]>
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, | |
409 | int offset_in_cluster, | |
410 | int bytes) | |
45aba42f | 411 | { |
ff99129a | 412 | BDRVQcow2State *s = bs->opaque; |
aef4acb6 SH |
413 | QEMUIOVector qiov; |
414 | struct iovec iov; | |
aaa4d20b | 415 | int ret; |
1b9f1491 | 416 | |
aaa4d20b | 417 | iov.iov_len = bytes; |
de82815d KW |
418 | iov.iov_base = qemu_try_blockalign(bs, iov.iov_len); |
419 | if (iov.iov_base == NULL) { | |
420 | return -ENOMEM; | |
421 | } | |
aef4acb6 SH |
422 | |
423 | qemu_iovec_init_external(&qiov, &iov, 1); | |
1b9f1491 | 424 | |
66f82cee | 425 | BLKDBG_EVENT(bs->file, BLKDBG_COW_READ); |
aef4acb6 | 426 | |
dba28555 | 427 | if (!bs->drv) { |
bd604369 KW |
428 | ret = -ENOMEDIUM; |
429 | goto out; | |
dba28555 HR |
430 | } |
431 | ||
aef4acb6 SH |
432 | /* Call .bdrv_co_readv() directly instead of using the public block-layer |
433 | * interface. This avoids double I/O throttling and request tracking, | |
434 | * which can lead to deadlock when block layer copy-on-read is enabled. | |
435 | */ | |
aaa4d20b KW |
436 | ret = bs->drv->bdrv_co_preadv(bs, src_cluster_offset + offset_in_cluster, |
437 | bytes, &qiov, 0); | |
1b9f1491 KW |
438 | if (ret < 0) { |
439 | goto out; | |
440 | } | |
441 | ||
8336aafa | 442 | if (bs->encrypted) { |
f6fa64f6 | 443 | Error *err = NULL; |
bb9f8dd0 | 444 | int64_t sector = (src_cluster_offset + offset_in_cluster) |
aaa4d20b | 445 | >> BDRV_SECTOR_BITS; |
f6fa64f6 | 446 | assert(s->cipher); |
aaa4d20b KW |
447 | assert((offset_in_cluster & ~BDRV_SECTOR_MASK) == 0); |
448 | assert((bytes & ~BDRV_SECTOR_MASK) == 0); | |
449 | if (qcow2_encrypt_sectors(s, sector, iov.iov_base, iov.iov_base, | |
450 | bytes >> BDRV_SECTOR_BITS, true, &err) < 0) { | |
f6fa64f6 DB |
451 | ret = -EIO; |
452 | error_free(err); | |
453 | goto out; | |
454 | } | |
45aba42f | 455 | } |
1b9f1491 | 456 | |
231bb267 | 457 | ret = qcow2_pre_write_overlap_check(bs, 0, |
aaa4d20b | 458 | cluster_offset + offset_in_cluster, bytes); |
cf93980e HR |
459 | if (ret < 0) { |
460 | goto out; | |
461 | } | |
462 | ||
66f82cee | 463 | BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE); |
a03ef88f | 464 | ret = bdrv_co_pwritev(bs->file, cluster_offset + offset_in_cluster, |
aaa4d20b | 465 | bytes, &qiov, 0); |
1b9f1491 KW |
466 | if (ret < 0) { |
467 | goto out; | |
468 | } | |
469 | ||
470 | ret = 0; | |
471 | out: | |
aef4acb6 | 472 | qemu_vfree(iov.iov_base); |
1b9f1491 | 473 | return ret; |
45aba42f KW |
474 | } |
475 | ||
476 | ||
477 | /* | |
478 | * get_cluster_offset | |
479 | * | |
ecfe1863 KW |
480 | * For a given offset of the virtual disk, find the cluster type and offset in |
481 | * the qcow2 file. The offset is stored in *cluster_offset. | |
45aba42f | 482 | * |
ecfe1863 KW |
483 | * On entry, *bytes is the maximum number of contiguous bytes starting at |
484 | * offset that we are interested in. | |
45aba42f | 485 | * |
ecfe1863 KW |
486 | * On exit, *bytes is the number of bytes starting at offset that have the same |
487 | * cluster type and (if applicable) are stored contiguously in the image file. | |
488 | * Compressed clusters are always returned one by one. | |
45aba42f | 489 | * |
68d000a3 KW |
490 | * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error |
491 | * cases. | |
45aba42f | 492 | */ |
1c46efaa | 493 | int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, |
ecfe1863 | 494 | unsigned int *bytes, uint64_t *cluster_offset) |
45aba42f | 495 | { |
ff99129a | 496 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
497 | unsigned int l2_index; |
498 | uint64_t l1_index, l2_offset, *l2_table; | |
45aba42f | 499 | int l1_bits, c; |
c834cba9 HR |
500 | unsigned int offset_in_cluster; |
501 | uint64_t bytes_available, bytes_needed, nb_clusters; | |
3ef95218 | 502 | QCow2ClusterType type; |
55c17e98 | 503 | int ret; |
45aba42f | 504 | |
b2f65d6b | 505 | offset_in_cluster = offset_into_cluster(s, offset); |
ecfe1863 | 506 | bytes_needed = (uint64_t) *bytes + offset_in_cluster; |
45aba42f | 507 | |
b2f65d6b | 508 | l1_bits = s->l2_bits + s->cluster_bits; |
45aba42f | 509 | |
b2f65d6b KW |
510 | /* compute how many bytes there are between the start of the cluster |
511 | * containing offset and the end of the l1 entry */ | |
512 | bytes_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1)) | |
513 | + offset_in_cluster; | |
45aba42f | 514 | |
b2f65d6b KW |
515 | if (bytes_needed > bytes_available) { |
516 | bytes_needed = bytes_available; | |
45aba42f KW |
517 | } |
518 | ||
1c46efaa | 519 | *cluster_offset = 0; |
45aba42f | 520 | |
b6af0975 | 521 | /* seek to the l2 offset in the l1 table */ |
45aba42f KW |
522 | |
523 | l1_index = offset >> l1_bits; | |
68d000a3 | 524 | if (l1_index >= s->l1_size) { |
3ef95218 | 525 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 526 | goto out; |
68d000a3 | 527 | } |
45aba42f | 528 | |
68d000a3 KW |
529 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
530 | if (!l2_offset) { | |
3ef95218 | 531 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 532 | goto out; |
68d000a3 | 533 | } |
45aba42f | 534 | |
a97c67ee HR |
535 | if (offset_into_cluster(s, l2_offset)) { |
536 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
537 | " unaligned (L1 index: %#" PRIx64 ")", | |
538 | l2_offset, l1_index); | |
539 | return -EIO; | |
540 | } | |
541 | ||
45aba42f KW |
542 | /* load the l2 table in memory */ |
543 | ||
55c17e98 KW |
544 | ret = l2_load(bs, l2_offset, &l2_table); |
545 | if (ret < 0) { | |
546 | return ret; | |
1c46efaa | 547 | } |
45aba42f KW |
548 | |
549 | /* find the cluster offset for the given disk offset */ | |
550 | ||
551 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
1c46efaa | 552 | *cluster_offset = be64_to_cpu(l2_table[l2_index]); |
b6d36def | 553 | |
b2f65d6b | 554 | nb_clusters = size_to_clusters(s, bytes_needed); |
c834cba9 HR |
555 | /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned |
556 | * integers; the minimum cluster size is 512, so this assertion is always | |
557 | * true */ | |
558 | assert(nb_clusters <= INT_MAX); | |
45aba42f | 559 | |
3ef95218 | 560 | type = qcow2_get_cluster_type(*cluster_offset); |
fdfab37d EB |
561 | if (s->qcow_version < 3 && (type == QCOW2_CLUSTER_ZERO_PLAIN || |
562 | type == QCOW2_CLUSTER_ZERO_ALLOC)) { | |
563 | qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found" | |
564 | " in pre-v3 image (L2 offset: %#" PRIx64 | |
565 | ", L2 index: %#x)", l2_offset, l2_index); | |
566 | ret = -EIO; | |
567 | goto fail; | |
568 | } | |
3ef95218 | 569 | switch (type) { |
68d000a3 KW |
570 | case QCOW2_CLUSTER_COMPRESSED: |
571 | /* Compressed clusters can only be processed one by one */ | |
572 | c = 1; | |
573 | *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK; | |
574 | break; | |
fdfab37d | 575 | case QCOW2_CLUSTER_ZERO_PLAIN: |
68d000a3 | 576 | case QCOW2_CLUSTER_UNALLOCATED: |
45aba42f | 577 | /* how many empty clusters ? */ |
4341df8a | 578 | c = count_contiguous_clusters_unallocated(nb_clusters, |
fdfab37d | 579 | &l2_table[l2_index], type); |
68d000a3 KW |
580 | *cluster_offset = 0; |
581 | break; | |
fdfab37d | 582 | case QCOW2_CLUSTER_ZERO_ALLOC: |
68d000a3 | 583 | case QCOW2_CLUSTER_NORMAL: |
45aba42f KW |
584 | /* how many allocated clusters ? */ |
585 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
fdfab37d | 586 | &l2_table[l2_index], QCOW_OFLAG_ZERO); |
68d000a3 | 587 | *cluster_offset &= L2E_OFFSET_MASK; |
a97c67ee | 588 | if (offset_into_cluster(s, *cluster_offset)) { |
fdfab37d EB |
589 | qcow2_signal_corruption(bs, true, -1, -1, |
