]>
Commit | Line | Data |
---|---|---|
45aba42f KW |
1 | /* |
2 | * Block driver for the QCOW version 2 format | |
3 | * | |
4 | * Copyright (c) 2004-2006 Fabrice Bellard | |
5 | * | |
6 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
7 | * of this software and associated documentation files (the "Software"), to deal | |
8 | * in the Software without restriction, including without limitation the rights | |
9 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
10 | * copies of the Software, and to permit persons to whom the Software is | |
11 | * furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
22 | * THE SOFTWARE. | |
23 | */ | |
24 | ||
80c71a24 | 25 | #include "qemu/osdep.h" |
45aba42f KW |
26 | #include <zlib.h> |
27 | ||
c9a442e4 | 28 | #include "qapi/error.h" |
0d8c41da | 29 | #include "qcow2.h" |
58369e22 | 30 | #include "qemu/bswap.h" |
3cce16f4 | 31 | #include "trace.h" |
45aba42f | 32 | |
46b732cd PB |
33 | int qcow2_shrink_l1_table(BlockDriverState *bs, uint64_t exact_size) |
34 | { | |
35 | BDRVQcow2State *s = bs->opaque; | |
36 | int new_l1_size, i, ret; | |
37 | ||
38 | if (exact_size >= s->l1_size) { | |
39 | return 0; | |
40 | } | |
41 | ||
42 | new_l1_size = exact_size; | |
43 | ||
44 | #ifdef DEBUG_ALLOC2 | |
45 | fprintf(stderr, "shrink l1_table from %d to %d\n", s->l1_size, new_l1_size); | |
46 | #endif | |
47 | ||
48 | BLKDBG_EVENT(bs->file, BLKDBG_L1_SHRINK_WRITE_TABLE); | |
49 | ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset + | |
50 | new_l1_size * sizeof(uint64_t), | |
51 | (s->l1_size - new_l1_size) * sizeof(uint64_t), 0); | |
52 | if (ret < 0) { | |
53 | goto fail; | |
54 | } | |
55 | ||
56 | ret = bdrv_flush(bs->file->bs); | |
57 | if (ret < 0) { | |
58 | goto fail; | |
59 | } | |
60 | ||
61 | BLKDBG_EVENT(bs->file, BLKDBG_L1_SHRINK_FREE_L2_CLUSTERS); | |
62 | for (i = s->l1_size - 1; i > new_l1_size - 1; i--) { | |
63 | if ((s->l1_table[i] & L1E_OFFSET_MASK) == 0) { | |
64 | continue; | |
65 | } | |
66 | qcow2_free_clusters(bs, s->l1_table[i] & L1E_OFFSET_MASK, | |
67 | s->cluster_size, QCOW2_DISCARD_ALWAYS); | |
68 | s->l1_table[i] = 0; | |
69 | } | |
70 | return 0; | |
71 | ||
72 | fail: | |
73 | /* | |
74 | * If the write in the l1_table failed the image may contain a partially | |
75 | * overwritten l1_table. In this case it would be better to clear the | |
76 | * l1_table in memory to avoid possible image corruption. | |
77 | */ | |
78 | memset(s->l1_table + new_l1_size, 0, | |
79 | (s->l1_size - new_l1_size) * sizeof(uint64_t)); | |
80 | return ret; | |
81 | } | |
82 | ||
2cf7cfa1 KW |
83 | int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size, |
84 | bool exact_size) | |
45aba42f | 85 | { |
ff99129a | 86 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 | 87 | int new_l1_size2, ret, i; |
45aba42f | 88 | uint64_t *new_l1_table; |
fda74f82 | 89 | int64_t old_l1_table_offset, old_l1_size; |
2cf7cfa1 | 90 | int64_t new_l1_table_offset, new_l1_size; |
45aba42f KW |
91 | uint8_t data[12]; |
92 | ||
72893756 | 93 | if (min_size <= s->l1_size) |
45aba42f | 94 | return 0; |
72893756 | 95 | |
b93f9950 HR |
96 | /* Do a sanity check on min_size before trying to calculate new_l1_size |
97 | * (this prevents overflows during the while loop for the calculation of | |
98 | * new_l1_size) */ | |
99 | if (min_size > INT_MAX / sizeof(uint64_t)) { | |
100 | return -EFBIG; | |
101 | } | |
102 | ||
72893756 SH |
103 | if (exact_size) { |
104 | new_l1_size = min_size; | |
105 | } else { | |
106 | /* Bump size up to reduce the number of times we have to grow */ | |
107 | new_l1_size = s->l1_size; | |
108 | if (new_l1_size == 0) { | |
109 | new_l1_size = 1; | |
110 | } | |
111 | while (min_size > new_l1_size) { | |
21cf3e12 | 112 | new_l1_size = DIV_ROUND_UP(new_l1_size * 3, 2); |
72893756 | 113 | } |
45aba42f | 114 | } |
72893756 | 115 | |
84c26520 HR |
116 | QEMU_BUILD_BUG_ON(QCOW_MAX_L1_SIZE > INT_MAX); |
117 | if (new_l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) { | |
2cf7cfa1 KW |
118 | return -EFBIG; |
119 | } | |
120 | ||
45aba42f | 121 | #ifdef DEBUG_ALLOC2 |
2cf7cfa1 KW |
122 | fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n", |
123 | s->l1_size, new_l1_size); | |
45aba42f KW |
124 | #endif |
125 | ||
126 | new_l1_size2 = sizeof(uint64_t) * new_l1_size; | |
ef97d608 | 127 | new_l1_table = qemu_try_blockalign(bs->file->bs, new_l1_size2); |
de82815d KW |
128 | if (new_l1_table == NULL) { |
129 | return -ENOMEM; | |
130 | } | |
ef97d608 | 131 | memset(new_l1_table, 0, new_l1_size2); |
de82815d | 132 | |
0647d47c SH |
133 | if (s->l1_size) { |
134 | memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t)); | |
135 | } | |
45aba42f KW |
136 | |
137 | /* write new table (align to cluster) */ | |
66f82cee | 138 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE); |
ed6ccf0f | 139 | new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2); |
5d757b56 | 140 | if (new_l1_table_offset < 0) { |
de82815d | 141 | qemu_vfree(new_l1_table); |
5d757b56 KW |
142 | return new_l1_table_offset; |
143 | } | |
29c1a730 KW |
144 | |
145 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
146 | if (ret < 0) { | |
80fa3341 | 147 | goto fail; |
29c1a730 | 148 | } |
45aba42f | 149 | |
cf93980e HR |
150 | /* the L1 position has not yet been updated, so these clusters must |
151 | * indeed be completely free */ | |
231bb267 | 152 | ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset, |
966b000f | 153 | new_l1_size2, false); |
cf93980e HR |
154 | if (ret < 0) { |
155 | goto fail; | |
156 | } | |
157 | ||
66f82cee | 158 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE); |
45aba42f KW |
159 | for(i = 0; i < s->l1_size; i++) |
160 | new_l1_table[i] = cpu_to_be64(new_l1_table[i]); | |
d9ca2ea2 | 161 | ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, |
9a4f4c31 | 162 | new_l1_table, new_l1_size2); |
8b3b7206 | 163 | if (ret < 0) |
45aba42f KW |
164 | goto fail; |
165 | for(i = 0; i < s->l1_size; i++) | |
166 | new_l1_table[i] = be64_to_cpu(new_l1_table[i]); | |
167 | ||
168 | /* set new table */ | |
66f82cee | 169 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE); |
f1f7a1dd | 170 | stl_be_p(data, new_l1_size); |
e4ef9f46 | 171 | stq_be_p(data + 4, new_l1_table_offset); |
d9ca2ea2 | 172 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), |
9a4f4c31 | 173 | data, sizeof(data)); |
8b3b7206 | 174 | if (ret < 0) { |
45aba42f | 175 | goto fail; |
fb8fa77c | 176 | } |
de82815d | 177 | qemu_vfree(s->l1_table); |
fda74f82 | 178 | old_l1_table_offset = s->l1_table_offset; |
45aba42f KW |
179 | s->l1_table_offset = new_l1_table_offset; |
180 | s->l1_table = new_l1_table; | |
fda74f82 | 181 | old_l1_size = s->l1_size; |
45aba42f | 182 | s->l1_size = new_l1_size; |
fda74f82 HR |
183 | qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * sizeof(uint64_t), |
184 | QCOW2_DISCARD_OTHER); | |
45aba42f KW |
185 | return 0; |
186 | fail: | |
de82815d | 187 | qemu_vfree(new_l1_table); |
6cfcb9b8 KW |
188 | qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2, |
189 | QCOW2_DISCARD_OTHER); | |
8b3b7206 | 190 | return ret; |
45aba42f KW |
191 | } |
192 | ||
45aba42f KW |
193 | /* |
194 | * l2_load | |
195 | * | |
e2b5713e AG |
196 | * @bs: The BlockDriverState |
197 | * @offset: A guest offset, used to calculate what slice of the L2 | |
198 | * table to load. | |
199 | * @l2_offset: Offset to the L2 table in the image file. | |
200 | * @l2_slice: Location to store the pointer to the L2 slice. | |
45aba42f | 201 | * |
e2b5713e AG |
202 | * Loads a L2 slice into memory (L2 slices are the parts of L2 tables |
203 | * that are loaded by the qcow2 cache). If the slice is in the cache, | |
204 | * the cache is used; otherwise the L2 slice is loaded from the image | |
205 | * file. | |
45aba42f | 206 | */ |
e2b5713e AG |
207 | static int l2_load(BlockDriverState *bs, uint64_t offset, |
208 | uint64_t l2_offset, uint64_t **l2_slice) | |
45aba42f | 209 | { |
ff99129a | 210 | BDRVQcow2State *s = bs->opaque; |
e2b5713e AG |
211 | int start_of_slice = sizeof(uint64_t) * |
212 | (offset_to_l2_index(s, offset) - offset_to_l2_slice_index(s, offset)); | |
45aba42f | 213 | |
e2b5713e AG |
214 | return qcow2_cache_get(bs, s->l2_table_cache, l2_offset + start_of_slice, |
215 | (void **)l2_slice); | |
45aba42f KW |
216 | } |
217 | ||
6583e3c7 | 218 | /* |
da86f8cb AG |
219 | * Writes an L1 entry to disk (note that depending on the alignment |
220 | * requirements this function may write more that just one entry in | |
221 | * order to prevent bdrv_pwrite from performing a read-modify-write) | |
6583e3c7 | 222 | */ |
e23e400e | 223 | int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index) |
6583e3c7 | 224 | { |
ff99129a | 225 | BDRVQcow2State *s = bs->opaque; |
6583e3c7 | 226 | int l1_start_index; |
f7defcb6 | 227 | int i, ret; |
da86f8cb AG |
228 | int bufsize = MAX(sizeof(uint64_t), |
229 | MIN(bs->file->bs->bl.request_alignment, s->cluster_size)); | |
230 | int nentries = bufsize / sizeof(uint64_t); | |
231 | g_autofree uint64_t *buf = g_try_new0(uint64_t, nentries); | |
6583e3c7 | 232 | |
da86f8cb AG |
233 | if (buf == NULL) { |
234 | return -ENOMEM; | |
235 | } | |
236 | ||
237 | l1_start_index = QEMU_ALIGN_DOWN(l1_index, nentries); | |
238 | for (i = 0; i < MIN(nentries, s->l1_size - l1_start_index); i++) { | |
6583e3c7 KW |
239 | buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]); |
240 | } | |
241 | ||
231bb267 | 242 | ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1, |
da86f8cb | 243 | s->l1_table_offset + 8 * l1_start_index, bufsize, false); |
cf93980e HR |
244 | if (ret < 0) { |
245 | return ret; | |
246 | } | |
247 | ||
66f82cee | 248 | BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); |
d9ca2ea2 | 249 | ret = bdrv_pwrite_sync(bs->file, |
9a4f4c31 | 250 | s->l1_table_offset + 8 * l1_start_index, |
da86f8cb | 251 | buf, bufsize); |
f7defcb6 KW |
252 | if (ret < 0) { |
253 | return ret; | |
6583e3c7 KW |
254 | } |
255 | ||
256 | return 0; | |
257 | } | |
258 | ||
45aba42f KW |
259 | /* |
260 | * l2_allocate | |
261 | * | |
262 | * Allocate a new l2 entry in the file. If l1_index points to an already | |
263 | * used entry in the L2 table (i.e. we are doing a copy on write for the L2 | |
264 | * table) copy the contents of the old L2 table into the newly allocated one. | |
265 | * Otherwise the new table is initialized with zeros. | |
266 | * | |
267 | */ | |
268 | ||
3861946a | 269 | static int l2_allocate(BlockDriverState *bs, int l1_index) |
45aba42f | 270 | { |
ff99129a | 271 | BDRVQcow2State *s = bs->opaque; |
6583e3c7 | 272 | uint64_t old_l2_offset; |
3861946a AG |
273 | uint64_t *l2_slice = NULL; |
274 | unsigned slice, slice_size2, n_slices; | |
f4f0d391 | 275 | int64_t l2_offset; |
c46e1167 | 276 | int ret; |
45aba42f KW |
277 | |
278 | old_l2_offset = s->l1_table[l1_index]; | |
279 | ||
3cce16f4 KW |
280 | trace_qcow2_l2_allocate(bs, l1_index); |
281 | ||
45aba42f KW |
282 | /* allocate a new l2 entry */ |
283 | ||
ed6ccf0f | 284 | l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t)); |
5d757b56 | 285 | if (l2_offset < 0) { |
be0b742e HR |
286 | ret = l2_offset; |
287 | goto fail; | |
5d757b56 | 288 | } |
29c1a730 | 289 | |
c1c43990 AG |
290 | /* The offset must fit in the offset field of the L1 table entry */ |
