]>
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 | ||
c9a442e4 | 28 | #include "qapi/error.h" |
0d8c41da | 29 | #include "qcow2.h" |
58369e22 | 30 | #include "qemu/bswap.h" |
3cce16f4 | 31 | #include "trace.h" |
45aba42f | 32 | |
46b732cd PB |
33 | int qcow2_shrink_l1_table(BlockDriverState *bs, uint64_t exact_size) |
34 | { | |
35 | BDRVQcow2State *s = bs->opaque; | |
36 | int new_l1_size, i, ret; | |
37 | ||
38 | if (exact_size >= s->l1_size) { | |
39 | return 0; | |
40 | } | |
41 | ||
42 | new_l1_size = exact_size; | |
43 | ||
44 | #ifdef DEBUG_ALLOC2 | |
45 | fprintf(stderr, "shrink l1_table from %d to %d\n", s->l1_size, new_l1_size); | |
46 | #endif | |
47 | ||
48 | BLKDBG_EVENT(bs->file, BLKDBG_L1_SHRINK_WRITE_TABLE); | |
49 | ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset + | |
50 | new_l1_size * sizeof(uint64_t), | |
51 | (s->l1_size - new_l1_size) * sizeof(uint64_t), 0); | |
52 | if (ret < 0) { | |
53 | goto fail; | |
54 | } | |
55 | ||
56 | ret = bdrv_flush(bs->file->bs); | |
57 | if (ret < 0) { | |
58 | goto fail; | |
59 | } | |
60 | ||
61 | BLKDBG_EVENT(bs->file, BLKDBG_L1_SHRINK_FREE_L2_CLUSTERS); | |
62 | for (i = s->l1_size - 1; i > new_l1_size - 1; i--) { | |
63 | if ((s->l1_table[i] & L1E_OFFSET_MASK) == 0) { | |
64 | continue; | |
65 | } | |
66 | qcow2_free_clusters(bs, s->l1_table[i] & L1E_OFFSET_MASK, | |
67 | s->cluster_size, QCOW2_DISCARD_ALWAYS); | |
68 | s->l1_table[i] = 0; | |
69 | } | |
70 | return 0; | |
71 | ||
72 | fail: | |
73 | /* | |
74 | * If the write in the l1_table failed the image may contain a partially | |
75 | * overwritten l1_table. In this case it would be better to clear the | |
76 | * l1_table in memory to avoid possible image corruption. | |
77 | */ | |
78 | memset(s->l1_table + new_l1_size, 0, | |
79 | (s->l1_size - new_l1_size) * sizeof(uint64_t)); | |
80 | return ret; | |
81 | } | |
82 | ||
2cf7cfa1 KW |
83 | int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size, |
84 | bool exact_size) | |
45aba42f | 85 | { |
ff99129a | 86 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 | 87 | int new_l1_size2, ret, i; |
45aba42f | 88 | uint64_t *new_l1_table; |
fda74f82 | 89 | int64_t old_l1_table_offset, old_l1_size; |
2cf7cfa1 | 90 | int64_t new_l1_table_offset, new_l1_size; |
45aba42f KW |
91 | uint8_t data[12]; |
92 | ||
72893756 | 93 | if (min_size <= s->l1_size) |
45aba42f | 94 | return 0; |
72893756 | 95 | |
b93f9950 HR |
96 | /* Do a sanity check on min_size before trying to calculate new_l1_size |
97 | * (this prevents overflows during the while loop for the calculation of | |
98 | * new_l1_size) */ | |
99 | if (min_size > INT_MAX / sizeof(uint64_t)) { | |
100 | return -EFBIG; | |
101 | } | |
102 | ||
72893756 SH |
103 | if (exact_size) { |
104 | new_l1_size = min_size; | |
105 | } else { | |
106 | /* Bump size up to reduce the number of times we have to grow */ | |
107 | new_l1_size = s->l1_size; | |
108 | if (new_l1_size == 0) { | |
109 | new_l1_size = 1; | |
110 | } | |
111 | while (min_size > new_l1_size) { | |
21cf3e12 | 112 | new_l1_size = DIV_ROUND_UP(new_l1_size * 3, 2); |
72893756 | 113 | } |
45aba42f | 114 | } |
72893756 | 115 | |
84c26520 HR |
116 | QEMU_BUILD_BUG_ON(QCOW_MAX_L1_SIZE > INT_MAX); |
117 | if (new_l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) { | |
2cf7cfa1 KW |
118 | return -EFBIG; |
119 | } | |
120 | ||
45aba42f | 121 | #ifdef DEBUG_ALLOC2 |
2cf7cfa1 KW |
122 | fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n", |
123 | s->l1_size, new_l1_size); | |
45aba42f KW |
124 | #endif |
125 | ||
126 | new_l1_size2 = sizeof(uint64_t) * new_l1_size; | |
ef97d608 | 127 | new_l1_table = qemu_try_blockalign(bs->file->bs, new_l1_size2); |
de82815d KW |
128 | if (new_l1_table == NULL) { |
129 | return -ENOMEM; | |
130 | } | |
ef97d608 | 131 | memset(new_l1_table, 0, new_l1_size2); |
de82815d | 132 | |
0647d47c SH |
133 | if (s->l1_size) { |
134 | memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t)); | |
135 | } | |
45aba42f KW |
136 | |
137 | /* write new table (align to cluster) */ | |
66f82cee | 138 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE); |
ed6ccf0f | 139 | new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2); |
5d757b56 | 140 | if (new_l1_table_offset < 0) { |
de82815d | 141 | qemu_vfree(new_l1_table); |
5d757b56 KW |
142 | return new_l1_table_offset; |
143 | } | |
29c1a730 KW |
144 | |
145 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
146 | if (ret < 0) { | |
80fa3341 | 147 | goto fail; |
29c1a730 | 148 | } |
45aba42f | 149 | |
cf93980e HR |
150 | /* the L1 position has not yet been updated, so these clusters must |
151 | * indeed be completely free */ | |
231bb267 | 152 | ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset, |
966b000f | 153 | new_l1_size2, false); |
cf93980e HR |
154 | if (ret < 0) { |
155 | goto fail; | |
156 | } | |
157 | ||
66f82cee | 158 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE); |
45aba42f KW |
159 | for(i = 0; i < s->l1_size; i++) |
160 | new_l1_table[i] = cpu_to_be64(new_l1_table[i]); | |
d9ca2ea2 | 161 | ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, |
9a4f4c31 | 162 | new_l1_table, new_l1_size2); |
8b3b7206 | 163 | if (ret < 0) |
45aba42f KW |
164 | goto fail; |
165 | for(i = 0; i < s->l1_size; i++) | |
166 | new_l1_table[i] = be64_to_cpu(new_l1_table[i]); | |
167 | ||
168 | /* set new table */ | |
66f82cee | 169 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE); |
f1f7a1dd | 170 | stl_be_p(data, new_l1_size); |
e4ef9f46 | 171 | stq_be_p(data + 4, new_l1_table_offset); |
d9ca2ea2 | 172 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), |
9a4f4c31 | 173 | data, sizeof(data)); |
8b3b7206 | 174 | if (ret < 0) { |
45aba42f | 175 | goto fail; |
fb8fa77c | 176 | } |
de82815d | 177 | qemu_vfree(s->l1_table); |
fda74f82 | 178 | old_l1_table_offset = s->l1_table_offset; |
45aba42f KW |
179 | s->l1_table_offset = new_l1_table_offset; |
180 | s->l1_table = new_l1_table; | |
fda74f82 | 181 | old_l1_size = s->l1_size; |
45aba42f | 182 | s->l1_size = new_l1_size; |
fda74f82 HR |
183 | qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * sizeof(uint64_t), |
184 | QCOW2_DISCARD_OTHER); | |
45aba42f KW |
185 | return 0; |
186 | fail: | |
de82815d | 187 | qemu_vfree(new_l1_table); |
6cfcb9b8 KW |
188 | qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2, |
189 | QCOW2_DISCARD_OTHER); | |
8b3b7206 | 190 | return ret; |
45aba42f KW |
191 | } |
192 | ||
45aba42f KW |
193 | /* |
194 | * l2_load | |
195 | * | |
e2b5713e AG |
196 | * @bs: The BlockDriverState |
197 | * @offset: A guest offset, used to calculate what slice of the L2 | |
198 | * table to load. | |
199 | * @l2_offset: Offset to the L2 table in the image file. | |
200 | * @l2_slice: Location to store the pointer to the L2 slice. | |
45aba42f | 201 | * |
e2b5713e AG |
202 | * Loads a L2 slice into memory (L2 slices are the parts of L2 tables |
203 | * that are loaded by the qcow2 cache). If the slice is in the cache, | |
204 | * the cache is used; otherwise the L2 slice is loaded from the image | |
205 | * file. | |
45aba42f | 206 | */ |
e2b5713e AG |
207 | static int l2_load(BlockDriverState *bs, uint64_t offset, |
208 | uint64_t l2_offset, uint64_t **l2_slice) | |
45aba42f | 209 | { |
ff99129a | 210 | BDRVQcow2State *s = bs->opaque; |
e2b5713e AG |
211 | int start_of_slice = sizeof(uint64_t) * |
212 | (offset_to_l2_index(s, offset) - offset_to_l2_slice_index(s, offset)); | |
45aba42f | 213 | |
e2b5713e AG |
214 | return qcow2_cache_get(bs, s->l2_table_cache, l2_offset + start_of_slice, |
215 | (void **)l2_slice); | |
45aba42f KW |
216 | } |
217 | ||
6583e3c7 KW |
218 | /* |
219 | * Writes one sector of the L1 table to the disk (can't update single entries | |
220 | * and we really don't want bdrv_pread to perform a read-modify-write) | |
221 | */ | |
222 | #define L1_ENTRIES_PER_SECTOR (512 / 8) | |
e23e400e | 223 | int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index) |
6583e3c7 | 224 | { |
ff99129a | 225 | BDRVQcow2State *s = bs->opaque; |
a1391444 | 226 | uint64_t buf[L1_ENTRIES_PER_SECTOR] = { 0 }; |
6583e3c7 | 227 | int l1_start_index; |
f7defcb6 | 228 | int i, ret; |
6583e3c7 KW |
229 | |
230 | l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1); | |
a1391444 HR |
231 | for (i = 0; i < L1_ENTRIES_PER_SECTOR && l1_start_index + i < s->l1_size; |
232 | i++) | |
233 | { | |
6583e3c7 KW |
234 | buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]); |
235 | } | |
236 | ||
231bb267 | 237 | ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1, |
966b000f | 238 | s->l1_table_offset + 8 * l1_start_index, sizeof(buf), false); |
cf93980e HR |
239 | if (ret < 0) { |
240 | return ret; | |
241 | } | |
242 | ||
66f82cee | 243 | BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); |
d9ca2ea2 | 244 | ret = bdrv_pwrite_sync(bs->file, |
9a4f4c31 KW |
245 | s->l1_table_offset + 8 * l1_start_index, |
246 | buf, sizeof(buf)); | |
f7defcb6 KW |
247 | if (ret < 0) { |
248 | return ret; | |
6583e3c7 KW |
249 | } |
250 | ||
251 | return 0; | |
252 | } | |
253 | ||
45aba42f KW |
254 | /* |
255 | * l2_allocate | |
256 | * | |
257 | * Allocate a new l2 entry in the file. If l1_index points to an already | |
258 | * used entry in the L2 table (i.e. we are doing a copy on write for the L2 | |
259 | * table) copy the contents of the old L2 table into the newly allocated one. | |
260 | * Otherwise the new table is initialized with zeros. | |
261 | * | |
262 | */ | |
263 | ||
3861946a | 264 | static int l2_allocate(BlockDriverState *bs, int l1_index) |
45aba42f | 265 | { |
ff99129a | 266 | BDRVQcow2State *s = bs->opaque; |
6583e3c7 | 267 | uint64_t old_l2_offset; |
3861946a AG |
268 | uint64_t *l2_slice = NULL; |
269 | unsigned slice, slice_size2, n_slices; | |
f4f0d391 | 270 | int64_t l2_offset; |
c46e1167 | 271 | int ret; |
45aba42f KW |
272 | |
273 | old_l2_offset = s->l1_table[l1_index]; | |
274 | ||
3cce16f4 KW |
275 | trace_qcow2_l2_allocate(bs, l1_index); |
276 | ||
45aba42f KW |
277 | /* allocate a new l2 entry */ |
278 | ||
ed6ccf0f | 279 | l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t)); |
5d757b56 | 280 | if (l2_offset < 0) { |
be0b742e HR |
281 | ret = l2_offset; |
282 | goto fail; | |
5d757b56 | 283 | } |
29c1a730 | 284 | |
c1c43990 AG |
285 | /* The offset must fit in the offset field of the L1 table entry */ |
286 | assert((l2_offset & L1E_OFFSET_MASK) == l2_offset); | |
287 | ||
98839750 AG |
288 | /* If we're allocating the table at offset 0 then something is wrong */ |
289 | if (l2_offset == 0) { | |
290 | qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid " | |
291 | "allocation of L2 table at offset 0"); | |
292 | ret = -EIO; | |
293 | goto fail; | |
294 | } | |
295 | ||
