linux-user: Use safe_syscall for open and openat system calls
[qemu.git] / block / qcow2-cluster.c
1 /*
2 * Block driver for the QCOW version 2 format
3 *
4 * Copyright (c) 2004-2006 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "qemu/osdep.h"
26 #include <zlib.h>
27
28 #include "qapi/error.h"
29 #include "qemu-common.h"
30 #include "block/block_int.h"
31 #include "block/qcow2.h"
32 #include "qemu/bswap.h"
33 #include "trace.h"
34
35 int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
36 bool exact_size)
37 {
38 BDRVQcow2State *s = bs->opaque;
39 int new_l1_size2, ret, i;
40 uint64_t *new_l1_table;
41 int64_t old_l1_table_offset, old_l1_size;
42 int64_t new_l1_table_offset, new_l1_size;
43 uint8_t data[12];
44
45 if (min_size <= s->l1_size)
46 return 0;
47
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
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 }
66 }
67
68 if (new_l1_size > INT_MAX / sizeof(uint64_t)) {
69 return -EFBIG;
70 }
71
72 #ifdef DEBUG_ALLOC2
73 fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n",
74 s->l1_size, new_l1_size);
75 #endif
76
77 new_l1_size2 = sizeof(uint64_t) * new_l1_size;
78 new_l1_table = qemu_try_blockalign(bs->file->bs,
79 align_offset(new_l1_size2, 512));
80 if (new_l1_table == NULL) {
81 return -ENOMEM;
82 }
83 memset(new_l1_table, 0, align_offset(new_l1_size2, 512));
84
85 memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
86
87 /* write new table (align to cluster) */
88 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE);
89 new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
90 if (new_l1_table_offset < 0) {
91 qemu_vfree(new_l1_table);
92 return new_l1_table_offset;
93 }
94
95 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
96 if (ret < 0) {
97 goto fail;
98 }
99
100 /* the L1 position has not yet been updated, so these clusters must
101 * indeed be completely free */
102 ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset,
103 new_l1_size2);
104 if (ret < 0) {
105 goto fail;
106 }
107
108 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE);
109 for(i = 0; i < s->l1_size; i++)
110 new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
111 ret = bdrv_pwrite_sync(bs->file->bs, new_l1_table_offset,
112 new_l1_table, new_l1_size2);
113 if (ret < 0)
114 goto fail;
115 for(i = 0; i < s->l1_size; i++)
116 new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
117
118 /* set new table */
119 BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE);
120 cpu_to_be32w((uint32_t*)data, new_l1_size);
121 stq_be_p(data + 4, new_l1_table_offset);
122 ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, l1_size),
123 data, sizeof(data));
124 if (ret < 0) {
125 goto fail;
126 }
127 qemu_vfree(s->l1_table);
128 old_l1_table_offset = s->l1_table_offset;
129 s->l1_table_offset = new_l1_table_offset;
130 s->l1_table = new_l1_table;
131 old_l1_size = s->l1_size;
132 s->l1_size = new_l1_size;
133 qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * sizeof(uint64_t),
134 QCOW2_DISCARD_OTHER);
135 return 0;
136 fail:
137 qemu_vfree(new_l1_table);
138 qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2,
139 QCOW2_DISCARD_OTHER);
140 return ret;
141 }
142
143 /*
144 * l2_load
145 *
146 * Loads a L2 table into memory. If the table is in the cache, the cache
147 * is used; otherwise the L2 table is loaded from the image file.
148 *
149 * Returns a pointer to the L2 table on success, or NULL if the read from
150 * the image file failed.
151 */
152
153 static int l2_load(BlockDriverState *bs, uint64_t l2_offset,
154 uint64_t **l2_table)
155 {
156 BDRVQcow2State *s = bs->opaque;
157 int ret;
158
159 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, (void**) l2_table);
160
161 return ret;
162 }
163
164 /*
165 * Writes one sector of the L1 table to the disk (can't update single entries
166 * and we really don't want bdrv_pread to perform a read-modify-write)
167 */
168 #define L1_ENTRIES_PER_SECTOR (512 / 8)
169 int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index)
170 {
171 BDRVQcow2State *s = bs->opaque;
172 uint64_t buf[L1_ENTRIES_PER_SECTOR] = { 0 };
173 int l1_start_index;
174 int i, ret;
175
176 l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
177 for (i = 0; i < L1_ENTRIES_PER_SECTOR && l1_start_index + i < s->l1_size;
178 i++)
179 {
180 buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
181 }
182
183 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1,
184 s->l1_table_offset + 8 * l1_start_index, sizeof(buf));
185 if (ret < 0) {
186 return ret;
187 }
188
189 BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
190 ret = bdrv_pwrite_sync(bs->file->bs,
191 s->l1_table_offset + 8 * l1_start_index,
192 buf, sizeof(buf));
193 if (ret < 0) {
194 return ret;
195 }
196
197 return 0;
198 }
199
200 /*
201 * l2_allocate
202 *
203 * Allocate a new l2 entry in the file. If l1_index points to an already
204 * used entry in the L2 table (i.e. we are doing a copy on write for the L2
205 * table) copy the contents of the old L2 table into the newly allocated one.
206 * Otherwise the new table is initialized with zeros.