291 | assert((l2_offset & L1E_OFFSET_MASK) == l2_offset); | |
292 | ||
98839750 AG |
293 | /* If we're allocating the table at offset 0 then something is wrong */ |
294 | if (l2_offset == 0) { | |
295 | qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid " | |
296 | "allocation of L2 table at offset 0"); | |
297 | ret = -EIO; | |
298 | goto fail; | |
299 | } | |
300 | ||
29c1a730 KW |
301 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
302 | if (ret < 0) { | |
303 | goto fail; | |
304 | } | |
45aba42f | 305 | |
45aba42f KW |
306 | /* allocate a new entry in the l2 cache */ |
307 | ||
3861946a AG |
308 | slice_size2 = s->l2_slice_size * sizeof(uint64_t); |
309 | n_slices = s->cluster_size / slice_size2; | |
310 | ||
3cce16f4 | 311 | trace_qcow2_l2_allocate_get_empty(bs, l1_index); |
3861946a | 312 | for (slice = 0; slice < n_slices; slice++) { |
6580bb09 | 313 | ret = qcow2_cache_get_empty(bs, s->l2_table_cache, |
3861946a AG |
314 | l2_offset + slice * slice_size2, |
315 | (void **) &l2_slice); | |
6580bb09 AG |
316 | if (ret < 0) { |
317 | goto fail; | |
318 | } | |
29c1a730 | 319 | |
6580bb09 | 320 | if ((old_l2_offset & L1E_OFFSET_MASK) == 0) { |
3861946a AG |
321 | /* if there was no old l2 table, clear the new slice */ |
322 | memset(l2_slice, 0, slice_size2); | |
6580bb09 | 323 | } else { |
3861946a AG |
324 | uint64_t *old_slice; |
325 | uint64_t old_l2_slice_offset = | |
326 | (old_l2_offset & L1E_OFFSET_MASK) + slice * slice_size2; | |
29c1a730 | 327 | |
3861946a | 328 | /* if there was an old l2 table, read a slice from the disk */ |
6580bb09 | 329 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ); |
3861946a AG |
330 | ret = qcow2_cache_get(bs, s->l2_table_cache, old_l2_slice_offset, |
331 | (void **) &old_slice); | |
6580bb09 AG |
332 | if (ret < 0) { |
333 | goto fail; | |
334 | } | |
335 | ||
3861946a | 336 | memcpy(l2_slice, old_slice, slice_size2); |
6580bb09 | 337 | |
3861946a | 338 | qcow2_cache_put(s->l2_table_cache, (void **) &old_slice); |
29c1a730 KW |
339 | } |
340 | ||
3861946a | 341 | /* write the l2 slice to the file */ |
6580bb09 | 342 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE); |
29c1a730 | 343 | |
6580bb09 | 344 | trace_qcow2_l2_allocate_write_l2(bs, l1_index); |
3861946a AG |
345 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
346 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
45aba42f | 347 | } |
29c1a730 | 348 | |
29c1a730 | 349 | ret = qcow2_cache_flush(bs, s->l2_table_cache); |
c46e1167 | 350 | if (ret < 0) { |
175e1152 KW |
351 | goto fail; |
352 | } | |
353 | ||
354 | /* update the L1 entry */ | |
3cce16f4 | 355 | trace_qcow2_l2_allocate_write_l1(bs, l1_index); |
175e1152 | 356 | s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; |
e23e400e | 357 | ret = qcow2_write_l1_entry(bs, l1_index); |
175e1152 KW |
358 | if (ret < 0) { |
359 | goto fail; | |
c46e1167 | 360 | } |
45aba42f | 361 | |
3cce16f4 | 362 | trace_qcow2_l2_allocate_done(bs, l1_index, 0); |
c46e1167 | 363 | return 0; |
175e1152 KW |
364 | |
365 | fail: | |
3cce16f4 | 366 | trace_qcow2_l2_allocate_done(bs, l1_index, ret); |
3861946a AG |
367 | if (l2_slice != NULL) { |
368 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
8585afd8 | 369 | } |
68dba0bf | 370 | s->l1_table[l1_index] = old_l2_offset; |
e3b21ef9 HR |
371 | if (l2_offset > 0) { |
372 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), | |
373 | QCOW2_DISCARD_ALWAYS); | |
374 | } | |
175e1152 | 375 | return ret; |
45aba42f KW |
376 | } |
377 | ||
2bfcc4a0 | 378 | /* |
13f893c4 | 379 | * Checks how many clusters in a given L2 slice are contiguous in the image |
2bfcc4a0 KW |
380 | * file. As soon as one of the flags in the bitmask stop_flags changes compared |
381 | * to the first cluster, the search is stopped and the cluster is not counted | |
382 | * as contiguous. (This allows it, for example, to stop at the first compressed | |
383 | * cluster which may require a different handling) | |
384 | */ | |
808c2bb4 | 385 | static int count_contiguous_clusters(BlockDriverState *bs, int nb_clusters, |
12c6aebe | 386 | int cluster_size, uint64_t *l2_slice, int l2_index, uint64_t stop_flags) |
45aba42f | 387 | { |
12c6aebe | 388 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 389 | int i; |
3ef95218 | 390 | QCow2ClusterType first_cluster_type; |
78a52ad5 | 391 | uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED; |
12c6aebe | 392 | uint64_t first_entry = get_l2_entry(s, l2_slice, l2_index); |
15684a47 | 393 | uint64_t offset = first_entry & mask; |
45aba42f | 394 | |
b8c8353a KW |
395 | first_cluster_type = qcow2_get_cluster_type(bs, first_entry); |
396 | if (first_cluster_type == QCOW2_CLUSTER_UNALLOCATED) { | |
45aba42f | 397 | return 0; |
564a6b69 | 398 | } |
45aba42f | 399 | |
564a6b69 | 400 | /* must be allocated */ |
564a6b69 | 401 | assert(first_cluster_type == QCOW2_CLUSTER_NORMAL || |
fdfab37d | 402 | first_cluster_type == QCOW2_CLUSTER_ZERO_ALLOC); |
15684a47 | 403 | |
61653008 | 404 | for (i = 0; i < nb_clusters; i++) { |
12c6aebe | 405 | uint64_t l2_entry = get_l2_entry(s, l2_slice, l2_index + i) & mask; |
2bfcc4a0 | 406 | if (offset + (uint64_t) i * cluster_size != l2_entry) { |
45aba42f | 407 | break; |
2bfcc4a0 KW |
408 | } |
409 | } | |
45aba42f | 410 | |
7d37435b | 411 | return i; |
45aba42f KW |
412 | } |
413 | ||
4341df8a EB |
414 | /* |
415 | * Checks how many consecutive unallocated clusters in a given L2 | |
c26f10ba | 416 | * slice have the same cluster type. |
4341df8a | 417 | */ |
808c2bb4 KW |
418 | static int count_contiguous_clusters_unallocated(BlockDriverState *bs, |
419 | int nb_clusters, | |
c26f10ba | 420 | uint64_t *l2_slice, |
12c6aebe | 421 | int l2_index, |
3ef95218 | 422 | QCow2ClusterType wanted_type) |
45aba42f | 423 | { |
12c6aebe | 424 | BDRVQcow2State *s = bs->opaque; |
2bfcc4a0 KW |
425 | int i; |
426 | ||
fdfab37d | 427 | assert(wanted_type == QCOW2_CLUSTER_ZERO_PLAIN || |
4341df8a | 428 | wanted_type == QCOW2_CLUSTER_UNALLOCATED); |
2bfcc4a0 | 429 | for (i = 0; i < nb_clusters; i++) { |
12c6aebe | 430 | uint64_t entry = get_l2_entry(s, l2_slice, l2_index + i); |
808c2bb4 | 431 | QCow2ClusterType type = qcow2_get_cluster_type(bs, entry); |
45aba42f | 432 | |
fdfab37d | 433 | if (type != wanted_type) { |
2bfcc4a0 KW |
434 | break; |
435 | } | |
436 | } | |
45aba42f KW |
437 | |
438 | return i; | |
439 | } | |
440 | ||
672f0f2c AG |
441 | static int coroutine_fn do_perform_cow_read(BlockDriverState *bs, |
442 | uint64_t src_cluster_offset, | |
443 | unsigned offset_in_cluster, | |
86b862c4 | 444 | QEMUIOVector *qiov) |
45aba42f | 445 | { |
aaa4d20b | 446 | int ret; |
1b9f1491 | 447 | |
86b862c4 | 448 | if (qiov->size == 0) { |
99450c6f AG |
449 | return 0; |
450 | } | |
451 | ||
66f82cee | 452 | BLKDBG_EVENT(bs->file, BLKDBG_COW_READ); |
aef4acb6 | 453 | |
dba28555 | 454 | if (!bs->drv) { |
672f0f2c | 455 | return -ENOMEDIUM; |
dba28555 HR |
456 | } |
457 | ||
aef4acb6 SH |
458 | /* Call .bdrv_co_readv() directly instead of using the public block-layer |
459 | * interface. This avoids double I/O throttling and request tracking, | |
460 | * which can lead to deadlock when block layer copy-on-read is enabled. | |
461 | */ | |
df893d25 VSO |
462 | ret = bs->drv->bdrv_co_preadv_part(bs, |
463 | src_cluster_offset + offset_in_cluster, | |
464 | qiov->size, qiov, 0, 0); | |
1b9f1491 | 465 | if (ret < 0) { |
672f0f2c | 466 | return ret; |
1b9f1491 KW |
467 | } |
468 | ||
672f0f2c AG |
469 | return 0; |
470 | } | |
471 | ||
672f0f2c AG |
472 | static int coroutine_fn do_perform_cow_write(BlockDriverState *bs, |
473 | uint64_t cluster_offset, | |
474 | unsigned offset_in_cluster, | |
86b862c4 | 475 | QEMUIOVector *qiov) |
672f0f2c | 476 | { |
966b000f | 477 | BDRVQcow2State *s = bs->opaque; |
672f0f2c AG |
478 | int ret; |
479 | ||
86b862c4 | 480 | if (qiov->size == 0) { |
672f0f2c AG |
481 | return 0; |
482 | } | |
483 | ||
231bb267 | 484 | ret = qcow2_pre_write_overlap_check(bs, 0, |
966b000f | 485 | cluster_offset + offset_in_cluster, qiov->size, true); |
cf93980e | 486 | if (ret < 0) { |
672f0f2c | 487 | return ret; |
cf93980e HR |
488 | } |
489 | ||
66f82cee | 490 | BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE); |
966b000f | 491 | ret = bdrv_co_pwritev(s->data_file, cluster_offset + offset_in_cluster, |
86b862c4 | 492 | qiov->size, qiov, 0); |
1b9f1491 | 493 | if (ret < 0) { |
672f0f2c | 494 | return ret; |
1b9f1491 KW |
495 | } |
496 | ||
672f0f2c | 497 | return 0; |
45aba42f KW |
498 | } |
499 | ||
500 | ||
501 | /* | |
388e5816 | 502 | * get_host_offset |
45aba42f | 503 | * |
388e5816 AG |
504 | * For a given offset of the virtual disk find the equivalent host |
505 | * offset in the qcow2 file and store it in *host_offset. Neither | |
506 | * offset needs to be aligned to a cluster boundary. | |
507 | * | |
508 | * If the cluster is unallocated then *host_offset will be 0. | |
509 | * If the cluster is compressed then *host_offset will contain the | |
510 | * complete compressed cluster descriptor. | |
45aba42f | 511 | * |
ecfe1863 KW |
512 | * On entry, *bytes is the maximum number of contiguous bytes starting at |
513 | * offset that we are interested in. | |
45aba42f | 514 | * |
ecfe1863 KW |
515 | * On exit, *bytes is the number of bytes starting at offset that have the same |
516 | * cluster type and (if applicable) are stored contiguously in the image file. | |
517 | * Compressed clusters are always returned one by one. | |
45aba42f | 518 | * |
68d000a3 KW |
519 | * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error |
520 | * cases. | |
45aba42f | 521 | */ |
388e5816 AG |
522 | int qcow2_get_host_offset(BlockDriverState *bs, uint64_t offset, |
523 | unsigned int *bytes, uint64_t *host_offset) | |
45aba42f | 524 | { |
ff99129a | 525 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 | 526 | unsigned int l2_index; |
388e5816 | 527 | uint64_t l1_index, l2_offset, *l2_slice, l2_entry; |
fd630039 | 528 | int c; |
c834cba9 HR |
529 | unsigned int offset_in_cluster; |
530 | uint64_t bytes_available, bytes_needed, nb_clusters; | |
3ef95218 | 531 | QCow2ClusterType type; |
55c17e98 | 532 | int ret; |
45aba42f | 533 | |
b2f65d6b | 534 | offset_in_cluster = offset_into_cluster(s, offset); |
ecfe1863 | 535 | bytes_needed = (uint64_t) *bytes + offset_in_cluster; |
45aba42f | 536 | |
b2f65d6b | 537 | /* compute how many bytes there are between the start of the cluster |
fd630039 AG |
538 | * containing offset and the end of the l2 slice that contains |
539 | * the entry pointing to it */ | |
540 | bytes_available = | |
541 | ((uint64_t) (s->l2_slice_size - offset_to_l2_slice_index(s, offset))) | |
542 | << s->cluster_bits; | |
45aba42f | 543 | |
b2f65d6b KW |
544 | if (bytes_needed > bytes_available) { |
545 | bytes_needed = bytes_available; | |
45aba42f KW |
546 | } |
547 | ||
388e5816 | 548 | *host_offset = 0; |
45aba42f | 549 | |
b6af0975 | 550 | /* seek to the l2 offset in the l1 table */ |
45aba42f | 551 | |
05b5b6ee | 552 | l1_index = offset_to_l1_index(s, offset); |
68d000a3 | 553 | if (l1_index >= s->l1_size) { |
3ef95218 | 554 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 555 | goto out; |
68d000a3 | 556 | } |
45aba42f | 557 | |
68d000a3 KW |
558 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
559 | if (!l2_offset) { | |
3ef95218 | 560 | type = QCOW2_CLUSTER_UNALLOCATED; |