29c1a730 KW |
296 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
297 | if (ret < 0) { | |
298 | goto fail; | |
299 | } | |
45aba42f | 300 | |
45aba42f KW |
301 | /* allocate a new entry in the l2 cache */ |
302 | ||
3861946a AG |
303 | slice_size2 = s->l2_slice_size * sizeof(uint64_t); |
304 | n_slices = s->cluster_size / slice_size2; | |
305 | ||
3cce16f4 | 306 | trace_qcow2_l2_allocate_get_empty(bs, l1_index); |
3861946a | 307 | for (slice = 0; slice < n_slices; slice++) { |
6580bb09 | 308 | ret = qcow2_cache_get_empty(bs, s->l2_table_cache, |
3861946a AG |
309 | l2_offset + slice * slice_size2, |
310 | (void **) &l2_slice); | |
6580bb09 AG |
311 | if (ret < 0) { |
312 | goto fail; | |
313 | } | |
29c1a730 | 314 | |
6580bb09 | 315 | if ((old_l2_offset & L1E_OFFSET_MASK) == 0) { |
3861946a AG |
316 | /* if there was no old l2 table, clear the new slice */ |
317 | memset(l2_slice, 0, slice_size2); | |
6580bb09 | 318 | } else { |
3861946a AG |
319 | uint64_t *old_slice; |
320 | uint64_t old_l2_slice_offset = | |
321 | (old_l2_offset & L1E_OFFSET_MASK) + slice * slice_size2; | |
29c1a730 | 322 | |
3861946a | 323 | /* if there was an old l2 table, read a slice from the disk */ |
6580bb09 | 324 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ); |
3861946a AG |
325 | ret = qcow2_cache_get(bs, s->l2_table_cache, old_l2_slice_offset, |
326 | (void **) &old_slice); | |
6580bb09 AG |
327 | if (ret < 0) { |
328 | goto fail; | |
329 | } | |
330 | ||
3861946a | 331 | memcpy(l2_slice, old_slice, slice_size2); |
6580bb09 | 332 | |
3861946a | 333 | qcow2_cache_put(s->l2_table_cache, (void **) &old_slice); |
29c1a730 KW |
334 | } |
335 | ||
3861946a | 336 | /* write the l2 slice to the file */ |
6580bb09 | 337 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE); |
29c1a730 | 338 | |
6580bb09 | 339 | trace_qcow2_l2_allocate_write_l2(bs, l1_index); |
3861946a AG |
340 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
341 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
45aba42f | 342 | } |
29c1a730 | 343 | |
29c1a730 | 344 | ret = qcow2_cache_flush(bs, s->l2_table_cache); |
c46e1167 | 345 | if (ret < 0) { |
175e1152 KW |
346 | goto fail; |
347 | } | |
348 | ||
349 | /* update the L1 entry */ | |
3cce16f4 | 350 | trace_qcow2_l2_allocate_write_l1(bs, l1_index); |
175e1152 | 351 | s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; |
e23e400e | 352 | ret = qcow2_write_l1_entry(bs, l1_index); |
175e1152 KW |
353 | if (ret < 0) { |
354 | goto fail; | |
c46e1167 | 355 | } |
45aba42f | 356 | |
3cce16f4 | 357 | trace_qcow2_l2_allocate_done(bs, l1_index, 0); |
c46e1167 | 358 | return 0; |
175e1152 KW |
359 | |
360 | fail: | |
3cce16f4 | 361 | trace_qcow2_l2_allocate_done(bs, l1_index, ret); |
3861946a AG |
362 | if (l2_slice != NULL) { |
363 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
8585afd8 | 364 | } |
68dba0bf | 365 | s->l1_table[l1_index] = old_l2_offset; |
e3b21ef9 HR |
366 | if (l2_offset > 0) { |
367 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), | |
368 | QCOW2_DISCARD_ALWAYS); | |
369 | } | |
175e1152 | 370 | return ret; |
45aba42f KW |
371 | } |
372 | ||
2bfcc4a0 | 373 | /* |
13f893c4 | 374 | * Checks how many clusters in a given L2 slice are contiguous in the image |
2bfcc4a0 KW |
375 | * file. As soon as one of the flags in the bitmask stop_flags changes compared |
376 | * to the first cluster, the search is stopped and the cluster is not counted | |
377 | * as contiguous. (This allows it, for example, to stop at the first compressed | |
378 | * cluster which may require a different handling) | |
379 | */ | |
808c2bb4 KW |
380 | static int count_contiguous_clusters(BlockDriverState *bs, int nb_clusters, |
381 | int cluster_size, uint64_t *l2_slice, uint64_t stop_flags) | |
45aba42f KW |
382 | { |
383 | int i; | |
3ef95218 | 384 | QCow2ClusterType first_cluster_type; |
78a52ad5 | 385 | uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED; |
13f893c4 | 386 | uint64_t first_entry = be64_to_cpu(l2_slice[0]); |
15684a47 | 387 | uint64_t offset = first_entry & mask; |
45aba42f | 388 | |
b8c8353a KW |
389 | first_cluster_type = qcow2_get_cluster_type(bs, first_entry); |
390 | if (first_cluster_type == QCOW2_CLUSTER_UNALLOCATED) { | |
45aba42f | 391 | return 0; |
564a6b69 | 392 | } |
45aba42f | 393 | |
564a6b69 | 394 | /* must be allocated */ |
564a6b69 | 395 | assert(first_cluster_type == QCOW2_CLUSTER_NORMAL || |
fdfab37d | 396 | first_cluster_type == QCOW2_CLUSTER_ZERO_ALLOC); |
15684a47 | 397 | |
61653008 | 398 | for (i = 0; i < nb_clusters; i++) { |
13f893c4 | 399 | uint64_t l2_entry = be64_to_cpu(l2_slice[i]) & mask; |
2bfcc4a0 | 400 | if (offset + (uint64_t) i * cluster_size != l2_entry) { |
45aba42f | 401 | break; |
2bfcc4a0 KW |
402 | } |
403 | } | |
45aba42f | 404 | |
7d37435b | 405 | return i; |
45aba42f KW |
406 | } |
407 | ||
4341df8a EB |
408 | /* |
409 | * Checks how many consecutive unallocated clusters in a given L2 | |
c26f10ba | 410 | * slice have the same cluster type. |
4341df8a | 411 | */ |
808c2bb4 KW |
412 | static int count_contiguous_clusters_unallocated(BlockDriverState *bs, |
413 | int nb_clusters, | |
c26f10ba | 414 | uint64_t *l2_slice, |
3ef95218 | 415 | QCow2ClusterType wanted_type) |
45aba42f | 416 | { |
2bfcc4a0 KW |
417 | int i; |
418 | ||
fdfab37d | 419 | assert(wanted_type == QCOW2_CLUSTER_ZERO_PLAIN || |
4341df8a | 420 | wanted_type == QCOW2_CLUSTER_UNALLOCATED); |
2bfcc4a0 | 421 | for (i = 0; i < nb_clusters; i++) { |
c26f10ba | 422 | uint64_t entry = be64_to_cpu(l2_slice[i]); |
808c2bb4 | 423 | QCow2ClusterType type = qcow2_get_cluster_type(bs, entry); |
45aba42f | 424 | |
fdfab37d | 425 | if (type != wanted_type) { |
2bfcc4a0 KW |
426 | break; |
427 | } | |
428 | } | |
45aba42f KW |
429 | |
430 | return i; | |
431 | } | |
432 | ||
672f0f2c AG |
433 | static int coroutine_fn do_perform_cow_read(BlockDriverState *bs, |
434 | uint64_t src_cluster_offset, | |
435 | unsigned offset_in_cluster, | |
86b862c4 | 436 | QEMUIOVector *qiov) |
45aba42f | 437 | { |
aaa4d20b | 438 | int ret; |
1b9f1491 | 439 | |
86b862c4 | 440 | if (qiov->size == 0) { |
99450c6f AG |
441 | return 0; |
442 | } | |
443 | ||
66f82cee | 444 | BLKDBG_EVENT(bs->file, BLKDBG_COW_READ); |
aef4acb6 | 445 | |
dba28555 | 446 | if (!bs->drv) { |
672f0f2c | 447 | return -ENOMEDIUM; |
dba28555 HR |
448 | } |
449 | ||
aef4acb6 SH |
450 | /* Call .bdrv_co_readv() directly instead of using the public block-layer |
451 | * interface. This avoids double I/O throttling and request tracking, | |
452 | * which can lead to deadlock when block layer copy-on-read is enabled. | |
453 | */ | |
df893d25 VSO |
454 | ret = bs->drv->bdrv_co_preadv_part(bs, |
455 | src_cluster_offset + offset_in_cluster, | |
456 | qiov->size, qiov, 0, 0); | |
1b9f1491 | 457 | if (ret < 0) { |
672f0f2c | 458 | return ret; |
1b9f1491 KW |
459 | } |
460 | ||
672f0f2c AG |
461 | return 0; |
462 | } | |
463 | ||
672f0f2c AG |
464 | static int coroutine_fn do_perform_cow_write(BlockDriverState *bs, |
465 | uint64_t cluster_offset, | |
466 | unsigned offset_in_cluster, | |
86b862c4 | 467 | QEMUIOVector *qiov) |
672f0f2c | 468 | { |
966b000f | 469 | BDRVQcow2State *s = bs->opaque; |
672f0f2c AG |
470 | int ret; |
471 | ||
86b862c4 | 472 | if (qiov->size == 0) { |
672f0f2c AG |
473 | return 0; |
474 | } | |
475 | ||
231bb267 | 476 | ret = qcow2_pre_write_overlap_check(bs, 0, |
966b000f | 477 | cluster_offset + offset_in_cluster, qiov->size, true); |
cf93980e | 478 | if (ret < 0) { |
672f0f2c | 479 | return ret; |
cf93980e HR |
480 | } |
481 | ||
66f82cee | 482 | BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE); |
966b000f | 483 | ret = bdrv_co_pwritev(s->data_file, cluster_offset + offset_in_cluster, |
86b862c4 | 484 | qiov->size, qiov, 0); |
1b9f1491 | 485 | if (ret < 0) { |
672f0f2c | 486 | return ret; |
1b9f1491 KW |
487 | } |
488 | ||
672f0f2c | 489 | return 0; |
45aba42f KW |
490 | } |
491 | ||
492 | ||
493 | /* | |
494 | * get_cluster_offset | |
495 | * | |
ecfe1863 KW |
496 | * For a given offset of the virtual disk, find the cluster type and offset in |
497 | * the qcow2 file. The offset is stored in *cluster_offset. | |
45aba42f | 498 | * |
ecfe1863 KW |
499 | * On entry, *bytes is the maximum number of contiguous bytes starting at |
500 | * offset that we are interested in. | |
45aba42f | 501 | * |
ecfe1863 KW |
502 | * On exit, *bytes is the number of bytes starting at offset that have the same |
503 | * cluster type and (if applicable) are stored contiguously in the image file. | |
504 | * Compressed clusters are always returned one by one. | |
45aba42f | 505 | * |
68d000a3 KW |
506 | * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error |
507 | * cases. | |
45aba42f | 508 | */ |
1c46efaa | 509 | int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, |
ecfe1863 | 510 | unsigned int *bytes, uint64_t *cluster_offset) |
45aba42f | 511 | { |
ff99129a | 512 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 | 513 | unsigned int l2_index; |
fd630039 AG |
514 | uint64_t l1_index, l2_offset, *l2_slice; |
515 | int c; | |
c834cba9 HR |
516 | unsigned int offset_in_cluster; |
517 | uint64_t bytes_available, bytes_needed, nb_clusters; | |
3ef95218 | 518 | QCow2ClusterType type; |
55c17e98 | 519 | int ret; |
45aba42f | 520 | |
b2f65d6b | 521 | offset_in_cluster = offset_into_cluster(s, offset); |
ecfe1863 | 522 | bytes_needed = (uint64_t) *bytes + offset_in_cluster; |
45aba42f | 523 | |
b2f65d6b | 524 | /* compute how many bytes there are between the start of the cluster |
fd630039 AG |
525 | * containing offset and the end of the l2 slice that contains |
526 | * the entry pointing to it */ | |
527 | bytes_available = | |
528 | ((uint64_t) (s->l2_slice_size - offset_to_l2_slice_index(s, offset))) | |
529 | << s->cluster_bits; | |
45aba42f | 530 | |
b2f65d6b KW |
531 | if (bytes_needed > bytes_available) { |
532 | bytes_needed = bytes_available; | |
45aba42f KW |
533 | } |
534 | ||
1c46efaa | 535 | *cluster_offset = 0; |
45aba42f | 536 | |
b6af0975 | 537 | /* seek to the l2 offset in the l1 table */ |
45aba42f | 538 | |
05b5b6ee | 539 | l1_index = offset_to_l1_index(s, offset); |
68d000a3 | 540 | if (l1_index >= s->l1_size) { |
3ef95218 | 541 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 542 | goto out; |
68d000a3 | 543 | } |
45aba42f | 544 | |
68d000a3 KW |
545 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
546 | if (!l2_offset) { | |
3ef95218 | 547 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 548 | goto out; |
68d000a3 | 549 | } |
45aba42f | 550 | |
a97c67ee HR |
551 | if (offset_into_cluster(s, l2_offset)) { |
552 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
553 | " unaligned (L1 index: %#" PRIx64 ")", | |