207 *
208 */
209
210 static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
211 {
212 BDRVQcow2State *s = bs->opaque;
213 uint64_t old_l2_offset;
214 uint64_t *l2_table = NULL;
215 int64_t l2_offset;
216 int ret;
217
218 old_l2_offset = s->l1_table[l1_index];
219
220 trace_qcow2_l2_allocate(bs, l1_index);
221
222 /* allocate a new l2 entry */
223
224 l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
225 if (l2_offset < 0) {
226 ret = l2_offset;
227 goto fail;
228 }
229
230 ret = qcow2_cache_flush(bs, s->refcount_block_cache);
231 if (ret < 0) {
232 goto fail;
233 }
234
235 /* allocate a new entry in the l2 cache */
236
237 trace_qcow2_l2_allocate_get_empty(bs, l1_index);
238 ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table);
239 if (ret < 0) {
240 goto fail;
241 }
242
243 l2_table = *table;
244
245 if ((old_l2_offset & L1E_OFFSET_MASK) == 0) {
246 /* if there was no old l2 table, clear the new table */
247 memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
248 } else {
249 uint64_t* old_table;
250
251 /* if there was an old l2 table, read it from the disk */
252 BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ);
253 ret = qcow2_cache_get(bs, s->l2_table_cache,
254 old_l2_offset & L1E_OFFSET_MASK,
255 (void**) &old_table);
256 if (ret < 0) {
257 goto fail;
258 }
259
260 memcpy(l2_table, old_table, s->cluster_size);
261
262 qcow2_cache_put(bs, s->l2_table_cache, (void **) &old_table);
263 }
264
265 /* write the l2 table to the file */
266 BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
267
268 trace_qcow2_l2_allocate_write_l2(bs, l1_index);
269 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
270 ret = qcow2_cache_flush(bs, s->l2_table_cache);
271 if (ret < 0) {
272 goto fail;
273 }
274
275 /* update the L1 entry */
276 trace_qcow2_l2_allocate_write_l1(bs, l1_index);
277 s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
278 ret = qcow2_write_l1_entry(bs, l1_index);
279 if (ret < 0) {
280 goto fail;
281 }
282
283 *table = l2_table;
284 trace_qcow2_l2_allocate_done(bs, l1_index, 0);
285 return 0;
286
287 fail:
288 trace_qcow2_l2_allocate_done(bs, l1_index, ret);
289 if (l2_table != NULL) {
290 qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
291 }
292 s->l1_table[l1_index] = old_l2_offset;
293 if (l2_offset > 0) {
294 qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
295 QCOW2_DISCARD_ALWAYS);
296 }
297 return ret;
298 }
299
300 /*
301 * Checks how many clusters in a given L2 table are contiguous in the image
302 * file. As soon as one of the flags in the bitmask stop_flags changes compared
303 * to the first cluster, the search is stopped and the cluster is not counted
304 * as contiguous. (This allows it, for example, to stop at the first compressed
305 * cluster which may require a different handling)
306 */
307 static int count_contiguous_clusters(int nb_clusters, int cluster_size,
308 uint64_t *l2_table, uint64_t stop_flags)
309 {
310 int i;
311 uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED;
312 uint64_t first_entry = be64_to_cpu(l2_table[0]);
313 uint64_t offset = first_entry & mask;
314
315 if (!offset)
316 return 0;
317
318 assert(qcow2_get_cluster_type(first_entry) == QCOW2_CLUSTER_NORMAL);
319
320 for (i = 0; i < nb_clusters; i++) {
321 uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask;
322 if (offset + (uint64_t) i * cluster_size != l2_entry) {
323 break;
324 }
325 }
326
327 return i;
328 }
329
330 static int count_contiguous_clusters_by_type(int nb_clusters,
331 uint64_t *l2_table,
332 int wanted_type)
333 {
334 int i;
335
336 for (i = 0; i < nb_clusters; i++) {
337 int type = qcow2_get_cluster_type(be64_to_cpu(l2_table[i]));
338
339 if (type != wanted_type) {
340 break;
341 }
342 }
343
344 return i;
345 }
346
347 /* The crypt function is compatible with the linux cryptoloop
348 algorithm for < 4 GB images. NOTE: out_buf == in_buf is
349 supported */
350 int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num,
351 uint8_t *out_buf, const uint8_t *in_buf,
352 int nb_sectors, bool enc,
353 Error **errp)
354 {
355 union {
356 uint64_t ll[2];
357 uint8_t b[16];
358 } ivec;
359 int i;
360 int ret;
361
362 for(i = 0; i < nb_sectors; i++) {
363 ivec.ll[0] = cpu_to_le64(sector_num);
364 ivec.ll[1] = 0;
365 if (qcrypto_cipher_setiv(s->cipher,
366 ivec.b, G_N_ELEMENTS(ivec.b),
367 errp) < 0) {
368 return -1;
369 }
370 if (enc) {
371 ret = qcrypto_cipher_encrypt(s->cipher,
372 in_buf,
373 out_buf,
374 512,
375 errp);
376 } else {
377 ret = qcrypto_cipher_decrypt(s->cipher,
378 in_buf,
379 out_buf,
380 512,
381 errp);
382 }
383 if (ret < 0) {
384 return -1;
385 }
386 sector_num++;
387 in_buf += 512;
388 out_buf += 512;
389 }
390 return 0;
391 }
392
393 static int coroutine_fn copy_sectors(BlockDriverState *bs,
394 uint64_t start_sect,
395 uint64_t cluster_offset,
396 int n_start, int n_end)
397 {
398 BDRVQcow2State *s = bs->opaque;
399 QEMUIOVector qiov;
400 struct iovec iov;
401 int n, ret;
402
403 n = n_end - n_start;
404 if (n <= 0) {
405 return 0;
406 }
407
408 iov.iov_len = n * BDRV_SECTOR_SIZE;
409 iov.iov_base = qemu_try_blockalign(bs, iov.iov_len);
410 if (iov.iov_base == NULL) {
411 return -ENOMEM;
412 }
413
414 qemu_iovec_init_external(&qiov, &iov, 1);
415
416 BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
417
418 if (!bs->drv) {
419 ret = -ENOMEDIUM;
420 goto out;
421 }
422
423 /* Call .bdrv_co_readv() directly instead of using the public block-layer
424 * interface. This avoids double I/O throttling and request tracking,
425 * which can lead to deadlock when block layer copy-on-read is enabled.
426 */
427 ret = bs->drv->bdrv_co_readv(bs, start_sect + n_start, n, &qiov);
428 if (ret < 0) {
429 goto out;
430 }
431
432 if (bs->encrypted) {
433 Error *err = NULL;
434 assert(s->cipher);
435 if (qcow2_encrypt_sectors(s, start_sect + n_start,
436 iov.iov_base, iov.iov_base, n,
437 true, &err) < 0) {
438 ret = -EIO;
439 error_free(err);
440 goto out;
441 }
442 }
443
444 ret = qcow2_pre_write_overlap_check(bs, 0,
445 cluster_offset + n_start * BDRV_SECTOR_SIZE, n * BDRV_SECTOR_SIZE);
446 if (ret < 0) {
447 goto out;
448 }
449
450 BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
451 ret = bdrv_co_writev(bs->file->bs, (cluster_offset >> 9) + n_start, n,
452 &qiov);
453 if (ret < 0) {
454 goto out;
455 }
456
457 ret = 0;
458 out:
459 qemu_vfree(iov.iov_base);
460 return ret;
461 }
462
463
464 /*
465 * get_cluster_offset
466 *
467 * For a given offset of the disk image, find the cluster offset in
468 * qcow2 file. The offset is stored in *cluster_offset.
469 *
470 * on entry, *num is the number of contiguous sectors we'd like to
471 * access following offset.
472 *
473 * on exit, *num is the number of contiguous sectors we can read.
474 *
475 * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error
476 * cases.
