]>
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 | ||
360 | /* The crypt function is compatible with the linux cryptoloop | |
361 | algorithm for < 4 GB images. NOTE: out_buf == in_buf is | |
362 | supported */ | |
ff99129a | 363 | int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num, |
f6fa64f6 DB |
364 | uint8_t *out_buf, const uint8_t *in_buf, |
365 | int nb_sectors, bool enc, | |
366 | Error **errp) | |
45aba42f KW |
367 | { |
368 | union { | |
369 | uint64_t ll[2]; | |
370 | uint8_t b[16]; | |
371 | } ivec; | |
372 | int i; | |
f6fa64f6 | 373 | int ret; |
45aba42f KW |
374 | |
375 | for(i = 0; i < nb_sectors; i++) { | |
376 | ivec.ll[0] = cpu_to_le64(sector_num); | |
377 | ivec.ll[1] = 0; | |
f6fa64f6 DB |
378 | if (qcrypto_cipher_setiv(s->cipher, |
379 | ivec.b, G_N_ELEMENTS(ivec.b), | |
380 | errp) < 0) { | |
381 | return -1; | |
382 | } | |
383 | if (enc) { | |
384 | ret = qcrypto_cipher_encrypt(s->cipher, | |
385 | in_buf, | |
386 | out_buf, | |
387 | 512, | |
388 | errp); | |
389 | } else { | |
390 | ret = qcrypto_cipher_decrypt(s->cipher, | |
391 | in_buf, | |
392 | out_buf, | |
393 | 512, | |
394 | errp); | |
395 | } | |
396 | if (ret < 0) { | |
397 | return -1; | |
398 | } | |
45aba42f KW |
399 | sector_num++; |
400 | in_buf += 512; | |
401 | out_buf += 512; | |
402 | } | |
f6fa64f6 | 403 | return 0; |
45aba42f KW |
404 | } |
405 | ||
aaa4d20b KW |
406 | static int coroutine_fn do_perform_cow(BlockDriverState *bs, |
407 | uint64_t src_cluster_offset, | |
408 | uint64_t cluster_offset, | |
e034f5bc AG |
409 | unsigned offset_in_cluster, |
410 | unsigned bytes) | |
45aba42f | 411 | { |
ff99129a | 412 | BDRVQcow2State *s = bs->opaque; |
aef4acb6 SH |
413 | QEMUIOVector qiov; |
414 | struct iovec iov; | |
aaa4d20b | 415 | int ret; |
1b9f1491 | 416 | |
99450c6f AG |
417 | if (bytes == 0) { |
418 | return 0; | |
419 | } | |
420 | ||
aaa4d20b | 421 | iov.iov_len = bytes; |
de82815d KW |
422 | iov.iov_base = qemu_try_blockalign(bs, iov.iov_len); |
423 | if (iov.iov_base == NULL) { | |
424 | return -ENOMEM; | |
425 | } | |
aef4acb6 SH |
426 | |
427 | qemu_iovec_init_external(&qiov, &iov, 1); | |
1b9f1491 | 428 | |
66f82cee | 429 | BLKDBG_EVENT(bs->file, BLKDBG_COW_READ); |
aef4acb6 | 430 | |
dba28555 | 431 | if (!bs->drv) { |
bd604369 KW |
432 | ret = -ENOMEDIUM; |
433 | goto out; | |
dba28555 HR |
434 | } |
435 | ||
aef4acb6 SH |
436 | /* Call .bdrv_co_readv() directly instead of using the public block-layer |
437 | * interface. This avoids double I/O throttling and request tracking, | |
438 | * which can lead to deadlock when block layer copy-on-read is enabled. | |
439 | */ | |
aaa4d20b KW |
440 | ret = bs->drv->bdrv_co_preadv(bs, src_cluster_offset + offset_in_cluster, |
441 | bytes, &qiov, 0); | |
1b9f1491 KW |
442 | if (ret < 0) { |
443 | goto out; | |
444 | } | |
445 | ||
8336aafa | 446 | if (bs->encrypted) { |
bb9f8dd0 | 447 | int64_t sector = (src_cluster_offset + offset_in_cluster) |
aaa4d20b | 448 | >> BDRV_SECTOR_BITS; |
f6fa64f6 | 449 | assert(s->cipher); |
aaa4d20b KW |
450 | assert((offset_in_cluster & ~BDRV_SECTOR_MASK) == 0); |
451 | assert((bytes & ~BDRV_SECTOR_MASK) == 0); | |
452 | if (qcow2_encrypt_sectors(s, sector, iov.iov_base, iov.iov_base, | |
026ac158 | 453 | bytes >> BDRV_SECTOR_BITS, true, NULL) < 0) { |
f6fa64f6 | 454 | ret = -EIO; |
f6fa64f6 DB |
455 | goto out; |
456 | } | |
45aba42f | 457 | } |
1b9f1491 | 458 | |
231bb267 | 459 | ret = qcow2_pre_write_overlap_check(bs, 0, |
aaa4d20b | 460 | cluster_offset + offset_in_cluster, bytes); |
cf93980e HR |
461 | if (ret < 0) { |
462 | goto out; | |
463 | } | |
464 | ||
66f82cee | 465 | BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE); |
a03ef88f | 466 | ret = bdrv_co_pwritev(bs->file, cluster_offset + offset_in_cluster, |
aaa4d20b | 467 | bytes, &qiov, 0); |
1b9f1491 KW |
468 | if (ret < 0) { |
469 | goto out; | |
470 | } | |
471 | ||
472 | ret = 0; | |
473 | out: | |
aef4acb6 | 474 | qemu_vfree(iov.iov_base); |
1b9f1491 | 475 | return ret; |
45aba42f KW |
476 | } |
477 | ||
478 | ||
479 | /* | |
480 | * get_cluster_offset | |
481 | * | |
ecfe1863 KW |
482 | * For a given offset of the virtual disk, find the cluster type and offset in |
483 | * the qcow2 file. The offset is stored in *cluster_offset. | |
45aba42f | 484 | * |
ecfe1863 KW |
485 | * On entry, *bytes is the maximum number of contiguous bytes starting at |
486 | * offset that we are interested in. | |
45aba42f | 487 | * |
ecfe1863 KW |
488 | * On exit, *bytes is the number of bytes starting at offset that have the same |
489 | * cluster type and (if applicable) are stored contiguously in the image file. | |
490 | * Compressed clusters are always returned one by one. | |
45aba42f | 491 | * |
68d000a3 KW |
492 | * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error |
493 | * cases. | |
45aba42f | 494 | */ |
1c46efaa | 495 | int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, |
ecfe1863 | 496 | unsigned int *bytes, uint64_t *cluster_offset) |
45aba42f | 497 | { |
ff99129a | 498 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
499 | unsigned int l2_index; |
500 | uint64_t l1_index, l2_offset, *l2_table; | |
45aba42f | 501 | int l1_bits, c; |
c834cba9 HR |
502 | unsigned int offset_in_cluster; |
503 | uint64_t bytes_available, bytes_needed, nb_clusters; | |
3ef95218 | 504 | QCow2ClusterType type; |
55c17e98 | 505 | int ret; |
45aba42f | 506 | |
b2f65d6b | 507 | offset_in_cluster = offset_into_cluster(s, offset); |
ecfe1863 | 508 | bytes_needed = (uint64_t) *bytes + offset_in_cluster; |
45aba42f | 509 | |
b2f65d6b | 510 | l1_bits = s->l2_bits + s->cluster_bits; |
45aba42f | 511 | |
b2f65d6b KW |
512 | /* compute how many bytes there are between the start of the cluster |
513 | * containing offset and the end of the l1 entry */ | |
514 | bytes_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1)) | |
515 | + offset_in_cluster; | |
45aba42f | 516 | |
b2f65d6b KW |
517 | if (bytes_needed > bytes_available) { |
518 | bytes_needed = bytes_available; | |
45aba42f KW |
519 | } |
520 | ||
1c46efaa | 521 | *cluster_offset = 0; |
45aba42f | 522 | |
b6af0975 | 523 | /* seek to the l2 offset in the l1 table */ |
45aba42f KW |
524 | |
525 | l1_index = offset >> l1_bits; | |
68d000a3 | 526 | if (l1_index >= s->l1_size) { |
3ef95218 | 527 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 528 | goto out; |
68d000a3 | 529 | } |
45aba42f | 530 | |
68d000a3 KW |
531 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
532 | if (!l2_offset) { | |
3ef95218 | 533 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 534 | goto out; |
68d000a3 | 535 | } |
45aba42f | 536 | |
a97c67ee HR |
537 | if (offset_into_cluster(s, l2_offset)) { |
538 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
539 | " unaligned (L1 index: %#" PRIx64 ")", | |
540 | l2_offset, l1_index); | |
541 | return -EIO; | |
542 | } | |
543 | ||
45aba42f KW |
544 | /* load the l2 table in memory */ |
545 | ||
55c17e98 KW |
546 | ret = l2_load(bs, l2_offset, &l2_table); |
547 | if (ret < 0) { | |
548 | return ret; | |
1c46efaa | 549 | } |
45aba42f KW |
550 | |
551 | /* find the cluster offset for the given disk offset */ | |
552 | ||
553 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
1c46efaa | 554 | *cluster_offset = be64_to_cpu(l2_table[l2_index]); |
b6d36def | 555 | |
b2f65d6b | 556 | nb_clusters = size_to_clusters(s, bytes_needed); |
c834cba9 HR |
557 | /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned |
558 | * integers; the minimum cluster size is 512, so this assertion is always | |
559 | * true */ | |
560 | assert(nb_clusters <= INT_MAX); | |
45aba42f | 561 | |
3ef95218 | 562 | type = qcow2_get_cluster_type(*cluster_offset); |
fdfab37d EB |
563 | if (s->qcow_version < 3 && (type == QCOW2_CLUSTER_ZERO_PLAIN || |
564 | type == QCOW2_CLUSTER_ZERO_ALLOC)) { | |
565 | qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found" | |