45aba42f | 561 | goto out; |
68d000a3 | 562 | } |
45aba42f | 563 | |
a97c67ee HR |
564 | if (offset_into_cluster(s, l2_offset)) { |
565 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
566 | " unaligned (L1 index: %#" PRIx64 ")", | |
567 | l2_offset, l1_index); | |
568 | return -EIO; | |
569 | } | |
570 | ||
fd630039 | 571 | /* load the l2 slice in memory */ |
45aba42f | 572 | |
fd630039 | 573 | ret = l2_load(bs, offset, l2_offset, &l2_slice); |
55c17e98 KW |
574 | if (ret < 0) { |
575 | return ret; | |
1c46efaa | 576 | } |
45aba42f KW |
577 | |
578 | /* find the cluster offset for the given disk offset */ | |
579 | ||
fd630039 | 580 | l2_index = offset_to_l2_slice_index(s, offset); |
12c6aebe | 581 | l2_entry = get_l2_entry(s, l2_slice, l2_index); |
b6d36def | 582 | |
b2f65d6b | 583 | nb_clusters = size_to_clusters(s, bytes_needed); |
c834cba9 HR |
584 | /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned |
585 | * integers; the minimum cluster size is 512, so this assertion is always | |
586 | * true */ | |
587 | assert(nb_clusters <= INT_MAX); | |
45aba42f | 588 | |
388e5816 | 589 | type = qcow2_get_cluster_type(bs, l2_entry); |
fdfab37d EB |
590 | if (s->qcow_version < 3 && (type == QCOW2_CLUSTER_ZERO_PLAIN || |
591 | type == QCOW2_CLUSTER_ZERO_ALLOC)) { | |
592 | qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found" | |
593 | " in pre-v3 image (L2 offset: %#" PRIx64 | |
594 | ", L2 index: %#x)", l2_offset, l2_index); | |
595 | ret = -EIO; | |
596 | goto fail; | |
597 | } | |
3ef95218 | 598 | switch (type) { |
68d000a3 | 599 | case QCOW2_CLUSTER_COMPRESSED: |
966b000f KW |
600 | if (has_data_file(bs)) { |
601 | qcow2_signal_corruption(bs, true, -1, -1, "Compressed cluster " | |
602 | "entry found in image with external data " | |
603 | "file (L2 offset: %#" PRIx64 ", L2 index: " | |
604 | "%#x)", l2_offset, l2_index); | |
605 | ret = -EIO; | |
606 | goto fail; | |
607 | } | |
68d000a3 KW |
608 | /* Compressed clusters can only be processed one by one */ |
609 | c = 1; | |
388e5816 | 610 | *host_offset = l2_entry & L2E_COMPRESSED_OFFSET_SIZE_MASK; |
68d000a3 | 611 | break; |
fdfab37d | 612 | case QCOW2_CLUSTER_ZERO_PLAIN: |
68d000a3 | 613 | case QCOW2_CLUSTER_UNALLOCATED: |
45aba42f | 614 | /* how many empty clusters ? */ |
808c2bb4 | 615 | c = count_contiguous_clusters_unallocated(bs, nb_clusters, |
12c6aebe | 616 | l2_slice, l2_index, type); |
68d000a3 | 617 | break; |
fdfab37d | 618 | case QCOW2_CLUSTER_ZERO_ALLOC: |
388e5816 AG |
619 | case QCOW2_CLUSTER_NORMAL: { |
620 | uint64_t host_cluster_offset = l2_entry & L2E_OFFSET_MASK; | |
621 | *host_offset = host_cluster_offset + offset_in_cluster; | |
45aba42f | 622 | /* how many allocated clusters ? */ |
808c2bb4 | 623 | c = count_contiguous_clusters(bs, nb_clusters, s->cluster_size, |
12c6aebe | 624 | l2_slice, l2_index, QCOW_OFLAG_ZERO); |
388e5816 | 625 | if (offset_into_cluster(s, host_cluster_offset)) { |
fdfab37d EB |
626 | qcow2_signal_corruption(bs, true, -1, -1, |
627 | "Cluster allocation offset %#" | |
a97c67ee | 628 | PRIx64 " unaligned (L2 offset: %#" PRIx64 |
388e5816 | 629 | ", L2 index: %#x)", host_cluster_offset, |
a97c67ee HR |
630 | l2_offset, l2_index); |
631 | ret = -EIO; | |
632 | goto fail; | |
633 | } | |
388e5816 | 634 | if (has_data_file(bs) && *host_offset != offset) { |
966b000f KW |
635 | qcow2_signal_corruption(bs, true, -1, -1, |
636 | "External data file host cluster offset %#" | |
637 | PRIx64 " does not match guest cluster " | |
638 | "offset: %#" PRIx64 | |
388e5816 | 639 | ", L2 index: %#x)", host_cluster_offset, |
966b000f KW |
640 | offset - offset_in_cluster, l2_index); |
641 | ret = -EIO; | |
642 | goto fail; | |
643 | } | |
68d000a3 | 644 | break; |
388e5816 | 645 | } |
1417d7e4 KW |
646 | default: |
647 | abort(); | |
45aba42f KW |
648 | } |
649 | ||
fd630039 | 650 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
29c1a730 | 651 | |
c834cba9 | 652 | bytes_available = (int64_t)c * s->cluster_size; |
68d000a3 | 653 | |
45aba42f | 654 | out: |
b2f65d6b KW |
655 | if (bytes_available > bytes_needed) { |
656 | bytes_available = bytes_needed; | |
657 | } | |
45aba42f | 658 | |
c834cba9 HR |
659 | /* bytes_available <= bytes_needed <= *bytes + offset_in_cluster; |
660 | * subtracting offset_in_cluster will therefore definitely yield something | |
661 | * not exceeding UINT_MAX */ | |
662 | assert(bytes_available - offset_in_cluster <= UINT_MAX); | |
ecfe1863 | 663 | *bytes = bytes_available - offset_in_cluster; |
45aba42f | 664 | |
3ef95218 | 665 | return type; |
a97c67ee HR |
666 | |
667 | fail: | |
fd630039 | 668 | qcow2_cache_put(s->l2_table_cache, (void **)&l2_slice); |
a97c67ee | 669 | return ret; |
45aba42f KW |
670 | } |
671 | ||
672 | /* | |
673 | * get_cluster_table | |
674 | * | |
675 | * for a given disk offset, load (and allocate if needed) | |
c03bfc5b | 676 | * the appropriate slice of its l2 table. |
45aba42f | 677 | * |
c03bfc5b | 678 | * the cluster index in the l2 slice is given to the caller. |
45aba42f | 679 | * |
1e3e8f1a | 680 | * Returns 0 on success, -errno in failure case |
45aba42f | 681 | */ |
45aba42f | 682 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
c03bfc5b | 683 | uint64_t **new_l2_slice, |
45aba42f KW |
684 | int *new_l2_index) |
685 | { | |
ff99129a | 686 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
687 | unsigned int l2_index; |
688 | uint64_t l1_index, l2_offset; | |
c03bfc5b | 689 | uint64_t *l2_slice = NULL; |
80ee15a6 | 690 | int ret; |
45aba42f | 691 | |
b6af0975 | 692 | /* seek to the l2 offset in the l1 table */ |
45aba42f | 693 | |
05b5b6ee | 694 | l1_index = offset_to_l1_index(s, offset); |
45aba42f | 695 | if (l1_index >= s->l1_size) { |
72893756 | 696 | ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
1e3e8f1a KW |
697 | if (ret < 0) { |
698 | return ret; | |
699 | } | |
45aba42f | 700 | } |
8e37f681 | 701 | |
2cf7cfa1 | 702 | assert(l1_index < s->l1_size); |
8e37f681 | 703 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
a97c67ee HR |
704 | if (offset_into_cluster(s, l2_offset)) { |
705 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
706 | " unaligned (L1 index: %#" PRIx64 ")", | |
707 | l2_offset, l1_index); | |
708 | return -EIO; | |
709 | } | |
45aba42f | 710 | |
05f9ee46 | 711 | if (!(s->l1_table[l1_index] & QCOW_OFLAG_COPIED)) { |
16fde5f2 | 712 | /* First allocate a new L2 table (and do COW if needed) */ |
3861946a | 713 | ret = l2_allocate(bs, l1_index); |
c46e1167 KW |
714 | if (ret < 0) { |
715 | return ret; | |
1e3e8f1a | 716 | } |
16fde5f2 KW |
717 | |
718 | /* Then decrease the refcount of the old table */ | |
719 | if (l2_offset) { | |
6cfcb9b8 KW |
720 | qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t), |
721 | QCOW2_DISCARD_OTHER); | |
16fde5f2 | 722 | } |
3861946a AG |
723 | |
724 | /* Get the offset of the newly-allocated l2 table */ | |
725 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; | |
726 | assert(offset_into_cluster(s, l2_offset) == 0); | |
05f9ee46 AG |
727 | } |
728 | ||
c03bfc5b AG |
729 | /* load the l2 slice in memory */ |
730 | ret = l2_load(bs, offset, l2_offset, &l2_slice); | |
05f9ee46 AG |
731 | if (ret < 0) { |
732 | return ret; | |
45aba42f KW |
733 | } |
734 | ||
735 | /* find the cluster offset for the given disk offset */ | |
736 | ||
c03bfc5b | 737 | l2_index = offset_to_l2_slice_index(s, offset); |
45aba42f | 738 | |
c03bfc5b | 739 | *new_l2_slice = l2_slice; |
45aba42f KW |
740 | *new_l2_index = l2_index; |
741 | ||
1e3e8f1a | 742 | return 0; |
45aba42f KW |
743 | } |
744 | ||
745 | /* | |
746 | * alloc_compressed_cluster_offset | |
747 | * | |
77e023ff KW |
748 | * For a given offset on the virtual disk, allocate a new compressed cluster |
749 | * and put the host offset of the cluster into *host_offset. If a cluster is | |
750 | * already allocated at the offset, return an error. | |
45aba42f | 751 | * |
77e023ff | 752 | * Return 0 on success and -errno in error cases |
45aba42f | 753 | */ |
77e023ff KW |
754 | int qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, |
755 | uint64_t offset, | |
756 | int compressed_size, | |
757 | uint64_t *host_offset) | |
45aba42f | 758 | { |
ff99129a | 759 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 760 | int l2_index, ret; |
e4e72548 | 761 | uint64_t *l2_slice; |
f4f0d391 | 762 | int64_t cluster_offset; |
45aba42f KW |
763 | int nb_csectors; |
764 | ||
966b000f KW |
765 | if (has_data_file(bs)) { |
766 | return 0; | |
767 | } | |
768 | ||
e4e72548 | 769 | ret = get_cluster_table(bs, offset, &l2_slice, &l2_index); |
1e3e8f1a | 770 | if (ret < 0) { |
77e023ff | 771 | return ret; |
1e3e8f1a | 772 | } |
45aba42f | 773 | |
b0b6862e KW |
774 | /* Compression can't overwrite anything. Fail if the cluster was already |
775 | * allocated. */ | |
12c6aebe | 776 | cluster_offset = get_l2_entry(s, l2_slice, l2_index); |
b0b6862e | 777 | if (cluster_offset & L2E_OFFSET_MASK) { |
e4e72548 | 778 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
77e023ff | 779 | return -EIO; |
8f1efd00 | 780 | } |
45aba42f | 781 | |
ed6ccf0f | 782 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
5d757b56 | 783 | if (cluster_offset < 0) { |
e4e72548 | 784 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
77e023ff | 785 | return cluster_offset; |
5d757b56 KW |
786 | } |
787 | ||
b6c24694 AG |
788 | nb_csectors = |
789 | (cluster_offset + compressed_size - 1) / QCOW2_COMPRESSED_SECTOR_SIZE - | |
790 | (cluster_offset / QCOW2_COMPRESSED_SECTOR_SIZE); | |
45aba42f | 791 | |
3a75a870 AG |
792 | /* The offset and size must fit in their fields of the L2 table entry */ |
793 | assert((cluster_offset & s->cluster_offset_mask) == cluster_offset); | |
794 | assert((nb_csectors & s->csize_mask) == nb_csectors); | |
795 | ||
45aba42f KW |
796 | cluster_offset |= QCOW_OFLAG_COMPRESSED | |
797 | ((uint64_t)nb_csectors << s->csize_shift); | |
798 | ||
799 | /* update L2 table */ | |
800 | ||
801 | /* compressed clusters never have the copied flag */ | |
802 | ||
66f82cee | 803 | BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
e4e72548 | 804 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
12c6aebe | 805 | set_l2_entry(s, l2_slice, l2_index, cluster_offset); |
e4e72548 | 806 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
4c1612d9 | 807 | |
77e023ff KW |
808 | *host_offset = cluster_offset & s->cluster_offset_mask; |
809 | return 0; | |
4c1612d9 KW |
810 | } |
811 | ||
99450c6f | 812 | static int perform_cow(BlockDriverState *bs, QCowL2Meta *m) |
593fb83c | 813 | { |
ff99129a | 814 | BDRVQcow2State *s = bs->opaque; |
99450c6f AG |
815 | Qcow2COWRegion *start = &m->cow_start; |
816 | Qcow2COWRegion *end = &m->cow_end; | |
672f0f2c | 817 | unsigned buffer_size; |
b3cf1c7c AG |
818 | unsigned data_bytes = end->offset - (start->offset + start->nb_bytes); |
819 | bool merge_reads; | |
672f0f2c | 820 | uint8_t *start_buffer, *end_buffer; |
86b862c4 | 821 | QEMUIOVector qiov; |
593fb83c KW |
822 | int ret; |
823 | ||
672f0f2c | 824 | assert(start->nb_bytes <= UINT_MAX - end->nb_bytes); |
b3cf1c7c AG |
825 | assert(start->nb_bytes + end->nb_bytes <= UINT_MAX - data_bytes); |
826 | assert(start->offset + start->nb_bytes <= end->offset); | |
672f0f2c | 827 | |
c8bb23cb | 828 | if ((start->nb_bytes == 0 && end->nb_bytes == 0) || m->skip_cow) { |