554 | l2_offset, l1_index); | |
555 | return -EIO; | |
556 | } | |
557 | ||
fd630039 | 558 | /* load the l2 slice in memory */ |
45aba42f | 559 | |
fd630039 | 560 | ret = l2_load(bs, offset, l2_offset, &l2_slice); |
55c17e98 KW |
561 | if (ret < 0) { |
562 | return ret; | |
1c46efaa | 563 | } |
45aba42f KW |
564 | |
565 | /* find the cluster offset for the given disk offset */ | |
566 | ||
fd630039 AG |
567 | l2_index = offset_to_l2_slice_index(s, offset); |
568 | *cluster_offset = be64_to_cpu(l2_slice[l2_index]); | |
b6d36def | 569 | |
b2f65d6b | 570 | nb_clusters = size_to_clusters(s, bytes_needed); |
c834cba9 HR |
571 | /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned |
572 | * integers; the minimum cluster size is 512, so this assertion is always | |
573 | * true */ | |
574 | assert(nb_clusters <= INT_MAX); | |
45aba42f | 575 | |
808c2bb4 | 576 | type = qcow2_get_cluster_type(bs, *cluster_offset); |
fdfab37d EB |
577 | if (s->qcow_version < 3 && (type == QCOW2_CLUSTER_ZERO_PLAIN || |
578 | type == QCOW2_CLUSTER_ZERO_ALLOC)) { | |
579 | qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found" | |
580 | " in pre-v3 image (L2 offset: %#" PRIx64 | |
581 | ", L2 index: %#x)", l2_offset, l2_index); | |
582 | ret = -EIO; | |
583 | goto fail; | |
584 | } | |
3ef95218 | 585 | switch (type) { |
68d000a3 | 586 | case QCOW2_CLUSTER_COMPRESSED: |
966b000f KW |
587 | if (has_data_file(bs)) { |
588 | qcow2_signal_corruption(bs, true, -1, -1, "Compressed cluster " | |
589 | "entry found in image with external data " | |
590 | "file (L2 offset: %#" PRIx64 ", L2 index: " | |
591 | "%#x)", l2_offset, l2_index); | |
592 | ret = -EIO; | |
593 | goto fail; | |
594 | } | |
68d000a3 KW |
595 | /* Compressed clusters can only be processed one by one */ |
596 | c = 1; | |
597 | *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK; | |
598 | break; | |
fdfab37d | 599 | case QCOW2_CLUSTER_ZERO_PLAIN: |
68d000a3 | 600 | case QCOW2_CLUSTER_UNALLOCATED: |
45aba42f | 601 | /* how many empty clusters ? */ |
808c2bb4 | 602 | c = count_contiguous_clusters_unallocated(bs, nb_clusters, |
fd630039 | 603 | &l2_slice[l2_index], type); |
68d000a3 KW |
604 | *cluster_offset = 0; |
605 | break; | |
fdfab37d | 606 | case QCOW2_CLUSTER_ZERO_ALLOC: |
68d000a3 | 607 | case QCOW2_CLUSTER_NORMAL: |
45aba42f | 608 | /* how many allocated clusters ? */ |
808c2bb4 | 609 | c = count_contiguous_clusters(bs, nb_clusters, s->cluster_size, |
fd630039 | 610 | &l2_slice[l2_index], QCOW_OFLAG_ZERO); |
68d000a3 | 611 | *cluster_offset &= L2E_OFFSET_MASK; |
a97c67ee | 612 | if (offset_into_cluster(s, *cluster_offset)) { |
fdfab37d EB |
613 | qcow2_signal_corruption(bs, true, -1, -1, |
614 | "Cluster allocation offset %#" | |
a97c67ee HR |
615 | PRIx64 " unaligned (L2 offset: %#" PRIx64 |
616 | ", L2 index: %#x)", *cluster_offset, | |
617 | l2_offset, l2_index); | |
618 | ret = -EIO; | |
619 | goto fail; | |
620 | } | |
966b000f KW |
621 | if (has_data_file(bs) && *cluster_offset != offset - offset_in_cluster) |
622 | { | |
623 | qcow2_signal_corruption(bs, true, -1, -1, | |
624 | "External data file host cluster offset %#" | |
625 | PRIx64 " does not match guest cluster " | |
626 | "offset: %#" PRIx64 | |
627 | ", L2 index: %#x)", *cluster_offset, | |
628 | offset - offset_in_cluster, l2_index); | |
629 | ret = -EIO; | |
630 | goto fail; | |
631 | } | |
68d000a3 | 632 | break; |
1417d7e4 KW |
633 | default: |
634 | abort(); | |
45aba42f KW |
635 | } |
636 | ||
fd630039 | 637 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
29c1a730 | 638 | |
c834cba9 | 639 | bytes_available = (int64_t)c * s->cluster_size; |
68d000a3 | 640 | |
45aba42f | 641 | out: |
b2f65d6b KW |
642 | if (bytes_available > bytes_needed) { |
643 | bytes_available = bytes_needed; | |
644 | } | |
45aba42f | 645 | |
c834cba9 HR |
646 | /* bytes_available <= bytes_needed <= *bytes + offset_in_cluster; |
647 | * subtracting offset_in_cluster will therefore definitely yield something | |
648 | * not exceeding UINT_MAX */ | |
649 | assert(bytes_available - offset_in_cluster <= UINT_MAX); | |
ecfe1863 | 650 | *bytes = bytes_available - offset_in_cluster; |
45aba42f | 651 | |
3ef95218 | 652 | return type; |
a97c67ee HR |
653 | |
654 | fail: | |
fd630039 | 655 | qcow2_cache_put(s->l2_table_cache, (void **)&l2_slice); |
a97c67ee | 656 | return ret; |
45aba42f KW |
657 | } |
658 | ||
659 | /* | |
660 | * get_cluster_table | |
661 | * | |
662 | * for a given disk offset, load (and allocate if needed) | |
c03bfc5b | 663 | * the appropriate slice of its l2 table. |
45aba42f | 664 | * |
c03bfc5b | 665 | * the cluster index in the l2 slice is given to the caller. |
45aba42f | 666 | * |
1e3e8f1a | 667 | * Returns 0 on success, -errno in failure case |
45aba42f | 668 | */ |
45aba42f | 669 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
c03bfc5b | 670 | uint64_t **new_l2_slice, |
45aba42f KW |
671 | int *new_l2_index) |
672 | { | |
ff99129a | 673 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
674 | unsigned int l2_index; |
675 | uint64_t l1_index, l2_offset; | |
c03bfc5b | 676 | uint64_t *l2_slice = NULL; |
80ee15a6 | 677 | int ret; |
45aba42f | 678 | |
b6af0975 | 679 | /* seek to the l2 offset in the l1 table */ |
45aba42f | 680 | |
05b5b6ee | 681 | l1_index = offset_to_l1_index(s, offset); |
45aba42f | 682 | if (l1_index >= s->l1_size) { |
72893756 | 683 | ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
1e3e8f1a KW |
684 | if (ret < 0) { |
685 | return ret; | |
686 | } | |
45aba42f | 687 | } |
8e37f681 | 688 | |
2cf7cfa1 | 689 | assert(l1_index < s->l1_size); |
8e37f681 | 690 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
a97c67ee HR |
691 | if (offset_into_cluster(s, l2_offset)) { |
692 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
693 | " unaligned (L1 index: %#" PRIx64 ")", | |
694 | l2_offset, l1_index); | |
695 | return -EIO; | |
696 | } | |
45aba42f | 697 | |
05f9ee46 | 698 | if (!(s->l1_table[l1_index] & QCOW_OFLAG_COPIED)) { |
16fde5f2 | 699 | /* First allocate a new L2 table (and do COW if needed) */ |
3861946a | 700 | ret = l2_allocate(bs, l1_index); |
c46e1167 KW |
701 | if (ret < 0) { |
702 | return ret; | |
1e3e8f1a | 703 | } |
16fde5f2 KW |
704 | |
705 | /* Then decrease the refcount of the old table */ | |
706 | if (l2_offset) { | |
6cfcb9b8 KW |
707 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), |
708 | QCOW2_DISCARD_OTHER); | |
16fde5f2 | 709 | } |
3861946a AG |
710 | |
711 | /* Get the offset of the newly-allocated l2 table */ | |
712 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; | |
713 | assert(offset_into_cluster(s, l2_offset) == 0); | |
05f9ee46 AG |
714 | } |
715 | ||
c03bfc5b AG |
716 | /* load the l2 slice in memory */ |
717 | ret = l2_load(bs, offset, l2_offset, &l2_slice); | |
05f9ee46 AG |
718 | if (ret < 0) { |
719 | return ret; | |
45aba42f KW |
720 | } |
721 | ||
722 | /* find the cluster offset for the given disk offset */ | |
723 | ||
c03bfc5b | 724 | l2_index = offset_to_l2_slice_index(s, offset); |
45aba42f | 725 | |
c03bfc5b | 726 | *new_l2_slice = l2_slice; |
45aba42f KW |
727 | *new_l2_index = l2_index; |
728 | ||
1e3e8f1a | 729 | return 0; |
45aba42f KW |
730 | } |
731 | ||
732 | /* | |
733 | * alloc_compressed_cluster_offset | |
734 | * | |
77e023ff KW |
735 | * For a given offset on the virtual disk, allocate a new compressed cluster |
736 | * and put the host offset of the cluster into *host_offset. If a cluster is | |
737 | * already allocated at the offset, return an error. | |
45aba42f | 738 | * |
77e023ff | 739 | * Return 0 on success and -errno in error cases |
45aba42f | 740 | */ |
77e023ff KW |
741 | int qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
742 | uint64_t offset, | |
743 | int compressed_size, | |
744 | uint64_t *host_offset) | |
45aba42f | 745 | { |
ff99129a | 746 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 747 | int l2_index, ret; |
e4e72548 | 748 | uint64_t *l2_slice; |
f4f0d391 | 749 | int64_t cluster_offset; |
45aba42f KW |
750 | int nb_csectors; |
751 | ||
966b000f KW |
752 | if (has_data_file(bs)) { |
753 | return 0; | |
754 | } | |
755 | ||
e4e72548 | 756 | ret = get_cluster_table(bs, offset, &l2_slice, &l2_index); |
1e3e8f1a | 757 | if (ret < 0) { |
77e023ff | 758 | return ret; |
1e3e8f1a | 759 | } |
45aba42f | 760 | |
b0b6862e KW |
761 | /* Compression can't overwrite anything. Fail if the cluster was already |
762 | * allocated. */ | |
e4e72548 | 763 | cluster_offset = be64_to_cpu(l2_slice[l2_index]); |
b0b6862e | 764 | if (cluster_offset & L2E_OFFSET_MASK) { |
e4e72548 | 765 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
77e023ff | 766 | return -EIO; |
8f1efd00 | 767 | } |
45aba42f | 768 | |
ed6ccf0f | 769 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
5d757b56 | 770 | if (cluster_offset < 0) { |
e4e72548 | 771 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
77e023ff | 772 | return cluster_offset; |
5d757b56 KW |
773 | } |
774 | ||
b6c24694 AG |
775 | nb_csectors = |
776 | (cluster_offset + compressed_size - 1) / QCOW2_COMPRESSED_SECTOR_SIZE - | |
777 | (cluster_offset / QCOW2_COMPRESSED_SECTOR_SIZE); | |
45aba42f | 778 | |
3a75a870 AG |
779 | /* The offset and size must fit in their fields of the L2 table entry */ |
780 | assert((cluster_offset & s->cluster_offset_mask) == cluster_offset); | |
781 | assert((nb_csectors & s->csize_mask) == nb_csectors); | |
782 | ||
45aba42f KW |
783 | cluster_offset |= QCOW_OFLAG_COMPRESSED | |
784 | ((uint64_t)nb_csectors << s->csize_shift); | |
785 | ||
786 | /* update L2 table */ | |
787 | ||
788 | /* compressed clusters never have the copied flag */ | |
789 | ||
66f82cee | 790 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
e4e72548 AG |
791 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
792 | l2_slice[l2_index] = cpu_to_be64(cluster_offset); | |
793 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
4c1612d9 | 794 | |
77e023ff KW |
795 | *host_offset = cluster_offset & s->cluster_offset_mask; |
796 | return 0; | |
4c1612d9 KW |
797 | } |
798 | ||
99450c6f | 799 | static int perform_cow(BlockDriverState *bs, QCowL2Meta *m) |
593fb83c | 800 | { |
ff99129a | 801 | BDRVQcow2State *s = bs->opaque; |
99450c6f AG |
802 | Qcow2COWRegion *start = &m->cow_start; |
803 | Qcow2COWRegion *end = &m->cow_end; | |
672f0f2c | 804 | unsigned buffer_size; |
b3cf1c7c AG |
805 | unsigned data_bytes = end->offset - (start->offset + start->nb_bytes); |
806 | bool merge_reads; | |
672f0f2c | 807 | uint8_t *start_buffer, *end_buffer; |
86b862c4 | 808 | QEMUIOVector qiov; |
593fb83c KW |
809 | int ret; |
810 | ||
672f0f2c | 811 | assert(start->nb_bytes <= UINT_MAX - end->nb_bytes); |
b3cf1c7c AG |
812 | assert(start->nb_bytes + end->nb_bytes <= UINT_MAX - data_bytes); |
