]>
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 | ||
da34e65c | 28 | #include "qapi/error.h" |
45aba42f | 29 | #include "qemu-common.h" |
737e150e | 30 | #include "block/block_int.h" |
45aba42f | 31 | #include "block/qcow2.h" |
58369e22 | 32 | #include "qemu/bswap.h" |
3cce16f4 | 33 | #include "trace.h" |
45aba42f | 34 | |
2cf7cfa1 KW |
35 | int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size, |
36 | bool exact_size) | |
45aba42f | 37 | { |
ff99129a | 38 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 | 39 | int new_l1_size2, ret, i; |
45aba42f | 40 | uint64_t *new_l1_table; |
fda74f82 | 41 | int64_t old_l1_table_offset, old_l1_size; |
2cf7cfa1 | 42 | int64_t new_l1_table_offset, new_l1_size; |
45aba42f KW |
43 | uint8_t data[12]; |
44 | ||
72893756 | 45 | if (min_size <= s->l1_size) |
45aba42f | 46 | return 0; |
72893756 | 47 | |
b93f9950 HR |
48 | /* Do a sanity check on min_size before trying to calculate new_l1_size |
49 | * (this prevents overflows during the while loop for the calculation of | |
50 | * new_l1_size) */ | |
51 | if (min_size > INT_MAX / sizeof(uint64_t)) { | |
52 | return -EFBIG; | |
53 | } | |
54 | ||
72893756 SH |
55 | if (exact_size) { |
56 | new_l1_size = min_size; | |
57 | } else { | |
58 | /* Bump size up to reduce the number of times we have to grow */ | |
59 | new_l1_size = s->l1_size; | |
60 | if (new_l1_size == 0) { | |
61 | new_l1_size = 1; | |
62 | } | |
63 | while (min_size > new_l1_size) { | |
64 | new_l1_size = (new_l1_size * 3 + 1) / 2; | |
65 | } | |
45aba42f | 66 | } |
72893756 | 67 | |
84c26520 HR |
68 | QEMU_BUILD_BUG_ON(QCOW_MAX_L1_SIZE > INT_MAX); |
69 | if (new_l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) { | |
2cf7cfa1 KW |
70 | return -EFBIG; |
71 | } | |
72 | ||
45aba42f | 73 | #ifdef DEBUG_ALLOC2 |
2cf7cfa1 KW |
74 | fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n", |
75 | s->l1_size, new_l1_size); | |
45aba42f KW |
76 | #endif |
77 | ||
78 | new_l1_size2 = sizeof(uint64_t) * new_l1_size; | |
9a4f4c31 | 79 | new_l1_table = qemu_try_blockalign(bs->file->bs, |
de82815d KW |
80 | align_offset(new_l1_size2, 512)); |
81 | if (new_l1_table == NULL) { | |
82 | return -ENOMEM; | |
83 | } | |
84 | memset(new_l1_table, 0, align_offset(new_l1_size2, 512)); | |
85 | ||
0647d47c SH |
86 | if (s->l1_size) { |
87 | memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t)); | |
88 | } | |
45aba42f KW |
89 | |
90 | /* write new table (align to cluster) */ | |
66f82cee | 91 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE); |
ed6ccf0f | 92 | new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2); |
5d757b56 | 93 | if (new_l1_table_offset < 0) { |
de82815d | 94 | qemu_vfree(new_l1_table); |
5d757b56 KW |
95 | return new_l1_table_offset; |
96 | } | |
29c1a730 KW |
97 | |
98 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
99 | if (ret < 0) { | |
80fa3341 | 100 | goto fail; |
29c1a730 | 101 | } |
45aba42f | 102 | |
cf93980e HR |
103 | /* the L1 position has not yet been updated, so these clusters must |
104 | * indeed be completely free */ | |
231bb267 HR |
105 | ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset, |
106 | new_l1_size2); | |
cf93980e HR |
107 | if (ret < 0) { |
108 | goto fail; | |
109 | } | |
110 | ||
66f82cee | 111 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE); |
45aba42f KW |
112 | for(i = 0; i < s->l1_size; i++) |
113 | new_l1_table[i] = cpu_to_be64(new_l1_table[i]); | |
d9ca2ea2 | 114 | ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, |
9a4f4c31 | 115 | new_l1_table, new_l1_size2); |
8b3b7206 | 116 | if (ret < 0) |
45aba42f KW |
117 | goto fail; |
118 | for(i = 0; i < s->l1_size; i++) | |
119 | new_l1_table[i] = be64_to_cpu(new_l1_table[i]); | |
120 | ||
121 | /* set new table */ | |
66f82cee | 122 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE); |
f1f7a1dd | 123 | stl_be_p(data, new_l1_size); |
e4ef9f46 | 124 | stq_be_p(data + 4, new_l1_table_offset); |
d9ca2ea2 | 125 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), |
9a4f4c31 | 126 | data, sizeof(data)); |
8b3b7206 | 127 | if (ret < 0) { |
45aba42f | 128 | goto fail; |
fb8fa77c | 129 | } |
de82815d | 130 | qemu_vfree(s->l1_table); |
fda74f82 | 131 | old_l1_table_offset = s->l1_table_offset; |
45aba42f KW |
132 | s->l1_table_offset = new_l1_table_offset; |
133 | s->l1_table = new_l1_table; | |
fda74f82 | 134 | old_l1_size = s->l1_size; |
45aba42f | 135 | s->l1_size = new_l1_size; |
fda74f82 HR |
136 | qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * sizeof(uint64_t), |
137 | QCOW2_DISCARD_OTHER); | |
45aba42f KW |
138 | return 0; |
139 | fail: | |
de82815d | 140 | qemu_vfree(new_l1_table); |
6cfcb9b8 KW |
141 | qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2, |
142 | QCOW2_DISCARD_OTHER); | |
8b3b7206 | 143 | return ret; |
45aba42f KW |
144 | } |
145 | ||
45aba42f KW |
146 | /* |
147 | * l2_load | |
148 | * | |
149 | * Loads a L2 table into memory. If the table is in the cache, the cache | |
150 | * is used; otherwise the L2 table is loaded from the image file. | |
151 | * | |
152 | * Returns a pointer to the L2 table on success, or NULL if the read from | |
153 | * the image file failed. | |
154 | */ | |
155 | ||
55c17e98 KW |
156 | static int l2_load(BlockDriverState *bs, uint64_t l2_offset, |
157 | uint64_t **l2_table) | |
45aba42f | 158 | { |
ff99129a | 159 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 160 | |
9be38598 EH |
161 | return qcow2_cache_get(bs, s->l2_table_cache, l2_offset, |
162 | (void **)l2_table); | |
45aba42f KW |
163 | } |
164 | ||
6583e3c7 KW |
165 | /* |
166 | * Writes one sector of the L1 table to the disk (can't update single entries | |
167 | * and we really don't want bdrv_pread to perform a read-modify-write) | |
168 | */ | |
169 | #define L1_ENTRIES_PER_SECTOR (512 / 8) | |
e23e400e | 170 | int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index) |
6583e3c7 | 171 | { |
ff99129a | 172 | BDRVQcow2State *s = bs->opaque; |
a1391444 | 173 | uint64_t buf[L1_ENTRIES_PER_SECTOR] = { 0 }; |
6583e3c7 | 174 | int l1_start_index; |
f7defcb6 | 175 | int i, ret; |
6583e3c7 KW |
176 | |
177 | l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1); | |
a1391444 HR |
178 | for (i = 0; i < L1_ENTRIES_PER_SECTOR && l1_start_index + i < s->l1_size; |
179 | i++) | |
180 | { | |
6583e3c7 KW |
181 | buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]); |
182 | } | |
183 | ||
231bb267 | 184 | ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1, |
cf93980e HR |
185 | s->l1_table_offset + 8 * l1_start_index, sizeof(buf)); |
186 | if (ret < 0) { | |
187 | return ret; | |
188 | } | |
189 | ||
66f82cee | 190 | BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); |
d9ca2ea2 | 191 | ret = bdrv_pwrite_sync(bs->file, |
9a4f4c31 KW |
192 | s->l1_table_offset + 8 * l1_start_index, |
193 | buf, sizeof(buf)); | |
f7defcb6 KW |
194 | if (ret < 0) { |
195 | return ret; | |
6583e3c7 KW |
196 | } |
197 | ||
198 | return 0; | |
199 | } | |
200 | ||
45aba42f KW |
201 | /* |
202 | * l2_allocate | |
203 | * | |
204 | * Allocate a new l2 entry in the file. If l1_index points to an already | |
205 | * used entry in the L2 table (i.e. we are doing a copy on write for the L2 | |
206 | * table) copy the contents of the old L2 table into the newly allocated one. | |
207 | * Otherwise the new table is initialized with zeros. | |
208 | * | |
209 | */ | |
210 | ||
c46e1167 | 211 | static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table) |
45aba42f | 212 | { |
ff99129a | 213 | BDRVQcow2State *s = bs->opaque; |
6583e3c7 | 214 | uint64_t old_l2_offset; |
8585afd8 | 215 | uint64_t *l2_table = NULL; |
f4f0d391 | 216 | int64_t l2_offset; |
c46e1167 | 217 | int ret; |
45aba42f KW |
218 | |
219 | old_l2_offset = s->l1_table[l1_index]; | |
220 | ||
3cce16f4 KW |
221 | trace_qcow2_l2_allocate(bs, l1_index); |
222 | ||
45aba42f KW |
223 | /* allocate a new l2 entry */ |
224 | ||
ed6ccf0f | 225 | l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t)); |
5d757b56 | 226 | if (l2_offset < 0) { |
be0b742e HR |
227 | ret = l2_offset; |
228 | goto fail; | |
5d757b56 | 229 | } |
29c1a730 KW |
230 | |
231 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
232 | if (ret < 0) { | |
233 | goto fail; | |
234 | } | |
45aba42f | 235 | |
45aba42f KW |
236 | /* allocate a new entry in the l2 cache */ |
237 | ||
3cce16f4 | 238 | trace_qcow2_l2_allocate_get_empty(bs, l1_index); |
29c1a730 KW |
239 | ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table); |
240 | if (ret < 0) { | |
be0b742e | 241 | goto fail; |
29c1a730 KW |
242 | } |
243 | ||
244 | l2_table = *table; | |
45aba42f | 245 | |
8e37f681 | 246 | if ((old_l2_offset & L1E_OFFSET_MASK) == 0) { |
45aba42f KW |
247 | /* if there was no old l2 table, clear the new table */ |
248 | memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); | |
249 | } else { | |
29c1a730 KW |
250 | uint64_t* old_table; |
251 | ||
45aba42f | 252 | /* if there was an old l2 table, read it from the disk */ |
66f82cee | 253 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ); |
8e37f681 KW |
254 | ret = qcow2_cache_get(bs, s->l2_table_cache, |
255 | old_l2_offset & L1E_OFFSET_MASK, | |
29c1a730 KW |
256 | (void**) &old_table); |
257 | if (ret < 0) { | |
258 | goto fail; | |
259 | } | |
260 | ||
261 | memcpy(l2_table, old_table, s->cluster_size); | |
262 | ||
a3f1afb4 | 263 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &old_table); |
45aba42f | 264 | } |
29c1a730 | 265 | |
45aba42f | 266 | /* write the l2 table to the file */ |
66f82cee | 267 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE); |
29c1a730 | 268 | |
3cce16f4 | 269 | trace_qcow2_l2_allocate_write_l2(bs, l1_index); |
72e80b89 | 270 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
29c1a730 | 271 | ret = qcow2_cache_flush(bs, s->l2_table_cache); |
c46e1167 | 272 | if (ret < 0) { |
175e1152 KW |
273 | goto fail; |
274 | } | |
275 | ||
276 | /* update the L1 entry */ | |
3cce16f4 | 277 | trace_qcow2_l2_allocate_write_l1(bs, l1_index); |