566 | " in pre-v3 image (L2 offset: %#" PRIx64 | |
567 | ", L2 index: %#x)", l2_offset, l2_index); | |
568 | ret = -EIO; | |
569 | goto fail; | |
570 | } | |
3ef95218 | 571 | switch (type) { |
68d000a3 KW |
572 | case QCOW2_CLUSTER_COMPRESSED: |
573 | /* Compressed clusters can only be processed one by one */ | |
574 | c = 1; | |
575 | *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK; | |
576 | break; | |
fdfab37d | 577 | case QCOW2_CLUSTER_ZERO_PLAIN: |
68d000a3 | 578 | case QCOW2_CLUSTER_UNALLOCATED: |
45aba42f | 579 | /* how many empty clusters ? */ |
4341df8a | 580 | c = count_contiguous_clusters_unallocated(nb_clusters, |
fdfab37d | 581 | &l2_table[l2_index], type); |
68d000a3 KW |
582 | *cluster_offset = 0; |
583 | break; | |
fdfab37d | 584 | case QCOW2_CLUSTER_ZERO_ALLOC: |
68d000a3 | 585 | case QCOW2_CLUSTER_NORMAL: |
45aba42f KW |
586 | /* how many allocated clusters ? */ |
587 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
fdfab37d | 588 | &l2_table[l2_index], QCOW_OFLAG_ZERO); |
68d000a3 | 589 | *cluster_offset &= L2E_OFFSET_MASK; |
a97c67ee | 590 | if (offset_into_cluster(s, *cluster_offset)) { |
fdfab37d EB |
591 | qcow2_signal_corruption(bs, true, -1, -1, |
592 | "Cluster allocation offset %#" | |
a97c67ee HR |
593 | PRIx64 " unaligned (L2 offset: %#" PRIx64 |
594 | ", L2 index: %#x)", *cluster_offset, | |
595 | l2_offset, l2_index); | |
596 | ret = -EIO; | |
597 | goto fail; | |
598 | } | |
68d000a3 | 599 | break; |
1417d7e4 KW |
600 | default: |
601 | abort(); | |
45aba42f KW |
602 | } |
603 | ||
29c1a730 KW |
604 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
605 | ||
c834cba9 | 606 | bytes_available = (int64_t)c * s->cluster_size; |
68d000a3 | 607 | |
45aba42f | 608 | out: |
b2f65d6b KW |
609 | if (bytes_available > bytes_needed) { |
610 | bytes_available = bytes_needed; | |
611 | } | |
45aba42f | 612 | |
c834cba9 HR |
613 | /* bytes_available <= bytes_needed <= *bytes + offset_in_cluster; |
614 | * subtracting offset_in_cluster will therefore definitely yield something | |
615 | * not exceeding UINT_MAX */ | |
616 | assert(bytes_available - offset_in_cluster <= UINT_MAX); | |
ecfe1863 | 617 | *bytes = bytes_available - offset_in_cluster; |
45aba42f | 618 | |
3ef95218 | 619 | return type; |
a97c67ee HR |
620 | |
621 | fail: | |
622 | qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table); | |
623 | return ret; | |
45aba42f KW |
624 | } |
625 | ||
626 | /* | |
627 | * get_cluster_table | |
628 | * | |
629 | * for a given disk offset, load (and allocate if needed) | |
630 | * the l2 table. | |
631 | * | |
632 | * the l2 table offset in the qcow2 file and the cluster index | |
633 | * in the l2 table are given to the caller. | |
634 | * | |
1e3e8f1a | 635 | * Returns 0 on success, -errno in failure case |
45aba42f | 636 | */ |
45aba42f KW |
637 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
638 | uint64_t **new_l2_table, | |
45aba42f KW |
639 | int *new_l2_index) |
640 | { | |
ff99129a | 641 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
642 | unsigned int l2_index; |
643 | uint64_t l1_index, l2_offset; | |
c46e1167 | 644 | uint64_t *l2_table = NULL; |
80ee15a6 | 645 | int ret; |
45aba42f | 646 | |
b6af0975 | 647 | /* seek to the l2 offset in the l1 table */ |
45aba42f KW |
648 | |
649 | l1_index = offset >> (s->l2_bits + s->cluster_bits); | |
650 | if (l1_index >= s->l1_size) { | |
72893756 | 651 | ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
1e3e8f1a KW |
652 | if (ret < 0) { |
653 | return ret; | |
654 | } | |
45aba42f | 655 | } |
8e37f681 | 656 | |
2cf7cfa1 | 657 | assert(l1_index < s->l1_size); |
8e37f681 | 658 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
a97c67ee HR |
659 | if (offset_into_cluster(s, l2_offset)) { |
660 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
661 | " unaligned (L1 index: %#" PRIx64 ")", | |
662 | l2_offset, l1_index); | |
663 | return -EIO; | |
664 | } | |
45aba42f KW |
665 | |
666 | /* seek the l2 table of the given l2 offset */ | |
667 | ||
8e37f681 | 668 | if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) { |
45aba42f | 669 | /* load the l2 table in memory */ |
55c17e98 KW |
670 | ret = l2_load(bs, l2_offset, &l2_table); |
671 | if (ret < 0) { | |
672 | return ret; | |
1e3e8f1a | 673 | } |
45aba42f | 674 | } else { |
16fde5f2 | 675 | /* First allocate a new L2 table (and do COW if needed) */ |
c46e1167 KW |
676 | ret = l2_allocate(bs, l1_index, &l2_table); |
677 | if (ret < 0) { | |
678 | return ret; | |
1e3e8f1a | 679 | } |
16fde5f2 KW |
680 | |
681 | /* Then decrease the refcount of the old table */ | |
682 | if (l2_offset) { | |
6cfcb9b8 KW |
683 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), |
684 | QCOW2_DISCARD_OTHER); | |
16fde5f2 | 685 | } |
45aba42f KW |
686 | } |
687 | ||
688 | /* find the cluster offset for the given disk offset */ | |
689 | ||
690 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
691 | ||
692 | *new_l2_table = l2_table; | |
45aba42f KW |
693 | *new_l2_index = l2_index; |
694 | ||
1e3e8f1a | 695 | return 0; |
45aba42f KW |
696 | } |
697 | ||
698 | /* | |
699 | * alloc_compressed_cluster_offset | |
700 | * | |
701 | * For a given offset of the disk image, return cluster offset in | |
702 | * qcow2 file. | |
703 | * | |
704 | * If the offset is not found, allocate a new compressed cluster. | |
705 | * | |
706 | * Return the cluster offset if successful, | |
707 | * Return 0, otherwise. | |
708 | * | |
709 | */ | |
710 | ||
ed6ccf0f KW |
711 | uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
712 | uint64_t offset, | |
713 | int compressed_size) | |
45aba42f | 714 | { |
ff99129a | 715 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 716 | int l2_index, ret; |
3948d1d4 | 717 | uint64_t *l2_table; |
f4f0d391 | 718 | int64_t cluster_offset; |
45aba42f KW |
719 | int nb_csectors; |
720 | ||
3948d1d4 | 721 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
1e3e8f1a | 722 | if (ret < 0) { |
45aba42f | 723 | return 0; |
1e3e8f1a | 724 | } |
45aba42f | 725 | |
b0b6862e KW |
726 | /* Compression can't overwrite anything. Fail if the cluster was already |
727 | * allocated. */ | |
45aba42f | 728 | cluster_offset = be64_to_cpu(l2_table[l2_index]); |
b0b6862e | 729 | if (cluster_offset & L2E_OFFSET_MASK) { |
8f1efd00 KW |
730 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
731 | return 0; | |
732 | } | |
45aba42f | 733 | |
ed6ccf0f | 734 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
5d757b56 | 735 | if (cluster_offset < 0) { |
29c1a730 | 736 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
5d757b56 KW |
737 | return 0; |
738 | } | |
739 | ||
45aba42f KW |
740 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
741 | (cluster_offset >> 9); | |
742 | ||
743 | cluster_offset |= QCOW_OFLAG_COMPRESSED | | |
744 | ((uint64_t)nb_csectors << s->csize_shift); | |
745 | ||
746 | /* update L2 table */ | |
747 | ||
748 | /* compressed clusters never have the copied flag */ | |
749 | ||
66f82cee | 750 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
72e80b89 | 751 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
45aba42f | 752 | l2_table[l2_index] = cpu_to_be64(cluster_offset); |
a3f1afb4 | 753 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
4c1612d9 | 754 | |
29c1a730 | 755 | return cluster_offset; |
4c1612d9 KW |
756 | } |
757 | ||
99450c6f | 758 | static int perform_cow(BlockDriverState *bs, QCowL2Meta *m) |
593fb83c | 759 | { |
ff99129a | 760 | BDRVQcow2State *s = bs->opaque; |
99450c6f AG |
761 | Qcow2COWRegion *start = &m->cow_start; |
762 | Qcow2COWRegion *end = &m->cow_end; | |
593fb83c KW |
763 | int ret; |
764 | ||
99450c6f | 765 | if (start->nb_bytes == 0 && end->nb_bytes == 0) { |
593fb83c KW |
766 | return 0; |
767 | } | |
768 | ||
769 | qemu_co_mutex_unlock(&s->lock); | |
99450c6f AG |
770 | ret = do_perform_cow(bs, m->offset, m->alloc_offset, |
771 | start->offset, start->nb_bytes); | |
593fb83c | 772 | if (ret < 0) { |
