |
1997 | } | |
1998 | goto fail; | |
320c7066 | 1999 | } |
32b6444d | 2000 | |
226494ff | 2001 | if (l2_refcount == 1) { |
415184f5 | 2002 | l2_slice[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED); |
226494ff | 2003 | } else { |
415184f5 | 2004 | l2_slice[j] = cpu_to_be64(offset); |
226494ff AG |
2005 | } |
2006 | l2_dirty = true; | |
e390cf5a | 2007 | } |
32b6444d | 2008 | |
226494ff AG |
2009 | if (is_active_l1) { |
2010 | if (l2_dirty) { | |
415184f5 | 2011 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
226494ff | 2012 | qcow2_cache_depends_on_flush(s->l2_table_cache); |
32b6444d | 2013 | } |
415184f5 | 2014 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
226494ff AG |
2015 | } else { |
2016 | if (l2_dirty) { | |
2017 | ret = qcow2_pre_write_overlap_check( | |
2018 | bs, QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, | |
415184f5 | 2019 | slice_offset, slice_size2); |
226494ff AG |
2020 | if (ret < 0) { |
2021 | goto fail; | |
2022 | } | |
32b6444d | 2023 | |
415184f5 AG |
2024 | ret = bdrv_pwrite(bs->file, slice_offset, |
2025 | l2_slice, slice_size2); | |
226494ff AG |
2026 | if (ret < 0) { |
2027 | goto fail; | |
2028 | } | |
32b6444d HR |
2029 | } |
2030 | } | |
2031 | } | |
4057a2b2 HR |
2032 | |
2033 | (*visited_l1_entries)++; | |
2034 | if (status_cb) { | |
8b13976d | 2035 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 2036 | } |
32b6444d HR |
2037 | } |
2038 | ||
2039 | ret = 0; | |
2040 | ||
2041 | fail: | |
415184f5 | 2042 | if (l2_slice) { |
32b6444d | 2043 | if (!is_active_l1) { |
415184f5 | 2044 | qemu_vfree(l2_slice); |
32b6444d | 2045 | } else { |
415184f5 | 2046 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
32b6444d HR |
2047 | } |
2048 | } | |
2049 | return ret; | |
2050 | } | |
2051 | ||
2052 | /* | |
2053 | * For backed images, expands all zero clusters on the image. For non-backed | |
2054 | * images, deallocates all non-pre-allocated zero clusters (and claims the | |
2055 | * allocation for pre-allocated ones). This is important for downgrading to a | |
2056 | * qcow2 version which doesn't yet support metadata zero clusters. | |
2057 | */ | |
4057a2b2 | 2058 | int qcow2_expand_zero_clusters(BlockDriverState *bs, |
8b13976d HR |
2059 | BlockDriverAmendStatusCB *status_cb, |
2060 | void *cb_opaque) | |
32b6444d | 2061 | { |
ff99129a | 2062 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 2063 | uint64_t *l1_table = NULL; |
4057a2b2 | 2064 | int64_t l1_entries = 0, visited_l1_entries = 0; |
32b6444d HR |
2065 | int ret; |
2066 | int i, j; | |
2067 | ||
4057a2b2 HR |
2068 | if (status_cb) { |
2069 | l1_entries = s->l1_size; | |
2070 | for (i = 0; i < s->nb_snapshots; i++) { | |
2071 | l1_entries += s->snapshots[i].l1_size; | |
2072 | } | |
2073 | } | |
2074 | ||
32b6444d | 2075 | ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size, |
4057a2b2 | 2076 | &visited_l1_entries, l1_entries, |
8b13976d | 2077 | status_cb, cb_opaque); |
32b6444d HR |
2078 | if (ret < 0) { |
2079 | goto fail; | |
2080 | } | |
2081 | ||
2082 | /* Inactive L1 tables may point to active L2 tables - therefore it is | |
2083 | * necessary to flush the L2 table cache before trying to access the L2 | |
2084 | * tables pointed to by inactive L1 entries (else we might try to expand | |
2085 | * zero clusters that have already been expanded); furthermore, it is also | |
2086 | * necessary to empty the L2 table cache, since it may contain tables which | |
2087 | * are now going to be modified directly on disk, bypassing the cache. | |
2088 | * qcow2_cache_empty() does both for us. */ | |
2089 | ret = qcow2_cache_empty(bs, s->l2_table_cache); | |
2090 | if (ret < 0) { | |
2091 | goto fail; | |
2092 | } | |
2093 | ||
2094 | for (i = 0; i < s->nb_snapshots; i++) { | |
d737b78c LV |
2095 | int l1_sectors = DIV_ROUND_UP(s->snapshots[i].l1_size * |
2096 | sizeof(uint64_t), BDRV_SECTOR_SIZE); | |
32b6444d | 2097 | |
de7269d2 AG |
2098 | uint64_t *new_l1_table = |
2099 | g_try_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE); | |
2100 | ||
2101 | if (!new_l1_table) { | |
2102 | ret = -ENOMEM; | |
2103 | goto fail; | |
2104 | } | |
2105 | ||
2106 | l1_table = new_l1_table; | |
32b6444d | 2107 | |
fbcbbf4e | 2108 | ret = bdrv_read(bs->file, |
9a4f4c31 KW |
2109 | s->snapshots[i].l1_table_offset / BDRV_SECTOR_SIZE, |
2110 | (void *)l1_table, l1_sectors); | |
32b6444d HR |
2111 | if (ret < 0) { |
2112 | goto fail; | |
2113 | } | |
2114 | ||
2115 | for (j = 0; j < s->snapshots[i].l1_size; j++) { | |
2116 | be64_to_cpus(&l1_table[j]); | |
2117 | } | |
2118 | ||
2119 | ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size, | |
4057a2b2 | 2120 | &visited_l1_entries, l1_entries, |
8b13976d | 2121 | status_cb, cb_opaque); |
32b6444d HR |
2122 | if (ret < 0) { |
2123 | goto fail; | |
2124 | } | |
2125 | } | |
2126 | ||
2127 | ret = 0; | |
2128 | ||
2129 | fail: | |
32b6444d HR |
2130 | g_free(l1_table); |
2131 | return ret; | |
2132 | } |