]>
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 | ||
e2c1c34f | 28 | #include "block/block-io.h" |
c9a442e4 | 29 | #include "qapi/error.h" |
0d8c41da | 30 | #include "qcow2.h" |
58369e22 | 31 | #include "qemu/bswap.h" |
5df022cf | 32 | #include "qemu/memalign.h" |
3cce16f4 | 33 | #include "trace.h" |
45aba42f | 34 | |
a1b4ecfd PB |
35 | int coroutine_fn qcow2_shrink_l1_table(BlockDriverState *bs, |
36 | uint64_t exact_size) | |
46b732cd PB |
37 | { |
38 | BDRVQcow2State *s = bs->opaque; | |
39 | int new_l1_size, i, ret; | |
40 | ||
41 | if (exact_size >= s->l1_size) { | |
42 | return 0; | |
43 | } | |
44 | ||
45 | new_l1_size = exact_size; | |
46 | ||
47 | #ifdef DEBUG_ALLOC2 | |
48 | fprintf(stderr, "shrink l1_table from %d to %d\n", s->l1_size, new_l1_size); | |
49 | #endif | |
50 | ||
17362398 | 51 | BLKDBG_CO_EVENT(bs->file, BLKDBG_L1_SHRINK_WRITE_TABLE); |
38505e2a AF |
52 | ret = bdrv_co_pwrite_zeroes(bs->file, |
53 | s->l1_table_offset + new_l1_size * L1E_SIZE, | |
54 | (s->l1_size - new_l1_size) * L1E_SIZE, 0); | |
46b732cd PB |
55 | if (ret < 0) { |
56 | goto fail; | |
57 | } | |
58 | ||
38505e2a | 59 | ret = bdrv_co_flush(bs->file->bs); |
46b732cd PB |
60 | if (ret < 0) { |
61 | goto fail; | |
62 | } | |
63 | ||
17362398 | 64 | BLKDBG_CO_EVENT(bs->file, BLKDBG_L1_SHRINK_FREE_L2_CLUSTERS); |
46b732cd PB |
65 | for (i = s->l1_size - 1; i > new_l1_size - 1; i--) { |
66 | if ((s->l1_table[i] & L1E_OFFSET_MASK) == 0) { | |
67 | continue; | |
68 | } | |
69 | qcow2_free_clusters(bs, s->l1_table[i] & L1E_OFFSET_MASK, | |
70 | s->cluster_size, QCOW2_DISCARD_ALWAYS); | |
71 | s->l1_table[i] = 0; | |
72 | } | |
73 | return 0; | |
74 | ||
75 | fail: | |
76 | /* | |
77 | * If the write in the l1_table failed the image may contain a partially | |
78 | * overwritten l1_table. In this case it would be better to clear the | |
79 | * l1_table in memory to avoid possible image corruption. | |
80 | */ | |
81 | memset(s->l1_table + new_l1_size, 0, | |
02b1ecfa | 82 | (s->l1_size - new_l1_size) * L1E_SIZE); |
46b732cd PB |
83 | return ret; |
84 | } | |
85 | ||
2cf7cfa1 KW |
86 | int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size, |
87 | bool exact_size) | |
45aba42f | 88 | { |
ff99129a | 89 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 | 90 | int new_l1_size2, ret, i; |
45aba42f | 91 | uint64_t *new_l1_table; |
fda74f82 | 92 | int64_t old_l1_table_offset, old_l1_size; |
2cf7cfa1 | 93 | int64_t new_l1_table_offset, new_l1_size; |
45aba42f KW |
94 | uint8_t data[12]; |
95 | ||
72893756 | 96 | if (min_size <= s->l1_size) |
45aba42f | 97 | return 0; |
72893756 | 98 | |
b93f9950 HR |
99 | /* Do a sanity check on min_size before trying to calculate new_l1_size |
100 | * (this prevents overflows during the while loop for the calculation of | |
101 | * new_l1_size) */ | |
02b1ecfa | 102 | if (min_size > INT_MAX / L1E_SIZE) { |
b93f9950 HR |
103 | return -EFBIG; |
104 | } | |
105 | ||
72893756 SH |
106 | if (exact_size) { |
107 | new_l1_size = min_size; | |
108 | } else { | |
109 | /* Bump size up to reduce the number of times we have to grow */ | |
110 | new_l1_size = s->l1_size; | |
111 | if (new_l1_size == 0) { | |
112 | new_l1_size = 1; | |
113 | } | |
114 | while (min_size > new_l1_size) { | |
21cf3e12 | 115 | new_l1_size = DIV_ROUND_UP(new_l1_size * 3, 2); |
72893756 | 116 | } |
45aba42f | 117 | } |
72893756 | 118 | |
84c26520 | 119 | QEMU_BUILD_BUG_ON(QCOW_MAX_L1_SIZE > INT_MAX); |
02b1ecfa | 120 | if (new_l1_size > QCOW_MAX_L1_SIZE / L1E_SIZE) { |
2cf7cfa1 KW |
121 | return -EFBIG; |
122 | } | |
123 | ||
45aba42f | 124 | #ifdef DEBUG_ALLOC2 |
2cf7cfa1 KW |
125 | fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n", |
126 | s->l1_size, new_l1_size); | |
45aba42f KW |
127 | #endif |
128 | ||
02b1ecfa | 129 | new_l1_size2 = L1E_SIZE * new_l1_size; |
ef97d608 | 130 | new_l1_table = qemu_try_blockalign(bs->file->bs, new_l1_size2); |
de82815d KW |
131 | if (new_l1_table == NULL) { |
132 | return -ENOMEM; | |
133 | } | |
ef97d608 | 134 | memset(new_l1_table, 0, new_l1_size2); |
de82815d | 135 | |
0647d47c | 136 | if (s->l1_size) { |
02b1ecfa | 137 | memcpy(new_l1_table, s->l1_table, s->l1_size * L1E_SIZE); |
0647d47c | 138 | } |
45aba42f KW |
139 | |
140 | /* write new table (align to cluster) */ | |
66f82cee | 141 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE); |
ed6ccf0f | 142 | new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2); |
5d757b56 | 143 | if (new_l1_table_offset < 0) { |
de82815d | 144 | qemu_vfree(new_l1_table); |
5d757b56 KW |
145 | return new_l1_table_offset; |
146 | } | |
29c1a730 KW |
147 | |
148 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); | |
149 | if (ret < 0) { | |
80fa3341 | 150 | goto fail; |
29c1a730 | 151 | } |
45aba42f | 152 | |
cf93980e HR |
153 | /* the L1 position has not yet been updated, so these clusters must |
154 | * indeed be completely free */ | |
231bb267 | 155 | ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset, |
966b000f | 156 | new_l1_size2, false); |
cf93980e HR |
157 | if (ret < 0) { |
158 | goto fail; | |
159 | } | |
160 | ||
66f82cee | 161 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE); |
45aba42f KW |
162 | for(i = 0; i < s->l1_size; i++) |
163 | new_l1_table[i] = cpu_to_be64(new_l1_table[i]); | |
32cc71de AF |
164 | ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, new_l1_size2, |
165 | new_l1_table, 0); | |
8b3b7206 | 166 | if (ret < 0) |
45aba42f KW |
167 | goto fail; |
168 | for(i = 0; i < s->l1_size; i++) | |
169 | new_l1_table[i] = be64_to_cpu(new_l1_table[i]); | |
170 | ||
171 | /* set new table */ | |
66f82cee | 172 | BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE); |
f1f7a1dd | 173 | stl_be_p(data, new_l1_size); |
e4ef9f46 | 174 | stq_be_p(data + 4, new_l1_table_offset); |
32cc71de AF |
175 | ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), |
176 | sizeof(data), data, 0); | |
8b3b7206 | 177 | if (ret < 0) { |
45aba42f | 178 | goto fail; |
fb8fa77c | 179 | } |
de82815d | 180 | qemu_vfree(s->l1_table); |
fda74f82 | 181 | old_l1_table_offset = s->l1_table_offset; |
45aba42f KW |
182 | s->l1_table_offset = new_l1_table_offset; |
183 | s->l1_table = new_l1_table; | |
fda74f82 | 184 | old_l1_size = s->l1_size; |
45aba42f | 185 | s->l1_size = new_l1_size; |
02b1ecfa | 186 | qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * L1E_SIZE, |
fda74f82 | 187 | QCOW2_DISCARD_OTHER); |
45aba42f KW |
188 | return 0; |
189 | fail: | |
de82815d | 190 | qemu_vfree(new_l1_table); |
6cfcb9b8 KW |
191 | qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2, |
192 | QCOW2_DISCARD_OTHER); | |
8b3b7206 | 193 | return ret; |
45aba42f KW |
194 | } |
195 | ||
45aba42f KW |
196 | /* |
197 | * l2_load | |
198 | * | |
e2b5713e AG |
199 | * @bs: The BlockDriverState |
200 | * @offset: A guest offset, used to calculate what slice of the L2 | |
201 | * table to load. | |
202 | * @l2_offset: Offset to the L2 table in the image file. | |
203 | * @l2_slice: Location to store the pointer to the L2 slice. | |
45aba42f | 204 | * |
e2b5713e AG |
205 | * Loads a L2 slice into memory (L2 slices are the parts of L2 tables |
206 | * that are loaded by the qcow2 cache). If the slice is in the cache, | |
207 | * the cache is used; otherwise the L2 slice is loaded from the image | |
208 | * file. | |
45aba42f | 209 | */ |
e2b5713e AG |
210 | static int l2_load(BlockDriverState *bs, uint64_t offset, |
211 | uint64_t l2_offset, uint64_t **l2_slice) | |
45aba42f | 212 | { |
ff99129a | 213 | BDRVQcow2State *s = bs->opaque; |
c8fd8554 | 214 | int start_of_slice = l2_entry_size(s) * |
e2b5713e | 215 | (offset_to_l2_index(s, offset) - offset_to_l2_slice_index(s, offset)); |
45aba42f | 216 | |
e2b5713e AG |
217 | return qcow2_cache_get(bs, s->l2_table_cache, l2_offset + start_of_slice, |
218 | (void **)l2_slice); | |
45aba42f KW |
219 | } |
220 | ||
6583e3c7 | 221 | /* |
da86f8cb AG |
222 | * Writes an L1 entry to disk (note that depending on the alignment |
223 | * requirements this function may write more that just one entry in | |
224 | * order to prevent bdrv_pwrite from performing a read-modify-write) | |
6583e3c7 | 225 | */ |
e23e400e | 226 | int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index) |
6583e3c7 | 227 | { |
ff99129a | 228 | BDRVQcow2State *s = bs->opaque; |
6583e3c7 | 229 | int l1_start_index; |
f7defcb6 | 230 | int i, ret; |
02b1ecfa | 231 | int bufsize = MAX(L1E_SIZE, |
da86f8cb | 232 | MIN(bs->file->bs->bl.request_alignment, s->cluster_size)); |
02b1ecfa | 233 | int nentries = bufsize / L1E_SIZE; |
da86f8cb | 234 | g_autofree uint64_t *buf = g_try_new0(uint64_t, nentries); |
6583e3c7 | 235 | |
da86f8cb AG |
236 | if (buf == NULL) { |
237 | return -ENOMEM; | |
238 | } | |
239 | ||
240 | l1_start_index = QEMU_ALIGN_DOWN(l1_index, nentries); | |
241 | for (i = 0; i < MIN(nentries, s->l1_size - l1_start_index); i++) { | |
6583e3c7 KW |
242 | buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]); |
243 | } | |
244 | ||
231bb267 | 245 | ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1, |
c508c73d | 246 | s->l1_table_offset + L1E_SIZE * l1_start_index, bufsize, false); |
cf93980e HR |
247 | if (ret < 0) { |
248 | return ret; | |
249 | } | |
250 | ||
66f82cee | 251 | BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE); |
d9ca2ea2 | 252 | ret = bdrv_pwrite_sync(bs->file, |
c508c73d | 253 | s->l1_table_offset + L1E_SIZE * l1_start_index, |
32cc71de | 254 | bufsize, buf, 0); |
f7defcb6 KW |
255 | if (ret < 0) { |
256 | return ret; | |
6583e3c7 KW |
257 | } |
258 | ||
259 | return 0; | |
260 | } | |
261 | ||
45aba42f KW |
262 | /* |
263 | * l2_allocate | |
264 | * | |
265 | * Allocate a new l2 entry in the file. If l1_index points to an already | |
266 | * used entry in the L2 table (i.e. we are doing a copy on write for the L2 | |
267 | * table) copy the contents of the old L2 table into the newly allocated one. | |
268 | * Otherwise the new table is initialized with zeros. | |
269 | * | |
270 | */ | |
271 | ||
3861946a | 272 | static int l2_allocate(BlockDriverState *bs, int l1_index) |
45aba42f | 273 | { |
ff99129a | 274 | BDRVQcow2State *s = bs->opaque; |
6583e3c7 | 275 | uint64_t old_l2_offset; |
3861946a AG |
276 | uint64_t *l2_slice = NULL; |
277 | unsigned slice, slice_size2, n_slices; | |
f4f0d391 | 278 | int64_t l2_offset; |
c46e1167 | 279 | int ret; |
45aba42f KW |
280 | |
281 | old_l2_offset = s->l1_table[l1_index]; | |
282 | ||
3cce16f4 KW |
283 | trace_qcow2_l2_allocate(bs, l1_index); |
284 | ||
45aba42f KW |
285 | /* allocate a new l2 entry */ |
286 | ||
c8fd8554 | 287 | l2_offset = qcow2_alloc_clusters(bs, s->l2_size * l2_entry_size(s)); |
5d757b56 | 288 | if (l2_offset < 0) { |
be0b742e HR |
289 | ret = l2_offset; |
290 | goto fail; | |
5d757b56 | 291 | } |
29c1a730 | 292 | |
c1c43990 AG |
293 | /* The offset must fit in the offset field of the L1 table entry */ |
294 | assert((l2_offset & L1E_OFFSET_MASK) == l2_offset); | |
295 | ||
98839750 AG |
296 | /* If we're allocating the table at offset 0 then something is wrong */ |
297 | if (l2_offset == 0) { | |
298 | qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid " | |
299 | "allocation of L2 table at offset 0"); | |
300 | ret = -EIO; | |
301 | goto fail; | |
302 | } | |
303 | ||
29c1a730 KW |
304 | ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
305 | if (ret < 0) { | |
306 | goto fail; | |
307 | } | |
45aba42f | 308 | |
45aba42f KW |
309 | /* allocate a new entry in the l2 cache */ |
310 | ||
c8fd8554 | 311 | slice_size2 = s->l2_slice_size * l2_entry_size(s); |
3861946a AG |
312 | n_slices = s->cluster_size / slice_size2; |
313 | ||
3cce16f4 | 314 | trace_qcow2_l2_allocate_get_empty(bs, l1_index); |
3861946a | 315 | for (slice = 0; slice < n_slices; slice++) { |
6580bb09 | 316 | ret = qcow2_cache_get_empty(bs, s->l2_table_cache, |
3861946a AG |
317 | l2_offset + slice * slice_size2, |
318 | (void **) &l2_slice); | |
6580bb09 AG |
319 | if (ret < 0) { |
320 | goto fail; | |
321 | } | |
29c1a730 | 322 | |
6580bb09 | 323 | if ((old_l2_offset & L1E_OFFSET_MASK) == 0) { |
3861946a AG |
324 | /* if there was no old l2 table, clear the new slice */ |
325 | memset(l2_slice, 0, slice_size2); | |
6580bb09 | 326 | } else { |
3861946a AG |
327 | uint64_t *old_slice; |
328 | uint64_t old_l2_slice_offset = | |
329 | (old_l2_offset & L1E_OFFSET_MASK) + slice * slice_size2; | |
29c1a730 | 330 | |
3861946a | 331 | /* if there was an old l2 table, read a slice from the disk */ |
6580bb09 | 332 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ); |
3861946a AG |
333 | ret = qcow2_cache_get(bs, s->l2_table_cache, old_l2_slice_offset, |
334 | (void **) &old_slice); | |
6580bb09 AG |
335 | if (ret < 0) { |
336 | goto fail; | |
337 | } | |
338 | ||
3861946a | 339 | memcpy(l2_slice, old_slice, slice_size2); |
6580bb09 | 340 | |
3861946a | 341 | qcow2_cache_put(s->l2_table_cache, (void **) &old_slice); |
29c1a730 KW |
342 | } |
343 | ||
3861946a | 344 | /* write the l2 slice to the file */ |
6580bb09 | 345 | BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE); |
29c1a730 | 346 | |
6580bb09 | 347 | trace_qcow2_l2_allocate_write_l2(bs, l1_index); |
3861946a AG |
348 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
