]> git.proxmox.com Git - mirror_qemu.git/blame - block/qcow2-cluster.c
virtio: decrement vq->inuse in virtqueue_discard()
[mirror_qemu.git] / block / qcow2-cluster.c
CommitLineData
45aba42f
KW
1/*
2 * Block driver for the QCOW version 2 format
3 *
4 * Copyright (c) 2004-2006 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
80c71a24 25#include "qemu/osdep.h"
45aba42f
KW
26#include <zlib.h>
27
da34e65c 28#include "qapi/error.h"
45aba42f 29#include "qemu-common.h"
737e150e 30#include "block/block_int.h"
45aba42f 31#include "block/qcow2.h"
58369e22 32#include "qemu/bswap.h"
3cce16f4 33#include "trace.h"
45aba42f 34
2cf7cfa1
KW
35int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
36 bool exact_size)
45aba42f 37{
ff99129a 38 BDRVQcow2State *s = bs->opaque;
2cf7cfa1 39 int new_l1_size2, ret, i;
45aba42f 40 uint64_t *new_l1_table;
fda74f82 41 int64_t old_l1_table_offset, old_l1_size;
2cf7cfa1 42 int64_t new_l1_table_offset, new_l1_size;
45aba42f
KW
43 uint8_t data[12];
44
72893756 45 if (min_size <= s->l1_size)
45aba42f 46 return 0;
72893756 47
b93f9950
HR
48 /* Do a sanity check on min_size before trying to calculate new_l1_size
49 * (this prevents overflows during the while loop for the calculation of
50 * new_l1_size) */
51 if (min_size > INT_MAX / sizeof(uint64_t)) {
52 return -EFBIG;
53 }
54
72893756
SH
55 if (exact_size) {
56 new_l1_size = min_size;
57 } else {
58 /* Bump size up to reduce the number of times we have to grow */
59 new_l1_size = s->l1_size;
60 if (new_l1_size == 0) {
61 new_l1_size = 1;
62 }
63 while (min_size > new_l1_size) {
64 new_l1_size = (new_l1_size * 3 + 1) / 2;
65 }
45aba42f 66 }
72893756 67
84c26520
HR
68 QEMU_BUILD_BUG_ON(QCOW_MAX_L1_SIZE > INT_MAX);
69 if (new_l1_size > QCOW_MAX_L1_SIZE / sizeof(uint64_t)) {
2cf7cfa1
KW
70 return -EFBIG;
71 }
72
45aba42f 73#ifdef DEBUG_ALLOC2
2cf7cfa1
KW
74 fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n",
75 s->l1_size, new_l1_size);
45aba42f
KW
76#endif
77
78 new_l1_size2 = sizeof(uint64_t) * new_l1_size;
9a4f4c31 79 new_l1_table = qemu_try_blockalign(bs->file->bs,
de82815d
KW
80 align_offset(new_l1_size2, 512));
81 if (new_l1_table == NULL) {
82 return -ENOMEM;
83 }
84 memset(new_l1_table, 0, align_offset(new_l1_size2, 512));
85
45aba42f
KW
86 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
87
88 /* write new table (align to cluster) */
66f82cee 89 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE);
ed6ccf0f 90 new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
5d757b56 91 if (new_l1_table_offset < 0) {
de82815d 92 qemu_vfree(new_l1_table);
5d757b56
KW
93 return new_l1_table_offset;
94 }
29c1a730
KW
95
96 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
97 if (ret < 0) {
80fa3341 98 goto fail;
29c1a730 99 }
45aba42f 100
cf93980e
HR
101 /* the L1 position has not yet been updated, so these clusters must
102 * indeed be completely free */
231bb267
HR
103 ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset,
104 new_l1_size2);
cf93980e
HR
105 if (ret < 0) {
106 goto fail;
107 }
108
66f82cee 109 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE);
45aba42f
KW
110 for(i = 0; i < s->l1_size; i++)
111 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
d9ca2ea2 112 ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset,
9a4f4c31 113 new_l1_table, new_l1_size2);
8b3b7206 114 if (ret < 0)
45aba42f
KW
115 goto fail;
116 for(i = 0; i < s->l1_size; i++)
117 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
118
119 /* set new table */
66f82cee 120 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE);
f1f7a1dd 121 stl_be_p(data, new_l1_size);
e4ef9f46 122 stq_be_p(data + 4, new_l1_table_offset);
d9ca2ea2 123 ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size),
9a4f4c31 124 data, sizeof(data));
8b3b7206 125 if (ret < 0) {
45aba42f 126 goto fail;
fb8fa77c 127 }
de82815d 128 qemu_vfree(s->l1_table);
fda74f82 129 old_l1_table_offset = s->l1_table_offset;
45aba42f
KW
130 s->l1_table_offset = new_l1_table_offset;
131 s->l1_table = new_l1_table;
fda74f82 132 old_l1_size = s->l1_size;
45aba42f 133 s->l1_size = new_l1_size;
fda74f82
HR
134 qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * sizeof(uint64_t),
135 QCOW2_DISCARD_OTHER);
45aba42f
KW
136 return 0;
137 fail:
de82815d 138 qemu_vfree(new_l1_table);
6cfcb9b8
KW
139 qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2,
140 QCOW2_DISCARD_OTHER);
8b3b7206 141 return ret;
45aba42f
KW
142}
143
45aba42f
KW
144/*
145 * l2_load
146 *
147 * Loads a L2 table into memory. If the table is in the cache, the cache
148 * is used; otherwise the L2 table is loaded from the image file.
149 *
150 * Returns a pointer to the L2 table on success, or NULL if the read from
151 * the image file failed.
152 */
153
55c17e98
KW
154static int l2_load(BlockDriverState *bs, uint64_t l2_offset,
155 uint64_t **l2_table)
45aba42f 156{
ff99129a 157 BDRVQcow2State *s = bs->opaque;
45aba42f 158
9be38598
EH
159 return qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
160 (void **)l2_table);
45aba42f
KW
161}
162
6583e3c7
KW
163/*
164 * Writes one sector of the L1 table to the disk (can't update single entries
165 * and we really don't want bdrv_pread to perform a read-modify-write)
166 */
167#define L1_ENTRIES_PER_SECTOR (512 / 8)
e23e400e 168int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index)
6583e3c7 169{
ff99129a 170 BDRVQcow2State *s = bs->opaque;
a1391444 171 uint64_t buf[L1_ENTRIES_PER_SECTOR] = { 0 };
6583e3c7 172 int l1_start_index;
f7defcb6 173 int i, ret;
6583e3c7
KW
174
175 l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
a1391444
HR
176 for (i = 0; i < L1_ENTRIES_PER_SECTOR && l1_start_index + i < s->l1_size;
177 i++)
178 {
6583e3c7
KW
179 buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
180 }
181
231bb267 182 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1,
cf93980e
HR
183 s->l1_table_offset + 8 * l1_start_index, sizeof(buf));
184 if (ret < 0) {
185 return ret;
186 }
187
66f82cee 188 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
d9ca2ea2 189 ret = bdrv_pwrite_sync(bs->file,
9a4f4c31
KW
190 s->l1_table_offset + 8 * l1_start_index,
191 buf, sizeof(buf));
f7defcb6
KW
192 if (ret < 0) {
193 return ret;
6583e3c7
KW
194 }
195
196 return 0;
197}
198
45aba42f
KW
199/*
200 * l2_allocate
201 *
202 * Allocate a new l2 entry in the file. If l1_index points to an already
203 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
204 * table) copy the contents of the old L2 table into the newly allocated one.
205 * Otherwise the new table is initialized with zeros.