593fb83c KW |
829 | return 0; |
830 | } | |
831 | ||
b3cf1c7c AG |
832 | /* If we have to read both the start and end COW regions and the |
833 | * middle region is not too large then perform just one read | |
834 | * operation */ | |
835 | merge_reads = start->nb_bytes && end->nb_bytes && data_bytes <= 16384; | |
836 | if (merge_reads) { | |
837 | buffer_size = start->nb_bytes + data_bytes + end->nb_bytes; | |
838 | } else { | |
839 | /* If we have to do two reads, add some padding in the middle | |
840 | * if necessary to make sure that the end region is optimally | |
841 | * aligned. */ | |
842 | size_t align = bdrv_opt_mem_align(bs); | |
843 | assert(align > 0 && align <= UINT_MAX); | |
844 | assert(QEMU_ALIGN_UP(start->nb_bytes, align) <= | |
845 | UINT_MAX - end->nb_bytes); | |
846 | buffer_size = QEMU_ALIGN_UP(start->nb_bytes, align) + end->nb_bytes; | |
847 | } | |
848 | ||
849 | /* Reserve a buffer large enough to store all the data that we're | |
850 | * going to read */ | |
672f0f2c AG |
851 | start_buffer = qemu_try_blockalign(bs, buffer_size); |
852 | if (start_buffer == NULL) { | |
853 | return -ENOMEM; | |
854 | } | |
855 | /* The part of the buffer where the end region is located */ | |
856 | end_buffer = start_buffer + buffer_size - end->nb_bytes; | |
857 | ||
5396234b VSO |
858 | qemu_iovec_init(&qiov, 2 + (m->data_qiov ? |
859 | qemu_iovec_subvec_niov(m->data_qiov, | |
860 | m->data_qiov_offset, | |
861 | data_bytes) | |
862 | : 0)); | |
86b862c4 | 863 | |
593fb83c | 864 | qemu_co_mutex_unlock(&s->lock); |
b3cf1c7c AG |
865 | /* First we read the existing data from both COW regions. We |
866 | * either read the whole region in one go, or the start and end | |
867 | * regions separately. */ | |
868 | if (merge_reads) { | |
86b862c4 AG |
869 | qemu_iovec_add(&qiov, start_buffer, buffer_size); |
870 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c | 871 | } else { |
86b862c4 AG |
872 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); |
873 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c AG |
874 | if (ret < 0) { |
875 | goto fail; | |
876 | } | |
672f0f2c | 877 | |
86b862c4 AG |
878 | qemu_iovec_reset(&qiov); |
879 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
880 | ret = do_perform_cow_read(bs, m->offset, end->offset, &qiov); | |
b3cf1c7c | 881 | } |
593fb83c | 882 | if (ret < 0) { |
99450c6f | 883 | goto fail; |
593fb83c KW |
884 | } |
885 | ||
672f0f2c AG |
886 | /* Encrypt the data if necessary before writing it */ |
887 | if (bs->encrypted) { | |
603fbd07 ML |
888 | ret = qcow2_co_encrypt(bs, |
889 | m->alloc_offset + start->offset, | |
890 | m->offset + start->offset, | |
891 | start_buffer, start->nb_bytes); | |
892 | if (ret < 0) { | |
893 | goto fail; | |
894 | } | |
895 | ||
896 | ret = qcow2_co_encrypt(bs, | |
897 | m->alloc_offset + end->offset, | |
898 | m->offset + end->offset, | |
899 | end_buffer, end->nb_bytes); | |
900 | if (ret < 0) { | |
672f0f2c AG |
901 | goto fail; |
902 | } | |
903 | } | |
904 | ||
ee22a9d8 AG |
905 | /* And now we can write everything. If we have the guest data we |
906 | * can write everything in one single operation */ | |
907 | if (m->data_qiov) { | |
908 | qemu_iovec_reset(&qiov); | |
909 | if (start->nb_bytes) { | |
910 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
911 | } | |
5396234b | 912 | qemu_iovec_concat(&qiov, m->data_qiov, m->data_qiov_offset, data_bytes); |
ee22a9d8 AG |
913 | if (end->nb_bytes) { |
914 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
915 | } | |
916 | /* NOTE: we have a write_aio blkdebug event here followed by | |
917 | * a cow_write one in do_perform_cow_write(), but there's only | |
918 | * one single I/O operation */ | |
919 | BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); | |
920 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); | |
921 | } else { | |
922 | /* If there's no guest data then write both COW regions separately */ | |
923 | qemu_iovec_reset(&qiov); | |
924 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
925 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); | |
926 | if (ret < 0) { | |
927 | goto fail; | |
928 | } | |
929 | ||
930 | qemu_iovec_reset(&qiov); | |
931 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
932 | ret = do_perform_cow_write(bs, m->alloc_offset, end->offset, &qiov); | |
672f0f2c | 933 | } |
99450c6f AG |
934 | |
935 | fail: | |
936 | qemu_co_mutex_lock(&s->lock); | |
937 | ||
593fb83c KW |
938 | /* |
939 | * Before we update the L2 table to actually point to the new cluster, we | |
940 | * need to be sure that the refcounts have been increased and COW was | |
941 | * handled. | |
942 | */ | |
99450c6f AG |
943 | if (ret == 0) { |
944 | qcow2_cache_depends_on_flush(s->l2_table_cache); | |
945 | } | |
593fb83c | 946 | |
672f0f2c | 947 | qemu_vfree(start_buffer); |
86b862c4 | 948 | qemu_iovec_destroy(&qiov); |
99450c6f | 949 | return ret; |
593fb83c KW |
950 | } |
951 | ||
148da7ea | 952 | int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) |
45aba42f | 953 | { |
ff99129a | 954 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 955 | int i, j = 0, l2_index, ret; |
a002c0b0 | 956 | uint64_t *old_cluster, *l2_slice; |
250196f1 | 957 | uint64_t cluster_offset = m->alloc_offset; |
45aba42f | 958 | |
3cce16f4 | 959 | trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
f50f88b9 | 960 | assert(m->nb_clusters > 0); |
45aba42f | 961 | |
5839e53b | 962 | old_cluster = g_try_new(uint64_t, m->nb_clusters); |
de82815d KW |
963 | if (old_cluster == NULL) { |
964 | ret = -ENOMEM; | |
965 | goto err; | |
966 | } | |
45aba42f KW |
967 | |
968 | /* copy content of unmodified sectors */ | |
99450c6f | 969 | ret = perform_cow(bs, m); |
593fb83c KW |
970 | if (ret < 0) { |
971 | goto err; | |
29c1a730 KW |
972 | } |
973 | ||
593fb83c | 974 | /* Update L2 table. */ |
74c4510a | 975 | if (s->use_lazy_refcounts) { |
280d3735 KW |
976 | qcow2_mark_dirty(bs); |
977 | } | |
bfe8043e SH |
978 | if (qcow2_need_accurate_refcounts(s)) { |
979 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
980 | s->refcount_block_cache); | |
981 | } | |
280d3735 | 982 | |
a002c0b0 | 983 | ret = get_cluster_table(bs, m->offset, &l2_slice, &l2_index); |
1e3e8f1a | 984 | if (ret < 0) { |
45aba42f | 985 | goto err; |
1e3e8f1a | 986 | } |
a002c0b0 | 987 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
45aba42f | 988 | |
a002c0b0 | 989 | assert(l2_index + m->nb_clusters <= s->l2_slice_size); |
45aba42f | 990 | for (i = 0; i < m->nb_clusters; i++) { |
348fcc4f | 991 | uint64_t offset = cluster_offset + ((uint64_t)i << s->cluster_bits); |
45aba42f | 992 | /* if two concurrent writes happen to the same unallocated cluster |
aaa4d20b KW |
993 | * each write allocates separate cluster and writes data concurrently. |
994 | * The first one to complete updates l2 table with pointer to its | |
995 | * cluster the second one has to do RMW (which is done above by | |
996 | * perform_cow()), update l2 table with its cluster pointer and free | |
997 | * old cluster. This is what this loop does */ | |
12c6aebe AG |
998 | if (get_l2_entry(s, l2_slice, l2_index + i) != 0) { |
999 | old_cluster[j++] = get_l2_entry(s, l2_slice, l2_index + i); | |
aaa4d20b | 1000 | } |
45aba42f | 1001 | |
3a75a870 AG |
1002 | /* The offset must fit in the offset field of the L2 table entry */ |
1003 | assert((offset & L2E_OFFSET_MASK) == offset); | |
1004 | ||
12c6aebe | 1005 | set_l2_entry(s, l2_slice, l2_index + i, offset | QCOW_OFLAG_COPIED); |
45aba42f KW |
1006 | } |
1007 | ||
9f8e668e | 1008 | |
a002c0b0 | 1009 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
45aba42f | 1010 | |
7ec5e6a4 KW |
1011 | /* |
1012 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
6cfcb9b8 KW |
1013 | * |
1014 | * Don't discard clusters that reach a refcount of 0 (e.g. compressed | |
1015 | * clusters), the next write will reuse them anyway. | |
7ec5e6a4 | 1016 | */ |
564a6b69 | 1017 | if (!m->keep_old_clusters && j != 0) { |
7ec5e6a4 | 1018 | for (i = 0; i < j; i++) { |
12c6aebe | 1019 | qcow2_free_any_clusters(bs, old_cluster[i], 1, QCOW2_DISCARD_NEVER); |
7ec5e6a4 KW |
1020 | } |
1021 | } | |
45aba42f KW |
1022 | |
1023 | ret = 0; | |
1024 | err: | |
7267c094 | 1025 | g_free(old_cluster); |
45aba42f KW |
1026 | return ret; |
1027 | } | |
1028 | ||
8b24cd14 KW |
1029 | /** |
1030 | * Frees the allocated clusters because the request failed and they won't | |
1031 | * actually be linked. | |
1032 | */ | |
1033 | void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m) | |
1034 | { | |
1035 | BDRVQcow2State *s = bs->opaque; | |
3ede935f | 1036 | if (!has_data_file(bs) && !m->keep_old_clusters) { |
c3b6658c KW |
1037 | qcow2_free_clusters(bs, m->alloc_offset, |
1038 | m->nb_clusters << s->cluster_bits, | |
1039 | QCOW2_DISCARD_NEVER); | |
1040 | } | |
8b24cd14 KW |
1041 | } |
1042 | ||
8f91d690 AG |
1043 | /* |
1044 | * For a given write request, create a new QCowL2Meta structure, add | |
57538c86 AG |
1045 | * it to @m and the BDRVQcow2State.cluster_allocs list. If the write |
1046 | * request does not need copy-on-write or changes to the L2 metadata | |
1047 | * then this function does nothing. | |
8f91d690 AG |
1048 | * |
1049 | * @host_cluster_offset points to the beginning of the first cluster. | |
1050 | * | |
1051 | * @guest_offset and @bytes indicate the offset and length of the | |
1052 | * request. | |
1053 | * | |
57538c86 AG |
1054 | * @l2_slice contains the L2 entries of all clusters involved in this |
1055 | * write request. | |
1056 | * | |
8f91d690 AG |
1057 | * If @keep_old is true it means that the clusters were already |
1058 | * allocated and will be overwritten. If false then the clusters are | |
1059 | * new and we have to decrease the reference count of the old ones. | |
1060 | */ | |
1061 | static void calculate_l2_meta(BlockDriverState *bs, | |
1062 | uint64_t host_cluster_offset, | |
1063 | uint64_t guest_offset, unsigned bytes, | |
57538c86 | 1064 | uint64_t *l2_slice, QCowL2Meta **m, bool keep_old) |
8f91d690 AG |
1065 | { |
1066 | BDRVQcow2State *s = bs->opaque; | |
57538c86 AG |
1067 | int l2_index = offset_to_l2_slice_index(s, guest_offset); |
1068 | uint64_t l2_entry; | |
1069 | unsigned cow_start_from, cow_end_to; | |
8f91d690 AG |
1070 | unsigned cow_start_to = offset_into_cluster(s, guest_offset); |
1071 | unsigned cow_end_from = cow_start_to + bytes; | |
8f91d690 AG |
1072 | unsigned nb_clusters = size_to_clusters(s, cow_end_from); |
1073 | QCowL2Meta *old_m = *m; | |
57538c86 AG |
1074 | QCow2ClusterType type; |
1075 | ||
1076 | assert(nb_clusters <= s->l2_slice_size - l2_index); | |
1077 | ||
1078 | /* Return if there's no COW (all clusters are normal and we keep them) */ | |
1079 | if (keep_old) { | |
1080 | int i; | |
1081 | for (i = 0; i < nb_clusters; i++) { | |
12c6aebe | 1082 | l2_entry = get_l2_entry(s, l2_slice, l2_index + i); |
57538c86 AG |
1083 | if (qcow2_get_cluster_type(bs, l2_entry) != QCOW2_CLUSTER_NORMAL) { |
1084 | break; | |
1085 | } | |
1086 | } | |
1087 | if (i == nb_clusters) { | |
1088 | return; | |
1089 | } | |
1090 | } | |
1091 | ||
1092 | /* Get the L2 entry of the first cluster */ | |
12c6aebe | 1093 | l2_entry = get_l2_entry(s, l2_slice, l2_index); |
57538c86 AG |
1094 | type = qcow2_get_cluster_type(bs, l2_entry); |
1095 | ||