349 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
45aba42f | 350 | } |
29c1a730 | 351 | |
29c1a730 | 352 | ret = qcow2_cache_flush(bs, s->l2_table_cache); |
c46e1167 | 353 | if (ret < 0) { |
175e1152 KW |
354 | goto fail; |
355 | } | |
356 | ||
357 | /* update the L1 entry */ | |
3cce16f4 | 358 | trace_qcow2_l2_allocate_write_l1(bs, l1_index); |
175e1152 | 359 | s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED; |
e23e400e | 360 | ret = qcow2_write_l1_entry(bs, l1_index); |
175e1152 KW |
361 | if (ret < 0) { |
362 | goto fail; | |
c46e1167 | 363 | } |
45aba42f | 364 | |
3cce16f4 | 365 | trace_qcow2_l2_allocate_done(bs, l1_index, 0); |
c46e1167 | 366 | return 0; |
175e1152 KW |
367 | |
368 | fail: | |
3cce16f4 | 369 | trace_qcow2_l2_allocate_done(bs, l1_index, ret); |
3861946a AG |
370 | if (l2_slice != NULL) { |
371 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
8585afd8 | 372 | } |
68dba0bf | 373 | s->l1_table[l1_index] = old_l2_offset; |
e3b21ef9 | 374 | if (l2_offset > 0) { |
c8fd8554 | 375 | qcow2_free_clusters(bs, l2_offset, s->l2_size * l2_entry_size(s), |
e3b21ef9 HR |
376 | QCOW2_DISCARD_ALWAYS); |
377 | } | |
175e1152 | 378 | return ret; |
45aba42f KW |
379 | } |
380 | ||
70d1cbae AG |
381 | /* |
382 | * For a given L2 entry, count the number of contiguous subclusters of | |
383 | * the same type starting from @sc_from. Compressed clusters are | |
384 | * treated as if they were divided into subclusters of size | |
385 | * s->subcluster_size. | |
386 | * | |
387 | * Return the number of contiguous subclusters and set @type to the | |
388 | * subcluster type. | |
389 | * | |
390 | * If the L2 entry is invalid return -errno and set @type to | |
391 | * QCOW2_SUBCLUSTER_INVALID. | |
392 | */ | |
70d1cbae AG |
393 | static int qcow2_get_subcluster_range_type(BlockDriverState *bs, |
394 | uint64_t l2_entry, | |
395 | uint64_t l2_bitmap, | |
396 | unsigned sc_from, | |
397 | QCow2SubclusterType *type) | |
398 | { | |
399 | BDRVQcow2State *s = bs->opaque; | |
400 | uint32_t val; | |
401 | ||
402 | *type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc_from); | |
403 | ||
404 | if (*type == QCOW2_SUBCLUSTER_INVALID) { | |
405 | return -EINVAL; | |
406 | } else if (!has_subclusters(s) || *type == QCOW2_SUBCLUSTER_COMPRESSED) { | |
407 | return s->subclusters_per_cluster - sc_from; | |
408 | } | |
409 | ||
410 | switch (*type) { | |
411 | case QCOW2_SUBCLUSTER_NORMAL: | |
412 | val = l2_bitmap | QCOW_OFLAG_SUB_ALLOC_RANGE(0, sc_from); | |
413 | return cto32(val) - sc_from; | |
414 | ||
415 | case QCOW2_SUBCLUSTER_ZERO_PLAIN: | |
416 | case QCOW2_SUBCLUSTER_ZERO_ALLOC: | |
417 | val = (l2_bitmap | QCOW_OFLAG_SUB_ZERO_RANGE(0, sc_from)) >> 32; | |
418 | return cto32(val) - sc_from; | |
419 | ||
420 | case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: | |
421 | case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: | |
422 | val = ((l2_bitmap >> 32) | l2_bitmap) | |
423 | & ~QCOW_OFLAG_SUB_ALLOC_RANGE(0, sc_from); | |
424 | return ctz32(val) - sc_from; | |
425 | ||
426 | default: | |
427 | g_assert_not_reached(); | |
428 | } | |
429 | } | |
430 | ||
2bfcc4a0 | 431 | /* |
3f9c6b3b AG |
432 | * Return the number of contiguous subclusters of the exact same type |
433 | * in a given L2 slice, starting from cluster @l2_index, subcluster | |
434 | * @sc_index. Allocated subclusters are required to be contiguous in | |
435 | * the image file. | |
436 | * At most @nb_clusters are checked (note that this means clusters, | |
437 | * not subclusters). | |
438 | * Compressed clusters are always processed one by one but for the | |
439 | * purpose of this count they are treated as if they were divided into | |
440 | * subclusters of size s->subcluster_size. | |
441 | * On failure return -errno and update @l2_index to point to the | |
442 | * invalid entry. | |
2bfcc4a0 | 443 | */ |
3f9c6b3b AG |
444 | static int count_contiguous_subclusters(BlockDriverState *bs, int nb_clusters, |
445 | unsigned sc_index, uint64_t *l2_slice, | |
446 | unsigned *l2_index) | |
45aba42f | 447 | { |
12c6aebe | 448 | BDRVQcow2State *s = bs->opaque; |
3f9c6b3b AG |
449 | int i, count = 0; |
450 | bool check_offset = false; | |
451 | uint64_t expected_offset = 0; | |
452 | QCow2SubclusterType expected_type = QCOW2_SUBCLUSTER_NORMAL, type; | |
45aba42f | 453 | |
3f9c6b3b | 454 | assert(*l2_index + nb_clusters <= s->l2_slice_size); |
15684a47 | 455 | |
61653008 | 456 | for (i = 0; i < nb_clusters; i++) { |
3f9c6b3b AG |
457 | unsigned first_sc = (i == 0) ? sc_index : 0; |
458 | uint64_t l2_entry = get_l2_entry(s, l2_slice, *l2_index + i); | |
459 | uint64_t l2_bitmap = get_l2_bitmap(s, l2_slice, *l2_index + i); | |
460 | int ret = qcow2_get_subcluster_range_type(bs, l2_entry, l2_bitmap, | |
461 | first_sc, &type); | |
462 | if (ret < 0) { | |
463 | *l2_index += i; /* Point to the invalid entry */ | |
464 | return -EIO; | |
465 | } | |
466 | if (i == 0) { | |
467 | if (type == QCOW2_SUBCLUSTER_COMPRESSED) { | |
468 | /* Compressed clusters are always processed one by one */ | |
469 | return ret; | |
470 | } | |
471 | expected_type = type; | |
472 | expected_offset = l2_entry & L2E_OFFSET_MASK; | |
473 | check_offset = (type == QCOW2_SUBCLUSTER_NORMAL || | |
474 | type == QCOW2_SUBCLUSTER_ZERO_ALLOC || | |
475 | type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC); | |
476 | } else if (type != expected_type) { | |
45aba42f | 477 | break; |
3f9c6b3b AG |
478 | } else if (check_offset) { |
479 | expected_offset += s->cluster_size; | |
480 | if (expected_offset != (l2_entry & L2E_OFFSET_MASK)) { | |
481 | break; | |
482 | } | |
2bfcc4a0 | 483 | } |
3f9c6b3b AG |
484 | count += ret; |
485 | /* Stop if there are type changes before the end of the cluster */ | |
486 | if (first_sc + ret < s->subclusters_per_cluster) { | |
2bfcc4a0 KW |
487 | break; |
488 | } | |
489 | } | |
45aba42f | 490 | |
3f9c6b3b | 491 | return count; |
45aba42f KW |
492 | } |
493 | ||
7b1fb72e KW |
494 | static int coroutine_fn GRAPH_RDLOCK |
495 | do_perform_cow_read(BlockDriverState *bs, uint64_t src_cluster_offset, | |
496 | unsigned offset_in_cluster, QEMUIOVector *qiov) | |
45aba42f | 497 | { |
aaa4d20b | 498 | int ret; |
1b9f1491 | 499 | |
86b862c4 | 500 | if (qiov->size == 0) { |
99450c6f AG |
501 | return 0; |
502 | } | |
503 | ||
17362398 | 504 | BLKDBG_CO_EVENT(bs->file, BLKDBG_COW_READ); |
aef4acb6 | 505 | |
dba28555 | 506 | if (!bs->drv) { |
672f0f2c | 507 | return -ENOMEDIUM; |
dba28555 HR |
508 | } |
509 | ||
f7ef38dd VSO |
510 | /* |
511 | * We never deal with requests that don't satisfy | |
512 | * bdrv_check_qiov_request(), and aligning requests to clusters never | |
513 | * breaks this condition. So, do some assertions before calling | |
514 | * bs->drv->bdrv_co_preadv_part() which has int64_t arguments. | |
515 | */ | |
516 | assert(src_cluster_offset <= INT64_MAX); | |
517 | assert(src_cluster_offset + offset_in_cluster <= INT64_MAX); | |
e7e588d4 HR |
518 | /* Cast qiov->size to uint64_t to silence a compiler warning on -m32 */ |
519 | assert((uint64_t)qiov->size <= INT64_MAX); | |
f7ef38dd VSO |
520 | bdrv_check_qiov_request(src_cluster_offset + offset_in_cluster, qiov->size, |
521 | qiov, 0, &error_abort); | |
522 | /* | |
523 | * Call .bdrv_co_readv() directly instead of using the public block-layer | |
aef4acb6 SH |
524 | * interface. This avoids double I/O throttling and request tracking, |
525 | * which can lead to deadlock when block layer copy-on-read is enabled. | |
526 | */ | |
df893d25 VSO |
527 | ret = bs->drv->bdrv_co_preadv_part(bs, |
528 | src_cluster_offset + offset_in_cluster, | |
529 | qiov->size, qiov, 0, 0); | |
1b9f1491 | 530 | if (ret < 0) { |
672f0f2c | 531 | return ret; |
1b9f1491 KW |
532 | } |
533 | ||
672f0f2c AG |
534 | return 0; |
535 | } | |
536 | ||
b9b10c35 KW |
537 | static int coroutine_fn GRAPH_RDLOCK |
538 | do_perform_cow_write(BlockDriverState *bs, uint64_t cluster_offset, | |
539 | unsigned offset_in_cluster, QEMUIOVector *qiov) | |
672f0f2c | 540 | { |
966b000f | 541 | BDRVQcow2State *s = bs->opaque; |
672f0f2c AG |
542 | int ret; |
543 | ||
86b862c4 | 544 | if (qiov->size == 0) { |
672f0f2c AG |
545 | return 0; |
546 | } | |
547 | ||
231bb267 | 548 | ret = qcow2_pre_write_overlap_check(bs, 0, |
966b000f | 549 | cluster_offset + offset_in_cluster, qiov->size, true); |
cf93980e | 550 | if (ret < 0) { |
672f0f2c | 551 | return ret; |
cf93980e HR |
552 | } |
553 | ||
17362398 | 554 | BLKDBG_CO_EVENT(bs->file, BLKDBG_COW_WRITE); |
966b000f | 555 | ret = bdrv_co_pwritev(s->data_file, cluster_offset + offset_in_cluster, |
86b862c4 | 556 | qiov->size, qiov, 0); |
1b9f1491 | 557 | if (ret < 0) { |
672f0f2c | 558 | return ret; |
1b9f1491 KW |
559 | } |
560 | ||
672f0f2c | 561 | return 0; |
45aba42f KW |
562 | } |
563 | ||
564 | ||
565 | /* | |
388e5816 | 566 | * get_host_offset |
45aba42f | 567 | * |
388e5816 AG |
568 | * For a given offset of the virtual disk find the equivalent host |
569 | * offset in the qcow2 file and store it in *host_offset. Neither | |
570 | * offset needs to be aligned to a cluster boundary. | |
571 | * | |
572 | * If the cluster is unallocated then *host_offset will be 0. | |
9a3978a4 | 573 | * If the cluster is compressed then *host_offset will contain the l2 entry. |
45aba42f | 574 | * |
ecfe1863 KW |
575 | * On entry, *bytes is the maximum number of contiguous bytes starting at |
576 | * offset that we are interested in. | |
45aba42f | 577 | * |
ecfe1863 | 578 | * On exit, *bytes is the number of bytes starting at offset that have the same |
10dabdc5 AG |
579 | * subcluster type and (if applicable) are stored contiguously in the image |
580 | * file. The subcluster type is stored in *subcluster_type. | |
581 | * Compressed clusters are always processed one by one. | |
45aba42f | 582 | * |
ca4a0bb8 | 583 | * Returns 0 on success, -errno in error cases. |
45aba42f | 584 | */ |
388e5816 | 585 | int qcow2_get_host_offset(BlockDriverState *bs, uint64_t offset, |
ca4a0bb8 | 586 | unsigned int *bytes, uint64_t *host_offset, |
10dabdc5 | 587 | QCow2SubclusterType *subcluster_type) |
45aba42f | 588 | { |
ff99129a | 589 | BDRVQcow2State *s = bs->opaque; |
3f9c6b3b AG |
590 | unsigned int l2_index, sc_index; |
591 | uint64_t l1_index, l2_offset, *l2_slice, l2_entry, l2_bitmap; | |
592 | int sc; | |
c834cba9 HR |
593 | unsigned int offset_in_cluster; |
594 | uint64_t bytes_available, bytes_needed, nb_clusters; | |
3f9c6b3b | 595 | QCow2SubclusterType type; |
55c17e98 | 596 | int ret; |
45aba42f | 597 | |
b2f65d6b | 598 | offset_in_cluster = offset_into_cluster(s, offset); |
ecfe1863 | 599 | bytes_needed = (uint64_t) *bytes + offset_in_cluster; |
45aba42f | 600 | |
b2f65d6b | 601 | /* compute how many bytes there are between the start of the cluster |
fd630039 AG |
602 | * containing offset and the end of the l2 slice that contains |
603 | * the entry pointing to it */ | |
604 | bytes_available = | |
605 | ((uint64_t) (s->l2_slice_size - offset_to_l2_slice_index(s, offset))) | |
606 | << s->cluster_bits; | |
45aba42f | 607 | |
b2f65d6b KW |
608 | if (bytes_needed > bytes_available) { |
609 | bytes_needed = bytes_available; | |
45aba42f KW |
610 | } |
611 | ||
388e5816 | 612 | *host_offset = 0; |
45aba42f | 613 | |
b6af0975 | 614 | /* seek to the l2 offset in the l1 table */ |
45aba42f | 615 | |
05b5b6ee | 616 | l1_index = offset_to_l1_index(s, offset); |
68d000a3 | 617 | if (l1_index >= s->l1_size) { |
3f9c6b3b | 618 | type = QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN; |
45aba42f | 619 | goto out; |
68d000a3 | 620 | } |
45aba42f | 621 | |
68d000a3 KW |
622 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
623 | if (!l2_offset) { | |
3f9c6b3b | 624 | type = QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN; |
45aba42f | 625 | goto out; |
68d000a3 | 626 | } |
45aba42f | 627 | |
a97c67ee HR |
628 | if (offset_into_cluster(s, l2_offset)) { |
629 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
630 | " unaligned (L1 index: %#" PRIx64 ")", | |
631 | l2_offset, l1_index); | |
632 | return -EIO; | |
633 | } | |
634 | ||
fd630039 | 635 | /* load the l2 slice in memory */ |
45aba42f | 636 | |
fd630039 | 637 | ret = l2_load(bs, offset, l2_offset, &l2_slice); |
55c17e98 KW |
638 | if (ret < 0) { |
639 | return ret; | |
1c46efaa | 640 | } |
45aba42f KW |
641 | |
642 | /* find the cluster offset for the given disk offset */ | |
643 | ||
fd630039 | 644 | l2_index = offset_to_l2_slice_index(s, offset); |
3f9c6b3b | 645 | sc_index = offset_to_sc_index(s, offset); |
12c6aebe | 646 | l2_entry = get_l2_entry(s, l2_slice, l2_index); |
3f9c6b3b | 647 | l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index); |
b6d36def | 648 | |
b2f65d6b | 649 | nb_clusters = size_to_clusters(s, bytes_needed); |
c834cba9 HR |
650 | /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned |
651 | * integers; the minimum cluster size is 512, so this assertion is always | |
652 | * true */ | |
653 | assert(nb_clusters <= INT_MAX); | |
45aba42f | 654 | |
3f9c6b3b AG |
655 | type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc_index); |
656 | if (s->qcow_version < 3 && (type == QCOW2_SUBCLUSTER_ZERO_PLAIN || | |
657 | type == QCOW2_SUBCLUSTER_ZERO_ALLOC)) { | |