206 *
207 */
208
c46e1167 209static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
45aba42f 210{
ff99129a 211 BDRVQcow2State *s = bs->opaque;
6583e3c7 212 uint64_t old_l2_offset;
8585afd8 213 uint64_t *l2_table = NULL;
f4f0d391 214 int64_t l2_offset;
c46e1167 215 int ret;
45aba42f
KW
216
217 old_l2_offset = s->l1_table[l1_index];
218
3cce16f4
KW
219 trace_qcow2_l2_allocate(bs, l1_index);
220
45aba42f
KW
221 /* allocate a new l2 entry */
222
ed6ccf0f 223 l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
5d757b56 224 if (l2_offset < 0) {
be0b742e
HR
225 ret = l2_offset;
226 goto fail;
5d757b56 227 }
29c1a730
KW
228
229 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
230 if (ret < 0) {
231 goto fail;
232 }
45aba42f 233
45aba42f
KW
234 /* allocate a new entry in the l2 cache */
235
3cce16f4 236 trace_qcow2_l2_allocate_get_empty(bs, l1_index);
29c1a730
KW
237 ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table);
238 if (ret < 0) {
be0b742e 239 goto fail;
29c1a730
KW
240 }
241
242 l2_table = *table;
45aba42f 243
8e37f681 244 if ((old_l2_offset & L1E_OFFSET_MASK) == 0) {
45aba42f
KW
245 /* if there was no old l2 table, clear the new table */
246 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
247 } else {
29c1a730
KW
248 uint64_t* old_table;
249
45aba42f 250 /* if there was an old l2 table, read it from the disk */
66f82cee 251 BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ);
8e37f681
KW
252 ret = qcow2_cache_get(bs, s->l2_table_cache,
253 old_l2_offset & L1E_OFFSET_MASK,
29c1a730
KW
254 (void**) &old_table);
255 if (ret < 0) {
256 goto fail;
257 }
258
259 memcpy(l2_table, old_table, s->cluster_size);
260
a3f1afb4 261 qcow2_cache_put(bs, s->l2_table_cache, (void **) &old_table);
45aba42f 262 }
29c1a730 263
45aba42f 264 /* write the l2 table to the file */
66f82cee 265 BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
29c1a730 266
3cce16f4 267 trace_qcow2_l2_allocate_write_l2(bs, l1_index);
72e80b89 268 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
29c1a730 269 ret = qcow2_cache_flush(bs, s->l2_table_cache);
c46e1167 270 if (ret < 0) {
175e1152
KW
271 goto fail;
272 }
273
274 /* update the L1 entry */
3cce16f4 275 trace_qcow2_l2_allocate_write_l1(bs, l1_index);
175e1152 276 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
e23e400e 277 ret = qcow2_write_l1_entry(bs, l1_index);
175e1152
KW
278 if (ret < 0) {
279 goto fail;
c46e1167 280 }
45aba42f 281
c46e1167 282 *table = l2_table;
3cce16f4 283 trace_qcow2_l2_allocate_done(bs, l1_index, 0);
c46e1167 284 return 0;
175e1152
KW
285
286fail:
3cce16f4 287 trace_qcow2_l2_allocate_done(bs, l1_index, ret);
8585afd8
HR
288 if (l2_table != NULL) {
289 qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
290 }
68dba0bf 291 s->l1_table[l1_index] = old_l2_offset;
e3b21ef9
HR
292 if (l2_offset > 0) {
293 qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
294 QCOW2_DISCARD_ALWAYS);
295 }
175e1152 296 return ret;
45aba42f
KW
297}
298
2bfcc4a0
KW
299/*
300 * Checks how many clusters in a given L2 table are contiguous in the image
301 * file. As soon as one of the flags in the bitmask stop_flags changes compared
302 * to the first cluster, the search is stopped and the cluster is not counted
303 * as contiguous. (This allows it, for example, to stop at the first compressed
304 * cluster which may require a different handling)
305 */
b6d36def 306static int count_contiguous_clusters(int nb_clusters, int cluster_size,
61653008 307 uint64_t *l2_table, uint64_t stop_flags)
45aba42f
KW
308{
309 int i;
78a52ad5 310 uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED;
15684a47
HR
311 uint64_t first_entry = be64_to_cpu(l2_table[0]);
312 uint64_t offset = first_entry & mask;
45aba42f
KW
313
314 if (!offset)
315 return 0;
316
a99dfb45 317 assert(qcow2_get_cluster_type(first_entry) == QCOW2_CLUSTER_NORMAL);
15684a47 318
61653008 319 for (i = 0; i < nb_clusters; i++) {
2bfcc4a0
KW
320 uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask;
321 if (offset + (uint64_t) i * cluster_size != l2_entry) {
45aba42f 322 break;
2bfcc4a0
KW
323 }
324 }
45aba42f 325
61653008 326 return i;
45aba42f
KW
327}
328
a99dfb45
KW
329static int count_contiguous_clusters_by_type(int nb_clusters,
330 uint64_t *l2_table,
331 int wanted_type)
45aba42f 332{
2bfcc4a0
KW
333 int i;
334
335 for (i = 0; i < nb_clusters; i++) {
336 int type = qcow2_get_cluster_type(be64_to_cpu(l2_table[i]));
45aba42f 337
a99dfb45 338 if (type != wanted_type) {
2bfcc4a0
KW
339 break;
340 }
341 }
45aba42f
KW
342
343 return i;
344}
345
346/* The crypt function is compatible with the linux cryptoloop
347 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
348 supported */
ff99129a 349int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num,
f6fa64f6
DB
350 uint8_t *out_buf, const uint8_t *in_buf,
351 int nb_sectors, bool enc,
352 Error **errp)
45aba42f
KW
353{
354 union {
355 uint64_t ll[2];
356 uint8_t b[16];
357 } ivec;
358 int i;
f6fa64f6 359 int ret;
45aba42f
KW
360
361 for(i = 0; i < nb_sectors; i++) {
362 ivec.ll[0] = cpu_to_le64(sector_num);
363 ivec.ll[1] = 0;
f6fa64f6
DB
364 if (qcrypto_cipher_setiv(s->cipher,
365 ivec.b, G_N_ELEMENTS(ivec.b),
366 errp) < 0) {
367 return -1;
368 }
369 if (enc) {
370 ret = qcrypto_cipher_encrypt(s->cipher,
371 in_buf,
372 out_buf,
373 512,
374 errp);
375 } else {
376 ret = qcrypto_cipher_decrypt(s->cipher,
377 in_buf,
378 out_buf,
379 512,
380 errp);
381 }
382 if (ret < 0) {
383 return -1;
384 }
45aba42f
KW
385 sector_num++;
386 in_buf += 512;
387 out_buf += 512;
388 }
f6fa64f6 389 return 0;
45aba42f
KW
390}
391
aaa4d20b
KW
392static int coroutine_fn do_perform_cow(BlockDriverState *bs,
393 uint64_t src_cluster_offset,
394 uint64_t cluster_offset,
395 int offset_in_cluster,
396 int bytes)
45aba42f 397{
ff99129a 398 BDRVQcow2State *s = bs->opaque;
aef4acb6
SH
399 QEMUIOVector qiov;
400 struct iovec iov;
aaa4d20b 401 int ret;
1b9f1491 402
aaa4d20b 403 iov.iov_len = bytes;
de82815d
KW
404 iov.iov_base = qemu_try_blockalign(bs, iov.iov_len);
405 if (iov.iov_base == NULL) {
406 return -ENOMEM;
407 }
aef4acb6
SH
408
409 qemu_iovec_init_external(&qiov, &iov, 1);
1b9f1491 410
66f82cee 411 BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
aef4acb6 412
dba28555 413 if (!bs->drv) {
bd604369
KW
414 ret = -ENOMEDIUM;
415 goto out;
dba28555
HR
416 }
417
aef4acb6
SH
418 /* Call .bdrv_co_readv() directly instead of using the public block-layer
419 * interface. This avoids double I/O throttling and request tracking,
420 * which can lead to deadlock when block layer copy-on-read is enabled.
421 */
aaa4d20b
KW
422 ret = bs->drv->bdrv_co_preadv(bs, src_cluster_offset + offset_in_cluster,
423 bytes, &qiov, 0);
1b9f1491
KW
424 if (ret < 0) {
425 goto out;
426 }
427
8336aafa 428 if (bs->encrypted) {
f6fa64f6 429 Error *err = NULL;
aaa4d20b
KW
430 int64_t sector = (cluster_offset + offset_in_cluster)
431 >> BDRV_SECTOR_BITS;
f6fa64f6 432 assert(s->cipher);
aaa4d20b
KW
433 assert((offset_in_cluster & ~BDRV_SECTOR_MASK) == 0);
434 assert((bytes & ~BDRV_SECTOR_MASK) == 0);
435 if (qcow2_encrypt_sectors(s, sector, iov.iov_base, iov.iov_base,
436 bytes >> BDRV_SECTOR_BITS, true, &err) < 0) {
f6fa64f6
DB
437 ret = -EIO;
438 error_free(err);
439 goto out;
440 }
45aba42f 441 }
1b9f1491 442
231bb267 443 ret = qcow2_pre_write_overlap_check(bs, 0,
aaa4d20b 444 cluster_offset + offset_in_cluster, bytes);
cf93980e
HR
445 if (ret < 0) {
446 goto out;
447 }
448
66f82cee 449 BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
a03ef88f 450 ret = bdrv_co_pwritev(bs->file, cluster_offset + offset_in_cluster,
aaa4d20b 451 bytes, &qiov, 0);
1b9f1491
KW
452 if (ret < 0) {
453 goto out;
454 }
455
456 ret = 0;
457out:
aef4acb6 458 qemu_vfree(iov.iov_base);
1b9f1491 459 return ret;
45aba42f
KW
460}
461
462
463/*
464 * get_cluster_offset
465 *
ecfe1863
KW
466 * For a given offset of the virtual disk, find the cluster type and offset in
467 * the qcow2 file. The offset is stored in *cluster_offset.
45aba42f 468 *
ecfe1863
KW
469 * On entry, *bytes is the maximum number of contiguous bytes starting at
470 * offset that we are interested in.
45aba42f 471 *
ecfe1863
KW
472 * On exit, *bytes is the number of bytes starting at offset that have the same
473 * cluster type and (if applicable) are stored contiguously in the image file.
474 * Compressed clusters are always returned one by one.
45aba42f 475 *
68d000a3
KW
476 * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error
477 * cases.