175e1152 | 278 | s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; |
e23e400e | 279 | ret = qcow2_write_l1_entry(bs, l1_index); |
175e1152 KW |
280 | if (ret < 0) { |
281 | goto fail; | |
c46e1167 | 282 | } |
45aba42f | 283 | |
c46e1167 | 284 | *table = l2_table; |
3cce16f4 | 285 | trace_qcow2_l2_allocate_done(bs, l1_index, 0); |
c46e1167 | 286 | return 0; |
175e1152 KW |
287 | |
288 | fail: | |
3cce16f4 | 289 | trace_qcow2_l2_allocate_done(bs, l1_index, ret); |
8585afd8 HR |
290 | if (l2_table != NULL) { |
291 | qcow2_cache_put(bs, s->l2_table_cache, (void**) table); | |
292 | } | |
68dba0bf | 293 | s->l1_table[l1_index] = old_l2_offset; |
e3b21ef9 HR |
294 | if (l2_offset > 0) { |
295 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), | |
296 | QCOW2_DISCARD_ALWAYS); | |
297 | } | |
175e1152 | 298 | return ret; |
45aba42f KW |
299 | } |
300 | ||
2bfcc4a0 KW |
301 | /* |
302 | * Checks how many clusters in a given L2 table are contiguous in the image | |
303 | * file. As soon as one of the flags in the bitmask stop_flags changes compared | |
304 | * to the first cluster, the search is stopped and the cluster is not counted | |
305 | * as contiguous. (This allows it, for example, to stop at the first compressed | |
306 | * cluster which may require a different handling) | |
307 | */ | |
b6d36def | 308 | static int count_contiguous_clusters(int nb_clusters, int cluster_size, |
61653008 | 309 | uint64_t *l2_table, uint64_t stop_flags) |
45aba42f KW |
310 | { |
311 | int i; | |
3ef95218 | 312 | QCow2ClusterType first_cluster_type; |
78a52ad5 | 313 | uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED; |
15684a47 HR |
314 | uint64_t first_entry = be64_to_cpu(l2_table[0]); |
315 | uint64_t offset = first_entry & mask; | |
45aba42f | 316 | |
564a6b69 | 317 | if (!offset) { |
45aba42f | 318 | return 0; |
564a6b69 | 319 | } |
45aba42f | 320 | |
564a6b69 HR |
321 | /* must be allocated */ |
322 | first_cluster_type = qcow2_get_cluster_type(first_entry); | |
323 | assert(first_cluster_type == QCOW2_CLUSTER_NORMAL || | |
fdfab37d | 324 | first_cluster_type == QCOW2_CLUSTER_ZERO_ALLOC); |
15684a47 | 325 | |
61653008 | 326 | for (i = 0; i < nb_clusters; i++) { |
2bfcc4a0 KW |
327 | uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask; |
328 | if (offset + (uint64_t) i * cluster_size != l2_entry) { | |
45aba42f | 329 | break; |
2bfcc4a0 KW |
330 | } |
331 | } | |
45aba42f | 332 | |
61653008 | 333 | return i; |
45aba42f KW |
334 | } |
335 | ||
4341df8a EB |
336 | /* |
337 | * Checks how many consecutive unallocated clusters in a given L2 | |
338 | * table have the same cluster type. | |
339 | */ | |
340 | static int count_contiguous_clusters_unallocated(int nb_clusters, | |
341 | uint64_t *l2_table, | |
3ef95218 | 342 | QCow2ClusterType wanted_type) |
45aba42f | 343 | { |
2bfcc4a0 KW |
344 | int i; |
345 | ||
fdfab37d | 346 | assert(wanted_type == QCOW2_CLUSTER_ZERO_PLAIN || |
4341df8a | 347 | wanted_type == QCOW2_CLUSTER_UNALLOCATED); |
2bfcc4a0 | 348 | for (i = 0; i < nb_clusters; i++) { |
4341df8a | 349 | uint64_t entry = be64_to_cpu(l2_table[i]); |
3ef95218 | 350 | QCow2ClusterType type = qcow2_get_cluster_type(entry); |
45aba42f | 351 | |
fdfab37d | 352 | if (type != wanted_type) { |
2bfcc4a0 KW |
353 | break; |
354 | } | |
355 | } | |
45aba42f KW |
356 | |
357 | return i; | |
358 | } | |
359 | ||
672f0f2c AG |
360 | static int coroutine_fn do_perform_cow_read(BlockDriverState *bs, |
361 | uint64_t src_cluster_offset, | |
362 | unsigned offset_in_cluster, | |
86b862c4 | 363 | QEMUIOVector *qiov) |
45aba42f | 364 | { |
aaa4d20b | 365 | int ret; |
1b9f1491 | 366 | |
86b862c4 | 367 | if (qiov->size == 0) { |
99450c6f AG |
368 | return 0; |
369 | } | |
370 | ||
66f82cee | 371 | BLKDBG_EVENT(bs->file, BLKDBG_COW_READ); |
aef4acb6 | 372 | |
dba28555 | 373 | if (!bs->drv) { |
672f0f2c | 374 | return -ENOMEDIUM; |
dba28555 HR |
375 | } |
376 | ||
aef4acb6 SH |
377 | /* Call .bdrv_co_readv() directly instead of using the public block-layer |
378 | * interface. This avoids double I/O throttling and request tracking, | |
379 | * which can lead to deadlock when block layer copy-on-read is enabled. | |
380 | */ | |
aaa4d20b | 381 | ret = bs->drv->bdrv_co_preadv(bs, src_cluster_offset + offset_in_cluster, |
86b862c4 | 382 | qiov->size, qiov, 0); |
1b9f1491 | 383 | if (ret < 0) { |
672f0f2c | 384 | return ret; |
1b9f1491 KW |
385 | } |
386 | ||
672f0f2c AG |
387 | return 0; |
388 | } | |
389 | ||
390 | static bool coroutine_fn do_perform_cow_encrypt(BlockDriverState *bs, | |
391 | uint64_t src_cluster_offset, | |
4652b8f3 | 392 | uint64_t cluster_offset, |
672f0f2c AG |
393 | unsigned offset_in_cluster, |
394 | uint8_t *buffer, | |
395 | unsigned bytes) | |
396 | { | |
397 | if (bytes && bs->encrypted) { | |
398 | BDRVQcow2State *s = bs->opaque; | |
4652b8f3 DB |
399 | int64_t sector = (s->crypt_physical_offset ? |
400 | (cluster_offset + offset_in_cluster) : | |
401 | (src_cluster_offset + offset_in_cluster)) | |
aaa4d20b | 402 | >> BDRV_SECTOR_BITS; |
aaa4d20b KW |
403 | assert((offset_in_cluster & ~BDRV_SECTOR_MASK) == 0); |
404 | assert((bytes & ~BDRV_SECTOR_MASK) == 0); | |
b25b387f DB |
405 | assert(s->crypto); |
406 | if (qcrypto_block_encrypt(s->crypto, sector, buffer, | |
407 | bytes, NULL) < 0) { | |
672f0f2c | 408 | return false; |
f6fa64f6 | 409 | } |
45aba42f | 410 | } |
672f0f2c AG |
411 | return true; |
412 | } | |
413 | ||
414 | static int coroutine_fn do_perform_cow_write(BlockDriverState *bs, | |
415 | uint64_t cluster_offset, | |
416 | unsigned offset_in_cluster, | |
86b862c4 | 417 | QEMUIOVector *qiov) |
672f0f2c | 418 | { |
672f0f2c AG |
419 | int ret; |
420 | ||
86b862c4 | 421 | if (qiov->size == 0) { |
672f0f2c AG |
422 | return 0; |
423 | } | |
424 | ||
231bb267 | 425 | ret = qcow2_pre_write_overlap_check(bs, 0, |
86b862c4 | 426 | cluster_offset + offset_in_cluster, qiov->size); |
cf93980e | 427 | if (ret < 0) { |
672f0f2c | 428 | return ret; |
cf93980e HR |
429 | } |
430 | ||
66f82cee | 431 | BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE); |
a03ef88f | 432 | ret = bdrv_co_pwritev(bs->file, cluster_offset + offset_in_cluster, |
86b862c4 | 433 | qiov->size, qiov, 0); |
1b9f1491 | 434 | if (ret < 0) { |
672f0f2c | 435 | return ret; |
1b9f1491 KW |
436 | } |
437 | ||
672f0f2c | 438 | return 0; |
45aba42f KW |
439 | } |
440 | ||
441 | ||
442 | /* | |
443 | * get_cluster_offset | |
444 | * | |
ecfe1863 KW |
445 | * For a given offset of the virtual disk, find the cluster type and offset in |
446 | * the qcow2 file. The offset is stored in *cluster_offset. | |
45aba42f | 447 | * |
ecfe1863 KW |
448 | * On entry, *bytes is the maximum number of contiguous bytes starting at |
449 | * offset that we are interested in. | |
45aba42f | 450 | * |
ecfe1863 KW |
451 | * On exit, *bytes is the number of bytes starting at offset that have the same |
452 | * cluster type and (if applicable) are stored contiguously in the image file. | |
453 | * Compressed clusters are always returned one by one. | |
45aba42f | 454 | * |
68d000a3 KW |
455 | * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error |
456 | * cases. | |
45aba42f | 457 | */ |
1c46efaa | 458 | int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, |
ecfe1863 | 459 | unsigned int *bytes, uint64_t *cluster_offset) |
45aba42f | 460 | { |
ff99129a | 461 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
462 | unsigned int l2_index; |
463 | uint64_t l1_index, l2_offset, *l2_table; | |
45aba42f | 464 | int l1_bits, c; |
c834cba9 HR |
465 | unsigned int offset_in_cluster; |
466 | uint64_t bytes_available, bytes_needed, nb_clusters; | |
3ef95218 | 467 | QCow2ClusterType type; |
55c17e98 | 468 | int ret; |
45aba42f | 469 | |
b2f65d6b | 470 | offset_in_cluster = offset_into_cluster(s, offset); |
ecfe1863 | 471 | bytes_needed = (uint64_t) *bytes + offset_in_cluster; |
45aba42f | 472 | |
b2f65d6b | 473 | l1_bits = s->l2_bits + s->cluster_bits; |
45aba42f | 474 | |
b2f65d6b KW |
475 | /* compute how many bytes there are between the start of the cluster |
476 | * containing offset and the end of the l1 entry */ | |
477 | bytes_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1)) | |
478 | + offset_in_cluster; | |
45aba42f | 479 | |
b2f65d6b KW |
480 | if (bytes_needed > bytes_available) { |
481 | bytes_needed = bytes_available; | |
45aba42f KW |
482 | } |
483 | ||
1c46efaa | 484 | *cluster_offset = 0; |
45aba42f | 485 | |
b6af0975 | 486 | /* seek to the l2 offset in the l1 table */ |
45aba42f KW |
487 | |
488 | l1_index = offset >> l1_bits; | |
68d000a3 | 489 | if (l1_index >= s->l1_size) { |
3ef95218 | 490 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 491 | goto out; |
68d000a3 | 492 | } |
45aba42f | 493 | |
68d000a3 KW |
494 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
495 | if (!l2_offset) { | |
3ef95218 | 496 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 497 | goto out; |
68d000a3 | 498 | } |
45aba42f | 499 | |
a97c67ee HR |
500 | if (offset_into_cluster(s, l2_offset)) { |
501 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
502 | " unaligned (L1 index: %#" PRIx64 ")", | |
503 | l2_offset, l1_index); | |
504 | return -EIO; | |
505 | } | |
506 | ||
45aba42f KW |
507 | /* load the l2 table in memory */ |
508 | ||
55c17e98 KW |
509 | ret = l2_load(bs, l2_offset, &l2_table); |
510 | if (ret < 0) { | |
511 | return ret; | |
1c46efaa | 512 | } |
45aba42f KW |
513 | |
514 | /* find the cluster offset for the given disk offset */ | |
515 | ||
24990c5b | 516 | l2_index = offset_to_l2_index(s, offset); |
1c46efaa | 517 | *cluster_offset = be64_to_cpu(l2_table[l2_index]); |
b6d36def | 518 | |
b2f65d6b | 519 | nb_clusters = size_to_clusters(s, bytes_needed); |
c834cba9 HR |
520 | /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned |
521 | * integers; the minimum cluster size is 512, so this assertion is always | |
522 | * true */ | |
523 | assert(nb_clusters <= INT_MAX); | |
45aba42f | 524 | |
3ef95218 | 525 | type = qcow2_get_cluster_type(*cluster_offset); |