590 | "Cluster allocation offset %#" | |
a97c67ee HR |
591 | PRIx64 " unaligned (L2 offset: %#" PRIx64 |
592 | ", L2 index: %#x)", *cluster_offset, | |
593 | l2_offset, l2_index); | |
594 | ret = -EIO; | |
595 | goto fail; | |
596 | } | |
68d000a3 | 597 | break; |
1417d7e4 KW |
598 | default: |
599 | abort(); | |
45aba42f KW |
600 | } |
601 | ||
29c1a730 KW |
602 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
603 | ||
c834cba9 | 604 | bytes_available = (int64_t)c * s->cluster_size; |
68d000a3 | 605 | |
45aba42f | 606 | out: |
b2f65d6b KW |
607 | if (bytes_available > bytes_needed) { |
608 | bytes_available = bytes_needed; | |
609 | } | |
45aba42f | 610 | |
c834cba9 HR |
611 | /* bytes_available <= bytes_needed <= *bytes + offset_in_cluster; |
612 | * subtracting offset_in_cluster will therefore definitely yield something | |
613 | * not exceeding UINT_MAX */ | |
614 | assert(bytes_available - offset_in_cluster <= UINT_MAX); | |
ecfe1863 | 615 | *bytes = bytes_available - offset_in_cluster; |
45aba42f | 616 | |
3ef95218 | 617 | return type; |
a97c67ee HR |
618 | |
619 | fail: | |
620 | qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table); | |
621 | return ret; | |
45aba42f KW |
622 | } |
623 | ||
624 | /* | |
625 | * get_cluster_table | |
626 | * | |
627 | * for a given disk offset, load (and allocate if needed) | |
628 | * the l2 table. | |
629 | * | |
630 | * the l2 table offset in the qcow2 file and the cluster index | |
631 | * in the l2 table are given to the caller. | |
632 | * | |
1e3e8f1a | 633 | * Returns 0 on success, -errno in failure case |
45aba42f | 634 | */ |
45aba42f KW |
635 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
636 | uint64_t **new_l2_table, | |
45aba42f KW |
637 | int *new_l2_index) |
638 | { | |
ff99129a | 639 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
640 | unsigned int l2_index; |
641 | uint64_t l1_index, l2_offset; | |
c46e1167 | 642 | uint64_t *l2_table = NULL; |
80ee15a6 | 643 | int ret; |
45aba42f | 644 | |
b6af0975 | 645 | /* seek to the l2 offset in the l1 table */ |
45aba42f KW |
646 | |
647 | l1_index = offset >> (s->l2_bits + s->cluster_bits); | |
648 | if (l1_index >= s->l1_size) { | |
72893756 | 649 | ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
1e3e8f1a KW |
650 | if (ret < 0) { |
651 | return ret; | |
652 | } | |
45aba42f | 653 | } |
8e37f681 | 654 | |
2cf7cfa1 | 655 | assert(l1_index < s->l1_size); |
8e37f681 | 656 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
a97c67ee HR |
657 | if (offset_into_cluster(s, l2_offset)) { |
658 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
659 | " unaligned (L1 index: %#" PRIx64 ")", | |
660 | l2_offset, l1_index); | |
661 | return -EIO; | |
662 | } | |
45aba42f KW |
663 | |
664 | /* seek the l2 table of the given l2 offset */ | |
665 | ||
8e37f681 | 666 | if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) { |
45aba42f | 667 | /* load the l2 table in memory */ |
55c17e98 KW |
668 | ret = l2_load(bs, l2_offset, &l2_table); |
669 | if (ret < 0) { | |
670 | return ret; | |
1e3e8f1a | 671 | } |
45aba42f | 672 | } else { |
16fde5f2 | 673 | /* First allocate a new L2 table (and do COW if needed) */ |
c46e1167 KW |
674 | ret = l2_allocate(bs, l1_index, &l2_table); |
675 | if (ret < 0) { | |
676 | return ret; | |
1e3e8f1a | 677 | } |
16fde5f2 KW |
678 | |
679 | /* Then decrease the refcount of the old table */ | |
680 | if (l2_offset) { | |
6cfcb9b8 KW |
681 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), |
682 | QCOW2_DISCARD_OTHER); | |
16fde5f2 | 683 | } |
45aba42f KW |
684 | } |
685 | ||
686 | /* find the cluster offset for the given disk offset */ | |
687 | ||
688 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
689 | ||
690 | *new_l2_table = l2_table; | |
45aba42f KW |
691 | *new_l2_index = l2_index; |
692 | ||
1e3e8f1a | 693 | return 0; |
45aba42f KW |
694 | } |
695 | ||
696 | /* | |
697 | * alloc_compressed_cluster_offset | |
698 | * | |
699 | * For a given offset of the disk image, return cluster offset in | |
700 | * qcow2 file. | |
701 | * | |
702 | * If the offset is not found, allocate a new compressed cluster. | |
703 | * | |
704 | * Return the cluster offset if successful, | |
705 | * Return 0, otherwise. | |
706 | * | |
707 | */ | |
708 | ||
ed6ccf0f KW |
709 | uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
710 | uint64_t offset, | |
711 | int compressed_size) | |
45aba42f | 712 | { |
ff99129a | 713 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 714 | int l2_index, ret; |
3948d1d4 | 715 | uint64_t *l2_table; |
f4f0d391 | 716 | int64_t cluster_offset; |
45aba42f KW |
717 | int nb_csectors; |
718 | ||
3948d1d4 | 719 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
1e3e8f1a | 720 | if (ret < 0) { |
45aba42f | 721 | return 0; |
1e3e8f1a | 722 | } |
45aba42f | 723 | |
b0b6862e KW |
724 | /* Compression can't overwrite anything. Fail if the cluster was already |
725 | * allocated. */ | |
45aba42f | 726 | cluster_offset = be64_to_cpu(l2_table[l2_index]); |
b0b6862e | 727 | if (cluster_offset & L2E_OFFSET_MASK) { |
8f1efd00 KW |
728 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
729 | return 0; | |
730 | } | |
45aba42f | 731 | |
ed6ccf0f | 732 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
5d757b56 | 733 | if (cluster_offset < 0) { |
29c1a730 | 734 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
5d757b56 KW |
735 | return 0; |
736 | } | |
737 | ||
45aba42f KW |
738 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
739 | (cluster_offset >> 9); | |
740 | ||
741 | cluster_offset |= QCOW_OFLAG_COMPRESSED | | |
742 | ((uint64_t)nb_csectors << s->csize_shift); | |
743 | ||
744 | /* update L2 table */ | |
745 | ||
746 | /* compressed clusters never have the copied flag */ | |
747 | ||
66f82cee | 748 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
72e80b89 | 749 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
45aba42f | 750 | l2_table[l2_index] = cpu_to_be64(cluster_offset); |
a3f1afb4 | 751 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
4c1612d9 | 752 | |
29c1a730 | 753 | return cluster_offset; |
4c1612d9 KW |
754 | } |
755 | ||
593fb83c KW |
756 | static int perform_cow(BlockDriverState *bs, QCowL2Meta *m, Qcow2COWRegion *r) |
757 | { | |
ff99129a | 758 | BDRVQcow2State *s = bs->opaque; |
593fb83c KW |
759 | int ret; |
760 | ||
85567393 | 761 | if (r->nb_bytes == 0) { |
593fb83c KW |
762 | return 0; |
763 | } | |
764 | ||
765 | qemu_co_mutex_unlock(&s->lock); | |
85567393 | 766 | ret = do_perform_cow(bs, m->offset, m->alloc_offset, r->offset, r->nb_bytes); |
593fb83c KW |
767 | qemu_co_mutex_lock(&s->lock); |
768 | ||
769 | if (ret < 0) { | |
770 | return ret; | |
771 | } | |
772 | ||
773 | /* | |
774 | * Before we update the L2 table to actually point to the new cluster, we | |
775 | * need to be sure that the refcounts have been increased and COW was | |
776 | * handled. | |
777 | */ | |
778 | qcow2_cache_depends_on_flush(s->l2_table_cache); | |
779 | ||
780 | return 0; | |
781 | } | |
782 | ||
148da7ea | 783 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) |
45aba42f | 784 | { |
ff99129a | 785 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 786 | int i, j = 0, l2_index, ret; |
593fb83c | 787 | uint64_t *old_cluster, *l2_table; |
250196f1 | 788 | uint64_t cluster_offset = m->alloc_offset; |
45aba42f | 789 | |
3cce16f4 | 790 | trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
f50f88b9 | 791 | assert(m->nb_clusters > 0); |
45aba42f | 792 | |
5839e53b | 793 | old_cluster = g_try_new(uint64_t, m->nb_clusters); |
de82815d KW |