fdfab37d EB |
658 | qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found" |
659 | " in pre-v3 image (L2 offset: %#" PRIx64 | |
660 | ", L2 index: %#x)", l2_offset, l2_index); | |
661 | ret = -EIO; | |
662 | goto fail; | |
663 | } | |
3ef95218 | 664 | switch (type) { |
3f9c6b3b AG |
665 | case QCOW2_SUBCLUSTER_INVALID: |
666 | break; /* This is handled by count_contiguous_subclusters() below */ | |
667 | case QCOW2_SUBCLUSTER_COMPRESSED: | |
966b000f KW |
668 | if (has_data_file(bs)) { |
669 | qcow2_signal_corruption(bs, true, -1, -1, "Compressed cluster " | |
670 | "entry found in image with external data " | |
671 | "file (L2 offset: %#" PRIx64 ", L2 index: " | |
672 | "%#x)", l2_offset, l2_index); | |
673 | ret = -EIO; | |
674 | goto fail; | |
675 | } | |
9a3978a4 | 676 | *host_offset = l2_entry; |
68d000a3 | 677 | break; |
3f9c6b3b AG |
678 | case QCOW2_SUBCLUSTER_ZERO_PLAIN: |
679 | case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: | |
68d000a3 | 680 | break; |
3f9c6b3b AG |
681 | case QCOW2_SUBCLUSTER_ZERO_ALLOC: |
682 | case QCOW2_SUBCLUSTER_NORMAL: | |
683 | case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: { | |
388e5816 AG |
684 | uint64_t host_cluster_offset = l2_entry & L2E_OFFSET_MASK; |
685 | *host_offset = host_cluster_offset + offset_in_cluster; | |
388e5816 | 686 | if (offset_into_cluster(s, host_cluster_offset)) { |
fdfab37d EB |
687 | qcow2_signal_corruption(bs, true, -1, -1, |
688 | "Cluster allocation offset %#" | |
a97c67ee | 689 | PRIx64 " unaligned (L2 offset: %#" PRIx64 |
388e5816 | 690 | ", L2 index: %#x)", host_cluster_offset, |
a97c67ee HR |
691 | l2_offset, l2_index); |
692 | ret = -EIO; | |
693 | goto fail; | |
694 | } | |
388e5816 | 695 | if (has_data_file(bs) && *host_offset != offset) { |
966b000f KW |
696 | qcow2_signal_corruption(bs, true, -1, -1, |
697 | "External data file host cluster offset %#" | |
698 | PRIx64 " does not match guest cluster " | |
699 | "offset: %#" PRIx64 | |
388e5816 | 700 | ", L2 index: %#x)", host_cluster_offset, |
966b000f KW |
701 | offset - offset_in_cluster, l2_index); |
702 | ret = -EIO; | |
703 | goto fail; | |
704 | } | |
68d000a3 | 705 | break; |
388e5816 | 706 | } |
1417d7e4 KW |
707 | default: |
708 | abort(); | |
45aba42f KW |
709 | } |
710 | ||
3f9c6b3b AG |
711 | sc = count_contiguous_subclusters(bs, nb_clusters, sc_index, |
712 | l2_slice, &l2_index); | |
713 | if (sc < 0) { | |
714 | qcow2_signal_corruption(bs, true, -1, -1, "Invalid cluster entry found " | |
715 | " (L2 offset: %#" PRIx64 ", L2 index: %#x)", | |
716 | l2_offset, l2_index); | |
717 | ret = -EIO; | |
718 | goto fail; | |
719 | } | |
fd630039 | 720 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
29c1a730 | 721 | |
3f9c6b3b | 722 | bytes_available = ((int64_t)sc + sc_index) << s->subcluster_bits; |
68d000a3 | 723 | |
45aba42f | 724 | out: |
b2f65d6b KW |
725 | if (bytes_available > bytes_needed) { |
726 | bytes_available = bytes_needed; | |
727 | } | |
45aba42f | 728 | |
c834cba9 HR |
729 | /* bytes_available <= bytes_needed <= *bytes + offset_in_cluster; |
730 | * subtracting offset_in_cluster will therefore definitely yield something | |
731 | * not exceeding UINT_MAX */ | |
732 | assert(bytes_available - offset_in_cluster <= UINT_MAX); | |
ecfe1863 | 733 | *bytes = bytes_available - offset_in_cluster; |
45aba42f | 734 | |
3f9c6b3b | 735 | *subcluster_type = type; |
ca4a0bb8 AG |
736 | |
737 | return 0; | |
a97c67ee HR |
738 | |
739 | fail: | |
fd630039 | 740 | qcow2_cache_put(s->l2_table_cache, (void **)&l2_slice); |
a97c67ee | 741 | return ret; |
45aba42f KW |
742 | } |
743 | ||
744 | /* | |
745 | * get_cluster_table | |
746 | * | |
747 | * for a given disk offset, load (and allocate if needed) | |
c03bfc5b | 748 | * the appropriate slice of its l2 table. |
45aba42f | 749 | * |
c03bfc5b | 750 | * the cluster index in the l2 slice is given to the caller. |
45aba42f | 751 | * |
1e3e8f1a | 752 | * Returns 0 on success, -errno in failure case |
45aba42f | 753 | */ |
45aba42f | 754 | static int get_cluster_table(BlockDriverState *bs, uint64_t offset, |
c03bfc5b | 755 | uint64_t **new_l2_slice, |
45aba42f KW |
756 | int *new_l2_index) |
757 | { | |
ff99129a | 758 | BDRVQcow2State *s = bs->opaque; |
2cf7cfa1 KW |
759 | unsigned int l2_index; |
760 | uint64_t l1_index, l2_offset; | |
c03bfc5b | 761 | uint64_t *l2_slice = NULL; |
80ee15a6 | 762 | int ret; |
45aba42f | 763 | |
b6af0975 | 764 | /* seek to the l2 offset in the l1 table */ |
45aba42f | 765 | |
05b5b6ee | 766 | l1_index = offset_to_l1_index(s, offset); |
45aba42f | 767 | if (l1_index >= s->l1_size) { |
72893756 | 768 | ret = qcow2_grow_l1_table(bs, l1_index + 1, false); |
1e3e8f1a KW |
769 | if (ret < 0) { |
770 | return ret; | |
771 | } | |
45aba42f | 772 | } |
8e37f681 | 773 | |
2cf7cfa1 | 774 | assert(l1_index < s->l1_size); |
8e37f681 | 775 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; |
a97c67ee HR |
776 | if (offset_into_cluster(s, l2_offset)) { |
777 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64 | |
778 | " unaligned (L1 index: %#" PRIx64 ")", | |
779 | l2_offset, l1_index); | |
780 | return -EIO; | |
781 | } | |
45aba42f | 782 | |
05f9ee46 | 783 | if (!(s->l1_table[l1_index] & QCOW_OFLAG_COPIED)) { |
16fde5f2 | 784 | /* First allocate a new L2 table (and do COW if needed) */ |
3861946a | 785 | ret = l2_allocate(bs, l1_index); |
c46e1167 KW |
786 | if (ret < 0) { |
787 | return ret; | |
1e3e8f1a | 788 | } |
16fde5f2 KW |
789 | |
790 | /* Then decrease the refcount of the old table */ | |
791 | if (l2_offset) { | |
c8fd8554 | 792 | qcow2_free_clusters(bs, l2_offset, s->l2_size * l2_entry_size(s), |
6cfcb9b8 | 793 | QCOW2_DISCARD_OTHER); |
16fde5f2 | 794 | } |
3861946a AG |
795 | |
796 | /* Get the offset of the newly-allocated l2 table */ | |
797 | l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; | |
798 | assert(offset_into_cluster(s, l2_offset) == 0); | |
05f9ee46 AG |
799 | } |
800 | ||
c03bfc5b AG |
801 | /* load the l2 slice in memory */ |
802 | ret = l2_load(bs, offset, l2_offset, &l2_slice); | |
05f9ee46 AG |
803 | if (ret < 0) { |
804 | return ret; | |
45aba42f KW |
805 | } |
806 | ||
807 | /* find the cluster offset for the given disk offset */ | |
808 | ||
c03bfc5b | 809 | l2_index = offset_to_l2_slice_index(s, offset); |
45aba42f | 810 | |
c03bfc5b | 811 | *new_l2_slice = l2_slice; |
45aba42f KW |
812 | *new_l2_index = l2_index; |
813 | ||
1e3e8f1a | 814 | return 0; |
45aba42f KW |
815 | } |
816 | ||
817 | /* | |
818 | * alloc_compressed_cluster_offset | |
819 | * | |
77e023ff KW |
820 | * For a given offset on the virtual disk, allocate a new compressed cluster |
821 | * and put the host offset of the cluster into *host_offset. If a cluster is | |
822 | * already allocated at the offset, return an error. | |
45aba42f | 823 | * |
77e023ff | 824 | * Return 0 on success and -errno in error cases |
45aba42f | 825 | */ |
70bacc44 PB |
826 | int coroutine_fn GRAPH_RDLOCK |
827 | qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, uint64_t offset, | |
828 | int compressed_size, uint64_t *host_offset) | |
45aba42f | 829 | { |
ff99129a | 830 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 831 | int l2_index, ret; |
e4e72548 | 832 | uint64_t *l2_slice; |
f4f0d391 | 833 | int64_t cluster_offset; |
45aba42f KW |
834 | int nb_csectors; |
835 | ||
966b000f KW |
836 | if (has_data_file(bs)) { |
837 | return 0; | |
838 | } | |
839 | ||
e4e72548 | 840 | ret = get_cluster_table(bs, offset, &l2_slice, &l2_index); |
1e3e8f1a | 841 | if (ret < 0) { |
77e023ff | 842 | return ret; |
1e3e8f1a | 843 | } |
45aba42f | 844 | |
b0b6862e KW |
845 | /* Compression can't overwrite anything. Fail if the cluster was already |
846 | * allocated. */ | |
12c6aebe | 847 | cluster_offset = get_l2_entry(s, l2_slice, l2_index); |
b0b6862e | 848 | if (cluster_offset & L2E_OFFSET_MASK) { |
e4e72548 | 849 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
77e023ff | 850 | return -EIO; |
8f1efd00 | 851 | } |
45aba42f | 852 | |
ed6ccf0f | 853 | cluster_offset = qcow2_alloc_bytes(bs, compressed_size); |
5d757b56 | 854 | if (cluster_offset < 0) { |
e4e72548 | 855 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
77e023ff | 856 | return cluster_offset; |
5d757b56 KW |
857 | } |
858 | ||
b6c24694 AG |
859 | nb_csectors = |
860 | (cluster_offset + compressed_size - 1) / QCOW2_COMPRESSED_SECTOR_SIZE - | |
861 | (cluster_offset / QCOW2_COMPRESSED_SECTOR_SIZE); | |
45aba42f | 862 | |
3a75a870 AG |
863 | /* The offset and size must fit in their fields of the L2 table entry */ |
864 | assert((cluster_offset & s->cluster_offset_mask) == cluster_offset); | |
865 | assert((nb_csectors & s->csize_mask) == nb_csectors); | |
866 | ||
45aba42f KW |
867 | cluster_offset |= QCOW_OFLAG_COMPRESSED | |
868 | ((uint64_t)nb_csectors << s->csize_shift); | |
869 | ||
870 | /* update L2 table */ | |
871 | ||
872 | /* compressed clusters never have the copied flag */ | |
873 | ||
17362398 | 874 | BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED); |
e4e72548 | 875 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
12c6aebe | 876 | set_l2_entry(s, l2_slice, l2_index, cluster_offset); |
ff4cdec7 AG |
877 | if (has_subclusters(s)) { |
878 | set_l2_bitmap(s, l2_slice, l2_index, 0); | |
879 | } | |
e4e72548 | 880 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
4c1612d9 | 881 | |
77e023ff KW |
882 | *host_offset = cluster_offset & s->cluster_offset_mask; |
883 | return 0; | |
4c1612d9 KW |
884 | } |
885 | ||
7b1fb72e KW |
886 | static int coroutine_fn GRAPH_RDLOCK |
887 | perform_cow(BlockDriverState *bs, QCowL2Meta *m) | |
593fb83c | 888 | { |
ff99129a | 889 | BDRVQcow2State *s = bs->opaque; |
99450c6f AG |
890 | Qcow2COWRegion *start = &m->cow_start; |
891 | Qcow2COWRegion *end = &m->cow_end; | |
672f0f2c | 892 | unsigned buffer_size; |
b3cf1c7c AG |
893 | unsigned data_bytes = end->offset - (start->offset + start->nb_bytes); |
894 | bool merge_reads; | |
672f0f2c | 895 | uint8_t *start_buffer, *end_buffer; |
86b862c4 | 896 | QEMUIOVector qiov; |
593fb83c KW |
897 | int ret; |
898 | ||
672f0f2c | 899 | assert(start->nb_bytes <= UINT_MAX - end->nb_bytes); |
b3cf1c7c AG |
900 | assert(start->nb_bytes + end->nb_bytes <= UINT_MAX - data_bytes); |
901 | assert(start->offset + start->nb_bytes <= end->offset); | |
672f0f2c | 902 | |
c8bb23cb | 903 | if ((start->nb_bytes == 0 && end->nb_bytes == 0) || m->skip_cow) { |
593fb83c KW |
904 | return 0; |
905 | } | |
906 | ||
b3cf1c7c AG |
907 | /* If we have to read both the start and end COW regions and the |
908 | * middle region is not too large then perform just one read | |
909 | * operation */ | |
910 | merge_reads = start->nb_bytes && end->nb_bytes && data_bytes <= 16384; | |
911 | if (merge_reads) { | |
912 | buffer_size = start->nb_bytes + data_bytes + end->nb_bytes; | |
913 | } else { | |
914 | /* If we have to do two reads, add some padding in the middle | |
915 | * if necessary to make sure that the end region is optimally | |
916 | * aligned. */ | |
917 | size_t align = bdrv_opt_mem_align(bs); | |
918 | assert(align > 0 && align <= UINT_MAX); | |
919 | assert(QEMU_ALIGN_UP(start->nb_bytes, align) <= | |
920 | UINT_MAX - end->nb_bytes); | |
921 | buffer_size = QEMU_ALIGN_UP(start->nb_bytes, align) + end->nb_bytes; | |
922 | } | |
923 | ||
924 | /* Reserve a buffer large enough to store all the data that we're | |
925 | * going to read */ | |
672f0f2c AG |
926 | start_buffer = qemu_try_blockalign(bs, buffer_size); |
927 | if (start_buffer == NULL) { | |
928 | return -ENOMEM; | |
929 | } | |
930 | /* The part of the buffer where the end region is located */ | |
931 | end_buffer = start_buffer + buffer_size - end->nb_bytes; | |
932 | ||
5396234b VSO |
933 | qemu_iovec_init(&qiov, 2 + (m->data_qiov ? |
934 | qemu_iovec_subvec_niov(m->data_qiov, | |
935 | m->data_qiov_offset, | |
936 | data_bytes) | |
937 | : 0)); | |
86b862c4 | 938 | |
593fb83c | 939 | qemu_co_mutex_unlock(&s->lock); |
b3cf1c7c AG |
940 | /* First we read the existing data from both COW regions. We |
941 | * either read the whole region in one go, or the start and end | |
942 | * regions separately. */ | |
943 | if (merge_reads) { | |
86b862c4 AG |
944 | qemu_iovec_add(&qiov, start_buffer, buffer_size); |
945 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c | 946 | } else { |
86b862c4 AG |
947 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); |
948 | ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov); | |
b3cf1c7c AG |
949 | if (ret < 0) { |
950 | goto fail; | |
951 | } | |
672f0f2c | 952 | |
86b862c4 AG |
953 | qemu_iovec_reset(&qiov); |
954 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
955 | ret = do_perform_cow_read(bs, m->offset, end->offset, &qiov); | |
b3cf1c7c | 956 | } |
593fb83c | 957 | if (ret < 0) { |
99450c6f | 958 | goto fail; |
593fb83c KW |
959 | } |
960 | ||
672f0f2c AG |
961 | /* Encrypt the data if necessary before writing it */ |
962 | if (bs->encrypted) { | |
603fbd07 ML |
963 | ret = qcow2_co_encrypt(bs, |
964 | m->alloc_offset + start->offset, | |
965 | m->offset + start->offset, | |
966 | start_buffer, start->nb_bytes); | |
967 | if (ret < 0) { | |
968 | goto fail; | |
969 | } | |
970 | ||
