]>
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 | ||
25 | #include <zlib.h> | |
26 | ||
27 | #include "qemu-common.h" | |
737e150e | 28 | #include "block/block_int.h" |
45aba42f | 29 | #include "block/qcow2.h" |
3cce16f4 | 30 | #include "trace.h" |
45aba42f | 31 | |
2cf7cfa1 KW |
32 | int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size, |
33 | bool exact_size) | |
45aba42f | 34 | { |
ff99129a | 35 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 | 36 | int new_l1_size2, ret, i; |
45aba42f | 37 | uint64_t *new_l1_table; |
fda74f82 | 38 | int64_t old_l1_table_offset, old_l1_size; |
2cf7cfa1 | 39 | int64_t new_l1_table_offset, new_l1_size; |
45aba42f KW |
40 | uint8_t data[12]; |
41 | ||
72893756 | 42 | if (min_size <= s->l1_size) |
45aba42f | 43 | return 0; |
72893756 | 44 | |
b93f9950 HR |
45 | /* Do a sanity check on min_size before trying to calculate new_l1_size |
46 | * (this prevents overflows during the while loop for the calculation of | |
47 | * new_l1_size) */ | |
48 | if (min_size > INT_MAX / sizeof(uint64_t)) { | |
49 | return -EFBIG; | |
50 | } | |
51 | ||
72893756 SH |
52 | if (exact_size) { |
53 | new_l1_size = min_size; | |
54 | } else { | |
55 | /* Bump size up to reduce the number of times we have to grow */ | |
56 | new_l1_size = s->l1_size; | |
57 | if (new_l1_size == 0) { | |
58 | new_l1_size = 1; | |
59 | } | |
60 | while (min_size > new_l1_size) { | |
61 | new_l1_size = (new_l1_size * 3 + 1) / 2; | |
62 | } | |
45aba42f | 63 | } |
72893756 | 64 | |
cab60de9 | 65 | if (new_l1_size > INT_MAX / sizeof(uint64_t)) { |
2cf7cfa1 KW |
66 | return -EFBIG; |
67 | } | |
68 | ||
45aba42f | 69 | #ifdef DEBUG_ALLOC2 |
2cf7cfa1 KW |
70 | fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n", |
71 | s->l1_size, new_l1_size); | |
45aba42f KW |
72 | #endif |
73 | ||
74 | new_l1_size2 = sizeof(uint64_t) * new_l1_size; | |
9a4f4c31 | 75 | new_l1_table = qemu_try_blockalign(bs->file->bs, |
de82815d KW |
76 | align_offset(new_l1_size2, 512)); |
77 | if (new_l1_table == NULL) { | |
78 | return -ENOMEM; | |
79 | } | |
80 | memset(new_l1_table, 0, align_offset(new_l1_size2, 512)); | |
81 | ||
45aba42f KW |
82 | memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t)); |
83 | ||
84 | /* write new table (align to cluster) */ | |
66f82cee | 85 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE); |
ed6ccf0f | 86 | new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2); |
5d757b56 | 87 | if (new_l1_table_offset < 0) { |
de82815d | 88 | qemu_vfree(new_l1_table); |
5d757b56 KW |
89 | return new_l1_table_offset; |
90 | } | |
29c1a730 KW |
91 | |
92 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
93 | if (ret < 0) { | |
80fa3341 | 94 | goto fail; |
29c1a730 | 95 | } |
45aba42f | 96 | |
cf93980e HR |
97 | /* the L1 position has not yet been updated, so these clusters must |
98 | * indeed be completely free */ | |
231bb267 HR |
99 | ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset, |
100 | new_l1_size2); | |
cf93980e HR |
101 | if (ret < 0) { |
102 | goto fail; | |
103 | } | |
104 | ||
66f82cee | 105 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE); |
45aba42f KW |
106 | for(i = 0; i < s->l1_size; i++) |
107 | new_l1_table[i] = cpu_to_be64(new_l1_table[i]); | |
9a4f4c31 KW |
108 | ret = bdrv_pwrite_sync(bs->file->bs, new_l1_table_offset, |
109 | new_l1_table, new_l1_size2); | |
8b3b7206 | 110 | if (ret < 0) |
45aba42f KW |
111 | goto fail; |
112 | for(i = 0; i < s->l1_size; i++) | |
113 | new_l1_table[i] = be64_to_cpu(new_l1_table[i]); | |
114 | ||
115 | /* set new table */ | |
66f82cee | 116 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE); |
45aba42f | 117 | cpu_to_be32w((uint32_t*)data, new_l1_size); |
e4ef9f46 | 118 | stq_be_p(data + 4, new_l1_table_offset); |
9a4f4c31 KW |
119 | ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, l1_size), |
120 | data, sizeof(data)); | |
8b3b7206 | 121 | if (ret < 0) { |
45aba42f | 122 | goto fail; |
fb8fa77c | 123 | } |
de82815d | 124 | qemu_vfree(s->l1_table); |
fda74f82 | 125 | old_l1_table_offset = s->l1_table_offset; |
45aba42f KW |
126 | s->l1_table_offset = new_l1_table_offset; |
127 | s->l1_table = new_l1_table; | |
fda74f82 | 128 | old_l1_size = s->l1_size; |
45aba42f | 129 | s->l1_size = new_l1_size; |
fda74f82 HR |
130 | qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * sizeof(uint64_t), |
131 | QCOW2_DISCARD_OTHER); | |
45aba42f KW |
132 | return 0; |
133 | fail: | |
de82815d | 134 | qemu_vfree(new_l1_table); |
6cfcb9b8 KW |
135 | qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2, |
136 | QCOW2_DISCARD_OTHER); | |
8b3b7206 | 137 | return ret; |
45aba42f KW |
138 | } |
139 | ||
45aba42f KW |
140 | /* |
141 | * l2_load | |
142 | * | |
143 | * Loads a L2 table into memory. If the table is in the cache, the cache | |
144 | * is used; otherwise the L2 table is loaded from the image file. | |
145 | * | |
146 | * Returns a pointer to the L2 table on success, or NULL if the read from | |
147 | * the image file failed. | |
148 | */ | |
149 | ||
55c17e98 KW |
150 | static int l2_load(BlockDriverState *bs, uint64_t l2_offset, |
151 | uint64_t **l2_table) | |
45aba42f | 152 | { |
ff99129a | 153 | BDRVQcow2State *s = bs->opaque; |
55c17e98 | 154 | int ret; |
45aba42f | 155 | |
29c1a730 | 156 | ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, (void**) l2_table); |
45aba42f | 157 | |
29c1a730 | 158 | return ret; |
45aba42f KW |
159 | } |
160 | ||
6583e3c7 KW |
161 | /* |
162 | * Writes one sector of the L1 table to the disk (can't update single entries | |
163 | * and we really don't want bdrv_pread to perform a read-modify-write) | |
164 | */ | |
165 | #define L1_ENTRIES_PER_SECTOR (512 / 8) | |
e23e400e | 166 | int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index) |
6583e3c7 | 167 | { |
ff99129a | 168 | BDRVQcow2State *s = bs->opaque; |
a1391444 | 169 | uint64_t buf[L1_ENTRIES_PER_SECTOR] = { 0 }; |
6583e3c7 | 170 | int l1_start_index; |
f7defcb6 | 171 | int i, ret; |
6583e3c7 KW |
172 | |
173 | l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1); | |
a1391444 HR |
174 | for (i = 0; i < L1_ENTRIES_PER_SECTOR && l1_start_index + i < s->l1_size; |
175 | i++) | |
176 | { | |
6583e3c7 KW |
177 | buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]); |
178 | } | |
179 | ||
231bb267 | 180 | ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1, |
cf93980e HR |
181 | s->l1_table_offset + 8 * l1_start_index, sizeof(buf)); |
182 | if (ret < 0) { | |
183 | return ret; | |
184 | } | |
185 | ||
66f82cee | 186 | BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); |
9a4f4c31 KW |
187 | ret = bdrv_pwrite_sync(bs->file->bs, |
188 | s->l1_table_offset + 8 * l1_start_index, | |
189 | buf, sizeof(buf)); | |
f7defcb6 KW |
190 | if (ret < 0) { |
191 | return ret; | |
6583e3c7 KW |
192 | } |
193 | ||
194 | return 0; | |
195 | } | |
196 | ||
45aba42f KW |
197 | /* |
198 | * l2_allocate | |
199 | * | |
200 | * Allocate a new l2 entry in the file. If l1_index points to an already | |
201 | * used entry in the L2 table (i.e. we are doing a copy on write for the L2 | |
202 | * table) copy the contents of the old L2 table into the newly allocated one. | |
203 | * Otherwise the new table is initialized with zeros. | |
204 | * | |
205 | */ | |
206 | ||
c46e1167 | 207 | static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table) |
45aba42f | 208 | { |
ff99129a | 209 | BDRVQcow2State *s = bs->opaque; |
6583e3c7 | 210 | uint64_t old_l2_offset; |
8585afd8 | 211 | uint64_t *l2_table = NULL; |
f4f0d391 | 212 | int64_t l2_offset; |
c46e1167 | 213 | int ret; |
45aba42f KW |
214 | |
215 | old_l2_offset = s->l1_table[l1_index]; | |
216 | ||
3cce16f4 KW |
217 | trace_qcow2_l2_allocate(bs, l1_index); |
218 | ||
45aba42f KW |
219 | /* allocate a new l2 entry */ |
220 | ||
ed6ccf0f | 221 | l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t)); |
5d757b56 | 222 | if (l2_offset < 0) { |
be0b742e HR |
223 | ret = l2_offset; |
224 | goto fail; | |
5d757b56 | 225 | } |
29c1a730 KW |
226 | |
227 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
228 | if (ret < 0) { | |
229 | goto fail; | |
230 | } | |
45aba42f | 231 | |
45aba42f KW |
232 | /* allocate a new entry in the l2 cache */ |
233 | ||
3cce16f4 | 234 | trace_qcow2_l2_allocate_get_empty(bs, l1_index); |
29c1a730 KW |
235 | ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table); |
236 | if (ret < 0) { | |
be0b742e | 237 | goto fail; |
29c1a730 KW |
238 | } |
239 | ||
240 | l2_table = *table; | |
45aba42f | 241 | |
8e37f681 | 242 | if ((old_l2_offset & L1E_OFFSET_MASK) == 0) { |
45aba42f KW |
243 | /* if there was no old l2 table, clear the new table */ |
244 | memset(l2_table, 0, s->l2_size * sizeof(uint64_t)); | |
245 | } else { | |
29c1a730 KW |
246 | uint64_t* old_table; |
247 | ||
45aba42f | 248 | /* if there was an old l2 table, read it from the disk */ |
66f82cee | 249 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ); |
8e37f681 KW |
250 | ret = qcow2_cache_get(bs, s->l2_table_cache, |
251 | old_l2_offset & L1E_OFFSET_MASK, | |
29c1a730 KW |