477 */
478 int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
479 int *num, uint64_t *cluster_offset)
480 {
481 BDRVQcow2State *s = bs->opaque;
482 unsigned int l2_index;
483 uint64_t l1_index, l2_offset, *l2_table;
484 int l1_bits, c;
485 unsigned int index_in_cluster, nb_clusters;
486 uint64_t nb_available, nb_needed;
487 int ret;
488
489 index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
490 nb_needed = *num + index_in_cluster;
491
492 l1_bits = s->l2_bits + s->cluster_bits;
493
494 /* compute how many bytes there are between the offset and
495 * the end of the l1 entry
496 */
497
498 nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
499
500 /* compute the number of available sectors */
501
502 nb_available = (nb_available >> 9) + index_in_cluster;
503
504 if (nb_needed > nb_available) {
505 nb_needed = nb_available;
506 }
507 assert(nb_needed <= INT_MAX);
508
509 *cluster_offset = 0;
510
511 /* seek to the l2 offset in the l1 table */
512
513 l1_index = offset >> l1_bits;
514 if (l1_index >= s->l1_size) {
515 ret = QCOW2_CLUSTER_UNALLOCATED;
516 goto out;
517 }
518
519 l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
520 if (!l2_offset) {
521 ret = QCOW2_CLUSTER_UNALLOCATED;
522 goto out;
523 }
524
525 if (offset_into_cluster(s, l2_offset)) {
526 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64
527 " unaligned (L1 index: %#" PRIx64 ")",
528 l2_offset, l1_index);
529 return -EIO;
530 }
531
532 /* load the l2 table in memory */
533
534 ret = l2_load(bs, l2_offset, &l2_table);
535 if (ret < 0) {
536 return ret;
537 }
538
539 /* find the cluster offset for the given disk offset */
540
541 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
542 *cluster_offset = be64_to_cpu(l2_table[l2_index]);
543
544 /* nb_needed <= INT_MAX, thus nb_clusters <= INT_MAX, too */
545 nb_clusters = size_to_clusters(s, nb_needed << 9);
546
547 ret = qcow2_get_cluster_type(*cluster_offset);
548 switch (ret) {
549 case QCOW2_CLUSTER_COMPRESSED:
550 /* Compressed clusters can only be processed one by one */
551 c = 1;
552 *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK;
553 break;
554 case QCOW2_CLUSTER_ZERO:
555 if (s->qcow_version < 3) {
556 qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found"
557 " in pre-v3 image (L2 offset: %#" PRIx64
558 ", L2 index: %#x)", l2_offset, l2_index);
559 ret = -EIO;
560 goto fail;
561 }
562 c = count_contiguous_clusters_by_type(nb_clusters, &l2_table[l2_index],
563 QCOW2_CLUSTER_ZERO);
564 *cluster_offset = 0;
565 break;
566 case QCOW2_CLUSTER_UNALLOCATED:
567 /* how many empty clusters ? */
568 c = count_contiguous_clusters_by_type(nb_clusters, &l2_table[l2_index],
569 QCOW2_CLUSTER_UNALLOCATED);
570 *cluster_offset = 0;
571 break;
572 case QCOW2_CLUSTER_NORMAL:
573 /* how many allocated clusters ? */
574 c = count_contiguous_clusters(nb_clusters, s->cluster_size,
575 &l2_table[l2_index], QCOW_OFLAG_ZERO);
576 *cluster_offset &= L2E_OFFSET_MASK;
577 if (offset_into_cluster(s, *cluster_offset)) {
578 qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset %#"
579 PRIx64 " unaligned (L2 offset: %#" PRIx64
580 ", L2 index: %#x)", *cluster_offset,
581 l2_offset, l2_index);
582 ret = -EIO;
583 goto fail;
584 }
585 break;
586 default:
587 abort();
588 }
589
590 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
591
592 nb_available = (c * s->cluster_sectors);
593
594 out:
595 if (nb_available > nb_needed)
596 nb_available = nb_needed;
597
598 *num = nb_available - index_in_cluster;
599
600 return ret;
601
602 fail:
603 qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table);
604 return ret;
605 }
606
607 /*
608 * get_cluster_table
609 *
610 * for a given disk offset, load (and allocate if needed)
611 * the l2 table.
612 *
613 * the l2 table offset in the qcow2 file and the cluster index
614 * in the l2 table are given to the caller.
615 *
616 * Returns 0 on success, -errno in failure case
617 */
618 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
619 uint64_t **new_l2_table,
620 int *new_l2_index)
621 {
622 BDRVQcow2State *s = bs->opaque;
623 unsigned int l2_index;
624 uint64_t l1_index, l2_offset;
625 uint64_t *l2_table = NULL;
626 int ret;
627
628 /* seek to the l2 offset in the l1 table */
629
630 l1_index = offset >> (s->l2_bits + s->cluster_bits);
631 if (l1_index >= s->l1_size) {
632 ret = qcow2_grow_l1_table(bs, l1_index + 1, false);
633 if (ret < 0) {
634 return ret;
635 }
636 }
637
638 assert(l1_index < s->l1_size);
639 l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
640 if (offset_into_cluster(s, l2_offset)) {
641 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64
642 " unaligned (L1 index: %#" PRIx64 ")",
643 l2_offset, l1_index);
644 return -EIO;
645 }
646
647 /* seek the l2 table of the given l2 offset */
648
649 if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) {
650 /* load the l2 table in memory */
651 ret = l2_load(bs, l2_offset, &l2_table);
652 if (ret < 0) {
653 return ret;
654 }
655 } else {
656 /* First allocate a new L2 table (and do COW if needed) */
657 ret = l2_allocate(bs, l1_index, &l2_table);
658 if (ret < 0) {
659 return ret;
660 }
661
662 /* Then decrease the refcount of the old table */
663 if (l2_offset) {
664 qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
665 QCOW2_DISCARD_OTHER);
666 }
667 }
668
669 /* find the cluster offset for the given disk offset */
670
671 l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
672
673 *new_l2_table = l2_table;
674 *new_l2_index = l2_index;
675
676 return 0;
677 }
678
679 /*
680 * alloc_compressed_cluster_offset
681 *
682 * For a given offset of the disk image, return cluster offset in
683 * qcow2 file.
684 *
685 * If the offset is not found, allocate a new compressed cluster.
686 *
687 * Return the cluster offset if successful,
688 * Return 0, otherwise.
689 *
690 */
691
692 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
693 uint64_t offset,
694 int compressed_size)
695 {
696 BDRVQcow2State *s = bs->opaque;
697 int l2_index, ret;
698 uint64_t *l2_table;
699 int64_t cluster_offset;
700 int nb_csectors;
701
702 ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
703 if (ret < 0) {
704 return 0;
705 }
706
707 /* Compression can't overwrite anything. Fail if the cluster was already
708 * allocated. */
709 cluster_offset = be64_to_cpu(l2_table[l2_index]);
710 if (cluster_offset & L2E_OFFSET_MASK) {
711 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
712 return 0;
713 }
714
715 cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
716 if (cluster_offset < 0) {
717 qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
718 return 0;
719 }
720
721 nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
722 (cluster_offset >> 9);
723
724 cluster_offset |= QCOW_OFLAG_COMPRESSED |
725 ((uint64_t)nb_csectors << s->csize_shift);
726
727 /* update L2 table */
728
729 /* compressed clusters never have the copied flag */
730
731 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
732 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
733 l2_table[l2_index] = cpu_to_be64(cluster_offset);
734 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
735
736 return cluster_offset;
737 }
738
739 static int perform_cow(BlockDriverState *bs, QCowL2Meta *m, Qcow2COWRegion *r)
740 {
741 BDRVQcow2State *s = bs->opaque;
742 int ret;
743
744 if (r->nb_sectors == 0) {
745 return 0;
746 }
747
748 qemu_co_mutex_unlock(&s->lock);
749 ret = copy_sectors(bs, m->offset / BDRV_SECTOR_SIZE, m->alloc_offset,
750 r->offset / BDRV_SECTOR_SIZE,
751 r->offset / BDRV_SECTOR_SIZE + r->nb_sectors);
752 qemu_co_mutex_lock(&s->lock);
753
754 if (ret < 0) {
755 return ret;
756 }
757
758 /*
759 * Before we update the L2 table to actually point to the new cluster, we
760 * need to be sure that the refcounts have been increased and COW was
761 * handled.