813 | assert(start->offset + start->nb_bytes <= end->offset); | |
672f0f2c | 814 | |
c8bb23cb | 815 | if ((start->nb_bytes == 0 && end->nb_bytes == 0) || m->skip_cow) { |
593fb83c KW |
816 | return 0; |
817 | } | |
818 | ||
b3cf1c7c AG |
819 | /* If we have to read both the start and end COW regions and the |
820 | * middle region is not too large then perform just one read | |
821 | * operation */ | |
822 | merge_reads = start->nb_bytes && end->nb_bytes && data_bytes <= 16384; | |
823 | if (merge_reads) { | |
824 | buffer_size = start->nb_bytes + data_bytes + end->nb_bytes; | |
825 | } else { | |
826 | /* If we have to do two reads, add some padding in the middle | |
827 | * if necessary to make sure that the end region is optimally | |
828 | * aligned. */ | |
829 | size_t align = bdrv_opt_mem_align(bs); | |
830 | assert(align > 0 && align <= UINT_MAX); | |
831 | assert(QEMU_ALIGN_UP(start->nb_bytes, align) <= | |
832 | UINT_MAX - end->nb_bytes); | |
833 | buffer_size = QEMU_ALIGN_UP(start->nb_bytes, align) + end->nb_bytes; | |
834 | } | |
835 | ||
836 | /* Reserve a buffer large enough to store all the data that we're | |
837 | * going to read */ | |
672f0f2c AG |
838 | start_buffer = qemu_try_blockalign(bs, buffer_size); |
839 | if (start_buffer == NULL) { | |
840 | return -ENOMEM; | |
841 | } | |
842 | /* The part of the buffer where the end region is located */ | |
843 | end_buffer = start_buffer + buffer_size - end->nb_bytes; | |
844 | ||
5396234b VSO |
845 | qemu_iovec_init(&qiov, 2 + (m->data_qiov ? |
846 | qemu_iovec_subvec_niov(m->data_qiov, | |
847 | m->data_qiov_offset, | |
848 | data_bytes) | |
849 | : 0)); | |
86b862c4 | 850 | |
593fb83c | 851 | qemu_co_mutex_unlock(&s->lock); |
b3cf1c7c AG |
852 | /* First we read the existing data from both COW regions. We |
853 | * either read the whole region in one go, or the start and end | |
854 | * regions separately. */ | |
855 | if (merge_reads) { | |
86b862c4 AG |
856 | qemu_iovec_add(&qiov, start_buffer, buffer_size); |
857 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c | 858 | } else { |
86b862c4 AG |
859 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); |
860 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c AG |
861 | if (ret < 0) { |
862 | goto fail; | |
863 | } | |
672f0f2c | 864 | |
86b862c4 AG |
865 | qemu_iovec_reset(&qiov); |
866 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
867 | ret = do_perform_cow_read(bs, m->offset, end->offset, &qiov); | |
b3cf1c7c | 868 | } |
593fb83c | 869 | if (ret < 0) { |
99450c6f | 870 | goto fail; |
593fb83c KW |
871 | } |
872 | ||
672f0f2c AG |
873 | /* Encrypt the data if necessary before writing it */ |
874 | if (bs->encrypted) { | |
603fbd07 ML |
875 | ret = qcow2_co_encrypt(bs, |
876 | m->alloc_offset + start->offset, | |
877 | m->offset + start->offset, | |
878 | start_buffer, start->nb_bytes); | |
879 | if (ret < 0) { | |
880 | goto fail; | |
881 | } | |
882 | ||
883 | ret = qcow2_co_encrypt(bs, | |
884 | m->alloc_offset + end->offset, | |
885 | m->offset + end->offset, | |
886 | end_buffer, end->nb_bytes); | |
887 | if (ret < 0) { | |
672f0f2c AG |
888 | goto fail; |
889 | } | |
890 | } | |
891 | ||
ee22a9d8 AG |
892 | /* And now we can write everything. If we have the guest data we |
893 | * can write everything in one single operation */ | |
894 | if (m->data_qiov) { | |
895 | qemu_iovec_reset(&qiov); | |
896 | if (start->nb_bytes) { | |
897 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
898 | } | |
5396234b | 899 | qemu_iovec_concat(&qiov, m->data_qiov, m->data_qiov_offset, data_bytes); |
ee22a9d8 AG |
900 | if (end->nb_bytes) { |
901 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
902 | } | |
903 | /* NOTE: we have a write_aio blkdebug event here followed by | |
904 | * a cow_write one in do_perform_cow_write(), but there's only | |
905 | * one single I/O operation */ | |
906 | BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); | |
907 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); | |
908 | } else { | |
909 | /* If there's no guest data then write both COW regions separately */ | |
910 | qemu_iovec_reset(&qiov); | |
911 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
912 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); | |
913 | if (ret < 0) { | |
914 | goto fail; | |
915 | } | |
916 | ||
917 | qemu_iovec_reset(&qiov); | |
918 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
919 | ret = do_perform_cow_write(bs, m->alloc_offset, end->offset, &qiov); | |
672f0f2c | 920 | } |
99450c6f AG |
921 | |
922 | fail: | |
923 | qemu_co_mutex_lock(&s->lock); | |
924 | ||
593fb83c KW |
925 | /* |
926 | * Before we update the L2 table to actually point to the new cluster, we | |
927 | * need to be sure that the refcounts have been increased and COW was | |
928 | * handled. | |
929 | */ | |
99450c6f AG |
930 | if (ret == 0) { |
931 | qcow2_cache_depends_on_flush(s->l2_table_cache); | |
932 | } | |
593fb83c | 933 | |
672f0f2c | 934 | qemu_vfree(start_buffer); |
86b862c4 | 935 | qemu_iovec_destroy(&qiov); |
99450c6f | 936 | return ret; |
593fb83c KW |
937 | } |
938 | ||
148da7ea | 939 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) |
45aba42f | 940 | { |
ff99129a | 941 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 942 | int i, j = 0, l2_index, ret; |
a002c0b0 | 943 | uint64_t *old_cluster, *l2_slice; |
250196f1 | 944 | uint64_t cluster_offset = m->alloc_offset; |
45aba42f | 945 | |
3cce16f4 | 946 | trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
f50f88b9 | 947 | assert(m->nb_clusters > 0); |
45aba42f | 948 | |
5839e53b | 949 | old_cluster = g_try_new(uint64_t, m->nb_clusters); |
de82815d KW |
950 | if (old_cluster == NULL) { |
951 | ret = -ENOMEM; | |
952 | goto err; | |
953 | } | |
45aba42f KW |
954 | |
955 | /* copy content of unmodified sectors */ | |
99450c6f | 956 | ret = perform_cow(bs, m); |
593fb83c KW |
957 | if (ret < 0) { |
958 | goto err; | |
29c1a730 KW |
959 | } |
960 | ||
593fb83c | 961 | /* Update L2 table. */ |
74c4510a | 962 | if (s->use_lazy_refcounts) { |
280d3735 KW |
963 | qcow2_mark_dirty(bs); |
964 | } | |
bfe8043e SH |
965 | if (qcow2_need_accurate_refcounts(s)) { |
966 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
967 | s->refcount_block_cache); | |
968 | } | |
280d3735 | 969 | |
a002c0b0 | 970 | ret = get_cluster_table(bs, m->offset, &l2_slice, &l2_index); |
1e3e8f1a | 971 | if (ret < 0) { |
45aba42f | 972 | goto err; |
1e3e8f1a | 973 | } |
a002c0b0 | 974 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
45aba42f | 975 | |
a002c0b0 | 976 | assert(l2_index + m->nb_clusters <= s->l2_slice_size); |
45aba42f | 977 | for (i = 0; i < m->nb_clusters; i++) { |
3a75a870 | 978 | uint64_t offset = cluster_offset + (i << s->cluster_bits); |
45aba42f | 979 | /* if two concurrent writes happen to the same unallocated cluster |
aaa4d20b KW |
980 | * each write allocates separate cluster and writes data concurrently. |
981 | * The first one to complete updates l2 table with pointer to its | |
982 | * cluster the second one has to do RMW (which is done above by | |
983 | * perform_cow()), update l2 table with its cluster pointer and free | |
984 | * old cluster. This is what this loop does */ | |
a002c0b0 AG |
985 | if (l2_slice[l2_index + i] != 0) { |
986 | old_cluster[j++] = l2_slice[l2_index + i]; | |
aaa4d20b | 987 | } |
45aba42f | 988 | |
3a75a870 AG |
989 | /* The offset must fit in the offset field of the L2 table entry */ |
990 | assert((offset & L2E_OFFSET_MASK) == offset); | |
991 | ||
992 | l2_slice[l2_index + i] = cpu_to_be64(offset | QCOW_OFLAG_COPIED); | |
45aba42f KW |
993 | } |
994 | ||
9f8e668e | 995 | |
a002c0b0 | 996 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
45aba42f | 997 | |
7ec5e6a4 KW |
998 | /* |
999 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
6cfcb9b8 KW |
1000 | * |
1001 | * Don't discard clusters that reach a refcount of 0 (e.g. compressed | |
1002 | * clusters), the next write will reuse them anyway. | |
7ec5e6a4 | 1003 | */ |
564a6b69 | 1004 | if (!m->keep_old_clusters && j != 0) { |
7ec5e6a4 | 1005 | for (i = 0; i < j; i++) { |
6cfcb9b8 KW |
1006 | qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1, |
1007 | QCOW2_DISCARD_NEVER); | |
7ec5e6a4 KW |
1008 | } |
1009 | } | |
45aba42f KW |
1010 | |
1011 | ret = 0; | |
1012 | err: | |
7267c094 | 1013 | g_free(old_cluster); |
45aba42f KW |
1014 | return ret; |
1015 | } | |
1016 | ||
8b24cd14 KW |
1017 | /** |
1018 | * Frees the allocated clusters because the request failed and they won't | |
1019 | * actually be linked. | |
1020 | */ | |
1021 | void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m) | |
1022 | { | |
1023 | BDRVQcow2State *s = bs->opaque; | |
1024 | qcow2_free_clusters(bs, m->alloc_offset, m->nb_clusters << s->cluster_bits, | |
1025 | QCOW2_DISCARD_NEVER); | |
1026 | } | |
1027 | ||
bf319ece KW |
1028 | /* |
1029 | * Returns the number of contiguous clusters that can be used for an allocating | |
1030 | * write, but require COW to be performed (this includes yet unallocated space, | |
1031 | * which must copy from the backing file) | |
1032 | */ | |
808c2bb4 | 1033 | static int count_cow_clusters(BlockDriverState *bs, int nb_clusters, |
dd32c881 | 1034 | uint64_t *l2_slice, int l2_index) |
bf319ece | 1035 | { |
143550a8 | 1036 | int i; |
bf319ece | 1037 | |
143550a8 | 1038 | for (i = 0; i < nb_clusters; i++) { |
dd32c881 | 1039 | uint64_t l2_entry = be64_to_cpu(l2_slice[l2_index + i]); |
808c2bb4 | 1040 | QCow2ClusterType cluster_type = qcow2_get_cluster_type(bs, l2_entry); |
143550a8 KW |
1041 | |
1042 | switch(cluster_type) { | |
1043 | case QCOW2_CLUSTER_NORMAL: | |
1044 | if (l2_entry & QCOW_OFLAG_COPIED) { | |
1045 | goto out; | |
1046 | } | |
bf319ece | 1047 | break; |
143550a8 KW |
1048 | case QCOW2_CLUSTER_UNALLOCATED: |
1049 | case QCOW2_CLUSTER_COMPRESSED: | |
fdfab37d EB |
1050 | case QCOW2_CLUSTER_ZERO_PLAIN: |
1051 | case QCOW2_CLUSTER_ZERO_ALLOC: | |
bf319ece | 1052 | break; |
143550a8 KW |
1053 | default: |
1054 | abort(); | |
1055 | } | |
bf319ece KW |
1056 | } |
1057 | ||
143550a8 | 1058 | out: |
bf319ece KW |
1059 | assert(i <= nb_clusters); |
1060 | return i; | |
1061 | } | |
1062 | ||
250196f1 | 1063 | /* |
226c3c26 KW |
1064 | * Check if there already is an AIO write request in flight which allocates |
1065 | * the same cluster. In this case we need to wait until the previous | |
1066 | * request has completed and updated the L2 table accordingly. | |
65eb2e35 KW |
1067 | * |
1068 | * Returns: | |
1069 | * 0 if there was no dependency. *cur_bytes indicates the number of | |
1070 | * bytes from guest_offset that can be read before the next | |
1071 | * dependency must be processed (or the request is complete) | |