794 | if (old_cluster == NULL) { |
795 | ret = -ENOMEM; | |
796 | goto err; | |
797 | } | |
45aba42f KW |
798 | |
799 | /* copy content of unmodified sectors */ | |
593fb83c KW |
800 | ret = perform_cow(bs, m, &m->cow_start); |
801 | if (ret < 0) { | |
802 | goto err; | |
45aba42f KW |
803 | } |
804 | ||
593fb83c KW |
805 | ret = perform_cow(bs, m, &m->cow_end); |
806 | if (ret < 0) { | |
807 | goto err; | |
29c1a730 KW |
808 | } |
809 | ||
593fb83c | 810 | /* Update L2 table. */ |
74c4510a | 811 | if (s->use_lazy_refcounts) { |
280d3735 KW |
812 | qcow2_mark_dirty(bs); |
813 | } | |
bfe8043e SH |
814 | if (qcow2_need_accurate_refcounts(s)) { |
815 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
816 | s->refcount_block_cache); | |
817 | } | |
280d3735 | 818 | |
3948d1d4 | 819 | ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index); |
1e3e8f1a | 820 | if (ret < 0) { |
45aba42f | 821 | goto err; |
1e3e8f1a | 822 | } |
72e80b89 | 823 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
45aba42f | 824 | |
c01dbccb | 825 | assert(l2_index + m->nb_clusters <= s->l2_size); |
45aba42f KW |
826 | for (i = 0; i < m->nb_clusters; i++) { |
827 | /* if two concurrent writes happen to the same unallocated cluster | |
aaa4d20b KW |
828 | * each write allocates separate cluster and writes data concurrently. |
829 | * The first one to complete updates l2 table with pointer to its | |
830 | * cluster the second one has to do RMW (which is done above by | |
831 | * perform_cow()), update l2 table with its cluster pointer and free | |
832 | * old cluster. This is what this loop does */ | |
833 | if (l2_table[l2_index + i] != 0) { | |
45aba42f | 834 | old_cluster[j++] = l2_table[l2_index + i]; |
aaa4d20b | 835 | } |
45aba42f KW |
836 | |
837 | l2_table[l2_index + i] = cpu_to_be64((cluster_offset + | |
838 | (i << s->cluster_bits)) | QCOW_OFLAG_COPIED); | |
839 | } | |
840 | ||
9f8e668e | 841 | |
a3f1afb4 | 842 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
45aba42f | 843 | |
7ec5e6a4 KW |
844 | /* |
845 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
6cfcb9b8 KW |
846 | * |
847 | * Don't discard clusters that reach a refcount of 0 (e.g. compressed | |
848 | * clusters), the next write will reuse them anyway. | |
7ec5e6a4 | 849 | */ |
564a6b69 | 850 | if (!m->keep_old_clusters && j != 0) { |
7ec5e6a4 | 851 | for (i = 0; i < j; i++) { |
6cfcb9b8 KW |
852 | qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1, |
853 | QCOW2_DISCARD_NEVER); | |
7ec5e6a4 KW |
854 | } |
855 | } | |
45aba42f KW |
856 | |
857 | ret = 0; | |
858 | err: | |
7267c094 | 859 | g_free(old_cluster); |
45aba42f KW |
860 | return ret; |
861 | } | |
862 | ||
bf319ece KW |
863 | /* |
864 | * Returns the number of contiguous clusters that can be used for an allocating | |
865 | * write, but require COW to be performed (this includes yet unallocated space, | |
866 | * which must copy from the backing file) | |
867 | */ | |
ff99129a | 868 | static int count_cow_clusters(BDRVQcow2State *s, int nb_clusters, |
bf319ece KW |
869 | uint64_t *l2_table, int l2_index) |
870 | { | |
143550a8 | 871 | int i; |
bf319ece | 872 | |
143550a8 KW |
873 | for (i = 0; i < nb_clusters; i++) { |
874 | uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]); | |
3ef95218 | 875 | QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry); |
143550a8 KW |
876 | |
877 | switch(cluster_type) { | |
878 | case QCOW2_CLUSTER_NORMAL: | |
879 | if (l2_entry & QCOW_OFLAG_COPIED) { | |
880 | goto out; | |
881 | } | |
bf319ece | 882 | break; |
143550a8 KW |
883 | case QCOW2_CLUSTER_UNALLOCATED: |
884 | case QCOW2_CLUSTER_COMPRESSED: | |
fdfab37d EB |
885 | case QCOW2_CLUSTER_ZERO_PLAIN: |
886 | case QCOW2_CLUSTER_ZERO_ALLOC: | |
bf319ece | 887 | break; |
143550a8 KW |
888 | default: |
889 | abort(); | |
890 | } | |
bf319ece KW |
891 | } |
892 | ||
143550a8 | 893 | out: |
bf319ece KW |
894 | assert(i <= nb_clusters); |
895 | return i; | |
896 | } | |
897 | ||
250196f1 | 898 | /* |
226c3c26 KW |
899 | * Check if there already is an AIO write request in flight which allocates |
900 | * the same cluster. In this case we need to wait until the previous | |
901 | * request has completed and updated the L2 table accordingly. | |
65eb2e35 KW |
902 | * |
903 | * Returns: | |
904 | * 0 if there was no dependency. *cur_bytes indicates the number of | |
905 | * bytes from guest_offset that can be read before the next | |
906 | * dependency must be processed (or the request is complete) | |
907 | * | |
908 | * -EAGAIN if we had to wait for another request, previously gathered | |
909 | * information on cluster allocation may be invalid now. The caller | |
910 | * must start over anyway, so consider *cur_bytes undefined. | |
250196f1 | 911 | */ |
226c3c26 | 912 | static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, |
ecdd5333 | 913 | uint64_t *cur_bytes, QCowL2Meta **m) |
250196f1 | 914 | { |
ff99129a | 915 | BDRVQcow2State *s = bs->opaque; |
250196f1 | 916 | QCowL2Meta *old_alloc; |
65eb2e35 | 917 | uint64_t bytes = *cur_bytes; |
250196f1 | 918 | |
250196f1 KW |
919 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
920 | ||
65eb2e35 KW |
921 | uint64_t start = guest_offset; |
922 | uint64_t end = start + bytes; | |
923 | uint64_t old_start = l2meta_cow_start(old_alloc); | |
924 | uint64_t old_end = l2meta_cow_end(old_alloc); | |
250196f1 | 925 | |
d9d74f41 | 926 | if (end <= old_start || start >= old_end) { |
250196f1 KW |
927 | /* No intersection */ |
928 | } else { | |
929 | if (start < old_start) { | |
930 | /* Stop at the start of a running allocation */ | |
65eb2e35 | 931 | bytes = old_start - start; |
250196f1 | 932 | } else { |
65eb2e35 | 933 | bytes = 0; |
250196f1 KW |
934 | } |
935 | ||
ecdd5333 KW |
936 | /* Stop if already an l2meta exists. After yielding, it wouldn't |
937 | * be valid any more, so we'd have to clean up the old L2Metas | |
938 | * and deal with requests depending on them before starting to | |
939 | * gather new ones. Not worth the trouble. */ | |
940 | if (bytes == 0 && *m) { | |
941 | *cur_bytes = 0; | |
942 | return 0; | |
943 | } | |
944 | ||
65eb2e35 | 945 | if (bytes == 0) { |
250196f1 KW |
946 | /* Wait for the dependency to complete. We need to recheck |
947 | * the free/allocated clusters when we continue. */ | |
1ace7cea | 948 | qemu_co_queue_wait(&old_alloc->dependent_requests, &s->lock); |
250196f1 KW |
949 | return -EAGAIN; |
950 | } | |
951 | } | |
952 | } | |
953 | ||
65eb2e35 KW |
954 | /* Make sure that existing clusters and new allocations are only used up to |
955 | * the next dependency if we shortened the request above */ | |
956 | *cur_bytes = bytes; | |
250196f1 | 957 | |
226c3c26 KW |
958 | return 0; |
959 | } | |
960 | ||
0af729ec KW |
961 | /* |
962 | * Checks how many already allocated clusters that don't require a copy on | |
963 | * write there are at the given guest_offset (up to *bytes). If | |
964 | * *host_offset is not zero, only physically contiguous clusters beginning at | |
965 | * this host offset are counted. | |
966 | * | |
411d62b0 KW |
967 | * Note that guest_offset may not be cluster aligned. In this case, the |
968 | * returned *host_offset points to exact byte referenced by guest_offset and | |
969 | * therefore isn't cluster aligned as well. | |
0af729ec KW |
970 | * |
971 | * Returns: | |
972 | * 0: if no allocated clusters are available at the given offset. | |
973 | * *bytes is normally unchanged. It is set to 0 if the cluster | |
974 | * is allocated and doesn't need COW, but doesn't have the right | |
975 | * physical offset. | |