45aba42f 478 */
1c46efaa 479int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
ecfe1863 480 unsigned int *bytes, uint64_t *cluster_offset)
45aba42f 481{
ff99129a 482 BDRVQcow2State *s = bs->opaque;
2cf7cfa1
KW
483 unsigned int l2_index;
484 uint64_t l1_index, l2_offset, *l2_table;
45aba42f 485 int l1_bits, c;
c834cba9
HR
486 unsigned int offset_in_cluster;
487 uint64_t bytes_available, bytes_needed, nb_clusters;
55c17e98 488 int ret;
45aba42f 489
b2f65d6b 490 offset_in_cluster = offset_into_cluster(s, offset);
ecfe1863 491 bytes_needed = (uint64_t) *bytes + offset_in_cluster;
45aba42f 492
b2f65d6b 493 l1_bits = s->l2_bits + s->cluster_bits;
45aba42f 494
b2f65d6b
KW
495 /* compute how many bytes there are between the start of the cluster
496 * containing offset and the end of the l1 entry */
497 bytes_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1))
498 + offset_in_cluster;
45aba42f 499
b2f65d6b
KW
500 if (bytes_needed > bytes_available) {
501 bytes_needed = bytes_available;
45aba42f
KW
502 }
503
1c46efaa 504 *cluster_offset = 0;
45aba42f 505
b6af0975 506 /* seek to the l2 offset in the l1 table */
45aba42f
KW
507
508 l1_index = offset >> l1_bits;
68d000a3
KW
509 if (l1_index >= s->l1_size) {
510 ret = QCOW2_CLUSTER_UNALLOCATED;
45aba42f 511 goto out;
68d000a3 512 }
45aba42f 513
68d000a3
KW
514 l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
515 if (!l2_offset) {
516 ret = QCOW2_CLUSTER_UNALLOCATED;
45aba42f 517 goto out;
68d000a3 518 }
45aba42f 519
a97c67ee
HR
520 if (offset_into_cluster(s, l2_offset)) {
521 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64
522 " unaligned (L1 index: %#" PRIx64 ")",
523 l2_offset, l1_index);
524 return -EIO;
525 }
526
45aba42f
KW
527 /* load the l2 table in memory */
528
55c17e98
KW
529 ret = l2_load(bs, l2_offset, &l2_table);
530 if (ret < 0) {
531 return ret;
1c46efaa 532 }
45aba42f
KW
533
534 /* find the cluster offset for the given disk offset */
535
536 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
1c46efaa 537 *cluster_offset = be64_to_cpu(l2_table[l2_index]);
b6d36def 538
b2f65d6b 539 nb_clusters = size_to_clusters(s, bytes_needed);
c834cba9
HR
540 /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned
541 * integers; the minimum cluster size is 512, so this assertion is always
542 * true */
543 assert(nb_clusters <= INT_MAX);
45aba42f 544
68d000a3
KW
545 ret = qcow2_get_cluster_type(*cluster_offset);
546 switch (ret) {
547 case QCOW2_CLUSTER_COMPRESSED:
548 /* Compressed clusters can only be processed one by one */
549 c = 1;
550 *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK;
551 break;
6377af48 552 case QCOW2_CLUSTER_ZERO:
381b487d 553 if (s->qcow_version < 3) {
a97c67ee
HR
554 qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found"
555 " in pre-v3 image (L2 offset: %#" PRIx64
556 ", L2 index: %#x)", l2_offset, l2_index);
557 ret = -EIO;
558 goto fail;
381b487d 559 }
a99dfb45
KW
560 c = count_contiguous_clusters_by_type(nb_clusters, &l2_table[l2_index],
561 QCOW2_CLUSTER_ZERO);
6377af48
KW
562 *cluster_offset = 0;
563 break;
68d000a3 564 case QCOW2_CLUSTER_UNALLOCATED:
45aba42f 565 /* how many empty clusters ? */
a99dfb45
KW
566 c = count_contiguous_clusters_by_type(nb_clusters, &l2_table[l2_index],
567 QCOW2_CLUSTER_UNALLOCATED);
68d000a3
KW
568 *cluster_offset = 0;
569 break;
570 case QCOW2_CLUSTER_NORMAL:
45aba42f
KW
571 /* how many allocated clusters ? */
572 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
61653008 573 &l2_table[l2_index], QCOW_OFLAG_ZERO);
68d000a3 574 *cluster_offset &= L2E_OFFSET_MASK;
a97c67ee
HR
575 if (offset_into_cluster(s, *cluster_offset)) {
576 qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset %#"
577 PRIx64 " unaligned (L2 offset: %#" PRIx64
578 ", L2 index: %#x)", *cluster_offset,
579 l2_offset, l2_index);
580 ret = -EIO;
581 goto fail;
582 }
68d000a3 583 break;
1417d7e4
KW
584 default:
585 abort();
45aba42f
KW
586 }
587
29c1a730
KW
588 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
589
c834cba9 590 bytes_available = (int64_t)c * s->cluster_size;
68d000a3 591
45aba42f 592out:
b2f65d6b
KW
593 if (bytes_available > bytes_needed) {
594 bytes_available = bytes_needed;
595 }
45aba42f 596
c834cba9
HR
597 /* bytes_available <= bytes_needed <= *bytes + offset_in_cluster;
598 * subtracting offset_in_cluster will therefore definitely yield something
599 * not exceeding UINT_MAX */
600 assert(bytes_available - offset_in_cluster <= UINT_MAX);
ecfe1863 601 *bytes = bytes_available - offset_in_cluster;
45aba42f 602
68d000a3 603 return ret;
a97c67ee
HR
604
605fail:
606 qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table);
607 return ret;
45aba42f
KW
608}
609
610/*
611 * get_cluster_table
612 *
613 * for a given disk offset, load (and allocate if needed)
614 * the l2 table.
615 *
616 * the l2 table offset in the qcow2 file and the cluster index
617 * in the l2 table are given to the caller.
618 *
1e3e8f1a 619 * Returns 0 on success, -errno in failure case
45aba42f 620 */
45aba42f
KW
621static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
622 uint64_t **new_l2_table,
45aba42f
KW
623 int *new_l2_index)
624{
ff99129a 625 BDRVQcow2State *s = bs->opaque;
2cf7cfa1
KW
626 unsigned int l2_index;
627 uint64_t l1_index, l2_offset;
c46e1167 628 uint64_t *l2_table = NULL;
80ee15a6 629 int ret;
45aba42f 630
b6af0975 631 /* seek to the l2 offset in the l1 table */
45aba42f
KW
632
633 l1_index = offset >> (s->l2_bits + s->cluster_bits);
634 if (l1_index >= s->l1_size) {
72893756 635 ret = qcow2_grow_l1_table(bs, l1_index + 1, false);
1e3e8f1a
KW
636 if (ret < 0) {
637 return ret;
638 }
45aba42f 639 }
8e37f681 640
2cf7cfa1 641 assert(l1_index < s->l1_size);
8e37f681 642 l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
a97c67ee
HR
643 if (offset_into_cluster(s, l2_offset)) {
644 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64
645 " unaligned (L1 index: %#" PRIx64 ")",
646 l2_offset, l1_index);
647 return -EIO;
648 }
45aba42f
KW
649
650 /* seek the l2 table of the given l2 offset */
651
8e37f681 652 if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) {
45aba42f 653 /* load the l2 table in memory */
55c17e98
KW
654 ret = l2_load(bs, l2_offset, &l2_table);
655 if (ret < 0) {
656 return ret;
1e3e8f1a 657 }
45aba42f 658 } else {
16fde5f2 659 /* First allocate a new L2 table (and do COW if needed) */
c46e1167
KW
660 ret = l2_allocate(bs, l1_index, &l2_table);
661 if (ret < 0) {
662 return ret;
1e3e8f1a 663 }
16fde5f2
KW
664
665 /* Then decrease the refcount of the old table */
666 if (l2_offset) {
6cfcb9b8
KW
667 qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
668 QCOW2_DISCARD_OTHER);
16fde5f2 669 }
45aba42f
KW
670 }
671
672 /* find the cluster offset for the given disk offset */
673
674 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
675
676 *new_l2_table = l2_table;
45aba42f
KW
677 *new_l2_index = l2_index;
678
1e3e8f1a 679 return 0;
45aba42f
KW
680}
681
682/*
683 * alloc_compressed_cluster_offset
684 *
685 * For a given offset of the disk image, return cluster offset in
686 * qcow2 file.
687 *
688 * If the offset is not found, allocate a new compressed cluster.
689 *
690 * Return the cluster offset if successful,
691 * Return 0, otherwise.