1072 | * | |
1073 | * -EAGAIN if we had to wait for another request, previously gathered | |
1074 | * information on cluster allocation may be invalid now. The caller | |
1075 | * must start over anyway, so consider *cur_bytes undefined. | |
250196f1 | 1076 | */ |
226c3c26 | 1077 | static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, |
ecdd5333 | 1078 | uint64_t *cur_bytes, QCowL2Meta **m) |
250196f1 | 1079 | { |
ff99129a | 1080 | BDRVQcow2State *s = bs->opaque; |
250196f1 | 1081 | QCowL2Meta *old_alloc; |
65eb2e35 | 1082 | uint64_t bytes = *cur_bytes; |
250196f1 | 1083 | |
250196f1 KW |
1084 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
1085 | ||
65eb2e35 KW |
1086 | uint64_t start = guest_offset; |
1087 | uint64_t end = start + bytes; | |
1088 | uint64_t old_start = l2meta_cow_start(old_alloc); | |
1089 | uint64_t old_end = l2meta_cow_end(old_alloc); | |
250196f1 | 1090 | |
d9d74f41 | 1091 | if (end <= old_start || start >= old_end) { |
250196f1 KW |
1092 | /* No intersection */ |
1093 | } else { | |
1094 | if (start < old_start) { | |
1095 | /* Stop at the start of a running allocation */ | |
65eb2e35 | 1096 | bytes = old_start - start; |
250196f1 | 1097 | } else { |
65eb2e35 | 1098 | bytes = 0; |
250196f1 KW |
1099 | } |
1100 | ||
ecdd5333 KW |
1101 | /* Stop if already an l2meta exists. After yielding, it wouldn't |
1102 | * be valid any more, so we'd have to clean up the old L2Metas | |
1103 | * and deal with requests depending on them before starting to | |
1104 | * gather new ones. Not worth the trouble. */ | |
1105 | if (bytes == 0 && *m) { | |
1106 | *cur_bytes = 0; | |
1107 | return 0; | |
1108 | } | |
1109 | ||
65eb2e35 | 1110 | if (bytes == 0) { |
250196f1 KW |
1111 | /* Wait for the dependency to complete. We need to recheck |
1112 | * the free/allocated clusters when we continue. */ | |
1ace7cea | 1113 | qemu_co_queue_wait(&old_alloc->dependent_requests, &s->lock); |
250196f1 KW |
1114 | return -EAGAIN; |
1115 | } | |
1116 | } | |
1117 | } | |
1118 | ||
65eb2e35 KW |
1119 | /* Make sure that existing clusters and new allocations are only used up to |
1120 | * the next dependency if we shortened the request above */ | |
1121 | *cur_bytes = bytes; | |
250196f1 | 1122 | |
226c3c26 KW |
1123 | return 0; |
1124 | } | |
1125 | ||
0af729ec KW |
1126 | /* |
1127 | * Checks how many already allocated clusters that don't require a copy on | |
c6d619cc KW |
1128 | * write there are at the given guest_offset (up to *bytes). If *host_offset is |
1129 | * not INV_OFFSET, only physically contiguous clusters beginning at this host | |
1130 | * offset are counted. | |
0af729ec | 1131 | * |
411d62b0 KW |
1132 | * Note that guest_offset may not be cluster aligned. In this case, the |
1133 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1134 | * therefore isn't cluster aligned as well. | |
0af729ec KW |
1135 | * |
1136 | * Returns: | |
1137 | * 0: if no allocated clusters are available at the given offset. | |
1138 | * *bytes is normally unchanged. It is set to 0 if the cluster | |
1139 | * is allocated and doesn't need COW, but doesn't have the right | |
1140 | * physical offset. | |
1141 | * | |
1142 | * 1: if allocated clusters that don't require a COW are available at | |
1143 | * the requested offset. *bytes may have decreased and describes | |
1144 | * the length of the area that can be written to. | |
1145 | * | |
1146 | * -errno: in error cases | |
0af729ec KW |
1147 | */ |
1148 | static int handle_copied(BlockDriverState *bs, uint64_t guest_offset, | |
c53ede9f | 1149 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
0af729ec | 1150 | { |
ff99129a | 1151 | BDRVQcow2State *s = bs->opaque; |
0af729ec KW |
1152 | int l2_index; |
1153 | uint64_t cluster_offset; | |
cde91766 | 1154 | uint64_t *l2_slice; |
b6d36def | 1155 | uint64_t nb_clusters; |
c53ede9f | 1156 | unsigned int keep_clusters; |
a3f1afb4 | 1157 | int ret; |
0af729ec KW |
1158 | |
1159 | trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset, | |
1160 | *bytes); | |
0af729ec | 1161 | |
c6d619cc KW |
1162 | assert(*host_offset == INV_OFFSET || offset_into_cluster(s, guest_offset) |
1163 | == offset_into_cluster(s, *host_offset)); | |
411d62b0 | 1164 | |
acb0467f | 1165 | /* |
cde91766 | 1166 | * Calculate the number of clusters to look for. We stop at L2 slice |
acb0467f KW |
1167 | * boundaries to keep things simple. |
1168 | */ | |
1169 | nb_clusters = | |
1170 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1171 | ||
cde91766 AG |
1172 | l2_index = offset_to_l2_slice_index(s, guest_offset); |
1173 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
b6d36def | 1174 | assert(nb_clusters <= INT_MAX); |
acb0467f | 1175 | |
0af729ec | 1176 | /* Find L2 entry for the first involved cluster */ |
cde91766 | 1177 | ret = get_cluster_table(bs, guest_offset, &l2_slice, &l2_index); |
0af729ec KW |
1178 | if (ret < 0) { |
1179 | return ret; | |
1180 | } | |
1181 | ||
cde91766 | 1182 | cluster_offset = be64_to_cpu(l2_slice[l2_index]); |
0af729ec KW |
1183 | |
1184 | /* Check how many clusters are already allocated and don't need COW */ | |
808c2bb4 | 1185 | if (qcow2_get_cluster_type(bs, cluster_offset) == QCOW2_CLUSTER_NORMAL |
0af729ec KW |
1186 | && (cluster_offset & QCOW_OFLAG_COPIED)) |
1187 | { | |
e62daaf6 KW |
1188 | /* If a specific host_offset is required, check it */ |
1189 | bool offset_matches = | |
1190 | (cluster_offset & L2E_OFFSET_MASK) == *host_offset; | |
1191 | ||
a97c67ee HR |
1192 | if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) { |
1193 | qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset " | |
1194 | "%#llx unaligned (guest offset: %#" PRIx64 | |
1195 | ")", cluster_offset & L2E_OFFSET_MASK, | |
1196 | guest_offset); | |
1197 | ret = -EIO; | |
1198 | goto out; | |
1199 | } | |
1200 | ||
c6d619cc | 1201 | if (*host_offset != INV_OFFSET && !offset_matches) { |
e62daaf6 KW |
1202 | *bytes = 0; |
1203 | ret = 0; | |
1204 | goto out; | |
1205 | } | |
1206 | ||
0af729ec | 1207 | /* We keep all QCOW_OFLAG_COPIED clusters */ |
c53ede9f | 1208 | keep_clusters = |
808c2bb4 | 1209 | count_contiguous_clusters(bs, nb_clusters, s->cluster_size, |
cde91766 | 1210 | &l2_slice[l2_index], |
0af729ec | 1211 | QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); |
c53ede9f KW |
1212 | assert(keep_clusters <= nb_clusters); |
1213 | ||
1214 | *bytes = MIN(*bytes, | |
1215 | keep_clusters * s->cluster_size | |
1216 | - offset_into_cluster(s, guest_offset)); | |
0af729ec KW |
1217 | |
1218 | ret = 1; | |
1219 | } else { | |
0af729ec KW |
1220 | ret = 0; |
1221 | } | |
1222 | ||
0af729ec | 1223 | /* Cleanup */ |
e62daaf6 | 1224 | out: |
cde91766 | 1225 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
0af729ec | 1226 | |
e62daaf6 KW |
1227 | /* Only return a host offset if we actually made progress. Otherwise we |
1228 | * would make requirements for handle_alloc() that it can't fulfill */ | |
a97c67ee | 1229 | if (ret > 0) { |
411d62b0 KW |
1230 | *host_offset = (cluster_offset & L2E_OFFSET_MASK) |
1231 | + offset_into_cluster(s, guest_offset); | |
e62daaf6 KW |
1232 | } |
1233 | ||
0af729ec KW |
1234 | return ret; |
1235 | } | |
1236 | ||
226c3c26 KW |
1237 | /* |
1238 | * Allocates new clusters for the given guest_offset. | |
1239 | * | |
1240 | * At most *nb_clusters are allocated, and on return *nb_clusters is updated to | |
1241 | * contain the number of clusters that have been allocated and are contiguous | |
1242 | * in the image file. | |
1243 | * | |
c6d619cc KW |
1244 | * If *host_offset is not INV_OFFSET, it specifies the offset in the image file |
1245 | * at which the new clusters must start. *nb_clusters can be 0 on return in | |
1246 | * this case if the cluster at host_offset is already in use. If *host_offset | |
1247 | * is INV_OFFSET, the clusters can be allocated anywhere in the image file. | |
226c3c26 KW |
1248 | * |
1249 | * *host_offset is updated to contain the offset into the image file at which | |
1250 | * the first allocated cluster starts. | |
1251 | * | |
1252 | * Return 0 on success and -errno in error cases. -EAGAIN means that the | |
1253 | * function has been waiting for another request and the allocation must be | |
1254 | * restarted, but the whole request should not be failed. | |
1255 | */ | |
1256 | static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, | |
b6d36def | 1257 | uint64_t *host_offset, uint64_t *nb_clusters) |
226c3c26 | 1258 | { |
ff99129a | 1259 | BDRVQcow2State *s = bs->opaque; |
226c3c26 KW |
1260 | |
1261 | trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, | |
1262 | *host_offset, *nb_clusters); | |
1263 | ||
966b000f KW |
1264 | if (has_data_file(bs)) { |
1265 | assert(*host_offset == INV_OFFSET || | |
1266 | *host_offset == start_of_cluster(s, guest_offset)); | |
1267 | *host_offset = start_of_cluster(s, guest_offset); | |
1268 | return 0; | |
1269 | } | |
1270 | ||
250196f1 KW |
1271 | /* Allocate new clusters */ |
1272 | trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); | |
c6d619cc | 1273 | if (*host_offset == INV_OFFSET) { |
df021791 KW |
1274 | int64_t cluster_offset = |
1275 | qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); | |
1276 | if (cluster_offset < 0) { | |
1277 | return cluster_offset; | |
1278 | } | |
1279 | *host_offset = cluster_offset; | |
1280 | return 0; | |
250196f1 | 1281 | } else { |
b6d36def | 1282 | int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); |
df021791 KW |
1283 | if (ret < 0) { |
1284 | return ret; | |
1285 | } | |
1286 | *nb_clusters = ret; | |
1287 | return 0; | |
250196f1 | 1288 | } |
250196f1 KW |
1289 | } |
1290 | ||
10f0ed8b KW |
1291 | /* |
1292 | * Allocates new clusters for an area that either is yet unallocated or needs a | |
c6d619cc KW |
1293 | * copy on write. If *host_offset is not INV_OFFSET, clusters are only |
1294 | * allocated if the new allocation can match the specified host offset. | |
10f0ed8b | 1295 | * |
411d62b0 KW |
1296 | * Note that guest_offset may not be cluster aligned. In this case, the |
1297 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1298 | * therefore isn't cluster aligned as well. | |
10f0ed8b KW |
1299 | * |
1300 | * Returns: | |
1301 | * 0: if no clusters could be allocated. *bytes is set to 0, | |
1302 | * *host_offset is left unchanged. | |
1303 | * | |
1304 | * 1: if new clusters were allocated. *bytes may be decreased if the | |
1305 | * new allocation doesn't cover all of the requested area. | |
1306 | * *host_offset is updated to contain the host offset of the first | |
1307 | * newly allocated cluster. | |
1308 | * | |
1309 | * -errno: in error cases | |
10f0ed8b KW |
1310 | */ |
1311 | static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, | |