252 | (void**) &old_table); |
253 | if (ret < 0) { | |
254 | goto fail; | |
255 | } | |
256 | ||
257 | memcpy(l2_table, old_table, s->cluster_size); | |
258 | ||
a3f1afb4 | 259 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &old_table); |
45aba42f | 260 | } |
29c1a730 | 261 | |
45aba42f | 262 | /* write the l2 table to the file */ |
66f82cee | 263 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE); |
29c1a730 | 264 | |
3cce16f4 | 265 | trace_qcow2_l2_allocate_write_l2(bs, l1_index); |
72e80b89 | 266 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
29c1a730 | 267 | ret = qcow2_cache_flush(bs, s->l2_table_cache); |
c46e1167 | 268 | if (ret < 0) { |
175e1152 KW |
269 | goto fail; |
270 | } | |
271 | ||
272 | /* update the L1 entry */ | |
3cce16f4 | 273 | trace_qcow2_l2_allocate_write_l1(bs, l1_index); |
175e1152 | 274 | s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; |
e23e400e | 275 | ret = qcow2_write_l1_entry(bs, l1_index); |
175e1152 KW |
276 | if (ret < 0) { |
277 | goto fail; | |
c46e1167 | 278 | } |
45aba42f | 279 | |
c46e1167 | 280 | *table = l2_table; |
3cce16f4 | 281 | trace_qcow2_l2_allocate_done(bs, l1_index, 0); |
c46e1167 | 282 | return 0; |
175e1152 KW |
283 | |
284 | fail: | |
3cce16f4 | 285 | trace_qcow2_l2_allocate_done(bs, l1_index, ret); |
8585afd8 HR |
286 | if (l2_table != NULL) { |
287 | qcow2_cache_put(bs, s->l2_table_cache, (void**) table); | |
288 | } | |
68dba0bf | 289 | s->l1_table[l1_index] = old_l2_offset; |
e3b21ef9 HR |
290 | if (l2_offset > 0) { |
291 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), | |
292 | QCOW2_DISCARD_ALWAYS); | |
293 | } | |
175e1152 | 294 | return ret; |
45aba42f KW |
295 | } |
296 | ||
2bfcc4a0 KW |
297 | /* |
298 | * Checks how many clusters in a given L2 table are contiguous in the image | |
299 | * file. As soon as one of the flags in the bitmask stop_flags changes compared | |
300 | * to the first cluster, the search is stopped and the cluster is not counted | |
301 | * as contiguous. (This allows it, for example, to stop at the first compressed | |
302 | * cluster which may require a different handling) | |
303 | */ | |
b6d36def | 304 | static int count_contiguous_clusters(int nb_clusters, int cluster_size, |
61653008 | 305 | uint64_t *l2_table, uint64_t stop_flags) |
45aba42f KW |
306 | { |
307 | int i; | |
78a52ad5 | 308 | uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED; |
15684a47 HR |
309 | uint64_t first_entry = be64_to_cpu(l2_table[0]); |
310 | uint64_t offset = first_entry & mask; | |
45aba42f KW |
311 | |
312 | if (!offset) | |
313 | return 0; | |
314 | ||
a99dfb45 | 315 | assert(qcow2_get_cluster_type(first_entry) == QCOW2_CLUSTER_NORMAL); |
15684a47 | 316 | |
61653008 | 317 | for (i = 0; i < nb_clusters; i++) { |
2bfcc4a0 KW |
318 | uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask; |
319 | if (offset + (uint64_t) i * cluster_size != l2_entry) { | |
45aba42f | 320 | break; |
2bfcc4a0 KW |
321 | } |
322 | } | |
45aba42f | 323 | |
61653008 | 324 | return i; |
45aba42f KW |
325 | } |
326 | ||
a99dfb45 KW |
327 | static int count_contiguous_clusters_by_type(int nb_clusters, |
328 | uint64_t *l2_table, | |
329 | int wanted_type) | |
45aba42f | 330 | { |
2bfcc4a0 KW |
331 | int i; |
332 | ||
333 | for (i = 0; i < nb_clusters; i++) { | |
334 | int type = qcow2_get_cluster_type(be64_to_cpu(l2_table[i])); | |
45aba42f | 335 | |
a99dfb45 | 336 | if (type != wanted_type) { |
2bfcc4a0 KW |
337 | break; |
338 | } | |
339 | } | |
45aba42f KW |
340 | |
341 | return i; | |
342 | } | |
343 | ||
344 | /* The crypt function is compatible with the linux cryptoloop | |
345 | algorithm for < 4 GB images. NOTE: out_buf == in_buf is | |
346 | supported */ | |
ff99129a | 347 | int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num, |
f6fa64f6 DB |
348 | uint8_t *out_buf, const uint8_t *in_buf, |
349 | int nb_sectors, bool enc, | |
350 | Error **errp) | |
45aba42f KW |
351 | { |
352 | union { | |
353 | uint64_t ll[2]; | |
354 | uint8_t b[16]; | |
355 | } ivec; | |
356 | int i; | |
f6fa64f6 | 357 | int ret; |
45aba42f KW |
358 | |
359 | for(i = 0; i < nb_sectors; i++) { | |
360 | ivec.ll[0] = cpu_to_le64(sector_num); | |
361 | ivec.ll[1] = 0; | |
f6fa64f6 DB |
362 | if (qcrypto_cipher_setiv(s->cipher, |
363 | ivec.b, G_N_ELEMENTS(ivec.b), | |
364 | errp) < 0) { | |
365 | return -1; | |
366 | } | |
367 | if (enc) { | |
368 | ret = qcrypto_cipher_encrypt(s->cipher, | |
369 | in_buf, | |
370 | out_buf, | |
371 | 512, | |
372 | errp); | |
373 | } else { | |
374 | ret = qcrypto_cipher_decrypt(s->cipher, | |
375 | in_buf, | |
376 | out_buf, | |
377 | 512, | |
378 | errp); | |
379 | } | |
380 | if (ret < 0) { | |
381 | return -1; | |
382 | } | |
45aba42f KW |
383 | sector_num++; |
384 | in_buf += 512; | |
385 | out_buf += 512; | |
386 | } | |
f6fa64f6 | 387 | return 0; |
45aba42f KW |
388 | } |
389 | ||
aef4acb6 SH |
390 | static int coroutine_fn copy_sectors(BlockDriverState *bs, |
391 | uint64_t start_sect, | |
392 | uint64_t cluster_offset, | |
393 | int n_start, int n_end) | |
45aba42f | 394 | { |
ff99129a | 395 | BDRVQcow2State *s = bs->opaque; |
aef4acb6 SH |
396 | QEMUIOVector qiov; |
397 | struct iovec iov; | |
45aba42f | 398 | int n, ret; |
1b9f1491 | 399 | |
45aba42f | 400 | n = n_end - n_start; |
1b9f1491 | 401 | if (n <= 0) { |
45aba42f | 402 | return 0; |
1b9f1491 KW |
403 | } |
404 | ||
aef4acb6 | 405 | iov.iov_len = n * BDRV_SECTOR_SIZE; |
de82815d KW |
406 | iov.iov_base = qemu_try_blockalign(bs, iov.iov_len); |
407 | if (iov.iov_base == NULL) { | |
408 | return -ENOMEM; | |
409 | } | |
aef4acb6 SH |
410 | |
411 | qemu_iovec_init_external(&qiov, &iov, 1); | |
1b9f1491 | 412 | |
66f82cee | 413 | BLKDBG_EVENT(bs->file, BLKDBG_COW_READ); |
aef4acb6 | 414 | |
dba28555 | 415 | if (!bs->drv) { |
bd604369 KW |
416 | ret = -ENOMEDIUM; |
417 | goto out; | |
dba28555 HR |
418 | } |
419 | ||
aef4acb6 SH |
420 | /* Call .bdrv_co_readv() directly instead of using the public block-layer |
421 | * interface. This avoids double I/O throttling and request tracking, | |
422 | * which can lead to deadlock when block layer copy-on-read is enabled. | |
423 | */ | |
424 | ret = bs->drv->bdrv_co_readv(bs, start_sect + n_start, n, &qiov); | |
1b9f1491 KW |
425 | if (ret < 0) { |
426 | goto out; | |
427 | } | |
428 | ||
8336aafa | 429 | if (bs->encrypted) { |
f6fa64f6 DB |
430 | Error *err = NULL; |
431 | assert(s->cipher); | |
432 | if (qcow2_encrypt_sectors(s, start_sect + n_start, | |
433 | iov.iov_base, iov.iov_base, n, | |
434 | true, &err) < 0) { | |
435 | ret = -EIO; | |
436 | error_free(err); | |
437 | goto out; | |
438 | } | |
45aba42f | 439 | } |
1b9f1491 | 440 | |
231bb267 | 441 | ret = qcow2_pre_write_overlap_check(bs, 0, |
cf93980e HR |
442 | cluster_offset + n_start * BDRV_SECTOR_SIZE, n * BDRV_SECTOR_SIZE); |
443 | if (ret < 0) { | |
444 | goto out; | |
445 | } | |
446 | ||
66f82cee | 447 | BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE); |
9a4f4c31 KW |
448 | ret = bdrv_co_writev(bs->file->bs, (cluster_offset >> 9) + n_start, n, |
449 | &qiov); | |
1b9f1491 KW |
450 | if (ret < 0) { |
451 | goto out; | |
452 | } | |
453 | ||
454 | ret = 0; | |
455 | out: | |
aef4acb6 | 456 | qemu_vfree(iov.iov_base); |
1b9f1491 | 457 | return ret; |
45aba42f KW |
458 | } |
459 | ||
460 | ||
461 | /* | |
462 | * get_cluster_offset | |
463 | * | |
1c46efaa KW |
464 | * For a given offset of the disk image, find the cluster offset in |
465 | * qcow2 file. The offset is stored in *cluster_offset. | |
45aba42f | 466 | * |
d57237f2 | 467 | * on entry, *num is the number of contiguous sectors we'd like to |
45aba42f KW |
468 | * access following offset. |
469 | * | |
d57237f2 | 470 | * on exit, *num is the number of contiguous sectors we can read. |
45aba42f | 471 | * |
68d000a3 KW |
472 | * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error |
473 | * cases. | |
45aba42f | 474 | */ |
1c46efaa KW |
475 | int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, |
476 | int *num, uint64_t *cluster_offset) | |
45aba42f | 477 | { |
ff99129a | 478 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
479 | unsigned int l2_index; |
480 | uint64_t l1_index, l2_offset, *l2_table; | |
45aba42f | 481 | int l1_bits, c; |
80ee15a6 KW |
482 | unsigned int index_in_cluster, nb_clusters; |
483 | uint64_t nb_available, nb_needed; | |
55c17e98 | 484 | int ret; |
45aba42f KW |
485 | |
486 | index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1); | |
487 | nb_needed = *num + index_in_cluster; | |
488 | ||
489 | l1_bits = s->l2_bits + s->cluster_bits; | |
490 | ||
491 | /* compute how many bytes there are between the offset and | |
492 | * the end of the l1 entry | |
493 | */ | |
494 | ||
80ee15a6 | 495 | nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1)); |
45aba42f KW |
496 | |
497 | /* compute the number of available sectors */ | |
498 | ||
499 | nb_available = (nb_available >> 9) + index_in_cluster; | |
500 | ||
501 | if (nb_needed > nb_available) { | |
502 | nb_needed = nb_available; | |
503 | } | |