971 | ret = qcow2_co_encrypt(bs, | |
972 | m->alloc_offset + end->offset, | |
973 | m->offset + end->offset, | |
974 | end_buffer, end->nb_bytes); | |
975 | if (ret < 0) { | |
672f0f2c AG |
976 | goto fail; |
977 | } | |
978 | } | |
979 | ||
ee22a9d8 AG |
980 | /* And now we can write everything. If we have the guest data we |
981 | * can write everything in one single operation */ | |
982 | if (m->data_qiov) { | |
983 | qemu_iovec_reset(&qiov); | |
984 | if (start->nb_bytes) { | |
985 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
986 | } | |
5396234b | 987 | qemu_iovec_concat(&qiov, m->data_qiov, m->data_qiov_offset, data_bytes); |
ee22a9d8 AG |
988 | if (end->nb_bytes) { |
989 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
990 | } | |
991 | /* NOTE: we have a write_aio blkdebug event here followed by | |
992 | * a cow_write one in do_perform_cow_write(), but there's only | |
993 | * one single I/O operation */ | |
17362398 | 994 | BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_AIO); |
ee22a9d8 AG |
995 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); |
996 | } else { | |
997 | /* If there's no guest data then write both COW regions separately */ | |
998 | qemu_iovec_reset(&qiov); | |
999 | qemu_iovec_add(&qiov, start_buffer, start->nb_bytes); | |
1000 | ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov); | |
1001 | if (ret < 0) { | |
1002 | goto fail; | |
1003 | } | |
1004 | ||
1005 | qemu_iovec_reset(&qiov); | |
1006 | qemu_iovec_add(&qiov, end_buffer, end->nb_bytes); | |
1007 | ret = do_perform_cow_write(bs, m->alloc_offset, end->offset, &qiov); | |
672f0f2c | 1008 | } |
99450c6f AG |
1009 | |
1010 | fail: | |
1011 | qemu_co_mutex_lock(&s->lock); | |
1012 | ||
593fb83c KW |
1013 | /* |
1014 | * Before we update the L2 table to actually point to the new cluster, we | |
1015 | * need to be sure that the refcounts have been increased and COW was | |
1016 | * handled. | |
1017 | */ | |
99450c6f AG |
1018 | if (ret == 0) { |
1019 | qcow2_cache_depends_on_flush(s->l2_table_cache); | |
1020 | } | |
593fb83c | 1021 | |
672f0f2c | 1022 | qemu_vfree(start_buffer); |
86b862c4 | 1023 | qemu_iovec_destroy(&qiov); |
99450c6f | 1024 | return ret; |
593fb83c KW |
1025 | } |
1026 | ||
050ed2e7 PB |
1027 | int coroutine_fn qcow2_alloc_cluster_link_l2(BlockDriverState *bs, |
1028 | QCowL2Meta *m) | |
45aba42f | 1029 | { |
ff99129a | 1030 | BDRVQcow2State *s = bs->opaque; |
45aba42f | 1031 | int i, j = 0, l2_index, ret; |
a002c0b0 | 1032 | uint64_t *old_cluster, *l2_slice; |
250196f1 | 1033 | uint64_t cluster_offset = m->alloc_offset; |
45aba42f | 1034 | |
3cce16f4 | 1035 | trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters); |
f50f88b9 | 1036 | assert(m->nb_clusters > 0); |
45aba42f | 1037 | |
5839e53b | 1038 | old_cluster = g_try_new(uint64_t, m->nb_clusters); |
de82815d KW |
1039 | if (old_cluster == NULL) { |
1040 | ret = -ENOMEM; | |
1041 | goto err; | |
1042 | } | |
45aba42f KW |
1043 | |
1044 | /* copy content of unmodified sectors */ | |
99450c6f | 1045 | ret = perform_cow(bs, m); |
593fb83c KW |
1046 | if (ret < 0) { |
1047 | goto err; | |
29c1a730 KW |
1048 | } |
1049 | ||
593fb83c | 1050 | /* Update L2 table. */ |
74c4510a | 1051 | if (s->use_lazy_refcounts) { |
280d3735 KW |
1052 | qcow2_mark_dirty(bs); |
1053 | } | |
bfe8043e SH |
1054 | if (qcow2_need_accurate_refcounts(s)) { |
1055 | qcow2_cache_set_dependency(bs, s->l2_table_cache, | |
1056 | s->refcount_block_cache); | |
1057 | } | |
280d3735 | 1058 | |
a002c0b0 | 1059 | ret = get_cluster_table(bs, m->offset, &l2_slice, &l2_index); |
1e3e8f1a | 1060 | if (ret < 0) { |
45aba42f | 1061 | goto err; |
1e3e8f1a | 1062 | } |
a002c0b0 | 1063 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
45aba42f | 1064 | |
a002c0b0 | 1065 | assert(l2_index + m->nb_clusters <= s->l2_slice_size); |
3441ad4b AG |
1066 | assert(m->cow_end.offset + m->cow_end.nb_bytes <= |
1067 | m->nb_clusters << s->cluster_bits); | |
45aba42f | 1068 | for (i = 0; i < m->nb_clusters; i++) { |
348fcc4f | 1069 | uint64_t offset = cluster_offset + ((uint64_t)i << s->cluster_bits); |
45aba42f | 1070 | /* if two concurrent writes happen to the same unallocated cluster |
aaa4d20b KW |
1071 | * each write allocates separate cluster and writes data concurrently. |
1072 | * The first one to complete updates l2 table with pointer to its | |
1073 | * cluster the second one has to do RMW (which is done above by | |
1074 | * perform_cow()), update l2 table with its cluster pointer and free | |
1075 | * old cluster. This is what this loop does */ | |
12c6aebe AG |
1076 | if (get_l2_entry(s, l2_slice, l2_index + i) != 0) { |
1077 | old_cluster[j++] = get_l2_entry(s, l2_slice, l2_index + i); | |
aaa4d20b | 1078 | } |
45aba42f | 1079 | |
3a75a870 AG |
1080 | /* The offset must fit in the offset field of the L2 table entry */ |
1081 | assert((offset & L2E_OFFSET_MASK) == offset); | |
1082 | ||
12c6aebe | 1083 | set_l2_entry(s, l2_slice, l2_index + i, offset | QCOW_OFLAG_COPIED); |
aca00cd9 AG |
1084 | |
1085 | /* Update bitmap with the subclusters that were just written */ | |
40dee943 | 1086 | if (has_subclusters(s) && !m->prealloc) { |
aca00cd9 AG |
1087 | uint64_t l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i); |
1088 | unsigned written_from = m->cow_start.offset; | |
3441ad4b | 1089 | unsigned written_to = m->cow_end.offset + m->cow_end.nb_bytes; |
aca00cd9 AG |
1090 | int first_sc, last_sc; |
1091 | /* Narrow written_from and written_to down to the current cluster */ | |
1092 | written_from = MAX(written_from, i << s->cluster_bits); | |
1093 | written_to = MIN(written_to, (i + 1) << s->cluster_bits); | |
1094 | assert(written_from < written_to); | |
1095 | first_sc = offset_to_sc_index(s, written_from); | |
1096 | last_sc = offset_to_sc_index(s, written_to - 1); | |
1097 | l2_bitmap |= QCOW_OFLAG_SUB_ALLOC_RANGE(first_sc, last_sc + 1); | |
1098 | l2_bitmap &= ~QCOW_OFLAG_SUB_ZERO_RANGE(first_sc, last_sc + 1); | |
1099 | set_l2_bitmap(s, l2_slice, l2_index + i, l2_bitmap); | |
1100 | } | |
45aba42f KW |
1101 | } |
1102 | ||
9f8e668e | 1103 | |
a002c0b0 | 1104 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
45aba42f | 1105 | |
7ec5e6a4 KW |
1106 | /* |
1107 | * If this was a COW, we need to decrease the refcount of the old cluster. | |
6cfcb9b8 KW |
1108 | * |
1109 | * Don't discard clusters that reach a refcount of 0 (e.g. compressed | |
1110 | * clusters), the next write will reuse them anyway. | |
7ec5e6a4 | 1111 | */ |
564a6b69 | 1112 | if (!m->keep_old_clusters && j != 0) { |
7ec5e6a4 | 1113 | for (i = 0; i < j; i++) { |
3fec237f | 1114 | qcow2_free_any_cluster(bs, old_cluster[i], QCOW2_DISCARD_NEVER); |
7ec5e6a4 KW |
1115 | } |
1116 | } | |
45aba42f KW |
1117 | |
1118 | ret = 0; | |
1119 | err: | |
7267c094 | 1120 | g_free(old_cluster); |
45aba42f KW |
1121 | return ret; |
1122 | } | |
1123 | ||
8b24cd14 KW |
1124 | /** |
1125 | * Frees the allocated clusters because the request failed and they won't | |
1126 | * actually be linked. | |
1127 | */ | |
a39bae4e | 1128 | void coroutine_fn qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m) |
8b24cd14 KW |
1129 | { |
1130 | BDRVQcow2State *s = bs->opaque; | |
3ede935f | 1131 | if (!has_data_file(bs) && !m->keep_old_clusters) { |
c3b6658c KW |
1132 | qcow2_free_clusters(bs, m->alloc_offset, |
1133 | m->nb_clusters << s->cluster_bits, | |
1134 | QCOW2_DISCARD_NEVER); | |
1135 | } | |
8b24cd14 KW |
1136 | } |
1137 | ||
8f91d690 AG |
1138 | /* |
1139 | * For a given write request, create a new QCowL2Meta structure, add | |
57538c86 AG |
1140 | * it to @m and the BDRVQcow2State.cluster_allocs list. If the write |
1141 | * request does not need copy-on-write or changes to the L2 metadata | |
1142 | * then this function does nothing. | |
8f91d690 AG |
1143 | * |
1144 | * @host_cluster_offset points to the beginning of the first cluster. | |
1145 | * | |
1146 | * @guest_offset and @bytes indicate the offset and length of the | |
1147 | * request. | |
1148 | * | |
57538c86 AG |
1149 | * @l2_slice contains the L2 entries of all clusters involved in this |
1150 | * write request. | |
1151 | * | |
8f91d690 AG |
1152 | * If @keep_old is true it means that the clusters were already |
1153 | * allocated and will be overwritten. If false then the clusters are | |
1154 | * new and we have to decrease the reference count of the old ones. | |
d53ec3d8 AG |
1155 | * |
1156 | * Returns 0 on success, -errno on failure. | |
8f91d690 | 1157 | */ |
a39bae4e PB |
1158 | static int coroutine_fn calculate_l2_meta(BlockDriverState *bs, |
1159 | uint64_t host_cluster_offset, | |
1160 | uint64_t guest_offset, unsigned bytes, | |
1161 | uint64_t *l2_slice, QCowL2Meta **m, | |
1162 | bool keep_old) | |
8f91d690 AG |
1163 | { |
1164 | BDRVQcow2State *s = bs->opaque; | |
d53ec3d8 AG |
1165 | int sc_index, l2_index = offset_to_l2_slice_index(s, guest_offset); |
1166 | uint64_t l2_entry, l2_bitmap; | |
57538c86 | 1167 | unsigned cow_start_from, cow_end_to; |
8f91d690 AG |
1168 | unsigned cow_start_to = offset_into_cluster(s, guest_offset); |
1169 | unsigned cow_end_from = cow_start_to + bytes; | |
8f91d690 AG |
1170 | unsigned nb_clusters = size_to_clusters(s, cow_end_from); |
1171 | QCowL2Meta *old_m = *m; | |
d53ec3d8 AG |
1172 | QCow2SubclusterType type; |
1173 | int i; | |
1174 | bool skip_cow = keep_old; | |
57538c86 AG |
1175 | |
1176 | assert(nb_clusters <= s->l2_slice_size - l2_index); | |
1177 | ||
d53ec3d8 AG |
1178 | /* Check the type of all affected subclusters */ |
1179 | for (i = 0; i < nb_clusters; i++) { | |
1180 | l2_entry = get_l2_entry(s, l2_slice, l2_index + i); | |
1181 | l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i); | |
1182 | if (skip_cow) { | |
1183 | unsigned write_from = MAX(cow_start_to, i << s->cluster_bits); | |
1184 | unsigned write_to = MIN(cow_end_from, (i + 1) << s->cluster_bits); | |
1185 | int first_sc = offset_to_sc_index(s, write_from); | |
1186 | int last_sc = offset_to_sc_index(s, write_to - 1); | |
1187 | int cnt = qcow2_get_subcluster_range_type(bs, l2_entry, l2_bitmap, | |
1188 | first_sc, &type); | |
1189 | /* Is any of the subclusters of type != QCOW2_SUBCLUSTER_NORMAL ? */ | |
1190 | if (type != QCOW2_SUBCLUSTER_NORMAL || first_sc + cnt <= last_sc) { | |
1191 | skip_cow = false; | |
57538c86 | 1192 | } |
d53ec3d8 AG |
1193 | } else { |
1194 | /* If we can't skip the cow we can still look for invalid entries */ | |
1195 | type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, 0); | |
57538c86 | 1196 | } |
d53ec3d8 AG |
1197 | if (type == QCOW2_SUBCLUSTER_INVALID) { |
1198 | int l1_index = offset_to_l1_index(s, guest_offset); | |
1199 | uint64_t l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK; | |
1200 | qcow2_signal_corruption(bs, true, -1, -1, "Invalid cluster " | |
1201 | "entry found (L2 offset: %#" PRIx64 | |
1202 | ", L2 index: %#x)", | |
1203 | l2_offset, l2_index + i); | |
1204 | return -EIO; | |
57538c86 AG |
1205 | } |
1206 | } | |
1207 | ||
d53ec3d8 AG |
1208 | if (skip_cow) { |
1209 | return 0; | |
1210 | } | |
1211 | ||
57538c86 | 1212 | /* Get the L2 entry of the first cluster */ |
12c6aebe | 1213 | l2_entry = get_l2_entry(s, l2_slice, l2_index); |
d53ec3d8 AG |
1214 | l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index); |
1215 | sc_index = offset_to_sc_index(s, guest_offset); | |
1216 | type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc_index); | |
1217 | ||
1218 | if (!keep_old) { | |
1219 | switch (type) { | |
1220 | case QCOW2_SUBCLUSTER_COMPRESSED: | |
1221 | cow_start_from = 0; | |
1222 | break; | |
1223 | case QCOW2_SUBCLUSTER_NORMAL: | |
1224 | case QCOW2_SUBCLUSTER_ZERO_ALLOC: | |
1225 | case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: | |
1226 | if (has_subclusters(s)) { | |
1227 | /* Skip all leading zero and unallocated subclusters */ | |
1228 | uint32_t alloc_bitmap = l2_bitmap & QCOW_L2_BITMAP_ALL_ALLOC; | |
1229 | cow_start_from = | |
1230 | MIN(sc_index, ctz32(alloc_bitmap)) << s->subcluster_bits; | |
1231 | } else { | |
1232 | cow_start_from = 0; | |
1233 | } | |
1234 | break; | |
1235 | case QCOW2_SUBCLUSTER_ZERO_PLAIN: | |
1236 | case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: | |
1237 | cow_start_from = sc_index << s->subcluster_bits; | |
1238 | break; | |
1239 | default: | |
1240 | g_assert_not_reached(); | |
1241 | } | |
57538c86 | 1242 | } else { |
d53ec3d8 AG |
1243 | switch (type) { |
1244 | case QCOW2_SUBCLUSTER_NORMAL: | |
1245 | cow_start_from = cow_start_to; | |
1246 | break; | |
1247 | case QCOW2_SUBCLUSTER_ZERO_ALLOC: | |
1248 | case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: | |
1249 | cow_start_from = sc_index << s->subcluster_bits; | |
1250 | break; | |
1251 | default: | |
1252 | g_assert_not_reached(); | |
1253 | } | |
57538c86 AG |
1254 | } |
1255 | ||
1256 | /* Get the L2 entry of the last cluster */ | |
d53ec3d8 AG |
1257 | l2_index += nb_clusters - 1; |
1258 | l2_entry = get_l2_entry(s, l2_slice, l2_index); | |
1259 | l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index); | |
1260 | sc_index = offset_to_sc_index(s, guest_offset + bytes - 1); | |
1261 | type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc_index); | |
1262 | ||