99450c6f | 773 | goto fail; |
593fb83c KW |
774 | } |
775 | ||
99450c6f AG |
776 | ret = do_perform_cow(bs, m->offset, m->alloc_offset, |
777 | end->offset, end->nb_bytes); | |
778 | ||
779 | fail: | |
780 | qemu_co_mutex_lock(&s->lock); | |
781 | ||
593fb83c KW |
782 | /* |
783 | * Before we update the L2 table to actually point to the new cluster, we | |
784 | * need to be sure that the refcounts have been increased and COW was | |
785 | * handled. | |
786 | */ | |
99450c6f AG |
787 | if (ret == 0) { |
788 | qcow2_cache_depends_on_flush(s->l2_table_cache); | |
789 | } | |
593fb83c | 790 | |
99450c6f | 791 | return ret; |
593fb83c KW |
792 | } |
793 | ||
148da7ea | 794 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) |
45aba42f | 795 | { |
ff99129a | 796 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 797 | int i, j = 0, l2_index, ret; |
593fb83c | 798 | uint64_t *old_cluster, *l2_table; |
250196f1 | 799 | uint64_t cluster_offset = m->alloc_offset; |
45aba42f | 800 | |
3cce16f4 | 801 | trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
f50f88b9 | 802 | assert(m->nb_clusters > 0); |
45aba42f | 803 | |
5839e53b | 804 | old_cluster = g_try_new(uint64_t, m->nb_clusters); |
de82815d KW |
805 | if (old_cluster == NULL) { |
806 | ret = -ENOMEM; | |
807 | goto err; | |
808 | } | |
45aba42f KW |
809 | |
810 | /* copy content of unmodified sectors */ | |
99450c6f | 811 | ret = perform_cow(bs, m); |
593fb83c KW |
812 | if (ret < 0) { |
813 | goto err; | |
29c1a730 KW |
814 | } |
815 | ||
593fb83c | 816 | /* Update L2 table. */ |
74c4510a | 817 | if (s->use_lazy_refcounts) { |
280d3735 KW |
818 | qcow2_mark_dirty(bs); |
819 | } | |
bfe8043e SH |
820 | if (qcow2_need_accurate_refcounts(s)) { |
821 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
822 | s->refcount_block_cache); | |
823 | } | |
280d3735 | 824 | |
3948d1d4 | 825 | ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index); |
1e3e8f1a | 826 | if (ret < 0) { |
45aba42f | 827 | goto err; |
1e3e8f1a | 828 | } |
72e80b89 | 829 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
45aba42f | 830 | |
c01dbccb | 831 | assert(l2_index + m->nb_clusters <= s->l2_size); |
45aba42f KW |
832 | for (i = 0; i < m->nb_clusters; i++) { |
833 | /* if two concurrent writes happen to the same unallocated cluster | |
aaa4d20b KW |
834 | * each write allocates separate cluster and writes data concurrently. |
835 | * The first one to complete updates l2 table with pointer to its | |
836 | * cluster the second one has to do RMW (which is done above by | |
837 | * perform_cow()), update l2 table with its cluster pointer and free | |
838 | * old cluster. This is what this loop does */ | |
839 | if (l2_table[l2_index + i] != 0) { | |
45aba42f | 840 | old_cluster[j++] = l2_table[l2_index + i]; |
aaa4d20b | 841 | } |
45aba42f KW |
842 | |
843 | l2_table[l2_index + i] = cpu_to_be64((cluster_offset + | |
844 | (i << s->cluster_bits)) | QCOW_OFLAG_COPIED); | |
845 | } | |
846 | ||
9f8e668e | 847 | |
a3f1afb4 | 848 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
45aba42f | 849 | |
7ec5e6a4 KW |
850 | /* |
851 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
6cfcb9b8 KW |
852 | * |
853 | * Don't discard clusters that reach a refcount of 0 (e.g. compressed | |
854 | * clusters), the next write will reuse them anyway. | |
7ec5e6a4 | 855 | */ |
564a6b69 | 856 | if (!m->keep_old_clusters && j != 0) { |
7ec5e6a4 | 857 | for (i = 0; i < j; i++) { |
6cfcb9b8 KW |
858 | qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1, |
859 | QCOW2_DISCARD_NEVER); | |
7ec5e6a4 KW |
860 | } |
861 | } | |
45aba42f KW |
862 | |
863 | ret = 0; | |
864 | err: | |
7267c094 | 865 | g_free(old_cluster); |
45aba42f KW |
866 | return ret; |
867 | } | |
868 | ||
bf319ece KW |
869 | /* |
870 | * Returns the number of contiguous clusters that can be used for an allocating | |
871 | * write, but require COW to be performed (this includes yet unallocated space, | |
872 | * which must copy from the backing file) | |
873 | */ | |
ff99129a | 874 | static int count_cow_clusters(BDRVQcow2State *s, int nb_clusters, |
bf319ece KW |
875 | uint64_t *l2_table, int l2_index) |
876 | { | |
143550a8 | 877 | int i; |
bf319ece | 878 | |
143550a8 KW |
879 | for (i = 0; i < nb_clusters; i++) { |
880 | uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]); | |
3ef95218 | 881 | QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry); |
143550a8 KW |
882 | |
883 | switch(cluster_type) { | |
884 | case QCOW2_CLUSTER_NORMAL: | |
885 | if (l2_entry & QCOW_OFLAG_COPIED) { | |
886 | goto out; | |
887 | } | |
bf319ece | 888 | break; |
143550a8 KW |
889 | case QCOW2_CLUSTER_UNALLOCATED: |
890 | case QCOW2_CLUSTER_COMPRESSED: | |
fdfab37d EB |
891 | case QCOW2_CLUSTER_ZERO_PLAIN: |
892 | case QCOW2_CLUSTER_ZERO_ALLOC: | |
bf319ece | 893 | break; |
143550a8 KW |
894 | default: |
895 | abort(); | |
896 | } | |
bf319ece KW |
897 | } |
898 | ||
143550a8 | 899 | out: |
bf319ece KW |
900 | assert(i <= nb_clusters); |
901 | return i; | |
902 | } | |
903 | ||
250196f1 | 904 | /* |
226c3c26 KW |
905 | * Check if there already is an AIO write request in flight which allocates |
906 | * the same cluster. In this case we need to wait until the previous | |
907 | * request has completed and updated the L2 table accordingly. | |
65eb2e35 KW |
908 | * |
909 | * Returns: | |
910 | * 0 if there was no dependency. *cur_bytes indicates the number of | |
911 | * bytes from guest_offset that can be read before the next | |
912 | * dependency must be processed (or the request is complete) | |
913 | * | |
914 | * -EAGAIN if we had to wait for another request, previously gathered | |
915 | * information on cluster allocation may be invalid now. The caller | |
916 | * must start over anyway, so consider *cur_bytes undefined. | |
250196f1 | 917 | */ |
226c3c26 | 918 | static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, |
ecdd5333 | 919 | uint64_t *cur_bytes, QCowL2Meta **m) |
250196f1 | 920 | { |
ff99129a | 921 | BDRVQcow2State *s = bs->opaque; |
250196f1 | 922 | QCowL2Meta *old_alloc; |
65eb2e35 | 923 | uint64_t bytes = *cur_bytes; |
250196f1 | 924 | |
250196f1 KW |
925 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
926 | ||
65eb2e35 KW |
927 | uint64_t start = guest_offset; |
928 | uint64_t end = start + bytes; | |
929 | uint64_t old_start = l2meta_cow_start(old_alloc); | |
930 | uint64_t old_end = l2meta_cow_end(old_alloc); | |
250196f1 | 931 | |
d9d74f41 | 932 | if (end <= old_start || start >= old_end) { |
250196f1 KW |
933 | /* No intersection */ |
934 | } else { | |
935 | if (start < old_start) { | |
936 | /* Stop at the start of a running allocation */ | |
65eb2e35 | 937 | bytes = old_start - start; |
250196f1 | 938 | } else { |
65eb2e35 | 939 | bytes = 0; |
250196f1 KW |
940 | } |
941 | ||
ecdd5333 KW |
942 | /* Stop if already an l2meta exists. After yielding, it wouldn't |
943 | * be valid any more, so we'd have to clean up the old L2Metas | |
944 | * and deal with requests depending on them before starting to | |
945 | * gather new ones. Not worth the trouble. */ | |
946 | if (bytes == 0 && *m) { | |
947 | *cur_bytes = 0; | |
948 | return 0; | |
949 | } | |
950 | ||
65eb2e35 | 951 | if (bytes == 0) { |
250196f1 KW |
952 | /* Wait for the dependency to complete. We need to recheck |
953 | * the free/allocated clusters when we continue. */ | |
1ace7cea | 954 | qemu_co_queue_wait(&old_alloc->dependent_requests, &s->lock); |
250196f1 KW |
955 | return -EAGAIN; |
956 | } | |
957 | } | |
958 | } | |
959 | ||
65eb2e35 KW |
960 | /* Make sure that existing clusters and new allocations are only used up to |
961 | * the next dependency if we shortened the request above */ | |
962 | *cur_bytes = bytes; | |
250196f1 | 963 | |
226c3c26 KW |
964 | return 0; |
965 | } | |
966 | ||
0af729ec KW |
967 | /* |
968 | * Checks how many already allocated clusters that don't require a copy on | |