b6d36def | 504 | assert(nb_needed <= INT_MAX); |
45aba42f | 505 | |
1c46efaa | 506 | *cluster_offset = 0; |
45aba42f | 507 | |
b6af0975 | 508 | /* seek to the l2 offset in the l1 table */ |
45aba42f KW |
509 | |
510 | l1_index = offset >> l1_bits; | |
68d000a3 KW |
511 | if (l1_index >= s->l1_size) { |
512 | ret = QCOW2_CLUSTER_UNALLOCATED; | |
45aba42f | 513 | goto out; |
68d000a3 | 514 | } |
45aba42f | 515 | |
68d000a3 KW |
516 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
517 | if (!l2_offset) { | |
518 | ret = QCOW2_CLUSTER_UNALLOCATED; | |
45aba42f | 519 | goto out; |
68d000a3 | 520 | } |
45aba42f | 521 | |
a97c67ee HR |
522 | if (offset_into_cluster(s, l2_offset)) { |
523 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
524 | " unaligned (L1 index: %#" PRIx64 ")", | |
525 | l2_offset, l1_index); | |
526 | return -EIO; | |
527 | } | |
528 | ||
45aba42f KW |
529 | /* load the l2 table in memory */ |
530 | ||
55c17e98 KW |
531 | ret = l2_load(bs, l2_offset, &l2_table); |
532 | if (ret < 0) { | |
533 | return ret; | |
1c46efaa | 534 | } |
45aba42f KW |
535 | |
536 | /* find the cluster offset for the given disk offset */ | |
537 | ||
538 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
1c46efaa | 539 | *cluster_offset = be64_to_cpu(l2_table[l2_index]); |
b6d36def HR |
540 | |
541 | /* nb_needed <= INT_MAX, thus nb_clusters <= INT_MAX, too */ | |
45aba42f KW |
542 | nb_clusters = size_to_clusters(s, nb_needed << 9); |
543 | ||
68d000a3 KW |
544 | ret = qcow2_get_cluster_type(*cluster_offset); |
545 | switch (ret) { | |
546 | case QCOW2_CLUSTER_COMPRESSED: | |
547 | /* Compressed clusters can only be processed one by one */ | |
548 | c = 1; | |
549 | *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK; | |
550 | break; | |
6377af48 | 551 | case QCOW2_CLUSTER_ZERO: |
381b487d | 552 | if (s->qcow_version < 3) { |
a97c67ee HR |
553 | qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found" |
554 | " in pre-v3 image (L2 offset: %#" PRIx64 | |
555 | ", L2 index: %#x)", l2_offset, l2_index); | |
556 | ret = -EIO; | |
557 | goto fail; | |
381b487d | 558 | } |
a99dfb45 KW |
559 | c = count_contiguous_clusters_by_type(nb_clusters, &l2_table[l2_index], |
560 | QCOW2_CLUSTER_ZERO); | |
6377af48 KW |
561 | *cluster_offset = 0; |
562 | break; | |
68d000a3 | 563 | case QCOW2_CLUSTER_UNALLOCATED: |
45aba42f | 564 | /* how many empty clusters ? */ |
a99dfb45 KW |
565 | c = count_contiguous_clusters_by_type(nb_clusters, &l2_table[l2_index], |
566 | QCOW2_CLUSTER_UNALLOCATED); | |
68d000a3 KW |
567 | *cluster_offset = 0; |
568 | break; | |
569 | case QCOW2_CLUSTER_NORMAL: | |
45aba42f KW |
570 | /* how many allocated clusters ? */ |
571 | c = count_contiguous_clusters(nb_clusters, s->cluster_size, | |
61653008 | 572 | &l2_table[l2_index], QCOW_OFLAG_ZERO); |
68d000a3 | 573 | *cluster_offset &= L2E_OFFSET_MASK; |
a97c67ee HR |
574 | if (offset_into_cluster(s, *cluster_offset)) { |
575 | qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset %#" | |
576 | PRIx64 " unaligned (L2 offset: %#" PRIx64 | |
577 | ", L2 index: %#x)", *cluster_offset, | |
578 | l2_offset, l2_index); | |
579 | ret = -EIO; | |
580 | goto fail; | |
581 | } | |
68d000a3 | 582 | break; |
1417d7e4 KW |
583 | default: |
584 | abort(); | |
45aba42f KW |
585 | } |
586 | ||
29c1a730 KW |
587 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
588 | ||
68d000a3 KW |
589 | nb_available = (c * s->cluster_sectors); |
590 | ||
45aba42f KW |
591 | out: |
592 | if (nb_available > nb_needed) | |
593 | nb_available = nb_needed; | |
594 | ||
595 | *num = nb_available - index_in_cluster; | |
596 | ||
68d000a3 | 597 | return ret; |
a97c67ee HR |
598 | |
599 | fail: | |
600 | qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table); | |
601 | return ret; | |
45aba42f KW |
602 | } |
603 | ||
604 | /* | |
605 | * get_cluster_table | |
606 | * | |
607 | * for a given disk offset, load (and allocate if needed) | |
608 | * the l2 table. | |
609 | * | |
610 | * the l2 table offset in the qcow2 file and the cluster index | |
611 | * in the l2 table are given to the caller. | |
612 | * | |
1e3e8f1a | 613 | * Returns 0 on success, -errno in failure case |
45aba42f | 614 | */ |
45aba42f KW |
615 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
616 | uint64_t **new_l2_table, | |
45aba42f KW |
617 | int *new_l2_index) |
618 | { | |
ff99129a | 619 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
620 | unsigned int l2_index; |
621 | uint64_t l1_index, l2_offset; | |
c46e1167 | 622 | uint64_t *l2_table = NULL; |
80ee15a6 | 623 | int ret; |
45aba42f | 624 | |
b6af0975 | 625 | /* seek to the l2 offset in the l1 table */ |
45aba42f KW |
626 | |
627 | l1_index = offset >> (s->l2_bits + s->cluster_bits); | |
628 | if (l1_index >= s->l1_size) { | |
72893756 | 629 | ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
1e3e8f1a KW |
630 | if (ret < 0) { |
631 | return ret; | |
632 | } | |
45aba42f | 633 | } |
8e37f681 | 634 | |
2cf7cfa1 | 635 | assert(l1_index < s->l1_size); |
8e37f681 | 636 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
a97c67ee HR |
637 | if (offset_into_cluster(s, l2_offset)) { |
638 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
639 | " unaligned (L1 index: %#" PRIx64 ")", | |
640 | l2_offset, l1_index); | |
641 | return -EIO; | |
642 | } | |
45aba42f KW |
643 | |
644 | /* seek the l2 table of the given l2 offset */ | |
645 | ||
8e37f681 | 646 | if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) { |
45aba42f | 647 | /* load the l2 table in memory */ |
55c17e98 KW |
648 | ret = l2_load(bs, l2_offset, &l2_table); |
649 | if (ret < 0) { | |
650 | return ret; | |
1e3e8f1a | 651 | } |
45aba42f | 652 | } else { |
16fde5f2 | 653 | /* First allocate a new L2 table (and do COW if needed) */ |
c46e1167 KW |
654 | ret = l2_allocate(bs, l1_index, &l2_table); |
655 | if (ret < 0) { | |
656 | return ret; | |
1e3e8f1a | 657 | } |
16fde5f2 KW |
658 | |
659 | /* Then decrease the refcount of the old table */ | |
660 | if (l2_offset) { | |
6cfcb9b8 KW |
661 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), |
662 | QCOW2_DISCARD_OTHER); | |
16fde5f2 | 663 | } |
45aba42f KW |
664 | } |
665 | ||
666 | /* find the cluster offset for the given disk offset */ | |
667 | ||
668 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); | |
669 | ||
670 | *new_l2_table = l2_table; | |
45aba42f KW |
671 | *new_l2_index = l2_index; |
672 | ||
1e3e8f1a | 673 | return 0; |
45aba42f KW |
674 | } |
675 | ||
676 | /* | |
677 | * alloc_compressed_cluster_offset | |
678 | * | |
679 | * For a given offset of the disk image, return cluster offset in | |
680 | * qcow2 file. | |
681 | * | |
682 | * If the offset is not found, allocate a new compressed cluster. | |
683 | * | |
684 | * Return the cluster offset if successful, | |
685 | * Return 0, otherwise. | |
686 | * | |
687 | */ | |
688 | ||
ed6ccf0f KW |
689 | uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
690 | uint64_t offset, | |
691 | int compressed_size) | |
45aba42f | 692 | { |
ff99129a | 693 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 694 | int l2_index, ret; |
3948d1d4 | 695 | uint64_t *l2_table; |
f4f0d391 | 696 | int64_t cluster_offset; |
45aba42f KW |
697 | int nb_csectors; |
698 | ||
3948d1d4 | 699 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
1e3e8f1a | 700 | if (ret < 0) { |
45aba42f | 701 | return 0; |
1e3e8f1a | 702 | } |
45aba42f | 703 | |
b0b6862e KW |
704 | /* Compression can't overwrite anything. Fail if the cluster was already |
705 | * allocated. */ | |
45aba42f | 706 | cluster_offset = be64_to_cpu(l2_table[l2_index]); |
b0b6862e | 707 | if (cluster_offset & L2E_OFFSET_MASK) { |
8f1efd00 KW |
708 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
709 | return 0; | |
710 | } | |
45aba42f | 711 | |
ed6ccf0f | 712 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
5d757b56 | 713 | if (cluster_offset < 0) { |
29c1a730 | 714 | qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); |
5d757b56 KW |
715 | return 0; |
716 | } | |
717 | ||
45aba42f KW |
718 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
719 | (cluster_offset >> 9); | |
720 | ||
721 | cluster_offset |= QCOW_OFLAG_COMPRESSED | | |
722 | ((uint64_t)nb_csectors << s->csize_shift); | |
723 | ||
724 | /* update L2 table */ | |
725 | ||
726 | /* compressed clusters never have the copied flag */ | |
727 | ||
66f82cee | 728 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
72e80b89 | 729 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
45aba42f | 730 | l2_table[l2_index] = cpu_to_be64(cluster_offset); |
a3f1afb4 | 731 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
4c1612d9 | 732 | |
29c1a730 | 733 | return cluster_offset; |
4c1612d9 KW |
734 | } |
735 | ||
593fb83c KW |
736 | static int perform_cow(BlockDriverState *bs, QCowL2Meta *m, Qcow2COWRegion *r) |
737 | { | |
ff99129a | 738 | BDRVQcow2State *s = bs->opaque; |
593fb83c KW |
739 | int ret; |
740 | ||
741 | if (r->nb_sectors == 0) { | |
742 | return 0; | |
743 | } | |
744 | ||
745 | qemu_co_mutex_unlock(&s->lock); | |