1263 | if (!keep_old) { | |
1264 | switch (type) { | |
1265 | case QCOW2_SUBCLUSTER_COMPRESSED: | |
1266 | cow_end_to = ROUND_UP(cow_end_from, s->cluster_size); | |
1267 | break; | |
1268 | case QCOW2_SUBCLUSTER_NORMAL: | |
1269 | case QCOW2_SUBCLUSTER_ZERO_ALLOC: | |
1270 | case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: | |
1271 | cow_end_to = ROUND_UP(cow_end_from, s->cluster_size); | |
1272 | if (has_subclusters(s)) { | |
1273 | /* Skip all trailing zero and unallocated subclusters */ | |
1274 | uint32_t alloc_bitmap = l2_bitmap & QCOW_L2_BITMAP_ALL_ALLOC; | |
1275 | cow_end_to -= | |
1276 | MIN(s->subclusters_per_cluster - sc_index - 1, | |
1277 | clz32(alloc_bitmap)) << s->subcluster_bits; | |
1278 | } | |
1279 | break; | |
1280 | case QCOW2_SUBCLUSTER_ZERO_PLAIN: | |
1281 | case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN: | |
1282 | cow_end_to = ROUND_UP(cow_end_from, s->subcluster_size); | |
1283 | break; | |
1284 | default: | |
1285 | g_assert_not_reached(); | |
1286 | } | |
57538c86 | 1287 | } else { |
d53ec3d8 AG |
1288 | switch (type) { |
1289 | case QCOW2_SUBCLUSTER_NORMAL: | |
1290 | cow_end_to = cow_end_from; | |
1291 | break; | |
1292 | case QCOW2_SUBCLUSTER_ZERO_ALLOC: | |
1293 | case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: | |
1294 | cow_end_to = ROUND_UP(cow_end_from, s->subcluster_size); | |
1295 | break; | |
1296 | default: | |
1297 | g_assert_not_reached(); | |
1298 | } | |
57538c86 | 1299 | } |
8f91d690 AG |
1300 | |
1301 | *m = g_malloc0(sizeof(**m)); | |
1302 | **m = (QCowL2Meta) { | |
1303 | .next = old_m, | |
1304 | ||
1305 | .alloc_offset = host_cluster_offset, | |
1306 | .offset = start_of_cluster(s, guest_offset), | |
1307 | .nb_clusters = nb_clusters, | |
1308 | ||
1309 | .keep_old_clusters = keep_old, | |
1310 | ||
1311 | .cow_start = { | |
1312 | .offset = cow_start_from, | |
1313 | .nb_bytes = cow_start_to - cow_start_from, | |
1314 | }, | |
1315 | .cow_end = { | |
1316 | .offset = cow_end_from, | |
1317 | .nb_bytes = cow_end_to - cow_end_from, | |
1318 | }, | |
1319 | }; | |
1320 | ||
1321 | qemu_co_queue_init(&(*m)->dependent_requests); | |
1322 | QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); | |
d53ec3d8 AG |
1323 | |
1324 | return 0; | |
8f91d690 AG |
1325 | } |
1326 | ||
57538c86 AG |
1327 | /* |
1328 | * Returns true if writing to the cluster pointed to by @l2_entry | |
1329 | * requires a new allocation (that is, if the cluster is unallocated | |
1330 | * or has refcount > 1 and therefore cannot be written in-place). | |
1331 | */ | |
1332 | static bool cluster_needs_new_alloc(BlockDriverState *bs, uint64_t l2_entry) | |
c1587d87 AG |
1333 | { |
1334 | switch (qcow2_get_cluster_type(bs, l2_entry)) { | |
1335 | case QCOW2_CLUSTER_NORMAL: | |
57538c86 | 1336 | case QCOW2_CLUSTER_ZERO_ALLOC: |
c1587d87 AG |
1337 | if (l2_entry & QCOW_OFLAG_COPIED) { |
1338 | return false; | |
1339 | } | |
b9be6fae | 1340 | /* fallthrough */ |
c1587d87 AG |
1341 | case QCOW2_CLUSTER_UNALLOCATED: |
1342 | case QCOW2_CLUSTER_COMPRESSED: | |
1343 | case QCOW2_CLUSTER_ZERO_PLAIN: | |
c1587d87 AG |
1344 | return true; |
1345 | default: | |
1346 | abort(); | |
1347 | } | |
1348 | } | |
1349 | ||
bf319ece | 1350 | /* |
57538c86 AG |
1351 | * Returns the number of contiguous clusters that can be written to |
1352 | * using one single write request, starting from @l2_index. | |
1353 | * At most @nb_clusters are checked. | |
1354 | * | |
1355 | * If @new_alloc is true this counts clusters that are either | |
1356 | * unallocated, or allocated but with refcount > 1 (so they need to be | |
1357 | * newly allocated and COWed). | |
1358 | * | |
1359 | * If @new_alloc is false this counts clusters that are already | |
1360 | * allocated and can be overwritten in-place (this includes clusters | |
1361 | * of type QCOW2_CLUSTER_ZERO_ALLOC). | |
bf319ece | 1362 | */ |
57538c86 AG |
1363 | static int count_single_write_clusters(BlockDriverState *bs, int nb_clusters, |
1364 | uint64_t *l2_slice, int l2_index, | |
1365 | bool new_alloc) | |
bf319ece | 1366 | { |
57538c86 | 1367 | BDRVQcow2State *s = bs->opaque; |
12c6aebe | 1368 | uint64_t l2_entry = get_l2_entry(s, l2_slice, l2_index); |
57538c86 | 1369 | uint64_t expected_offset = l2_entry & L2E_OFFSET_MASK; |
143550a8 | 1370 | int i; |
bf319ece | 1371 | |
143550a8 | 1372 | for (i = 0; i < nb_clusters; i++) { |
12c6aebe | 1373 | l2_entry = get_l2_entry(s, l2_slice, l2_index + i); |
57538c86 | 1374 | if (cluster_needs_new_alloc(bs, l2_entry) != new_alloc) { |
bf319ece | 1375 | break; |
143550a8 | 1376 | } |
57538c86 AG |
1377 | if (!new_alloc) { |
1378 | if (expected_offset != (l2_entry & L2E_OFFSET_MASK)) { | |
1379 | break; | |
1380 | } | |
1381 | expected_offset += s->cluster_size; | |
1382 | } | |
bf319ece KW |
1383 | } |
1384 | ||
1385 | assert(i <= nb_clusters); | |
1386 | return i; | |
1387 | } | |
1388 | ||
250196f1 | 1389 | /* |
226c3c26 KW |
1390 | * Check if there already is an AIO write request in flight which allocates |
1391 | * the same cluster. In this case we need to wait until the previous | |
1392 | * request has completed and updated the L2 table accordingly. | |
65eb2e35 KW |
1393 | * |
1394 | * Returns: | |
1395 | * 0 if there was no dependency. *cur_bytes indicates the number of | |
1396 | * bytes from guest_offset that can be read before the next | |
1397 | * dependency must be processed (or the request is complete) | |
1398 | * | |
1399 | * -EAGAIN if we had to wait for another request, previously gathered | |
1400 | * information on cluster allocation may be invalid now. The caller | |
1401 | * must start over anyway, so consider *cur_bytes undefined. | |
250196f1 | 1402 | */ |
050ed2e7 PB |
1403 | static int coroutine_fn handle_dependencies(BlockDriverState *bs, |
1404 | uint64_t guest_offset, | |
1405 | uint64_t *cur_bytes, QCowL2Meta **m) | |
250196f1 | 1406 | { |
ff99129a | 1407 | BDRVQcow2State *s = bs->opaque; |
250196f1 | 1408 | QCowL2Meta *old_alloc; |
65eb2e35 | 1409 | uint64_t bytes = *cur_bytes; |
250196f1 | 1410 | |
250196f1 KW |
1411 | QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { |
1412 | ||
65eb2e35 KW |
1413 | uint64_t start = guest_offset; |
1414 | uint64_t end = start + bytes; | |
d53ec3d8 AG |
1415 | uint64_t old_start = start_of_cluster(s, l2meta_cow_start(old_alloc)); |
1416 | uint64_t old_end = ROUND_UP(l2meta_cow_end(old_alloc), s->cluster_size); | |
250196f1 | 1417 | |
d9d74f41 | 1418 | if (end <= old_start || start >= old_end) { |
250196f1 | 1419 | /* No intersection */ |
6d207d35 VSO |
1420 | continue; |
1421 | } | |
1422 | ||
ff812c55 VSO |
1423 | if (old_alloc->keep_old_clusters && |
1424 | (end <= l2meta_cow_start(old_alloc) || | |
1425 | start >= l2meta_cow_end(old_alloc))) | |
1426 | { | |
1427 | /* | |
1428 | * Clusters intersect but COW areas don't. And cluster itself is | |
1429 | * already allocated. So, there is no actual conflict. | |
1430 | */ | |
1431 | continue; | |
1432 | } | |
1433 | ||
6d207d35 VSO |
1434 | /* Conflict */ |
1435 | ||
1436 | if (start < old_start) { | |
1437 | /* Stop at the start of a running allocation */ | |
1438 | bytes = old_start - start; | |
250196f1 | 1439 | } else { |
6d207d35 VSO |
1440 | bytes = 0; |
1441 | } | |
250196f1 | 1442 | |
6d207d35 VSO |
1443 | /* |
1444 | * Stop if an l2meta already exists. After yielding, it wouldn't | |
1445 | * be valid any more, so we'd have to clean up the old L2Metas | |
1446 | * and deal with requests depending on them before starting to | |
1447 | * gather new ones. Not worth the trouble. | |
1448 | */ | |
1449 | if (bytes == 0 && *m) { | |
1450 | *cur_bytes = 0; | |
1451 | return 0; | |
1452 | } | |
ecdd5333 | 1453 | |
6d207d35 VSO |
1454 | if (bytes == 0) { |
1455 | /* | |
1456 | * Wait for the dependency to complete. We need to recheck | |
1457 | * the free/allocated clusters when we continue. | |
1458 | */ | |
1459 | qemu_co_queue_wait(&old_alloc->dependent_requests, &s->lock); | |
1460 | return -EAGAIN; | |
250196f1 KW |
1461 | } |
1462 | } | |
1463 | ||
65eb2e35 KW |
1464 | /* Make sure that existing clusters and new allocations are only used up to |
1465 | * the next dependency if we shortened the request above */ | |
1466 | *cur_bytes = bytes; | |
250196f1 | 1467 | |
226c3c26 KW |
1468 | return 0; |
1469 | } | |
1470 | ||
0af729ec | 1471 | /* |
57538c86 AG |
1472 | * Checks how many already allocated clusters that don't require a new |
1473 | * allocation there are at the given guest_offset (up to *bytes). | |
1474 | * If *host_offset is not INV_OFFSET, only physically contiguous clusters | |
1475 | * beginning at this host offset are counted. | |
0af729ec | 1476 | * |
411d62b0 KW |
1477 | * Note that guest_offset may not be cluster aligned. In this case, the |
1478 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1479 | * therefore isn't cluster aligned as well. | |
0af729ec KW |
1480 | * |
1481 | * Returns: | |
1482 | * 0: if no allocated clusters are available at the given offset. | |
1483 | * *bytes is normally unchanged. It is set to 0 if the cluster | |
57538c86 AG |
1484 | * is allocated and can be overwritten in-place but doesn't have |
1485 | * the right physical offset. | |
0af729ec | 1486 | * |
57538c86 AG |
1487 | * 1: if allocated clusters that can be overwritten in place are |
1488 | * available at the requested offset. *bytes may have decreased | |
1489 | * and describes the length of the area that can be written to. | |
0af729ec KW |
1490 | * |
1491 | * -errno: in error cases | |
0af729ec | 1492 | */ |
a1b4ecfd PB |
1493 | static int coroutine_fn handle_copied(BlockDriverState *bs, |
1494 | uint64_t guest_offset, uint64_t *host_offset, uint64_t *bytes, | |
1495 | QCowL2Meta **m) | |
0af729ec | 1496 | { |
ff99129a | 1497 | BDRVQcow2State *s = bs->opaque; |
0af729ec | 1498 | int l2_index; |
57538c86 | 1499 | uint64_t l2_entry, cluster_offset; |
cde91766 | 1500 | uint64_t *l2_slice; |
b6d36def | 1501 | uint64_t nb_clusters; |
c53ede9f | 1502 | unsigned int keep_clusters; |
a3f1afb4 | 1503 | int ret; |
0af729ec KW |
1504 | |
1505 | trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset, | |
1506 | *bytes); | |
0af729ec | 1507 | |
c6d619cc KW |
1508 | assert(*host_offset == INV_OFFSET || offset_into_cluster(s, guest_offset) |
1509 | == offset_into_cluster(s, *host_offset)); | |
411d62b0 | 1510 | |
acb0467f | 1511 | /* |
cde91766 | 1512 | * Calculate the number of clusters to look for. We stop at L2 slice |
acb0467f KW |
1513 | * boundaries to keep things simple. |
1514 | */ | |
1515 | nb_clusters = | |
1516 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1517 | ||
cde91766 AG |
1518 | l2_index = offset_to_l2_slice_index(s, guest_offset); |
1519 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
57538c86 AG |
1520 | /* Limit total byte count to BDRV_REQUEST_MAX_BYTES */ |
1521 | nb_clusters = MIN(nb_clusters, BDRV_REQUEST_MAX_BYTES >> s->cluster_bits); | |
acb0467f | 1522 | |
0af729ec | 1523 | /* Find L2 entry for the first involved cluster */ |
cde91766 | 1524 | ret = get_cluster_table(bs, guest_offset, &l2_slice, &l2_index); |
0af729ec KW |
1525 | if (ret < 0) { |
1526 | return ret; | |
1527 | } | |
1528 | ||
12c6aebe | 1529 | l2_entry = get_l2_entry(s, l2_slice, l2_index); |
57538c86 AG |
1530 | cluster_offset = l2_entry & L2E_OFFSET_MASK; |
1531 | ||
1532 | if (!cluster_needs_new_alloc(bs, l2_entry)) { | |
1533 | if (offset_into_cluster(s, cluster_offset)) { | |
1534 | qcow2_signal_corruption(bs, true, -1, -1, "%s cluster offset " | |
1535 | "%#" PRIx64 " unaligned (guest offset: %#" | |
1536 | PRIx64 ")", l2_entry & QCOW_OFLAG_ZERO ? | |
1537 | "Preallocated zero" : "Data", | |
1538 | cluster_offset, guest_offset); | |
a97c67ee HR |
1539 | ret = -EIO; |
1540 | goto out; | |
1541 | } | |
1542 | ||
57538c86 AG |
1543 | /* If a specific host_offset is required, check it */ |
1544 | if (*host_offset != INV_OFFSET && cluster_offset != *host_offset) { | |
e62daaf6 KW |
1545 | *bytes = 0; |
1546 | ret = 0; | |
1547 | goto out; | |
1548 | } | |
1549 | ||
0af729ec | 1550 | /* We keep all QCOW_OFLAG_COPIED clusters */ |
57538c86 AG |
1551 | keep_clusters = count_single_write_clusters(bs, nb_clusters, l2_slice, |
1552 | l2_index, false); | |
c53ede9f KW |
1553 | assert(keep_clusters <= nb_clusters); |
1554 | ||
1555 | *bytes = MIN(*bytes, | |
1556 | keep_clusters * s->cluster_size | |
1557 | - offset_into_cluster(s, guest_offset)); | |
57538c86 AG |
1558 | assert(*bytes != 0); |
1559 | ||
d53ec3d8 AG |
1560 | ret = calculate_l2_meta(bs, cluster_offset, guest_offset, |
1561 | *bytes, l2_slice, m, true); | |
1562 | if (ret < 0) { | |
1563 | goto out; | |
1564 | } | |
0af729ec KW |
1565 | |
1566 | ret = 1; | |
1567 | } else { | |
0af729ec KW |
1568 | ret = 0; |
1569 | } | |
1570 | ||
0af729ec | 1571 | /* Cleanup */ |
e62daaf6 | 1572 | out: |
cde91766 | 1573 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
0af729ec | 1574 | |
e62daaf6 KW |
1575 | /* Only return a host offset if we actually made progress. Otherwise we |
1576 | * would make requirements for handle_alloc() that it can't fulfill */ | |
a97c67ee | 1577 | if (ret > 0) { |
57538c86 | 1578 | *host_offset = cluster_offset + offset_into_cluster(s, guest_offset); |
e62daaf6 KW |