969 | * write there are at the given guest_offset (up to *bytes). If | |
970 | * *host_offset is not zero, only physically contiguous clusters beginning at | |
971 | * this host offset are counted. | |
972 | * | |
411d62b0 KW |
973 | * Note that guest_offset may not be cluster aligned. In this case, the |
974 | * returned *host_offset points to exact byte referenced by guest_offset and | |
975 | * therefore isn't cluster aligned as well. | |
0af729ec KW |
976 | * |
977 | * Returns: | |
978 | * 0: if no allocated clusters are available at the given offset. | |
979 | * *bytes is normally unchanged. It is set to 0 if the cluster | |
980 | * is allocated and doesn't need COW, but doesn't have the right | |
981 | * physical offset. | |
982 | * | |
983 | * 1: if allocated clusters that don't require a COW are available at | |
984 | * the requested offset. *bytes may have decreased and describes | |
985 | * the length of the area that can be written to. | |
986 | * | |
987 | * -errno: in error cases | |
0af729ec KW |
988 | */ |
989 | static int handle_copied(BlockDriverState *bs, uint64_t guest_offset, | |
c53ede9f | 990 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
0af729ec | 991 | { |
ff99129a | 992 | BDRVQcow2State *s = bs->opaque; |
0af729ec KW |
993 | int l2_index; |
994 | uint64_t cluster_offset; | |
995 | uint64_t *l2_table; | |
b6d36def | 996 | uint64_t nb_clusters; |
c53ede9f | 997 | unsigned int keep_clusters; |
a3f1afb4 | 998 | int ret; |
0af729ec KW |
999 | |
1000 | trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset, | |
1001 | *bytes); | |
0af729ec | 1002 | |
411d62b0 KW |
1003 | assert(*host_offset == 0 || offset_into_cluster(s, guest_offset) |
1004 | == offset_into_cluster(s, *host_offset)); | |
1005 | ||
acb0467f KW |
1006 | /* |
1007 | * Calculate the number of clusters to look for. We stop at L2 table | |
1008 | * boundaries to keep things simple. | |
1009 | */ | |
1010 | nb_clusters = | |
1011 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1012 | ||
1013 | l2_index = offset_to_l2_index(s, guest_offset); | |
1014 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1015 | assert(nb_clusters <= INT_MAX); |
acb0467f | 1016 | |
0af729ec KW |
1017 | /* Find L2 entry for the first involved cluster */ |
1018 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
1019 | if (ret < 0) { | |
1020 | return ret; | |
1021 | } | |
1022 | ||
1023 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
1024 | ||
1025 | /* Check how many clusters are already allocated and don't need COW */ | |
1026 | if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL | |
1027 | && (cluster_offset & QCOW_OFLAG_COPIED)) | |
1028 | { | |
e62daaf6 KW |
1029 | /* If a specific host_offset is required, check it */ |
1030 | bool offset_matches = | |
1031 | (cluster_offset & L2E_OFFSET_MASK) == *host_offset; | |
1032 | ||
a97c67ee HR |
1033 | if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) { |
1034 | qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset " | |
1035 | "%#llx unaligned (guest offset: %#" PRIx64 | |
1036 | ")", cluster_offset & L2E_OFFSET_MASK, | |
1037 | guest_offset); | |
1038 | ret = -EIO; | |
1039 | goto out; | |
1040 | } | |
1041 | ||
e62daaf6 KW |
1042 | if (*host_offset != 0 && !offset_matches) { |
1043 | *bytes = 0; | |
1044 | ret = 0; | |
1045 | goto out; | |
1046 | } | |
1047 | ||
0af729ec | 1048 | /* We keep all QCOW_OFLAG_COPIED clusters */ |
c53ede9f | 1049 | keep_clusters = |
acb0467f | 1050 | count_contiguous_clusters(nb_clusters, s->cluster_size, |
61653008 | 1051 | &l2_table[l2_index], |
0af729ec | 1052 | QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); |
c53ede9f KW |
1053 | assert(keep_clusters <= nb_clusters); |
1054 | ||
1055 | *bytes = MIN(*bytes, | |
1056 | keep_clusters * s->cluster_size | |
1057 | - offset_into_cluster(s, guest_offset)); | |
0af729ec KW |
1058 | |
1059 | ret = 1; | |
1060 | } else { | |
0af729ec KW |
1061 | ret = 0; |
1062 | } | |
1063 | ||
0af729ec | 1064 | /* Cleanup */ |
e62daaf6 | 1065 | out: |
a3f1afb4 | 1066 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
0af729ec | 1067 | |
e62daaf6 KW |
1068 | /* Only return a host offset if we actually made progress. Otherwise we |
1069 | * would make requirements for handle_alloc() that it can't fulfill */ | |
a97c67ee | 1070 | if (ret > 0) { |
411d62b0 KW |
1071 | *host_offset = (cluster_offset & L2E_OFFSET_MASK) |
1072 | + offset_into_cluster(s, guest_offset); | |
e62daaf6 KW |
1073 | } |
1074 | ||
0af729ec KW |
1075 | return ret; |
1076 | } | |
1077 | ||
226c3c26 KW |
1078 | /* |
1079 | * Allocates new clusters for the given guest_offset. | |
1080 | * | |
1081 | * At most *nb_clusters are allocated, and on return *nb_clusters is updated to | |
1082 | * contain the number of clusters that have been allocated and are contiguous | |
1083 | * in the image file. | |
1084 | * | |
1085 | * If *host_offset is non-zero, it specifies the offset in the image file at | |
1086 | * which the new clusters must start. *nb_clusters can be 0 on return in this | |
1087 | * case if the cluster at host_offset is already in use. If *host_offset is | |
1088 | * zero, the clusters can be allocated anywhere in the image file. | |
1089 | * | |
1090 | * *host_offset is updated to contain the offset into the image file at which | |
1091 | * the first allocated cluster starts. | |
1092 | * | |
1093 | * Return 0 on success and -errno in error cases. -EAGAIN means that the | |
1094 | * function has been waiting for another request and the allocation must be | |
1095 | * restarted, but the whole request should not be failed. | |
1096 | */ | |
1097 | static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, | |
b6d36def | 1098 | uint64_t *host_offset, uint64_t *nb_clusters) |
226c3c26 | 1099 | { |
ff99129a | 1100 | BDRVQcow2State *s = bs->opaque; |
226c3c26 KW |
1101 | |
1102 | trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, | |
1103 | *host_offset, *nb_clusters); | |
1104 | ||
250196f1 KW |
1105 | /* Allocate new clusters */ |
1106 | trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); | |
1107 | if (*host_offset == 0) { | |
df021791 KW |
1108 | int64_t cluster_offset = |
1109 | qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); | |
1110 | if (cluster_offset < 0) { | |
1111 | return cluster_offset; | |
1112 | } | |
1113 | *host_offset = cluster_offset; | |
1114 | return 0; | |
250196f1 | 1115 | } else { |
b6d36def | 1116 | int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); |
df021791 KW |
1117 | if (ret < 0) { |
1118 | return ret; | |
1119 | } | |
1120 | *nb_clusters = ret; | |
1121 | return 0; | |
250196f1 | 1122 | } |
250196f1 KW |
1123 | } |
1124 | ||
10f0ed8b KW |
1125 | /* |
1126 | * Allocates new clusters for an area that either is yet unallocated or needs a | |
1127 | * copy on write. If *host_offset is non-zero, clusters are only allocated if | |
1128 | * the new allocation can match the specified host offset. | |
1129 | * | |
411d62b0 KW |
1130 | * Note that guest_offset may not be cluster aligned. In this case, the |
1131 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1132 | * therefore isn't cluster aligned as well. | |
10f0ed8b KW |
1133 | * |
1134 | * Returns: | |
1135 | * 0: if no clusters could be allocated. *bytes is set to 0, | |
1136 | * *host_offset is left unchanged. | |
1137 | * | |
1138 | * 1: if new clusters were allocated. *bytes may be decreased if the | |
1139 | * new allocation doesn't cover all of the requested area. | |
1140 | * *host_offset is updated to contain the host offset of the first | |
1141 | * newly allocated cluster. | |
1142 | * | |
1143 | * -errno: in error cases | |
10f0ed8b KW |
1144 | */ |
1145 | static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, | |
c37f4cd7 | 1146 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
10f0ed8b | 1147 | { |
ff99129a | 1148 | BDRVQcow2State *s = bs->opaque; |
10f0ed8b KW |
1149 | int l2_index; |