746 | ret = copy_sectors(bs, m->offset / BDRV_SECTOR_SIZE, m->alloc_offset, | |
747 | r->offset / BDRV_SECTOR_SIZE, | |
748 | r->offset / BDRV_SECTOR_SIZE + r->nb_sectors); | |
749 | qemu_co_mutex_lock(&s->lock); | |
750 | ||
751 | if (ret < 0) { | |
752 | return ret; | |
753 | } | |
754 | ||
755 | /* | |
756 | * Before we update the L2 table to actually point to the new cluster, we | |
757 | * need to be sure that the refcounts have been increased and COW was | |
758 | * handled. | |
759 | */ | |
760 | qcow2_cache_depends_on_flush(s->l2_table_cache); | |
761 | ||
762 | return 0; | |
763 | } | |
764 | ||
148da7ea | 765 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) |
45aba42f | 766 | { |
ff99129a | 767 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 768 | int i, j = 0, l2_index, ret; |
593fb83c | 769 | uint64_t *old_cluster, *l2_table; |
250196f1 | 770 | uint64_t cluster_offset = m->alloc_offset; |
45aba42f | 771 | |
3cce16f4 | 772 | trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
f50f88b9 | 773 | assert(m->nb_clusters > 0); |
45aba42f | 774 | |
5839e53b | 775 | old_cluster = g_try_new(uint64_t, m->nb_clusters); |
de82815d KW |
776 | if (old_cluster == NULL) { |
777 | ret = -ENOMEM; | |
778 | goto err; | |
779 | } | |
45aba42f KW |
780 | |
781 | /* copy content of unmodified sectors */ | |
593fb83c KW |
782 | ret = perform_cow(bs, m, &m->cow_start); |
783 | if (ret < 0) { | |
784 | goto err; | |
45aba42f KW |
785 | } |
786 | ||
593fb83c KW |
787 | ret = perform_cow(bs, m, &m->cow_end); |
788 | if (ret < 0) { | |
789 | goto err; | |
29c1a730 KW |
790 | } |
791 | ||
593fb83c | 792 | /* Update L2 table. */ |
74c4510a | 793 | if (s->use_lazy_refcounts) { |
280d3735 KW |
794 | qcow2_mark_dirty(bs); |
795 | } | |
bfe8043e SH |
796 | if (qcow2_need_accurate_refcounts(s)) { |
797 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
798 | s->refcount_block_cache); | |
799 | } | |
280d3735 | 800 | |
3948d1d4 | 801 | ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index); |
1e3e8f1a | 802 | if (ret < 0) { |
45aba42f | 803 | goto err; |
1e3e8f1a | 804 | } |
72e80b89 | 805 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
45aba42f | 806 | |
c01dbccb | 807 | assert(l2_index + m->nb_clusters <= s->l2_size); |
45aba42f KW |
808 | for (i = 0; i < m->nb_clusters; i++) { |
809 | /* if two concurrent writes happen to the same unallocated cluster | |
810 | * each write allocates separate cluster and writes data concurrently. | |
811 | * The first one to complete updates l2 table with pointer to its | |
812 | * cluster the second one has to do RMW (which is done above by | |
813 | * copy_sectors()), update l2 table with its cluster pointer and free | |
814 | * old cluster. This is what this loop does */ | |
815 | if(l2_table[l2_index + i] != 0) | |
816 | old_cluster[j++] = l2_table[l2_index + i]; | |
817 | ||
818 | l2_table[l2_index + i] = cpu_to_be64((cluster_offset + | |
819 | (i << s->cluster_bits)) | QCOW_OFLAG_COPIED); | |
820 | } | |
821 | ||
9f8e668e | 822 | |
a3f1afb4 | 823 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
45aba42f | 824 | |
7ec5e6a4 KW |
825 | /* |
826 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
6cfcb9b8 KW |
827 | * |
828 | * Don't discard clusters that reach a refcount of 0 (e.g. compressed | |
829 | * clusters), the next write will reuse them anyway. | |
7ec5e6a4 KW |
830 | */ |
831 | if (j != 0) { | |
7ec5e6a4 | 832 | for (i = 0; i < j; i++) { |
6cfcb9b8 KW |
833 | qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1, |
834 | QCOW2_DISCARD_NEVER); | |
7ec5e6a4 KW |
835 | } |
836 | } | |
45aba42f KW |
837 | |
838 | ret = 0; | |
839 | err: | |
7267c094 | 840 | g_free(old_cluster); |
45aba42f KW |
841 | return ret; |
842 | } | |
843 | ||
bf319ece KW |
844 | /* |
845 | * Returns the number of contiguous clusters that can be used for an allocating | |
846 | * write, but require COW to be performed (this includes yet unallocated space, | |
847 | * which must copy from the backing file) | |
848 | */ | |
ff99129a | 849 | static int count_cow_clusters(BDRVQcow2State *s, int nb_clusters, |
bf319ece KW |
850 | uint64_t *l2_table, int l2_index) |
851 | { | |
143550a8 | 852 | int i; |
bf319ece | 853 | |
143550a8 KW |
854 | for (i = 0; i < nb_clusters; i++) { |
855 | uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]); | |
856 | int cluster_type = qcow2_get_cluster_type(l2_entry); | |
857 | ||
858 | switch(cluster_type) { | |
859 | case QCOW2_CLUSTER_NORMAL: | |
860 | if (l2_entry & QCOW_OFLAG_COPIED) { | |
861 | goto out; | |
862 | } | |
bf319ece | 863 | break; |
143550a8 KW |
864 | case QCOW2_CLUSTER_UNALLOCATED: |
865 | case QCOW2_CLUSTER_COMPRESSED: | |
6377af48 | 866 | case QCOW2_CLUSTER_ZERO: |
bf319ece | 867 | break; |
143550a8 KW |
868 | default: |
869 | abort(); | |
870 | } | |
bf319ece KW |
871 | } |
872 | ||
143550a8 | 873 | out: |
bf319ece KW |
874 | assert(i <= nb_clusters); |
875 | return i; | |
876 | } | |
877 | ||
250196f1 | 878 | /* |
226c3c26 KW |
879 | * Check if there already is an AIO write request in flight which allocates |
880 | * the same cluster. In this case we need to wait until the previous | |
881 | * request has completed and updated the L2 table accordingly. | |
65eb2e35 KW |
882 | * |
883 | * Returns: | |
884 | * 0 if there was no dependency. *cur_bytes indicates the number of | |
885 | * bytes from guest_offset that can be read before the next | |
886 | * dependency must be processed (or the request is complete) | |
887 | * | |
888 | * -EAGAIN if we had to wait for another request, previously gathered | |
889 | * information on cluster allocation may be invalid now. The caller | |
890 | * must start over anyway, so consider *cur_bytes undefined. | |
250196f1 | 891 | */ |
226c3c26 | 892 | static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, |
ecdd5333 | 893 | uint64_t *cur_bytes, QCowL2Meta **m) |
250196f1 | 894 | { |
ff99129a | 895 | BDRVQcow2State *s = bs->opaque; |
250196f1 | 896 | QCowL2Meta *old_alloc; |
65eb2e35 | 897 | uint64_t bytes = *cur_bytes; |
250196f1 | 898 | |
250196f1 KW |
899 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
900 | ||
65eb2e35 KW |
901 | uint64_t start = guest_offset; |
902 | uint64_t end = start + bytes; | |
903 | uint64_t old_start = l2meta_cow_start(old_alloc); | |
904 | uint64_t old_end = l2meta_cow_end(old_alloc); | |
250196f1 | 905 | |
d9d74f41 | 906 | if (end <= old_start || start >= old_end) { |
250196f1 KW |
907 | /* No intersection */ |
908 | } else { | |
909 | if (start < old_start) { | |
910 | /* Stop at the start of a running allocation */ | |
65eb2e35 | 911 | bytes = old_start - start; |
250196f1 | 912 | } else { |
65eb2e35 | 913 | bytes = 0; |
250196f1 KW |
914 | } |
915 | ||
ecdd5333 KW |
916 | /* Stop if already an l2meta exists. After yielding, it wouldn't |
917 | * be valid any more, so we'd have to clean up the old L2Metas | |
918 | * and deal with requests depending on them before starting to | |
919 | * gather new ones. Not worth the trouble. */ | |
920 | if (bytes == 0 && *m) { | |
921 | *cur_bytes = 0; | |
922 | return 0; | |
923 | } | |
924 | ||
65eb2e35 | 925 | if (bytes == 0) { |
250196f1 KW |
926 | /* Wait for the dependency to complete. We need to recheck |
927 | * the free/allocated clusters when we continue. */ | |
928 | qemu_co_mutex_unlock(&s->lock); | |
929 | qemu_co_queue_wait(&old_alloc->dependent_requests); | |
930 | qemu_co_mutex_lock(&s->lock); | |
931 | return -EAGAIN; | |
932 | } | |
933 | } | |
934 | } | |
935 | ||
65eb2e35 KW |
936 | /* Make sure that existing clusters and new allocations are only used up to |
937 | * the next dependency if we shortened the request above */ | |
938 | *cur_bytes = bytes; | |
250196f1 | 939 | |
226c3c26 KW |
940 | return 0; |
941 | } | |
942 | ||
0af729ec KW |
943 | /* |
944 | * Checks how many already allocated clusters that don't require a copy on | |
945 | * write there are at the given guest_offset (up to *bytes). If | |
946 | * *host_offset is not zero, only physically contiguous clusters beginning at | |
947 | * this host offset are counted. | |
948 | * | |
411d62b0 KW |
949 | * Note that guest_offset may not be cluster aligned. In this case, the |
950 | * returned *host_offset points to exact byte referenced by guest_offset and | |
951 | * therefore isn't cluster aligned as well. | |
0af729ec KW |
952 | * |
953 | * Returns: | |
954 | * 0: if no allocated clusters are available at the given offset. | |
955 | * *bytes is normally unchanged. It is set to 0 if the cluster | |
956 | * is allocated and doesn't need COW, but doesn't have the right | |
957 | * physical offset. | |
958 | * | |
959 | * 1: if allocated clusters that don't require a COW are available at | |
960 | * the requested offset. *bytes may have decreased and describes | |
961 | * the length of the area that can be written to. | |
962 | * | |
963 | * -errno: in error cases | |
0af729ec KW |
964 | */ |
965 | static int handle_copied(BlockDriverState *bs, uint64_t guest_offset, | |
c53ede9f | 966 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