1096 | if (type == QCOW2_CLUSTER_NORMAL && keep_old) { | |
1097 | cow_start_from = cow_start_to; | |
1098 | } else { | |
1099 | cow_start_from = 0; | |
1100 | } | |
1101 | ||
1102 | /* Get the L2 entry of the last cluster */ | |
12c6aebe | 1103 | l2_entry = get_l2_entry(s, l2_slice, l2_index + nb_clusters - 1); |
57538c86 AG |
1104 | type = qcow2_get_cluster_type(bs, l2_entry); |
1105 | ||
1106 | if (type == QCOW2_CLUSTER_NORMAL && keep_old) { | |
1107 | cow_end_to = cow_end_from; | |
1108 | } else { | |
1109 | cow_end_to = ROUND_UP(cow_end_from, s->cluster_size); | |
1110 | } | |
8f91d690 AG |
1111 | |
1112 | *m = g_malloc0(sizeof(**m)); | |
1113 | **m = (QCowL2Meta) { | |
1114 | .next = old_m, | |
1115 | ||
1116 | .alloc_offset = host_cluster_offset, | |
1117 | .offset = start_of_cluster(s, guest_offset), | |
1118 | .nb_clusters = nb_clusters, | |
1119 | ||
1120 | .keep_old_clusters = keep_old, | |
1121 | ||
1122 | .cow_start = { | |
1123 | .offset = cow_start_from, | |
1124 | .nb_bytes = cow_start_to - cow_start_from, | |
1125 | }, | |
1126 | .cow_end = { | |
1127 | .offset = cow_end_from, | |
1128 | .nb_bytes = cow_end_to - cow_end_from, | |
1129 | }, | |
1130 | }; | |
1131 | ||
1132 | qemu_co_queue_init(&(*m)->dependent_requests); | |
1133 | QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); | |
1134 | } | |
1135 | ||
57538c86 AG |
1136 | /* |
1137 | * Returns true if writing to the cluster pointed to by @l2_entry | |
1138 | * requires a new allocation (that is, if the cluster is unallocated | |
1139 | * or has refcount > 1 and therefore cannot be written in-place). | |
1140 | */ | |
1141 | static bool cluster_needs_new_alloc(BlockDriverState *bs, uint64_t l2_entry) | |
c1587d87 AG |
1142 | { |
1143 | switch (qcow2_get_cluster_type(bs, l2_entry)) { | |
1144 | case QCOW2_CLUSTER_NORMAL: | |
57538c86 | 1145 | case QCOW2_CLUSTER_ZERO_ALLOC: |
c1587d87 AG |
1146 | if (l2_entry & QCOW_OFLAG_COPIED) { |
1147 | return false; | |
1148 | } | |
1149 | case QCOW2_CLUSTER_UNALLOCATED: | |
1150 | case QCOW2_CLUSTER_COMPRESSED: | |
1151 | case QCOW2_CLUSTER_ZERO_PLAIN: | |
c1587d87 AG |
1152 | return true; |
1153 | default: | |
1154 | abort(); | |
1155 | } | |
1156 | } | |
1157 | ||
bf319ece | 1158 | /* |
57538c86 AG |
1159 | * Returns the number of contiguous clusters that can be written to |
1160 | * using one single write request, starting from @l2_index. | |
1161 | * At most @nb_clusters are checked. | |
1162 | * | |
1163 | * If @new_alloc is true this counts clusters that are either | |
1164 | * unallocated, or allocated but with refcount > 1 (so they need to be | |
1165 | * newly allocated and COWed). | |
1166 | * | |
1167 | * If @new_alloc is false this counts clusters that are already | |
1168 | * allocated and can be overwritten in-place (this includes clusters | |
1169 | * of type QCOW2_CLUSTER_ZERO_ALLOC). | |
bf319ece | 1170 | */ |
57538c86 AG |
1171 | static int count_single_write_clusters(BlockDriverState *bs, int nb_clusters, |
1172 | uint64_t *l2_slice, int l2_index, | |
1173 | bool new_alloc) | |
bf319ece | 1174 | { |
57538c86 | 1175 | BDRVQcow2State *s = bs->opaque; |
12c6aebe | 1176 | uint64_t l2_entry = get_l2_entry(s, l2_slice, l2_index); |
57538c86 | 1177 | uint64_t expected_offset = l2_entry & L2E_OFFSET_MASK; |
143550a8 | 1178 | int i; |
bf319ece | 1179 | |
143550a8 | 1180 | for (i = 0; i < nb_clusters; i++) { |
12c6aebe | 1181 | l2_entry = get_l2_entry(s, l2_slice, l2_index + i); |
57538c86 | 1182 | if (cluster_needs_new_alloc(bs, l2_entry) != new_alloc) { |
bf319ece | 1183 | break; |
143550a8 | 1184 | } |
57538c86 AG |
1185 | if (!new_alloc) { |
1186 | if (expected_offset != (l2_entry & L2E_OFFSET_MASK)) { | |
1187 | break; | |
1188 | } | |
1189 | expected_offset += s->cluster_size; | |
1190 | } | |
bf319ece KW |
1191 | } |
1192 | ||
1193 | assert(i <= nb_clusters); | |
1194 | return i; | |
1195 | } | |
1196 | ||
250196f1 | 1197 | /* |
226c3c26 KW |
1198 | * Check if there already is an AIO write request in flight which allocates |
1199 | * the same cluster. In this case we need to wait until the previous | |
1200 | * request has completed and updated the L2 table accordingly. | |
65eb2e35 KW |
1201 | * |
1202 | * Returns: | |
1203 | * 0 if there was no dependency. *cur_bytes indicates the number of | |
1204 | * bytes from guest_offset that can be read before the next | |
1205 | * dependency must be processed (or the request is complete) | |
1206 | * | |
1207 | * -EAGAIN if we had to wait for another request, previously gathered | |
1208 | * information on cluster allocation may be invalid now. The caller | |
1209 | * must start over anyway, so consider *cur_bytes undefined. | |
250196f1 | 1210 | */ |
226c3c26 | 1211 | static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, |
ecdd5333 | 1212 | uint64_t *cur_bytes, QCowL2Meta **m) |
250196f1 | 1213 | { |
ff99129a | 1214 | BDRVQcow2State *s = bs->opaque; |
250196f1 | 1215 | QCowL2Meta *old_alloc; |
65eb2e35 | 1216 | uint64_t bytes = *cur_bytes; |
250196f1 | 1217 | |
250196f1 KW |
1218 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
1219 | ||
65eb2e35 KW |
1220 | uint64_t start = guest_offset; |
1221 | uint64_t end = start + bytes; | |
1222 | uint64_t old_start = l2meta_cow_start(old_alloc); | |
1223 | uint64_t old_end = l2meta_cow_end(old_alloc); | |
250196f1 | 1224 | |
d9d74f41 | 1225 | if (end <= old_start || start >= old_end) { |
250196f1 KW |
1226 | /* No intersection */ |
1227 | } else { | |
1228 | if (start < old_start) { | |
1229 | /* Stop at the start of a running allocation */ | |
65eb2e35 | 1230 | bytes = old_start - start; |
250196f1 | 1231 | } else { |
65eb2e35 | 1232 | bytes = 0; |
250196f1 KW |
1233 | } |
1234 | ||
ecdd5333 KW |
1235 | /* Stop if already an l2meta exists. After yielding, it wouldn't |
1236 | * be valid any more, so we'd have to clean up the old L2Metas | |
1237 | * and deal with requests depending on them before starting to | |
1238 | * gather new ones. Not worth the trouble. */ | |
1239 | if (bytes == 0 && *m) { | |
1240 | *cur_bytes = 0; | |
1241 | return 0; | |
1242 | } | |
1243 | ||
65eb2e35 | 1244 | if (bytes == 0) { |
250196f1 KW |
1245 | /* Wait for the dependency to complete. We need to recheck |
1246 | * the free/allocated clusters when we continue. */ | |
1ace7cea | 1247 | qemu_co_queue_wait(&old_alloc->dependent_requests, &s->lock); |
250196f1 KW |
1248 | return -EAGAIN; |
1249 | } | |
1250 | } | |
1251 | } | |
1252 | ||
65eb2e35 KW |
1253 | /* Make sure that existing clusters and new allocations are only used up to |
1254 | * the next dependency if we shortened the request above */ | |
1255 | *cur_bytes = bytes; | |
250196f1 | 1256 | |
226c3c26 KW |
1257 | return 0; |
1258 | } | |
1259 | ||
0af729ec | 1260 | /* |
57538c86 AG |
1261 | * Checks how many already allocated clusters that don't require a new |
1262 | * allocation there are at the given guest_offset (up to *bytes). | |
1263 | * If *host_offset is not INV_OFFSET, only physically contiguous clusters | |
1264 | * beginning at this host offset are counted. | |
0af729ec | 1265 | * |
411d62b0 KW |
1266 | * Note that guest_offset may not be cluster aligned. In this case, the |
1267 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1268 | * therefore isn't cluster aligned as well. | |
0af729ec KW |
1269 | * |
1270 | * Returns: | |
1271 | * 0: if no allocated clusters are available at the given offset. | |
1272 | * *bytes is normally unchanged. It is set to 0 if the cluster | |
57538c86 AG |
1273 | * is allocated and can be overwritten in-place but doesn't have |
1274 | * the right physical offset. | |
0af729ec | 1275 | * |
57538c86 AG |
1276 | * 1: if allocated clusters that can be overwritten in place are |
1277 | * available at the requested offset. *bytes may have decreased | |
1278 | * and describes the length of the area that can be written to. | |
0af729ec KW |
1279 | * |
1280 | * -errno: in error cases | |
0af729ec KW |
1281 | */ |
1282 | static int handle_copied(BlockDriverState *bs, uint64_t guest_offset, | |
c53ede9f | 1283 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
0af729ec | 1284 | { |
ff99129a | 1285 | BDRVQcow2State *s = bs->opaque; |
0af729ec | 1286 | int l2_index; |
57538c86 | 1287 | uint64_t l2_entry, cluster_offset; |
cde91766 | 1288 | uint64_t *l2_slice; |
b6d36def | 1289 | uint64_t nb_clusters; |
c53ede9f | 1290 | unsigned int keep_clusters; |
a3f1afb4 | 1291 | int ret; |
0af729ec KW |
1292 | |
1293 | trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset, | |
1294 | *bytes); | |
0af729ec | 1295 | |
c6d619cc KW |
1296 | assert(*host_offset == INV_OFFSET || offset_into_cluster(s, guest_offset) |
1297 | == offset_into_cluster(s, *host_offset)); | |
411d62b0 | 1298 | |
acb0467f | 1299 | /* |
cde91766 | 1300 | * Calculate the number of clusters to look for. We stop at L2 slice |
acb0467f KW |
1301 | * boundaries to keep things simple. |
1302 | */ | |
1303 | nb_clusters = | |
1304 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1305 | ||
cde91766 AG |
1306 | l2_index = offset_to_l2_slice_index(s, guest_offset); |
1307 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
57538c86 AG |
1308 | /* Limit total byte count to BDRV_REQUEST_MAX_BYTES */ |
1309 | nb_clusters = MIN(nb_clusters, BDRV_REQUEST_MAX_BYTES >> s->cluster_bits); | |
acb0467f | 1310 | |
0af729ec | 1311 | /* Find L2 entry for the first involved cluster */ |
cde91766 | 1312 | ret = get_cluster_table(bs, guest_offset, &l2_slice, &l2_index); |
0af729ec KW |
1313 | if (ret < 0) { |
1314 | return ret; | |
1315 | } | |
1316 | ||
12c6aebe | 1317 | l2_entry = get_l2_entry(s, l2_slice, l2_index); |
57538c86 AG |
1318 | cluster_offset = l2_entry & L2E_OFFSET_MASK; |
1319 | ||
1320 | if (!cluster_needs_new_alloc(bs, l2_entry)) { | |
1321 | if (offset_into_cluster(s, cluster_offset)) { | |
1322 | qcow2_signal_corruption(bs, true, -1, -1, "%s cluster offset " | |
1323 | "%#" PRIx64 " unaligned (guest offset: %#" | |
1324 | PRIx64 ")", l2_entry & QCOW_OFLAG_ZERO ? | |
1325 | "Preallocated zero" : "Data", | |
1326 | cluster_offset, guest_offset); | |
a97c67ee HR |
1327 | ret = -EIO; |
1328 | goto out; | |
1329 | } | |
1330 | ||
57538c86 AG |
1331 | /* If a specific host_offset is required, check it */ |
1332 | if (*host_offset != INV_OFFSET && cluster_offset != *host_offset) { | |
e62daaf6 KW |
1333 | *bytes = 0; |
1334 | ret = 0; | |
1335 | goto out; | |
1336 | } | |
1337 | ||
0af729ec | 1338 | /* We keep all QCOW_OFLAG_COPIED clusters */ |
57538c86 AG |
1339 | keep_clusters = count_single_write_clusters(bs, nb_clusters, l2_slice, |
1340 | l2_index, false); | |
c53ede9f KW |
1341 | assert(keep_clusters <= nb_clusters); |
1342 | ||
1343 | *bytes = MIN(*bytes, | |
1344 | keep_clusters * s->cluster_size | |
1345 | - offset_into_cluster(s, guest_offset)); | |
57538c86 AG |
1346 | assert(*bytes != 0); |
1347 | ||
1348 | calculate_l2_meta(bs, cluster_offset, guest_offset, | |
1349 | *bytes, l2_slice, m, true); | |
0af729ec KW |
1350 | |
1351 | ret = 1; | |
1352 | } else { | |
0af729ec KW |
1353 | ret = 0; |
1354 | } | |
1355 | ||
0af729ec | 1356 | /* Cleanup */ |
e62daaf6 | 1357 | out: |
cde91766 | 1358 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
0af729ec | 1359 | |
e62daaf6 KW |
1360 | /* Only return a host offset if we actually made progress. Otherwise we |