c37f4cd7 | 1312 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
10f0ed8b | 1313 | { |
ff99129a | 1314 | BDRVQcow2State *s = bs->opaque; |
10f0ed8b | 1315 | int l2_index; |
6d99a344 | 1316 | uint64_t *l2_slice; |
10f0ed8b | 1317 | uint64_t entry; |
b6d36def | 1318 | uint64_t nb_clusters; |
10f0ed8b | 1319 | int ret; |
564a6b69 | 1320 | bool keep_old_clusters = false; |
10f0ed8b | 1321 | |
c6d619cc | 1322 | uint64_t alloc_cluster_offset = INV_OFFSET; |
10f0ed8b KW |
1323 | |
1324 | trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset, | |
1325 | *bytes); | |
1326 | assert(*bytes > 0); | |
1327 | ||
f5bc6350 | 1328 | /* |
6d99a344 | 1329 | * Calculate the number of clusters to look for. We stop at L2 slice |
f5bc6350 KW |
1330 | * boundaries to keep things simple. |
1331 | */ | |
c37f4cd7 KW |
1332 | nb_clusters = |
1333 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1334 | ||
6d99a344 AG |
1335 | l2_index = offset_to_l2_slice_index(s, guest_offset); |
1336 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
b6d36def | 1337 | assert(nb_clusters <= INT_MAX); |
f5bc6350 | 1338 | |
d1b9d19f HR |
1339 | /* Limit total allocation byte count to INT_MAX */ |
1340 | nb_clusters = MIN(nb_clusters, INT_MAX >> s->cluster_bits); | |
1341 | ||
10f0ed8b | 1342 | /* Find L2 entry for the first involved cluster */ |
6d99a344 | 1343 | ret = get_cluster_table(bs, guest_offset, &l2_slice, &l2_index); |
10f0ed8b KW |
1344 | if (ret < 0) { |
1345 | return ret; | |
1346 | } | |
1347 | ||
6d99a344 | 1348 | entry = be64_to_cpu(l2_slice[l2_index]); |
bf3d78ae | 1349 | nb_clusters = count_cow_clusters(bs, nb_clusters, l2_slice, l2_index); |
10f0ed8b | 1350 | |
ecdd5333 KW |
1351 | /* This function is only called when there were no non-COW clusters, so if |
1352 | * we can't find any unallocated or COW clusters either, something is | |
1353 | * wrong with our code. */ | |
1354 | assert(nb_clusters > 0); | |
1355 | ||
808c2bb4 | 1356 | if (qcow2_get_cluster_type(bs, entry) == QCOW2_CLUSTER_ZERO_ALLOC && |
fdfab37d | 1357 | (entry & QCOW_OFLAG_COPIED) && |
c6d619cc | 1358 | (*host_offset == INV_OFFSET || |
564a6b69 HR |
1359 | start_of_cluster(s, *host_offset) == (entry & L2E_OFFSET_MASK))) |
1360 | { | |
93bbaf03 HR |
1361 | int preallocated_nb_clusters; |
1362 | ||
1363 | if (offset_into_cluster(s, entry & L2E_OFFSET_MASK)) { | |
1364 | qcow2_signal_corruption(bs, true, -1, -1, "Preallocated zero " | |
1365 | "cluster offset %#llx unaligned (guest " | |
1366 | "offset: %#" PRIx64 ")", | |
1367 | entry & L2E_OFFSET_MASK, guest_offset); | |
1368 | ret = -EIO; | |
1369 | goto fail; | |
1370 | } | |
1371 | ||
564a6b69 HR |
1372 | /* Try to reuse preallocated zero clusters; contiguous normal clusters |
1373 | * would be fine, too, but count_cow_clusters() above has limited | |
1374 | * nb_clusters already to a range of COW clusters */ | |
93bbaf03 | 1375 | preallocated_nb_clusters = |
808c2bb4 | 1376 | count_contiguous_clusters(bs, nb_clusters, s->cluster_size, |
6d99a344 | 1377 | &l2_slice[l2_index], QCOW_OFLAG_COPIED); |
564a6b69 | 1378 | assert(preallocated_nb_clusters > 0); |
10f0ed8b | 1379 | |
564a6b69 HR |
1380 | nb_clusters = preallocated_nb_clusters; |
1381 | alloc_cluster_offset = entry & L2E_OFFSET_MASK; | |
10f0ed8b | 1382 | |
564a6b69 HR |
1383 | /* We want to reuse these clusters, so qcow2_alloc_cluster_link_l2() |
1384 | * should not free them. */ | |
1385 | keep_old_clusters = true; | |
10f0ed8b KW |
1386 | } |
1387 | ||
6d99a344 | 1388 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
564a6b69 | 1389 | |
c6d619cc | 1390 | if (alloc_cluster_offset == INV_OFFSET) { |
564a6b69 | 1391 | /* Allocate, if necessary at a given offset in the image file */ |
c6d619cc KW |
1392 | alloc_cluster_offset = *host_offset == INV_OFFSET ? INV_OFFSET : |
1393 | start_of_cluster(s, *host_offset); | |
564a6b69 HR |
1394 | ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset, |
1395 | &nb_clusters); | |
1396 | if (ret < 0) { | |
1397 | goto fail; | |
1398 | } | |
1399 | ||
1400 | /* Can't extend contiguous allocation */ | |
1401 | if (nb_clusters == 0) { | |
1402 | *bytes = 0; | |
1403 | return 0; | |
1404 | } | |
1405 | ||
c6d619cc | 1406 | assert(alloc_cluster_offset != INV_OFFSET); |
ff52aab2 HR |
1407 | } |
1408 | ||
83baa9a4 KW |
1409 | /* |
1410 | * Save info needed for meta data update. | |
1411 | * | |
85567393 | 1412 | * requested_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1413 | * newly allocated cluster to the end of the (possibly shortened |
1414 | * before) write request. | |
1415 | * | |
85567393 | 1416 | * avail_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1417 | * newly allocated to the end of the last newly allocated cluster. |
1418 | * | |
85567393 | 1419 | * nb_bytes: The number of bytes from the start of the first |
83baa9a4 KW |
1420 | * newly allocated cluster to the end of the area that the write |
1421 | * request actually writes to (excluding COW at the end) | |
1422 | */ | |
85567393 | 1423 | uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset); |
d1b9d19f | 1424 | int avail_bytes = nb_clusters << s->cluster_bits; |
85567393 | 1425 | int nb_bytes = MIN(requested_bytes, avail_bytes); |
88c6588c | 1426 | QCowL2Meta *old_m = *m; |
83baa9a4 | 1427 | |
83baa9a4 KW |
1428 | *m = g_malloc0(sizeof(**m)); |
1429 | ||
1430 | **m = (QCowL2Meta) { | |
88c6588c KW |
1431 | .next = old_m, |
1432 | ||
411d62b0 | 1433 | .alloc_offset = alloc_cluster_offset, |
83baa9a4 KW |
1434 | .offset = start_of_cluster(s, guest_offset), |
1435 | .nb_clusters = nb_clusters, | |
83baa9a4 | 1436 | |
564a6b69 HR |
1437 | .keep_old_clusters = keep_old_clusters, |
1438 | ||
83baa9a4 KW |
1439 | .cow_start = { |
1440 | .offset = 0, | |
85567393 | 1441 | .nb_bytes = offset_into_cluster(s, guest_offset), |
83baa9a4 KW |
1442 | }, |
1443 | .cow_end = { | |
85567393 KW |
1444 | .offset = nb_bytes, |
1445 | .nb_bytes = avail_bytes - nb_bytes, | |
83baa9a4 KW |
1446 | }, |
1447 | }; | |
1448 | qemu_co_queue_init(&(*m)->dependent_requests); | |
1449 | QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); | |
1450 | ||
411d62b0 | 1451 | *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset); |
85567393 | 1452 | *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset)); |
83baa9a4 KW |
1453 | assert(*bytes != 0); |
1454 | ||
10f0ed8b KW |
1455 | return 1; |
1456 | ||
1457 | fail: | |
1458 | if (*m && (*m)->nb_clusters > 0) { | |
1459 | QLIST_REMOVE(*m, next_in_flight); | |
1460 | } | |
1461 | return ret; | |
1462 | } | |
1463 | ||
45aba42f KW |
1464 | /* |
1465 | * alloc_cluster_offset | |
1466 | * | |
250196f1 KW |
1467 | * For a given offset on the virtual disk, find the cluster offset in qcow2 |
1468 | * file. If the offset is not found, allocate a new cluster. | |
45aba42f | 1469 | * |
250196f1 | 1470 | * If the cluster was already allocated, m->nb_clusters is set to 0 and |
a7912369 | 1471 | * other fields in m are meaningless. |
148da7ea KW |
1472 | * |
1473 | * If the cluster is newly allocated, m->nb_clusters is set to the number of | |
68d100e9 KW |
1474 | * contiguous clusters that have been allocated. In this case, the other |
1475 | * fields of m are valid and contain information about the first allocated | |
1476 | * cluster. | |
45aba42f | 1477 | * |
68d100e9 KW |
1478 | * If the request conflicts with another write request in flight, the coroutine |
1479 | * is queued and will be reentered when the dependency has completed. | |
148da7ea KW |
1480 | * |
1481 | * Return 0 on success and -errno in error cases | |
45aba42f | 1482 | */ |
f4f0d391 | 1483 | int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, |
d46a0bb2 KW |
1484 | unsigned int *bytes, uint64_t *host_offset, |
1485 | QCowL2Meta **m) | |
45aba42f | 1486 | { |
ff99129a | 1487 | BDRVQcow2State *s = bs->opaque; |
710c2496 | 1488 | uint64_t start, remaining; |
250196f1 | 1489 | uint64_t cluster_offset; |
65eb2e35 | 1490 | uint64_t cur_bytes; |
710c2496 | 1491 | int ret; |
45aba42f | 1492 | |
d46a0bb2 | 1493 | trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *bytes); |
710c2496 | 1494 | |
72424114 | 1495 | again: |
16f0587e | 1496 | start = offset; |
d46a0bb2 | 1497 | remaining = *bytes; |
c6d619cc KW |
1498 | cluster_offset = INV_OFFSET; |
1499 | *host_offset = INV_OFFSET; | |
ecdd5333 KW |
1500 | cur_bytes = 0; |
1501 | *m = NULL; | |
0af729ec | 1502 | |
2c3b32d2 | 1503 | while (true) { |
ecdd5333 | 1504 | |
c6d619cc | 1505 | if (*host_offset == INV_OFFSET && cluster_offset != INV_OFFSET) { |
ecdd5333 KW |
1506 | *host_offset = start_of_cluster(s, cluster_offset); |
1507 | } | |
1508 | ||
1509 | assert(remaining >= cur_bytes); | |
1510 | ||
1511 | start += cur_bytes; | |
1512 | remaining -= cur_bytes; | |
c6d619cc KW |
1513 | |
1514 | if (cluster_offset != INV_OFFSET) { | |
1515 | cluster_offset += cur_bytes; | |
1516 | } | |
ecdd5333 KW |
1517 | |
1518 | if (remaining == 0) { | |
1519 | break; | |
1520 | } | |
1521 | ||
1522 | cur_bytes = remaining; | |
1523 | ||
2c3b32d2 KW |
1524 | /* |
1525 | * Now start gathering as many contiguous clusters as possible: | |
1526 | * | |
1527 | * 1. Check for overlaps with in-flight allocations | |
1528 | * | |
1529 | * a) Overlap not in the first cluster -> shorten this request and | |
1530 | * let the caller handle the rest in its next loop iteration. | |
1531 | * | |
1532 | * b) Real overlaps of two requests. Yield and restart the search | |
1533 | * for contiguous clusters (the situation could have changed | |
1534 | * while we were sleeping) | |
1535 | * | |
1536 | * c) TODO: Request starts in the same cluster as the in-flight | |
1537 | * allocation ends. Shorten the COW of the in-fight allocation, | |
1538 | * set cluster_offset to write to the same cluster and set up | |
1539 | * the right synchronisation between the in-flight request and | |
1540 | * the new one. | |
1541 | */ | |
ecdd5333 | 1542 | ret = handle_dependencies(bs, start, &cur_bytes, m); |
2c3b32d2 | 1543 | if (ret == -EAGAIN) { |
ecdd5333 KW |
1544 | /* Currently handle_dependencies() doesn't yield if we already had |
1545 | * an allocation. If it did, we would have to clean up the L2Meta | |
1546 | * structs before starting over. */ | |
1547 | assert(*m == NULL); | |
2c3b32d2 KW |
1548 | goto again; |
1549 | } else if (ret < 0) { | |
1550 | return ret; | |
ecdd5333 KW |
1551 | } else if (cur_bytes == 0) { |
1552 | break; | |
2c3b32d2 KW |
1553 | } else { |
1554 | /* handle_dependencies() may have decreased cur_bytes (shortened | |
1555 | * the allocations below) so that the next dependency is processed | |
1556 | * correctly during the next loop iteration. */ | |
0af729ec | 1557 | } |
710c2496 | 1558 | |
2c3b32d2 KW |
1559 | /* |