976 | * | |
977 | * 1: if allocated clusters that don't require a COW are available at | |
978 | * the requested offset. *bytes may have decreased and describes | |
979 | * the length of the area that can be written to. | |
980 | * | |
981 | * -errno: in error cases | |
0af729ec KW |
982 | */ |
983 | static int handle_copied(BlockDriverState *bs, uint64_t guest_offset, | |
c53ede9f | 984 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
0af729ec | 985 | { |
ff99129a | 986 | BDRVQcow2State *s = bs->opaque; |
0af729ec KW |
987 | int l2_index; |
988 | uint64_t cluster_offset; | |
989 | uint64_t *l2_table; | |
b6d36def | 990 | uint64_t nb_clusters; |
c53ede9f | 991 | unsigned int keep_clusters; |
a3f1afb4 | 992 | int ret; |
0af729ec KW |
993 | |
994 | trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset, | |
995 | *bytes); | |
0af729ec | 996 | |
411d62b0 KW |
997 | assert(*host_offset == 0 || offset_into_cluster(s, guest_offset) |
998 | == offset_into_cluster(s, *host_offset)); | |
999 | ||
acb0467f KW |
1000 | /* |
1001 | * Calculate the number of clusters to look for. We stop at L2 table | |
1002 | * boundaries to keep things simple. | |
1003 | */ | |
1004 | nb_clusters = | |
1005 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1006 | ||
1007 | l2_index = offset_to_l2_index(s, guest_offset); | |
1008 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1009 | assert(nb_clusters <= INT_MAX); |
acb0467f | 1010 | |
0af729ec KW |
1011 | /* Find L2 entry for the first involved cluster */ |
1012 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
1013 | if (ret < 0) { | |
1014 | return ret; | |
1015 | } | |
1016 | ||
1017 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
1018 | ||
1019 | /* Check how many clusters are already allocated and don't need COW */ | |
1020 | if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL | |
1021 | && (cluster_offset & QCOW_OFLAG_COPIED)) | |
1022 | { | |
e62daaf6 KW |
1023 | /* If a specific host_offset is required, check it */ |
1024 | bool offset_matches = | |
1025 | (cluster_offset & L2E_OFFSET_MASK) == *host_offset; | |
1026 | ||
a97c67ee HR |
1027 | if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) { |
1028 | qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset " | |
1029 | "%#llx unaligned (guest offset: %#" PRIx64 | |
1030 | ")", cluster_offset & L2E_OFFSET_MASK, | |
1031 | guest_offset); | |
1032 | ret = -EIO; | |
1033 | goto out; | |
1034 | } | |
1035 | ||
e62daaf6 KW |
1036 | if (*host_offset != 0 && !offset_matches) { |
1037 | *bytes = 0; | |
1038 | ret = 0; | |
1039 | goto out; | |
1040 | } | |
1041 | ||
0af729ec | 1042 | /* We keep all QCOW_OFLAG_COPIED clusters */ |
c53ede9f | 1043 | keep_clusters = |
acb0467f | 1044 | count_contiguous_clusters(nb_clusters, s->cluster_size, |
61653008 | 1045 | &l2_table[l2_index], |
0af729ec | 1046 | QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); |
c53ede9f KW |
1047 | assert(keep_clusters <= nb_clusters); |
1048 | ||
1049 | *bytes = MIN(*bytes, | |
1050 | keep_clusters * s->cluster_size | |
1051 | - offset_into_cluster(s, guest_offset)); | |
0af729ec KW |
1052 | |
1053 | ret = 1; | |
1054 | } else { | |
0af729ec KW |
1055 | ret = 0; |
1056 | } | |
1057 | ||
0af729ec | 1058 | /* Cleanup */ |
e62daaf6 | 1059 | out: |
a3f1afb4 | 1060 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
0af729ec | 1061 | |
e62daaf6 KW |
1062 | /* Only return a host offset if we actually made progress. Otherwise we |
1063 | * would make requirements for handle_alloc() that it can't fulfill */ | |
a97c67ee | 1064 | if (ret > 0) { |
411d62b0 KW |
1065 | *host_offset = (cluster_offset & L2E_OFFSET_MASK) |
1066 | + offset_into_cluster(s, guest_offset); | |
e62daaf6 KW |
1067 | } |
1068 | ||
0af729ec KW |
1069 | return ret; |
1070 | } | |
1071 | ||
226c3c26 KW |
1072 | /* |
1073 | * Allocates new clusters for the given guest_offset. | |
1074 | * | |
1075 | * At most *nb_clusters are allocated, and on return *nb_clusters is updated to | |
1076 | * contain the number of clusters that have been allocated and are contiguous | |
1077 | * in the image file. | |
1078 | * | |
1079 | * If *host_offset is non-zero, it specifies the offset in the image file at | |
1080 | * which the new clusters must start. *nb_clusters can be 0 on return in this | |
1081 | * case if the cluster at host_offset is already in use. If *host_offset is | |
1082 | * zero, the clusters can be allocated anywhere in the image file. | |
1083 | * | |
1084 | * *host_offset is updated to contain the offset into the image file at which | |
1085 | * the first allocated cluster starts. | |
1086 | * | |
1087 | * Return 0 on success and -errno in error cases. -EAGAIN means that the | |
1088 | * function has been waiting for another request and the allocation must be | |
1089 | * restarted, but the whole request should not be failed. | |
1090 | */ | |
1091 | static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, | |
b6d36def | 1092 | uint64_t *host_offset, uint64_t *nb_clusters) |
226c3c26 | 1093 | { |
ff99129a | 1094 | BDRVQcow2State *s = bs->opaque; |
226c3c26 KW |
1095 | |
1096 | trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, | |
1097 | *host_offset, *nb_clusters); | |
1098 | ||
250196f1 KW |
1099 | /* Allocate new clusters */ |
1100 | trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); | |
1101 | if (*host_offset == 0) { | |
df021791 KW |
1102 | int64_t cluster_offset = |
1103 | qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); | |
1104 | if (cluster_offset < 0) { | |
1105 | return cluster_offset; | |
1106 | } | |
1107 | *host_offset = cluster_offset; | |
1108 | return 0; | |
250196f1 | 1109 | } else { |
b6d36def | 1110 | int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); |
df021791 KW |
1111 | if (ret < 0) { |
1112 | return ret; | |
1113 | } | |
1114 | *nb_clusters = ret; | |
1115 | return 0; | |
250196f1 | 1116 | } |
250196f1 KW |
1117 | } |
1118 | ||
10f0ed8b KW |
1119 | /* |
1120 | * Allocates new clusters for an area that either is yet unallocated or needs a | |
1121 | * copy on write. If *host_offset is non-zero, clusters are only allocated if | |
1122 | * the new allocation can match the specified host offset. | |
1123 | * | |
411d62b0 KW |
1124 | * Note that guest_offset may not be cluster aligned. In this case, the |
1125 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1126 | * therefore isn't cluster aligned as well. | |
10f0ed8b KW |
1127 | * |
1128 | * Returns: | |
1129 | * 0: if no clusters could be allocated. *bytes is set to 0, | |
1130 | * *host_offset is left unchanged. | |
1131 | * | |
1132 | * 1: if new clusters were allocated. *bytes may be decreased if the | |
1133 | * new allocation doesn't cover all of the requested area. | |
1134 | * *host_offset is updated to contain the host offset of the first | |
1135 | * newly allocated cluster. | |
1136 | * | |
1137 | * -errno: in error cases | |
10f0ed8b KW |
1138 | */ |
1139 | static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, | |
c37f4cd7 | 1140 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
10f0ed8b | 1141 | { |
ff99129a | 1142 | BDRVQcow2State *s = bs->opaque; |
10f0ed8b KW |
1143 | int l2_index; |
1144 | uint64_t *l2_table; | |
1145 | uint64_t entry; | |
b6d36def | 1146 | uint64_t nb_clusters; |
10f0ed8b | 1147 | int ret; |
564a6b69 | 1148 | bool keep_old_clusters = false; |
10f0ed8b | 1149 | |
564a6b69 | 1150 | uint64_t alloc_cluster_offset = 0; |
10f0ed8b KW |
1151 | |
1152 | trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset, | |
1153 | *bytes); | |