1361 | * would make requirements for handle_alloc() that it can't fulfill */ | |
a97c67ee | 1362 | if (ret > 0) { |
57538c86 | 1363 | *host_offset = cluster_offset + offset_into_cluster(s, guest_offset); |
e62daaf6 KW |
1364 | } |
1365 | ||
0af729ec KW |
1366 | return ret; |
1367 | } | |
1368 | ||
226c3c26 KW |
1369 | /* |
1370 | * Allocates new clusters for the given guest_offset. | |
1371 | * | |
1372 | * At most *nb_clusters are allocated, and on return *nb_clusters is updated to | |
1373 | * contain the number of clusters that have been allocated and are contiguous | |
1374 | * in the image file. | |
1375 | * | |
c6d619cc KW |
1376 | * If *host_offset is not INV_OFFSET, it specifies the offset in the image file |
1377 | * at which the new clusters must start. *nb_clusters can be 0 on return in | |
1378 | * this case if the cluster at host_offset is already in use. If *host_offset | |
1379 | * is INV_OFFSET, the clusters can be allocated anywhere in the image file. | |
226c3c26 KW |
1380 | * |
1381 | * *host_offset is updated to contain the offset into the image file at which | |
1382 | * the first allocated cluster starts. | |
1383 | * | |
1384 | * Return 0 on success and -errno in error cases. -EAGAIN means that the | |
1385 | * function has been waiting for another request and the allocation must be | |
1386 | * restarted, but the whole request should not be failed. | |
1387 | */ | |
1388 | static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, | |
b6d36def | 1389 | uint64_t *host_offset, uint64_t *nb_clusters) |
226c3c26 | 1390 | { |
ff99129a | 1391 | BDRVQcow2State *s = bs->opaque; |
226c3c26 KW |
1392 | |
1393 | trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, | |
1394 | *host_offset, *nb_clusters); | |
1395 | ||
966b000f KW |
1396 | if (has_data_file(bs)) { |
1397 | assert(*host_offset == INV_OFFSET || | |
1398 | *host_offset == start_of_cluster(s, guest_offset)); | |
1399 | *host_offset = start_of_cluster(s, guest_offset); | |
1400 | return 0; | |
1401 | } | |
1402 | ||
250196f1 KW |
1403 | /* Allocate new clusters */ |
1404 | trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); | |
c6d619cc | 1405 | if (*host_offset == INV_OFFSET) { |
df021791 KW |
1406 | int64_t cluster_offset = |
1407 | qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); | |
1408 | if (cluster_offset < 0) { | |
1409 | return cluster_offset; | |
1410 | } | |
1411 | *host_offset = cluster_offset; | |
1412 | return 0; | |
250196f1 | 1413 | } else { |
b6d36def | 1414 | int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); |
df021791 KW |
1415 | if (ret < 0) { |
1416 | return ret; | |
1417 | } | |
1418 | *nb_clusters = ret; | |
1419 | return 0; | |
250196f1 | 1420 | } |
250196f1 KW |
1421 | } |
1422 | ||
10f0ed8b | 1423 | /* |
57538c86 AG |
1424 | * Allocates new clusters for an area that is either still unallocated or |
1425 | * cannot be overwritten in-place. If *host_offset is not INV_OFFSET, | |
1426 | * clusters are only allocated if the new allocation can match the specified | |
1427 | * host offset. | |
10f0ed8b | 1428 | * |
411d62b0 KW |
1429 | * Note that guest_offset may not be cluster aligned. In this case, the |
1430 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1431 | * therefore isn't cluster aligned as well. | |
10f0ed8b KW |
1432 | * |
1433 | * Returns: | |
1434 | * 0: if no clusters could be allocated. *bytes is set to 0, | |
1435 | * *host_offset is left unchanged. | |
1436 | * | |
1437 | * 1: if new clusters were allocated. *bytes may be decreased if the | |
1438 | * new allocation doesn't cover all of the requested area. | |
1439 | * *host_offset is updated to contain the host offset of the first | |
1440 | * newly allocated cluster. | |
1441 | * | |
1442 | * -errno: in error cases | |
10f0ed8b KW |
1443 | */ |
1444 | static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, | |
c37f4cd7 | 1445 | uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) |
10f0ed8b | 1446 | { |
ff99129a | 1447 | BDRVQcow2State *s = bs->opaque; |
10f0ed8b | 1448 | int l2_index; |
6d99a344 | 1449 | uint64_t *l2_slice; |
b6d36def | 1450 | uint64_t nb_clusters; |
10f0ed8b KW |
1451 | int ret; |
1452 | ||
57538c86 | 1453 | uint64_t alloc_cluster_offset; |
10f0ed8b KW |
1454 | |
1455 | trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset, | |
1456 | *bytes); | |
1457 | assert(*bytes > 0); | |
1458 | ||
f5bc6350 | 1459 | /* |
6d99a344 | 1460 | * Calculate the number of clusters to look for. We stop at L2 slice |
f5bc6350 KW |
1461 | * boundaries to keep things simple. |
1462 | */ | |
c37f4cd7 KW |
1463 | nb_clusters = |
1464 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1465 | ||
6d99a344 AG |
1466 | l2_index = offset_to_l2_slice_index(s, guest_offset); |
1467 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
57538c86 AG |
1468 | /* Limit total allocation byte count to BDRV_REQUEST_MAX_BYTES */ |
1469 | nb_clusters = MIN(nb_clusters, BDRV_REQUEST_MAX_BYTES >> s->cluster_bits); | |
d1b9d19f | 1470 | |
10f0ed8b | 1471 | /* Find L2 entry for the first involved cluster */ |
6d99a344 | 1472 | ret = get_cluster_table(bs, guest_offset, &l2_slice, &l2_index); |
10f0ed8b KW |
1473 | if (ret < 0) { |
1474 | return ret; | |
1475 | } | |
1476 | ||
57538c86 AG |
1477 | nb_clusters = count_single_write_clusters(bs, nb_clusters, |
1478 | l2_slice, l2_index, true); | |
10f0ed8b | 1479 | |
ecdd5333 KW |
1480 | /* This function is only called when there were no non-COW clusters, so if |
1481 | * we can't find any unallocated or COW clusters either, something is | |
1482 | * wrong with our code. */ | |
1483 | assert(nb_clusters > 0); | |
1484 | ||
57538c86 AG |
1485 | /* Allocate at a given offset in the image file */ |
1486 | alloc_cluster_offset = *host_offset == INV_OFFSET ? INV_OFFSET : | |
1487 | start_of_cluster(s, *host_offset); | |
1488 | ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset, | |
1489 | &nb_clusters); | |
1490 | if (ret < 0) { | |
1491 | goto out; | |
10f0ed8b KW |
1492 | } |
1493 | ||
57538c86 AG |
1494 | /* Can't extend contiguous allocation */ |
1495 | if (nb_clusters == 0) { | |
1496 | *bytes = 0; | |
1497 | ret = 0; | |
1498 | goto out; | |
ff52aab2 HR |
1499 | } |
1500 | ||
57538c86 AG |
1501 | assert(alloc_cluster_offset != INV_OFFSET); |
1502 | ||
83baa9a4 KW |
1503 | /* |
1504 | * Save info needed for meta data update. | |
1505 | * | |
85567393 | 1506 | * requested_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1507 | * newly allocated cluster to the end of the (possibly shortened |
1508 | * before) write request. | |
1509 | * | |
85567393 | 1510 | * avail_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1511 | * newly allocated to the end of the last newly allocated cluster. |
1512 | * | |
85567393 | 1513 | * nb_bytes: The number of bytes from the start of the first |
83baa9a4 KW |
1514 | * newly allocated cluster to the end of the area that the write |
1515 | * request actually writes to (excluding COW at the end) | |
1516 | */ | |
85567393 | 1517 | uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset); |
d1b9d19f | 1518 | int avail_bytes = nb_clusters << s->cluster_bits; |
85567393 | 1519 | int nb_bytes = MIN(requested_bytes, avail_bytes); |
83baa9a4 | 1520 | |
411d62b0 | 1521 | *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset); |
85567393 | 1522 | *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset)); |
83baa9a4 KW |
1523 | assert(*bytes != 0); |
1524 | ||
57538c86 AG |
1525 | calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes, l2_slice, |
1526 | m, false); | |
8f91d690 | 1527 | |
57538c86 | 1528 | ret = 1; |
10f0ed8b | 1529 | |
57538c86 AG |
1530 | out: |
1531 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
1532 | if (ret < 0 && *m && (*m)->nb_clusters > 0) { | |
10f0ed8b KW |
1533 | QLIST_REMOVE(*m, next_in_flight); |
1534 | } | |
1535 | return ret; | |
1536 | } | |
1537 | ||
45aba42f KW |
1538 | /* |
1539 | * alloc_cluster_offset | |
1540 | * | |
250196f1 KW |
1541 | * For a given offset on the virtual disk, find the cluster offset in qcow2 |
1542 | * file. If the offset is not found, allocate a new cluster. | |
45aba42f | 1543 | * |
250196f1 | 1544 | * If the cluster was already allocated, m->nb_clusters is set to 0 and |
a7912369 | 1545 | * other fields in m are meaningless. |
148da7ea KW |
1546 | * |
1547 | * If the cluster is newly allocated, m->nb_clusters is set to the number of | |
68d100e9 KW |
1548 | * contiguous clusters that have been allocated. In this case, the other |
1549 | * fields of m are valid and contain information about the first allocated | |
1550 | * cluster. | |
45aba42f | 1551 | * |
68d100e9 KW |
1552 | * If the request conflicts with another write request in flight, the coroutine |
1553 | * is queued and will be reentered when the dependency has completed. | |
148da7ea KW |
1554 | * |
1555 | * Return 0 on success and -errno in error cases | |
45aba42f | 1556 | */ |
f4f0d391 | 1557 | int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, |
d46a0bb2 KW |
1558 | unsigned int *bytes, uint64_t *host_offset, |
1559 | QCowL2Meta **m) | |
45aba42f | 1560 | { |
ff99129a | 1561 | BDRVQcow2State *s = bs->opaque; |
710c2496 | 1562 | uint64_t start, remaining; |
250196f1 | 1563 | uint64_t cluster_offset; |
65eb2e35 | 1564 | uint64_t cur_bytes; |
710c2496 | 1565 | int ret; |
45aba42f | 1566 | |
d46a0bb2 | 1567 | trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *bytes); |
710c2496 | 1568 | |
72424114 | 1569 | again: |
16f0587e | 1570 | start = offset; |
d46a0bb2 | 1571 | remaining = *bytes; |
c6d619cc KW |
1572 | cluster_offset = INV_OFFSET; |
1573 | *host_offset = INV_OFFSET; | |
ecdd5333 KW |
1574 | cur_bytes = 0; |
1575 | *m = NULL; | |
0af729ec | 1576 | |
2c3b32d2 | 1577 | while (true) { |
ecdd5333 | 1578 | |
c6d619cc | 1579 | if (*host_offset == INV_OFFSET && cluster_offset != INV_OFFSET) { |
ecdd5333 KW |
1580 | *host_offset = start_of_cluster(s, cluster_offset); |
1581 | } | |
1582 | ||
1583 | assert(remaining >= cur_bytes); | |
1584 | ||
1585 | start += cur_bytes; | |
1586 | remaining -= cur_bytes; | |
c6d619cc KW |
1587 | |
1588 | if (cluster_offset != INV_OFFSET) { | |
1589 | cluster_offset += cur_bytes; | |
1590 | } | |
ecdd5333 KW |
1591 | |
1592 | if (remaining == 0) { | |
1593 | break; | |
1594 | } | |
1595 | ||
1596 | cur_bytes = remaining; | |
1597 | ||
2c3b32d2 KW |
1598 | /* |
1599 | * Now start gathering as many contiguous clusters as possible: | |
1600 | * | |
1601 | * 1. Check for overlaps with in-flight allocations | |
1602 | * | |
1603 | * a) Overlap not in the first cluster -> shorten this request and | |
1604 | * let the caller handle the rest in its next loop iteration. | |
1605 | * | |
1606 | * b) Real overlaps of two requests. Yield and restart the search | |
1607 | * for contiguous clusters (the situation could have changed | |
1608 | * while we were sleeping) | |
1609 | * | |
1610 | * c) TODO: Request starts in the same cluster as the in-flight | |
1611 | * allocation ends. Shorten the COW of the in-fight allocation, | |
1612 | * set cluster_offset to write to the same cluster and set up | |
1613 | * the right synchronisation between the in-flight request and | |
1614 | * the new one. | |
1615 | */ | |
ecdd5333 | 1616 | ret = handle_dependencies(bs, start, &cur_bytes, m); |
2c3b32d2 | 1617 | if (ret == -EAGAIN) { |