0af729ec | 967 | { |
ff99129a | 968 | BDRVQcow2State *s = bs->opaque; |
0af729ec KW |
969 | int l2_index; |
970 | uint64_t cluster_offset; | |
971 | uint64_t *l2_table; | |
b6d36def | 972 | uint64_t nb_clusters; |
c53ede9f | 973 | unsigned int keep_clusters; |
a3f1afb4 | 974 | int ret; |
0af729ec KW |
975 | |
976 | trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset, | |
977 | *bytes); | |
0af729ec | 978 | |
411d62b0 KW |
979 | assert(*host_offset == 0 || offset_into_cluster(s, guest_offset) |
980 | == offset_into_cluster(s, *host_offset)); | |
981 | ||
acb0467f KW |
982 | /* |
983 | * Calculate the number of clusters to look for. We stop at L2 table | |
984 | * boundaries to keep things simple. | |
985 | */ | |
986 | nb_clusters = | |
987 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
988 | ||
989 | l2_index = offset_to_l2_index(s, guest_offset); | |
990 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 991 | assert(nb_clusters <= INT_MAX); |
acb0467f | 992 | |
0af729ec KW |
993 | /* Find L2 entry for the first involved cluster */ |
994 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
995 | if (ret < 0) { | |
996 | return ret; | |
997 | } | |
998 | ||
999 | cluster_offset = be64_to_cpu(l2_table[l2_index]); | |
1000 | ||
1001 | /* Check how many clusters are already allocated and don't need COW */ | |
1002 | if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL | |
1003 | && (cluster_offset & QCOW_OFLAG_COPIED)) | |
1004 | { | |
e62daaf6 KW |
1005 | /* If a specific host_offset is required, check it */ |
1006 | bool offset_matches = | |
1007 | (cluster_offset & L2E_OFFSET_MASK) == *host_offset; | |
1008 | ||
a97c67ee HR |
1009 | if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) { |
1010 | qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset " | |
1011 | "%#llx unaligned (guest offset: %#" PRIx64 | |
1012 | ")", cluster_offset & L2E_OFFSET_MASK, | |
1013 | guest_offset); | |
1014 | ret = -EIO; | |
1015 | goto out; | |
1016 | } | |
1017 | ||
e62daaf6 KW |
1018 | if (*host_offset != 0 && !offset_matches) { |
1019 | *bytes = 0; | |
1020 | ret = 0; | |
1021 | goto out; | |
1022 | } | |
1023 | ||
0af729ec | 1024 | /* We keep all QCOW_OFLAG_COPIED clusters */ |
c53ede9f | 1025 | keep_clusters = |
acb0467f | 1026 | count_contiguous_clusters(nb_clusters, s->cluster_size, |
61653008 | 1027 | &l2_table[l2_index], |
0af729ec | 1028 | QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); |
c53ede9f KW |
1029 | assert(keep_clusters <= nb_clusters); |
1030 | ||
1031 | *bytes = MIN(*bytes, | |
1032 | keep_clusters * s->cluster_size | |
1033 | - offset_into_cluster(s, guest_offset)); | |
0af729ec KW |
1034 | |
1035 | ret = 1; | |
1036 | } else { | |
0af729ec KW |
1037 | ret = 0; |
1038 | } | |
1039 | ||
0af729ec | 1040 | /* Cleanup */ |
e62daaf6 | 1041 | out: |
a3f1afb4 | 1042 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
0af729ec | 1043 | |
e62daaf6 KW |
1044 | /* Only return a host offset if we actually made progress. Otherwise we |
1045 | * would make requirements for handle_alloc() that it can't fulfill */ | |
a97c67ee | 1046 | if (ret > 0) { |
411d62b0 KW |
1047 | *host_offset = (cluster_offset & L2E_OFFSET_MASK) |
1048 | + offset_into_cluster(s, guest_offset); | |
e62daaf6 KW |
1049 | } |
1050 | ||
0af729ec KW |
1051 | return ret; |
1052 | } | |
1053 | ||
226c3c26 KW |
1054 | /* |
1055 | * Allocates new clusters for the given guest_offset. | |
1056 | * | |
1057 | * At most *nb_clusters are allocated, and on return *nb_clusters is updated to | |
1058 | * contain the number of clusters that have been allocated and are contiguous | |
1059 | * in the image file. | |
1060 | * | |
1061 | * If *host_offset is non-zero, it specifies the offset in the image file at | |
1062 | * which the new clusters must start. *nb_clusters can be 0 on return in this | |
1063 | * case if the cluster at host_offset is already in use. If *host_offset is | |
1064 | * zero, the clusters can be allocated anywhere in the image file. | |
1065 | * | |
1066 | * *host_offset is updated to contain the offset into the image file at which | |
1067 | * the first allocated cluster starts. | |
1068 | * | |
1069 | * Return 0 on success and -errno in error cases. -EAGAIN means that the | |
1070 | * function has been waiting for another request and the allocation must be | |
1071 | * restarted, but the whole request should not be failed. | |
1072 | */ | |
1073 | static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, | |
b6d36def | 1074 | uint64_t *host_offset, uint64_t *nb_clusters) |
226c3c26 | 1075 | { |
ff99129a | 1076 | BDRVQcow2State *s = bs->opaque; |
226c3c26 KW |
1077 | |
1078 | trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, | |
1079 | *host_offset, *nb_clusters); | |
1080 | ||
250196f1 KW |
1081 | /* Allocate new clusters */ |
1082 | trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); | |
1083 | if (*host_offset == 0) { | |
df021791 KW |
1084 | int64_t cluster_offset = |
1085 | qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); | |
1086 | if (cluster_offset < 0) { | |
1087 | return cluster_offset; | |
1088 | } | |
1089 | *host_offset = cluster_offset; | |
1090 | return 0; | |
250196f1 | 1091 | } else { |
b6d36def | 1092 | int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); |
df021791 KW |
1093 | if (ret < 0) { |
1094 | return ret; | |
1095 | } | |
1096 | *nb_clusters = ret; | |
1097 | return 0; | |
250196f1 | 1098 | } |
250196f1 KW |
1099 | } |
1100 | ||
10f0ed8b KW |
1101 | /* |
1102 | * Allocates new clusters for an area that either is yet unallocated or needs a | |
1103 | * copy on write. If *host_offset is non-zero, clusters are only allocated if | |
1104 | * the new allocation can match the specified host offset. | |
1105 | * | |
411d62b0 KW |
1106 | * Note that guest_offset may not be cluster aligned. In this case, the |
1107 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1108 | * therefore isn't cluster aligned as well. | |
10f0ed8b KW |
1109 | * |
1110 | * Returns: | |
1111 | * 0: if no clusters could be allocated. *bytes is set to 0, | |
1112 | * *host_offset is left unchanged. | |
1113 | * | |
1114 | * 1: if new clusters were allocated. *bytes may be decreased if the | |
1115 | * new allocation doesn't cover all of the requested area. | |
1116 | * *host_offset is updated to contain the host offset of the first | |
1117 | * newly allocated cluster. | |
1118 | * | |
1119 | * -errno: in error cases | |
10f0ed8b KW |
1120 | */ |
1121 | static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, | |
c37f4cd7 | 1122 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
10f0ed8b | 1123 | { |
ff99129a | 1124 | BDRVQcow2State *s = bs->opaque; |
10f0ed8b KW |
1125 | int l2_index; |
1126 | uint64_t *l2_table; | |
1127 | uint64_t entry; | |
b6d36def | 1128 | uint64_t nb_clusters; |
10f0ed8b KW |
1129 | int ret; |
1130 | ||
10f0ed8b | 1131 | uint64_t alloc_cluster_offset; |
10f0ed8b KW |
1132 | |
1133 | trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset, | |
1134 | *bytes); | |
1135 | assert(*bytes > 0); | |
1136 | ||
f5bc6350 KW |
1137 | /* |
1138 | * Calculate the number of clusters to look for. We stop at L2 table | |
1139 | * boundaries to keep things simple. | |
1140 | */ | |
c37f4cd7 KW |
1141 | nb_clusters = |
1142 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1143 | ||
f5bc6350 | 1144 | l2_index = offset_to_l2_index(s, guest_offset); |
c37f4cd7 | 1145 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); |
b6d36def | 1146 | assert(nb_clusters <= INT_MAX); |
f5bc6350 | 1147 | |
10f0ed8b KW |
1148 | /* Find L2 entry for the first involved cluster */ |
1149 | ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); | |
1150 | if (ret < 0) { | |
1151 | return ret; | |
1152 | } | |
1153 | ||
3b8e2e26 | 1154 | entry = be64_to_cpu(l2_table[l2_index]); |
10f0ed8b KW |
1155 | |
1156 | /* For the moment, overwrite compressed clusters one by one */ | |
1157 | if (entry & QCOW_OFLAG_COMPRESSED) { | |
1158 | nb_clusters = 1; | |
1159 | } else { | |
3b8e2e26 | 1160 | nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index); |
10f0ed8b KW |
1161 | } |
1162 | ||
ecdd5333 KW |
1163 | /* This function is only called when there were no non-COW clusters, so if |
1164 | * we can't find any unallocated or COW clusters either, something is | |
1165 | * wrong with our code. */ | |
1166 | assert(nb_clusters > 0); | |
1167 | ||
a3f1afb4 | 1168 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
10f0ed8b | 1169 | |
10f0ed8b | 1170 | /* Allocate, if necessary at a given offset in the image file */ |
411d62b0 | 1171 | alloc_cluster_offset = start_of_cluster(s, *host_offset); |
83baa9a4 | 1172 | ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset, |
10f0ed8b KW |
1173 | &nb_clusters); |
1174 | if (ret < 0) { | |
1175 | goto fail; | |
1176 | } | |
1177 | ||
83baa9a4 KW |
1178 | /* Can't extend contiguous allocation */ |
1179 | if (nb_clusters == 0) { | |
10f0ed8b KW |
1180 | *bytes = 0; |
1181 | return 0; | |
1182 | } | |
1183 | ||
ff52aab2 HR |