692 *
693 */
694
ed6ccf0f
KW
695uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
696 uint64_t offset,
697 int compressed_size)
45aba42f 698{
ff99129a 699 BDRVQcow2State *s = bs->opaque;
45aba42f 700 int l2_index, ret;
3948d1d4 701 uint64_t *l2_table;
f4f0d391 702 int64_t cluster_offset;
45aba42f
KW
703 int nb_csectors;
704
3948d1d4 705 ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1e3e8f1a 706 if (ret < 0) {
45aba42f 707 return 0;
1e3e8f1a 708 }
45aba42f 709
b0b6862e
KW
710 /* Compression can't overwrite anything. Fail if the cluster was already
711 * allocated. */
45aba42f 712 cluster_offset = be64_to_cpu(l2_table[l2_index]);
b0b6862e 713 if (cluster_offset & L2E_OFFSET_MASK) {
8f1efd00
KW
714 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
715 return 0;
716 }
45aba42f 717
ed6ccf0f 718 cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
5d757b56 719 if (cluster_offset < 0) {
29c1a730 720 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
5d757b56
KW
721 return 0;
722 }
723
45aba42f
KW
724 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
725 (cluster_offset >> 9);
726
727 cluster_offset |= QCOW_OFLAG_COMPRESSED |
728 ((uint64_t)nb_csectors << s->csize_shift);
729
730 /* update L2 table */
731
732 /* compressed clusters never have the copied flag */
733
66f82cee 734 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
72e80b89 735 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
45aba42f 736 l2_table[l2_index] = cpu_to_be64(cluster_offset);
a3f1afb4 737 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
4c1612d9 738
29c1a730 739 return cluster_offset;
4c1612d9
KW
740}
741
593fb83c
KW
742static int perform_cow(BlockDriverState *bs, QCowL2Meta *m, Qcow2COWRegion *r)
743{
ff99129a 744 BDRVQcow2State *s = bs->opaque;
593fb83c
KW
745 int ret;
746
85567393 747 if (r->nb_bytes == 0) {
593fb83c
KW
748 return 0;
749 }
750
751 qemu_co_mutex_unlock(&s->lock);
85567393 752 ret = do_perform_cow(bs, m->offset, m->alloc_offset, r->offset, r->nb_bytes);
593fb83c
KW
753 qemu_co_mutex_lock(&s->lock);
754
755 if (ret < 0) {
756 return ret;
757 }
758
759 /*
760 * Before we update the L2 table to actually point to the new cluster, we
761 * need to be sure that the refcounts have been increased and COW was
762 * handled.
763 */
764 qcow2_cache_depends_on_flush(s->l2_table_cache);
765
766 return 0;
767}
768
148da7ea 769int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
45aba42f 770{
ff99129a 771 BDRVQcow2State *s = bs->opaque;
45aba42f 772 int i, j = 0, l2_index, ret;
593fb83c 773 uint64_t *old_cluster, *l2_table;
250196f1 774 uint64_t cluster_offset = m->alloc_offset;
45aba42f 775
3cce16f4 776 trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters);
f50f88b9 777 assert(m->nb_clusters > 0);
45aba42f 778
5839e53b 779 old_cluster = g_try_new(uint64_t, m->nb_clusters);
de82815d
KW
780 if (old_cluster == NULL) {
781 ret = -ENOMEM;
782 goto err;
783 }
45aba42f
KW
784
785 /* copy content of unmodified sectors */
593fb83c
KW
786 ret = perform_cow(bs, m, &m->cow_start);
787 if (ret < 0) {
788 goto err;
45aba42f
KW
789 }
790
593fb83c
KW
791 ret = perform_cow(bs, m, &m->cow_end);
792 if (ret < 0) {
793 goto err;
29c1a730
KW
794 }
795
593fb83c 796 /* Update L2 table. */
74c4510a 797 if (s->use_lazy_refcounts) {
280d3735
KW
798 qcow2_mark_dirty(bs);
799 }
bfe8043e
SH
800 if (qcow2_need_accurate_refcounts(s)) {
801 qcow2_cache_set_dependency(bs, s->l2_table_cache,
802 s->refcount_block_cache);
803 }
280d3735 804
3948d1d4 805 ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index);
1e3e8f1a 806 if (ret < 0) {
45aba42f 807 goto err;
1e3e8f1a 808 }
72e80b89 809 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
45aba42f 810
c01dbccb 811 assert(l2_index + m->nb_clusters <= s->l2_size);
45aba42f
KW
812 for (i = 0; i < m->nb_clusters; i++) {
813 /* if two concurrent writes happen to the same unallocated cluster
aaa4d20b
KW
814 * each write allocates separate cluster and writes data concurrently.
815 * The first one to complete updates l2 table with pointer to its
816 * cluster the second one has to do RMW (which is done above by
817 * perform_cow()), update l2 table with its cluster pointer and free
818 * old cluster. This is what this loop does */
819 if (l2_table[l2_index + i] != 0) {
45aba42f 820 old_cluster[j++] = l2_table[l2_index + i];
aaa4d20b 821 }
45aba42f
KW
822
823 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
824 (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
825 }
826
9f8e668e 827
a3f1afb4 828 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
45aba42f 829
7ec5e6a4
KW
830 /*
831 * If this was a COW, we need to decrease the refcount of the old cluster.
6cfcb9b8
KW
832 *
833 * Don't discard clusters that reach a refcount of 0 (e.g. compressed
834 * clusters), the next write will reuse them anyway.
7ec5e6a4
KW
835 */
836 if (j != 0) {
7ec5e6a4 837 for (i = 0; i < j; i++) {
6cfcb9b8
KW
838 qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1,
839 QCOW2_DISCARD_NEVER);
7ec5e6a4
KW
840 }
841 }
45aba42f
KW
842
843 ret = 0;
844err:
7267c094 845 g_free(old_cluster);
45aba42f
KW
846 return ret;
847 }
848
bf319ece
KW
849/*
850 * Returns the number of contiguous clusters that can be used for an allocating
851 * write, but require COW to be performed (this includes yet unallocated space,
852 * which must copy from the backing file)
853 */
ff99129a 854static int count_cow_clusters(BDRVQcow2State *s, int nb_clusters,
bf319ece
KW
855 uint64_t *l2_table, int l2_index)
856{
143550a8 857 int i;
bf319ece 858
143550a8
KW
859 for (i = 0; i < nb_clusters; i++) {
860 uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]);
861 int cluster_type = qcow2_get_cluster_type(l2_entry);
862
863 switch(cluster_type) {
864 case QCOW2_CLUSTER_NORMAL:
865 if (l2_entry & QCOW_OFLAG_COPIED) {
866 goto out;
867 }
bf319ece 868 break;
143550a8
KW
869 case QCOW2_CLUSTER_UNALLOCATED:
870 case QCOW2_CLUSTER_COMPRESSED:
6377af48 871 case QCOW2_CLUSTER_ZERO:
bf319ece 872 break;
143550a8
KW
873 default:
874 abort();
875 }
bf319ece
KW
876 }
877
143550a8 878out:
bf319ece
KW
879 assert(i <= nb_clusters);
880 return i;
881}
882
250196f1 883/*
226c3c26
KW
884 * Check if there already is an AIO write request in flight which allocates
885 * the same cluster. In this case we need to wait until the previous
886 * request has completed and updated the L2 table accordingly.
65eb2e35
KW
887 *
888 * Returns:
889 * 0 if there was no dependency. *cur_bytes indicates the number of
890 * bytes from guest_offset that can be read before the next
891 * dependency must be processed (or the request is complete)
892 *
893 * -EAGAIN if we had to wait for another request, previously gathered
894 * information on cluster allocation may be invalid now. The caller
895 * must start over anyway, so consider *cur_bytes undefined.
250196f1 896 */
226c3c26 897static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset,
ecdd5333 898 uint64_t *cur_bytes, QCowL2Meta **m)
250196f1 899{
ff99129a 900 BDRVQcow2State *s = bs->opaque;
250196f1 901 QCowL2Meta *old_alloc;
65eb2e35 902 uint64_t bytes = *cur_bytes;
250196f1 903
250196f1
KW
904 QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
905
65eb2e35
KW
906 uint64_t start = guest_offset;
907 uint64_t end = start + bytes;
908 uint64_t old_start = l2meta_cow_start(old_alloc);
909 uint64_t old_end = l2meta_cow_end(old_alloc);
250196f1 910
d9d74f41 911 if (end <= old_start || start >= old_end) {
250196f1
KW
912 /* No intersection */
913 } else {
914 if (start < old_start) {
915 /* Stop at the start of a running allocation */
65eb2e35 916 bytes = old_start - start;
250196f1 917 } else {
65eb2e35 918 bytes = 0;
250196f1
KW
919 }
920
ecdd5333
KW
921 /* Stop if already an l2meta exists. After yielding, it wouldn't
922 * be valid any more, so we'd have to clean up the old L2Metas
923 * and deal with requests depending on them before starting to
924 * gather new ones. Not worth the trouble. */
925 if (bytes == 0 && *m) {
926 *cur_bytes = 0;
927 return 0;
928 }
929
65eb2e35 930 if (bytes == 0) {
250196f1
KW
931 /* Wait for the dependency to complete. We need to recheck
932 * the free/allocated clusters when we continue. */
933 qemu_co_mutex_unlock(&s->lock);
934 qemu_co_queue_wait(&old_alloc->dependent_requests);
935 qemu_co_mutex_lock(&s->lock);
936 return -EAGAIN;
937 }
938 }
939 }
940
65eb2e35
KW
941 /* Make sure that existing clusters and new allocations are only used up to
942 * the next dependency if we shortened the request above */
943 *cur_bytes = bytes;
250196f1 944
226c3c26
KW
945 return 0;
946}
947
0af729ec
KW
948/*
949 * Checks how many already allocated clusters that don't require a copy on
950 * write there are at the given guest_offset (up to *bytes). If
951 * *host_offset is not zero, only physically contiguous clusters beginning at
952 * this host offset are counted.