ecdd5333 KW |
1618 | /* Currently handle_dependencies() doesn't yield if we already had |
1619 | * an allocation. If it did, we would have to clean up the L2Meta | |
1620 | * structs before starting over. */ | |
1621 | assert(*m == NULL); | |
2c3b32d2 KW |
1622 | goto again; |
1623 | } else if (ret < 0) { | |
1624 | return ret; | |
ecdd5333 KW |
1625 | } else if (cur_bytes == 0) { |
1626 | break; | |
2c3b32d2 KW |
1627 | } else { |
1628 | /* handle_dependencies() may have decreased cur_bytes (shortened | |
1629 | * the allocations below) so that the next dependency is processed | |
1630 | * correctly during the next loop iteration. */ | |
0af729ec | 1631 | } |
710c2496 | 1632 | |
2c3b32d2 KW |
1633 | /* |
1634 | * 2. Count contiguous COPIED clusters. | |
1635 | */ | |
1636 | ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m); | |
1637 | if (ret < 0) { | |
1638 | return ret; | |
1639 | } else if (ret) { | |
ecdd5333 | 1640 | continue; |
2c3b32d2 KW |
1641 | } else if (cur_bytes == 0) { |
1642 | break; | |
1643 | } | |
060bee89 | 1644 | |
2c3b32d2 KW |
1645 | /* |
1646 | * 3. If the request still hasn't completed, allocate new clusters, | |
1647 | * considering any cluster_offset of steps 1c or 2. | |
1648 | */ | |
1649 | ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m); | |
1650 | if (ret < 0) { | |
1651 | return ret; | |
1652 | } else if (ret) { | |
ecdd5333 | 1653 | continue; |
2c3b32d2 KW |
1654 | } else { |
1655 | assert(cur_bytes == 0); | |
1656 | break; | |
1657 | } | |
f5bc6350 | 1658 | } |
10f0ed8b | 1659 | |
d46a0bb2 KW |
1660 | *bytes -= remaining; |
1661 | assert(*bytes > 0); | |
c6d619cc | 1662 | assert(*host_offset != INV_OFFSET); |
45aba42f | 1663 | |
148da7ea | 1664 | return 0; |
45aba42f KW |
1665 | } |
1666 | ||
5ea929e3 KW |
1667 | /* |
1668 | * This discards as many clusters of nb_clusters as possible at once (i.e. | |
21ab3add | 1669 | * all clusters in the same L2 slice) and returns the number of discarded |
5ea929e3 KW |
1670 | * clusters. |
1671 | */ | |
21ab3add AG |
1672 | static int discard_in_l2_slice(BlockDriverState *bs, uint64_t offset, |
1673 | uint64_t nb_clusters, | |
1674 | enum qcow2_discard_type type, bool full_discard) | |
5ea929e3 | 1675 | { |
ff99129a | 1676 | BDRVQcow2State *s = bs->opaque; |
21ab3add | 1677 | uint64_t *l2_slice; |
5ea929e3 KW |
1678 | int l2_index; |
1679 | int ret; | |
1680 | int i; | |
1681 | ||
21ab3add | 1682 | ret = get_cluster_table(bs, offset, &l2_slice, &l2_index); |
5ea929e3 KW |
1683 | if (ret < 0) { |
1684 | return ret; | |
1685 | } | |
1686 | ||
21ab3add AG |
1687 | /* Limit nb_clusters to one L2 slice */ |
1688 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
b6d36def | 1689 | assert(nb_clusters <= INT_MAX); |
5ea929e3 KW |
1690 | |
1691 | for (i = 0; i < nb_clusters; i++) { | |
c883db0d | 1692 | uint64_t old_l2_entry; |
5ea929e3 | 1693 | |
12c6aebe | 1694 | old_l2_entry = get_l2_entry(s, l2_slice, l2_index + i); |
a71835a0 KW |
1695 | |
1696 | /* | |
808c4b6f HR |
1697 | * If full_discard is false, make sure that a discarded area reads back |
1698 | * as zeroes for v3 images (we cannot do it for v2 without actually | |
1699 | * writing a zero-filled buffer). We can skip the operation if the | |
1700 | * cluster is already marked as zero, or if it's unallocated and we | |
1701 | * don't have a backing file. | |
a71835a0 | 1702 | * |
237d78f8 | 1703 | * TODO We might want to use bdrv_block_status(bs) here, but we're |
a71835a0 | 1704 | * holding s->lock, so that doesn't work today. |
808c4b6f HR |
1705 | * |
1706 | * If full_discard is true, the sector should not read back as zeroes, | |
1707 | * but rather fall through to the backing file. | |
a71835a0 | 1708 | */ |
808c2bb4 | 1709 | switch (qcow2_get_cluster_type(bs, old_l2_entry)) { |
bbd995d8 EB |
1710 | case QCOW2_CLUSTER_UNALLOCATED: |
1711 | if (full_discard || !bs->backing) { | |
1712 | continue; | |
1713 | } | |
1714 | break; | |
1715 | ||
fdfab37d EB |
1716 | case QCOW2_CLUSTER_ZERO_PLAIN: |
1717 | if (!full_discard) { | |
bbd995d8 EB |
1718 | continue; |
1719 | } | |
1720 | break; | |
1721 | ||
fdfab37d | 1722 | case QCOW2_CLUSTER_ZERO_ALLOC: |
bbd995d8 EB |
1723 | case QCOW2_CLUSTER_NORMAL: |
1724 | case QCOW2_CLUSTER_COMPRESSED: | |
1725 | break; | |
1726 | ||
1727 | default: | |
1728 | abort(); | |
5ea929e3 KW |
1729 | } |
1730 | ||
1731 | /* First remove L2 entries */ | |
21ab3add | 1732 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
808c4b6f | 1733 | if (!full_discard && s->qcow_version >= 3) { |
12c6aebe | 1734 | set_l2_entry(s, l2_slice, l2_index + i, QCOW_OFLAG_ZERO); |
a71835a0 | 1735 | } else { |
12c6aebe | 1736 | set_l2_entry(s, l2_slice, l2_index + i, 0); |
a71835a0 | 1737 | } |
5ea929e3 KW |
1738 | |
1739 | /* Then decrease the refcount */ | |
c883db0d | 1740 | qcow2_free_any_clusters(bs, old_l2_entry, 1, type); |
5ea929e3 KW |
1741 | } |
1742 | ||
21ab3add | 1743 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
5ea929e3 KW |
1744 | |
1745 | return nb_clusters; | |
1746 | } | |
1747 | ||
d2cb36af EB |
1748 | int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset, |
1749 | uint64_t bytes, enum qcow2_discard_type type, | |
1750 | bool full_discard) | |
5ea929e3 | 1751 | { |
ff99129a | 1752 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1753 | uint64_t end_offset = offset + bytes; |
b6d36def | 1754 | uint64_t nb_clusters; |
d2cb36af | 1755 | int64_t cleared; |
5ea929e3 KW |
1756 | int ret; |
1757 | ||
f10ee139 | 1758 | /* Caller must pass aligned values, except at image end */ |
0c1bd469 | 1759 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); |
f10ee139 EB |
1760 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || |
1761 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
5ea929e3 | 1762 | |
d2cb36af | 1763 | nb_clusters = size_to_clusters(s, bytes); |
5ea929e3 | 1764 | |
0b919fae KW |
1765 | s->cache_discards = true; |
1766 | ||
21ab3add | 1767 | /* Each L2 slice is handled by its own loop iteration */ |
5ea929e3 | 1768 | while (nb_clusters > 0) { |
21ab3add AG |
1769 | cleared = discard_in_l2_slice(bs, offset, nb_clusters, type, |
1770 | full_discard); | |
d2cb36af EB |
1771 | if (cleared < 0) { |
1772 | ret = cleared; | |
0b919fae | 1773 | goto fail; |
5ea929e3 KW |
1774 | } |
1775 | ||
d2cb36af EB |
1776 | nb_clusters -= cleared; |
1777 | offset += (cleared * s->cluster_size); | |
5ea929e3 KW |
1778 | } |
1779 | ||
0b919fae KW |
1780 | ret = 0; |
1781 | fail: | |
1782 | s->cache_discards = false; | |
1783 | qcow2_process_discards(bs, ret); | |
1784 | ||
1785 | return ret; | |
5ea929e3 | 1786 | } |
621f0589 KW |
1787 | |
1788 | /* | |
1789 | * This zeroes as many clusters of nb_clusters as possible at once (i.e. | |
a9a9f8f0 | 1790 | * all clusters in the same L2 slice) and returns the number of zeroed |
621f0589 KW |
1791 | * clusters. |
1792 | */ | |
a9a9f8f0 AG |
1793 | static int zero_in_l2_slice(BlockDriverState *bs, uint64_t offset, |
1794 | uint64_t nb_clusters, int flags) | |
621f0589 | 1795 | { |
ff99129a | 1796 | BDRVQcow2State *s = bs->opaque; |
a9a9f8f0 | 1797 | uint64_t *l2_slice; |
621f0589 KW |
1798 | int l2_index; |
1799 | int ret; | |
1800 | int i; | |
06cc5e2b | 1801 | bool unmap = !!(flags & BDRV_REQ_MAY_UNMAP); |
621f0589 | 1802 | |
a9a9f8f0 | 1803 | ret = get_cluster_table(bs, offset, &l2_slice, &l2_index); |
621f0589 KW |
1804 | if (ret < 0) { |
1805 | return ret; | |
1806 | } | |
1807 | ||
a9a9f8f0 AG |
1808 | /* Limit nb_clusters to one L2 slice */ |
1809 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
b6d36def | 1810 | assert(nb_clusters <= INT_MAX); |
621f0589 KW |
1811 | |
1812 | for (i = 0; i < nb_clusters; i++) { | |
1813 | uint64_t old_offset; | |
06cc5e2b | 1814 | QCow2ClusterType cluster_type; |
621f0589 | 1815 | |
12c6aebe | 1816 | old_offset = get_l2_entry(s, l2_slice, l2_index + i); |
621f0589 | 1817 | |
06cc5e2b EB |
1818 | /* |
1819 | * Minimize L2 changes if the cluster already reads back as | |
1820 | * zeroes with correct allocation. | |
1821 | */ | |
808c2bb4 | 1822 | cluster_type = qcow2_get_cluster_type(bs, old_offset); |
06cc5e2b EB |
1823 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN || |
1824 | (cluster_type == QCOW2_CLUSTER_ZERO_ALLOC && !unmap)) { | |
1825 | continue; | |
1826 | } | |
1827 | ||
a9a9f8f0 | 1828 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
06cc5e2b | 1829 | if (cluster_type == QCOW2_CLUSTER_COMPRESSED || unmap) { |
12c6aebe | 1830 | set_l2_entry(s, l2_slice, l2_index + i, QCOW_OFLAG_ZERO); |
6cfcb9b8 | 1831 | qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST); |
621f0589 | 1832 | } else { |
12c6aebe AG |
1833 | uint64_t entry = get_l2_entry(s, l2_slice, l2_index + i); |
1834 | set_l2_entry(s, l2_slice, l2_index + i, entry | QCOW_OFLAG_ZERO); | |
621f0589 KW |
1835 | } |
1836 | } | |
1837 | ||
a9a9f8f0 | 1838 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
621f0589 KW |
1839 | |
1840 | return nb_clusters; | |
1841 | } | |
1842 | ||
d2cb36af EB |
1843 | int qcow2_cluster_zeroize(BlockDriverState *bs, uint64_t offset, |
1844 | uint64_t bytes, int flags) | |
621f0589 | 1845 | { |
ff99129a | 1846 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 1847 | uint64_t end_offset = offset + bytes; |
b6d36def | 1848 | uint64_t nb_clusters; |
d2cb36af | 1849 | int64_t cleared; |
621f0589 KW |
1850 | int ret; |
1851 | ||
6c3944dc KW |
1852 | /* If we have to stay in sync with an external data file, zero out |
1853 | * s->data_file first. */ | |
1854 | if (data_file_is_raw(bs)) { | |
1855 | assert(has_data_file(bs)); | |
1856 | ret = bdrv_co_pwrite_zeroes(s->data_file, offset, bytes, flags); | |
1857 | if (ret < 0) { | |
1858 | return ret; | |
1859 | } | |
1860 | } | |
1861 | ||
f10ee139 EB |
1862 | /* Caller must pass aligned values, except at image end */ |
1863 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); | |
1864 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || | |
f01643fb | 1865 | end_offset >= bs->total_sectors << BDRV_SECTOR_BITS); |
f10ee139 | 1866 | |
61b30439 KW |
1867 | /* |
1868 | * The zero flag is only supported by version 3 and newer. However, if we | |
1869 | * have no backing file, we can resort to discard in version 2. | |
1870 | */ | |
621f0589 | 1871 | if (s->qcow_version < 3) { |
61b30439 KW |
1872 | if (!bs->backing) { |
1873 | return qcow2_cluster_discard(bs, offset, bytes, | |
1874 | QCOW2_DISCARD_REQUEST, false); | |
1875 | } | |
621f0589 KW |
1876 | return -ENOTSUP; |
1877 | } | |
1878 | ||
a9a9f8f0 | 1879 | /* Each L2 slice is handled by its own loop iteration */ |
d2cb36af | 1880 | nb_clusters = size_to_clusters(s, bytes); |
621f0589 | 1881 | |
0b919fae KW |
1882 | s->cache_discards = true; |
1883 | ||
621f0589 | 1884 | while (nb_clusters > 0) { |
a9a9f8f0 | 1885 | cleared = zero_in_l2_slice(bs, offset, nb_clusters, flags); |
d2cb36af EB |
1886 | if (cleared < 0) { |
1887 | ret = cleared; | |
0b919fae | 1888 | goto fail; |
621f0589 KW |
1889 | } |
1890 | ||
d2cb36af EB |
1891 | nb_clusters -= cleared; |
1892 | offset += (cleared * s->cluster_size); | |
621f0589 KW |
1893 | } |
1894 | ||
0b919fae KW |
1895 | ret = 0; |
1896 | fail: | |
1897 | s->cache_discards = false; | |
1898 | qcow2_process_discards(bs, ret); | |
1899 | ||
1900 | return ret; | |
621f0589 | 1901 | } |
32b6444d HR |