1184 | /* !*host_offset would overwrite the image header and is reserved for "no |
1185 | * host offset preferred". If 0 was a valid host offset, it'd trigger the | |
1186 | * following overlap check; do that now to avoid having an invalid value in | |
1187 | * *host_offset. */ | |
1188 | if (!alloc_cluster_offset) { | |
1189 | ret = qcow2_pre_write_overlap_check(bs, 0, alloc_cluster_offset, | |
1190 | nb_clusters * s->cluster_size); | |
1191 | assert(ret < 0); | |
1192 | goto fail; | |
1193 | } | |
1194 | ||
83baa9a4 KW |
1195 | /* |
1196 | * Save info needed for meta data update. | |
1197 | * | |
1198 | * requested_sectors: Number of sectors from the start of the first | |
1199 | * newly allocated cluster to the end of the (possibly shortened | |
1200 | * before) write request. | |
1201 | * | |
1202 | * avail_sectors: Number of sectors from the start of the first | |
1203 | * newly allocated to the end of the last newly allocated cluster. | |
1204 | * | |
1205 | * nb_sectors: The number of sectors from the start of the first | |
1206 | * newly allocated cluster to the end of the area that the write | |
1207 | * request actually writes to (excluding COW at the end) | |
1208 | */ | |
1209 | int requested_sectors = | |
1210 | (*bytes + offset_into_cluster(s, guest_offset)) | |
1211 | >> BDRV_SECTOR_BITS; | |
1212 | int avail_sectors = nb_clusters | |
1213 | << (s->cluster_bits - BDRV_SECTOR_BITS); | |
1214 | int alloc_n_start = offset_into_cluster(s, guest_offset) | |
1215 | >> BDRV_SECTOR_BITS; | |
1216 | int nb_sectors = MIN(requested_sectors, avail_sectors); | |
88c6588c | 1217 | QCowL2Meta *old_m = *m; |
83baa9a4 | 1218 | |
83baa9a4 KW |
1219 | *m = g_malloc0(sizeof(**m)); |
1220 | ||
1221 | **m = (QCowL2Meta) { | |
88c6588c KW |
1222 | .next = old_m, |
1223 | ||
411d62b0 | 1224 | .alloc_offset = alloc_cluster_offset, |
83baa9a4 KW |
1225 | .offset = start_of_cluster(s, guest_offset), |
1226 | .nb_clusters = nb_clusters, | |
1227 | .nb_available = nb_sectors, | |
1228 | ||
1229 | .cow_start = { | |
1230 | .offset = 0, | |
1231 | .nb_sectors = alloc_n_start, | |
1232 | }, | |
1233 | .cow_end = { | |
1234 | .offset = nb_sectors * BDRV_SECTOR_SIZE, | |
1235 | .nb_sectors = avail_sectors - nb_sectors, | |
1236 | }, | |
1237 | }; | |
1238 | qemu_co_queue_init(&(*m)->dependent_requests); | |
1239 | QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); | |
1240 | ||
411d62b0 | 1241 | *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset); |
83baa9a4 KW |
1242 | *bytes = MIN(*bytes, (nb_sectors * BDRV_SECTOR_SIZE) |
1243 | - offset_into_cluster(s, guest_offset)); | |
1244 | assert(*bytes != 0); | |
1245 | ||
10f0ed8b KW |
1246 | return 1; |
1247 | ||
1248 | fail: | |
1249 | if (*m && (*m)->nb_clusters > 0) { | |
1250 | QLIST_REMOVE(*m, next_in_flight); | |
1251 | } | |
1252 | return ret; | |
1253 | } | |
1254 | ||
45aba42f KW |
1255 | /* |
1256 | * alloc_cluster_offset | |
1257 | * | |
250196f1 KW |
1258 | * For a given offset on the virtual disk, find the cluster offset in qcow2 |
1259 | * file. If the offset is not found, allocate a new cluster. | |
45aba42f | 1260 | * |
250196f1 | 1261 | * If the cluster was already allocated, m->nb_clusters is set to 0 and |
a7912369 | 1262 | * other fields in m are meaningless. |
148da7ea KW |
1263 | * |
1264 | * If the cluster is newly allocated, m->nb_clusters is set to the number of | |
68d100e9 KW |
1265 | * contiguous clusters that have been allocated. In this case, the other |
1266 | * fields of m are valid and contain information about the first allocated | |
1267 | * cluster. | |
45aba42f | 1268 | * |
68d100e9 KW |
1269 | * If the request conflicts with another write request in flight, the coroutine |
1270 | * is queued and will be reentered when the dependency has completed. | |
148da7ea KW |
1271 | * |
1272 | * Return 0 on success and -errno in error cases | |
45aba42f | 1273 | */ |
f4f0d391 | 1274 | int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, |
16f0587e | 1275 | int *num, uint64_t *host_offset, QCowL2Meta **m) |
45aba42f | 1276 | { |
ff99129a | 1277 | BDRVQcow2State *s = bs->opaque; |
710c2496 | 1278 | uint64_t start, remaining; |
250196f1 | 1279 | uint64_t cluster_offset; |
65eb2e35 | 1280 | uint64_t cur_bytes; |
710c2496 | 1281 | int ret; |
45aba42f | 1282 | |
16f0587e | 1283 | trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *num); |
3cce16f4 | 1284 | |
16f0587e | 1285 | assert((offset & ~BDRV_SECTOR_MASK) == 0); |
710c2496 | 1286 | |
72424114 | 1287 | again: |
16f0587e | 1288 | start = offset; |
11c89769 | 1289 | remaining = (uint64_t)*num << BDRV_SECTOR_BITS; |
0af729ec KW |
1290 | cluster_offset = 0; |
1291 | *host_offset = 0; | |
ecdd5333 KW |
1292 | cur_bytes = 0; |
1293 | *m = NULL; | |
0af729ec | 1294 | |
2c3b32d2 | 1295 | while (true) { |
ecdd5333 KW |
1296 | |
1297 | if (!*host_offset) { | |
1298 | *host_offset = start_of_cluster(s, cluster_offset); | |
1299 | } | |
1300 | ||
1301 | assert(remaining >= cur_bytes); | |
1302 | ||
1303 | start += cur_bytes; | |
1304 | remaining -= cur_bytes; | |
1305 | cluster_offset += cur_bytes; | |
1306 | ||
1307 | if (remaining == 0) { | |
1308 | break; | |
1309 | } | |
1310 | ||
1311 | cur_bytes = remaining; | |
1312 | ||
2c3b32d2 KW |
1313 | /* |
1314 | * Now start gathering as many contiguous clusters as possible: | |
1315 | * | |
1316 | * 1. Check for overlaps with in-flight allocations | |
1317 | * | |
1318 | * a) Overlap not in the first cluster -> shorten this request and | |
1319 | * let the caller handle the rest in its next loop iteration. | |
1320 | * | |
1321 | * b) Real overlaps of two requests. Yield and restart the search | |
1322 | * for contiguous clusters (the situation could have changed | |
1323 | * while we were sleeping) | |
1324 | * | |
1325 | * c) TODO: Request starts in the same cluster as the in-flight | |
1326 | * allocation ends. Shorten the COW of the in-fight allocation, | |
1327 | * set cluster_offset to write to the same cluster and set up | |
1328 | * the right synchronisation between the in-flight request and | |
1329 | * the new one. | |
1330 | */ | |
ecdd5333 | 1331 | ret = handle_dependencies(bs, start, &cur_bytes, m); |
2c3b32d2 | 1332 | if (ret == -EAGAIN) { |
ecdd5333 KW |
1333 | /* Currently handle_dependencies() doesn't yield if we already had |
1334 | * an allocation. If it did, we would have to clean up the L2Meta | |
1335 | * structs before starting over. */ | |
1336 | assert(*m == NULL); | |
2c3b32d2 KW |
1337 | goto again; |
1338 | } else if (ret < 0) { | |
1339 | return ret; | |
ecdd5333 KW |
1340 | } else if (cur_bytes == 0) { |
1341 | break; | |
2c3b32d2 KW |
1342 | } else { |
1343 | /* handle_dependencies() may have decreased cur_bytes (shortened | |
1344 | * the allocations below) so that the next dependency is processed | |
1345 | * correctly during the next loop iteration. */ | |
0af729ec | 1346 | } |
710c2496 | 1347 | |
2c3b32d2 KW |
1348 | /* |
1349 | * 2. Count contiguous COPIED clusters. | |
1350 | */ | |
1351 | ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m); | |
1352 | if (ret < 0) { | |
1353 | return ret; | |
1354 | } else if (ret) { | |
ecdd5333 | 1355 | continue; |
2c3b32d2 KW |
1356 | } else if (cur_bytes == 0) { |
1357 | break; | |
1358 | } | |
060bee89 | 1359 | |
2c3b32d2 KW |
1360 | /* |
1361 | * 3. If the request still hasn't completed, allocate new clusters, | |
1362 | * considering any cluster_offset of steps 1c or 2. | |
1363 | */ | |
1364 | ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m); | |
1365 | if (ret < 0) { | |
1366 | return ret; | |
1367 | } else if (ret) { | |
ecdd5333 | 1368 | continue; |
2c3b32d2 KW |
1369 | } else { |
1370 | assert(cur_bytes == 0); | |
1371 | break; | |
1372 | } | |
f5bc6350 | 1373 | } |
10f0ed8b | 1374 | |
16f0587e | 1375 | *num -= remaining >> BDRV_SECTOR_BITS; |
710c2496 KW |
1376 | assert(*num > 0); |
1377 | assert(*host_offset != 0); | |
45aba42f | 1378 | |
148da7ea | 1379 | return 0; |
45aba42f KW |
1380 | } |
1381 | ||
1382 | static int decompress_buffer(uint8_t *out_buf, int out_buf_size, | |
1383 | const uint8_t *buf, int buf_size) | |
1384 | { | |
1385 | z_stream strm1, *strm = &strm1; | |
1386 | int ret, out_len; | |
1387 | ||
1388 | memset(strm, 0, sizeof(*strm)); | |
1389 | ||
1390 | strm->next_in = (uint8_t *)buf; | |
1391 | strm->avail_in = buf_size; | |
1392 | strm->next_out = out_buf; | |
1393 | strm->avail_out = out_buf_size; | |
1394 | ||
1395 | ret = inflateInit2(strm, -12); | |
1396 | if (ret != Z_OK) | |
1397 | return -1; | |
1398 | ret = inflate(strm, Z_FINISH); | |
1399 | out_len = strm->next_out - out_buf; | |
1400 | if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) || | |
1401 | out_len != out_buf_size) { | |
1402 | inflateEnd(strm); | |
1403 | return -1; | |
1404 | } | |
1405 | inflateEnd(strm); | |
1406 | return 0; | |
1407 | } | |
1408 | ||
66f82cee | 1409 | int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset) |
45aba42f | 1410 | { |
ff99129a | 1411 | BDRVQcow2State *s = bs->opaque; |
45aba42f KW |
1412 | int ret, csize, nb_csectors, sector_offset; |