953 *
411d62b0
KW
954 * Note that guest_offset may not be cluster aligned. In this case, the
955 * returned *host_offset points to exact byte referenced by guest_offset and
956 * therefore isn't cluster aligned as well.
0af729ec
KW
957 *
958 * Returns:
959 * 0: if no allocated clusters are available at the given offset.
960 * *bytes is normally unchanged. It is set to 0 if the cluster
961 * is allocated and doesn't need COW, but doesn't have the right
962 * physical offset.
963 *
964 * 1: if allocated clusters that don't require a COW are available at
965 * the requested offset. *bytes may have decreased and describes
966 * the length of the area that can be written to.
967 *
968 * -errno: in error cases
0af729ec
KW
969 */
970static int handle_copied(BlockDriverState *bs, uint64_t guest_offset,
c53ede9f 971 uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
0af729ec 972{
ff99129a 973 BDRVQcow2State *s = bs->opaque;
0af729ec
KW
974 int l2_index;
975 uint64_t cluster_offset;
976 uint64_t *l2_table;
b6d36def 977 uint64_t nb_clusters;
c53ede9f 978 unsigned int keep_clusters;
a3f1afb4 979 int ret;
0af729ec
KW
980
981 trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset,
982 *bytes);
0af729ec 983
411d62b0
KW
984 assert(*host_offset == 0 || offset_into_cluster(s, guest_offset)
985 == offset_into_cluster(s, *host_offset));
986
acb0467f
KW
987 /*
988 * Calculate the number of clusters to look for. We stop at L2 table
989 * boundaries to keep things simple.
990 */
991 nb_clusters =
992 size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
993
994 l2_index = offset_to_l2_index(s, guest_offset);
995 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
b6d36def 996 assert(nb_clusters <= INT_MAX);
acb0467f 997
0af729ec
KW
998 /* Find L2 entry for the first involved cluster */
999 ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index);
1000 if (ret < 0) {
1001 return ret;
1002 }
1003
1004 cluster_offset = be64_to_cpu(l2_table[l2_index]);
1005
1006 /* Check how many clusters are already allocated and don't need COW */
1007 if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL
1008 && (cluster_offset & QCOW_OFLAG_COPIED))
1009 {
e62daaf6
KW
1010 /* If a specific host_offset is required, check it */
1011 bool offset_matches =
1012 (cluster_offset & L2E_OFFSET_MASK) == *host_offset;
1013
a97c67ee
HR
1014 if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) {
1015 qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset "
1016 "%#llx unaligned (guest offset: %#" PRIx64
1017 ")", cluster_offset & L2E_OFFSET_MASK,
1018 guest_offset);
1019 ret = -EIO;
1020 goto out;
1021 }
1022
e62daaf6
KW
1023 if (*host_offset != 0 && !offset_matches) {
1024 *bytes = 0;
1025 ret = 0;
1026 goto out;
1027 }
1028
0af729ec 1029 /* We keep all QCOW_OFLAG_COPIED clusters */
c53ede9f 1030 keep_clusters =
acb0467f 1031 count_contiguous_clusters(nb_clusters, s->cluster_size,
61653008 1032 &l2_table[l2_index],
0af729ec 1033 QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO);
c53ede9f
KW
1034 assert(keep_clusters <= nb_clusters);
1035
1036 *bytes = MIN(*bytes,
1037 keep_clusters * s->cluster_size
1038 - offset_into_cluster(s, guest_offset));
0af729ec
KW
1039
1040 ret = 1;
1041 } else {
0af729ec
KW
1042 ret = 0;
1043 }
1044
0af729ec 1045 /* Cleanup */
e62daaf6 1046out:
a3f1afb4 1047 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
0af729ec 1048
e62daaf6
KW
1049 /* Only return a host offset if we actually made progress. Otherwise we
1050 * would make requirements for handle_alloc() that it can't fulfill */
a97c67ee 1051 if (ret > 0) {
411d62b0
KW
1052 *host_offset = (cluster_offset & L2E_OFFSET_MASK)
1053 + offset_into_cluster(s, guest_offset);
e62daaf6
KW
1054 }
1055
0af729ec
KW
1056 return ret;
1057}
1058
226c3c26
KW
1059/*
1060 * Allocates new clusters for the given guest_offset.
1061 *
1062 * At most *nb_clusters are allocated, and on return *nb_clusters is updated to
1063 * contain the number of clusters that have been allocated and are contiguous
1064 * in the image file.
1065 *
1066 * If *host_offset is non-zero, it specifies the offset in the image file at
1067 * which the new clusters must start. *nb_clusters can be 0 on return in this
1068 * case if the cluster at host_offset is already in use. If *host_offset is
1069 * zero, the clusters can be allocated anywhere in the image file.
1070 *
1071 * *host_offset is updated to contain the offset into the image file at which
1072 * the first allocated cluster starts.
1073 *
1074 * Return 0 on success and -errno in error cases. -EAGAIN means that the
1075 * function has been waiting for another request and the allocation must be
1076 * restarted, but the whole request should not be failed.
1077 */
1078static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset,
b6d36def 1079 uint64_t *host_offset, uint64_t *nb_clusters)
226c3c26 1080{
ff99129a 1081 BDRVQcow2State *s = bs->opaque;
226c3c26
KW
1082
1083 trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset,
1084 *host_offset, *nb_clusters);
1085
250196f1
KW
1086 /* Allocate new clusters */
1087 trace_qcow2_cluster_alloc_phys(qemu_coroutine_self());
1088 if (*host_offset == 0) {
df021791
KW
1089 int64_t cluster_offset =
1090 qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size);
1091 if (cluster_offset < 0) {
1092 return cluster_offset;
1093 }
1094 *host_offset = cluster_offset;
1095 return 0;
250196f1 1096 } else {
b6d36def 1097 int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters);
df021791
KW
1098 if (ret < 0) {
1099 return ret;
1100 }
1101 *nb_clusters = ret;
1102 return 0;
250196f1 1103 }
250196f1
KW
1104}
1105
10f0ed8b
KW
1106/*
1107 * Allocates new clusters for an area that either is yet unallocated or needs a
1108 * copy on write. If *host_offset is non-zero, clusters are only allocated if
1109 * the new allocation can match the specified host offset.
1110 *
411d62b0
KW
1111 * Note that guest_offset may not be cluster aligned. In this case, the
1112 * returned *host_offset points to exact byte referenced by guest_offset and
1113 * therefore isn't cluster aligned as well.
10f0ed8b
KW
1114 *
1115 * Returns:
1116 * 0: if no clusters could be allocated. *bytes is set to 0,
1117 * *host_offset is left unchanged.
1118 *
1119 * 1: if new clusters were allocated. *bytes may be decreased if the
1120 * new allocation doesn't cover all of the requested area.
1121 * *host_offset is updated to contain the host offset of the first
1122 * newly allocated cluster.
1123 *
1124 * -errno: in error cases
10f0ed8b
KW
1125 */
1126static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
c37f4cd7 1127 uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
10f0ed8b 1128{
ff99129a 1129 BDRVQcow2State *s = bs->opaque;
10f0ed8b
KW
1130 int l2_index;
1131 uint64_t *l2_table;
1132 uint64_t entry;
b6d36def 1133 uint64_t nb_clusters;
10f0ed8b
KW
1134 int ret;
1135
10f0ed8b 1136 uint64_t alloc_cluster_offset;
10f0ed8b
KW
1137
1138 trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset,
1139 *bytes);
1140 assert(*bytes > 0);
1141
f5bc6350
KW
1142 /*
1143 * Calculate the number of clusters to look for. We stop at L2 table
1144 * boundaries to keep things simple.
1145 */
c37f4cd7
KW
1146 nb_clusters =
1147 size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
1148
f5bc6350 1149 l2_index = offset_to_l2_index(s, guest_offset);
c37f4cd7 1150 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
b6d36def 1151 assert(nb_clusters <= INT_MAX);
f5bc6350 1152
10f0ed8b
KW
1153 /* Find L2 entry for the first involved cluster */
1154 ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index);
1155 if (ret < 0) {
1156 return ret;
1157 }
1158
3b8e2e26 1159 entry = be64_to_cpu(l2_table[l2_index]);
10f0ed8b
KW
1160
1161 /* For the moment, overwrite compressed clusters one by one */
1162 if (entry & QCOW_OFLAG_COMPRESSED) {
1163 nb_clusters = 1;
1164 } else {
3b8e2e26 1165 nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index);
10f0ed8b
KW
1166 }
1167
ecdd5333
KW
1168 /* This function is only called when there were no non-COW clusters, so if
1169 * we can't find any unallocated or COW clusters either, something is
1170 * wrong with our code. */
1171 assert(nb_clusters > 0);
1172
a3f1afb4 1173 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
10f0ed8b 1174
10f0ed8b 1175 /* Allocate, if necessary at a given offset in the image file */
411d62b0 1176 alloc_cluster_offset = start_of_cluster(s, *host_offset);
83baa9a4 1177 ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset,
10f0ed8b
KW
1178 &nb_clusters);
1179 if (ret < 0) {
1180 goto fail;
1181 }
1182
83baa9a4
KW
1183 /* Can't extend contiguous allocation */
1184 if (nb_clusters == 0) {
10f0ed8b
KW
1185 *bytes = 0;
1186 return 0;
1187 }
1188
ff52aab2
HR
1189 /* !*host_offset would overwrite the image header and is reserved for "no
1190 * host offset preferred". If 0 was a valid host offset, it'd trigger the
1191 * following overlap check; do that now to avoid having an invalid value in
1192 * *host_offset. */
1193 if (!alloc_cluster_offset) {
1194 ret = qcow2_pre_write_overlap_check(bs, 0, alloc_cluster_offset,
1195 nb_clusters * s->cluster_size);
1196 assert(ret < 0);
1197 goto fail;
1198 }
1199
83baa9a4
KW
1200 /*
1201 * Save info needed for meta data update.