fdfab37d EB |
526 | if (s->qcow_version < 3 && (type == QCOW2_CLUSTER_ZERO_PLAIN || |
527 | type == QCOW2_CLUSTER_ZERO_ALLOC)) { | |
528 | qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found" | |
529 | " in pre-v3 image (L2 offset: %#" PRIx64 | |
530 | ", L2 index: %#x)", l2_offset, l2_index); | |
531 | ret = -EIO; | |
532 | goto fail; | |
533 | } | |
3ef95218 | 534 | switch (type) { |
68d000a3 KW |
535 | case QCOW2_CLUSTER_COMPRESSED: |
536 | /* Compressed clusters can only be processed one by one */ | |
537 | c = 1; | |
538 | *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK; | |
539 | break; | |
fdfab37d | 540 | case QCOW2_CLUSTER_ZERO_PLAIN: |
68d000a3 | 541 | case QCOW2_CLUSTER_UNALLOCATED: |
45aba42f | 542 | /* how many empty clusters ? */ |
4341df8a | 543 | c = count_contiguous_clusters_unallocated(nb_clusters, |
fdfab37d | 544 | &l2_table[l2_index], type); |
68d000a3 KW |
545 | *cluster_offset = 0; |
546 | break; | |
fdfab37d | 547 | case QCOW2_CLUSTER_ZERO_ALLOC: |
68d000a3 | 548 | case QCOW2_CLUSTER_NORMAL: |
45aba42f KW |
549 | /* how many allocated clusters ? */ |
550 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
fdfab37d | 551 | &l2_table[l2_index], QCOW_OFLAG_ZERO); |
68d000a3 | 552 | *cluster_offset &= L2E_OFFSET_MASK; |
a97c67ee | 553 | if (offset_into_cluster(s, *cluster_offset)) { |
fdfab37d EB |
554 | qcow2_signal_corruption(bs, true, -1, -1, |
555 | "Cluster allocation offset %#" | |
a97c67ee HR |
556 | PRIx64 " unaligned (L2 offset: %#" PRIx64 |
557 | ", L2 index: %#x)", *cluster_offset, | |
558 | l2_offset, l2_index); | |
559 | ret = -EIO; | |
560 | goto fail; | |
561 | } | |
68d000a3 | 562 | break; |
1417d7e4 KW |
563 | default: |
564 | abort(); | |
45aba42f KW |
565 | } |
566 | ||
29c1a730 KW |
567 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
568 | ||
c834cba9 | 569 | bytes_available = (int64_t)c * s->cluster_size; |
68d000a3 | 570 | |
45aba42f | 571 | out: |
b2f65d6b KW |
572 | if (bytes_available > bytes_needed) { |
573 | bytes_available = bytes_needed; | |
574 | } | |
45aba42f | 575 | |
c834cba9 HR |
576 | /* bytes_available <= bytes_needed <= *bytes + offset_in_cluster; |
577 | * subtracting offset_in_cluster will therefore definitely yield something | |
578 | * not exceeding UINT_MAX */ | |
579 | assert(bytes_available - offset_in_cluster <= UINT_MAX); | |
ecfe1863 | 580 | *bytes = bytes_available - offset_in_cluster; |
45aba42f | 581 | |
3ef95218 | 582 | return type; |
a97c67ee HR |
583 | |
584 | fail: | |
585 | qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table); | |
586 | return ret; | |
45aba42f KW |
587 | } |
588 | ||
589 | /* | |
590 | * get_cluster_table | |
591 | * | |
592 | * for a given disk offset, load (and allocate if needed) | |
593 | * the l2 table. | |
594 | * | |
595 | * the l2 table offset in the qcow2 file and the cluster index | |
596 | * in the l2 table are given to the caller. | |
597 | * | |
1e3e8f1a | 598 | * Returns 0 on success, -errno in failure case |
45aba42f | 599 | */ |
45aba42f KW |
600 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
601 | uint64_t **new_l2_table, | |
45aba42f KW |
602 | int *new_l2_index) |
603 | { | |
ff99129a | 604 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
605 | unsigned int l2_index; |
606 | uint64_t l1_index, l2_offset; | |
c46e1167 | 607 | uint64_t *l2_table = NULL; |
80ee15a6 | 608 | int ret; |
45aba42f | 609 | |
b6af0975 | 610 | /* seek to the l2 offset in the l1 table */ |
45aba42f KW |
611 | |
612 | l1_index = offset >> (s->l2_bits + s->cluster_bits); | |
613 | if (l1_index >= s->l1_size) { | |
72893756 | 614 | ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
1e3e8f1a KW |
615 | if (ret < 0) { |
616 | return ret; | |
617 | } | |
45aba42f | 618 | } |
8e37f681 | 619 | |
2cf7cfa1 | 620 | assert(l1_index < s->l1_size); |
8e37f681 | 621 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
a97c67ee HR |
622 | if (offset_into_cluster(s, l2_offset)) { |
623 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
624 | " unaligned (L1 index: %#" PRIx64 ")", | |
625 | l2_offset, l1_index); | |
626 | return -EIO; | |
627 | } | |
45aba42f KW |
628 | |
629 | /* seek the l2 table of the given l2 offset */ | |
630 | ||
8e37f681 | 631 | if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) { |
45aba42f | 632 | /* load the l2 table in memory */ |
55c17e98 KW |
633 | ret = l2_load(bs, l2_offset, &l2_table); |
634 | if (ret < 0) { | |
635 | return ret; | |
1e3e8f1a | 636 | } |
45aba42f | 637 | } else { |
16fde5f2 | 638 | /* First allocate a new L2 table (and do COW if needed) */ |
c46e1167 KW |
639 | ret = l2_allocate(bs, l1_index, &l2_table); |
640 | if (ret < 0) { | |
641 | return ret; | |
1e3e8f1a | 642 | } |
16fde5f2 KW |
643 | |
644 | /* Then decrease the refcount of the old table */ | |
645 | if (l2_offset) { | |
6cfcb9b8 KW |
646 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), |
647 | QCOW2_DISCARD_OTHER); | |
16fde5f2 | 648 | } |
45aba42f KW |
649 | } |
650 | ||
651 | /* find the cluster offset for the given disk offset */ | |
652 | ||
24990c5b | 653 | l2_index = offset_to_l2_index(s, offset); |
45aba42f KW |
654 | |
655 | *new_l2_table = l2_table; | |
45aba42f KW |
656 | *new_l2_index = l2_index; |
657 | ||
1e3e8f1a | 658 | return 0; |
45aba42f KW |
659 | } |
660 | ||
661 | /* | |
662 | * alloc_compressed_cluster_offset | |
663 | * | |
664 | * For a given offset of the disk image, return cluster offset in | |
665 | * qcow2 file. | |
666 | * | |
667 | * If the offset is not found, allocate a new compressed cluster. | |
668 | * | |
669 | * Return the cluster offset if successful, | |
670 | * Return 0, otherwise. | |
671 | * | |
672 | */ | |
673 | ||
ed6ccf0f KW |
674 | uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
675 | uint64_t offset, | |
676 | int compressed_size) | |
45aba42f | 677 | { |
ff99129a | 678 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 679 | int l2_index, ret; |
3948d1d4 | 680 | uint64_t *l2_table; |
f4f0d391 | 681 | int64_t cluster_offset; |
45aba42f KW |
682 | int nb_csectors; |
683 | ||
3948d1d4 | 684 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
1e3e8f1a | 685 | if (ret < 0) { |
45aba42f | 686 | return 0; |
1e3e8f1a | 687 | } |
45aba42f | 688 | |
b0b6862e KW |
689 | /* Compression can't overwrite anything. Fail if the cluster was already |
690 | * allocated. */ | |
45aba42f | 691 | cluster_offset = be64_to_cpu(l2_table[l2_index]); |
b0b6862e | 692 | if (cluster_offset & L2E_OFFSET_MASK) { |
8f1efd00 KW |
693 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
694 | return 0; | |
695 | } | |
45aba42f | 696 | |
ed6ccf0f | 697 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
5d757b56 | 698 | if (cluster_offset < 0) { |
29c1a730 | 699 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
5d757b56 KW |
700 | return 0; |
701 | } | |
702 | ||
45aba42f KW |
703 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
704 | (cluster_offset >> 9); | |
705 | ||
706 | cluster_offset |= QCOW_OFLAG_COMPRESSED | | |
707 | ((uint64_t)nb_csectors << s->csize_shift); | |
708 | ||
709 | /* update L2 table */ | |
710 | ||
711 | /* compressed clusters never have the copied flag */ | |
712 | ||
66f82cee | 713 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
72e80b89 | 714 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
45aba42f | 715 | l2_table[l2_index] = cpu_to_be64(cluster_offset); |
a3f1afb4 | 716 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
4c1612d9 | 717 | |
29c1a730 | 718 | return cluster_offset; |
4c1612d9 KW |
719 | } |
720 | ||
99450c6f | 721 | static int perform_cow(BlockDriverState *bs, QCowL2Meta *m) |
593fb83c | 722 | { |
ff99129a | 723 | BDRVQcow2State *s = bs->opaque; |
99450c6f AG |
724 | Qcow2COWRegion *start = &m->cow_start; |
725 | Qcow2COWRegion *end = &m->cow_end; | |
672f0f2c | 726 | unsigned buffer_size; |
b3cf1c7c AG |
727 | unsigned data_bytes = end->offset - (start->offset + start->nb_bytes); |
728 | bool merge_reads; | |
672f0f2c | 729 | uint8_t *start_buffer, *end_buffer; |
86b862c4 | 730 | QEMUIOVector qiov; |
593fb83c KW |
731 | int ret; |
732 | ||
672f0f2c | 733 | assert(start->nb_bytes <= UINT_MAX - end->nb_bytes); |
b3cf1c7c AG |
734 | assert(start->nb_bytes + end->nb_bytes <= UINT_MAX - data_bytes); |
735 | assert(start->offset + start->nb_bytes <= end->offset); | |
ee22a9d8 | 736 | assert(!m->data_qiov || m->data_qiov->size == data_bytes); |
672f0f2c | 737 | |
99450c6f | 738 | if (start->nb_bytes == 0 && end->nb_bytes == 0) { |
593fb83c KW |
739 | return 0; |
740 | } | |
741 | ||
b3cf1c7c AG |
742 | /* If we have to read both the start and end COW regions and the |
743 | * middle region is not too large then perform just one read | |
744 | * operation */ | |
745 | merge_reads = start->nb_bytes && end->nb_bytes && data_bytes <= 16384; | |
746 | if (merge_reads) { | |
747 | buffer_size = start->nb_bytes + data_bytes + end->nb_bytes; | |
748 | } else { | |
749 | /* If we have to do two reads, add some padding in the middle | |
750 | * if necessary to make sure that the end region is optimally | |
751 | * aligned. */ | |
752 | size_t align = bdrv_opt_mem_align(bs); | |
753 | assert(align > 0 && align <= UINT_MAX); | |
754 | assert(QEMU_ALIGN_UP(start->nb_bytes, align) <= | |
755 | UINT_MAX - end->nb_bytes); | |
756 | buffer_size = QEMU_ALIGN_UP(start->nb_bytes, align) + end->nb_bytes; | |
757 | } | |
758 | ||
759 | /* Reserve a buffer large enough to store all the data that we're | |
760 | * going to read */ | |
672f0f2c AG |
761 | start_buffer = qemu_try_blockalign(bs, buffer_size); |
762 | if (start_buffer == NULL) { | |
763 | return -ENOMEM; | |
764 | } | |
765 | /* The part of the buffer where the end region is located */ | |
766 | end_buffer = start_buffer + buffer_size - end->nb_bytes; | |
767 | ||
ee22a9d8 | 768 | qemu_iovec_init(&qiov, 2 + (m->data_qiov ? m->data_qiov->niov : 0)); |
86b862c4 | 769 | |
593fb83c | 770 | qemu_co_mutex_unlock(&s->lock); |
b3cf1c7c AG |
771 | /* First we read the existing data from both COW regions. We |