762 */
763 qcow2_cache_depends_on_flush(s->l2_table_cache);
764
765 return 0;
766 }
767
768 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
769 {
770 BDRVQcow2State *s = bs->opaque;
771 int i, j = 0, l2_index, ret;
772 uint64_t *old_cluster, *l2_table;
773 uint64_t cluster_offset = m->alloc_offset;
774
775 trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters);
776 assert(m->nb_clusters > 0);
777
778 old_cluster = g_try_new(uint64_t, m->nb_clusters);
779 if (old_cluster == NULL) {
780 ret = -ENOMEM;
781 goto err;
782 }
783
784 /* copy content of unmodified sectors */
785 ret = perform_cow(bs, m, &m->cow_start);
786 if (ret < 0) {
787 goto err;
788 }
789
790 ret = perform_cow(bs, m, &m->cow_end);
791 if (ret < 0) {
792 goto err;
793 }
794
795 /* Update L2 table. */
796 if (s->use_lazy_refcounts) {
797 qcow2_mark_dirty(bs);
798 }
799 if (qcow2_need_accurate_refcounts(s)) {
800 qcow2_cache_set_dependency(bs, s->l2_table_cache,
801 s->refcount_block_cache);
802 }
803
804 ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index);
805 if (ret < 0) {
806 goto err;
807 }
808 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
809
810 assert(l2_index + m->nb_clusters <= s->l2_size);
811 for (i = 0; i < m->nb_clusters; i++) {
812 /* if two concurrent writes happen to the same unallocated cluster
813 * each write allocates separate cluster and writes data concurrently.
814 * The first one to complete updates l2 table with pointer to its
815 * cluster the second one has to do RMW (which is done above by
816 * copy_sectors()), update l2 table with its cluster pointer and free
817 * old cluster. This is what this loop does */
818 if(l2_table[l2_index + i] != 0)
819 old_cluster[j++] = l2_table[l2_index + i];
820
821 l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
822 (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
823 }
824
825
826 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
827
828 /*
829 * If this was a COW, we need to decrease the refcount of the old cluster.
830 *
831 * Don't discard clusters that reach a refcount of 0 (e.g. compressed
832 * clusters), the next write will reuse them anyway.
833 */
834 if (j != 0) {
835 for (i = 0; i < j; i++) {
836 qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1,
837 QCOW2_DISCARD_NEVER);
838 }
839 }
840
841 ret = 0;
842 err:
843 g_free(old_cluster);
844 return ret;
845 }
846
847 /*
848 * Returns the number of contiguous clusters that can be used for an allocating
849 * write, but require COW to be performed (this includes yet unallocated space,
850 * which must copy from the backing file)
851 */
852 static int count_cow_clusters(BDRVQcow2State *s, int nb_clusters,
853 uint64_t *l2_table, int l2_index)
854 {
855 int i;
856
857 for (i = 0; i < nb_clusters; i++) {
858 uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]);
859 int cluster_type = qcow2_get_cluster_type(l2_entry);
860
861 switch(cluster_type) {
862 case QCOW2_CLUSTER_NORMAL:
863 if (l2_entry & QCOW_OFLAG_COPIED) {
864 goto out;
865 }
866 break;
867 case QCOW2_CLUSTER_UNALLOCATED:
868 case QCOW2_CLUSTER_COMPRESSED:
869 case QCOW2_CLUSTER_ZERO:
870 break;
871 default:
872 abort();
873 }
874 }
875
876 out:
877 assert(i <= nb_clusters);
878 return i;
879 }
880
881 /*
882 * Check if there already is an AIO write request in flight which allocates
883 * the same cluster. In this case we need to wait until the previous
884 * request has completed and updated the L2 table accordingly.
885 *
886 * Returns:
887 * 0 if there was no dependency. *cur_bytes indicates the number of
888 * bytes from guest_offset that can be read before the next
889 * dependency must be processed (or the request is complete)
890 *
891 * -EAGAIN if we had to wait for another request, previously gathered
892 * information on cluster allocation may be invalid now. The caller
893 * must start over anyway, so consider *cur_bytes undefined.
894 */
895 static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset,
896 uint64_t *cur_bytes, QCowL2Meta **m)
897 {
898 BDRVQcow2State *s = bs->opaque;
899 QCowL2Meta *old_alloc;
900 uint64_t bytes = *cur_bytes;
901
902 QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
903
904 uint64_t start = guest_offset;
905 uint64_t end = start + bytes;
906 uint64_t old_start = l2meta_cow_start(old_alloc);
907 uint64_t old_end = l2meta_cow_end(old_alloc);
908
909 if (end <= old_start || start >= old_end) {
910 /* No intersection */
911 } else {
912 if (start < old_start) {
913 /* Stop at the start of a running allocation */
914 bytes = old_start - start;
915 } else {
916 bytes = 0;
917 }
918
919 /* Stop if already an l2meta exists. After yielding, it wouldn't
920 * be valid any more, so we'd have to clean up the old L2Metas
921 * and deal with requests depending on them before starting to
922 * gather new ones. Not worth the trouble. */
923 if (bytes == 0 && *m) {
924 *cur_bytes = 0;
925 return 0;
926 }
927
928 if (bytes == 0) {
929 /* Wait for the dependency to complete. We need to recheck
930 * the free/allocated clusters when we continue. */
931 qemu_co_mutex_unlock(&s->lock);
932 qemu_co_queue_wait(&old_alloc->dependent_requests);
933 qemu_co_mutex_lock(&s->lock);
934 return -EAGAIN;
935 }
936 }
937 }
938
939 /* Make sure that existing clusters and new allocations are only used up to
940 * the next dependency if we shortened the request above */
941 *cur_bytes = bytes;
942
943 return 0;
944 }
945
946 /*
947 * Checks how many already allocated clusters that don't require a copy on
948 * write there are at the given guest_offset (up to *bytes). If
949 * *host_offset is not zero, only physically contiguous clusters beginning at
950 * this host offset are counted.
951 *
952 * Note that guest_offset may not be cluster aligned. In this case, the
953 * returned *host_offset points to exact byte referenced by guest_offset and
954 * therefore isn't cluster aligned as well.
955 *
956 * Returns:
957 * 0: if no allocated clusters are available at the given offset.
958 * *bytes is normally unchanged. It is set to 0 if the cluster
959 * is allocated and doesn't need COW, but doesn't have the right
960 * physical offset.
961 *
962 * 1: if allocated clusters that don't require a COW are available at
963 * the requested offset. *bytes may have decreased and describes
964 * the length of the area that can be written to.
965 *
966 * -errno: in error cases
967 */
968 static int handle_copied(BlockDriverState *bs, uint64_t guest_offset,
969 uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
970 {
971 BDRVQcow2State *s = bs->opaque;
972 int l2_index;
973 uint64_t cluster_offset;
974 uint64_t *l2_table;
975 uint64_t nb_clusters;
976 unsigned int keep_clusters;
977 int ret;
978
979 trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset,
980 *bytes);
981
982 assert(*host_offset == 0 || offset_into_cluster(s, guest_offset)
983 == offset_into_cluster(s, *host_offset));
984
985 /*
986 * Calculate the number of clusters to look for. We stop at L2 table
987 * boundaries to keep things simple.
