xref: /openbmc/qemu/block/qcow2-cluster.c (revision dffacd46)
1 /*
2  * Block driver for the QCOW version 2 format
3  *
4  * Copyright (c) 2004-2006 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 
25 #include <zlib.h>
26 
27 #include "qemu-common.h"
28 #include "block/block_int.h"
29 #include "block/qcow2.h"
30 #include "trace.h"
31 
32 int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
33                         bool exact_size)
34 {
35     BDRVQcowState *s = bs->opaque;
36     int new_l1_size2, ret, i;
37     uint64_t *new_l1_table;
38     int64_t old_l1_table_offset, old_l1_size;
39     int64_t new_l1_table_offset, new_l1_size;
40     uint8_t data[12];
41 
42     if (min_size <= s->l1_size)
43         return 0;
44 
45     if (exact_size) {
46         new_l1_size = min_size;
47     } else {
48         /* Bump size up to reduce the number of times we have to grow */
49         new_l1_size = s->l1_size;
50         if (new_l1_size == 0) {
51             new_l1_size = 1;
52         }
53         while (min_size > new_l1_size) {
54             new_l1_size = (new_l1_size * 3 + 1) / 2;
55         }
56     }
57 
58     if (new_l1_size > INT_MAX / sizeof(uint64_t)) {
59         return -EFBIG;
60     }
61 
62 #ifdef DEBUG_ALLOC2
63     fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n",
64             s->l1_size, new_l1_size);
65 #endif
66 
67     new_l1_size2 = sizeof(uint64_t) * new_l1_size;
68     new_l1_table = g_malloc0(align_offset(new_l1_size2, 512));
69     memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
70 
71     /* write new table (align to cluster) */
72     BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE);
73     new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
74     if (new_l1_table_offset < 0) {
75         g_free(new_l1_table);
76         return new_l1_table_offset;
77     }
78 
79     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
80     if (ret < 0) {
81         goto fail;
82     }
83 
84     /* the L1 position has not yet been updated, so these clusters must
85      * indeed be completely free */
86     ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset,
87                                         new_l1_size2);
88     if (ret < 0) {
89         goto fail;
90     }
91 
92     BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE);
93     for(i = 0; i < s->l1_size; i++)
94         new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
95     ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, new_l1_table, new_l1_size2);
96     if (ret < 0)
97         goto fail;
98     for(i = 0; i < s->l1_size; i++)
99         new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
100 
101     /* set new table */
102     BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE);
103     cpu_to_be32w((uint32_t*)data, new_l1_size);
104     stq_be_p(data + 4, new_l1_table_offset);
105     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), data,sizeof(data));
106     if (ret < 0) {
107         goto fail;
108     }
109     g_free(s->l1_table);
110     old_l1_table_offset = s->l1_table_offset;
111     s->l1_table_offset = new_l1_table_offset;
112     s->l1_table = new_l1_table;
113     old_l1_size = s->l1_size;
114     s->l1_size = new_l1_size;
115     qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * sizeof(uint64_t),
116                         QCOW2_DISCARD_OTHER);
117     return 0;
118  fail:
119     g_free(new_l1_table);
120     qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2,
121                         QCOW2_DISCARD_OTHER);
122     return ret;
123 }
124 
125 /*
126  * l2_load
127  *
128  * Loads a L2 table into memory. If the table is in the cache, the cache
129  * is used; otherwise the L2 table is loaded from the image file.
130  *
131  * Returns a pointer to the L2 table on success, or NULL if the read from
132  * the image file failed.
133  */
134 
135 static int l2_load(BlockDriverState *bs, uint64_t l2_offset,
136     uint64_t **l2_table)
137 {
138     BDRVQcowState *s = bs->opaque;
139     int ret;
140 
141     ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, (void**) l2_table);
142 
143     return ret;
144 }
145 
146 /*
147  * Writes one sector of the L1 table to the disk (can't update single entries
148  * and we really don't want bdrv_pread to perform a read-modify-write)
149  */
150 #define L1_ENTRIES_PER_SECTOR (512 / 8)
151 int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index)
152 {
153     BDRVQcowState *s = bs->opaque;
154     uint64_t buf[L1_ENTRIES_PER_SECTOR];
155     int l1_start_index;
156     int i, ret;
157 
158     l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
159     for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) {
160         buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
161     }
162 
163     ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1,
164             s->l1_table_offset + 8 * l1_start_index, sizeof(buf));
165     if (ret < 0) {
166         return ret;
167     }
168 
169     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
170     ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset + 8 * l1_start_index,
171         buf, sizeof(buf));
172     if (ret < 0) {
173         return ret;
174     }
175 
176     return 0;
177 }
178 
179 /*
180  * l2_allocate
181  *
182  * Allocate a new l2 entry in the file. If l1_index points to an already
183  * used entry in the L2 table (i.e. we are doing a copy on write for the L2
184  * table) copy the contents of the old L2 table into the newly allocated one.