1202 *
85567393 1203 * requested_bytes: Number of bytes from the start of the first
83baa9a4
KW
1204 * newly allocated cluster to the end of the (possibly shortened
1205 * before) write request.
1206 *
85567393 1207 * avail_bytes: Number of bytes from the start of the first
83baa9a4
KW
1208 * newly allocated to the end of the last newly allocated cluster.
1209 *
85567393 1210 * nb_bytes: The number of bytes from the start of the first
83baa9a4
KW
1211 * newly allocated cluster to the end of the area that the write
1212 * request actually writes to (excluding COW at the end)
1213 */
85567393
KW
1214 uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset);
1215 int avail_bytes = MIN(INT_MAX, nb_clusters << s->cluster_bits);
1216 int nb_bytes = MIN(requested_bytes, avail_bytes);
88c6588c 1217 QCowL2Meta *old_m = *m;
83baa9a4 1218
83baa9a4
KW
1219 *m = g_malloc0(sizeof(**m));
1220
1221 **m = (QCowL2Meta) {
88c6588c
KW
1222 .next = old_m,
1223
411d62b0 1224 .alloc_offset = alloc_cluster_offset,
83baa9a4
KW
1225 .offset = start_of_cluster(s, guest_offset),
1226 .nb_clusters = nb_clusters,
83baa9a4
KW
1227
1228 .cow_start = {
1229 .offset = 0,
85567393 1230 .nb_bytes = offset_into_cluster(s, guest_offset),
83baa9a4
KW
1231 },
1232 .cow_end = {
85567393
KW
1233 .offset = nb_bytes,
1234 .nb_bytes = avail_bytes - nb_bytes,
83baa9a4
KW
1235 },
1236 };
1237 qemu_co_queue_init(&(*m)->dependent_requests);
1238 QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
1239
411d62b0 1240 *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset);
85567393 1241 *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset));
83baa9a4
KW
1242 assert(*bytes != 0);
1243
10f0ed8b
KW
1244 return 1;
1245
1246fail:
1247 if (*m && (*m)->nb_clusters > 0) {
1248 QLIST_REMOVE(*m, next_in_flight);
1249 }
1250 return ret;
1251}
1252
45aba42f
KW
1253/*
1254 * alloc_cluster_offset
1255 *
250196f1
KW
1256 * For a given offset on the virtual disk, find the cluster offset in qcow2
1257 * file. If the offset is not found, allocate a new cluster.
45aba42f 1258 *
250196f1 1259 * If the cluster was already allocated, m->nb_clusters is set to 0 and
a7912369 1260 * other fields in m are meaningless.
148da7ea
KW
1261 *
1262 * If the cluster is newly allocated, m->nb_clusters is set to the number of
68d100e9
KW
1263 * contiguous clusters that have been allocated. In this case, the other
1264 * fields of m are valid and contain information about the first allocated
1265 * cluster.
45aba42f 1266 *
68d100e9
KW
1267 * If the request conflicts with another write request in flight, the coroutine
1268 * is queued and will be reentered when the dependency has completed.
148da7ea
KW
1269 *
1270 * Return 0 on success and -errno in error cases
45aba42f 1271 */
f4f0d391 1272int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
d46a0bb2
KW
1273 unsigned int *bytes, uint64_t *host_offset,
1274 QCowL2Meta **m)
45aba42f 1275{
ff99129a 1276 BDRVQcow2State *s = bs->opaque;
710c2496 1277 uint64_t start, remaining;
250196f1 1278 uint64_t cluster_offset;
65eb2e35 1279 uint64_t cur_bytes;
710c2496 1280 int ret;
45aba42f 1281
d46a0bb2 1282 trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *bytes);
710c2496 1283
72424114 1284again:
16f0587e 1285 start = offset;
d46a0bb2 1286 remaining = *bytes;
0af729ec
KW
1287 cluster_offset = 0;
1288 *host_offset = 0;
ecdd5333
KW
1289 cur_bytes = 0;
1290 *m = NULL;
0af729ec 1291
2c3b32d2 1292 while (true) {
ecdd5333
KW
1293
1294 if (!*host_offset) {
1295 *host_offset = start_of_cluster(s, cluster_offset);
1296 }
1297
1298 assert(remaining >= cur_bytes);
1299
1300 start += cur_bytes;
1301 remaining -= cur_bytes;
1302 cluster_offset += cur_bytes;
1303
1304 if (remaining == 0) {
1305 break;
1306 }
1307
1308 cur_bytes = remaining;
1309
2c3b32d2
KW
1310 /*
1311 * Now start gathering as many contiguous clusters as possible:
1312 *
1313 * 1. Check for overlaps with in-flight allocations
1314 *
1315 * a) Overlap not in the first cluster -> shorten this request and
1316 * let the caller handle the rest in its next loop iteration.
1317 *
1318 * b) Real overlaps of two requests. Yield and restart the search
1319 * for contiguous clusters (the situation could have changed
1320 * while we were sleeping)
1321 *
1322 * c) TODO: Request starts in the same cluster as the in-flight
1323 * allocation ends. Shorten the COW of the in-fight allocation,
1324 * set cluster_offset to write to the same cluster and set up
1325 * the right synchronisation between the in-flight request and
1326 * the new one.
1327 */
ecdd5333 1328 ret = handle_dependencies(bs, start, &cur_bytes, m);
2c3b32d2 1329 if (ret == -EAGAIN) {
ecdd5333
KW
1330 /* Currently handle_dependencies() doesn't yield if we already had
1331 * an allocation. If it did, we would have to clean up the L2Meta
1332 * structs before starting over. */
1333 assert(*m == NULL);
2c3b32d2
KW
1334 goto again;
1335 } else if (ret < 0) {
1336 return ret;
ecdd5333
KW
1337 } else if (cur_bytes == 0) {
1338 break;
2c3b32d2
KW
1339 } else {
1340 /* handle_dependencies() may have decreased cur_bytes (shortened
1341 * the allocations below) so that the next dependency is processed
1342 * correctly during the next loop iteration. */
0af729ec 1343 }
710c2496 1344
2c3b32d2
KW
1345 /*
1346 * 2. Count contiguous COPIED clusters.
1347 */
1348 ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m);
1349 if (ret < 0) {
1350 return ret;
1351 } else if (ret) {
ecdd5333 1352 continue;
2c3b32d2
KW
1353 } else if (cur_bytes == 0) {
1354 break;
1355 }
060bee89 1356
2c3b32d2
KW
1357 /*
1358 * 3. If the request still hasn't completed, allocate new clusters,
1359 * considering any cluster_offset of steps 1c or 2.
1360 */
1361 ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m);
1362 if (ret < 0) {
1363 return ret;
1364 } else if (ret) {
ecdd5333 1365 continue;
2c3b32d2
KW
1366 } else {
1367 assert(cur_bytes == 0);
1368 break;
1369 }
f5bc6350 1370 }
10f0ed8b 1371
d46a0bb2
KW
1372 *bytes -= remaining;
1373 assert(*bytes > 0);
710c2496 1374 assert(*host_offset != 0);
45aba42f 1375
148da7ea 1376 return 0;
45aba42f
KW
1377}
1378
1379static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1380 const uint8_t *buf, int buf_size)
1381{
1382 z_stream strm1, *strm = &strm1;
1383 int ret, out_len;
1384
1385 memset(strm, 0, sizeof(*strm));
1386
1387 strm->next_in = (uint8_t *)buf;
1388 strm->avail_in = buf_size;
1389 strm->next_out = out_buf;
1390 strm->avail_out = out_buf_size;
1391
1392 ret = inflateInit2(strm, -12);
1393 if (ret != Z_OK)
1394 return -1;
1395 ret = inflate(strm, Z_FINISH);
1396 out_len = strm->next_out - out_buf;
1397 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1398 out_len != out_buf_size) {
1399 inflateEnd(strm);
1400 return -1;
1401 }
1402 inflateEnd(strm);
1403 return 0;
1404}
1405
66f82cee 1406int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
45aba42f 1407{
ff99129a 1408 BDRVQcow2State *s = bs->opaque;
45aba42f
KW
1409 int ret, csize, nb_csectors, sector_offset;
1410 uint64_t coffset;
1411
1412 coffset = cluster_offset & s->cluster_offset_mask;
1413 if (s->cluster_cache_offset != coffset) {
1414 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1415 sector_offset = coffset & 511;
1416 csize = nb_csectors * 512 - sector_offset;
66f82cee 1417 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
fbcbbf4e 1418 ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data,
9a4f4c31 1419 nb_csectors);
45aba42f 1420 if (ret < 0) {
8af36488 1421 return ret;
45aba42f
KW
1422 }
1423 if (decompress_buffer(s->cluster_cache, s->cluster_size,
1424 s->cluster_data + sector_offset, csize) < 0) {
8af36488 1425 return -EIO;
45aba42f
KW
1426 }
1427 s->cluster_cache_offset = coffset;
1428 }
1429 return 0;
1430}
5ea929e3
KW
1431
1432/*
1433 * This discards as many clusters of nb_clusters as possible at once (i.e.