1154 | assert(*bytes > 0); | |
1155 | ||
f5bc6350 KW |
1156 | /* |
1157 | * Calculate the number of clusters to look for. We stop at L2 table | |
1158 | * boundaries to keep things simple. | |
1159 | */ | |
c37f4cd7 KW |
1160 | nb_clusters = |
1161 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1162 | ||
f5bc6350 | 1163 | l2_index = offset_to_l2_index(s, guest_offset); |
c37f4cd7 | 1164 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
b6d36def | 1165 | assert(nb_clusters <= INT_MAX); |
f5bc6350 | 1166 | |
10f0ed8b KW |
1167 | /* Find L2 entry for the first involved cluster */ |
1168 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
1169 | if (ret < 0) { | |
1170 | return ret; | |
1171 | } | |
1172 | ||
3b8e2e26 | 1173 | entry = be64_to_cpu(l2_table[l2_index]); |
10f0ed8b KW |
1174 | |
1175 | /* For the moment, overwrite compressed clusters one by one */ | |
1176 | if (entry & QCOW_OFLAG_COMPRESSED) { | |
1177 | nb_clusters = 1; | |
1178 | } else { | |
3b8e2e26 | 1179 | nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index); |
10f0ed8b KW |
1180 | } |
1181 | ||
ecdd5333 KW |
1182 | /* This function is only called when there were no non-COW clusters, so if |
1183 | * we can't find any unallocated or COW clusters either, something is | |
1184 | * wrong with our code. */ | |
1185 | assert(nb_clusters > 0); | |
1186 | ||
fdfab37d EB |
1187 | if (qcow2_get_cluster_type(entry) == QCOW2_CLUSTER_ZERO_ALLOC && |
1188 | (entry & QCOW_OFLAG_COPIED) && | |
564a6b69 HR |
1189 | (!*host_offset || |
1190 | start_of_cluster(s, *host_offset) == (entry & L2E_OFFSET_MASK))) | |
1191 | { | |
1192 | /* Try to reuse preallocated zero clusters; contiguous normal clusters | |
1193 | * would be fine, too, but count_cow_clusters() above has limited | |
1194 | * nb_clusters already to a range of COW clusters */ | |
1195 | int preallocated_nb_clusters = | |
1196 | count_contiguous_clusters(nb_clusters, s->cluster_size, | |
1197 | &l2_table[l2_index], QCOW_OFLAG_COPIED); | |
1198 | assert(preallocated_nb_clusters > 0); | |
10f0ed8b | 1199 | |
564a6b69 HR |
1200 | nb_clusters = preallocated_nb_clusters; |
1201 | alloc_cluster_offset = entry & L2E_OFFSET_MASK; | |
10f0ed8b | 1202 | |
564a6b69 HR |
1203 | /* We want to reuse these clusters, so qcow2_alloc_cluster_link_l2() |
1204 | * should not free them. */ | |
1205 | keep_old_clusters = true; | |
10f0ed8b KW |
1206 | } |
1207 | ||
564a6b69 HR |
1208 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
1209 | ||
ff52aab2 | 1210 | if (!alloc_cluster_offset) { |
564a6b69 HR |
1211 | /* Allocate, if necessary at a given offset in the image file */ |
1212 | alloc_cluster_offset = start_of_cluster(s, *host_offset); | |
1213 | ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset, | |
1214 | &nb_clusters); | |
1215 | if (ret < 0) { | |
1216 | goto fail; | |
1217 | } | |
1218 | ||
1219 | /* Can't extend contiguous allocation */ | |
1220 | if (nb_clusters == 0) { | |
1221 | *bytes = 0; | |
1222 | return 0; | |
1223 | } | |
1224 | ||
1225 | /* !*host_offset would overwrite the image header and is reserved for | |
1226 | * "no host offset preferred". If 0 was a valid host offset, it'd | |
1227 | * trigger the following overlap check; do that now to avoid having an | |
1228 | * invalid value in *host_offset. */ | |
1229 | if (!alloc_cluster_offset) { | |
1230 | ret = qcow2_pre_write_overlap_check(bs, 0, alloc_cluster_offset, | |
1231 | nb_clusters * s->cluster_size); | |
1232 | assert(ret < 0); | |
1233 | goto fail; | |
1234 | } | |
ff52aab2 HR |
1235 | } |
1236 | ||
83baa9a4 KW |
1237 | /* |
1238 | * Save info needed for meta data update. | |
1239 | * | |
85567393 | 1240 | * requested_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1241 | * newly allocated cluster to the end of the (possibly shortened |
1242 | * before) write request. | |
1243 | * | |
85567393 | 1244 | * avail_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1245 | * newly allocated to the end of the last newly allocated cluster. |
1246 | * | |
85567393 | 1247 | * nb_bytes: The number of bytes from the start of the first |
83baa9a4 KW |
1248 | * newly allocated cluster to the end of the area that the write |
1249 | * request actually writes to (excluding COW at the end) | |
1250 | */ | |
85567393 KW |
1251 | uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset); |
1252 | int avail_bytes = MIN(INT_MAX, nb_clusters << s->cluster_bits); | |
1253 | int nb_bytes = MIN(requested_bytes, avail_bytes); | |
88c6588c | 1254 | QCowL2Meta *old_m = *m; |
83baa9a4 | 1255 | |
83baa9a4 KW |
1256 | *m = g_malloc0(sizeof(**m)); |
1257 | ||
1258 | **m = (QCowL2Meta) { | |
88c6588c KW |
1259 | .next = old_m, |
1260 | ||
411d62b0 | 1261 | .alloc_offset = alloc_cluster_offset, |
83baa9a4 KW |
1262 | .offset = start_of_cluster(s, guest_offset), |
1263 | .nb_clusters = nb_clusters, | |
83baa9a4 | 1264 | |
564a6b69 HR |
1265 | .keep_old_clusters = keep_old_clusters, |
1266 | ||
83baa9a4 KW |
1267 | .cow_start = { |
1268 | .offset = 0, | |
85567393 | 1269 | .nb_bytes = offset_into_cluster(s, guest_offset), |
83baa9a4 KW |
1270 | }, |
1271 | .cow_end = { | |
85567393 KW |
1272 | .offset = nb_bytes, |
1273 | .nb_bytes = avail_bytes - nb_bytes, | |
83baa9a4 KW |
1274 | }, |
1275 | }; | |
1276 | qemu_co_queue_init(&(*m)->dependent_requests); | |
1277 | QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); | |
1278 | ||
411d62b0 | 1279 | *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset); |
85567393 | 1280 | *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset)); |
83baa9a4 KW |
1281 | assert(*bytes != 0); |
1282 | ||
10f0ed8b KW |
1283 | return 1; |
1284 | ||
1285 | fail: | |
1286 | if (*m && (*m)->nb_clusters > 0) { | |
1287 | QLIST_REMOVE(*m, next_in_flight); | |
1288 | } | |
1289 | return ret; | |
1290 | } | |
1291 | ||
45aba42f KW |
1292 | /* |
1293 | * alloc_cluster_offset | |
1294 | * | |
250196f1 KW |
1295 | * For a given offset on the virtual disk, find the cluster offset in qcow2 |
1296 | * file. If the offset is not found, allocate a new cluster. | |
45aba42f | 1297 | * |
250196f1 | 1298 | * If the cluster was already allocated, m->nb_clusters is set to 0 and |
a7912369 | 1299 | * other fields in m are meaningless. |
148da7ea KW |
1300 | * |
1301 | * If the cluster is newly allocated, m->nb_clusters is set to the number of | |
68d100e9 KW |
1302 | * contiguous clusters that have been allocated. In this case, the other |
1303 | * fields of m are valid and contain information about the first allocated | |
1304 | * cluster. | |
45aba42f | 1305 | * |
68d100e9 KW |
1306 | * If the request conflicts with another write request in flight, the coroutine |
1307 | * is queued and will be reentered when the dependency has completed. | |
148da7ea KW |
1308 | * |
1309 | * Return 0 on success and -errno in error cases | |
45aba42f | 1310 | */ |
f4f0d391 | 1311 | int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, |
d46a0bb2 KW |
1312 | unsigned int *bytes, uint64_t *host_offset, |
1313 | QCowL2Meta **m) | |
45aba42f | 1314 | { |
ff99129a | 1315 | BDRVQcow2State *s = bs->opaque; |
710c2496 | 1316 | uint64_t start, remaining; |
250196f1 | 1317 | uint64_t cluster_offset; |
65eb2e35 | 1318 | uint64_t cur_bytes; |
710c2496 | 1319 | int ret; |
45aba42f | 1320 | |
d46a0bb2 | 1321 | trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *bytes); |
710c2496 | 1322 | |
72424114 | 1323 | again: |
16f0587e | 1324 | start = offset; |
d46a0bb2 | 1325 | remaining = *bytes; |
0af729ec KW |
1326 | cluster_offset = 0; |
1327 | *host_offset = 0; | |
ecdd5333 KW |