988 */
989 nb_clusters =
990 size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
991
992 l2_index = offset_to_l2_index(s, guest_offset);
993 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
994 assert(nb_clusters <= INT_MAX);
995
996 /* Find L2 entry for the first involved cluster */
997 ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index);
998 if (ret < 0) {
999 return ret;
1000 }
1001
1002 cluster_offset = be64_to_cpu(l2_table[l2_index]);
1003
1004 /* Check how many clusters are already allocated and don't need COW */
1005 if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL
1006 && (cluster_offset & QCOW_OFLAG_COPIED))
1007 {
1008 /* If a specific host_offset is required, check it */
1009 bool offset_matches =
1010 (cluster_offset & L2E_OFFSET_MASK) == *host_offset;
1011
1012 if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) {
1013 qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset "
1014 "%#llx unaligned (guest offset: %#" PRIx64
1015 ")", cluster_offset & L2E_OFFSET_MASK,
1016 guest_offset);
1017 ret = -EIO;
1018 goto out;
1019 }
1020
1021 if (*host_offset != 0 && !offset_matches) {
1022 *bytes = 0;
1023 ret = 0;
1024 goto out;
1025 }
1026
1027 /* We keep all QCOW_OFLAG_COPIED clusters */
1028 keep_clusters =
1029 count_contiguous_clusters(nb_clusters, s->cluster_size,
1030 &l2_table[l2_index],
1031 QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO);
1032 assert(keep_clusters <= nb_clusters);
1033
1034 *bytes = MIN(*bytes,
1035 keep_clusters * s->cluster_size
1036 - offset_into_cluster(s, guest_offset));
1037
1038 ret = 1;
1039 } else {
1040 ret = 0;
1041 }
1042
1043 /* Cleanup */
1044 out:
1045 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1046
1047 /* Only return a host offset if we actually made progress. Otherwise we
1048 * would make requirements for handle_alloc() that it can't fulfill */
1049 if (ret > 0) {
1050 *host_offset = (cluster_offset & L2E_OFFSET_MASK)
1051 + offset_into_cluster(s, guest_offset);
1052 }
1053
1054 return ret;
1055 }
1056
1057 /*
1058 * Allocates new clusters for the given guest_offset.
1059 *
1060 * At most *nb_clusters are allocated, and on return *nb_clusters is updated to
1061 * contain the number of clusters that have been allocated and are contiguous
1062 * in the image file.
1063 *
1064 * If *host_offset is non-zero, it specifies the offset in the image file at
1065 * which the new clusters must start. *nb_clusters can be 0 on return in this
1066 * case if the cluster at host_offset is already in use. If *host_offset is
1067 * zero, the clusters can be allocated anywhere in the image file.
1068 *
1069 * *host_offset is updated to contain the offset into the image file at which
1070 * the first allocated cluster starts.
1071 *
1072 * Return 0 on success and -errno in error cases. -EAGAIN means that the
1073 * function has been waiting for another request and the allocation must be
1074 * restarted, but the whole request should not be failed.
1075 */
1076 static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset,
1077 uint64_t *host_offset, uint64_t *nb_clusters)
1078 {
1079 BDRVQcow2State *s = bs->opaque;
1080
1081 trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset,
1082 *host_offset, *nb_clusters);
1083
1084 /* Allocate new clusters */
1085 trace_qcow2_cluster_alloc_phys(qemu_coroutine_self());
1086 if (*host_offset == 0) {
1087 int64_t cluster_offset =
1088 qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size);
1089 if (cluster_offset < 0) {
1090 return cluster_offset;
1091 }
1092 *host_offset = cluster_offset;
1093 return 0;
1094 } else {
1095 int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters);
1096 if (ret < 0) {
1097 return ret;
1098 }
1099 *nb_clusters = ret;
1100 return 0;
1101 }
1102 }
1103
1104 /*
1105 * Allocates new clusters for an area that either is yet unallocated or needs a
1106 * copy on write. If *host_offset is non-zero, clusters are only allocated if
1107 * the new allocation can match the specified host offset.
1108 *
1109 * Note that guest_offset may not be cluster aligned. In this case, the
1110 * returned *host_offset points to exact byte referenced by guest_offset and
1111 * therefore isn't cluster aligned as well.
1112 *
1113 * Returns:
1114 * 0: if no clusters could be allocated. *bytes is set to 0,
1115 * *host_offset is left unchanged.
1116 *
1117 * 1: if new clusters were allocated. *bytes may be decreased if the
1118 * new allocation doesn't cover all of the requested area.
1119 * *host_offset is updated to contain the host offset of the first
1120 * newly allocated cluster.
1121 *
1122 * -errno: in error cases
1123 */
1124 static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
1125 uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
1126 {
1127 BDRVQcow2State *s = bs->opaque;
1128 int l2_index;
1129 uint64_t *l2_table;
1130 uint64_t entry;
1131 uint64_t nb_clusters;
1132 int ret;
1133
1134 uint64_t alloc_cluster_offset;
1135
1136 trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset,
1137 *bytes);
1138 assert(*bytes > 0);
1139
1140 /*
1141 * Calculate the number of clusters to look for. We stop at L2 table
1142 * boundaries to keep things simple.
1143 */
1144 nb_clusters =
1145 size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
1146
1147 l2_index = offset_to_l2_index(s, guest_offset);
1148 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1149 assert(nb_clusters <= INT_MAX);
1150
1151 /* Find L2 entry for the first involved cluster */
1152 ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index);
1153 if (ret < 0) {
1154 return ret;
1155 }
1156
1157 entry = be64_to_cpu(l2_table[l2_index]);
1158
1159 /* For the moment, overwrite compressed clusters one by one */
1160 if (entry & QCOW_OFLAG_COMPRESSED) {
1161 nb_clusters = 1;
1162 } else {
1163 nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index);
1164 }
1165
1166 /* This function is only called when there were no non-COW clusters, so if
1167 * we can't find any unallocated or COW clusters either, something is
1168 * wrong with our code. */
1169 assert(nb_clusters > 0);
1170
1171 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1172
1173 /* Allocate, if necessary at a given offset in the image file */
1174 alloc_cluster_offset = start_of_cluster(s, *host_offset);
1175 ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset,
1176 &nb_clusters);
1177 if (ret < 0) {
1178 goto fail;
1179 }
1180
1181 /* Can't extend contiguous allocation */
1182 if (nb_clusters == 0) {
1183 *bytes = 0;
1184 return 0;
1185 }
1186
1187 /* !*host_offset would overwrite the image header and is reserved for "no
1188 * host offset preferred". If 0 was a valid host offset, it'd trigger the
1189 * following overlap check; do that now to avoid having an invalid value in
1190 * *host_offset. */
1191 if (!alloc_cluster_offset) {
1192 ret = qcow2_pre_write_overlap_check(bs, 0, alloc_cluster_offset,
1193 nb_clusters * s->cluster_size);
1194 assert(ret < 0);
1195 goto fail;
1196 }
1197
1198 /*
1199 * Save info needed for meta data update.
1200 *
1201 * requested_sectors: Number of sectors from the start of the first
1202 * newly allocated cluster to the end of the (possibly shortened
1203 * before) write request.
1204 *
1205 * avail_sectors: Number of sectors from the start of the first
1206 * newly allocated to the end of the last newly allocated cluster.