1434 * all clusters in the same L2 table) and returns the number of discarded
1435 * clusters.
1436 */
1437static int discard_single_l2(BlockDriverState *bs, uint64_t offset,
b6d36def
HR
1438 uint64_t nb_clusters, enum qcow2_discard_type type,
1439 bool full_discard)
5ea929e3 1440{
ff99129a 1441 BDRVQcow2State *s = bs->opaque;
3948d1d4 1442 uint64_t *l2_table;
5ea929e3
KW
1443 int l2_index;
1444 int ret;
1445 int i;
1446
3948d1d4 1447 ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
5ea929e3
KW
1448 if (ret < 0) {
1449 return ret;
1450 }
1451
1452 /* Limit nb_clusters to one L2 table */
1453 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
b6d36def 1454 assert(nb_clusters <= INT_MAX);
5ea929e3
KW
1455
1456 for (i = 0; i < nb_clusters; i++) {
c883db0d 1457 uint64_t old_l2_entry;
5ea929e3 1458
c883db0d 1459 old_l2_entry = be64_to_cpu(l2_table[l2_index + i]);
a71835a0
KW
1460
1461 /*
808c4b6f
HR
1462 * If full_discard is false, make sure that a discarded area reads back
1463 * as zeroes for v3 images (we cannot do it for v2 without actually
1464 * writing a zero-filled buffer). We can skip the operation if the
1465 * cluster is already marked as zero, or if it's unallocated and we
1466 * don't have a backing file.
a71835a0
KW
1467 *
1468 * TODO We might want to use bdrv_get_block_status(bs) here, but we're
1469 * holding s->lock, so that doesn't work today.
808c4b6f
HR
1470 *
1471 * If full_discard is true, the sector should not read back as zeroes,
1472 * but rather fall through to the backing file.
a71835a0 1473 */
c883db0d
HR
1474 switch (qcow2_get_cluster_type(old_l2_entry)) {
1475 case QCOW2_CLUSTER_UNALLOCATED:
760e0063 1476 if (full_discard || !bs->backing) {
c883db0d
HR
1477 continue;
1478 }
1479 break;
1480
1481 case QCOW2_CLUSTER_ZERO:
808c4b6f
HR
1482 if (!full_discard) {
1483 continue;
1484 }
1485 break;
c883db0d
HR
1486
1487 case QCOW2_CLUSTER_NORMAL:
1488 case QCOW2_CLUSTER_COMPRESSED:
1489 break;
1490
1491 default:
1492 abort();
5ea929e3
KW
1493 }
1494
1495 /* First remove L2 entries */
72e80b89 1496 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
808c4b6f 1497 if (!full_discard && s->qcow_version >= 3) {
a71835a0
KW
1498 l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
1499 } else {
1500 l2_table[l2_index + i] = cpu_to_be64(0);
1501 }
5ea929e3
KW
1502
1503 /* Then decrease the refcount */
c883db0d 1504 qcow2_free_any_clusters(bs, old_l2_entry, 1, type);
5ea929e3
KW
1505 }
1506
a3f1afb4 1507 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
5ea929e3
KW
1508
1509 return nb_clusters;
1510}
1511
1512int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
808c4b6f 1513 int nb_sectors, enum qcow2_discard_type type, bool full_discard)
5ea929e3 1514{
ff99129a 1515 BDRVQcow2State *s = bs->opaque;
5ea929e3 1516 uint64_t end_offset;
b6d36def 1517 uint64_t nb_clusters;
5ea929e3
KW
1518 int ret;
1519
1520 end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS);
1521
1522 /* Round start up and end down */
1523 offset = align_offset(offset, s->cluster_size);
ac95acdb 1524 end_offset = start_of_cluster(s, end_offset);
5ea929e3
KW
1525
1526 if (offset > end_offset) {
1527 return 0;
1528 }
1529
1530 nb_clusters = size_to_clusters(s, end_offset - offset);
1531
0b919fae
KW
1532 s->cache_discards = true;
1533
5ea929e3
KW
1534 /* Each L2 table is handled by its own loop iteration */
1535 while (nb_clusters > 0) {
808c4b6f 1536 ret = discard_single_l2(bs, offset, nb_clusters, type, full_discard);
5ea929e3 1537 if (ret < 0) {
0b919fae 1538 goto fail;
5ea929e3
KW
1539 }
1540
1541 nb_clusters -= ret;
1542 offset += (ret * s->cluster_size);
1543 }
1544
0b919fae
KW
1545 ret = 0;
1546fail:
1547 s->cache_discards = false;
1548 qcow2_process_discards(bs, ret);
1549
1550 return ret;
5ea929e3 1551}
621f0589
KW
1552
1553/*
1554 * This zeroes as many clusters of nb_clusters as possible at once (i.e.
1555 * all clusters in the same L2 table) and returns the number of zeroed
1556 * clusters.
1557 */
1558static int zero_single_l2(BlockDriverState *bs, uint64_t offset,
b6d36def 1559 uint64_t nb_clusters)
621f0589 1560{
ff99129a 1561 BDRVQcow2State *s = bs->opaque;
621f0589
KW
1562 uint64_t *l2_table;
1563 int l2_index;
1564 int ret;
1565 int i;
1566
1567 ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1568 if (ret < 0) {
1569 return ret;
1570 }
1571
1572 /* Limit nb_clusters to one L2 table */
1573 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
b6d36def 1574 assert(nb_clusters <= INT_MAX);
621f0589
KW
1575
1576 for (i = 0; i < nb_clusters; i++) {
1577 uint64_t old_offset;
1578
1579 old_offset = be64_to_cpu(l2_table[l2_index + i]);
1580
1581 /* Update L2 entries */
72e80b89 1582 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
621f0589
KW
1583 if (old_offset & QCOW_OFLAG_COMPRESSED) {
1584 l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
6cfcb9b8 1585 qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST);
621f0589
KW
1586 } else {
1587 l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO);
1588 }
1589 }
1590
a3f1afb4 1591 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
621f0589
KW
1592
1593 return nb_clusters;
1594}
1595
1596int qcow2_zero_clusters(BlockDriverState *bs, uint64_t offset, int nb_sectors)
1597{
ff99129a 1598 BDRVQcow2State *s = bs->opaque;
b6d36def 1599 uint64_t nb_clusters;
621f0589
KW
1600 int ret;
1601
1602 /* The zero flag is only supported by version 3 and newer */
1603 if (s->qcow_version < 3) {
1604 return -ENOTSUP;
1605 }
1606
1607 /* Each L2 table is handled by its own loop iteration */
1608 nb_clusters = size_to_clusters(s, nb_sectors << BDRV_SECTOR_BITS);
1609
0b919fae
KW
1610 s->cache_discards = true;
1611
621f0589
KW
1612 while (nb_clusters > 0) {
1613 ret = zero_single_l2(bs, offset, nb_clusters);
1614 if (ret < 0) {
0b919fae 1615 goto fail;
621f0589
KW
1616 }
1617
1618 nb_clusters -= ret;
1619 offset += (ret * s->cluster_size);
1620 }
1621
0b919fae
KW
1622 ret = 0;
1623fail:
1624 s->cache_discards = false;
1625 qcow2_process_discards(bs, ret);
1626
1627 return ret;
621f0589 1628}
32b6444d
HR
1629
1630/*
1631 * Expands all zero clusters in a specific L1 table (or deallocates them, for
1632 * non-backed non-pre-allocated zero clusters).
1633 *
4057a2b2
HR
1634 * l1_entries and *visited_l1_entries are used to keep track of progress for
1635 * status_cb(). l1_entries contains the total number of L1 entries and
1636 * *visited_l1_entries counts all visited L1 entries.