772 | * either read the whole region in one go, or the start and end | |
773 | * regions separately. */ | |
774 | if (merge_reads) { | |
86b862c4 AG |
775 | qemu_iovec_add(&qiov, start_buffer, buffer_size); |
776 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c | 777 | } else { |
86b862c4 AG |
778 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); |
779 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c AG |
780 | if (ret < 0) { |
781 | goto fail; | |
782 | } | |
672f0f2c | 783 | |
86b862c4 AG |
784 | qemu_iovec_reset(&qiov); |
785 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
786 | ret = do_perform_cow_read(bs, m->offset, end->offset, &qiov); | |
b3cf1c7c | 787 | } |
593fb83c | 788 | if (ret < 0) { |
99450c6f | 789 | goto fail; |
593fb83c KW |
790 | } |
791 | ||
672f0f2c AG |
792 | /* Encrypt the data if necessary before writing it */ |
793 | if (bs->encrypted) { | |
4652b8f3 DB |
794 | if (!do_perform_cow_encrypt(bs, m->offset, m->alloc_offset, |
795 | start->offset, start_buffer, | |
796 | start->nb_bytes) || | |
797 | !do_perform_cow_encrypt(bs, m->offset, m->alloc_offset, | |
798 | end->offset, end_buffer, end->nb_bytes)) { | |
672f0f2c AG |
799 | ret = -EIO; |
800 | goto fail; | |
801 | } | |
802 | } | |
803 | ||
ee22a9d8 AG |
804 | /* And now we can write everything. If we have the guest data we |
805 | * can write everything in one single operation */ | |
806 | if (m->data_qiov) { | |
807 | qemu_iovec_reset(&qiov); | |
808 | if (start->nb_bytes) { | |
809 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
810 | } | |
811 | qemu_iovec_concat(&qiov, m->data_qiov, 0, data_bytes); | |
812 | if (end->nb_bytes) { | |
813 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
814 | } | |
815 | /* NOTE: we have a write_aio blkdebug event here followed by | |
816 | * a cow_write one in do_perform_cow_write(), but there's only | |
817 | * one single I/O operation */ | |
818 | BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); | |
819 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); | |
820 | } else { | |
821 | /* If there's no guest data then write both COW regions separately */ | |
822 | qemu_iovec_reset(&qiov); | |
823 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
824 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); | |
825 | if (ret < 0) { | |
826 | goto fail; | |
827 | } | |
828 | ||
829 | qemu_iovec_reset(&qiov); | |
830 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
831 | ret = do_perform_cow_write(bs, m->alloc_offset, end->offset, &qiov); | |
672f0f2c | 832 | } |
99450c6f AG |
833 | |
834 | fail: | |
835 | qemu_co_mutex_lock(&s->lock); | |
836 | ||
593fb83c KW |
837 | /* |
838 | * Before we update the L2 table to actually point to the new cluster, we | |
839 | * need to be sure that the refcounts have been increased and COW was | |
840 | * handled. | |
841 | */ | |
99450c6f AG |
842 | if (ret == 0) { |
843 | qcow2_cache_depends_on_flush(s->l2_table_cache); | |
844 | } | |
593fb83c | 845 | |
672f0f2c | 846 | qemu_vfree(start_buffer); |
86b862c4 | 847 | qemu_iovec_destroy(&qiov); |
99450c6f | 848 | return ret; |
593fb83c KW |
849 | } |
850 | ||
148da7ea | 851 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) |
45aba42f | 852 | { |
ff99129a | 853 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 854 | int i, j = 0, l2_index, ret; |
593fb83c | 855 | uint64_t *old_cluster, *l2_table; |
250196f1 | 856 | uint64_t cluster_offset = m->alloc_offset; |
45aba42f | 857 | |
3cce16f4 | 858 | trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
f50f88b9 | 859 | assert(m->nb_clusters > 0); |
45aba42f | 860 | |
5839e53b | 861 | old_cluster = g_try_new(uint64_t, m->nb_clusters); |
de82815d KW |
862 | if (old_cluster == NULL) { |
863 | ret = -ENOMEM; | |
864 | goto err; | |
865 | } | |
45aba42f KW |
866 | |
867 | /* copy content of unmodified sectors */ | |
99450c6f | 868 | ret = perform_cow(bs, m); |
593fb83c KW |
869 | if (ret < 0) { |
870 | goto err; | |
29c1a730 KW |
871 | } |
872 | ||
593fb83c | 873 | /* Update L2 table. */ |
74c4510a | 874 | if (s->use_lazy_refcounts) { |
280d3735 KW |
875 | qcow2_mark_dirty(bs); |
876 | } | |
bfe8043e SH |
877 | if (qcow2_need_accurate_refcounts(s)) { |
878 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
879 | s->refcount_block_cache); | |
880 | } | |
280d3735 | 881 | |
3948d1d4 | 882 | ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index); |
1e3e8f1a | 883 | if (ret < 0) { |
45aba42f | 884 | goto err; |
1e3e8f1a | 885 | } |
72e80b89 | 886 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
45aba42f | 887 | |
c01dbccb | 888 | assert(l2_index + m->nb_clusters <= s->l2_size); |
45aba42f KW |
889 | for (i = 0; i < m->nb_clusters; i++) { |
890 | /* if two concurrent writes happen to the same unallocated cluster | |
aaa4d20b KW |
891 | * each write allocates separate cluster and writes data concurrently. |
892 | * The first one to complete updates l2 table with pointer to its | |
893 | * cluster the second one has to do RMW (which is done above by | |
894 | * perform_cow()), update l2 table with its cluster pointer and free | |
895 | * old cluster. This is what this loop does */ | |
896 | if (l2_table[l2_index + i] != 0) { | |
45aba42f | 897 | old_cluster[j++] = l2_table[l2_index + i]; |
aaa4d20b | 898 | } |
45aba42f KW |
899 | |
900 | l2_table[l2_index + i] = cpu_to_be64((cluster_offset + | |
901 | (i << s->cluster_bits)) | QCOW_OFLAG_COPIED); | |
902 | } | |
903 | ||
9f8e668e | 904 | |
a3f1afb4 | 905 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
45aba42f | 906 | |
7ec5e6a4 KW |
907 | /* |
908 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
6cfcb9b8 KW |
909 | * |
910 | * Don't discard clusters that reach a refcount of 0 (e.g. compressed | |
911 | * clusters), the next write will reuse them anyway. | |
7ec5e6a4 | 912 | */ |
564a6b69 | 913 | if (!m->keep_old_clusters && j != 0) { |
7ec5e6a4 | 914 | for (i = 0; i < j; i++) { |
6cfcb9b8 KW |
915 | qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1, |
916 | QCOW2_DISCARD_NEVER); | |
7ec5e6a4 KW |
917 | } |
918 | } | |
45aba42f KW |
919 | |
920 | ret = 0; | |
921 | err: | |
7267c094 | 922 | g_free(old_cluster); |
45aba42f KW |
923 | return ret; |
924 | } | |
925 | ||
bf319ece KW |
926 | /* |
927 | * Returns the number of contiguous clusters that can be used for an allocating | |
928 | * write, but require COW to be performed (this includes yet unallocated space, | |
929 | * which must copy from the backing file) | |
930 | */ | |
ff99129a | 931 | static int count_cow_clusters(BDRVQcow2State *s, int nb_clusters, |
bf319ece KW |
932 | uint64_t *l2_table, int l2_index) |
933 | { | |
143550a8 | 934 | int i; |
bf319ece | 935 | |
143550a8 KW |
936 | for (i = 0; i < nb_clusters; i++) { |
937 | uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]); | |
3ef95218 | 938 | QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry); |
143550a8 KW |
939 | |
940 | switch(cluster_type) { | |
941 | case QCOW2_CLUSTER_NORMAL: | |
942 | if (l2_entry & QCOW_OFLAG_COPIED) { | |
943 | goto out; | |
944 | } | |
bf319ece | 945 | break; |
143550a8 KW |
946 | case QCOW2_CLUSTER_UNALLOCATED: |
947 | case QCOW2_CLUSTER_COMPRESSED: | |
fdfab37d EB |
948 | case QCOW2_CLUSTER_ZERO_PLAIN: |
949 | case QCOW2_CLUSTER_ZERO_ALLOC: | |
bf319ece | 950 | break; |
143550a8 KW |
951 | default: |
952 | abort(); | |
953 | } | |
bf319ece KW |
954 | } |
955 | ||
143550a8 | 956 | out: |
bf319ece KW |
957 | assert(i <= nb_clusters); |
958 | return i; | |
959 | } | |
960 | ||
250196f1 | 961 | /* |
226c3c26 KW |
962 | * Check if there already is an AIO write request in flight which allocates |
963 | * the same cluster. In this case we need to wait until the previous | |
964 | * request has completed and updated the L2 table accordingly. | |
65eb2e35 KW |
965 | * |
966 | * Returns: | |
967 | * 0 if there was no dependency. *cur_bytes indicates the number of | |
968 | * bytes from guest_offset that can be read before the next | |
969 | * dependency must be processed (or the request is complete) | |
970 | * | |
971 | * -EAGAIN if we had to wait for another request, previously gathered | |
972 | * information on cluster allocation may be invalid now. The caller | |
973 | * must start over anyway, so consider *cur_bytes undefined. | |
250196f1 | 974 | */ |
226c3c26 | 975 | static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, |
ecdd5333 | 976 | uint64_t *cur_bytes, QCowL2Meta **m) |
250196f1 | 977 | { |
ff99129a | 978 | BDRVQcow2State *s = bs->opaque; |
250196f1 | 979 | QCowL2Meta *old_alloc; |
65eb2e35 | 980 | uint64_t bytes = *cur_bytes; |
250196f1 | 981 | |
250196f1 KW |
982 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
983 | ||
65eb2e35 KW |
984 | uint64_t start = guest_offset; |
985 | uint64_t end = start + bytes; | |
986 | uint64_t old_start = l2meta_cow_start(old_alloc); | |
987 | uint64_t old_end = l2meta_cow_end(old_alloc); | |
250196f1 | 988 | |
d9d74f41 | 989 | if (end <= old_start || start >= old_end) { |
250196f1 KW |
990 | /* No intersection */ |
991 | } else { | |
992 | if (start < old_start) { | |
993 | /* Stop at the start of a running allocation */ | |
65eb2e35 | 994 | bytes = old_start - start; |
250196f1 | 995 | } else { |
65eb2e35 | 996 | bytes = 0; |
250196f1 KW |
997 | } |
998 | ||
ecdd5333 KW |
999 | /* Stop if already an l2meta exists. After yielding, it wouldn't |
1000 | * be valid any more, so we'd have to clean up the old L2Metas | |
1001 | * and deal with requests depending on them before starting to | |
1002 | * gather new ones. Not worth the trouble. */ | |
1003 | if (bytes == 0 && *m) { | |
1004 | *cur_bytes = 0; | |
1005 | return 0; | |
1006 | } | |
1007 | ||
65eb2e35 | 1008 | if (bytes == 0) { |
250196f1 KW |
1009 | /* Wait for the dependency to complete. We need to recheck |
1010 | * the free/allocated clusters when we continue. */ | |
1ace7cea | 1011 | qemu_co_queue_wait(&old_alloc->dependent_requests, &s->lock); |
250196f1 KW |
1012 | return -EAGAIN; |