1328 | cur_bytes = 0; |
1329 | *m = NULL; | |
0af729ec | 1330 | |
2c3b32d2 | 1331 | while (true) { |
ecdd5333 KW |
1332 | |
1333 | if (!*host_offset) { | |
1334 | *host_offset = start_of_cluster(s, cluster_offset); | |
1335 | } | |
1336 | ||
1337 | assert(remaining >= cur_bytes); | |
1338 | ||
1339 | start += cur_bytes; | |
1340 | remaining -= cur_bytes; | |
1341 | cluster_offset += cur_bytes; | |
1342 | ||
1343 | if (remaining == 0) { | |
1344 | break; | |
1345 | } | |
1346 | ||
1347 | cur_bytes = remaining; | |
1348 | ||
2c3b32d2 KW |
1349 | /* |
1350 | * Now start gathering as many contiguous clusters as possible: | |
1351 | * | |
1352 | * 1. Check for overlaps with in-flight allocations | |
1353 | * | |
1354 | * a) Overlap not in the first cluster -> shorten this request and | |
1355 | * let the caller handle the rest in its next loop iteration. | |
1356 | * | |
1357 | * b) Real overlaps of two requests. Yield and restart the search | |
1358 | * for contiguous clusters (the situation could have changed | |
1359 | * while we were sleeping) | |
1360 | * | |
1361 | * c) TODO: Request starts in the same cluster as the in-flight | |
1362 | * allocation ends. Shorten the COW of the in-fight allocation, | |
1363 | * set cluster_offset to write to the same cluster and set up | |
1364 | * the right synchronisation between the in-flight request and | |
1365 | * the new one. | |
1366 | */ | |
ecdd5333 | 1367 | ret = handle_dependencies(bs, start, &cur_bytes, m); |
2c3b32d2 | 1368 | if (ret == -EAGAIN) { |
ecdd5333 KW |
1369 | /* Currently handle_dependencies() doesn't yield if we already had |
1370 | * an allocation. If it did, we would have to clean up the L2Meta | |
1371 | * structs before starting over. */ | |
1372 | assert(*m == NULL); | |
2c3b32d2 KW |
1373 | goto again; |
1374 | } else if (ret < 0) { | |
1375 | return ret; | |
ecdd5333 KW |
1376 | } else if (cur_bytes == 0) { |
1377 | break; | |
2c3b32d2 KW |
1378 | } else { |
1379 | /* handle_dependencies() may have decreased cur_bytes (shortened | |
1380 | * the allocations below) so that the next dependency is processed | |
1381 | * correctly during the next loop iteration. */ | |
0af729ec | 1382 | } |
710c2496 | 1383 | |
2c3b32d2 KW |
1384 | /* |
1385 | * 2. Count contiguous COPIED clusters. | |
1386 | */ | |
1387 | ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m); | |
1388 | if (ret < 0) { | |
1389 | return ret; | |
1390 | } else if (ret) { | |
ecdd5333 | 1391 | continue; |
2c3b32d2 KW |
1392 | } else if (cur_bytes == 0) { |
1393 | break; | |
1394 | } | |
060bee89 | 1395 | |
2c3b32d2 KW |
1396 | /* |
1397 | * 3. If the request still hasn't completed, allocate new clusters, | |
1398 | * considering any cluster_offset of steps 1c or 2. | |
1399 | */ | |
1400 | ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m); | |
1401 | if (ret < 0) { | |
1402 | return ret; | |
1403 | } else if (ret) { | |
ecdd5333 | 1404 | continue; |
2c3b32d2 KW |
1405 | } else { |
1406 | assert(cur_bytes == 0); | |
1407 | break; | |
1408 | } | |
f5bc6350 | 1409 | } |
10f0ed8b | 1410 | |
d46a0bb2 KW |
1411 | *bytes -= remaining; |
1412 | assert(*bytes > 0); | |
710c2496 | 1413 | assert(*host_offset != 0); |
45aba42f | 1414 | |
148da7ea | 1415 | return 0; |
45aba42f KW |
1416 | } |
1417 | ||
1418 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
1419 | const uint8_t *buf, int buf_size) | |
1420 | { | |
1421 | z_stream strm1, *strm = &strm1; | |
1422 | int ret, out_len; | |
1423 | ||
1424 | memset(strm, 0, sizeof(*strm)); | |
1425 | ||
1426 | strm->next_in = (uint8_t *)buf; | |
1427 | strm->avail_in = buf_size; | |
1428 | strm->next_out = out_buf; | |
1429 | strm->avail_out = out_buf_size; | |
1430 | ||
1431 | ret = inflateInit2(strm, -12); | |
1432 | if (ret != Z_OK) | |
1433 | return -1; | |
1434 | ret = inflate(strm, Z_FINISH); | |
1435 | out_len = strm->next_out - out_buf; | |
1436 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
1437 | out_len != out_buf_size) { | |
1438 | inflateEnd(strm); | |
1439 | return -1; | |
1440 | } | |
1441 | inflateEnd(strm); | |
1442 | return 0; | |
1443 | } | |
1444 | ||
66f82cee | 1445 | int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) |
45aba42f | 1446 | { |
ff99129a | 1447 | BDRVQcow2State *s = bs->opaque; |
45aba42f KW |
1448 | int ret, csize, nb_csectors, sector_offset; |
1449 | uint64_t coffset; | |
1450 | ||
1451 | coffset = cluster_offset & s->cluster_offset_mask; | |
1452 | if (s->cluster_cache_offset != coffset) { | |
1453 | nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1; | |
1454 | sector_offset = coffset & 511; | |
1455 | csize = nb_csectors * 512 - sector_offset; | |
66f82cee | 1456 | BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
fbcbbf4e | 1457 | ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, |
9a4f4c31 | 1458 | nb_csectors); |
45aba42f | 1459 | if (ret < 0) { |
8af36488 | 1460 | return ret; |
45aba42f KW |
1461 | } |
1462 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
1463 | s->cluster_data + sector_offset, csize) < 0) { | |
8af36488 | 1464 | return -EIO; |
45aba42f KW |
1465 | } |
1466 | s->cluster_cache_offset = coffset; | |
1467 | } | |
1468 | return 0; | |
1469 | } | |
5ea929e3 KW |
1470 | |
1471 | /* | |
1472 | * This discards as many clusters of nb_clusters as possible at once (i.e. | |
1473 | * all clusters in the same L2 table) and returns the number of discarded | |
1474 | * clusters. | |
1475 | */ | |
1476 | static int discard_single_l2(BlockDriverState *bs, uint64_t offset, | |
b6d36def HR |
1477 | uint64_t nb_clusters, enum qcow2_discard_type type, |
1478 | bool full_discard) | |
5ea929e3 | 1479 | { |
ff99129a | 1480 | BDRVQcow2State *s = bs->opaque; |
3948d1d4 | 1481 | uint64_t *l2_table; |
5ea929e3 KW |
1482 | int l2_index; |
1483 | int ret; | |
1484 | int i; | |
1485 | ||
3948d1d4 | 1486 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
5ea929e3 KW |
1487 | if (ret < 0) { |
1488 | return ret; | |
1489 | } | |
1490 | ||
1491 | /* Limit nb_clusters to one L2 table */ | |
1492 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1493 | assert(nb_clusters <= INT_MAX); |
5ea929e3 KW |
1494 | |
1495 | for (i = 0; i < nb_clusters; i++) { | |
c883db0d | 1496 | uint64_t old_l2_entry; |
5ea929e3 | 1497 | |
c883db0d | 1498 | old_l2_entry = be64_to_cpu(l2_table[l2_index + i]); |
a71835a0 KW |
1499 | |
1500 | /* | |
808c4b6f HR |
1501 | * If full_discard is false, make sure that a discarded area reads back |
1502 | * as zeroes for v3 images (we cannot do it for v2 without actually | |
1503 | * writing a zero-filled buffer). We can skip the operation if the | |
1504 | * cluster is already marked as zero, or if it's unallocated and we | |
1505 | * don't have a backing file. | |
a71835a0 KW |
1506 | * |
1507 | * TODO We might want to use bdrv_get_block_status(bs) here, but we're | |
1508 | * holding s->lock, so that doesn't work today. | |
808c4b6f HR |
1509 | * |
1510 | * If full_discard is true, the sector should not read back as zeroes, | |
1511 | * but rather fall through to the backing file. | |
a71835a0 | 1512 | */ |
c883db0d | 1513 | switch (qcow2_get_cluster_type(old_l2_entry)) { |
bbd995d8 EB |
1514 | case QCOW2_CLUSTER_UNALLOCATED: |
1515 | if (full_discard || !bs->backing) { | |
1516 | continue; | |
1517 | } | |
1518 | break; | |
1519 | ||
fdfab37d EB |
1520 | case QCOW2_CLUSTER_ZERO_PLAIN: |
1521 | if (!full_discard) { | |
bbd995d8 EB |
1522 | continue; |
1523 | } | |
1524 | break; | |
1525 | ||
fdfab37d | 1526 | case QCOW2_CLUSTER_ZERO_ALLOC: |
bbd995d8 EB |
1527 | case QCOW2_CLUSTER_NORMAL: |