32b6444d
HR
1637 */
1638static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
ecf58777 1639 int l1_size, int64_t *visited_l1_entries,
4057a2b2 1640 int64_t l1_entries,
8b13976d
HR
1641 BlockDriverAmendStatusCB *status_cb,
1642 void *cb_opaque)
32b6444d 1643{
ff99129a 1644 BDRVQcow2State *s = bs->opaque;
32b6444d
HR
1645 bool is_active_l1 = (l1_table == s->l1_table);
1646 uint64_t *l2_table = NULL;
1647 int ret;
1648 int i, j;
1649
1650 if (!is_active_l1) {
1651 /* inactive L2 tables require a buffer to be stored in when loading
1652 * them from disk */
9a4f4c31 1653 l2_table = qemu_try_blockalign(bs->file->bs, s->cluster_size);
de82815d
KW
1654 if (l2_table == NULL) {
1655 return -ENOMEM;
1656 }
32b6444d
HR
1657 }
1658
1659 for (i = 0; i < l1_size; i++) {
1660 uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK;
1661 bool l2_dirty = false;
0e06528e 1662 uint64_t l2_refcount;
32b6444d
HR
1663
1664 if (!l2_offset) {
1665 /* unallocated */
4057a2b2
HR
1666 (*visited_l1_entries)++;
1667 if (status_cb) {
8b13976d 1668 status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque);
4057a2b2 1669 }
32b6444d
HR
1670 continue;
1671 }
1672
8dd93d93
HR
1673 if (offset_into_cluster(s, l2_offset)) {
1674 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#"
1675 PRIx64 " unaligned (L1 index: %#x)",
1676 l2_offset, i);
1677 ret = -EIO;
1678 goto fail;
1679 }
1680
32b6444d
HR
1681 if (is_active_l1) {
1682 /* get active L2 tables from cache */
1683 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
1684 (void **)&l2_table);
1685 } else {
1686 /* load inactive L2 tables from disk */
fbcbbf4e 1687 ret = bdrv_read(bs->file, l2_offset / BDRV_SECTOR_SIZE,
9a4f4c31 1688 (void *)l2_table, s->cluster_sectors);
32b6444d
HR
1689 }
1690 if (ret < 0) {
1691 goto fail;
1692 }
1693
7324c10f
HR
1694 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1695 &l2_refcount);
1696 if (ret < 0) {
ecf58777
HR
1697 goto fail;
1698 }
1699
32b6444d
HR
1700 for (j = 0; j < s->l2_size; j++) {
1701 uint64_t l2_entry = be64_to_cpu(l2_table[j]);
ecf58777 1702 int64_t offset = l2_entry & L2E_OFFSET_MASK;
32b6444d 1703 int cluster_type = qcow2_get_cluster_type(l2_entry);
320c7066 1704 bool preallocated = offset != 0;
32b6444d 1705
ecf58777 1706 if (cluster_type != QCOW2_CLUSTER_ZERO) {
32b6444d
HR
1707 continue;
1708 }
1709
320c7066 1710 if (!preallocated) {
760e0063 1711 if (!bs->backing) {
32b6444d
HR
1712 /* not backed; therefore we can simply deallocate the
1713 * cluster */
1714 l2_table[j] = 0;
1715 l2_dirty = true;
1716 continue;
1717 }
1718
1719 offset = qcow2_alloc_clusters(bs, s->cluster_size);
1720 if (offset < 0) {
1721 ret = offset;
1722 goto fail;
1723 }
ecf58777
HR
1724
1725 if (l2_refcount > 1) {
1726 /* For shared L2 tables, set the refcount accordingly (it is
1727 * already 1 and needs to be l2_refcount) */
1728 ret = qcow2_update_cluster_refcount(bs,
2aabe7c7
HR
1729 offset >> s->cluster_bits,
1730 refcount_diff(1, l2_refcount), false,
ecf58777
HR
1731 QCOW2_DISCARD_OTHER);
1732 if (ret < 0) {
1733 qcow2_free_clusters(bs, offset, s->cluster_size,
1734 QCOW2_DISCARD_OTHER);
1735 goto fail;
1736 }
1737 }
32b6444d
HR
1738 }
1739
8dd93d93
HR
1740 if (offset_into_cluster(s, offset)) {
1741 qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset "
1742 "%#" PRIx64 " unaligned (L2 offset: %#"
1743 PRIx64 ", L2 index: %#x)", offset,
1744 l2_offset, j);
1745 if (!preallocated) {
1746 qcow2_free_clusters(bs, offset, s->cluster_size,
1747 QCOW2_DISCARD_ALWAYS);
1748 }
1749 ret = -EIO;
1750 goto fail;
1751 }
1752
231bb267 1753 ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size);
32b6444d 1754 if (ret < 0) {
320c7066
HR
1755 if (!preallocated) {
1756 qcow2_free_clusters(bs, offset, s->cluster_size,
1757 QCOW2_DISCARD_ALWAYS);
1758 }
32b6444d
HR
1759 goto fail;
1760 }
1761
720ff280 1762 ret = bdrv_pwrite_zeroes(bs->file, offset, s->cluster_size, 0);
32b6444d 1763 if (ret < 0) {
320c7066
HR
1764 if (!preallocated) {
1765 qcow2_free_clusters(bs, offset, s->cluster_size,
1766 QCOW2_DISCARD_ALWAYS);
1767 }
32b6444d
HR
1768 goto fail;
1769 }
1770
ecf58777
HR
1771 if (l2_refcount == 1) {
1772 l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED);
1773 } else {
1774 l2_table[j] = cpu_to_be64(offset);
e390cf5a 1775 }
ecf58777 1776 l2_dirty = true;
32b6444d
HR
1777 }
1778
1779 if (is_active_l1) {
1780 if (l2_dirty) {
72e80b89 1781 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
32b6444d
HR
1782 qcow2_cache_depends_on_flush(s->l2_table_cache);
1783 }
a3f1afb4 1784 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
32b6444d
HR
1785 } else {
1786 if (l2_dirty) {
231bb267
HR
1787 ret = qcow2_pre_write_overlap_check(bs,
1788 QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, l2_offset,
32b6444d
HR
1789 s->cluster_size);
1790 if (ret < 0) {
1791 goto fail;
1792 }
1793
18d51c4b 1794 ret = bdrv_write(bs->file, l2_offset / BDRV_SECTOR_SIZE,
9a4f4c31 1795 (void *)l2_table, s->cluster_sectors);
32b6444d
HR
1796 if (ret < 0) {
1797 goto fail;
1798 }
1799 }
1800 }
4057a2b2
HR
1801
1802 (*visited_l1_entries)++;
1803 if (status_cb) {
8b13976d 1804 status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque);
4057a2b2 1805 }
32b6444d
HR
1806 }
1807
1808 ret = 0;
1809
1810fail:
1811 if (l2_table) {
1812 if (!is_active_l1) {
1813 qemu_vfree(l2_table);
1814 } else {
a3f1afb4 1815 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
32b6444d
HR
1816 }
1817 }
1818 return ret;
1819}
1820
1821/*
1822 * For backed images, expands all zero clusters on the image. For non-backed
1823 * images, deallocates all non-pre-allocated zero clusters (and claims the
1824 * allocation for pre-allocated ones). This is important for downgrading to a
1825 * qcow2 version which doesn't yet support metadata zero clusters.
1826 */
4057a2b2 1827int qcow2_expand_zero_clusters(BlockDriverState *bs,
8b13976d
HR
1828 BlockDriverAmendStatusCB *status_cb,
1829 void *cb_opaque)
32b6444d 1830{
ff99129a 1831 BDRVQcow2State *s = bs->opaque;
32b6444d 1832 uint64_t *l1_table = NULL;
4057a2b2 1833 int64_t l1_entries = 0, visited_l1_entries = 0;
32b6444d
HR
1834 int ret;
1835 int i, j;
1836
4057a2b2
HR
1837 if (status_cb) {
1838 l1_entries = s->l1_size;
1839 for (i = 0; i < s->nb_snapshots; i++) {
1840 l1_entries += s->snapshots[i].l1_size;
1841 }
1842 }
1843
32b6444d 1844 ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size,
4057a2b2 1845 &visited_l1_entries, l1_entries,
8b13976d 1846 status_cb, cb_opaque);
32b6444d
HR
1847 if (ret < 0) {
1848 goto fail;
1849 }
1850
1851 /* Inactive L1 tables may point to active L2 tables - therefore it is
1852 * necessary to flush the L2 table cache before trying to access the L2
1853 * tables pointed to by inactive L1 entries (else we might try to expand
1854 * zero clusters that have already been expanded); furthermore, it is also
1855 * necessary to empty the L2 table cache, since it may contain tables which
1856 * are now going to be modified directly on disk, bypassing the cache.
1857 * qcow2_cache_empty() does both for us. */
1858 ret = qcow2_cache_empty(bs, s->l2_table_cache);
1859 if (ret < 0) {
1860 goto fail;
1861 }
1862
1863 for (i = 0; i < s->nb_snapshots; i++) {
d737b78c
LV
1864 int l1_sectors = DIV_ROUND_UP(s->snapshots[i].l1_size *
1865 sizeof(uint64_t), BDRV_SECTOR_SIZE);
32b6444d
HR
1866
1867 l1_table = g_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE);
1868
fbcbbf4e 1869 ret = bdrv_read(bs->file,
9a4f4c31
KW
1870 s->snapshots[i].l1_table_offset / BDRV_SECTOR_SIZE,
1871 (void *)l1_table, l1_sectors);
32b6444d
HR
1872 if (ret < 0) {
1873 goto fail;
1874 }
1875
1876 for (j = 0; j < s->snapshots[i].l1_size; j++) {
1877 be64_to_cpus(&l1_table[j]);
1878 }
1879
1880 ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size,
4057a2b2 1881 &visited_l1_entries, l1_entries,
8b13976d 1882 status_cb, cb_opaque);
32b6444d
HR
1883 if (ret < 0) {
1884 goto fail;
1885 }
1886 }
1887
1888 ret = 0;
1889
1890fail:
32b6444d
HR
1891 g_free(l1_table);
1892 return ret;
1893}