1579 | } |
1580 | ||
0af729ec KW |
1581 | return ret; |
1582 | } | |
1583 | ||
226c3c26 KW |
1584 | /* |
1585 | * Allocates new clusters for the given guest_offset. | |
1586 | * | |
1587 | * At most *nb_clusters are allocated, and on return *nb_clusters is updated to | |
1588 | * contain the number of clusters that have been allocated and are contiguous | |
1589 | * in the image file. | |
1590 | * | |
c6d619cc KW |
1591 | * If *host_offset is not INV_OFFSET, it specifies the offset in the image file |
1592 | * at which the new clusters must start. *nb_clusters can be 0 on return in | |
1593 | * this case if the cluster at host_offset is already in use. If *host_offset | |
1594 | * is INV_OFFSET, the clusters can be allocated anywhere in the image file. | |
226c3c26 KW |
1595 | * |
1596 | * *host_offset is updated to contain the offset into the image file at which | |
1597 | * the first allocated cluster starts. | |
1598 | * | |
1599 | * Return 0 on success and -errno in error cases. -EAGAIN means that the | |
1600 | * function has been waiting for another request and the allocation must be | |
1601 | * restarted, but the whole request should not be failed. | |
1602 | */ | |
a39bae4e PB |
1603 | static int coroutine_fn do_alloc_cluster_offset(BlockDriverState *bs, |
1604 | uint64_t guest_offset, | |
1605 | uint64_t *host_offset, | |
1606 | uint64_t *nb_clusters) | |
226c3c26 | 1607 | { |
ff99129a | 1608 | BDRVQcow2State *s = bs->opaque; |
226c3c26 KW |
1609 | |
1610 | trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, | |
1611 | *host_offset, *nb_clusters); | |
1612 | ||
966b000f KW |
1613 | if (has_data_file(bs)) { |
1614 | assert(*host_offset == INV_OFFSET || | |
1615 | *host_offset == start_of_cluster(s, guest_offset)); | |
1616 | *host_offset = start_of_cluster(s, guest_offset); | |
1617 | return 0; | |
1618 | } | |
1619 | ||
250196f1 KW |
1620 | /* Allocate new clusters */ |
1621 | trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); | |
c6d619cc | 1622 | if (*host_offset == INV_OFFSET) { |
df021791 KW |
1623 | int64_t cluster_offset = |
1624 | qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size); | |
1625 | if (cluster_offset < 0) { | |
1626 | return cluster_offset; | |
1627 | } | |
1628 | *host_offset = cluster_offset; | |
1629 | return 0; | |
250196f1 | 1630 | } else { |
b6d36def | 1631 | int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); |
df021791 KW |
1632 | if (ret < 0) { |
1633 | return ret; | |
1634 | } | |
1635 | *nb_clusters = ret; | |
1636 | return 0; | |
250196f1 | 1637 | } |
250196f1 KW |
1638 | } |
1639 | ||
10f0ed8b | 1640 | /* |
57538c86 AG |
1641 | * Allocates new clusters for an area that is either still unallocated or |
1642 | * cannot be overwritten in-place. If *host_offset is not INV_OFFSET, | |
1643 | * clusters are only allocated if the new allocation can match the specified | |
1644 | * host offset. | |
10f0ed8b | 1645 | * |
411d62b0 KW |
1646 | * Note that guest_offset may not be cluster aligned. In this case, the |
1647 | * returned *host_offset points to exact byte referenced by guest_offset and | |
1648 | * therefore isn't cluster aligned as well. | |
10f0ed8b KW |
1649 | * |
1650 | * Returns: | |
1651 | * 0: if no clusters could be allocated. *bytes is set to 0, | |
1652 | * *host_offset is left unchanged. | |
1653 | * | |
1654 | * 1: if new clusters were allocated. *bytes may be decreased if the | |
1655 | * new allocation doesn't cover all of the requested area. | |
1656 | * *host_offset is updated to contain the host offset of the first | |
1657 | * newly allocated cluster. | |
1658 | * | |
1659 | * -errno: in error cases | |
10f0ed8b | 1660 | */ |
a1b4ecfd PB |
1661 | static int coroutine_fn handle_alloc(BlockDriverState *bs, |
1662 | uint64_t guest_offset, uint64_t *host_offset, uint64_t *bytes, | |
1663 | QCowL2Meta **m) | |
10f0ed8b | 1664 | { |
ff99129a | 1665 | BDRVQcow2State *s = bs->opaque; |
10f0ed8b | 1666 | int l2_index; |
6d99a344 | 1667 | uint64_t *l2_slice; |
b6d36def | 1668 | uint64_t nb_clusters; |
10f0ed8b KW |
1669 | int ret; |
1670 | ||
57538c86 | 1671 | uint64_t alloc_cluster_offset; |
10f0ed8b KW |
1672 | |
1673 | trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset, | |
1674 | *bytes); | |
1675 | assert(*bytes > 0); | |
1676 | ||
f5bc6350 | 1677 | /* |
6d99a344 | 1678 | * Calculate the number of clusters to look for. We stop at L2 slice |
f5bc6350 KW |
1679 | * boundaries to keep things simple. |
1680 | */ | |
c37f4cd7 KW |
1681 | nb_clusters = |
1682 | size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); | |
1683 | ||
6d99a344 AG |
1684 | l2_index = offset_to_l2_slice_index(s, guest_offset); |
1685 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
57538c86 AG |
1686 | /* Limit total allocation byte count to BDRV_REQUEST_MAX_BYTES */ |
1687 | nb_clusters = MIN(nb_clusters, BDRV_REQUEST_MAX_BYTES >> s->cluster_bits); | |
d1b9d19f | 1688 | |
10f0ed8b | 1689 | /* Find L2 entry for the first involved cluster */ |
6d99a344 | 1690 | ret = get_cluster_table(bs, guest_offset, &l2_slice, &l2_index); |
10f0ed8b KW |
1691 | if (ret < 0) { |
1692 | return ret; | |
1693 | } | |
1694 | ||
57538c86 AG |
1695 | nb_clusters = count_single_write_clusters(bs, nb_clusters, |
1696 | l2_slice, l2_index, true); | |
10f0ed8b | 1697 | |
ecdd5333 KW |
1698 | /* This function is only called when there were no non-COW clusters, so if |
1699 | * we can't find any unallocated or COW clusters either, something is | |
1700 | * wrong with our code. */ | |
1701 | assert(nb_clusters > 0); | |
1702 | ||
57538c86 AG |
1703 | /* Allocate at a given offset in the image file */ |
1704 | alloc_cluster_offset = *host_offset == INV_OFFSET ? INV_OFFSET : | |
1705 | start_of_cluster(s, *host_offset); | |
1706 | ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset, | |
1707 | &nb_clusters); | |
1708 | if (ret < 0) { | |
1709 | goto out; | |
10f0ed8b KW |
1710 | } |
1711 | ||
57538c86 AG |
1712 | /* Can't extend contiguous allocation */ |
1713 | if (nb_clusters == 0) { | |
1714 | *bytes = 0; | |
1715 | ret = 0; | |
1716 | goto out; | |
ff52aab2 HR |
1717 | } |
1718 | ||
57538c86 AG |
1719 | assert(alloc_cluster_offset != INV_OFFSET); |
1720 | ||
83baa9a4 KW |
1721 | /* |
1722 | * Save info needed for meta data update. | |
1723 | * | |
85567393 | 1724 | * requested_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1725 | * newly allocated cluster to the end of the (possibly shortened |
1726 | * before) write request. | |
1727 | * | |
85567393 | 1728 | * avail_bytes: Number of bytes from the start of the first |
83baa9a4 KW |
1729 | * newly allocated to the end of the last newly allocated cluster. |
1730 | * | |
85567393 | 1731 | * nb_bytes: The number of bytes from the start of the first |
83baa9a4 KW |
1732 | * newly allocated cluster to the end of the area that the write |
1733 | * request actually writes to (excluding COW at the end) | |
1734 | */ | |
85567393 | 1735 | uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset); |
d1b9d19f | 1736 | int avail_bytes = nb_clusters << s->cluster_bits; |
85567393 | 1737 | int nb_bytes = MIN(requested_bytes, avail_bytes); |
83baa9a4 | 1738 | |
411d62b0 | 1739 | *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset); |
85567393 | 1740 | *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset)); |
83baa9a4 KW |
1741 | assert(*bytes != 0); |
1742 | ||
d53ec3d8 AG |
1743 | ret = calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes, |
1744 | l2_slice, m, false); | |
1745 | if (ret < 0) { | |
1746 | goto out; | |
1747 | } | |
8f91d690 | 1748 | |
57538c86 | 1749 | ret = 1; |
10f0ed8b | 1750 | |
57538c86 AG |
1751 | out: |
1752 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
10f0ed8b KW |
1753 | return ret; |
1754 | } | |
1755 | ||
45aba42f | 1756 | /* |
2b60c5b9 AG |
1757 | * For a given area on the virtual disk defined by @offset and @bytes, |
1758 | * find the corresponding area on the qcow2 image, allocating new | |
1759 | * clusters (or subclusters) if necessary. The result can span a | |
1760 | * combination of allocated and previously unallocated clusters. | |
45aba42f | 1761 | * |
bfd0989a AG |
1762 | * Note that offset may not be cluster aligned. In this case, the returned |
1763 | * *host_offset points to exact byte referenced by offset and therefore | |
1764 | * isn't cluster aligned as well. | |
1765 | * | |
2b60c5b9 AG |
1766 | * On return, @host_offset is set to the beginning of the requested |
1767 | * area. This area is guaranteed to be contiguous on the qcow2 file | |
1768 | * but it can be smaller than initially requested. In this case @bytes | |
1769 | * is updated with the actual size. | |
45aba42f | 1770 | * |
2b60c5b9 AG |
1771 | * If any clusters or subclusters were allocated then @m contains a |
1772 | * list with the information of all the affected regions. Note that | |
1773 | * this can happen regardless of whether this function succeeds or | |
1774 | * not. The caller is responsible for updating the L2 metadata of the | |
1775 | * allocated clusters (on success) or freeing them (on failure), and | |
1776 | * for clearing the contents of @m afterwards in both cases. | |
45aba42f | 1777 | * |
68d100e9 KW |
1778 | * If the request conflicts with another write request in flight, the coroutine |
1779 | * is queued and will be reentered when the dependency has completed. | |
148da7ea KW |
1780 | * |
1781 | * Return 0 on success and -errno in error cases | |
45aba42f | 1782 | */ |
050ed2e7 PB |
1783 | int coroutine_fn qcow2_alloc_host_offset(BlockDriverState *bs, uint64_t offset, |
1784 | unsigned int *bytes, | |
1785 | uint64_t *host_offset, | |
1786 | QCowL2Meta **m) | |
45aba42f | 1787 | { |
ff99129a | 1788 | BDRVQcow2State *s = bs->opaque; |
710c2496 | 1789 | uint64_t start, remaining; |
250196f1 | 1790 | uint64_t cluster_offset; |
65eb2e35 | 1791 | uint64_t cur_bytes; |
710c2496 | 1792 | int ret; |
45aba42f | 1793 | |
d46a0bb2 | 1794 | trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *bytes); |
710c2496 | 1795 | |
72424114 | 1796 | again: |
16f0587e | 1797 | start = offset; |
d46a0bb2 | 1798 | remaining = *bytes; |
c6d619cc KW |
1799 | cluster_offset = INV_OFFSET; |
1800 | *host_offset = INV_OFFSET; | |
ecdd5333 KW |
1801 | cur_bytes = 0; |
1802 | *m = NULL; | |
0af729ec | 1803 | |
2c3b32d2 | 1804 | while (true) { |
ecdd5333 | 1805 | |
c6d619cc | 1806 | if (*host_offset == INV_OFFSET && cluster_offset != INV_OFFSET) { |
bfd0989a | 1807 | *host_offset = cluster_offset; |
ecdd5333 KW |
1808 | } |
1809 | ||
1810 | assert(remaining >= cur_bytes); | |
1811 | ||
1812 | start += cur_bytes; | |
1813 | remaining -= cur_bytes; | |
c6d619cc KW |
1814 | |
1815 | if (cluster_offset != INV_OFFSET) { | |
1816 | cluster_offset += cur_bytes; | |
1817 | } | |
ecdd5333 KW |
1818 | |
1819 | if (remaining == 0) { | |
1820 | break; | |
1821 | } | |
1822 | ||
1823 | cur_bytes = remaining; | |
1824 | ||
2c3b32d2 KW |
1825 | /* |
1826 | * Now start gathering as many contiguous clusters as possible: | |
1827 | * | |
1828 | * 1. Check for overlaps with in-flight allocations | |
1829 | * | |
1830 | * a) Overlap not in the first cluster -> shorten this request and | |
1831 | * let the caller handle the rest in its next loop iteration. | |
1832 | * | |
1833 | * b) Real overlaps of two requests. Yield and restart the search | |
1834 | * for contiguous clusters (the situation could have changed | |
1835 | * while we were sleeping) | |
1836 | * | |
1837 | * c) TODO: Request starts in the same cluster as the in-flight | |
1838 | * allocation ends. Shorten the COW of the in-fight allocation, | |
1839 | * set cluster_offset to write to the same cluster and set up | |
1840 | * the right synchronisation between the in-flight request and | |
1841 | * the new one. | |
1842 | */ | |
ecdd5333 | 1843 | ret = handle_dependencies(bs, start, &cur_bytes, m); |
2c3b32d2 | 1844 | if (ret == -EAGAIN) { |
ecdd5333 KW |
1845 | /* Currently handle_dependencies() doesn't yield if we already had |
1846 | * an allocation. If it did, we would have to clean up the L2Meta | |
1847 | * structs before starting over. */ | |
1848 | assert(*m == NULL); | |
2c3b32d2 KW |
1849 | goto again; |
1850 | } else if (ret < 0) { | |
1851 | return ret; | |
ecdd5333 KW |
1852 | } else if (cur_bytes == 0) { |
1853 | break; | |
2c3b32d2 KW |
1854 | } else { |
1855 | /* handle_dependencies() may have decreased cur_bytes (shortened | |
1856 | * the allocations below) so that the next dependency is processed | |
1857 | * correctly during the next loop iteration. */ | |
0af729ec | 1858 | } |
710c2496 | 1859 | |
2c3b32d2 KW |
1860 | /* |
1861 | * 2. Count contiguous COPIED clusters. | |
1862 | */ | |
1863 | ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m); | |
1864 | if (ret < 0) { | |
1865 | return ret; | |
1866 | } else if (ret) { | |
ecdd5333 | 1867 | continue; |
2c3b32d2 KW |
1868 | } else if (cur_bytes == 0) { |
1869 | break; | |
1870 | } | |
060bee89 | 1871 | |
2c3b32d2 KW |
1872 | /* |
1873 | * 3. If the request still hasn't completed, allocate new clusters, | |
1874 | * considering any cluster_offset of steps 1c or 2. | |
1875 | */ | |
1876 | ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m); | |