1150 | uint64_t *l2_table; | |
1151 | uint64_t entry; | |
b6d36def | 1152 | uint64_t nb_clusters; |
10f0ed8b | 1153 | int ret; |
564a6b69 | 1154 | bool keep_old_clusters = false; |
10f0ed8b | 1155 | |
564a6b69 | 1156 | uint64_t alloc_cluster_offset = 0; |
10f0ed8b KW |
1157 | |
1158 | trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset, | |
1159 | *bytes); | |
1160 | assert(*bytes > 0); | |
1161 | ||
f5bc6350 KW |
1162 | /* |
1163 | * Calculate the number of clusters to look for. We stop at L2 table | |
1164 | * boundaries to keep things simple. | |
1165 | */ | |
c37f4cd7 KW |
1166 | nb_clusters = |
1167 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1168 | ||
f5bc6350 | 1169 | l2_index = offset_to_l2_index(s, guest_offset); |
c37f4cd7 | 1170 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
b6d36def | 1171 | assert(nb_clusters <= INT_MAX); |
f5bc6350 | 1172 | |
10f0ed8b KW |
1173 | /* Find L2 entry for the first involved cluster */ |
1174 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
1175 | if (ret < 0) { | |
1176 | return ret; | |
1177 | } | |
1178 | ||
3b8e2e26 | 1179 | entry = be64_to_cpu(l2_table[l2_index]); |
10f0ed8b KW |
1180 | |
1181 | /* For the moment, overwrite compressed clusters one by one */ | |
1182 | if (entry & QCOW_OFLAG_COMPRESSED) { | |
1183 | nb_clusters = 1; | |
1184 | } else { | |
3b8e2e26 | 1185 | nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index); |
10f0ed8b KW |
1186 | } |
1187 | ||
ecdd5333 KW |
1188 | /* This function is only called when there were no non-COW clusters, so if |
1189 | * we can't find any unallocated or COW clusters either, something is | |
1190 | * wrong with our code. */ | |
1191 | assert(nb_clusters > 0); | |
1192 | ||
fdfab37d EB |
1193 | if (qcow2_get_cluster_type(entry) == QCOW2_CLUSTER_ZERO_ALLOC && |
1194 | (entry & QCOW_OFLAG_COPIED) && | |
564a6b69 HR |
1195 | (!*host_offset || |
1196 | start_of_cluster(s, *host_offset) == (entry & L2E_OFFSET_MASK))) | |
1197 | { | |
1198 | /* Try to reuse preallocated zero clusters; contiguous normal clusters | |
1199 | * would be fine, too, but count_cow_clusters() above has limited | |
1200 | * nb_clusters already to a range of COW clusters */ | |
1201 | int preallocated_nb_clusters = | |
1202 | count_contiguous_clusters(nb_clusters, s->cluster_size, | |
1203 | &l2_table[l2_index], QCOW_OFLAG_COPIED); | |
1204 | assert(preallocated_nb_clusters > 0); | |
10f0ed8b | 1205 | |
564a6b69 HR |
1206 | nb_clusters = preallocated_nb_clusters; |
1207 | alloc_cluster_offset = entry & L2E_OFFSET_MASK; | |
10f0ed8b | 1208 | |
564a6b69 HR |
1209 | /* We want to reuse these clusters, so qcow2_alloc_cluster_link_l2() |
1210 | * should not free them. */ | |
1211 | keep_old_clusters = true; | |
10f0ed8b KW |
1212 | } |
1213 | ||
564a6b69 HR |
1214 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
1215 | ||
ff52aab2 | 1216 | if (!alloc_cluster_offset) { |
564a6b69 HR |
1217 | /* Allocate, if necessary at a given offset in the image file */ |
1218 | alloc_cluster_offset = start_of_cluster(s, *host_offset); | |
1219 | ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset, | |
1220 | &nb_clusters); | |
1221 | if (ret < 0) { | |
1222 | goto fail; | |
1223 | } | |
1224 | ||
1225 | /* Can't extend contiguous allocation */ | |
1226 | if (nb_clusters == 0) { | |
1227 | *bytes = 0; | |
1228 | return 0; | |
1229 | } | |
1230 | ||
1231 | /* !*host_offset would overwrite the image header and is reserved for | |
1232 | * "no host offset preferred". If 0 was a valid host offset, it'd | |
1233 | * trigger the following overlap check; do that now to avoid having an | |
1234 | * invalid value in *host_offset. */ | |
1235 | if (!alloc_cluster_offset) { | |
1236 | ret = qcow2_pre_write_overlap_check(bs, 0, alloc_cluster_offset, | |
1237 | nb_clusters * s->cluster_size); | |
1238 | assert(ret < 0); | |
1239 | goto fail; | |
1240 | } | |
ff52aab2 HR |
1241 | } |
1242 | ||
83baa9a4 KW |
1243 | /* |
1244 | * Save info needed for meta data update. | |
1245 | * | |
85567393 | 1246 | * requested_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1247 | * newly allocated cluster to the end of the (possibly shortened |
1248 | * before) write request. | |
1249 | * | |
85567393 | 1250 | * avail_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1251 | * newly allocated to the end of the last newly allocated cluster. |
1252 | * | |
85567393 | 1253 | * nb_bytes: The number of bytes from the start of the first |
83baa9a4 KW |
1254 | * newly allocated cluster to the end of the area that the write |
1255 | * request actually writes to (excluding COW at the end) | |
1256 | */ | |
85567393 KW |
1257 | uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset); |
1258 | int avail_bytes = MIN(INT_MAX, nb_clusters << s->cluster_bits); | |
1259 | int nb_bytes = MIN(requested_bytes, avail_bytes); | |
88c6588c | 1260 | QCowL2Meta *old_m = *m; |
83baa9a4 | 1261 | |
83baa9a4 KW |
1262 | *m = g_malloc0(sizeof(**m)); |
1263 | ||
1264 | **m = (QCowL2Meta) { | |
88c6588c KW |
1265 | .next = old_m, |
1266 | ||
411d62b0 | 1267 | .alloc_offset = alloc_cluster_offset, |
83baa9a4 KW |
1268 | .offset = start_of_cluster(s, guest_offset), |
1269 | .nb_clusters = nb_clusters, | |
83baa9a4 | 1270 | |
564a6b69 HR |
1271 | .keep_old_clusters = keep_old_clusters, |
1272 | ||
83baa9a4 KW |
1273 | .cow_start = { |
1274 | .offset = 0, | |
85567393 | 1275 | .nb_bytes = offset_into_cluster(s, guest_offset), |
83baa9a4 KW |
1276 | }, |
1277 | .cow_end = { | |
85567393 KW |
1278 | .offset = nb_bytes, |
1279 | .nb_bytes = avail_bytes - nb_bytes, | |
83baa9a4 KW |
1280 | }, |
1281 | }; | |
1282 | qemu_co_queue_init(&(*m)->dependent_requests); | |
1283 | QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); | |
1284 | ||
411d62b0 | 1285 | *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset); |
85567393 | 1286 | *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset)); |
83baa9a4 KW |
1287 | assert(*bytes != 0); |
1288 | ||
10f0ed8b KW |
1289 | return 1; |
1290 | ||
1291 | fail: | |
1292 | if (*m && (*m)->nb_clusters > 0) { | |
1293 | QLIST_REMOVE(*m, next_in_flight); | |
1294 | } | |
1295 | return ret; | |
1296 | } | |
1297 | ||
45aba42f KW |
1298 | /* |
1299 | * alloc_cluster_offset | |
1300 | * | |
250196f1 KW |
1301 | * For a given offset on the virtual disk, find the cluster offset in qcow2 |
1302 | * file. If the offset is not found, allocate a new cluster. | |
45aba42f | 1303 | * |
250196f1 | 1304 | * If the cluster was already allocated, m->nb_clusters is set to 0 and |
a7912369 | 1305 | * other fields in m are meaningless. |
148da7ea KW |
1306 | * |
1307 | * If the cluster is newly allocated, m->nb_clusters is set to the number of | |
68d100e9 KW |
1308 | * contiguous clusters that have been allocated. In this case, the other |
1309 | * fields of m are valid and contain information about the first allocated | |
1310 | * cluster. | |
45aba42f | 1311 | * |
68d100e9 KW |
1312 | * If the request conflicts with another write request in flight, the coroutine |
1313 | * is queued and will be reentered when the dependency has completed. | |
148da7ea KW |
1314 | * |
1315 | * Return 0 on success and -errno in error cases | |
45aba42f | 1316 | */ |
f4f0d391 | 1317 | int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, |
d46a0bb2 KW |
1318 | unsigned int *bytes, uint64_t *host_offset, |
1319 | QCowL2Meta **m) | |
45aba42f | 1320 | { |
ff99129a | 1321 | BDRVQcow2State *s = bs->opaque; |
710c2496 | 1322 | uint64_t start, remaining; |
250196f1 | 1323 | uint64_t cluster_offset; |
65eb2e35 | 1324 | uint64_t cur_bytes; |
710c2496 | 1325 | int ret; |
45aba42f | 1326 | |
d46a0bb2 | 1327 | trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *bytes); |
710c2496 | 1328 | |
72424114 | 1329 | again: |
16f0587e | 1330 | start = offset; |
d46a0bb2 | 1331 | remaining = *bytes; |
0af729ec KW |
1332 | cluster_offset = 0; |
1333 | *host_offset = 0; | |