1207 *
1208 * nb_sectors: The number of sectors from the start of the first
1209 * newly allocated cluster to the end of the area that the write
1210 * request actually writes to (excluding COW at the end)
1211 */
1212 int requested_sectors =
1213 (*bytes + offset_into_cluster(s, guest_offset))
1214 >> BDRV_SECTOR_BITS;
1215 int avail_sectors = nb_clusters
1216 << (s->cluster_bits - BDRV_SECTOR_BITS);
1217 int alloc_n_start = offset_into_cluster(s, guest_offset)
1218 >> BDRV_SECTOR_BITS;
1219 int nb_sectors = MIN(requested_sectors, avail_sectors);
1220 QCowL2Meta *old_m = *m;
1221
1222 *m = g_malloc0(sizeof(**m));
1223
1224 **m = (QCowL2Meta) {
1225 .next = old_m,
1226
1227 .alloc_offset = alloc_cluster_offset,
1228 .offset = start_of_cluster(s, guest_offset),
1229 .nb_clusters = nb_clusters,
1230 .nb_available = nb_sectors,
1231
1232 .cow_start = {
1233 .offset = 0,
1234 .nb_sectors = alloc_n_start,
1235 },
1236 .cow_end = {
1237 .offset = nb_sectors * BDRV_SECTOR_SIZE,
1238 .nb_sectors = avail_sectors - nb_sectors,
1239 },
1240 };
1241 qemu_co_queue_init(&(*m)->dependent_requests);
1242 QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
1243
1244 *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset);
1245 *bytes = MIN(*bytes, (nb_sectors * BDRV_SECTOR_SIZE)
1246 - offset_into_cluster(s, guest_offset));
1247 assert(*bytes != 0);
1248
1249 return 1;
1250
1251 fail:
1252 if (*m && (*m)->nb_clusters > 0) {
1253 QLIST_REMOVE(*m, next_in_flight);
1254 }
1255 return ret;
1256 }
1257
1258 /*
1259 * alloc_cluster_offset
1260 *
1261 * For a given offset on the virtual disk, find the cluster offset in qcow2
1262 * file. If the offset is not found, allocate a new cluster.
1263 *
1264 * If the cluster was already allocated, m->nb_clusters is set to 0 and
1265 * other fields in m are meaningless.
1266 *
1267 * If the cluster is newly allocated, m->nb_clusters is set to the number of
1268 * contiguous clusters that have been allocated. In this case, the other
1269 * fields of m are valid and contain information about the first allocated
1270 * cluster.
1271 *
1272 * If the request conflicts with another write request in flight, the coroutine
1273 * is queued and will be reentered when the dependency has completed.
1274 *
1275 * Return 0 on success and -errno in error cases
1276 */
1277 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
1278 int *num, uint64_t *host_offset, QCowL2Meta **m)
1279 {
1280 BDRVQcow2State *s = bs->opaque;
1281 uint64_t start, remaining;
1282 uint64_t cluster_offset;
1283 uint64_t cur_bytes;
1284 int ret;
1285
1286 trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *num);
1287
1288 assert((offset & ~BDRV_SECTOR_MASK) == 0);
1289
1290 again:
1291 start = offset;
1292 remaining = (uint64_t)*num << BDRV_SECTOR_BITS;
1293 cluster_offset = 0;
1294 *host_offset = 0;
1295 cur_bytes = 0;
1296 *m = NULL;
1297
1298 while (true) {
1299
1300 if (!*host_offset) {
1301 *host_offset = start_of_cluster(s, cluster_offset);
1302 }
1303
1304 assert(remaining >= cur_bytes);
1305
1306 start += cur_bytes;
1307 remaining -= cur_bytes;
1308 cluster_offset += cur_bytes;
1309
1310 if (remaining == 0) {
1311 break;
1312 }
1313
1314 cur_bytes = remaining;
1315
1316 /*
1317 * Now start gathering as many contiguous clusters as possible:
1318 *
1319 * 1. Check for overlaps with in-flight allocations
1320 *
1321 * a) Overlap not in the first cluster -> shorten this request and
1322 * let the caller handle the rest in its next loop iteration.
1323 *
1324 * b) Real overlaps of two requests. Yield and restart the search
1325 * for contiguous clusters (the situation could have changed
1326 * while we were sleeping)
1327 *
1328 * c) TODO: Request starts in the same cluster as the in-flight
1329 * allocation ends. Shorten the COW of the in-fight allocation,
1330 * set cluster_offset to write to the same cluster and set up
1331 * the right synchronisation between the in-flight request and
1332 * the new one.
1333 */
1334 ret = handle_dependencies(bs, start, &cur_bytes, m);
1335 if (ret == -EAGAIN) {
1336 /* Currently handle_dependencies() doesn't yield if we already had
1337 * an allocation. If it did, we would have to clean up the L2Meta
1338 * structs before starting over. */
1339 assert(*m == NULL);
1340 goto again;
1341 } else if (ret < 0) {
1342 return ret;
1343 } else if (cur_bytes == 0) {
1344 break;
1345 } else {
1346 /* handle_dependencies() may have decreased cur_bytes (shortened
1347 * the allocations below) so that the next dependency is processed
1348 * correctly during the next loop iteration. */
1349 }
1350
1351 /*
1352 * 2. Count contiguous COPIED clusters.
1353 */
1354 ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m);
1355 if (ret < 0) {
1356 return ret;
1357 } else if (ret) {
1358 continue;
1359 } else if (cur_bytes == 0) {
1360 break;
1361 }
1362
1363 /*
1364 * 3. If the request still hasn't completed, allocate new clusters,
1365 * considering any cluster_offset of steps 1c or 2.
1366 */
1367 ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m);
1368 if (ret < 0) {
1369 return ret;
1370 } else if (ret) {
1371 continue;
1372 } else {
1373 assert(cur_bytes == 0);
1374 break;
1375 }
1376 }
1377
1378 *num -= remaining >> BDRV_SECTOR_BITS;
1379 assert(*num > 0);
1380 assert(*host_offset != 0);
1381
1382 return 0;
1383 }
1384
1385 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1386 const uint8_t *buf, int buf_size)
1387 {
1388 z_stream strm1, *strm = &strm1;
1389 int ret, out_len;
1390
1391 memset(strm, 0, sizeof(*strm));
1392
1393 strm->next_in = (uint8_t *)buf;
1394 strm->avail_in = buf_size;
1395 strm->next_out = out_buf;
1396 strm->avail_out = out_buf_size;
1397
1398 ret = inflateInit2(strm, -12);
1399 if (ret != Z_OK)
1400 return -1;
1401 ret = inflate(strm, Z_FINISH);
1402 out_len = strm->next_out - out_buf;
1403 if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1404 out_len != out_buf_size) {
1405 inflateEnd(strm);
1406 return -1;
1407 }
1408 inflateEnd(strm);
1409 return 0;
1410 }
1411
1412 int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
1413 {
1414 BDRVQcow2State *s = bs->opaque;
1415 int ret, csize, nb_csectors, sector_offset;
1416 uint64_t coffset;
1417
1418 coffset = cluster_offset & s->cluster_offset_mask;
1419 if (s->cluster_cache_offset != coffset) {
1420 nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1421 sector_offset = coffset & 511;
1422 csize = nb_csectors * 512 - sector_offset;
1423 BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
1424 ret = bdrv_read(bs->file->bs, coffset >> 9, s->cluster_data,
1425 nb_csectors);
1426 if (ret < 0) {
1427 return ret;
1428 }
1429 if (decompress_buffer(s->cluster_cache, s->cluster_size,
1430 s->cluster_data + sector_offset, csize) < 0) {
1431 return -EIO;
1432 }
1433 s->cluster_cache_offset = coffset;
1434 }
1435 return 0;
1436 }
1437
1438 /*
1439 * This discards as many clusters of nb_clusters as possible at once (i.e.
1440 * all clusters in the same L2 table) and returns the number of discarded
1441 * clusters.