185  * Otherwise the new table is initialized with zeros.
186  *
187  */
188 
189 static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
190 {
191     BDRVQcowState *s = bs->opaque;
192     uint64_t old_l2_offset;
193     uint64_t *l2_table = NULL;
194     int64_t l2_offset;
195     int ret;
196 
197     old_l2_offset = s->l1_table[l1_index];
198 
199     trace_qcow2_l2_allocate(bs, l1_index);
200 
201     /* allocate a new l2 entry */
202 
203     l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
204     if (l2_offset < 0) {
205         ret = l2_offset;
206         goto fail;
207     }
208 
209     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
210     if (ret < 0) {
211         goto fail;
212     }
213 
214     /* allocate a new entry in the l2 cache */
215 
216     trace_qcow2_l2_allocate_get_empty(bs, l1_index);
217     ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table);
218     if (ret < 0) {
219         goto fail;
220     }
221 
222     l2_table = *table;
223 
224     if ((old_l2_offset & L1E_OFFSET_MASK) == 0) {
225         /* if there was no old l2 table, clear the new table */
226         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
227     } else {
228         uint64_t* old_table;
229 
230         /* if there was an old l2 table, read it from the disk */
231         BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ);
232         ret = qcow2_cache_get(bs, s->l2_table_cache,
233             old_l2_offset & L1E_OFFSET_MASK,
234             (void**) &old_table);
235         if (ret < 0) {
236             goto fail;
237         }
238 
239         memcpy(l2_table, old_table, s->cluster_size);
240 
241         ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &old_table);
242         if (ret < 0) {
243             goto fail;
244         }
245     }
246 
247     /* write the l2 table to the file */
248     BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
249 
250     trace_qcow2_l2_allocate_write_l2(bs, l1_index);
251     qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
252     ret = qcow2_cache_flush(bs, s->l2_table_cache);
253     if (ret < 0) {
254         goto fail;
255     }
256 
257     /* update the L1 entry */
258     trace_qcow2_l2_allocate_write_l1(bs, l1_index);
259     s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
260     ret = qcow2_write_l1_entry(bs, l1_index);
261     if (ret < 0) {
262         goto fail;
263     }
264 
265     *table = l2_table;
266     trace_qcow2_l2_allocate_done(bs, l1_index, 0);
267     return 0;
268 
269 fail:
270     trace_qcow2_l2_allocate_done(bs, l1_index, ret);
271     if (l2_table != NULL) {
272         qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
273     }
274     s->l1_table[l1_index] = old_l2_offset;
275     if (l2_offset > 0) {
276         qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
277                             QCOW2_DISCARD_ALWAYS);
278     }
279     return ret;
280 }
281 
282 /*
283  * Checks how many clusters in a given L2 table are contiguous in the image
284  * file. As soon as one of the flags in the bitmask stop_flags changes compared
285  * to the first cluster, the search is stopped and the cluster is not counted
286  * as contiguous. (This allows it, for example, to stop at the first compressed
287  * cluster which may require a different handling)
288  */
289 static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
290         uint64_t *l2_table, uint64_t stop_flags)
291 {
292     int i;
293     uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED;
294     uint64_t first_entry = be64_to_cpu(l2_table[0]);
295     uint64_t offset = first_entry & mask;
296 
297     if (!offset)
298         return 0;
299 
300     assert(qcow2_get_cluster_type(first_entry) != QCOW2_CLUSTER_COMPRESSED);
301 
302     for (i = 0; i < nb_clusters; i++) {
303         uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask;
304         if (offset + (uint64_t) i * cluster_size != l2_entry) {
305             break;
306         }
307     }
308 
309 	return i;
310 }
311 
312 static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
313 {
314     int i;
315 
316     for (i = 0; i < nb_clusters; i++) {
317         int type = qcow2_get_cluster_type(be64_to_cpu(l2_table[i]));
318 
319         if (type != QCOW2_CLUSTER_UNALLOCATED) {
320             break;
321         }
322     }
323 
324     return i;
325 }
326 
327 /* The crypt function is compatible with the linux cryptoloop