1013 | } | |
1014 | } | |
1015 | } | |
1016 | ||
65eb2e35 KW |
1017 | /* Make sure that existing clusters and new allocations are only used up to |
1018 | * the next dependency if we shortened the request above */ | |
1019 | *cur_bytes = bytes; | |
250196f1 | 1020 | |
226c3c26 KW |
1021 | return 0; |
1022 | } | |
1023 | ||
0af729ec KW |
1024 | /* |
1025 | * Checks how many already allocated clusters that don't require a copy on | |
1026 | * write there are at the given guest_offset (up to *bytes). If | |
1027 | * *host_offset is not zero, only physically contiguous clusters beginning at | |
1028 | * this host offset are counted. | |
1029 | * | |
411d62b0 KW |
1030 | * Note that guest_offset may not be cluster aligned. In this case, the |
1031 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1032 | * therefore isn't cluster aligned as well. | |
0af729ec KW |
1033 | * |
1034 | * Returns: | |
1035 | * 0: if no allocated clusters are available at the given offset. | |
1036 | * *bytes is normally unchanged. It is set to 0 if the cluster | |
1037 | * is allocated and doesn't need COW, but doesn't have the right | |
1038 | * physical offset. | |
1039 | * | |
1040 | * 1: if allocated clusters that don't require a COW are available at | |
1041 | * the requested offset. *bytes may have decreased and describes | |
1042 | * the length of the area that can be written to. | |
1043 | * | |
1044 | * -errno: in error cases | |
0af729ec KW |
1045 | */ |
1046 | static int handle_copied(BlockDriverState *bs, uint64_t guest_offset, | |
c53ede9f | 1047 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
0af729ec | 1048 | { |
ff99129a | 1049 | BDRVQcow2State *s = bs->opaque; |
0af729ec KW |
1050 | int l2_index; |
1051 | uint64_t cluster_offset; | |
1052 | uint64_t *l2_table; | |
b6d36def | 1053 | uint64_t nb_clusters; |
c53ede9f | 1054 | unsigned int keep_clusters; |
a3f1afb4 | 1055 | int ret; |
0af729ec KW |
1056 | |
1057 | trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset, | |
1058 | *bytes); | |
0af729ec | 1059 | |
411d62b0 KW |
1060 | assert(*host_offset == 0 || offset_into_cluster(s, guest_offset) |
1061 | == offset_into_cluster(s, *host_offset)); | |
1062 | ||
acb0467f KW |
1063 | /* |
1064 | * Calculate the number of clusters to look for. We stop at L2 table | |
1065 | * boundaries to keep things simple. | |
1066 | */ | |
1067 | nb_clusters = | |
1068 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1069 | ||
1070 | l2_index = offset_to_l2_index(s, guest_offset); | |
1071 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1072 | assert(nb_clusters <= INT_MAX); |
acb0467f | 1073 | |
0af729ec KW |
1074 | /* Find L2 entry for the first involved cluster */ |
1075 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
1076 | if (ret < 0) { | |
1077 | return ret; | |
1078 | } | |
1079 | ||
1080 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
1081 | ||
1082 | /* Check how many clusters are already allocated and don't need COW */ | |
1083 | if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL | |
1084 | && (cluster_offset & QCOW_OFLAG_COPIED)) | |
1085 | { | |
e62daaf6 KW |
1086 | /* If a specific host_offset is required, check it */ |
1087 | bool offset_matches = | |
1088 | (cluster_offset & L2E_OFFSET_MASK) == *host_offset; | |
1089 | ||
a97c67ee HR |
1090 | if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) { |
1091 | qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset " | |
1092 | "%#llx unaligned (guest offset: %#" PRIx64 | |
1093 | ")", cluster_offset & L2E_OFFSET_MASK, | |
1094 | guest_offset); | |
1095 | ret = -EIO; | |
1096 | goto out; | |
1097 | } | |
1098 | ||
e62daaf6 KW |
1099 | if (*host_offset != 0 && !offset_matches) { |
1100 | *bytes = 0; | |
1101 | ret = 0; | |
1102 | goto out; | |
1103 | } | |
1104 | ||
0af729ec | 1105 | /* We keep all QCOW_OFLAG_COPIED clusters */ |
c53ede9f | 1106 | keep_clusters = |
acb0467f | 1107 | count_contiguous_clusters(nb_clusters, s->cluster_size, |
61653008 | 1108 | &l2_table[l2_index], |
0af729ec | 1109 | QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); |
c53ede9f KW |
1110 | assert(keep_clusters <= nb_clusters); |
1111 | ||
1112 | *bytes = MIN(*bytes, | |
1113 | keep_clusters * s->cluster_size | |
1114 | - offset_into_cluster(s, guest_offset)); | |
0af729ec KW |
1115 | |
1116 | ret = 1; | |
1117 | } else { | |
0af729ec KW |
1118 | ret = 0; |
1119 | } | |
1120 | ||
0af729ec | 1121 | /* Cleanup */ |
e62daaf6 | 1122 | out: |
a3f1afb4 | 1123 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
0af729ec | 1124 | |
e62daaf6 KW |
1125 | /* Only return a host offset if we actually made progress. Otherwise we |
1126 | * would make requirements for handle_alloc() that it can't fulfill */ | |
a97c67ee | 1127 | if (ret > 0) { |
411d62b0 KW |
1128 | *host_offset = (cluster_offset & L2E_OFFSET_MASK) |
1129 | + offset_into_cluster(s, guest_offset); | |
e62daaf6 KW |
1130 | } |
1131 | ||
0af729ec KW |
1132 | return ret; |
1133 | } | |
1134 | ||
226c3c26 KW |
1135 | /* |
1136 | * Allocates new clusters for the given guest_offset. | |
1137 | * | |
1138 | * At most *nb_clusters are allocated, and on return *nb_clusters is updated to | |
1139 | * contain the number of clusters that have been allocated and are contiguous | |
1140 | * in the image file. | |
1141 | * | |
1142 | * If *host_offset is non-zero, it specifies the offset in the image file at | |
1143 | * which the new clusters must start. *nb_clusters can be 0 on return in this | |
1144 | * case if the cluster at host_offset is already in use. If *host_offset is | |
1145 | * zero, the clusters can be allocated anywhere in the image file. | |
1146 | * | |
1147 | * *host_offset is updated to contain the offset into the image file at which | |
1148 | * the first allocated cluster starts. | |
1149 | * | |
1150 | * Return 0 on success and -errno in error cases. -EAGAIN means that the | |
1151 | * function has been waiting for another request and the allocation must be | |
1152 | * restarted, but the whole request should not be failed. | |
1153 | */ | |
1154 | static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, | |
b6d36def | 1155 | uint64_t *host_offset, uint64_t *nb_clusters) |
226c3c26 | 1156 | { |
ff99129a | 1157 | BDRVQcow2State *s = bs->opaque; |
226c3c26 KW |
1158 | |
1159 | trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, | |
1160 | *host_offset, *nb_clusters); | |
1161 | ||
250196f1 KW |
1162 | /* Allocate new clusters */ |
1163 | trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); | |
1164 | if (*host_offset == 0) { | |
df021791 KW |
1165 | int64_t cluster_offset = |
1166 | qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); | |
1167 | if (cluster_offset < 0) { | |
1168 | return cluster_offset; | |
1169 | } | |
1170 | *host_offset = cluster_offset; | |
1171 | return 0; | |
250196f1 | 1172 | } else { |
b6d36def | 1173 | int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); |
df021791 KW |
1174 | if (ret < 0) { |
1175 | return ret; | |
1176 | } | |
1177 | *nb_clusters = ret; | |
1178 | return 0; | |
250196f1 | 1179 | } |
250196f1 KW |
1180 | } |
1181 | ||
10f0ed8b KW |
1182 | /* |
1183 | * Allocates new clusters for an area that either is yet unallocated or needs a | |
1184 | * copy on write. If *host_offset is non-zero, clusters are only allocated if | |
1185 | * the new allocation can match the specified host offset. | |
1186 | * | |
411d62b0 KW |
1187 | * Note that guest_offset may not be cluster aligned. In this case, the |
1188 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1189 | * therefore isn't cluster aligned as well. | |
10f0ed8b KW |
1190 | * |
1191 | * Returns: | |
1192 | * 0: if no clusters could be allocated. *bytes is set to 0, | |
1193 | * *host_offset is left unchanged. | |
1194 | * | |
1195 | * 1: if new clusters were allocated. *bytes may be decreased if the | |
1196 | * new allocation doesn't cover all of the requested area. | |
1197 | * *host_offset is updated to contain the host offset of the first | |
1198 | * newly allocated cluster. | |
1199 | * | |
1200 | * -errno: in error cases | |
10f0ed8b KW |
1201 | */ |
1202 | static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, | |
c37f4cd7 | 1203 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
10f0ed8b | 1204 | { |
ff99129a | 1205 | BDRVQcow2State *s = bs->opaque; |
10f0ed8b KW |
1206 | int l2_index; |
1207 | uint64_t *l2_table; | |
1208 | uint64_t entry; | |
b6d36def | 1209 | uint64_t nb_clusters; |
10f0ed8b | 1210 | int ret; |
564a6b69 | 1211 | bool keep_old_clusters = false; |
10f0ed8b | 1212 | |
564a6b69 | 1213 | uint64_t alloc_cluster_offset = 0; |
10f0ed8b KW |
1214 | |
1215 | trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset, | |
1216 | *bytes); | |
1217 | assert(*bytes > 0); | |
1218 | ||
f5bc6350 KW |
1219 | /* |
1220 | * Calculate the number of clusters to look for. We stop at L2 table | |
1221 | * boundaries to keep things simple. | |
1222 | */ | |
c37f4cd7 KW |
1223 | nb_clusters = |
1224 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1225 | ||
f5bc6350 | 1226 | l2_index = offset_to_l2_index(s, guest_offset); |
c37f4cd7 | 1227 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
b6d36def | 1228 | assert(nb_clusters <= INT_MAX); |
f5bc6350 | 1229 | |
10f0ed8b KW |
1230 | /* Find L2 entry for the first involved cluster */ |
1231 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
1232 | if (ret < 0) { | |
1233 | return ret; | |
1234 | } | |
1235 | ||
3b8e2e26 | 1236 | entry = be64_to_cpu(l2_table[l2_index]); |
10f0ed8b KW |
1237 | |
1238 | /* For the moment, overwrite compressed clusters one by one */ | |
1239 | if (entry & QCOW_OFLAG_COMPRESSED) { | |
1240 | nb_clusters = 1; | |
1241 | } else { | |
3b8e2e26 | 1242 | nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index); |
10f0ed8b KW |
1243 | } |
1244 | ||
ecdd5333 KW |
1245 | /* This function is only called when there were no non-COW clusters, so if |
1246 | * we can't find any unallocated or COW clusters either, something is | |