1877 | if (ret < 0) { | |
1878 | return ret; | |
1879 | } else if (ret) { | |
ecdd5333 | 1880 | continue; |
2c3b32d2 KW |
1881 | } else { |
1882 | assert(cur_bytes == 0); | |
1883 | break; | |
1884 | } | |
f5bc6350 | 1885 | } |
10f0ed8b | 1886 | |
d46a0bb2 KW |
1887 | *bytes -= remaining; |
1888 | assert(*bytes > 0); | |
c6d619cc | 1889 | assert(*host_offset != INV_OFFSET); |
bfd0989a AG |
1890 | assert(offset_into_cluster(s, *host_offset) == |
1891 | offset_into_cluster(s, offset)); | |
45aba42f | 1892 | |
148da7ea | 1893 | return 0; |
45aba42f KW |
1894 | } |
1895 | ||
5ea929e3 KW |
1896 | /* |
1897 | * This discards as many clusters of nb_clusters as possible at once (i.e. | |
21ab3add | 1898 | * all clusters in the same L2 slice) and returns the number of discarded |
5ea929e3 KW |
1899 | * clusters. |
1900 | */ | |
21ab3add AG |
1901 | static int discard_in_l2_slice(BlockDriverState *bs, uint64_t offset, |
1902 | uint64_t nb_clusters, | |
1903 | enum qcow2_discard_type type, bool full_discard) | |
5ea929e3 | 1904 | { |
ff99129a | 1905 | BDRVQcow2State *s = bs->opaque; |
21ab3add | 1906 | uint64_t *l2_slice; |
5ea929e3 KW |
1907 | int l2_index; |
1908 | int ret; | |
1909 | int i; | |
1910 | ||
21ab3add | 1911 | ret = get_cluster_table(bs, offset, &l2_slice, &l2_index); |
5ea929e3 KW |
1912 | if (ret < 0) { |
1913 | return ret; | |
1914 | } | |
1915 | ||
21ab3add AG |
1916 | /* Limit nb_clusters to one L2 slice */ |
1917 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
b6d36def | 1918 | assert(nb_clusters <= INT_MAX); |
5ea929e3 KW |
1919 | |
1920 | for (i = 0; i < nb_clusters; i++) { | |
a68cd703 AG |
1921 | uint64_t old_l2_entry = get_l2_entry(s, l2_slice, l2_index + i); |
1922 | uint64_t old_l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i); | |
1923 | uint64_t new_l2_entry = old_l2_entry; | |
1924 | uint64_t new_l2_bitmap = old_l2_bitmap; | |
1925 | QCow2ClusterType cluster_type = | |
1926 | qcow2_get_cluster_type(bs, old_l2_entry); | |
42a2890a JLD |
1927 | bool keep_reference = (cluster_type != QCOW2_CLUSTER_COMPRESSED) && |
1928 | !full_discard && | |
1929 | (s->discard_no_unref && | |
1930 | type == QCOW2_DISCARD_REQUEST); | |
a71835a0 KW |
1931 | |
1932 | /* | |
a68cd703 AG |
1933 | * If full_discard is true, the cluster should not read back as zeroes, |
1934 | * but rather fall through to the backing file. | |
1935 | * | |
808c4b6f HR |
1936 | * If full_discard is false, make sure that a discarded area reads back |
1937 | * as zeroes for v3 images (we cannot do it for v2 without actually | |
1938 | * writing a zero-filled buffer). We can skip the operation if the | |
1939 | * cluster is already marked as zero, or if it's unallocated and we | |
1940 | * don't have a backing file. | |
a71835a0 | 1941 | * |
237d78f8 | 1942 | * TODO We might want to use bdrv_block_status(bs) here, but we're |
a71835a0 KW |
1943 | * holding s->lock, so that doesn't work today. |
1944 | */ | |
a68cd703 AG |
1945 | if (full_discard) { |
1946 | new_l2_entry = new_l2_bitmap = 0; | |
1947 | } else if (bs->backing || qcow2_cluster_is_allocated(cluster_type)) { | |
1948 | if (has_subclusters(s)) { | |
42a2890a JLD |
1949 | if (keep_reference) { |
1950 | new_l2_entry = old_l2_entry; | |
1951 | } else { | |
1952 | new_l2_entry = 0; | |
1953 | } | |
a68cd703 AG |
1954 | new_l2_bitmap = QCOW_L2_BITMAP_ALL_ZEROES; |
1955 | } else { | |
42a2890a JLD |
1956 | if (s->qcow_version >= 3) { |
1957 | if (keep_reference) { | |
1958 | new_l2_entry |= QCOW_OFLAG_ZERO; | |
1959 | } else { | |
1960 | new_l2_entry = QCOW_OFLAG_ZERO; | |
1961 | } | |
1962 | } else { | |
1963 | new_l2_entry = 0; | |
1964 | } | |
bbd995d8 | 1965 | } |
a68cd703 | 1966 | } |
bbd995d8 | 1967 | |
a68cd703 AG |
1968 | if (old_l2_entry == new_l2_entry && old_l2_bitmap == new_l2_bitmap) { |
1969 | continue; | |
5ea929e3 KW |
1970 | } |
1971 | ||
1972 | /* First remove L2 entries */ | |
21ab3add | 1973 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
a68cd703 AG |
1974 | set_l2_entry(s, l2_slice, l2_index + i, new_l2_entry); |
1975 | if (has_subclusters(s)) { | |
1976 | set_l2_bitmap(s, l2_slice, l2_index + i, new_l2_bitmap); | |
a71835a0 | 1977 | } |
42a2890a JLD |
1978 | if (!keep_reference) { |
1979 | /* Then decrease the refcount */ | |
1980 | qcow2_free_any_cluster(bs, old_l2_entry, type); | |
1981 | } else if (s->discard_passthrough[type] && | |
1982 | (cluster_type == QCOW2_CLUSTER_NORMAL || | |
1983 | cluster_type == QCOW2_CLUSTER_ZERO_ALLOC)) { | |
1984 | /* If we keep the reference, pass on the discard still */ | |
1985 | bdrv_pdiscard(s->data_file, old_l2_entry & L2E_OFFSET_MASK, | |
1986 | s->cluster_size); | |
1987 | } | |
5ea929e3 KW |
1988 | } |
1989 | ||
21ab3add | 1990 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
5ea929e3 KW |
1991 | |
1992 | return nb_clusters; | |
1993 | } | |
1994 | ||
d2cb36af EB |
1995 | int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset, |
1996 | uint64_t bytes, enum qcow2_discard_type type, | |
1997 | bool full_discard) | |
5ea929e3 | 1998 | { |
ff99129a | 1999 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 2000 | uint64_t end_offset = offset + bytes; |
b6d36def | 2001 | uint64_t nb_clusters; |
d2cb36af | 2002 | int64_t cleared; |
5ea929e3 KW |
2003 | int ret; |
2004 | ||
f10ee139 | 2005 | /* Caller must pass aligned values, except at image end */ |
0c1bd469 | 2006 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); |
f10ee139 EB |
2007 | assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) || |
2008 | end_offset == bs->total_sectors << BDRV_SECTOR_BITS); | |
5ea929e3 | 2009 | |
d2cb36af | 2010 | nb_clusters = size_to_clusters(s, bytes); |
5ea929e3 | 2011 | |
0b919fae KW |
2012 | s->cache_discards = true; |
2013 | ||
21ab3add | 2014 | /* Each L2 slice is handled by its own loop iteration */ |
5ea929e3 | 2015 | while (nb_clusters > 0) { |
21ab3add AG |
2016 | cleared = discard_in_l2_slice(bs, offset, nb_clusters, type, |
2017 | full_discard); | |
d2cb36af EB |
2018 | if (cleared < 0) { |
2019 | ret = cleared; | |
0b919fae | 2020 | goto fail; |
5ea929e3 KW |
2021 | } |
2022 | ||
d2cb36af EB |
2023 | nb_clusters -= cleared; |
2024 | offset += (cleared * s->cluster_size); | |
5ea929e3 KW |
2025 | } |
2026 | ||
0b919fae KW |
2027 | ret = 0; |
2028 | fail: | |
2029 | s->cache_discards = false; | |
2030 | qcow2_process_discards(bs, ret); | |
2031 | ||
2032 | return ret; | |
5ea929e3 | 2033 | } |
621f0589 KW |
2034 | |
2035 | /* | |
2036 | * This zeroes as many clusters of nb_clusters as possible at once (i.e. | |
a9a9f8f0 | 2037 | * all clusters in the same L2 slice) and returns the number of zeroed |
621f0589 KW |
2038 | * clusters. |
2039 | */ | |
70bacc44 PB |
2040 | static int coroutine_fn |
2041 | zero_in_l2_slice(BlockDriverState *bs, uint64_t offset, | |
2042 | uint64_t nb_clusters, int flags) | |
621f0589 | 2043 | { |
ff99129a | 2044 | BDRVQcow2State *s = bs->opaque; |
a9a9f8f0 | 2045 | uint64_t *l2_slice; |
621f0589 KW |
2046 | int l2_index; |
2047 | int ret; | |
2048 | int i; | |
2049 | ||
a9a9f8f0 | 2050 | ret = get_cluster_table(bs, offset, &l2_slice, &l2_index); |
621f0589 KW |
2051 | if (ret < 0) { |
2052 | return ret; | |
2053 | } | |
2054 | ||
a9a9f8f0 AG |
2055 | /* Limit nb_clusters to one L2 slice */ |
2056 | nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index); | |
b6d36def | 2057 | assert(nb_clusters <= INT_MAX); |
621f0589 KW |
2058 | |
2059 | for (i = 0; i < nb_clusters; i++) { | |
205fa507 AG |
2060 | uint64_t old_l2_entry = get_l2_entry(s, l2_slice, l2_index + i); |
2061 | uint64_t old_l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i); | |
2062 | QCow2ClusterType type = qcow2_get_cluster_type(bs, old_l2_entry); | |
2063 | bool unmap = (type == QCOW2_CLUSTER_COMPRESSED) || | |
2064 | ((flags & BDRV_REQ_MAY_UNMAP) && qcow2_cluster_is_allocated(type)); | |
2065 | uint64_t new_l2_entry = unmap ? 0 : old_l2_entry; | |
2066 | uint64_t new_l2_bitmap = old_l2_bitmap; | |
2067 | ||
2068 | if (has_subclusters(s)) { | |
2069 | new_l2_bitmap = QCOW_L2_BITMAP_ALL_ZEROES; | |
2070 | } else { | |
2071 | new_l2_entry |= QCOW_OFLAG_ZERO; | |
2072 | } | |
621f0589 | 2073 | |
205fa507 | 2074 | if (old_l2_entry == new_l2_entry && old_l2_bitmap == new_l2_bitmap) { |
06cc5e2b EB |
2075 | continue; |
2076 | } | |
2077 | ||
c8bf9a91 | 2078 | /* First update L2 entries */ |
a9a9f8f0 | 2079 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
205fa507 AG |
2080 | set_l2_entry(s, l2_slice, l2_index + i, new_l2_entry); |
2081 | if (has_subclusters(s)) { | |
2082 | set_l2_bitmap(s, l2_slice, l2_index + i, new_l2_bitmap); | |
621f0589 | 2083 | } |
c8bf9a91 ML |
2084 | |
2085 | /* Then decrease the refcount */ | |
2086 | if (unmap) { | |
2087 | qcow2_free_any_cluster(bs, old_l2_entry, QCOW2_DISCARD_REQUEST); | |
2088 | } | |
621f0589 KW |
2089 | } |
2090 | ||
a9a9f8f0 | 2091 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
621f0589 KW |
2092 | |
2093 | return nb_clusters; | |
2094 | } | |
2095 | ||
a39bae4e PB |
2096 | static int coroutine_fn |
2097 | zero_l2_subclusters(BlockDriverState *bs, uint64_t offset, | |
2098 | unsigned nb_subclusters) | |
a6841a2d AG |
2099 | { |
2100 | BDRVQcow2State *s = bs->opaque; | |
2101 | uint64_t *l2_slice; | |
2102 | uint64_t old_l2_bitmap, l2_bitmap; | |
2103 | int l2_index, ret, sc = offset_to_sc_index(s, offset); | |
2104 | ||
2105 | /* For full clusters use zero_in_l2_slice() instead */ | |
2106 | assert(nb_subclusters > 0 && nb_subclusters < s->subclusters_per_cluster); | |
2107 | assert(sc + nb_subclusters <= s->subclusters_per_cluster); | |
2108 | assert(offset_into_subcluster(s, offset) == 0); | |
2109 | ||
2110 | ret = get_cluster_table(bs, offset, &l2_slice, &l2_index); | |
2111 | if (ret < 0) { | |
2112 | return ret; | |
2113 | } | |
2114 | ||
2115 | switch (qcow2_get_cluster_type(bs, get_l2_entry(s, l2_slice, l2_index))) { | |
2116 | case QCOW2_CLUSTER_COMPRESSED: | |
2117 | ret = -ENOTSUP; /* We cannot partially zeroize compressed clusters */ | |
2118 | goto out; | |
2119 | case QCOW2_CLUSTER_NORMAL: | |
2120 | case QCOW2_CLUSTER_UNALLOCATED: | |
2121 | break; | |
2122 | default: | |
2123 | g_assert_not_reached(); | |
2124 | } | |
2125 | ||
2126 | old_l2_bitmap = l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index); | |
2127 | ||
2128 | l2_bitmap |= QCOW_OFLAG_SUB_ZERO_RANGE(sc, sc + nb_subclusters); | |
2129 | l2_bitmap &= ~QCOW_OFLAG_SUB_ALLOC_RANGE(sc, sc + nb_subclusters); | |
2130 | ||
2131 | if (old_l2_bitmap != l2_bitmap) { | |
2132 | set_l2_bitmap(s, l2_slice, l2_index, l2_bitmap); | |
2133 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); | |
2134 | } | |
2135 | ||
2136 | ret = 0; | |
2137 | out: | |
2138 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); | |
2139 | ||
2140 | return ret; | |
2141 | } | |
2142 | ||
050ed2e7 PB |
2143 | int coroutine_fn qcow2_subcluster_zeroize(BlockDriverState *bs, uint64_t offset, |
2144 | uint64_t bytes, int flags) | |
621f0589 | 2145 | { |
ff99129a | 2146 | BDRVQcow2State *s = bs->opaque; |
d2cb36af | 2147 | uint64_t end_offset = offset + bytes; |
b6d36def | 2148 | uint64_t nb_clusters; |
a6841a2d | 2149 | unsigned head, tail; |
d2cb36af | 2150 | int64_t cleared; |
621f0589 KW |
2151 | int ret; |
2152 | ||
6c3944dc KW |
2153 | /* If we have to stay in sync with an external data file, zero out |
2154 | * s->data_file first. */ | |
2155 | if (data_file_is_raw(bs)) { | |
2156 | assert(has_data_file(bs)); | |
2157 | ret = bdrv_co_pwrite_zeroes(s->data_file, offset, bytes, flags); | |
2158 | if (ret < 0) { | |
2159 | return ret; | |
2160 | } | |
2161 | } | |
2162 | ||
f10ee139 | 2163 | /* Caller must pass aligned values, except at image end */ |
a6841a2d AG |
2164 | assert(offset_into_subcluster(s, offset) == 0); |
2165 | assert(offset_into_subcluster(s, end_offset) == 0 || | |
f01643fb | 2166 | end_offset >= bs->total_sectors << BDRV_SECTOR_BITS); |
f10ee139 | 2167 | |
61b30439 KW |
2168 | /* |
2169 | * The zero flag is only supported by version 3 and newer. However, if we | |
2170 | * have no backing file, we can resort to discard in version 2. | |
2171 | */ | |
621f0589 | 2172 | if (s->qcow_version < 3) { |
61b30439 KW |
2173 | if (!bs->backing) { |
2174 | return qcow2_cluster_discard(bs, offset, bytes, | |
2175 | QCOW2_DISCARD_REQUEST, false); | |
2176 | } | |
621f0589 KW |
2177 | return -ENOTSUP; |
2178 | } | |
2179 | ||
a6841a2d AG |
2180 | head = MIN(end_offset, ROUND_UP(offset, s->cluster_size)) - offset; |
2181 | offset += head; | |
2182 | ||
2183 | tail = (end_offset >= bs->total_sectors << BDRV_SECTOR_BITS) ? 0 : | |
2184 | end_offset - MAX(offset, start_of_cluster(s, end_offset)); | |
2185 | end_offset -= tail; | |
621f0589 | 2186 | |
0b919fae KW |
2187 | s->cache_discards = true; |
2188 | ||
a6841a2d AG |
2189 | if (head) { |
2190 | ret = zero_l2_subclusters(bs, offset - head, | |
2191 | size_to_subclusters(s, head)); | |
2192 | if (ret < 0) { | |
2193 | goto fail; | |
2194 | } | |
2195 | } | |
2196 | ||
2197 | /* Each L2 slice is handled by its own loop iteration */ | |
2198 | nb_clusters = size_to_clusters(s, end_offset - offset); | |
2199 | ||
621f0589 | 2200 | while (nb_clusters > 0) { |
a9a9f8f0 | 2201 | cleared = zero_in_l2_slice(bs, offset, nb_clusters, flags); |