1560 | * 2. Count contiguous COPIED clusters. | |
1561 | */ | |
1562 | ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m); | |
1563 | if (ret < 0) { | |
1564 | return ret; | |
1565 | } else if (ret) { | |
ecdd5333 | 1566 | continue; |
2c3b32d2 KW |
1567 | } else if (cur_bytes == 0) { |
1568 | break; | |
1569 | } | |
060bee89 | 1570 | |
2c3b32d2 KW |
1571 | /* |
1572 | * 3. If the request still hasn't completed, allocate new clusters, | |
1573 | * considering any cluster_offset of steps 1c or 2. | |
1574 | */ | |
1575 | ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m); | |
1576 | if (ret < 0) { | |
1577 | return ret; | |
1578 | } else if (ret) { | |
ecdd5333 | 1579 | continue; |
2c3b32d2 KW |
1580 | } else { |
1581 | assert(cur_bytes == 0); | |
1582 | break; | |
1583 | } | |
f5bc6350 | 1584 | } |
10f0ed8b | 1585 | |
d46a0bb2 KW |
1586 | *bytes -= remaining; |
1587 | assert(*bytes > 0); | |
c6d619cc | 1588 | assert(*host_offset != INV_OFFSET); |
45aba42f | 1589 | |
148da7ea | 1590 | return 0; |
45aba42f KW |
1591 | } |
1592 | ||
5ea929e3 KW |
1593 | /* |
1594 | * This discards as many clusters of nb_clusters as possible at once (i.e. | |
21ab3add | 1595 | * all clusters in the same L2 slice) and returns the number of discarded |
5ea929e3 KW |
1596 | * clusters. |
1597 | */ | |
21ab3add AG |
1598 | static int discard_in_l2_slice(BlockDriverState *bs, uint64_t offset, |
1599 | uint64_t nb_clusters, | |
1600 | enum qcow2_discard_type type, bool full_discard) | |
5ea929e3 | 1601 | { |
ff99129a | 1602 | BDRVQcow2State *s = bs->opaque; |
21ab3add | 1603 | uint64_t *l2_slice; |
5ea929e3 KW |
1604 | int l2_index; |
1605 | int ret; | |
1606 | int i; | |
1607 | ||
21ab3add | 1608 | ret = get_cluster_table(bs, offset, &l2_slice, &l2_index); |
5ea929e3 KW |
1609 | if (ret < 0) { |
1610 | return ret; | |
1611 | } | |
1612 | ||
21ab3add AG |
1613 | /* Limit nb_clusters to one L2 slice */ |
1614 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
b6d36def | 1615 | assert(nb_clusters <= INT_MAX); |
5ea929e3 KW |
1616 | |
1617 | for (i = 0; i < nb_clusters; i++) { | |
c883db0d | 1618 | uint64_t old_l2_entry; |
5ea929e3 | 1619 | |
21ab3add | 1620 | old_l2_entry = be64_to_cpu(l2_slice[l2_index + i]); |
a71835a0 KW |
1621 | |
1622 | /* | |
808c4b6f HR |
1623 | * If full_discard is false, make sure that a discarded area reads back |
1624 | * as zeroes for v3 images (we cannot do it for v2 without actually | |
1625 | * writing a zero-filled buffer). We can skip the operation if the | |
1626 | * cluster is already marked as zero, or if it's unallocated and we | |
1627 | * don't have a backing file. | |
a71835a0 | 1628 | * |
237d78f8 | 1629 | * TODO We might want to use bdrv_block_status(bs) here, but we're |
a71835a0 | 1630 | * holding s->lock, so that doesn't work today. |
808c4b6f HR |
1631 | * |
1632 | * If full_discard is true, the sector should not read back as zeroes, | |
1633 | * but rather fall through to the backing file. | |
a71835a0 | 1634 | */ |
808c2bb4 | 1635 | switch (qcow2_get_cluster_type(bs, old_l2_entry)) { |
bbd995d8 EB |
1636 | case QCOW2_CLUSTER_UNALLOCATED: |
1637 | if (full_discard || !bs->backing) { | |
1638 | continue; | |
1639 | } | |
1640 | break; | |
1641 | ||
fdfab37d EB |
1642 | case QCOW2_CLUSTER_ZERO_PLAIN: |
1643 | if (!full_discard) { | |
bbd995d8 EB |
1644 | continue; |
1645 | } | |
1646 | break; | |
1647 | ||
fdfab37d | 1648 | case QCOW2_CLUSTER_ZERO_ALLOC: |
bbd995d8 EB |
1649 | case QCOW2_CLUSTER_NORMAL: |
1650 | case QCOW2_CLUSTER_COMPRESSED: | |
1651 | break; | |
1652 | ||
1653 | default: | |
1654 | abort(); | |
5ea929e3 KW |
1655 | } |
1656 | ||
1657 | /* First remove L2 entries */ | |
21ab3add | 1658 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
808c4b6f | 1659 | if (!full_discard && s->qcow_version >= 3) { |
21ab3add | 1660 | l2_slice[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
a71835a0 | 1661 | } else { |
21ab3add | 1662 | l2_slice[l2_index + i] = cpu_to_be64(0); |
a71835a0 | 1663 | } |
5ea929e3 KW |
1664 | |
1665 | /* Then decrease the refcount */ | |
c883db0d | 1666 | qcow2_free_any_clusters(bs, old_l2_entry, 1, type); |
5ea929e3 KW |
1667 | } |
1668 | ||
21ab3add | 1669 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
5ea929e3 KW |
1670 | |
1671 | return nb_clusters; | |
1672 | } | |
1673 | ||
d2cb36af EB |
1674 | int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset, |
1675 | uint64_t bytes, enum qcow2_discard_type type, | |
1676 | bool full_discard) | |
5ea929e3 | 1677 | { |
ff99129a | 1678 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1679 | uint64_t end_offset = offset + bytes; |
b6d36def | 1680 | uint64_t nb_clusters; |
d2cb36af | 1681 | int64_t cleared; |
5ea929e3 KW |
1682 | int ret; |
1683 | ||
f10ee139 | 1684 | /* Caller must pass aligned values, except at image end */ |
0c1bd469 | 1685 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); |
f10ee139 EB |
1686 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || |
1687 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
5ea929e3 | 1688 | |
d2cb36af | 1689 | nb_clusters = size_to_clusters(s, bytes); |
5ea929e3 | 1690 | |
0b919fae KW |
1691 | s->cache_discards = true; |
1692 | ||
21ab3add | 1693 | /* Each L2 slice is handled by its own loop iteration */ |
5ea929e3 | 1694 | while (nb_clusters > 0) { |
21ab3add AG |
1695 | cleared = discard_in_l2_slice(bs, offset, nb_clusters, type, |
1696 | full_discard); | |
d2cb36af EB |
1697 | if (cleared < 0) { |
1698 | ret = cleared; | |
0b919fae | 1699 | goto fail; |
5ea929e3 KW |
1700 | } |
1701 | ||
d2cb36af EB |
1702 | nb_clusters -= cleared; |
1703 | offset += (cleared * s->cluster_size); | |
5ea929e3 KW |
1704 | } |
1705 | ||
0b919fae KW |
1706 | ret = 0; |
1707 | fail: | |
1708 | s->cache_discards = false; | |
1709 | qcow2_process_discards(bs, ret); | |
1710 | ||
1711 | return ret; | |
5ea929e3 | 1712 | } |
621f0589 KW |
1713 | |
1714 | /* | |
1715 | * This zeroes as many clusters of nb_clusters as possible at once (i.e. | |
a9a9f8f0 | 1716 | * all clusters in the same L2 slice) and returns the number of zeroed |
621f0589 KW |
1717 | * clusters. |
1718 | */ | |
a9a9f8f0 AG |
1719 | static int zero_in_l2_slice(BlockDriverState *bs, uint64_t offset, |
1720 | uint64_t nb_clusters, int flags) | |
621f0589 | 1721 | { |
ff99129a | 1722 | BDRVQcow2State *s = bs->opaque; |
a9a9f8f0 | 1723 | uint64_t *l2_slice; |
621f0589 KW |
1724 | int l2_index; |
1725 | int ret; | |
1726 | int i; | |
06cc5e2b | 1727 | bool unmap = !!(flags & BDRV_REQ_MAY_UNMAP); |
621f0589 | 1728 | |
a9a9f8f0 | 1729 | ret = get_cluster_table(bs, offset, &l2_slice, &l2_index); |
621f0589 KW |
1730 | if (ret < 0) { |
1731 | return ret; | |
1732 | } | |
1733 | ||
a9a9f8f0 AG |
1734 | /* Limit nb_clusters to one L2 slice */ |
1735 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
b6d36def | 1736 | assert(nb_clusters <= INT_MAX); |
621f0589 KW |
1737 | |
1738 | for (i = 0; i < nb_clusters; i++) { | |
1739 | uint64_t old_offset; | |
06cc5e2b | 1740 | QCow2ClusterType cluster_type; |
621f0589 | 1741 | |
a9a9f8f0 | 1742 | old_offset = be64_to_cpu(l2_slice[l2_index + i]); |
621f0589 | 1743 | |
06cc5e2b EB |
1744 | /* |
1745 | * Minimize L2 changes if the cluster already reads back as | |
1746 | * zeroes with correct allocation. | |
1747 | */ | |
808c2bb4 | 1748 | cluster_type = qcow2_get_cluster_type(bs, old_offset); |
06cc5e2b EB |
1749 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN || |
1750 | (cluster_type == QCOW2_CLUSTER_ZERO_ALLOC && !unmap)) { | |
1751 | continue; | |
1752 | } | |
1753 | ||
a9a9f8f0 | 1754 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
06cc5e2b | 1755 | if (cluster_type == QCOW2_CLUSTER_COMPRESSED || unmap) { |
a9a9f8f0 | 1756 | l2_slice[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
6cfcb9b8 | 1757 | qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST); |
621f0589 | 1758 | } else { |
a9a9f8f0 | 1759 | l2_slice[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO); |
621f0589 KW |
1760 | } |
1761 | } | |
1762 | ||
a9a9f8f0 | 1763 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
621f0589 KW |
1764 | |
1765 | return nb_clusters; | |
1766 | } | |
1767 | ||
d2cb36af EB |
1768 | int qcow2_cluster_zeroize(BlockDriverState *bs, uint64_t offset, |
1769 | uint64_t bytes, int flags) | |
621f0589 | 1770 | { |
ff99129a | 1771 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1772 | uint64_t end_offset = offset + bytes; |
b6d36def | 1773 | uint64_t nb_clusters; |
d2cb36af | 1774 | int64_t cleared; |
621f0589 KW |
1775 | int ret; |
1776 | ||
6c3944dc KW |
1777 | /* If we have to stay in sync with an external data file, zero out |
1778 | * s->data_file first. */ | |
1779 | if (data_file_is_raw(bs)) { | |
1780 | assert(has_data_file(bs)); | |
1781 | ret = bdrv_co_pwrite_zeroes(s->data_file, offset, bytes, flags); | |
1782 | if (ret < 0) { | |
1783 | return ret; | |
1784 | } | |
1785 | } | |
1786 | ||
f10ee139 EB |
1787 | /* Caller must pass aligned values, except at image end */ |
1788 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); | |
1789 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || | |
1790 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
1791 | ||
621f0589 KW |
1792 | /* The zero flag is only supported by version 3 and newer */ |
1793 | if (s->qcow_version < 3) { | |
1794 | return -ENOTSUP; | |
1795 | } | |
1796 | ||
a9a9f8f0 | 1797 | /* Each L2 slice is handled by its own loop iteration */ |
d2cb36af | 1798 | nb_clusters = size_to_clusters(s, bytes); |
621f0589 | 1799 | |
0b919fae KW |
1800 | s->cache_discards = true; |
1801 | ||
621f0589 | 1802 | while (nb_clusters > 0) { |
a9a9f8f0 | 1803 | cleared = zero_in_l2_slice(bs, offset, nb_clusters, flags); |
d2cb36af EB |
1804 | if (cleared < 0) { |
1805 | ret = cleared; | |
0b919fae | 1806 | goto fail; |
621f0589 KW |
1807 | } |
1808 | ||
d2cb36af EB |
1809 | nb_clusters -= cleared; |
1810 | offset += (cleared * s->cluster_size); | |
621f0589 KW |
1811 | } |
1812 | ||
0b919fae KW |
1813 | ret = 0; |
1814 | fail: | |
1815 | s->cache_discards = false; | |
1816 | qcow2_process_discards(bs, ret); | |
1817 | ||
1818 | return ret; | |
621f0589 | 1819 | } |
32b6444d HR |
1820 | |
1821 | /* | |
1822 | * Expands all zero clusters in a specific L1 table (or deallocates them, for | |
1823 | * non-backed non-pre-allocated zero clusters). | |
1824 | * | |
4057a2b2 HR |
1825 | * l1_entries and *visited_l1_entries are used to keep track of progress for |
1826 | * status_cb(). l1_entries contains the total number of L1 entries and | |
1827 | * *visited_l1_entries counts all visited L1 entries. | |
32b6444d HR |
1828 | */ |
1829 | static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table, | |
ecf58777 | 1830 | int l1_size, int64_t *visited_l1_entries, |