1247 | * wrong with our code. */ | |
1248 | assert(nb_clusters > 0); | |
1249 | ||
fdfab37d EB |
1250 | if (qcow2_get_cluster_type(entry) == QCOW2_CLUSTER_ZERO_ALLOC && |
1251 | (entry & QCOW_OFLAG_COPIED) && | |
564a6b69 HR |
1252 | (!*host_offset || |
1253 | start_of_cluster(s, *host_offset) == (entry & L2E_OFFSET_MASK))) | |
1254 | { | |
1255 | /* Try to reuse preallocated zero clusters; contiguous normal clusters | |
1256 | * would be fine, too, but count_cow_clusters() above has limited | |
1257 | * nb_clusters already to a range of COW clusters */ | |
1258 | int preallocated_nb_clusters = | |
1259 | count_contiguous_clusters(nb_clusters, s->cluster_size, | |
1260 | &l2_table[l2_index], QCOW_OFLAG_COPIED); | |
1261 | assert(preallocated_nb_clusters > 0); | |
10f0ed8b | 1262 | |
564a6b69 HR |
1263 | nb_clusters = preallocated_nb_clusters; |
1264 | alloc_cluster_offset = entry & L2E_OFFSET_MASK; | |
10f0ed8b | 1265 | |
564a6b69 HR |
1266 | /* We want to reuse these clusters, so qcow2_alloc_cluster_link_l2() |
1267 | * should not free them. */ | |
1268 | keep_old_clusters = true; | |
10f0ed8b KW |
1269 | } |
1270 | ||
564a6b69 HR |
1271 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
1272 | ||
ff52aab2 | 1273 | if (!alloc_cluster_offset) { |
564a6b69 HR |
1274 | /* Allocate, if necessary at a given offset in the image file */ |
1275 | alloc_cluster_offset = start_of_cluster(s, *host_offset); | |
1276 | ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset, | |
1277 | &nb_clusters); | |
1278 | if (ret < 0) { | |
1279 | goto fail; | |
1280 | } | |
1281 | ||
1282 | /* Can't extend contiguous allocation */ | |
1283 | if (nb_clusters == 0) { | |
1284 | *bytes = 0; | |
1285 | return 0; | |
1286 | } | |
1287 | ||
1288 | /* !*host_offset would overwrite the image header and is reserved for | |
1289 | * "no host offset preferred". If 0 was a valid host offset, it'd | |
1290 | * trigger the following overlap check; do that now to avoid having an | |
1291 | * invalid value in *host_offset. */ | |
1292 | if (!alloc_cluster_offset) { | |
1293 | ret = qcow2_pre_write_overlap_check(bs, 0, alloc_cluster_offset, | |
1294 | nb_clusters * s->cluster_size); | |
1295 | assert(ret < 0); | |
1296 | goto fail; | |
1297 | } | |
ff52aab2 HR |
1298 | } |
1299 | ||
83baa9a4 KW |
1300 | /* |
1301 | * Save info needed for meta data update. | |
1302 | * | |
85567393 | 1303 | * requested_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1304 | * newly allocated cluster to the end of the (possibly shortened |
1305 | * before) write request. | |
1306 | * | |
85567393 | 1307 | * avail_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1308 | * newly allocated to the end of the last newly allocated cluster. |
1309 | * | |
85567393 | 1310 | * nb_bytes: The number of bytes from the start of the first |
83baa9a4 KW |
1311 | * newly allocated cluster to the end of the area that the write |
1312 | * request actually writes to (excluding COW at the end) | |
1313 | */ | |
85567393 KW |
1314 | uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset); |
1315 | int avail_bytes = MIN(INT_MAX, nb_clusters << s->cluster_bits); | |
1316 | int nb_bytes = MIN(requested_bytes, avail_bytes); | |
88c6588c | 1317 | QCowL2Meta *old_m = *m; |
83baa9a4 | 1318 | |
83baa9a4 KW |
1319 | *m = g_malloc0(sizeof(**m)); |
1320 | ||
1321 | **m = (QCowL2Meta) { | |
88c6588c KW |
1322 | .next = old_m, |
1323 | ||
411d62b0 | 1324 | .alloc_offset = alloc_cluster_offset, |
83baa9a4 KW |
1325 | .offset = start_of_cluster(s, guest_offset), |
1326 | .nb_clusters = nb_clusters, | |
83baa9a4 | 1327 | |
564a6b69 HR |
1328 | .keep_old_clusters = keep_old_clusters, |
1329 | ||
83baa9a4 KW |
1330 | .cow_start = { |
1331 | .offset = 0, | |
85567393 | 1332 | .nb_bytes = offset_into_cluster(s, guest_offset), |
83baa9a4 KW |
1333 | }, |
1334 | .cow_end = { | |
85567393 KW |
1335 | .offset = nb_bytes, |
1336 | .nb_bytes = avail_bytes - nb_bytes, | |
83baa9a4 KW |
1337 | }, |
1338 | }; | |
1339 | qemu_co_queue_init(&(*m)->dependent_requests); | |
1340 | QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); | |
1341 | ||
411d62b0 | 1342 | *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset); |
85567393 | 1343 | *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset)); |
83baa9a4 KW |
1344 | assert(*bytes != 0); |
1345 | ||
10f0ed8b KW |
1346 | return 1; |
1347 | ||
1348 | fail: | |
1349 | if (*m && (*m)->nb_clusters > 0) { | |
1350 | QLIST_REMOVE(*m, next_in_flight); | |
1351 | } | |
1352 | return ret; | |
1353 | } | |
1354 | ||
45aba42f KW |
1355 | /* |
1356 | * alloc_cluster_offset | |
1357 | * | |
250196f1 KW |
1358 | * For a given offset on the virtual disk, find the cluster offset in qcow2 |
1359 | * file. If the offset is not found, allocate a new cluster. | |
45aba42f | 1360 | * |
250196f1 | 1361 | * If the cluster was already allocated, m->nb_clusters is set to 0 and |
a7912369 | 1362 | * other fields in m are meaningless. |
148da7ea KW |
1363 | * |
1364 | * If the cluster is newly allocated, m->nb_clusters is set to the number of | |
68d100e9 KW |
1365 | * contiguous clusters that have been allocated. In this case, the other |
1366 | * fields of m are valid and contain information about the first allocated | |
1367 | * cluster. | |
45aba42f | 1368 | * |
68d100e9 KW |
1369 | * If the request conflicts with another write request in flight, the coroutine |
1370 | * is queued and will be reentered when the dependency has completed. | |
148da7ea KW |
1371 | * |
1372 | * Return 0 on success and -errno in error cases | |
45aba42f | 1373 | */ |
f4f0d391 | 1374 | int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, |
d46a0bb2 KW |
1375 | unsigned int *bytes, uint64_t *host_offset, |
1376 | QCowL2Meta **m) | |
45aba42f | 1377 | { |
ff99129a | 1378 | BDRVQcow2State *s = bs->opaque; |
710c2496 | 1379 | uint64_t start, remaining; |
250196f1 | 1380 | uint64_t cluster_offset; |
65eb2e35 | 1381 | uint64_t cur_bytes; |
710c2496 | 1382 | int ret; |
45aba42f | 1383 | |
d46a0bb2 | 1384 | trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *bytes); |
710c2496 | 1385 | |
72424114 | 1386 | again: |
16f0587e | 1387 | start = offset; |
d46a0bb2 | 1388 | remaining = *bytes; |
0af729ec KW |
1389 | cluster_offset = 0; |
1390 | *host_offset = 0; | |
ecdd5333 KW |
1391 | cur_bytes = 0; |
1392 | *m = NULL; | |
0af729ec | 1393 | |
2c3b32d2 | 1394 | while (true) { |
ecdd5333 KW |
1395 | |
1396 | if (!*host_offset) { | |
1397 | *host_offset = start_of_cluster(s, cluster_offset); | |
1398 | } | |
1399 | ||
1400 | assert(remaining >= cur_bytes); | |
1401 | ||
1402 | start += cur_bytes; | |
1403 | remaining -= cur_bytes; | |
1404 | cluster_offset += cur_bytes; | |
1405 | ||
1406 | if (remaining == 0) { | |
1407 | break; | |
1408 | } | |
1409 | ||
1410 | cur_bytes = remaining; | |
1411 | ||
2c3b32d2 KW |
1412 | /* |
1413 | * Now start gathering as many contiguous clusters as possible: | |
1414 | * | |
1415 | * 1. Check for overlaps with in-flight allocations | |
1416 | * | |
1417 | * a) Overlap not in the first cluster -> shorten this request and | |
1418 | * let the caller handle the rest in its next loop iteration. | |
1419 | * | |
1420 | * b) Real overlaps of two requests. Yield and restart the search | |
1421 | * for contiguous clusters (the situation could have changed | |
1422 | * while we were sleeping) | |
1423 | * | |
1424 | * c) TODO: Request starts in the same cluster as the in-flight | |
1425 | * allocation ends. Shorten the COW of the in-fight allocation, | |
1426 | * set cluster_offset to write to the same cluster and set up | |
1427 | * the right synchronisation between the in-flight request and | |
1428 | * the new one. | |
1429 | */ | |
ecdd5333 | 1430 | ret = handle_dependencies(bs, start, &cur_bytes, m); |
2c3b32d2 | 1431 | if (ret == -EAGAIN) { |
ecdd5333 KW |
1432 | /* Currently handle_dependencies() doesn't yield if we already had |
1433 | * an allocation. If it did, we would have to clean up the L2Meta | |
1434 | * structs before starting over. */ | |
1435 | assert(*m == NULL); | |
2c3b32d2 KW |
1436 | goto again; |
1437 | } else if (ret < 0) { | |
1438 | return ret; | |
ecdd5333 KW |
1439 | } else if (cur_bytes == 0) { |
1440 | break; | |
2c3b32d2 KW |
1441 | } else { |
1442 | /* handle_dependencies() may have decreased cur_bytes (shortened | |
1443 | * the allocations below) so that the next dependency is processed | |
1444 | * correctly during the next loop iteration. */ | |
0af729ec | 1445 | } |
710c2496 | 1446 | |
2c3b32d2 KW |
1447 | /* |
1448 | * 2. Count contiguous COPIED clusters. | |
1449 | */ | |
1450 | ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m); | |
1451 | if (ret < 0) { | |
1452 | return ret; | |
1453 | } else if (ret) { | |
ecdd5333 | 1454 | continue; |
2c3b32d2 KW |
1455 | } else if (cur_bytes == 0) { |
1456 | break; | |
1457 | } | |
060bee89 | 1458 | |
2c3b32d2 KW |
1459 | /* |
1460 | * 3. If the request still hasn't completed, allocate new clusters, | |
1461 | * considering any cluster_offset of steps 1c or 2. | |
1462 | */ | |
1463 | ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m); | |
1464 | if (ret < 0) { | |
1465 | return ret; | |
1466 | } else if (ret) { | |
ecdd5333 | 1467 | continue; |
2c3b32d2 KW |
1468 | } else { |
1469 | assert(cur_bytes == 0); | |
1470 | break; | |
1471 | } | |
f5bc6350 | 1472 | } |
10f0ed8b | 1473 | |
d46a0bb2 KW |
1474 | *bytes -= remaining; |
1475 | assert(*bytes > 0); | |
710c2496 | 1476 | assert(*host_offset != 0); |
45aba42f | 1477 | |
148da7ea | 1478 | return 0; |
45aba42f KW |
1479 | } |
1480 | ||
1481 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
1482 | const uint8_t *buf, int buf_size) | |
1483 | { | |
1484 | z_stream strm1, *strm = &strm1; | |
1485 | int ret, out_len; | |
1486 | ||
1487 | memset(strm, 0, sizeof(*strm)); | |
1488 | ||
1489 | strm->next_in = (uint8_t *)buf; | |
1490 | strm->avail_in = buf_size; | |
1491 | strm->next_out = out_buf; | |
1492 | strm->avail_out = out_buf_size; | |
1493 | ||
1494 | ret = inflateInit2(strm, -12); | |