1413 | uint64_t coffset; | |
1414 | ||
1415 | coffset = cluster_offset & s->cluster_offset_mask; | |
1416 | if (s->cluster_cache_offset != coffset) { | |
1417 | nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1; | |
1418 | sector_offset = coffset & 511; | |
1419 | csize = nb_csectors * 512 - sector_offset; | |
66f82cee | 1420 | BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); |
9a4f4c31 KW |
1421 | ret = bdrv_read(bs->file->bs, coffset >> 9, s->cluster_data, |
1422 | nb_csectors); | |
45aba42f | 1423 | if (ret < 0) { |
8af36488 | 1424 | return ret; |
45aba42f KW |
1425 | } |
1426 | if (decompress_buffer(s->cluster_cache, s->cluster_size, | |
1427 | s->cluster_data + sector_offset, csize) < 0) { | |
8af36488 | 1428 | return -EIO; |
45aba42f KW |
1429 | } |
1430 | s->cluster_cache_offset = coffset; | |
1431 | } | |
1432 | return 0; | |
1433 | } | |
5ea929e3 KW |
1434 | |
1435 | /* | |
1436 | * This discards as many clusters of nb_clusters as possible at once (i.e. | |
1437 | * all clusters in the same L2 table) and returns the number of discarded | |
1438 | * clusters. | |
1439 | */ | |
1440 | static int discard_single_l2(BlockDriverState *bs, uint64_t offset, | |
b6d36def HR |
1441 | uint64_t nb_clusters, enum qcow2_discard_type type, |
1442 | bool full_discard) | |
5ea929e3 | 1443 | { |
ff99129a | 1444 | BDRVQcow2State *s = bs->opaque; |
3948d1d4 | 1445 | uint64_t *l2_table; |
5ea929e3 KW |
1446 | int l2_index; |
1447 | int ret; | |
1448 | int i; | |
1449 | ||
3948d1d4 | 1450 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); |
5ea929e3 KW |
1451 | if (ret < 0) { |
1452 | return ret; | |
1453 | } | |
1454 | ||
1455 | /* Limit nb_clusters to one L2 table */ | |
1456 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1457 | assert(nb_clusters <= INT_MAX); |
5ea929e3 KW |
1458 | |
1459 | for (i = 0; i < nb_clusters; i++) { | |
c883db0d | 1460 | uint64_t old_l2_entry; |
5ea929e3 | 1461 | |
c883db0d | 1462 | old_l2_entry = be64_to_cpu(l2_table[l2_index + i]); |
a71835a0 KW |
1463 | |
1464 | /* | |
808c4b6f HR |
1465 | * If full_discard is false, make sure that a discarded area reads back |
1466 | * as zeroes for v3 images (we cannot do it for v2 without actually | |
1467 | * writing a zero-filled buffer). We can skip the operation if the | |
1468 | * cluster is already marked as zero, or if it's unallocated and we | |
1469 | * don't have a backing file. | |
a71835a0 KW |
1470 | * |
1471 | * TODO We might want to use bdrv_get_block_status(bs) here, but we're | |
1472 | * holding s->lock, so that doesn't work today. | |
808c4b6f HR |
1473 | * |
1474 | * If full_discard is true, the sector should not read back as zeroes, | |
1475 | * but rather fall through to the backing file. | |
a71835a0 | 1476 | */ |
c883db0d HR |
1477 | switch (qcow2_get_cluster_type(old_l2_entry)) { |
1478 | case QCOW2_CLUSTER_UNALLOCATED: | |
760e0063 | 1479 | if (full_discard || !bs->backing) { |
c883db0d HR |
1480 | continue; |
1481 | } | |
1482 | break; | |
1483 | ||
1484 | case QCOW2_CLUSTER_ZERO: | |
808c4b6f HR |
1485 | if (!full_discard) { |
1486 | continue; | |
1487 | } | |
1488 | break; | |
c883db0d HR |
1489 | |
1490 | case QCOW2_CLUSTER_NORMAL: | |
1491 | case QCOW2_CLUSTER_COMPRESSED: | |
1492 | break; | |
1493 | ||
1494 | default: | |
1495 | abort(); | |
5ea929e3 KW |
1496 | } |
1497 | ||
1498 | /* First remove L2 entries */ | |
72e80b89 | 1499 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
808c4b6f | 1500 | if (!full_discard && s->qcow_version >= 3) { |
a71835a0 KW |
1501 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); |
1502 | } else { | |
1503 | l2_table[l2_index + i] = cpu_to_be64(0); | |
1504 | } | |
5ea929e3 KW |
1505 | |
1506 | /* Then decrease the refcount */ | |
c883db0d | 1507 | qcow2_free_any_clusters(bs, old_l2_entry, 1, type); |
5ea929e3 KW |
1508 | } |
1509 | ||
a3f1afb4 | 1510 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
5ea929e3 KW |
1511 | |
1512 | return nb_clusters; | |
1513 | } | |
1514 | ||
1515 | int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset, | |
808c4b6f | 1516 | int nb_sectors, enum qcow2_discard_type type, bool full_discard) |
5ea929e3 | 1517 | { |
ff99129a | 1518 | BDRVQcow2State *s = bs->opaque; |
5ea929e3 | 1519 | uint64_t end_offset; |
b6d36def | 1520 | uint64_t nb_clusters; |
5ea929e3 KW |
1521 | int ret; |
1522 | ||
1523 | end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS); | |
1524 | ||
1525 | /* Round start up and end down */ | |
1526 | offset = align_offset(offset, s->cluster_size); | |
ac95acdb | 1527 | end_offset = start_of_cluster(s, end_offset); |
5ea929e3 KW |
1528 | |
1529 | if (offset > end_offset) { | |
1530 | return 0; | |
1531 | } | |
1532 | ||
1533 | nb_clusters = size_to_clusters(s, end_offset - offset); | |
1534 | ||
0b919fae KW |
1535 | s->cache_discards = true; |
1536 | ||
5ea929e3 KW |
1537 | /* Each L2 table is handled by its own loop iteration */ |
1538 | while (nb_clusters > 0) { | |
808c4b6f | 1539 | ret = discard_single_l2(bs, offset, nb_clusters, type, full_discard); |
5ea929e3 | 1540 | if (ret < 0) { |
0b919fae | 1541 | goto fail; |
5ea929e3 KW |
1542 | } |
1543 | ||
1544 | nb_clusters -= ret; | |
1545 | offset += (ret * s->cluster_size); | |
1546 | } | |
1547 | ||
0b919fae KW |
1548 | ret = 0; |
1549 | fail: | |
1550 | s->cache_discards = false; | |
1551 | qcow2_process_discards(bs, ret); | |
1552 | ||
1553 | return ret; | |
5ea929e3 | 1554 | } |
621f0589 KW |
1555 | |
1556 | /* | |
1557 | * This zeroes as many clusters of nb_clusters as possible at once (i.e. | |
1558 | * all clusters in the same L2 table) and returns the number of zeroed | |
1559 | * clusters. | |
1560 | */ | |
1561 | static int zero_single_l2(BlockDriverState *bs, uint64_t offset, | |
b6d36def | 1562 | uint64_t nb_clusters) |
621f0589 | 1563 | { |
ff99129a | 1564 | BDRVQcow2State *s = bs->opaque; |
621f0589 KW |
1565 | uint64_t *l2_table; |
1566 | int l2_index; | |
1567 | int ret; | |
1568 | int i; | |
1569 | ||
1570 | ret = get_cluster_table(bs, offset, &l2_table, &l2_index); | |
1571 | if (ret < 0) { | |
1572 | return ret; | |
1573 | } | |
1574 | ||
1575 | /* Limit nb_clusters to one L2 table */ | |
1576 | nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); | |
b6d36def | 1577 | assert(nb_clusters <= INT_MAX); |
621f0589 KW |
1578 | |
1579 | for (i = 0; i < nb_clusters; i++) { | |
1580 | uint64_t old_offset; | |
1581 | ||
1582 | old_offset = be64_to_cpu(l2_table[l2_index + i]); | |
1583 | ||
1584 | /* Update L2 entries */ | |
72e80b89 | 1585 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
621f0589 KW |
1586 | if (old_offset & QCOW_OFLAG_COMPRESSED) { |
1587 | l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO); | |
6cfcb9b8 | 1588 | qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST); |
621f0589 KW |
1589 | } else { |
1590 | l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO); | |
1591 | } | |
1592 | } | |
1593 | ||
a3f1afb4 | 1594 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
621f0589 KW |
1595 | |
1596 | return nb_clusters; | |
1597 | } | |
1598 | ||
1599 | int qcow2_zero_clusters(BlockDriverState *bs, uint64_t offset, int nb_sectors) | |
1600 | { | |
ff99129a | 1601 | BDRVQcow2State *s = bs->opaque; |
b6d36def | 1602 | uint64_t nb_clusters; |
621f0589 KW |
1603 | int ret; |
1604 | ||
1605 | /* The zero flag is only supported by version 3 and newer */ | |
1606 | if (s->qcow_version < 3) { | |
1607 | return -ENOTSUP; | |
1608 | } | |
1609 | ||
1610 | /* Each L2 table is handled by its own loop iteration */ | |
1611 | nb_clusters = size_to_clusters(s, nb_sectors << BDRV_SECTOR_BITS); | |
1612 | ||
0b919fae KW |
1613 | s->cache_discards = true; |
1614 | ||
621f0589 KW |
1615 | while (nb_clusters > 0) { |
1616 | ret = zero_single_l2(bs, offset, nb_clusters); | |
1617 | if (ret < 0) { | |
0b919fae | 1618 | goto fail; |
621f0589 KW |
1619 | } |
1620 | ||
1621 | nb_clusters -= ret; | |
1622 | offset += (ret * s->cluster_size); | |
1623 | } | |
1624 | ||
0b919fae KW |
1625 | ret = 0; |
1626 | fail: | |
1627 | s->cache_discards = false; | |
1628 | qcow2_process_discards(bs, ret); | |
1629 | ||
1630 | return ret; | |
621f0589 | 1631 | } |
32b6444d HR |
1632 | |
1633 | /* | |
1634 | * Expands all zero clusters in a specific L1 table (or deallocates them, for | |
1635 | * non-backed non-pre-allocated zero clusters). | |
1636 | * | |
4057a2b2 HR |
1637 | * l1_entries and *visited_l1_entries are used to keep track of progress for |
1638 | * status_cb(). l1_entries contains the total number of L1 entries and | |
1639 | * *visited_l1_entries counts all visited L1 entries. | |
32b6444d HR |
1640 | */ |
1641 | static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table, | |
ecf58777 | 1642 | int l1_size, int64_t *visited_l1_entries, |
4057a2b2 | 1643 | int64_t l1_entries, |
8b13976d HR |
1644 | BlockDriverAmendStatusCB *status_cb, |
1645 | void *cb_opaque) | |
32b6444d | 1646 | { |
ff99129a | 1647 | BDRVQcow2State *s = bs->opaque; |
32b6444d HR |
1648 | bool is_active_l1 = (l1_table == s->l1_table); |