d2cb36af EB |
2202 | if (cleared < 0) { |
2203 | ret = cleared; | |
0b919fae | 2204 | goto fail; |
621f0589 KW |
2205 | } |
2206 | ||
d2cb36af EB |
2207 | nb_clusters -= cleared; |
2208 | offset += (cleared * s->cluster_size); | |
621f0589 KW |
2209 | } |
2210 | ||
a6841a2d AG |
2211 | if (tail) { |
2212 | ret = zero_l2_subclusters(bs, end_offset, size_to_subclusters(s, tail)); | |
2213 | if (ret < 0) { | |
2214 | goto fail; | |
2215 | } | |
2216 | } | |
2217 | ||
0b919fae KW |
2218 | ret = 0; |
2219 | fail: | |
2220 | s->cache_discards = false; | |
2221 | qcow2_process_discards(bs, ret); | |
2222 | ||
2223 | return ret; | |
621f0589 | 2224 | } |
32b6444d HR |
2225 | |
2226 | /* | |
2227 | * Expands all zero clusters in a specific L1 table (or deallocates them, for | |
2228 | * non-backed non-pre-allocated zero clusters). | |
2229 | * | |
4057a2b2 HR |
2230 | * l1_entries and *visited_l1_entries are used to keep track of progress for |
2231 | * status_cb(). l1_entries contains the total number of L1 entries and | |
2232 | * *visited_l1_entries counts all visited L1 entries. | |
32b6444d HR |
2233 | */ |
2234 | static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table, | |
ecf58777 | 2235 | int l1_size, int64_t *visited_l1_entries, |
4057a2b2 | 2236 | int64_t l1_entries, |
8b13976d HR |
2237 | BlockDriverAmendStatusCB *status_cb, |
2238 | void *cb_opaque) | |
32b6444d | 2239 | { |
ff99129a | 2240 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 2241 | bool is_active_l1 = (l1_table == s->l1_table); |
415184f5 AG |
2242 | uint64_t *l2_slice = NULL; |
2243 | unsigned slice, slice_size2, n_slices; | |
32b6444d HR |
2244 | int ret; |
2245 | int i, j; | |
2246 | ||
7bbb5920 AG |
2247 | /* qcow2_downgrade() is not allowed in images with subclusters */ |
2248 | assert(!has_subclusters(s)); | |
2249 | ||
c8fd8554 | 2250 | slice_size2 = s->l2_slice_size * l2_entry_size(s); |
415184f5 AG |
2251 | n_slices = s->cluster_size / slice_size2; |
2252 | ||
32b6444d HR |
2253 | if (!is_active_l1) { |
2254 | /* inactive L2 tables require a buffer to be stored in when loading | |
2255 | * them from disk */ | |
415184f5 AG |
2256 | l2_slice = qemu_try_blockalign(bs->file->bs, slice_size2); |
2257 | if (l2_slice == NULL) { | |
de82815d KW |
2258 | return -ENOMEM; |
2259 | } | |
32b6444d HR |
2260 | } |
2261 | ||
2262 | for (i = 0; i < l1_size; i++) { | |
2263 | uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK; | |
0e06528e | 2264 | uint64_t l2_refcount; |
32b6444d HR |
2265 | |
2266 | if (!l2_offset) { | |
2267 | /* unallocated */ | |
4057a2b2 HR |
2268 | (*visited_l1_entries)++; |
2269 | if (status_cb) { | |
8b13976d | 2270 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 2271 | } |
32b6444d HR |
2272 | continue; |
2273 | } | |
2274 | ||
8dd93d93 HR |
2275 | if (offset_into_cluster(s, l2_offset)) { |
2276 | qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" | |
2277 | PRIx64 " unaligned (L1 index: %#x)", | |
2278 | l2_offset, i); | |
2279 | ret = -EIO; | |
2280 | goto fail; | |
2281 | } | |
2282 | ||
9b765486 AG |
2283 | ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits, |
2284 | &l2_refcount); | |
2285 | if (ret < 0) { | |
2286 | goto fail; | |
2287 | } | |
2288 | ||
415184f5 AG |
2289 | for (slice = 0; slice < n_slices; slice++) { |
2290 | uint64_t slice_offset = l2_offset + slice * slice_size2; | |
2291 | bool l2_dirty = false; | |
226494ff AG |
2292 | if (is_active_l1) { |
2293 | /* get active L2 tables from cache */ | |
415184f5 AG |
2294 | ret = qcow2_cache_get(bs, s->l2_table_cache, slice_offset, |
2295 | (void **)&l2_slice); | |
226494ff AG |
2296 | } else { |
2297 | /* load inactive L2 tables from disk */ | |
32cc71de AF |
2298 | ret = bdrv_pread(bs->file, slice_offset, slice_size2, |
2299 | l2_slice, 0); | |
226494ff AG |
2300 | } |
2301 | if (ret < 0) { | |
2302 | goto fail; | |
32b6444d HR |
2303 | } |
2304 | ||
415184f5 | 2305 | for (j = 0; j < s->l2_slice_size; j++) { |
12c6aebe | 2306 | uint64_t l2_entry = get_l2_entry(s, l2_slice, j); |
226494ff AG |
2307 | int64_t offset = l2_entry & L2E_OFFSET_MASK; |
2308 | QCow2ClusterType cluster_type = | |
808c2bb4 | 2309 | qcow2_get_cluster_type(bs, l2_entry); |
226494ff AG |
2310 | |
2311 | if (cluster_type != QCOW2_CLUSTER_ZERO_PLAIN && | |
2312 | cluster_type != QCOW2_CLUSTER_ZERO_ALLOC) { | |
32b6444d HR |
2313 | continue; |
2314 | } | |
2315 | ||
226494ff AG |
2316 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
2317 | if (!bs->backing) { | |
7bbb5920 AG |
2318 | /* |
2319 | * not backed; therefore we can simply deallocate the | |
2320 | * cluster. No need to call set_l2_bitmap(), this | |
2321 | * function doesn't support images with subclusters. | |
2322 | */ | |
12c6aebe | 2323 | set_l2_entry(s, l2_slice, j, 0); |
226494ff AG |
2324 | l2_dirty = true; |
2325 | continue; | |
2326 | } | |
2327 | ||
2328 | offset = qcow2_alloc_clusters(bs, s->cluster_size); | |
2329 | if (offset < 0) { | |
2330 | ret = offset; | |
2331 | goto fail; | |
2332 | } | |
ecf58777 | 2333 | |
3a75a870 AG |
2334 | /* The offset must fit in the offset field */ |
2335 | assert((offset & L2E_OFFSET_MASK) == offset); | |
2336 | ||
226494ff AG |
2337 | if (l2_refcount > 1) { |
2338 | /* For shared L2 tables, set the refcount accordingly | |
2339 | * (it is already 1 and needs to be l2_refcount) */ | |
2340 | ret = qcow2_update_cluster_refcount( | |
2341 | bs, offset >> s->cluster_bits, | |
2aabe7c7 | 2342 | refcount_diff(1, l2_refcount), false, |
ecf58777 | 2343 | QCOW2_DISCARD_OTHER); |
226494ff AG |
2344 | if (ret < 0) { |
2345 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
2346 | QCOW2_DISCARD_OTHER); | |
2347 | goto fail; | |
2348 | } | |
ecf58777 HR |
2349 | } |
2350 | } | |
32b6444d | 2351 | |
226494ff | 2352 | if (offset_into_cluster(s, offset)) { |
415184f5 | 2353 | int l2_index = slice * s->l2_slice_size + j; |
226494ff AG |
2354 | qcow2_signal_corruption( |
2355 | bs, true, -1, -1, | |
2356 | "Cluster allocation offset " | |
2357 | "%#" PRIx64 " unaligned (L2 offset: %#" | |
2358 | PRIx64 ", L2 index: %#x)", offset, | |
415184f5 | 2359 | l2_offset, l2_index); |
226494ff AG |
2360 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { |
2361 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
2362 | QCOW2_DISCARD_ALWAYS); | |
2363 | } | |
2364 | ret = -EIO; | |
2365 | goto fail; | |
8dd93d93 | 2366 | } |
8dd93d93 | 2367 | |
226494ff | 2368 | ret = qcow2_pre_write_overlap_check(bs, 0, offset, |
966b000f | 2369 | s->cluster_size, true); |
226494ff AG |
2370 | if (ret < 0) { |
2371 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { | |
2372 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
2373 | QCOW2_DISCARD_ALWAYS); | |
2374 | } | |
2375 | goto fail; | |
320c7066 | 2376 | } |
32b6444d | 2377 | |
966b000f KW |
2378 | ret = bdrv_pwrite_zeroes(s->data_file, offset, |
2379 | s->cluster_size, 0); | |
226494ff AG |
2380 | if (ret < 0) { |
2381 | if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) { | |
2382 | qcow2_free_clusters(bs, offset, s->cluster_size, | |
2383 | QCOW2_DISCARD_ALWAYS); | |
2384 | } | |
2385 | goto fail; | |
320c7066 | 2386 | } |
32b6444d | 2387 | |
226494ff | 2388 | if (l2_refcount == 1) { |
12c6aebe | 2389 | set_l2_entry(s, l2_slice, j, offset | QCOW_OFLAG_COPIED); |
226494ff | 2390 | } else { |
12c6aebe | 2391 | set_l2_entry(s, l2_slice, j, offset); |
226494ff | 2392 | } |
7bbb5920 AG |
2393 | /* |
2394 | * No need to call set_l2_bitmap() after set_l2_entry() because | |
2395 | * this function doesn't support images with subclusters. | |
2396 | */ | |
226494ff | 2397 | l2_dirty = true; |
e390cf5a | 2398 | } |
32b6444d | 2399 | |
226494ff AG |
2400 | if (is_active_l1) { |
2401 | if (l2_dirty) { | |
415184f5 | 2402 | qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice); |
226494ff | 2403 | qcow2_cache_depends_on_flush(s->l2_table_cache); |
32b6444d | 2404 | } |
415184f5 | 2405 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
226494ff AG |
2406 | } else { |
2407 | if (l2_dirty) { | |
2408 | ret = qcow2_pre_write_overlap_check( | |
2409 | bs, QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, | |
966b000f | 2410 | slice_offset, slice_size2, false); |
226494ff AG |
2411 | if (ret < 0) { |
2412 | goto fail; | |
2413 | } | |
32b6444d | 2414 | |
32cc71de AF |
2415 | ret = bdrv_pwrite(bs->file, slice_offset, slice_size2, |
2416 | l2_slice, 0); | |
226494ff AG |
2417 | if (ret < 0) { |
2418 | goto fail; | |
2419 | } | |
32b6444d HR |
2420 | } |
2421 | } | |
2422 | } | |
4057a2b2 HR |
2423 | |
2424 | (*visited_l1_entries)++; | |
2425 | if (status_cb) { | |
8b13976d | 2426 | status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque); |
4057a2b2 | 2427 | } |
32b6444d HR |
2428 | } |
2429 | ||
2430 | ret = 0; | |
2431 | ||
2432 | fail: | |
415184f5 | 2433 | if (l2_slice) { |
32b6444d | 2434 | if (!is_active_l1) { |
415184f5 | 2435 | qemu_vfree(l2_slice); |
32b6444d | 2436 | } else { |
415184f5 | 2437 | qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice); |
32b6444d HR |
2438 | } |
2439 | } | |
2440 | return ret; | |
2441 | } | |
2442 | ||
2443 | /* | |
2444 | * For backed images, expands all zero clusters on the image. For non-backed | |
2445 | * images, deallocates all non-pre-allocated zero clusters (and claims the | |
2446 | * allocation for pre-allocated ones). This is important for downgrading to a | |
2447 | * qcow2 version which doesn't yet support metadata zero clusters. | |
2448 | */ | |
4057a2b2 | 2449 | int qcow2_expand_zero_clusters(BlockDriverState *bs, |
8b13976d HR |
2450 | BlockDriverAmendStatusCB *status_cb, |
2451 | void *cb_opaque) | |
32b6444d | 2452 | { |
ff99129a | 2453 | BDRVQcow2State *s = bs->opaque; |
32b6444d | 2454 | uint64_t *l1_table = NULL; |
4057a2b2 | 2455 | int64_t l1_entries = 0, visited_l1_entries = 0; |
32b6444d HR |
2456 | int ret; |
2457 | int i, j; | |
2458 | ||
4057a2b2 HR |
2459 | if (status_cb) { |
2460 | l1_entries = s->l1_size; | |
2461 | for (i = 0; i < s->nb_snapshots; i++) { | |
2462 | l1_entries += s->snapshots[i].l1_size; | |
2463 | } | |
2464 | } | |
2465 | ||
32b6444d | 2466 | ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size, |
4057a2b2 | 2467 | &visited_l1_entries, l1_entries, |
8b13976d | 2468 | status_cb, cb_opaque); |
32b6444d HR |
2469 | if (ret < 0) { |
2470 | goto fail; | |
2471 | } | |
2472 | ||
2473 | /* Inactive L1 tables may point to active L2 tables - therefore it is | |
2474 | * necessary to flush the L2 table cache before trying to access the L2 | |
2475 | * tables pointed to by inactive L1 entries (else we might try to expand | |
2476 | * zero clusters that have already been expanded); furthermore, it is also | |
2477 | * necessary to empty the L2 table cache, since it may contain tables which | |
2478 | * are now going to be modified directly on disk, bypassing the cache. | |
2479 | * qcow2_cache_empty() does both for us. */ | |
2480 | ret = qcow2_cache_empty(bs, s->l2_table_cache); | |
2481 | if (ret < 0) { | |
2482 | goto fail; | |
2483 | } | |
2484 | ||
2485 | for (i = 0; i < s->nb_snapshots; i++) { | |
c9a442e4 AG |
2486 | int l1_size2; |
2487 | uint64_t *new_l1_table; | |
2488 | Error *local_err = NULL; | |
2489 | ||
2490 | ret = qcow2_validate_table(bs, s->snapshots[i].l1_table_offset, | |
02b1ecfa | 2491 | s->snapshots[i].l1_size, L1E_SIZE, |
c9a442e4 AG |
2492 | QCOW_MAX_L1_SIZE, "Snapshot L1 table", |
2493 | &local_err); | |
2494 | if (ret < 0) { | |
2495 | error_report_err(local_err); | |
2496 | goto fail; | |
2497 | } | |
32b6444d | 2498 | |
02b1ecfa | 2499 | l1_size2 = s->snapshots[i].l1_size * L1E_SIZE; |
c9a442e4 | 2500 | new_l1_table = g_try_realloc(l1_table, l1_size2); |
de7269d2 AG |
2501 | |
2502 | if (!new_l1_table) { | |
2503 | ret = -ENOMEM; | |
2504 | goto fail; | |
2505 | } | |
2506 | ||
2507 | l1_table = new_l1_table; | |
32b6444d | 2508 | |
32cc71de AF |
2509 | ret = bdrv_pread(bs->file, s->snapshots[i].l1_table_offset, l1_size2, |
2510 | l1_table, 0); | |
32b6444d HR |
2511 | if (ret < 0) { |
2512 | goto fail; | |
2513 | } | |
2514 | ||
2515 | for (j = 0; j < s->snapshots[i].l1_size; j++) { | |
2516 | be64_to_cpus(&l1_table[j]); | |
2517 | } | |
2518 | ||
2519 | ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size, | |
4057a2b2 | 2520 | &visited_l1_entries, l1_entries, |
8b13976d | 2521 | status_cb, cb_opaque); |
32b6444d HR |
2522 | if (ret < 0) { |
2523 | goto fail; | |
2524 | } | |
2525 | } | |
2526 | ||
2527 | ret = 0; | |
2528 | ||
2529 | fail: | |
32b6444d HR |
2530 | g_free(l1_table); |
2531 | return ret; | |
2532 | } | |
a6e09846 VSO |
2533 | |
2534 | void qcow2_parse_compressed_l2_entry(BlockDriverState *bs, uint64_t l2_entry, | |
2535 | uint64_t *coffset, int *csize) | |
2536 | { | |
2537 | BDRVQcow2State *s = bs->opaque; | |
2538 | int nb_csectors; | |
2539 | ||
2540 | assert(qcow2_get_cluster_type(bs, l2_entry) == QCOW2_CLUSTER_COMPRESSED); | |
2541 | ||
2542 | *coffset = l2_entry & s->cluster_offset_mask; | |
2543 | ||
2544 | nb_csectors = ((l2_entry >> s->csize_shift) & s->csize_mask) + 1; | |
2545 | *csize = nb_csectors * QCOW2_COMPRESSED_SECTOR_SIZE - | |
2546 | (*coffset & (QCOW2_COMPRESSED_SECTOR_SIZE - 1)); | |
2547 | } |