ecdd5333 KW |
1334 | cur_bytes = 0; |
1335 | *m = NULL; | |
0af729ec | 1336 | |
2c3b32d2 | 1337 | while (true) { |
ecdd5333 KW |
1338 | |
1339 | if (!*host_offset) { | |
1340 | *host_offset = start_of_cluster(s, cluster_offset); | |
1341 | } | |
1342 | ||
1343 | assert(remaining >= cur_bytes); | |
1344 | ||
1345 | start += cur_bytes; | |
1346 | remaining -= cur_bytes; | |
1347 | cluster_offset += cur_bytes; | |
1348 | ||
1349 | if (remaining == 0) { | |
1350 | break; | |
1351 | } | |
1352 | ||
1353 | cur_bytes = remaining; | |
1354 | ||
2c3b32d2 KW |
1355 | /* |
1356 | * Now start gathering as many contiguous clusters as possible: | |
1357 | * | |
1358 | * 1. Check for overlaps with in-flight allocations | |
1359 | * | |
1360 | * a) Overlap not in the first cluster -> shorten this request and | |
1361 | * let the caller handle the rest in its next loop iteration. | |
1362 | * | |
1363 | * b) Real overlaps of two requests. Yield and restart the search | |
1364 | * for contiguous clusters (the situation could have changed | |
1365 | * while we were sleeping) | |
1366 | * | |
1367 | * c) TODO: Request starts in the same cluster as the in-flight | |
1368 | * allocation ends. Shorten the COW of the in-fight allocation, | |
1369 | * set cluster_offset to write to the same cluster and set up | |
1370 | * the right synchronisation between the in-flight request and | |
1371 | * the new one. | |
1372 | */ | |
ecdd5333 | 1373 | ret = handle_dependencies(bs, start, &cur_bytes, m); |
2c3b32d2 | 1374 | if (ret == -EAGAIN) { |
ecdd5333 KW |
1375 | /* Currently handle_dependencies() doesn't yield if we already had |
1376 | * an allocation. If it did, we would have to clean up the L2Meta | |
1377 | * structs before starting over. */ | |
1378 | assert(*m == NULL); | |
2c3b32d2 KW |
1379 | goto again; |
1380 | } else if (ret < 0) { | |
1381 | return ret; | |
ecdd5333 KW |
1382 | } else if (cur_bytes == 0) { |
1383 | break; | |
2c3b32d2 KW |
1384 | } else { |
1385 | /* handle_dependencies() may have decreased cur_bytes (shortened | |
1386 | * the allocations below) so that the next dependency is processed | |
1387 | * correctly during the next loop iteration. */ | |
0af729ec | 1388 | } |
710c2496 | 1389 | |
2c3b32d2 KW |
1390 | /* |
1391 | * 2. Count contiguous COPIED clusters. | |
1392 | */ | |
1393 | ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m); | |
1394 | if (ret < 0) { | |
1395 | return ret; | |
1396 | } else if (ret) { | |
ecdd5333 | 1397 | continue; |
2c3b32d2 KW |
1398 | } else if (cur_bytes == 0) { |
1399 | break; | |
1400 | } | |
060bee89 | 1401 | |
2c3b32d2 KW |
1402 | /* |
1403 | * 3. If the request still hasn't completed, allocate new clusters, | |
1404 | * considering any cluster_offset of steps 1c or 2. | |
1405 | */ | |
1406 | ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m); | |
1407 | if (ret < 0) { | |
1408 | return ret; | |
1409 | } else if (ret) { | |
ecdd5333 | 1410 | continue; |
2c3b32d2 KW |
1411 | } else { |
1412 | assert(cur_bytes == 0); | |
1413 | break; | |
1414 | } | |
f5bc6350 | 1415 | } |
10f0ed8b | 1416 | |
d46a0bb2 KW |
1417 | *bytes -= remaining; |
1418 | assert(*bytes > 0); | |
710c2496 | 1419 | assert(*host_offset != 0); |
45aba42f | 1420 | |
148da7ea | 1421 | return 0; |
45aba42f KW |
1422 | } |
1423 | ||
1424 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
1425 | const uint8_t *buf, int buf_size) | |
1426 | { | |
1427 | z_stream strm1, *strm = &strm1; | |
1428 | int ret, out_len; | |
1429 | ||
1430 | memset(strm, 0, sizeof(*strm)); | |
1431 | ||
1432 | strm->next_in = (uint8_t *)buf; | |
1433 | strm->avail_in = buf_size; | |
1434 | strm->next_out = out_buf; | |
1435 | strm->avail_out = out_buf_size; | |
1436 | ||
1437 | ret = inflateInit2(strm, -12); | |
1438 | if (ret != Z_OK) | |
1439 | return -1; | |
1440 | ret = inflate(strm, Z_FINISH); | |
1441 | out_len = strm->next_out - out_buf; | |
1442 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
1443 | out_len != out_buf_size) { | |
1444 | inflateEnd(strm); | |
1445 | return -1; | |
1446 | } | |
1447 | inflateEnd(strm); | |
1448 | return 0; | |
1449 | } | |
1450 | ||
66f82cee | 1451 | int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) |
45aba42f | 1452 | { |
ff99129a | 1453 | BDRVQcow2State *s = bs->opaque; |
45aba42f KW |
1454 | int ret, csize, nb_csectors, sector_offset; |
1455 | uint64_t coffset; | |
1456 | ||
1457 | coffset = cluster_offset & s->cluster_offset_mask; | |
1458 | if (s->cluster_cache_offset != coffset) { | |
1459 | nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1; | |
1460 | sector_offset = coffset & 511; | |
1461 | csize = nb_csectors * 512 - sector_offset; | |
66f82cee | 1462 | BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
fbcbbf4e | 1463 | ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, |
9a4f4c31 | 1464 | nb_csectors); |
45aba42f | 1465 | if (ret < 0) { |
8af36488 | 1466 | return ret; |
45aba42f KW |
1467 | } |
1468 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
1469 | s->cluster_data + sector_offset, csize) < 0) { | |
8af36488 | 1470 | return -EIO; |
45aba42f KW |
1471 | } |
1472 | s->cluster_cache_offset = coffset; | |
1473 | } | |
1474 | return 0; | |
1475 | } | |
5ea929e3 KW |
1476 | |
1477 | /* | |
1478 | * This discards as many clusters of nb_clusters as possible at once (i.e. | |
1479 | * all clusters in the same L2 table) and returns the number of discarded | |
1480 | * clusters. | |
1481 | */ | |
1482 | static int discard_single_l2(BlockDriverState *bs, uint64_t offset, | |
b6d36def HR |
1483 | uint64_t nb_clusters, enum qcow2_discard_type type, |
1484 | bool full_discard) | |
5ea929e3 | 1485 | { |
ff99129a | 1486 | BDRVQcow2State *s = bs->opaque; |
3948d1d4 | 1487 | uint64_t *l2_table; |
5ea929e3 KW |
1488 | int l2_index; |
1489 | int ret; | |
1490 | int i; | |
1491 | ||
3948d1d4 | 1492 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
5ea929e3 KW |
1493 | if (ret < 0) { |
1494 | return ret; | |
1495 | } | |
1496 | ||
1497 | /* Limit nb_clusters to one L2 table */ | |
1498 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1499 | assert(nb_clusters <= INT_MAX); |
5ea929e3 KW |
1500 | |
1501 | for (i = 0; i < nb_clusters; i++) { | |
c883db0d | 1502 | uint64_t old_l2_entry; |
5ea929e3 | 1503 | |
c883db0d | 1504 | old_l2_entry = be64_to_cpu(l2_table[l2_index + i]); |
a71835a0 KW |
1505 | |
1506 | /* | |
808c4b6f HR |
1507 | * If full_discard is false, make sure that a discarded area reads back |
1508 | * as zeroes for v3 images (we cannot do it for v2 without actually | |
1509 | * writing a zero-filled buffer). We can skip the operation if the | |
1510 | * cluster is already marked as zero, or if it's unallocated and we | |
1511 | * don't have a backing file. | |
a71835a0 KW |
1512 | * |
1513 | * TODO We might want to use bdrv_get_block_status(bs) here, but we're | |
1514 | * holding s->lock, so that doesn't work today. | |
808c4b6f HR |
1515 | * |
1516 | * If full_discard is true, the sector should not read back as zeroes, | |
1517 | * but rather fall through to the backing file. | |
a71835a0 | 1518 | */ |
c883db0d | 1519 | switch (qcow2_get_cluster_type(old_l2_entry)) { |
bbd995d8 EB |
1520 | case QCOW2_CLUSTER_UNALLOCATED: |
1521 | if (full_discard || !bs->backing) { | |
1522 | continue; | |
1523 | } | |
1524 | break; | |
1525 | ||
fdfab37d EB |
1526 | case QCOW2_CLUSTER_ZERO_PLAIN: |
1527 | if (!full_discard) { | |
bbd995d8 EB |
1528 | continue; |
1529 | } | |
1530 | break; | |
1531 | ||
fdfab37d | 1532 | case QCOW2_CLUSTER_ZERO_ALLOC: |
bbd995d8 EB |
1533 | case QCOW2_CLUSTER_NORMAL: |
1534 | case QCOW2_CLUSTER_COMPRESSED: | |
1535 | break; | |
1536 | ||
1537 | default: | |
1538 | abort(); | |
5ea929e3 KW |
1539 | } |
1540 | ||
1541 | /* First remove L2 entries */ | |
72e80b89 | 1542 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
808c4b6f | 1543 | if (!full_discard && s->qcow_version >= 3) { |
a71835a0 KW |