1442 */
1443 static int discard_single_l2(BlockDriverState *bs, uint64_t offset,
1444 uint64_t nb_clusters, enum qcow2_discard_type type,
1445 bool full_discard)
1446 {
1447 BDRVQcow2State *s = bs->opaque;
1448 uint64_t *l2_table;
1449 int l2_index;
1450 int ret;
1451 int i;
1452
1453 ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1454 if (ret < 0) {
1455 return ret;
1456 }
1457
1458 /* Limit nb_clusters to one L2 table */
1459 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1460 assert(nb_clusters <= INT_MAX);
1461
1462 for (i = 0; i < nb_clusters; i++) {
1463 uint64_t old_l2_entry;
1464
1465 old_l2_entry = be64_to_cpu(l2_table[l2_index + i]);
1466
1467 /*
1468 * If full_discard is false, make sure that a discarded area reads back
1469 * as zeroes for v3 images (we cannot do it for v2 without actually
1470 * writing a zero-filled buffer). We can skip the operation if the
1471 * cluster is already marked as zero, or if it's unallocated and we
1472 * don't have a backing file.
1473 *
1474 * TODO We might want to use bdrv_get_block_status(bs) here, but we're
1475 * holding s->lock, so that doesn't work today.
1476 *
1477 * If full_discard is true, the sector should not read back as zeroes,
1478 * but rather fall through to the backing file.
1479 */
1480 switch (qcow2_get_cluster_type(old_l2_entry)) {
1481 case QCOW2_CLUSTER_UNALLOCATED:
1482 if (full_discard || !bs->backing) {
1483 continue;
1484 }
1485 break;
1486
1487 case QCOW2_CLUSTER_ZERO:
1488 if (!full_discard) {
1489 continue;
1490 }
1491 break;
1492
1493 case QCOW2_CLUSTER_NORMAL:
1494 case QCOW2_CLUSTER_COMPRESSED:
1495 break;
1496
1497 default:
1498 abort();
1499 }
1500
1501 /* First remove L2 entries */
1502 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
1503 if (!full_discard && s->qcow_version >= 3) {
1504 l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
1505 } else {
1506 l2_table[l2_index + i] = cpu_to_be64(0);
1507 }
1508
1509 /* Then decrease the refcount */
1510 qcow2_free_any_clusters(bs, old_l2_entry, 1, type);
1511 }
1512
1513 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1514
1515 return nb_clusters;
1516 }
1517
1518 int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
1519 int nb_sectors, enum qcow2_discard_type type, bool full_discard)
1520 {
1521 BDRVQcow2State *s = bs->opaque;
1522 uint64_t end_offset;
1523 uint64_t nb_clusters;
1524 int ret;
1525
1526 end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS);
1527
1528 /* Round start up and end down */
1529 offset = align_offset(offset, s->cluster_size);
1530 end_offset = start_of_cluster(s, end_offset);
1531
1532 if (offset > end_offset) {
1533 return 0;
1534 }
1535
1536 nb_clusters = size_to_clusters(s, end_offset - offset);
1537
1538 s->cache_discards = true;
1539
1540 /* Each L2 table is handled by its own loop iteration */
1541 while (nb_clusters > 0) {
1542 ret = discard_single_l2(bs, offset, nb_clusters, type, full_discard);
1543 if (ret < 0) {
1544 goto fail;
1545 }
1546
1547 nb_clusters -= ret;
1548 offset += (ret * s->cluster_size);
1549 }
1550
1551 ret = 0;
1552 fail:
1553 s->cache_discards = false;
1554 qcow2_process_discards(bs, ret);
1555
1556 return ret;
1557 }
1558
1559 /*
1560 * This zeroes as many clusters of nb_clusters as possible at once (i.e.
1561 * all clusters in the same L2 table) and returns the number of zeroed
1562 * clusters.
1563 */
1564 static int zero_single_l2(BlockDriverState *bs, uint64_t offset,
1565 uint64_t nb_clusters)
1566 {
1567 BDRVQcow2State *s = bs->opaque;
1568 uint64_t *l2_table;
1569 int l2_index;
1570 int ret;
1571 int i;
1572
1573 ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1574 if (ret < 0) {
1575 return ret;
1576 }
1577
1578 /* Limit nb_clusters to one L2 table */
1579 nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1580 assert(nb_clusters <= INT_MAX);
1581
1582 for (i = 0; i < nb_clusters; i++) {
1583 uint64_t old_offset;
1584
1585 old_offset = be64_to_cpu(l2_table[l2_index + i]);
1586
1587 /* Update L2 entries */
1588 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
1589 if (old_offset & QCOW_OFLAG_COMPRESSED) {
1590 l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
1591 qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST);
1592 } else {
1593 l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO);
1594 }
1595 }
1596
1597 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1598
1599 return nb_clusters;
1600 }
1601
1602 int qcow2_zero_clusters(BlockDriverState *bs, uint64_t offset, int nb_sectors)
1603 {
1604 BDRVQcow2State *s = bs->opaque;
1605 uint64_t nb_clusters;
1606 int ret;
1607
1608 /* The zero flag is only supported by version 3 and newer */
1609 if (s->qcow_version < 3) {
1610 return -ENOTSUP;
1611 }
1612
1613 /* Each L2 table is handled by its own loop iteration */
1614 nb_clusters = size_to_clusters(s, nb_sectors << BDRV_SECTOR_BITS);
1615
1616 s->cache_discards = true;
1617
1618 while (nb_clusters > 0) {
1619 ret = zero_single_l2(bs, offset, nb_clusters);
1620 if (ret < 0) {
1621 goto fail;
1622 }
1623
1624 nb_clusters -= ret;
1625 offset += (ret * s->cluster_size);
1626 }
1627
1628 ret = 0;
1629 fail:
1630 s->cache_discards = false;
1631 qcow2_process_discards(bs, ret);
1632
1633 return ret;
1634 }
1635
1636 /*
1637 * Expands all zero clusters in a specific L1 table (or deallocates them, for
1638 * non-backed non-pre-allocated zero clusters).
1639 *
1640 * l1_entries and *visited_l1_entries are used to keep track of progress for
1641 * status_cb(). l1_entries contains the total number of L1 entries and
1642 * *visited_l1_entries counts all visited L1 entries.