1528 | case QCOW2_CLUSTER_COMPRESSED: | |
1529 | break; | |
1530 | ||
1531 | default: | |
1532 | abort(); | |
5ea929e3 KW |
1533 | } |
1534 | ||
1535 | /* First remove L2 entries */ | |
72e80b89 | 1536 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
808c4b6f | 1537 | if (!full_discard && s->qcow_version >= 3) { |
a71835a0 KW |
1538 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
1539 | } else { | |
1540 | l2_table[l2_index + i] = cpu_to_be64(0); | |
1541 | } | |
5ea929e3 KW |
1542 | |
1543 | /* Then decrease the refcount */ | |
c883db0d | 1544 | qcow2_free_any_clusters(bs, old_l2_entry, 1, type); |
5ea929e3 KW |
1545 | } |
1546 | ||
a3f1afb4 | 1547 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
5ea929e3 KW |
1548 | |
1549 | return nb_clusters; | |
1550 | } | |
1551 | ||
1552 | int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset, | |
808c4b6f | 1553 | int nb_sectors, enum qcow2_discard_type type, bool full_discard) |
5ea929e3 | 1554 | { |
ff99129a | 1555 | BDRVQcow2State *s = bs->opaque; |
5ea929e3 | 1556 | uint64_t end_offset; |
b6d36def | 1557 | uint64_t nb_clusters; |
5ea929e3 KW |
1558 | int ret; |
1559 | ||
1560 | end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS); | |
1561 | ||
0c1bd469 EB |
1562 | /* The caller must cluster-align start; round end down except at EOF */ |
1563 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); | |
1564 | if (end_offset != bs->total_sectors * BDRV_SECTOR_SIZE) { | |
1565 | end_offset = start_of_cluster(s, end_offset); | |
5ea929e3 KW |
1566 | } |
1567 | ||
1568 | nb_clusters = size_to_clusters(s, end_offset - offset); | |
1569 | ||
0b919fae KW |
1570 | s->cache_discards = true; |
1571 | ||
5ea929e3 KW |
1572 | /* Each L2 table is handled by its own loop iteration */ |
1573 | while (nb_clusters > 0) { | |
808c4b6f | 1574 | ret = discard_single_l2(bs, offset, nb_clusters, type, full_discard); |
5ea929e3 | 1575 | if (ret < 0) { |
0b919fae | 1576 | goto fail; |
5ea929e3 KW |
1577 | } |
1578 | ||
1579 | nb_clusters -= ret; | |
1580 | offset += (ret * s->cluster_size); | |
1581 | } | |
1582 | ||
0b919fae KW |
1583 | ret = 0; |
1584 | fail: | |
1585 | s->cache_discards = false; | |
1586 | qcow2_process_discards(bs, ret); | |
1587 | ||
1588 | return ret; | |
5ea929e3 | 1589 | } |
621f0589 KW |
1590 | |
1591 | /* | |
1592 | * This zeroes as many clusters of nb_clusters as possible at once (i.e. | |
1593 | * all clusters in the same L2 table) and returns the number of zeroed | |
1594 | * clusters. | |
1595 | */ | |
1596 | static int zero_single_l2(BlockDriverState *bs, uint64_t offset, | |
170f4b2e | 1597 | uint64_t nb_clusters, int flags) |
621f0589 | 1598 | { |
ff99129a | 1599 | BDRVQcow2State *s = bs->opaque; |
621f0589 KW |
1600 | uint64_t *l2_table; |
1601 | int l2_index; | |
1602 | int ret; | |
1603 | int i; | |
1604 | ||
1605 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); | |
1606 | if (ret < 0) { | |
1607 | return ret; | |
1608 | } | |
1609 | ||
1610 | /* Limit nb_clusters to one L2 table */ | |
1611 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1612 | assert(nb_clusters <= INT_MAX); |
621f0589 KW |
1613 | |
1614 | for (i = 0; i < nb_clusters; i++) { | |
1615 | uint64_t old_offset; | |
1616 | ||
1617 | old_offset = be64_to_cpu(l2_table[l2_index + i]); | |
1618 | ||
1619 | /* Update L2 entries */ | |
72e80b89 | 1620 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
170f4b2e | 1621 | if (old_offset & QCOW_OFLAG_COMPRESSED || flags & BDRV_REQ_MAY_UNMAP) { |
621f0589 | 1622 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
6cfcb9b8 | 1623 | qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST); |
621f0589 KW |
1624 | } else { |
1625 | l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO); | |
1626 | } | |
1627 | } | |
1628 | ||
a3f1afb4 | 1629 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
621f0589 KW |
1630 | |
1631 | return nb_clusters; | |
1632 | } | |
1633 | ||
170f4b2e FZ |
1634 | int qcow2_zero_clusters(BlockDriverState *bs, uint64_t offset, int nb_sectors, |
1635 | int flags) | |
621f0589 | 1636 | { |
ff99129a | 1637 | BDRVQcow2State *s = bs->opaque; |
b6d36def | 1638 | uint64_t nb_clusters; |
621f0589 KW |
1639 | int ret; |
1640 | ||
1641 | /* The zero flag is only supported by version 3 and newer */ | |
1642 | if (s->qcow_version < 3) { | |
1643 | return -ENOTSUP; | |
1644 | } | |
1645 | ||
1646 | /* Each L2 table is handled by its own loop iteration */ | |
1647 | nb_clusters = size_to_clusters(s, nb_sectors << BDRV_SECTOR_BITS); | |
1648 | ||
0b919fae KW |
1649 | s->cache_discards = true; |
1650 | ||
621f0589 | 1651 | while (nb_clusters > 0) { |
170f4b2e | 1652 | ret = zero_single_l2(bs, offset, nb_clusters, flags); |
621f0589 | 1653 | if (ret < 0) { |
0b919fae | 1654 | goto fail; |
621f0589 KW |
1655 | } |
1656 | ||
1657 | nb_clusters -= ret; | |
1658 | offset += (ret * s->cluster_size); | |
1659 | } | |
1660 | ||
0b919fae KW |
1661 | ret = 0; |
1662 | fail: | |
1663 | s->cache_discards = false; | |
1664 | qcow2_process_discards(bs, ret); | |
1665 | ||
1666 | return ret; | |
621f0589 | 1667 | } |
32b6444d HR |
1668 | |
1669 | /* | |
1670 | * Expands all zero clusters in a specific L1 table (or deallocates them, for | |
1671 | * non-backed non-pre-allocated zero clusters). | |
1672 | * | |
4057a2b2 HR |
1673 | * l1_entries and *visited_l1_entries are used to keep track of progress for |
1674 | * status_cb(). l1_entries contains the total number of L1 entries and | |
1675 | * *visited_l1_entries counts all visited L1 entries. | |
32b6444d HR |
1676 | */ |
1677 | static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table, | |
ecf58777 | 1678 | int l1_size, int64_t *visited_l1_entries, |
4057a2b2 | 1679 | int64_t l1_entries, |
8b13976d HR |
1680 | BlockDriverAmendStatusCB *status_cb, |
1681 | void *cb_opaque) | |
32b6444d | 1682 | { |
ff99129a | 1683 | BDRVQcow2State *s = bs->opaque; |
32b6444d HR |
1684 | bool is_active_l1 = (l1_table == s->l1_table); |
1685 | uint64_t *l2_table = NULL; | |
1686 | int ret; | |
1687 | int i, j; | |
1688 | ||
1689 | if (!is_active_l1) { | |
1690 | /* inactive L2 tables require a buffer to be stored in when loading | |
1691 | * them from disk */ | |
9a4f4c31 | 1692 | l2_table = qemu_try_blockalign(bs->file->bs, s->cluster_size); |
de82815d KW |
1693 | if (l2_table == NULL) { |
1694 | return -ENOMEM; | |
1695 | } | |
32b6444d HR |
1696 | } |
1697 | ||
1698 | for (i = 0; i < l1_size; i++) { | |
1699 | uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK; | |
1700 | bool l2_dirty = false; | |
0e06528e | 1701 | uint64_t l2_refcount; |
32b6444d HR |
1702 | |
1703 | if (!l2_offset) { | |
1704 | /* unallocated */ | |
4057a2b2 HR |
1705 | (*visited_l1_entries)++; |
1706 | if (status_cb) { | |
8b13976d | 1707 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 1708 | } |
32b6444d HR |
1709 | continue; |
1710 | } | |
1711 | ||
8dd93d93 HR |
1712 | if (offset_into_cluster(s, l2_offset)) { |
1713 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" | |
1714 | PRIx64 " unaligned (L1 index: %#x)", | |
1715 | l2_offset, i); | |
1716 | ret = -EIO; | |
1717 | goto fail; | |
1718 | } | |
1719 | ||
32b6444d HR |
1720 | if (is_active_l1) { |
1721 | /* get active L2 tables from cache */ | |
1722 | ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, | |
1723 | (void **)&l2_table); | |
1724 | } else { | |
1725 | /* load inactive L2 tables from disk */ | |
fbcbbf4e | 1726 | ret = bdrv_read(bs->file, l2_offset / BDRV_SECTOR_SIZE, |
9a4f4c31 | 1727 | (void *)l2_table, s->cluster_sectors); |
32b6444d HR |
1728 | } |