1544 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
1545 | } else { | |
1546 | l2_table[l2_index + i] = cpu_to_be64(0); | |
1547 | } | |
5ea929e3 KW |
1548 | |
1549 | /* Then decrease the refcount */ | |
c883db0d | 1550 | qcow2_free_any_clusters(bs, old_l2_entry, 1, type); |
5ea929e3 KW |
1551 | } |
1552 | ||
a3f1afb4 | 1553 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
5ea929e3 KW |
1554 | |
1555 | return nb_clusters; | |
1556 | } | |
1557 | ||
d2cb36af EB |
1558 | int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset, |
1559 | uint64_t bytes, enum qcow2_discard_type type, | |
1560 | bool full_discard) | |
5ea929e3 | 1561 | { |
ff99129a | 1562 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1563 | uint64_t end_offset = offset + bytes; |
b6d36def | 1564 | uint64_t nb_clusters; |
d2cb36af | 1565 | int64_t cleared; |
5ea929e3 KW |
1566 | int ret; |
1567 | ||
f10ee139 | 1568 | /* Caller must pass aligned values, except at image end */ |
0c1bd469 | 1569 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); |
f10ee139 EB |
1570 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || |
1571 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
5ea929e3 | 1572 | |
d2cb36af | 1573 | nb_clusters = size_to_clusters(s, bytes); |
5ea929e3 | 1574 | |
0b919fae KW |
1575 | s->cache_discards = true; |
1576 | ||
5ea929e3 KW |
1577 | /* Each L2 table is handled by its own loop iteration */ |
1578 | while (nb_clusters > 0) { | |
d2cb36af EB |
1579 | cleared = discard_single_l2(bs, offset, nb_clusters, type, |
1580 | full_discard); | |
1581 | if (cleared < 0) { | |
1582 | ret = cleared; | |
0b919fae | 1583 | goto fail; |
5ea929e3 KW |
1584 | } |
1585 | ||
d2cb36af EB |
1586 | nb_clusters -= cleared; |
1587 | offset += (cleared * s->cluster_size); | |
5ea929e3 KW |
1588 | } |
1589 | ||
0b919fae KW |
1590 | ret = 0; |
1591 | fail: | |
1592 | s->cache_discards = false; | |
1593 | qcow2_process_discards(bs, ret); | |
1594 | ||
1595 | return ret; | |
5ea929e3 | 1596 | } |
621f0589 KW |
1597 | |
1598 | /* | |
1599 | * This zeroes as many clusters of nb_clusters as possible at once (i.e. | |
1600 | * all clusters in the same L2 table) and returns the number of zeroed | |
1601 | * clusters. | |
1602 | */ | |
1603 | static int zero_single_l2(BlockDriverState *bs, uint64_t offset, | |
170f4b2e | 1604 | uint64_t nb_clusters, int flags) |
621f0589 | 1605 | { |
ff99129a | 1606 | BDRVQcow2State *s = bs->opaque; |
621f0589 KW |
1607 | uint64_t *l2_table; |
1608 | int l2_index; | |
1609 | int ret; | |
1610 | int i; | |
06cc5e2b | 1611 | bool unmap = !!(flags & BDRV_REQ_MAY_UNMAP); |
621f0589 KW |
1612 | |
1613 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); | |
1614 | if (ret < 0) { | |
1615 | return ret; | |
1616 | } | |
1617 | ||
1618 | /* Limit nb_clusters to one L2 table */ | |
1619 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1620 | assert(nb_clusters <= INT_MAX); |
621f0589 KW |
1621 | |
1622 | for (i = 0; i < nb_clusters; i++) { | |
1623 | uint64_t old_offset; | |
06cc5e2b | 1624 | QCow2ClusterType cluster_type; |
621f0589 KW |
1625 | |
1626 | old_offset = be64_to_cpu(l2_table[l2_index + i]); | |
1627 | ||
06cc5e2b EB |
1628 | /* |
1629 | * Minimize L2 changes if the cluster already reads back as | |
1630 | * zeroes with correct allocation. | |
1631 | */ | |
1632 | cluster_type = qcow2_get_cluster_type(old_offset); | |
1633 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN || | |
1634 | (cluster_type == QCOW2_CLUSTER_ZERO_ALLOC && !unmap)) { | |
1635 | continue; | |
1636 | } | |
1637 | ||
72e80b89 | 1638 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
06cc5e2b | 1639 | if (cluster_type == QCOW2_CLUSTER_COMPRESSED || unmap) { |
621f0589 | 1640 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
6cfcb9b8 | 1641 | qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST); |
621f0589 KW |
1642 | } else { |
1643 | l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO); | |
1644 | } | |
1645 | } | |
1646 | ||
a3f1afb4 | 1647 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
621f0589 KW |
1648 | |
1649 | return nb_clusters; | |
1650 | } | |
1651 | ||
d2cb36af EB |
1652 | int qcow2_cluster_zeroize(BlockDriverState *bs, uint64_t offset, |
1653 | uint64_t bytes, int flags) | |
621f0589 | 1654 | { |
ff99129a | 1655 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1656 | uint64_t end_offset = offset + bytes; |
b6d36def | 1657 | uint64_t nb_clusters; |
d2cb36af | 1658 | int64_t cleared; |
621f0589 KW |
1659 | int ret; |
1660 | ||
f10ee139 EB |
1661 | /* Caller must pass aligned values, except at image end */ |
1662 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); | |
1663 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || | |
1664 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
1665 | ||
621f0589 KW |
1666 | /* The zero flag is only supported by version 3 and newer */ |
1667 | if (s->qcow_version < 3) { | |
1668 | return -ENOTSUP; | |
1669 | } | |
1670 | ||
1671 | /* Each L2 table is handled by its own loop iteration */ | |
d2cb36af | 1672 | nb_clusters = size_to_clusters(s, bytes); |
621f0589 | 1673 | |
0b919fae KW |
1674 | s->cache_discards = true; |
1675 | ||
621f0589 | 1676 | while (nb_clusters > 0) { |
d2cb36af EB |
1677 | cleared = zero_single_l2(bs, offset, nb_clusters, flags); |
1678 | if (cleared < 0) { | |
1679 | ret = cleared; | |
0b919fae | 1680 | goto fail; |
621f0589 KW |
1681 | } |
1682 | ||
d2cb36af EB |
1683 | nb_clusters -= cleared; |
1684 | offset += (cleared * s->cluster_size); | |
621f0589 KW |
1685 | } |
1686 | ||
0b919fae KW |
1687 | ret = 0; |
1688 | fail: | |
1689 | s->cache_discards = false; | |
1690 | qcow2_process_discards(bs, ret); | |
1691 | ||
1692 | return ret; | |
621f0589 | 1693 | } |
32b6444d HR |
1694 | |
1695 | /* | |
1696 | * Expands all zero clusters in a specific L1 table (or deallocates them, for | |
1697 | * non-backed non-pre-allocated zero clusters). | |
1698 | * | |
4057a2b2 HR |
1699 | * l1_entries and *visited_l1_entries are used to keep track of progress for |
1700 | * status_cb(). l1_entries contains the total number of L1 entries and | |
1701 | * *visited_l1_entries counts all visited L1 entries. | |
32b6444d HR |
1702 | */ |
1703 | static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table, | |
ecf58777 | 1704 | int l1_size, int64_t *visited_l1_entries, |
4057a2b2 | 1705 | int64_t l1_entries, |
8b13976d HR |
1706 | BlockDriverAmendStatusCB *status_cb, |
1707 | void *cb_opaque) | |
32b6444d | 1708 | { |
ff99129a | 1709 | BDRVQcow2State *s = bs->opaque; |
32b6444d HR |
1710 | bool is_active_l1 = (l1_table == s->l1_table); |
1711 | uint64_t *l2_table = NULL; | |
1712 | int ret; | |
1713 | int i, j; | |
1714 | ||
1715 | if (!is_active_l1) { | |
1716 | /* inactive L2 tables require a buffer to be stored in when loading | |
1717 | * them from disk */ | |
9a4f4c31 | 1718 | l2_table = qemu_try_blockalign(bs->file->bs, s->cluster_size); |
de82815d KW |
1719 | if (l2_table == NULL) { |
1720 | return -ENOMEM; | |
1721 | } | |
32b6444d HR |
1722 | } |
1723 | ||
1724 | for (i = 0; i < l1_size; i++) { | |
1725 | uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK; | |
1726 | bool l2_dirty = false; | |
0e06528e | 1727 | uint64_t l2_refcount; |
32b6444d HR |
1728 | |
1729 | if (!l2_offset) { | |
1730 | /* unallocated */ | |
4057a2b2 HR |
1731 | (*visited_l1_entries)++; |
1732 | if (status_cb) { | |
8b13976d | 1733 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 1734 | } |
32b6444d HR |
1735 | continue; |
1736 | } | |
1737 | ||
8dd93d93 HR |
1738 | if (offset_into_cluster(s, l2_offset)) { |
1739 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" | |
1740 | PRIx64 " unaligned (L1 index: %#x)", | |
1741 | l2_offset, i); | |
1742 | ret = -EIO; | |
1743 | goto fail; | |
1744 | } | |
1745 | ||
32b6444d HR |
1746 | if (is_active_l1) { |
1747 | /* get active L2 tables from cache */ | |