1643 */
1644 static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
1645 int l1_size, int64_t *visited_l1_entries,
1646 int64_t l1_entries,
1647 BlockDriverAmendStatusCB *status_cb,
1648 void *cb_opaque)
1649 {
1650 BDRVQcow2State *s = bs->opaque;
1651 bool is_active_l1 = (l1_table == s->l1_table);
1652 uint64_t *l2_table = NULL;
1653 int ret;
1654 int i, j;
1655
1656 if (!is_active_l1) {
1657 /* inactive L2 tables require a buffer to be stored in when loading
1658 * them from disk */
1659 l2_table = qemu_try_blockalign(bs->file->bs, s->cluster_size);
1660 if (l2_table == NULL) {
1661 return -ENOMEM;
1662 }
1663 }
1664
1665 for (i = 0; i < l1_size; i++) {
1666 uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK;
1667 bool l2_dirty = false;
1668 uint64_t l2_refcount;
1669
1670 if (!l2_offset) {
1671 /* unallocated */
1672 (*visited_l1_entries)++;
1673 if (status_cb) {
1674 status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque);
1675 }
1676 continue;
1677 }
1678
1679 if (offset_into_cluster(s, l2_offset)) {
1680 qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#"
1681 PRIx64 " unaligned (L1 index: %#x)",
1682 l2_offset, i);
1683 ret = -EIO;
1684 goto fail;
1685 }
1686
1687 if (is_active_l1) {
1688 /* get active L2 tables from cache */
1689 ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
1690 (void **)&l2_table);
1691 } else {
1692 /* load inactive L2 tables from disk */
1693 ret = bdrv_read(bs->file->bs, l2_offset / BDRV_SECTOR_SIZE,
1694 (void *)l2_table, s->cluster_sectors);
1695 }
1696 if (ret < 0) {
1697 goto fail;
1698 }
1699
1700 ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1701 &l2_refcount);
1702 if (ret < 0) {
1703 goto fail;
1704 }
1705
1706 for (j = 0; j < s->l2_size; j++) {
1707 uint64_t l2_entry = be64_to_cpu(l2_table[j]);
1708 int64_t offset = l2_entry & L2E_OFFSET_MASK;
1709 int cluster_type = qcow2_get_cluster_type(l2_entry);
1710 bool preallocated = offset != 0;
1711
1712 if (cluster_type != QCOW2_CLUSTER_ZERO) {
1713 continue;
1714 }
1715
1716 if (!preallocated) {
1717 if (!bs->backing) {
1718 /* not backed; therefore we can simply deallocate the
1719 * cluster */
1720 l2_table[j] = 0;
1721 l2_dirty = true;
1722 continue;
1723 }
1724
1725 offset = qcow2_alloc_clusters(bs, s->cluster_size);
1726 if (offset < 0) {
1727 ret = offset;
1728 goto fail;
1729 }
1730
1731 if (l2_refcount > 1) {
1732 /* For shared L2 tables, set the refcount accordingly (it is
1733 * already 1 and needs to be l2_refcount) */
1734 ret = qcow2_update_cluster_refcount(bs,
1735 offset >> s->cluster_bits,
1736 refcount_diff(1, l2_refcount), false,
1737 QCOW2_DISCARD_OTHER);
1738 if (ret < 0) {
1739 qcow2_free_clusters(bs, offset, s->cluster_size,
1740 QCOW2_DISCARD_OTHER);
1741 goto fail;
1742 }
1743 }
1744 }
1745
1746 if (offset_into_cluster(s, offset)) {
1747 qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset "
1748 "%#" PRIx64 " unaligned (L2 offset: %#"
1749 PRIx64 ", L2 index: %#x)", offset,
1750 l2_offset, j);
1751 if (!preallocated) {
1752 qcow2_free_clusters(bs, offset, s->cluster_size,
1753 QCOW2_DISCARD_ALWAYS);
1754 }
1755 ret = -EIO;
1756 goto fail;
1757 }
1758
1759 ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size);
1760 if (ret < 0) {
1761 if (!preallocated) {
1762 qcow2_free_clusters(bs, offset, s->cluster_size,
1763 QCOW2_DISCARD_ALWAYS);
1764 }
1765 goto fail;
1766 }
1767
1768 ret = bdrv_write_zeroes(bs->file->bs, offset / BDRV_SECTOR_SIZE,
1769 s->cluster_sectors, 0);
1770 if (ret < 0) {
1771 if (!preallocated) {
1772 qcow2_free_clusters(bs, offset, s->cluster_size,
1773 QCOW2_DISCARD_ALWAYS);
1774 }
1775 goto fail;
1776 }
1777
1778 if (l2_refcount == 1) {
1779 l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED);
1780 } else {
1781 l2_table[j] = cpu_to_be64(offset);
1782 }
1783 l2_dirty = true;
1784 }
1785
1786 if (is_active_l1) {
1787 if (l2_dirty) {
1788 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
1789 qcow2_cache_depends_on_flush(s->l2_table_cache);
1790 }
1791 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1792 } else {
1793 if (l2_dirty) {
1794 ret = qcow2_pre_write_overlap_check(bs,
1795 QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, l2_offset,
1796 s->cluster_size);
1797 if (ret < 0) {
1798 goto fail;
1799 }
1800
1801 ret = bdrv_write(bs->file->bs, l2_offset / BDRV_SECTOR_SIZE,
1802 (void *)l2_table, s->cluster_sectors);
1803 if (ret < 0) {
1804 goto fail;
1805 }
1806 }
1807 }
1808
1809 (*visited_l1_entries)++;
1810 if (status_cb) {
1811 status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque);
1812 }
1813 }
1814
1815 ret = 0;
1816
1817 fail:
1818 if (l2_table) {
1819 if (!is_active_l1) {
1820 qemu_vfree(l2_table);
1821 } else {
1822 qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1823 }
1824 }
1825 return ret;
1826 }
1827
1828 /*
1829 * For backed images, expands all zero clusters on the image. For non-backed
1830 * images, deallocates all non-pre-allocated zero clusters (and claims the
1831 * allocation for pre-allocated ones). This is important for downgrading to a
1832 * qcow2 version which doesn't yet support metadata zero clusters.
1833 */
1834 int qcow2_expand_zero_clusters(BlockDriverState *bs,
1835 BlockDriverAmendStatusCB *status_cb,
1836 void *cb_opaque)
1837 {
1838 BDRVQcow2State *s = bs->opaque;
1839 uint64_t *l1_table = NULL;
1840 int64_t l1_entries = 0, visited_l1_entries = 0;
1841 int ret;
1842 int i, j;
1843
1844 if (status_cb) {
1845 l1_entries = s->l1_size;
1846 for (i = 0; i < s->nb_snapshots; i++) {
1847 l1_entries += s->snapshots[i].l1_size;
1848 }
1849 }
1850
1851 ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size,
1852 &visited_l1_entries, l1_entries,
1853 status_cb, cb_opaque);
1854 if (ret < 0) {
1855 goto fail;
1856 }
1857
1858 /* Inactive L1 tables may point to active L2 tables - therefore it is
1859 * necessary to flush the L2 table cache before trying to access the L2
1860 * tables pointed to by inactive L1 entries (else we might try to expand
1861 * zero clusters that have already been expanded); furthermore, it is also
1862 * necessary to empty the L2 table cache, since it may contain tables which
1863 * are now going to be modified directly on disk, bypassing the cache.
1864 * qcow2_cache_empty() does both for us. */
1865 ret = qcow2_cache_empty(bs, s->l2_table_cache);
1866 if (ret < 0) {
1867 goto fail;
1868 }
1869
1870 for (i = 0; i < s->nb_snapshots; i++) {
1871 int l1_sectors = (s->snapshots[i].l1_size * sizeof(uint64_t) +
1872 BDRV_SECTOR_SIZE - 1) / BDRV_SECTOR_SIZE;
1873
1874 l1_table = g_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE);
1875
1876 ret = bdrv_read(bs->file->bs,
1877 s->snapshots[i].l1_table_offset / BDRV_SECTOR_SIZE,
1878 (void *)l1_table, l1_sectors);
1879 if (ret < 0) {
1880 goto fail;
1881 }
1882
1883 for (j = 0; j < s->snapshots[i].l1_size; j++) {
1884 be64_to_cpus(&l1_table[j]);
1885 }
1886
1887 ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size,
1888 &visited_l1_entries, l1_entries,
1889 status_cb, cb_opaque);
1890 if (ret < 0) {
1891 goto fail;
1892 }
1893 }
1894
1895 ret = 0;
1896
1897 fail:
1898 g_free(l1_table);
1899 return ret;
1900 }