1495 | if (ret != Z_OK) | |
1496 | return -1; | |
1497 | ret = inflate(strm, Z_FINISH); | |
1498 | out_len = strm->next_out - out_buf; | |
1499 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
1500 | out_len != out_buf_size) { | |
1501 | inflateEnd(strm); | |
1502 | return -1; | |
1503 | } | |
1504 | inflateEnd(strm); | |
1505 | return 0; | |
1506 | } | |
1507 | ||
66f82cee | 1508 | int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) |
45aba42f | 1509 | { |
ff99129a | 1510 | BDRVQcow2State *s = bs->opaque; |
45aba42f KW |
1511 | int ret, csize, nb_csectors, sector_offset; |
1512 | uint64_t coffset; | |
1513 | ||
1514 | coffset = cluster_offset & s->cluster_offset_mask; | |
1515 | if (s->cluster_cache_offset != coffset) { | |
1516 | nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1; | |
1517 | sector_offset = coffset & 511; | |
1518 | csize = nb_csectors * 512 - sector_offset; | |
66f82cee | 1519 | BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
fbcbbf4e | 1520 | ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, |
9a4f4c31 | 1521 | nb_csectors); |
45aba42f | 1522 | if (ret < 0) { |
8af36488 | 1523 | return ret; |
45aba42f KW |
1524 | } |
1525 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
1526 | s->cluster_data + sector_offset, csize) < 0) { | |
8af36488 | 1527 | return -EIO; |
45aba42f KW |
1528 | } |
1529 | s->cluster_cache_offset = coffset; | |
1530 | } | |
1531 | return 0; | |
1532 | } | |
5ea929e3 KW |
1533 | |
1534 | /* | |
1535 | * This discards as many clusters of nb_clusters as possible at once (i.e. | |
1536 | * all clusters in the same L2 table) and returns the number of discarded | |
1537 | * clusters. | |
1538 | */ | |
1539 | static int discard_single_l2(BlockDriverState *bs, uint64_t offset, | |
b6d36def HR |
1540 | uint64_t nb_clusters, enum qcow2_discard_type type, |
1541 | bool full_discard) | |
5ea929e3 | 1542 | { |
ff99129a | 1543 | BDRVQcow2State *s = bs->opaque; |
3948d1d4 | 1544 | uint64_t *l2_table; |
5ea929e3 KW |
1545 | int l2_index; |
1546 | int ret; | |
1547 | int i; | |
1548 | ||
3948d1d4 | 1549 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
5ea929e3 KW |
1550 | if (ret < 0) { |
1551 | return ret; | |
1552 | } | |
1553 | ||
1554 | /* Limit nb_clusters to one L2 table */ | |
1555 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1556 | assert(nb_clusters <= INT_MAX); |
5ea929e3 KW |
1557 | |
1558 | for (i = 0; i < nb_clusters; i++) { | |
c883db0d | 1559 | uint64_t old_l2_entry; |
5ea929e3 | 1560 | |
c883db0d | 1561 | old_l2_entry = be64_to_cpu(l2_table[l2_index + i]); |
a71835a0 KW |
1562 | |
1563 | /* | |
808c4b6f HR |
1564 | * If full_discard is false, make sure that a discarded area reads back |
1565 | * as zeroes for v3 images (we cannot do it for v2 without actually | |
1566 | * writing a zero-filled buffer). We can skip the operation if the | |
1567 | * cluster is already marked as zero, or if it's unallocated and we | |
1568 | * don't have a backing file. | |
a71835a0 KW |
1569 | * |
1570 | * TODO We might want to use bdrv_get_block_status(bs) here, but we're | |
1571 | * holding s->lock, so that doesn't work today. | |
808c4b6f HR |
1572 | * |
1573 | * If full_discard is true, the sector should not read back as zeroes, | |
1574 | * but rather fall through to the backing file. | |
a71835a0 | 1575 | */ |
c883db0d | 1576 | switch (qcow2_get_cluster_type(old_l2_entry)) { |
bbd995d8 EB |
1577 | case QCOW2_CLUSTER_UNALLOCATED: |
1578 | if (full_discard || !bs->backing) { | |
1579 | continue; | |
1580 | } | |
1581 | break; | |
1582 | ||
fdfab37d EB |
1583 | case QCOW2_CLUSTER_ZERO_PLAIN: |
1584 | if (!full_discard) { | |
bbd995d8 EB |
1585 | continue; |
1586 | } | |
1587 | break; | |
1588 | ||
fdfab37d | 1589 | case QCOW2_CLUSTER_ZERO_ALLOC: |
bbd995d8 EB |
1590 | case QCOW2_CLUSTER_NORMAL: |
1591 | case QCOW2_CLUSTER_COMPRESSED: | |
1592 | break; | |
1593 | ||
1594 | default: | |
1595 | abort(); | |
5ea929e3 KW |
1596 | } |
1597 | ||
1598 | /* First remove L2 entries */ | |
72e80b89 | 1599 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
808c4b6f | 1600 | if (!full_discard && s->qcow_version >= 3) { |
a71835a0 KW |
1601 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
1602 | } else { | |
1603 | l2_table[l2_index + i] = cpu_to_be64(0); | |
1604 | } | |
5ea929e3 KW |
1605 | |
1606 | /* Then decrease the refcount */ | |
c883db0d | 1607 | qcow2_free_any_clusters(bs, old_l2_entry, 1, type); |
5ea929e3 KW |
1608 | } |
1609 | ||
a3f1afb4 | 1610 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
5ea929e3 KW |
1611 | |
1612 | return nb_clusters; | |
1613 | } | |
1614 | ||
d2cb36af EB |
1615 | int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset, |
1616 | uint64_t bytes, enum qcow2_discard_type type, | |
1617 | bool full_discard) | |
5ea929e3 | 1618 | { |
ff99129a | 1619 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1620 | uint64_t end_offset = offset + bytes; |
b6d36def | 1621 | uint64_t nb_clusters; |
d2cb36af | 1622 | int64_t cleared; |
5ea929e3 KW |
1623 | int ret; |
1624 | ||
f10ee139 | 1625 | /* Caller must pass aligned values, except at image end */ |
0c1bd469 | 1626 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); |
f10ee139 EB |
1627 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || |
1628 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
5ea929e3 | 1629 | |
d2cb36af | 1630 | nb_clusters = size_to_clusters(s, bytes); |
5ea929e3 | 1631 | |
0b919fae KW |
1632 | s->cache_discards = true; |
1633 | ||
5ea929e3 KW |
1634 | /* Each L2 table is handled by its own loop iteration */ |
1635 | while (nb_clusters > 0) { | |
d2cb36af EB |
1636 | cleared = discard_single_l2(bs, offset, nb_clusters, type, |
1637 | full_discard); | |
1638 | if (cleared < 0) { | |
1639 | ret = cleared; | |
0b919fae | 1640 | goto fail; |
5ea929e3 KW |
1641 | } |
1642 | ||
d2cb36af EB |
1643 | nb_clusters -= cleared; |
1644 | offset += (cleared * s->cluster_size); | |
5ea929e3 KW |
1645 | } |
1646 | ||
0b919fae KW |
1647 | ret = 0; |
1648 | fail: | |
1649 | s->cache_discards = false; | |
1650 | qcow2_process_discards(bs, ret); | |
1651 | ||
1652 | return ret; | |
5ea929e3 | 1653 | } |
621f0589 KW |
1654 | |
1655 | /* | |
1656 | * This zeroes as many clusters of nb_clusters as possible at once (i.e. | |
1657 | * all clusters in the same L2 table) and returns the number of zeroed | |
1658 | * clusters. | |
1659 | */ | |
1660 | static int zero_single_l2(BlockDriverState *bs, uint64_t offset, | |
170f4b2e | 1661 | uint64_t nb_clusters, int flags) |
621f0589 | 1662 | { |
ff99129a | 1663 | BDRVQcow2State *s = bs->opaque; |
621f0589 KW |
1664 | uint64_t *l2_table; |
1665 | int l2_index; | |
1666 | int ret; | |
1667 | int i; | |
06cc5e2b | 1668 | bool unmap = !!(flags & BDRV_REQ_MAY_UNMAP); |
621f0589 KW |
1669 | |
1670 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); | |
1671 | if (ret < 0) { | |
1672 | return ret; | |
1673 | } | |
1674 | ||
1675 | /* Limit nb_clusters to one L2 table */ | |
1676 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1677 | assert(nb_clusters <= INT_MAX); |
621f0589 KW |
1678 | |
1679 | for (i = 0; i < nb_clusters; i++) { | |
1680 | uint64_t old_offset; | |
06cc5e2b | 1681 | QCow2ClusterType cluster_type; |
621f0589 KW |
1682 | |
1683 | old_offset = be64_to_cpu(l2_table[l2_index + i]); | |
1684 | ||
06cc5e2b EB |
1685 | /* |
1686 | * Minimize L2 changes if the cluster already reads back as | |
1687 | * zeroes with correct allocation. | |
1688 | */ | |
1689 | cluster_type = qcow2_get_cluster_type(old_offset); | |
1690 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN || | |
1691 | (cluster_type == QCOW2_CLUSTER_ZERO_ALLOC && !unmap)) { | |
1692 | continue; | |
1693 | } | |
1694 | ||
72e80b89 | 1695 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
06cc5e2b | 1696 | if (cluster_type == QCOW2_CLUSTER_COMPRESSED || unmap) { |
621f0589 | 1697 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
6cfcb9b8 | 1698 | qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST); |
621f0589 KW |
1699 | } else { |
1700 | l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO); | |
1701 | } | |
1702 | } | |
1703 | ||
a3f1afb4 | 1704 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
621f0589 KW |
1705 | |
1706 | return nb_clusters; | |
1707 | } | |
1708 | ||
d2cb36af EB |
1709 | int qcow2_cluster_zeroize(BlockDriverState *bs, uint64_t offset, |
1710 | uint64_t bytes, int flags) | |
621f0589 | 1711 | { |
ff99129a | 1712 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1713 | uint64_t end_offset = offset + bytes; |
b6d36def | 1714 | uint64_t nb_clusters; |
d2cb36af | 1715 | int64_t cleared; |
621f0589 KW |
1716 | int ret; |
1717 | ||
f10ee139 EB |
1718 | /* Caller must pass aligned values, except at image end */ |
1719 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); | |
1720 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || | |
1721 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
1722 | ||
621f0589 KW |
1723 | /* The zero flag is only supported by version 3 and newer */ |
1724 | if (s->qcow_version < 3) { | |
1725 | return -ENOTSUP; | |
1726 | } | |
1727 | ||
1728 | /* Each L2 table is handled by its own loop iteration */ | |
d2cb36af | 1729 | nb_clusters = size_to_clusters(s, bytes); |
621f0589 | 1730 | |
0b919fae KW |
1731 | s->cache_discards = true; |
1732 | ||
621f0589 | 1733 | while (nb_clusters > 0) { |
d2cb36af EB |
1734 | cleared = zero_single_l2(bs, offset, nb_clusters, flags); |
1735 | if (cleared < 0) { | |
1736 | ret = cleared; | |
0b919fae | 1737 | goto fail; |
621f0589 KW |
1738 | } |
1739 | ||
d2cb36af EB |
1740 | nb_clusters -= cleared; |
1741 | offset += (cleared * s->cluster_size); | |
621f0589 KW |
1742 | } |
1743 | ||
0b919fae KW |
1744 | ret = 0; |
1745 | fail: | |
1746 | s->cache_discards = false; | |
1747 | qcow2_process_discards(bs, ret); | |
1748 | ||
1749 | return ret; | |
621f0589 | 1750 | } |
32b6444d HR |
1751 | |
1752 | /* | |
1753 | * Expands all zero clusters in a specific L1 table (or deallocates them, for | |