1729 | if (ret < 0) { | |
1730 | goto fail; | |
1731 | } | |
1732 | ||
7324c10f HR |
1733 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
1734 | &l2_refcount); | |
1735 | if (ret < 0) { | |
ecf58777 HR |
1736 | goto fail; |
1737 | } | |
1738 | ||
32b6444d HR |
1739 | for (j = 0; j < s->l2_size; j++) { |
1740 | uint64_t l2_entry = be64_to_cpu(l2_table[j]); | |
ecf58777 | 1741 | int64_t offset = l2_entry & L2E_OFFSET_MASK; |
3ef95218 | 1742 | QCow2ClusterType cluster_type = qcow2_get_cluster_type(l2_entry); |
32b6444d | 1743 | |
fdfab37d EB |
1744 | if (cluster_type != QCOW2_CLUSTER_ZERO_PLAIN && |
1745 | cluster_type != QCOW2_CLUSTER_ZERO_ALLOC) { | |
32b6444d HR |
1746 | continue; |
1747 | } | |
1748 | ||
fdfab37d | 1749 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
760e0063 | 1750 | if (!bs->backing) { |
32b6444d HR |
1751 | /* not backed; therefore we can simply deallocate the |
1752 | * cluster */ | |
1753 | l2_table[j] = 0; | |
1754 | l2_dirty = true; | |
1755 | continue; | |
1756 | } | |
1757 | ||
1758 | offset = qcow2_alloc_clusters(bs, s->cluster_size); | |
1759 | if (offset < 0) { | |
1760 | ret = offset; | |
1761 | goto fail; | |
1762 | } | |
ecf58777 HR |
1763 | |
1764 | if (l2_refcount > 1) { | |
1765 | /* For shared L2 tables, set the refcount accordingly (it is | |
1766 | * already 1 and needs to be l2_refcount) */ | |
1767 | ret = qcow2_update_cluster_refcount(bs, | |
2aabe7c7 HR |
1768 | offset >> s->cluster_bits, |
1769 | refcount_diff(1, l2_refcount), false, | |
ecf58777 HR |
1770 | QCOW2_DISCARD_OTHER); |
1771 | if (ret < 0) { | |
1772 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1773 | QCOW2_DISCARD_OTHER); | |
1774 | goto fail; | |
1775 | } | |
1776 | } | |
32b6444d HR |
1777 | } |
1778 | ||
8dd93d93 HR |
1779 | if (offset_into_cluster(s, offset)) { |
1780 | qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset " | |
1781 | "%#" PRIx64 " unaligned (L2 offset: %#" | |
1782 | PRIx64 ", L2 index: %#x)", offset, | |
1783 | l2_offset, j); | |
fdfab37d | 1784 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
8dd93d93 HR |
1785 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1786 | QCOW2_DISCARD_ALWAYS); | |
1787 | } | |
1788 | ret = -EIO; | |
1789 | goto fail; | |
1790 | } | |
1791 | ||
231bb267 | 1792 | ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size); |
32b6444d | 1793 | if (ret < 0) { |
fdfab37d | 1794 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
320c7066 HR |
1795 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1796 | QCOW2_DISCARD_ALWAYS); | |
1797 | } | |
32b6444d HR |
1798 | goto fail; |
1799 | } | |
1800 | ||
720ff280 | 1801 | ret = bdrv_pwrite_zeroes(bs->file, offset, s->cluster_size, 0); |
32b6444d | 1802 | if (ret < 0) { |
fdfab37d | 1803 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
320c7066 HR |
1804 | qcow2_free_clusters(bs, offset, s->cluster_size, |
1805 | QCOW2_DISCARD_ALWAYS); | |
1806 | } | |
32b6444d HR |
1807 | goto fail; |
1808 | } | |
1809 | ||
ecf58777 HR |
1810 | if (l2_refcount == 1) { |
1811 | l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED); | |
1812 | } else { | |
1813 | l2_table[j] = cpu_to_be64(offset); | |
e390cf5a | 1814 | } |
ecf58777 | 1815 | l2_dirty = true; |
32b6444d HR |
1816 | } |
1817 | ||
1818 | if (is_active_l1) { | |
1819 | if (l2_dirty) { | |
72e80b89 | 1820 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
32b6444d HR |
1821 | qcow2_cache_depends_on_flush(s->l2_table_cache); |
1822 | } | |
a3f1afb4 | 1823 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
32b6444d HR |
1824 | } else { |
1825 | if (l2_dirty) { | |
231bb267 HR |
1826 | ret = qcow2_pre_write_overlap_check(bs, |
1827 | QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, l2_offset, | |
32b6444d HR |
1828 | s->cluster_size); |
1829 | if (ret < 0) { | |
1830 | goto fail; | |
1831 | } | |
1832 | ||
18d51c4b | 1833 | ret = bdrv_write(bs->file, l2_offset / BDRV_SECTOR_SIZE, |
9a4f4c31 | 1834 | (void *)l2_table, s->cluster_sectors); |
32b6444d HR |
1835 | if (ret < 0) { |
1836 | goto fail; | |
1837 | } | |
1838 | } | |
1839 | } | |
4057a2b2 HR |
1840 | |
1841 | (*visited_l1_entries)++; | |
1842 | if (status_cb) { | |
8b13976d | 1843 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 1844 | } |
32b6444d HR |
1845 | } |
1846 | ||
1847 | ret = 0; | |
1848 | ||
1849 | fail: | |
1850 | if (l2_table) { | |
1851 | if (!is_active_l1) { | |
1852 | qemu_vfree(l2_table); | |
1853 | } else { | |
a3f1afb4 | 1854 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
32b6444d HR |
1855 | } |
1856 | } | |
1857 | return ret; | |
1858 | } | |
1859 | ||
1860 | /* | |
1861 | * For backed images, expands all zero clusters on the image. For non-backed | |
1862 | * images, deallocates all non-pre-allocated zero clusters (and claims the | |
1863 | * allocation for pre-allocated ones). This is important for downgrading to a | |
1864 | * qcow2 version which doesn't yet support metadata zero clusters. | |
1865 | */ | |
4057a2b2 | 1866 | int qcow2_expand_zero_clusters(BlockDriverState *bs, |
8b13976d HR |
1867 | BlockDriverAmendStatusCB *status_cb, |
1868 | void *cb_opaque) | |
32b6444d | 1869 | { |
ff99129a | 1870 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 1871 | uint64_t *l1_table = NULL; |
4057a2b2 | 1872 | int64_t l1_entries = 0, visited_l1_entries = 0; |
32b6444d HR |
1873 | int ret; |
1874 | int i, j; | |
1875 | ||
4057a2b2 HR |
1876 | if (status_cb) { |
1877 | l1_entries = s->l1_size; | |
1878 | for (i = 0; i < s->nb_snapshots; i++) { | |
1879 | l1_entries += s->snapshots[i].l1_size; | |
1880 | } | |
1881 | } | |
1882 | ||
32b6444d | 1883 | ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size, |
4057a2b2 | 1884 | &visited_l1_entries, l1_entries, |
8b13976d | 1885 | status_cb, cb_opaque); |
32b6444d HR |
1886 | if (ret < 0) { |
1887 | goto fail; | |
1888 | } | |
1889 | ||
1890 | /* Inactive L1 tables may point to active L2 tables - therefore it is | |
1891 | * necessary to flush the L2 table cache before trying to access the L2 | |
1892 | * tables pointed to by inactive L1 entries (else we might try to expand | |
1893 | * zero clusters that have already been expanded); furthermore, it is also | |
1894 | * necessary to empty the L2 table cache, since it may contain tables which | |
1895 | * are now going to be modified directly on disk, bypassing the cache. | |
1896 | * qcow2_cache_empty() does both for us. */ | |
1897 | ret = qcow2_cache_empty(bs, s->l2_table_cache); | |
1898 | if (ret < 0) { | |
1899 | goto fail; | |
1900 | } | |
1901 | ||
1902 | for (i = 0; i < s->nb_snapshots; i++) { | |
d737b78c LV |
1903 | int l1_sectors = DIV_ROUND_UP(s->snapshots[i].l1_size * |
1904 | sizeof(uint64_t), BDRV_SECTOR_SIZE); | |
32b6444d HR |
1905 | |
1906 | l1_table = g_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE); | |
1907 | ||
fbcbbf4e | 1908 | ret = bdrv_read(bs->file, |
9a4f4c31 KW |
1909 | s->snapshots[i].l1_table_offset / BDRV_SECTOR_SIZE, |
1910 | (void *)l1_table, l1_sectors); | |
32b6444d HR |
1911 | if (ret < 0) { |
1912 | goto fail; | |
1913 | } | |
1914 | ||
1915 | for (j = 0; j < s->snapshots[i].l1_size; j++) { | |
1916 | be64_to_cpus(&l1_table[j]); | |
1917 | } | |
1918 | ||
1919 | ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size, | |
4057a2b2 | 1920 | &visited_l1_entries, l1_entries, |
8b13976d | 1921 | status_cb, cb_opaque); |
32b6444d HR |
1922 | if (ret < 0) { |
1923 | goto fail; | |
1924 | } | |
1925 | } | |
1926 | ||
1927 | ret = 0; | |
1928 | ||
1929 | fail: | |
32b6444d HR |
1930 | g_free(l1_table); |
1931 | return ret; | |
1932 | } |