1902 | |
1903 | /* | |
1904 | * Expands all zero clusters in a specific L1 table (or deallocates them, for | |
1905 | * non-backed non-pre-allocated zero clusters). | |
1906 | * | |
4057a2b2 HR |
1907 | * l1_entries and *visited_l1_entries are used to keep track of progress for |
1908 | * status_cb(). l1_entries contains the total number of L1 entries and | |
1909 | * *visited_l1_entries counts all visited L1 entries. | |
32b6444d HR |
1910 | */ |
1911 | static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table, | |
ecf58777 | 1912 | int l1_size, int64_t *visited_l1_entries, |
4057a2b2 | 1913 | int64_t l1_entries, |
8b13976d HR |
1914 | BlockDriverAmendStatusCB *status_cb, |
1915 | void *cb_opaque) | |
32b6444d | 1916 | { |
ff99129a | 1917 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 1918 | bool is_active_l1 = (l1_table == s->l1_table); |
415184f5 AG |
1919 | uint64_t *l2_slice = NULL; |
1920 | unsigned slice, slice_size2, n_slices; | |
32b6444d HR |
1921 | int ret; |
1922 | int i, j; | |
1923 | ||
415184f5 AG |
1924 | slice_size2 = s->l2_slice_size * sizeof(uint64_t); |
1925 | n_slices = s->cluster_size / slice_size2; | |
1926 | ||
32b6444d HR |
1927 | if (!is_active_l1) { |
1928 | /* inactive L2 tables require a buffer to be stored in when loading | |
1929 | * them from disk */ | |
415184f5 AG |
1930 | l2_slice = qemu_try_blockalign(bs->file->bs, slice_size2); |
1931 | if (l2_slice == NULL) { | |
de82815d KW |
1932 | return -ENOMEM; |
1933 | } | |
32b6444d HR |
1934 | } |
1935 | ||
1936 | for (i = 0; i < l1_size; i++) { | |
1937 | uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK; | |
0e06528e | 1938 | uint64_t l2_refcount; |
32b6444d HR |
1939 | |
1940 | if (!l2_offset) { | |
1941 | /* unallocated */ | |
4057a2b2 HR |
1942 | (*visited_l1_entries)++; |
1943 | if (status_cb) { | |
8b13976d | 1944 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 1945 | } |
32b6444d HR |
1946 | continue; |
1947 | } | |
1948 | ||
8dd93d93 HR |
1949 | if (offset_into_cluster(s, l2_offset)) { |
1950 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" | |
1951 | PRIx64 " unaligned (L1 index: %#x)", | |
1952 | l2_offset, i); | |
1953 | ret = -EIO; | |
1954 | goto fail; | |
1955 | } | |
1956 | ||
9b765486 AG |
1957 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
1958 | &l2_refcount); | |
1959 | if (ret < 0) { | |
1960 | goto fail; | |
1961 | } | |
1962 | ||
415184f5 AG |
1963 | for (slice = 0; slice < n_slices; slice++) { |
1964 | uint64_t slice_offset = l2_offset + slice * slice_size2; | |
1965 | bool l2_dirty = false; | |
226494ff AG |
1966 | if (is_active_l1) { |
1967 | /* get active L2 tables from cache */ | |
415184f5 AG |
1968 | ret = qcow2_cache_get(bs, s->l2_table_cache, slice_offset, |
1969 | (void **)&l2_slice); | |
226494ff AG |
1970 | } else { |
1971 | /* load inactive L2 tables from disk */ | |
415184f5 | 1972 | ret = bdrv_pread(bs->file, slice_offset, l2_slice, slice_size2); |
226494ff AG |
1973 | } |
1974 | if (ret < 0) { | |
1975 | goto fail; | |
32b6444d HR |
1976 | } |
1977 | ||
415184f5 | 1978 | for (j = 0; j < s->l2_slice_size; j++) { |
12c6aebe | 1979 | uint64_t l2_entry = get_l2_entry(s, l2_slice, j); |
226494ff AG |
1980 | int64_t offset = l2_entry & L2E_OFFSET_MASK; |
1981 | QCow2ClusterType cluster_type = | |
808c2bb4 | 1982 | qcow2_get_cluster_type(bs, l2_entry); |
226494ff AG |
1983 | |
1984 | if (cluster_type != QCOW2_CLUSTER_ZERO_PLAIN && | |
1985 | cluster_type != QCOW2_CLUSTER_ZERO_ALLOC) { | |
32b6444d HR |
1986 | continue; |
1987 | } | |
1988 | ||
226494ff AG |
1989 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
1990 | if (!bs->backing) { | |
1991 | /* not backed; therefore we can simply deallocate the | |
1992 | * cluster */ | |
12c6aebe | 1993 | set_l2_entry(s, l2_slice, j, 0); |
226494ff AG |
1994 | l2_dirty = true; |
1995 | continue; | |
1996 | } | |
1997 | ||
1998 | offset = qcow2_alloc_clusters(bs, s->cluster_size); | |
1999 | if (offset < 0) { | |
2000 | ret = offset; | |
2001 | goto fail; | |
2002 | } | |
ecf58777 | 2003 | |
3a75a870 AG |
2004 | /* The offset must fit in the offset field */ |
2005 | assert((offset & L2E_OFFSET_MASK) == offset); | |
2006 | ||
226494ff AG |
2007 | if (l2_refcount > 1) { |
2008 | /* For shared L2 tables, set the refcount accordingly | |
2009 | * (it is already 1 and needs to be l2_refcount) */ | |
2010 | ret = qcow2_update_cluster_refcount( | |
2011 | bs, offset >> s->cluster_bits, | |
2aabe7c7 | 2012 | refcount_diff(1, l2_refcount), false, |
ecf58777 | 2013 | QCOW2_DISCARD_OTHER); |
226494ff AG |
2014 | if (ret < 0) { |
2015 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
2016 | QCOW2_DISCARD_OTHER); | |
2017 | goto fail; | |
2018 | } | |
ecf58777 HR |
2019 | } |
2020 | } | |
32b6444d | 2021 | |
226494ff | 2022 | if (offset_into_cluster(s, offset)) { |
415184f5 | 2023 | int l2_index = slice * s->l2_slice_size + j; |
226494ff AG |
2024 | qcow2_signal_corruption( |
2025 | bs, true, -1, -1, | |
2026 | "Cluster allocation offset " | |
2027 | "%#" PRIx64 " unaligned (L2 offset: %#" | |
2028 | PRIx64 ", L2 index: %#x)", offset, | |
415184f5 | 2029 | l2_offset, l2_index); |
226494ff AG |
2030 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
2031 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
2032 | QCOW2_DISCARD_ALWAYS); | |
2033 | } | |
2034 | ret = -EIO; | |
2035 | goto fail; | |
8dd93d93 | 2036 | } |
8dd93d93 | 2037 | |
226494ff | 2038 | ret = qcow2_pre_write_overlap_check(bs, 0, offset, |
966b000f | 2039 | s->cluster_size, true); |
226494ff AG |
2040 | if (ret < 0) { |
2041 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { | |
2042 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
2043 | QCOW2_DISCARD_ALWAYS); | |
2044 | } | |
2045 | goto fail; | |
320c7066 | 2046 | } |
32b6444d | 2047 | |
966b000f KW |
2048 | ret = bdrv_pwrite_zeroes(s->data_file, offset, |
2049 | s->cluster_size, 0); | |
226494ff AG |
2050 | if (ret < 0) { |
2051 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { | |
2052 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
2053 | QCOW2_DISCARD_ALWAYS); | |
2054 | } | |
2055 | goto fail; | |
320c7066 | 2056 | } |
32b6444d | 2057 | |
226494ff | 2058 | if (l2_refcount == 1) { |
12c6aebe | 2059 | set_l2_entry(s, l2_slice, j, offset | QCOW_OFLAG_COPIED); |
226494ff | 2060 | } else { |
12c6aebe | 2061 | set_l2_entry(s, l2_slice, j, offset); |
226494ff AG |
2062 | } |
2063 | l2_dirty = true; | |
e390cf5a | 2064 | } |
32b6444d | 2065 | |
226494ff AG |
2066 | if (is_active_l1) { |
2067 | if (l2_dirty) { | |
415184f5 | 2068 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
226494ff | 2069 | qcow2_cache_depends_on_flush(s->l2_table_cache); |
32b6444d | 2070 | } |
415184f5 | 2071 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
226494ff AG |
2072 | } else { |
2073 | if (l2_dirty) { | |
2074 | ret = qcow2_pre_write_overlap_check( | |
2075 | bs, QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, | |
966b000f | 2076 | slice_offset, slice_size2, false); |
226494ff AG |
2077 | if (ret < 0) { |
2078 | goto fail; | |
2079 | } | |
32b6444d | 2080 | |
415184f5 AG |
2081 | ret = bdrv_pwrite(bs->file, slice_offset, |
2082 | l2_slice, slice_size2); | |
226494ff AG |
2083 | if (ret < 0) { |
2084 | goto fail; | |
2085 | } | |
32b6444d HR |
2086 | } |
2087 | } | |
2088 | } | |
4057a2b2 HR |
2089 | |
2090 | (*visited_l1_entries)++; | |
2091 | if (status_cb) { | |
8b13976d | 2092 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 2093 | } |
32b6444d HR |
2094 | } |
2095 | ||
2096 | ret = 0; | |
2097 | ||
2098 | fail: | |
415184f5 | 2099 | if (l2_slice) { |
32b6444d | 2100 | if (!is_active_l1) { |
415184f5 | 2101 | qemu_vfree(l2_slice); |
32b6444d | 2102 | } else { |
415184f5 | 2103 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
32b6444d HR |
2104 | } |
2105 | } | |
2106 | return ret; | |
2107 | } | |
2108 | ||
2109 | /* | |
2110 | * For backed images, expands all zero clusters on the image. For non-backed | |
2111 | * images, deallocates all non-pre-allocated zero clusters (and claims the | |
2112 | * allocation for pre-allocated ones). This is important for downgrading to a | |
2113 | * qcow2 version which doesn't yet support metadata zero clusters. | |
2114 | */ | |
4057a2b2 | 2115 | int qcow2_expand_zero_clusters(BlockDriverState *bs, |
8b13976d HR |
2116 | BlockDriverAmendStatusCB *status_cb, |
2117 | void *cb_opaque) | |
32b6444d | 2118 | { |
ff99129a | 2119 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 2120 | uint64_t *l1_table = NULL; |
4057a2b2 | 2121 | int64_t l1_entries = 0, visited_l1_entries = 0; |
32b6444d HR |
2122 | int ret; |
2123 | int i, j; | |
2124 | ||
4057a2b2 HR |
2125 | if (status_cb) { |
2126 | l1_entries = s->l1_size; | |
2127 | for (i = 0; i < s->nb_snapshots; i++) { | |
2128 | l1_entries += s->snapshots[i].l1_size; | |
2129 | } | |
2130 | } | |
2131 | ||
32b6444d | 2132 | ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size, |
4057a2b2 | 2133 | &visited_l1_entries, l1_entries, |
8b13976d | 2134 | status_cb, cb_opaque); |
32b6444d HR |
2135 | if (ret < 0) { |
2136 | goto fail; | |
2137 | } | |
2138 | ||
2139 | /* Inactive L1 tables may point to active L2 tables - therefore it is | |
2140 | * necessary to flush the L2 table cache before trying to access the L2 | |
2141 | * tables pointed to by inactive L1 entries (else we might try to expand | |
2142 | * zero clusters that have already been expanded); furthermore, it is also | |
2143 | * necessary to empty the L2 table cache, since it may contain tables which | |
2144 | * are now going to be modified directly on disk, bypassing the cache. | |
2145 | * qcow2_cache_empty() does both for us. */ | |
2146 | ret = qcow2_cache_empty(bs, s->l2_table_cache); | |
2147 | if (ret < 0) { | |
2148 | goto fail; | |
2149 | } | |
2150 | ||
2151 | for (i = 0; i < s->nb_snapshots; i++) { | |
c9a442e4 AG |
2152 | int l1_size2; |
2153 | uint64_t *new_l1_table; | |
2154 | Error *local_err = NULL; | |
2155 | ||
2156 | ret = qcow2_validate_table(bs, s->snapshots[i].l1_table_offset, | |
2157 | s->snapshots[i].l1_size, sizeof(uint64_t), | |
2158 | QCOW_MAX_L1_SIZE, "Snapshot L1 table", | |
2159 | &local_err); | |
2160 | if (ret < 0) { | |
2161 | error_report_err(local_err); | |
2162 | goto fail; | |
2163 | } | |
32b6444d | 2164 | |
c9a442e4 AG |
2165 | l1_size2 = s->snapshots[i].l1_size * sizeof(uint64_t); |
2166 | new_l1_table = g_try_realloc(l1_table, l1_size2); | |
de7269d2 AG |
2167 | |
2168 | if (!new_l1_table) { | |
2169 | ret = -ENOMEM; | |
2170 | goto fail; | |
2171 | } | |
2172 | ||
2173 | l1_table = new_l1_table; | |
32b6444d | 2174 | |
c9a442e4 AG |
2175 | ret = bdrv_pread(bs->file, s->snapshots[i].l1_table_offset, |
2176 | l1_table, l1_size2); | |
32b6444d HR |
2177 | if (ret < 0) { |
2178 | goto fail; | |
2179 | } | |
2180 | ||
2181 | for (j = 0; j < s->snapshots[i].l1_size; j++) { | |
2182 | be64_to_cpus(&l1_table[j]); | |
2183 | } | |
2184 | ||
2185 | ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size, | |
4057a2b2 | 2186 | &visited_l1_entries, l1_entries, |
8b13976d | 2187 | status_cb, cb_opaque); |
32b6444d HR |
2188 | if (ret < 0) { |
2189 | goto fail; | |
2190 | } | |
2191 | } | |
2192 | ||
2193 | ret = 0; | |
2194 | ||
2195 | fail: | |
32b6444d HR |
2196 | g_free(l1_table); |
2197 | return ret; | |
2198 | } |