1748 | ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, | |
1749 | (void **)&l2_table); | |
1750 | } else { | |
1751 | /* load inactive L2 tables from disk */ | |
fbcbbf4e | 1752 | ret = bdrv_read(bs->file, l2_offset / BDRV_SECTOR_SIZE, |
9a4f4c31 | 1753 | (void *)l2_table, s->cluster_sectors); |
32b6444d HR |
1754 | } |
1755 | if (ret < 0) { | |
1756 | goto fail; | |
1757 | } | |
1758 | ||
7324c10f HR |
1759 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
1760 | &l2_refcount); | |
1761 | if (ret < 0) { | |
ecf58777 HR |
1762 | goto fail; |
1763 | } | |
1764 | ||
32b6444d HR |
1765 | for (j = 0; j < s->l2_size; j++) { |
1766 | uint64_t l2_entry = be64_to_cpu(l2_table[j]); | |
ecf58777 | 1767 | int64_t offset = l2_entry & L2E_OFFSET_MASK; |
3ef95218 | 1768 | QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry); |
32b6444d | 1769 | |
fdfab37d EB |
1770 | if (cluster_type != QCOW2_CLUSTER_ZERO_PLAIN && |
1771 | cluster_type != QCOW2_CLUSTER_ZERO_ALLOC) { | |
32b6444d HR |
1772 | continue; |
1773 | } | |
1774 | ||
fdfab37d | 1775 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
760e0063 | 1776 | if (!bs->backing) { |
32b6444d HR |
1777 | /* not backed; therefore we can simply deallocate the |
1778 | * cluster */ | |
1779 | l2_table[j] = 0; | |
1780 | l2_dirty = true; | |
1781 | continue; | |
1782 | } | |
1783 | ||
1784 | offset = qcow2_alloc_clusters(bs, s->cluster_size); | |
1785 | if (offset < 0) { | |
1786 | ret = offset; | |
1787 | goto fail; | |
1788 | } | |
ecf58777 HR |
1789 | |
1790 | if (l2_refcount > 1) { | |
1791 | /* For shared L2 tables, set the refcount accordingly (it is | |
1792 | * already 1 and needs to be l2_refcount) */ | |
1793 | ret = qcow2_update_cluster_refcount(bs, | |
2aabe7c7 HR |
1794 | offset >> s->cluster_bits, |
1795 | refcount_diff(1, l2_refcount), false, | |
ecf58777 HR |
1796 | QCOW2_DISCARD_OTHER); |
1797 | if (ret < 0) { | |
1798 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1799 | QCOW2_DISCARD_OTHER); | |
1800 | goto fail; | |
1801 | } | |
1802 | } | |
32b6444d HR |
1803 | } |
1804 | ||
8dd93d93 | 1805 | if (offset_into_cluster(s, offset)) { |
bcb07dba EB |
1806 | qcow2_signal_corruption(bs, true, -1, -1, |
1807 | "Cluster allocation offset " | |
8dd93d93 HR |
1808 | "%#" PRIx64 " unaligned (L2 offset: %#" |
1809 | PRIx64 ", L2 index: %#x)", offset, | |
1810 | l2_offset, j); | |
fdfab37d | 1811 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
8dd93d93 HR |
1812 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1813 | QCOW2_DISCARD_ALWAYS); | |
1814 | } | |
1815 | ret = -EIO; | |
1816 | goto fail; | |
1817 | } | |
1818 | ||
231bb267 | 1819 | ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size); |
32b6444d | 1820 | if (ret < 0) { |
fdfab37d | 1821 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
320c7066 HR |
1822 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1823 | QCOW2_DISCARD_ALWAYS); | |
1824 | } | |
32b6444d HR |
1825 | goto fail; |
1826 | } | |
1827 | ||
720ff280 | 1828 | ret = bdrv_pwrite_zeroes(bs->file, offset, s->cluster_size, 0); |
32b6444d | 1829 | if (ret < 0) { |
fdfab37d | 1830 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
320c7066 HR |
1831 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1832 | QCOW2_DISCARD_ALWAYS); | |
1833 | } | |
32b6444d HR |
1834 | goto fail; |
1835 | } | |
1836 | ||
ecf58777 HR |
1837 | if (l2_refcount == 1) { |
1838 | l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED); | |
1839 | } else { | |
1840 | l2_table[j] = cpu_to_be64(offset); | |
e390cf5a | 1841 | } |
ecf58777 | 1842 | l2_dirty = true; |
32b6444d HR |
1843 | } |
1844 | ||
1845 | if (is_active_l1) { | |
1846 | if (l2_dirty) { | |
72e80b89 | 1847 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
32b6444d HR |
1848 | qcow2_cache_depends_on_flush(s->l2_table_cache); |
1849 | } | |
a3f1afb4 | 1850 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
32b6444d HR |
1851 | } else { |
1852 | if (l2_dirty) { | |
231bb267 HR |
1853 | ret = qcow2_pre_write_overlap_check(bs, |
1854 | QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, l2_offset, | |
32b6444d HR |
1855 | s->cluster_size); |
1856 | if (ret < 0) { | |
1857 | goto fail; | |
1858 | } | |
1859 | ||
18d51c4b | 1860 | ret = bdrv_write(bs->file, l2_offset / BDRV_SECTOR_SIZE, |
9a4f4c31 | 1861 | (void *)l2_table, s->cluster_sectors); |
32b6444d HR |
1862 | if (ret < 0) { |
1863 | goto fail; | |
1864 | } | |
1865 | } | |
1866 | } | |
4057a2b2 HR |
1867 | |
1868 | (*visited_l1_entries)++; | |
1869 | if (status_cb) { | |
8b13976d | 1870 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 1871 | } |
32b6444d HR |
1872 | } |
1873 | ||
1874 | ret = 0; | |
1875 | ||
1876 | fail: | |
1877 | if (l2_table) { | |
1878 | if (!is_active_l1) { | |
1879 | qemu_vfree(l2_table); | |
1880 | } else { | |
a3f1afb4 | 1881 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
32b6444d HR |
1882 | } |
1883 | } | |
1884 | return ret; | |
1885 | } | |
1886 | ||
1887 | /* | |
1888 | * For backed images, expands all zero clusters on the image. For non-backed | |
1889 | * images, deallocates all non-pre-allocated zero clusters (and claims the | |
1890 | * allocation for pre-allocated ones). This is important for downgrading to a | |
1891 | * qcow2 version which doesn't yet support metadata zero clusters. | |
1892 | */ | |
4057a2b2 | 1893 | int qcow2_expand_zero_clusters(BlockDriverState *bs, |
8b13976d HR |
1894 | BlockDriverAmendStatusCB *status_cb, |
1895 | void *cb_opaque) | |
32b6444d | 1896 | { |
ff99129a | 1897 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 1898 | uint64_t *l1_table = NULL; |
4057a2b2 | 1899 | int64_t l1_entries = 0, visited_l1_entries = 0; |
32b6444d HR |
1900 | int ret; |
1901 | int i, j; | |
1902 | ||
4057a2b2 HR |
1903 | if (status_cb) { |
1904 | l1_entries = s->l1_size; | |
1905 | for (i = 0; i < s->nb_snapshots; i++) { | |
1906 | l1_entries += s->snapshots[i].l1_size; | |
1907 | } | |
1908 | } | |
1909 | ||
32b6444d | 1910 | ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size, |
4057a2b2 | 1911 | &visited_l1_entries, l1_entries, |
8b13976d | 1912 | status_cb, cb_opaque); |
32b6444d HR |
1913 | if (ret < 0) { |
1914 | goto fail; | |
1915 | } | |
1916 | ||
1917 | /* Inactive L1 tables may point to active L2 tables - therefore it is | |
1918 | * necessary to flush the L2 table cache before trying to access the L2 | |
1919 | * tables pointed to by inactive L1 entries (else we might try to expand | |
1920 | * zero clusters that have already been expanded); furthermore, it is also | |
1921 | * necessary to empty the L2 table cache, since it may contain tables which | |
1922 | * are now going to be modified directly on disk, bypassing the cache. | |
1923 | * qcow2_cache_empty() does both for us. */ | |
1924 | ret = qcow2_cache_empty(bs, s->l2_table_cache); | |
1925 | if (ret < 0) { | |
1926 | goto fail; | |
1927 | } | |
1928 | ||
1929 | for (i = 0; i < s->nb_snapshots; i++) { | |
d737b78c LV |
1930 | int l1_sectors = DIV_ROUND_UP(s->snapshots[i].l1_size * |
1931 | sizeof(uint64_t), BDRV_SECTOR_SIZE); | |
32b6444d HR |
1932 | |
1933 | l1_table = g_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE); | |
1934 | ||
fbcbbf4e | 1935 | ret = bdrv_read(bs->file, |
9a4f4c31 KW |
1936 | s->snapshots[i].l1_table_offset / BDRV_SECTOR_SIZE, |
1937 | (void *)l1_table, l1_sectors); | |
32b6444d HR |
1938 | if (ret < 0) { |
1939 | goto fail; | |
1940 | } | |
1941 | ||
1942 | for (j = 0; j < s->snapshots[i].l1_size; j++) { | |
1943 | be64_to_cpus(&l1_table[j]); | |
1944 | } | |
1945 | ||
1946 | ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size, | |
4057a2b2 | 1947 | &visited_l1_entries, l1_entries, |
8b13976d | 1948 | status_cb, cb_opaque); |
32b6444d HR |
1949 | if (ret < 0) { |
1950 | goto fail; | |
1951 | } | |
1952 | } | |
1953 | ||
1954 | ret = 0; | |
1955 | ||
1956 | fail: | |
32b6444d HR |
1957 | g_free(l1_table); |
1958 | return ret; | |
1959 | } |