4057a2b2 | 1831 | int64_t l1_entries, |
8b13976d HR |
1832 | BlockDriverAmendStatusCB *status_cb, |
1833 | void *cb_opaque) | |
32b6444d | 1834 | { |
ff99129a | 1835 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 1836 | bool is_active_l1 = (l1_table == s->l1_table); |
415184f5 AG |
1837 | uint64_t *l2_slice = NULL; |
1838 | unsigned slice, slice_size2, n_slices; | |
32b6444d HR |
1839 | int ret; |
1840 | int i, j; | |
1841 | ||
415184f5 AG |
1842 | slice_size2 = s->l2_slice_size * sizeof(uint64_t); |
1843 | n_slices = s->cluster_size / slice_size2; | |
1844 | ||
32b6444d HR |
1845 | if (!is_active_l1) { |
1846 | /* inactive L2 tables require a buffer to be stored in when loading | |
1847 | * them from disk */ | |
415184f5 AG |
1848 | l2_slice = qemu_try_blockalign(bs->file->bs, slice_size2); |
1849 | if (l2_slice == NULL) { | |
de82815d KW |
1850 | return -ENOMEM; |
1851 | } | |
32b6444d HR |
1852 | } |
1853 | ||
1854 | for (i = 0; i < l1_size; i++) { | |
1855 | uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK; | |
0e06528e | 1856 | uint64_t l2_refcount; |
32b6444d HR |
1857 | |
1858 | if (!l2_offset) { | |
1859 | /* unallocated */ | |
4057a2b2 HR |
1860 | (*visited_l1_entries)++; |
1861 | if (status_cb) { | |
8b13976d | 1862 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 1863 | } |
32b6444d HR |
1864 | continue; |
1865 | } | |
1866 | ||
8dd93d93 HR |
1867 | if (offset_into_cluster(s, l2_offset)) { |
1868 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" | |
1869 | PRIx64 " unaligned (L1 index: %#x)", | |
1870 | l2_offset, i); | |
1871 | ret = -EIO; | |
1872 | goto fail; | |
1873 | } | |
1874 | ||
9b765486 AG |
1875 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
1876 | &l2_refcount); | |
1877 | if (ret < 0) { | |
1878 | goto fail; | |
1879 | } | |
1880 | ||
415184f5 AG |
1881 | for (slice = 0; slice < n_slices; slice++) { |
1882 | uint64_t slice_offset = l2_offset + slice * slice_size2; | |
1883 | bool l2_dirty = false; | |
226494ff AG |
1884 | if (is_active_l1) { |
1885 | /* get active L2 tables from cache */ | |
415184f5 AG |
1886 | ret = qcow2_cache_get(bs, s->l2_table_cache, slice_offset, |
1887 | (void **)&l2_slice); | |
226494ff AG |
1888 | } else { |
1889 | /* load inactive L2 tables from disk */ | |
415184f5 | 1890 | ret = bdrv_pread(bs->file, slice_offset, l2_slice, slice_size2); |
226494ff AG |
1891 | } |
1892 | if (ret < 0) { | |
1893 | goto fail; | |
32b6444d HR |
1894 | } |
1895 | ||
415184f5 AG |
1896 | for (j = 0; j < s->l2_slice_size; j++) { |
1897 | uint64_t l2_entry = be64_to_cpu(l2_slice[j]); | |
226494ff AG |
1898 | int64_t offset = l2_entry & L2E_OFFSET_MASK; |
1899 | QCow2ClusterType cluster_type = | |
808c2bb4 | 1900 | qcow2_get_cluster_type(bs, l2_entry); |
226494ff AG |
1901 | |
1902 | if (cluster_type != QCOW2_CLUSTER_ZERO_PLAIN && | |
1903 | cluster_type != QCOW2_CLUSTER_ZERO_ALLOC) { | |
32b6444d HR |
1904 | continue; |
1905 | } | |
1906 | ||
226494ff AG |
1907 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
1908 | if (!bs->backing) { | |
1909 | /* not backed; therefore we can simply deallocate the | |
1910 | * cluster */ | |
415184f5 | 1911 | l2_slice[j] = 0; |
226494ff AG |
1912 | l2_dirty = true; |
1913 | continue; | |
1914 | } | |
1915 | ||
1916 | offset = qcow2_alloc_clusters(bs, s->cluster_size); | |
1917 | if (offset < 0) { | |
1918 | ret = offset; | |
1919 | goto fail; | |
1920 | } | |
ecf58777 | 1921 | |
3a75a870 AG |
1922 | /* The offset must fit in the offset field */ |
1923 | assert((offset & L2E_OFFSET_MASK) == offset); | |
1924 | ||
226494ff AG |
1925 | if (l2_refcount > 1) { |
1926 | /* For shared L2 tables, set the refcount accordingly | |
1927 | * (it is already 1 and needs to be l2_refcount) */ | |
1928 | ret = qcow2_update_cluster_refcount( | |
1929 | bs, offset >> s->cluster_bits, | |
2aabe7c7 | 1930 | refcount_diff(1, l2_refcount), false, |
ecf58777 | 1931 | QCOW2_DISCARD_OTHER); |
226494ff AG |
1932 | if (ret < 0) { |
1933 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1934 | QCOW2_DISCARD_OTHER); | |
1935 | goto fail; | |
1936 | } | |
ecf58777 HR |
1937 | } |
1938 | } | |
32b6444d | 1939 | |
226494ff | 1940 | if (offset_into_cluster(s, offset)) { |
415184f5 | 1941 | int l2_index = slice * s->l2_slice_size + j; |
226494ff AG |
1942 | qcow2_signal_corruption( |
1943 | bs, true, -1, -1, | |
1944 | "Cluster allocation offset " | |
1945 | "%#" PRIx64 " unaligned (L2 offset: %#" | |
1946 | PRIx64 ", L2 index: %#x)", offset, | |
415184f5 | 1947 | l2_offset, l2_index); |
226494ff AG |
1948 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
1949 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1950 | QCOW2_DISCARD_ALWAYS); | |
1951 | } | |
1952 | ret = -EIO; | |
1953 | goto fail; | |
8dd93d93 | 1954 | } |
8dd93d93 | 1955 | |
226494ff | 1956 | ret = qcow2_pre_write_overlap_check(bs, 0, offset, |
966b000f | 1957 | s->cluster_size, true); |
226494ff AG |
1958 | if (ret < 0) { |
1959 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { | |
1960 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1961 | QCOW2_DISCARD_ALWAYS); | |
1962 | } | |
1963 | goto fail; | |
320c7066 | 1964 | } |
32b6444d | 1965 | |
966b000f KW |
1966 | ret = bdrv_pwrite_zeroes(s->data_file, offset, |
1967 | s->cluster_size, 0); | |
226494ff AG |
1968 | if (ret < 0) { |
1969 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { | |
1970 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1971 | QCOW2_DISCARD_ALWAYS); | |
1972 | } | |
1973 | goto fail; | |
320c7066 | 1974 | } |
32b6444d | 1975 | |
226494ff | 1976 | if (l2_refcount == 1) { |
415184f5 | 1977 | l2_slice[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED); |
226494ff | 1978 | } else { |
415184f5 | 1979 | l2_slice[j] = cpu_to_be64(offset); |
226494ff AG |
1980 | } |
1981 | l2_dirty = true; | |
e390cf5a | 1982 | } |
32b6444d | 1983 | |
226494ff AG |
1984 | if (is_active_l1) { |
1985 | if (l2_dirty) { | |
415184f5 | 1986 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
226494ff | 1987 | qcow2_cache_depends_on_flush(s->l2_table_cache); |
32b6444d | 1988 | } |
415184f5 | 1989 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
226494ff AG |
1990 | } else { |
1991 | if (l2_dirty) { | |
1992 | ret = qcow2_pre_write_overlap_check( | |
1993 | bs, QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, | |
966b000f | 1994 | slice_offset, slice_size2, false); |
226494ff AG |
1995 | if (ret < 0) { |
1996 | goto fail; | |
1997 | } | |
32b6444d | 1998 | |
415184f5 AG |
1999 | ret = bdrv_pwrite(bs->file, slice_offset, |
2000 | l2_slice, slice_size2); | |
226494ff AG |
2001 | if (ret < 0) { |
2002 | goto fail; | |
2003 | } | |
32b6444d HR |
2004 | } |
2005 | } | |
2006 | } | |
4057a2b2 HR |
2007 | |
2008 | (*visited_l1_entries)++; | |
2009 | if (status_cb) { | |
8b13976d | 2010 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 2011 | } |
32b6444d HR |
2012 | } |
2013 | ||
2014 | ret = 0; | |
2015 | ||
2016 | fail: | |
415184f5 | 2017 | if (l2_slice) { |
32b6444d | 2018 | if (!is_active_l1) { |
415184f5 | 2019 | qemu_vfree(l2_slice); |
32b6444d | 2020 | } else { |
415184f5 | 2021 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
32b6444d HR |
2022 | } |
2023 | } | |
2024 | return ret; | |
2025 | } | |
2026 | ||
2027 | /* | |
2028 | * For backed images, expands all zero clusters on the image. For non-backed | |
2029 | * images, deallocates all non-pre-allocated zero clusters (and claims the | |
2030 | * allocation for pre-allocated ones). This is important for downgrading to a | |
2031 | * qcow2 version which doesn't yet support metadata zero clusters. | |
2032 | */ | |
4057a2b2 | 2033 | int qcow2_expand_zero_clusters(BlockDriverState *bs, |
8b13976d HR |
2034 | BlockDriverAmendStatusCB *status_cb, |
2035 | void *cb_opaque) | |
32b6444d | 2036 | { |
ff99129a | 2037 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 2038 | uint64_t *l1_table = NULL; |
4057a2b2 | 2039 | int64_t l1_entries = 0, visited_l1_entries = 0; |
32b6444d HR |
2040 | int ret; |
2041 | int i, j; | |
2042 | ||
4057a2b2 HR |
2043 | if (status_cb) { |
2044 | l1_entries = s->l1_size; | |
2045 | for (i = 0; i < s->nb_snapshots; i++) { | |
2046 | l1_entries += s->snapshots[i].l1_size; | |
2047 | } | |
2048 | } | |
2049 | ||
32b6444d | 2050 | ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size, |
4057a2b2 | 2051 | &visited_l1_entries, l1_entries, |
8b13976d | 2052 | status_cb, cb_opaque); |
32b6444d HR |
2053 | if (ret < 0) { |
2054 | goto fail; | |
2055 | } | |
2056 | ||
2057 | /* Inactive L1 tables may point to active L2 tables - therefore it is | |
2058 | * necessary to flush the L2 table cache before trying to access the L2 | |
2059 | * tables pointed to by inactive L1 entries (else we might try to expand | |
2060 | * zero clusters that have already been expanded); furthermore, it is also | |
2061 | * necessary to empty the L2 table cache, since it may contain tables which | |
2062 | * are now going to be modified directly on disk, bypassing the cache. | |
2063 | * qcow2_cache_empty() does both for us. */ | |
2064 | ret = qcow2_cache_empty(bs, s->l2_table_cache); | |
2065 | if (ret < 0) { | |
2066 | goto fail; | |
2067 | } | |
2068 | ||
2069 | for (i = 0; i < s->nb_snapshots; i++) { | |
c9a442e4 AG |
2070 | int l1_size2; |
2071 | uint64_t *new_l1_table; | |
2072 | Error *local_err = NULL; | |
2073 | ||
2074 | ret = qcow2_validate_table(bs, s->snapshots[i].l1_table_offset, | |
2075 | s->snapshots[i].l1_size, sizeof(uint64_t), | |
2076 | QCOW_MAX_L1_SIZE, "Snapshot L1 table", | |
2077 | &local_err); | |
2078 | if (ret < 0) { | |
2079 | error_report_err(local_err); | |
2080 | goto fail; | |
2081 | } | |
32b6444d | 2082 | |
c9a442e4 AG |
2083 | l1_size2 = s->snapshots[i].l1_size * sizeof(uint64_t); |
2084 | new_l1_table = g_try_realloc(l1_table, l1_size2); | |
de7269d2 AG |
2085 | |
2086 | if (!new_l1_table) { | |
2087 | ret = -ENOMEM; | |
2088 | goto fail; | |
2089 | } | |
2090 | ||
2091 | l1_table = new_l1_table; | |
32b6444d | 2092 | |
c9a442e4 AG |
2093 | ret = bdrv_pread(bs->file, s->snapshots[i].l1_table_offset, |
2094 | l1_table, l1_size2); | |
32b6444d HR |
2095 | if (ret < 0) { |
2096 | goto fail; | |
2097 | } | |
2098 | ||
2099 | for (j = 0; j < s->snapshots[i].l1_size; j++) { | |
2100 | be64_to_cpus(&l1_table[j]); | |
2101 | } | |
2102 | ||
2103 | ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size, | |
4057a2b2 | 2104 | &visited_l1_entries, l1_entries, |
8b13976d | 2105 | status_cb, cb_opaque); |
32b6444d HR |
2106 | if (ret < 0) { |
2107 | goto fail; | |
2108 | } | |
2109 | } | |
2110 | ||
2111 | ret = 0; | |
2112 | ||
2113 | fail: | |
32b6444d HR |
2114 | g_free(l1_table); |
2115 | return ret; | |
2116 | } |