1754 | * non-backed non-pre-allocated zero clusters). | |
1755 | * | |
4057a2b2 HR |
1756 | * l1_entries and *visited_l1_entries are used to keep track of progress for |
1757 | * status_cb(). l1_entries contains the total number of L1 entries and | |
1758 | * *visited_l1_entries counts all visited L1 entries. | |
32b6444d HR |
1759 | */ |
1760 | static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table, | |
ecf58777 | 1761 | int l1_size, int64_t *visited_l1_entries, |
4057a2b2 | 1762 | int64_t l1_entries, |
8b13976d HR |
1763 | BlockDriverAmendStatusCB *status_cb, |
1764 | void *cb_opaque) | |
32b6444d | 1765 | { |
ff99129a | 1766 | BDRVQcow2State *s = bs->opaque; |
32b6444d HR |
1767 | bool is_active_l1 = (l1_table == s->l1_table); |
1768 | uint64_t *l2_table = NULL; | |
1769 | int ret; | |
1770 | int i, j; | |
1771 | ||
1772 | if (!is_active_l1) { | |
1773 | /* inactive L2 tables require a buffer to be stored in when loading | |
1774 | * them from disk */ | |
9a4f4c31 | 1775 | l2_table = qemu_try_blockalign(bs->file->bs, s->cluster_size); |
de82815d KW |
1776 | if (l2_table == NULL) { |
1777 | return -ENOMEM; | |
1778 | } | |
32b6444d HR |
1779 | } |
1780 | ||
1781 | for (i = 0; i < l1_size; i++) { | |
1782 | uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK; | |
1783 | bool l2_dirty = false; | |
0e06528e | 1784 | uint64_t l2_refcount; |
32b6444d HR |
1785 | |
1786 | if (!l2_offset) { | |
1787 | /* unallocated */ | |
4057a2b2 HR |
1788 | (*visited_l1_entries)++; |
1789 | if (status_cb) { | |
8b13976d | 1790 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 1791 | } |
32b6444d HR |
1792 | continue; |
1793 | } | |
1794 | ||
8dd93d93 HR |
1795 | if (offset_into_cluster(s, l2_offset)) { |
1796 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" | |
1797 | PRIx64 " unaligned (L1 index: %#x)", | |
1798 | l2_offset, i); | |
1799 | ret = -EIO; | |
1800 | goto fail; | |
1801 | } | |
1802 | ||
32b6444d HR |
1803 | if (is_active_l1) { |
1804 | /* get active L2 tables from cache */ | |
1805 | ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, | |
1806 | (void **)&l2_table); | |
1807 | } else { | |
1808 | /* load inactive L2 tables from disk */ | |
fbcbbf4e | 1809 | ret = bdrv_read(bs->file, l2_offset / BDRV_SECTOR_SIZE, |
9a4f4c31 | 1810 | (void *)l2_table, s->cluster_sectors); |
32b6444d HR |
1811 | } |
1812 | if (ret < 0) { | |
1813 | goto fail; | |
1814 | } | |
1815 | ||
7324c10f HR |
1816 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
1817 | &l2_refcount); | |
1818 | if (ret < 0) { | |
ecf58777 HR |
1819 | goto fail; |
1820 | } | |
1821 | ||
32b6444d HR |
1822 | for (j = 0; j < s->l2_size; j++) { |
1823 | uint64_t l2_entry = be64_to_cpu(l2_table[j]); | |
ecf58777 | 1824 | int64_t offset = l2_entry & L2E_OFFSET_MASK; |
3ef95218 | 1825 | QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry); |
32b6444d | 1826 | |
fdfab37d EB |
1827 | if (cluster_type != QCOW2_CLUSTER_ZERO_PLAIN && |
1828 | cluster_type != QCOW2_CLUSTER_ZERO_ALLOC) { | |
32b6444d HR |
1829 | continue; |
1830 | } | |
1831 | ||
fdfab37d | 1832 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
760e0063 | 1833 | if (!bs->backing) { |
32b6444d HR |
1834 | /* not backed; therefore we can simply deallocate the |
1835 | * cluster */ | |
1836 | l2_table[j] = 0; | |
1837 | l2_dirty = true; | |
1838 | continue; | |
1839 | } | |
1840 | ||
1841 | offset = qcow2_alloc_clusters(bs, s->cluster_size); | |
1842 | if (offset < 0) { | |
1843 | ret = offset; | |
1844 | goto fail; | |
1845 | } | |
ecf58777 HR |
1846 | |
1847 | if (l2_refcount > 1) { | |
1848 | /* For shared L2 tables, set the refcount accordingly (it is | |
1849 | * already 1 and needs to be l2_refcount) */ | |
1850 | ret = qcow2_update_cluster_refcount(bs, | |
2aabe7c7 HR |
1851 | offset >> s->cluster_bits, |
1852 | refcount_diff(1, l2_refcount), false, | |
ecf58777 HR |
1853 | QCOW2_DISCARD_OTHER); |
1854 | if (ret < 0) { | |
1855 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1856 | QCOW2_DISCARD_OTHER); | |
1857 | goto fail; | |
1858 | } | |
1859 | } | |
32b6444d HR |
1860 | } |
1861 | ||
8dd93d93 | 1862 | if (offset_into_cluster(s, offset)) { |
bcb07dba EB |
1863 | qcow2_signal_corruption(bs, true, -1, -1, |
1864 | "Cluster allocation offset " | |
8dd93d93 HR |
1865 | "%#" PRIx64 " unaligned (L2 offset: %#" |
1866 | PRIx64 ", L2 index: %#x)", offset, | |
1867 | l2_offset, j); | |
fdfab37d | 1868 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
8dd93d93 HR |
1869 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1870 | QCOW2_DISCARD_ALWAYS); | |
1871 | } | |
1872 | ret = -EIO; | |
1873 | goto fail; | |
1874 | } | |
1875 | ||
231bb267 | 1876 | ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size); |
32b6444d | 1877 | if (ret < 0) { |
fdfab37d | 1878 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
320c7066 HR |
1879 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1880 | QCOW2_DISCARD_ALWAYS); | |
1881 | } | |
32b6444d HR |
1882 | goto fail; |
1883 | } | |
1884 | ||
720ff280 | 1885 | ret = bdrv_pwrite_zeroes(bs->file, offset, s->cluster_size, 0); |
32b6444d | 1886 | if (ret < 0) { |
fdfab37d | 1887 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
320c7066 HR |
1888 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1889 | QCOW2_DISCARD_ALWAYS); | |
1890 | } | |
32b6444d HR |
1891 | goto fail; |
1892 | } | |
1893 | ||
ecf58777 HR |
1894 | if (l2_refcount == 1) { |
1895 | l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED); | |
1896 | } else { | |
1897 | l2_table[j] = cpu_to_be64(offset); | |
e390cf5a | 1898 | } |
ecf58777 | 1899 | l2_dirty = true; |
32b6444d HR |
1900 | } |
1901 | ||
1902 | if (is_active_l1) { | |
1903 | if (l2_dirty) { | |
72e80b89 | 1904 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
32b6444d HR |
1905 | qcow2_cache_depends_on_flush(s->l2_table_cache); |
1906 | } | |
a3f1afb4 | 1907 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
32b6444d HR |
1908 | } else { |
1909 | if (l2_dirty) { | |
231bb267 HR |
1910 | ret = qcow2_pre_write_overlap_check(bs, |
1911 | QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, l2_offset, | |
32b6444d HR |
1912 | s->cluster_size); |
1913 | if (ret < 0) { | |
1914 | goto fail; | |
1915 | } | |
1916 | ||
18d51c4b | 1917 | ret = bdrv_write(bs->file, l2_offset / BDRV_SECTOR_SIZE, |
9a4f4c31 | 1918 | (void *)l2_table, s->cluster_sectors); |
32b6444d HR |
1919 | if (ret < 0) { |
1920 | goto fail; | |
1921 | } | |
1922 | } | |
1923 | } | |
4057a2b2 HR |
1924 | |
1925 | (*visited_l1_entries)++; | |
1926 | if (status_cb) { | |
8b13976d | 1927 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 1928 | } |
32b6444d HR |
1929 | } |
1930 | ||
1931 | ret = 0; | |
1932 | ||
1933 | fail: | |
1934 | if (l2_table) { | |
1935 | if (!is_active_l1) { | |
1936 | qemu_vfree(l2_table); | |
1937 | } else { | |
a3f1afb4 | 1938 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
32b6444d HR |
1939 | } |
1940 | } | |
1941 | return ret; | |
1942 | } | |
1943 | ||
1944 | /* | |
1945 | * For backed images, expands all zero clusters on the image. For non-backed | |
1946 | * images, deallocates all non-pre-allocated zero clusters (and claims the | |
1947 | * allocation for pre-allocated ones). This is important for downgrading to a | |
1948 | * qcow2 version which doesn't yet support metadata zero clusters. | |
1949 | */ | |
4057a2b2 | 1950 | int qcow2_expand_zero_clusters(BlockDriverState *bs, |
8b13976d HR |
1951 | BlockDriverAmendStatusCB *status_cb, |
1952 | void *cb_opaque) | |
32b6444d | 1953 | { |
ff99129a | 1954 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 1955 | uint64_t *l1_table = NULL; |
4057a2b2 | 1956 | int64_t l1_entries = 0, visited_l1_entries = 0; |
32b6444d HR |
1957 | int ret; |
1958 | int i, j; | |
1959 | ||
4057a2b2 HR |
1960 | if (status_cb) { |
1961 | l1_entries = s->l1_size; | |
1962 | for (i = 0; i < s->nb_snapshots; i++) { | |
1963 | l1_entries += s->snapshots[i].l1_size; | |
1964 | } | |
1965 | } | |
1966 | ||
32b6444d | 1967 | ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size, |
4057a2b2 | 1968 | &visited_l1_entries, l1_entries, |
8b13976d | 1969 | status_cb, cb_opaque); |
32b6444d HR |
1970 | if (ret < 0) { |
1971 | goto fail; | |
1972 | } | |
1973 | ||
1974 | /* Inactive L1 tables may point to active L2 tables - therefore it is | |
1975 | * necessary to flush the L2 table cache before trying to access the L2 | |
1976 | * tables pointed to by inactive L1 entries (else we might try to expand | |
1977 | * zero clusters that have already been expanded); furthermore, it is also | |
1978 | * necessary to empty the L2 table cache, since it may contain tables which | |
1979 | * are now going to be modified directly on disk, bypassing the cache. | |
1980 | * qcow2_cache_empty() does both for us. */ | |
1981 | ret = qcow2_cache_empty(bs, s->l2_table_cache); | |
1982 | if (ret < 0) { | |
1983 | goto fail; | |
1984 | } | |
1985 | ||
1986 | for (i = 0; i < s->nb_snapshots; i++) { | |
d737b78c LV |
1987 | int l1_sectors = DIV_ROUND_UP(s->snapshots[i].l1_size * |
1988 | sizeof(uint64_t), BDRV_SECTOR_SIZE); | |
32b6444d HR |
1989 | |
1990 | l1_table = g_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE); | |
1991 | ||
fbcbbf4e | 1992 | ret = bdrv_read(bs->file, |
9a4f4c31 KW |
1993 | s->snapshots[i].l1_table_offset / BDRV_SECTOR_SIZE, |
1994 | (void *)l1_table, l1_sectors); | |
32b6444d HR |
1995 | if (ret < 0) { |
1996 | goto fail; | |
1997 | } | |
1998 | ||
1999 | for (j = 0; j < s->snapshots[i].l1_size; j++) { | |
2000 | be64_to_cpus(&l1_table[j]); | |
2001 | } | |
2002 | ||
2003 | ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size, | |
4057a2b2 | 2004 | &visited_l1_entries, l1_entries, |
8b13976d | 2005 | status_cb, cb_opaque); |
32b6444d HR |
2006 | if (ret < 0) { |
2007 | goto fail; | |
2008 | } | |
2009 | } | |
2010 | ||
2011 | ret = 0; | |
2012 | ||
2013 | fail: | |
32b6444d HR |
2014 | g_free(l1_table); |
2015 | return ret; | |
2016 | } |