1649 | uint64_t *l2_table = NULL; | |
1650 | int ret; | |
1651 | int i, j; | |
1652 | ||
1653 | if (!is_active_l1) { | |
1654 | /* inactive L2 tables require a buffer to be stored in when loading | |
1655 | * them from disk */ | |
9a4f4c31 | 1656 | l2_table = qemu_try_blockalign(bs->file->bs, s->cluster_size); |
de82815d KW |
1657 | if (l2_table == NULL) { |
1658 | return -ENOMEM; | |
1659 | } | |
32b6444d HR |
1660 | } |
1661 | ||
1662 | for (i = 0; i < l1_size; i++) { | |
1663 | uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK; | |
1664 | bool l2_dirty = false; | |
0e06528e | 1665 | uint64_t l2_refcount; |
32b6444d HR |
1666 | |
1667 | if (!l2_offset) { | |
1668 | /* unallocated */ | |
4057a2b2 HR |
1669 | (*visited_l1_entries)++; |
1670 | if (status_cb) { | |
8b13976d | 1671 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 1672 | } |
32b6444d HR |
1673 | continue; |
1674 | } | |
1675 | ||
8dd93d93 HR |
1676 | if (offset_into_cluster(s, l2_offset)) { |
1677 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" | |
1678 | PRIx64 " unaligned (L1 index: %#x)", | |
1679 | l2_offset, i); | |
1680 | ret = -EIO; | |
1681 | goto fail; | |
1682 | } | |
1683 | ||
32b6444d HR |
1684 | if (is_active_l1) { |
1685 | /* get active L2 tables from cache */ | |
1686 | ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, | |
1687 | (void **)&l2_table); | |
1688 | } else { | |
1689 | /* load inactive L2 tables from disk */ | |
9a4f4c31 KW |
1690 | ret = bdrv_read(bs->file->bs, l2_offset / BDRV_SECTOR_SIZE, |
1691 | (void *)l2_table, s->cluster_sectors); | |
32b6444d HR |
1692 | } |
1693 | if (ret < 0) { | |
1694 | goto fail; | |
1695 | } | |
1696 | ||
7324c10f HR |
1697 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
1698 | &l2_refcount); | |
1699 | if (ret < 0) { | |
ecf58777 HR |
1700 | goto fail; |
1701 | } | |
1702 | ||
32b6444d HR |
1703 | for (j = 0; j < s->l2_size; j++) { |
1704 | uint64_t l2_entry = be64_to_cpu(l2_table[j]); | |
ecf58777 | 1705 | int64_t offset = l2_entry & L2E_OFFSET_MASK; |
32b6444d | 1706 | int cluster_type = qcow2_get_cluster_type(l2_entry); |
320c7066 | 1707 | bool preallocated = offset != 0; |
32b6444d | 1708 | |
ecf58777 | 1709 | if (cluster_type != QCOW2_CLUSTER_ZERO) { |
32b6444d HR |
1710 | continue; |
1711 | } | |
1712 | ||
320c7066 | 1713 | if (!preallocated) { |
760e0063 | 1714 | if (!bs->backing) { |
32b6444d HR |
1715 | /* not backed; therefore we can simply deallocate the |
1716 | * cluster */ | |
1717 | l2_table[j] = 0; | |
1718 | l2_dirty = true; | |
1719 | continue; | |
1720 | } | |
1721 | ||
1722 | offset = qcow2_alloc_clusters(bs, s->cluster_size); | |
1723 | if (offset < 0) { | |
1724 | ret = offset; | |
1725 | goto fail; | |
1726 | } | |
ecf58777 HR |
1727 | |
1728 | if (l2_refcount > 1) { | |
1729 | /* For shared L2 tables, set the refcount accordingly (it is | |
1730 | * already 1 and needs to be l2_refcount) */ | |
1731 | ret = qcow2_update_cluster_refcount(bs, | |
2aabe7c7 HR |
1732 | offset >> s->cluster_bits, |
1733 | refcount_diff(1, l2_refcount), false, | |
ecf58777 HR |
1734 | QCOW2_DISCARD_OTHER); |
1735 | if (ret < 0) { | |
1736 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1737 | QCOW2_DISCARD_OTHER); | |
1738 | goto fail; | |
1739 | } | |
1740 | } | |
32b6444d HR |
1741 | } |
1742 | ||
8dd93d93 HR |
1743 | if (offset_into_cluster(s, offset)) { |
1744 | qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset " | |
1745 | "%#" PRIx64 " unaligned (L2 offset: %#" | |
1746 | PRIx64 ", L2 index: %#x)", offset, | |
1747 | l2_offset, j); | |
1748 | if (!preallocated) { | |
1749 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1750 | QCOW2_DISCARD_ALWAYS); | |
1751 | } | |
1752 | ret = -EIO; | |
1753 | goto fail; | |
1754 | } | |
1755 | ||
231bb267 | 1756 | ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size); |
32b6444d | 1757 | if (ret < 0) { |
320c7066 HR |
1758 | if (!preallocated) { |
1759 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1760 | QCOW2_DISCARD_ALWAYS); | |
1761 | } | |
32b6444d HR |
1762 | goto fail; |
1763 | } | |
1764 | ||
9a4f4c31 | 1765 | ret = bdrv_write_zeroes(bs->file->bs, offset / BDRV_SECTOR_SIZE, |
aa7bfbff | 1766 | s->cluster_sectors, 0); |
32b6444d | 1767 | if (ret < 0) { |
320c7066 HR |
1768 | if (!preallocated) { |
1769 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
1770 | QCOW2_DISCARD_ALWAYS); | |
1771 | } | |
32b6444d HR |
1772 | goto fail; |
1773 | } | |
1774 | ||
ecf58777 HR |
1775 | if (l2_refcount == 1) { |
1776 | l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED); | |
1777 | } else { | |
1778 | l2_table[j] = cpu_to_be64(offset); | |
e390cf5a | 1779 | } |
ecf58777 | 1780 | l2_dirty = true; |
32b6444d HR |
1781 | } |
1782 | ||
1783 | if (is_active_l1) { | |
1784 | if (l2_dirty) { | |
72e80b89 | 1785 | qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table); |
32b6444d HR |
1786 | qcow2_cache_depends_on_flush(s->l2_table_cache); |
1787 | } | |
a3f1afb4 | 1788 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
32b6444d HR |
1789 | } else { |
1790 | if (l2_dirty) { | |
231bb267 HR |
1791 | ret = qcow2_pre_write_overlap_check(bs, |
1792 | QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, l2_offset, | |
32b6444d HR |
1793 | s->cluster_size); |
1794 | if (ret < 0) { | |
1795 | goto fail; | |
1796 | } | |
1797 | ||
9a4f4c31 KW |
1798 | ret = bdrv_write(bs->file->bs, l2_offset / BDRV_SECTOR_SIZE, |
1799 | (void *)l2_table, s->cluster_sectors); | |
32b6444d HR |
1800 | if (ret < 0) { |
1801 | goto fail; | |
1802 | } | |
1803 | } | |
1804 | } | |
4057a2b2 HR |
1805 | |
1806 | (*visited_l1_entries)++; | |
1807 | if (status_cb) { | |
8b13976d | 1808 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 1809 | } |
32b6444d HR |
1810 | } |
1811 | ||
1812 | ret = 0; | |
1813 | ||
1814 | fail: | |
1815 | if (l2_table) { | |
1816 | if (!is_active_l1) { | |
1817 | qemu_vfree(l2_table); | |
1818 | } else { | |
a3f1afb4 | 1819 | qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table); |
32b6444d HR |
1820 | } |
1821 | } | |
1822 | return ret; | |
1823 | } | |
1824 | ||
1825 | /* | |
1826 | * For backed images, expands all zero clusters on the image. For non-backed | |
1827 | * images, deallocates all non-pre-allocated zero clusters (and claims the | |
1828 | * allocation for pre-allocated ones). This is important for downgrading to a | |
1829 | * qcow2 version which doesn't yet support metadata zero clusters. | |
1830 | */ | |
4057a2b2 | 1831 | int qcow2_expand_zero_clusters(BlockDriverState *bs, |
8b13976d HR |
1832 | BlockDriverAmendStatusCB *status_cb, |
1833 | void *cb_opaque) | |
32b6444d | 1834 | { |
ff99129a | 1835 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 1836 | uint64_t *l1_table = NULL; |
4057a2b2 | 1837 | int64_t l1_entries = 0, visited_l1_entries = 0; |
32b6444d HR |
1838 | int ret; |
1839 | int i, j; | |
1840 | ||
4057a2b2 HR |
1841 | if (status_cb) { |
1842 | l1_entries = s->l1_size; | |
1843 | for (i = 0; i < s->nb_snapshots; i++) { | |
1844 | l1_entries += s->snapshots[i].l1_size; | |
1845 | } | |
1846 | } | |
1847 | ||
32b6444d | 1848 | ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size, |
4057a2b2 | 1849 | &visited_l1_entries, l1_entries, |
8b13976d | 1850 | status_cb, cb_opaque); |
32b6444d HR |
1851 | if (ret < 0) { |
1852 | goto fail; | |
1853 | } | |
1854 | ||
1855 | /* Inactive L1 tables may point to active L2 tables - therefore it is | |
1856 | * necessary to flush the L2 table cache before trying to access the L2 | |
1857 | * tables pointed to by inactive L1 entries (else we might try to expand | |
1858 | * zero clusters that have already been expanded); furthermore, it is also | |
1859 | * necessary to empty the L2 table cache, since it may contain tables which | |
1860 | * are now going to be modified directly on disk, bypassing the cache. | |
1861 | * qcow2_cache_empty() does both for us. */ | |
1862 | ret = qcow2_cache_empty(bs, s->l2_table_cache); | |
1863 | if (ret < 0) { | |
1864 | goto fail; | |
1865 | } | |
1866 | ||
1867 | for (i = 0; i < s->nb_snapshots; i++) { | |
1868 | int l1_sectors = (s->snapshots[i].l1_size * sizeof(uint64_t) + | |
1869 | BDRV_SECTOR_SIZE - 1) / BDRV_SECTOR_SIZE; | |
1870 | ||
1871 | l1_table = g_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE); | |
1872 | ||
9a4f4c31 KW |
1873 | ret = bdrv_read(bs->file->bs, |
1874 | s->snapshots[i].l1_table_offset / BDRV_SECTOR_SIZE, | |
1875 | (void *)l1_table, l1_sectors); | |
32b6444d HR |
1876 | if (ret < 0) { |
1877 | goto fail; | |
1878 | } | |
1879 | ||
1880 | for (j = 0; j < s->snapshots[i].l1_size; j++) { | |
1881 | be64_to_cpus(&l1_table[j]); | |
1882 | } | |
1883 | ||
1884 | ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size, | |
4057a2b2 | 1885 | &visited_l1_entries, l1_entries, |
8b13976d | 1886 | status_cb, cb_opaque); |
32b6444d HR |
1887 | if (ret < 0) { |
1888 | goto fail; | |
1889 | } | |
1890 | } | |
1891 | ||
1892 | ret = 0; | |
1893 | ||
1894 | fail: | |
32b6444d HR |
1895 | g_free(l1_table); |
1896 | return ret; | |
1897 | } |