328    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
329    supported */
330 void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
331                            uint8_t *out_buf, const uint8_t *in_buf,
332                            int nb_sectors, int enc,
333                            const AES_KEY *key)
334 {
335     union {
336         uint64_t ll[2];
337         uint8_t b[16];
338     } ivec;
339     int i;
340 
341     for(i = 0; i < nb_sectors; i++) {
342         ivec.ll[0] = cpu_to_le64(sector_num);
343         ivec.ll[1] = 0;
344         AES_cbc_encrypt(in_buf, out_buf, 512, key,
345                         ivec.b, enc);
346         sector_num++;
347         in_buf += 512;
348         out_buf += 512;
349     }
350 }
351 
352 static int coroutine_fn copy_sectors(BlockDriverState *bs,
353                                      uint64_t start_sect,
354                                      uint64_t cluster_offset,
355                                      int n_start, int n_end)
356 {
357     BDRVQcowState *s = bs->opaque;
358     QEMUIOVector qiov;
359     struct iovec iov;
360     int n, ret;
361 
362     n = n_end - n_start;
363     if (n <= 0) {
364         return 0;
365     }
366 
367     iov.iov_len = n * BDRV_SECTOR_SIZE;
368     iov.iov_base = qemu_blockalign(bs, iov.iov_len);
369 
370     qemu_iovec_init_external(&qiov, &iov, 1);
371 
372     BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
373 
374     if (!bs->drv) {
375         return -ENOMEDIUM;
376     }
377 
378     /* Call .bdrv_co_readv() directly instead of using the public block-layer
379      * interface.  This avoids double I/O throttling and request tracking,
380      * which can lead to deadlock when block layer copy-on-read is enabled.
381      */
382     ret = bs->drv->bdrv_co_readv(bs, start_sect + n_start, n, &qiov);
383     if (ret < 0) {
384         goto out;
385     }
386 
387     if (s->crypt_method) {
388         qcow2_encrypt_sectors(s, start_sect + n_start,
389                         iov.iov_base, iov.iov_base, n, 1,
390                         &s->aes_encrypt_key);
391     }
392 
393     ret = qcow2_pre_write_overlap_check(bs, 0,
394             cluster_offset + n_start * BDRV_SECTOR_SIZE, n * BDRV_SECTOR_SIZE);
395     if (ret < 0) {
396         goto out;
397     }
398 
399     BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
400     ret = bdrv_co_writev(bs->file, (cluster_offset >> 9) + n_start, n, &qiov);
401     if (ret < 0) {
402         goto out;
403     }
404 
405     ret = 0;
406 out:
407     qemu_vfree(iov.iov_base);
408     return ret;
409 }
410 
411 
412 /*
413  * get_cluster_offset
414  *
415  * For a given offset of the disk image, find the cluster offset in
416  * qcow2 file. The offset is stored in *cluster_offset.
417  *
418  * on entry, *num is the number of contiguous sectors we'd like to
419  * access following offset.
420  *
421  * on exit, *num is the number of contiguous sectors we can read.
422  *
423  * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error
424  * cases.
425  */
426 int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
427     int *num, uint64_t *cluster_offset)
428 {
429     BDRVQcowState *s = bs->opaque;
430     unsigned int l2_index;
431     uint64_t l1_index, l2_offset, *l2_table;
432     int l1_bits, c;
433     unsigned int index_in_cluster, nb_clusters;
434     uint64_t nb_available, nb_needed;
435     int ret;
436 
437     index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
438     nb_needed = *num + index_in_cluster;
439 
440     l1_bits = s->l2_bits + s->cluster_bits;
441 
442     /* compute how many bytes there are between the offset and
443      * the end of the l1 entry
444      */
445 
446     nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
447 
448     /* compute the number of available sectors */
449 
450     nb_available = (nb_available >> 9) + index_in_cluster;
451 
452     if (nb_needed > nb_available) {
453         nb_needed = nb_available;
454     }
455 
456     *cluster_offset = 0;
457 
458     /* seek the the l2 offset in the l1 table */
459 
460     l1_index = offset >> l1_bits;
461     if (l1_index >= s->l1_size) {
462         ret = QCOW2_CLUSTER_UNALLOCATED;
463         goto out;
464     }
465 
466     l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
467     if (!l2_offset) {
468         ret = QCOW2_CLUSTER_UNALLOCATED;
469         goto out;
470     }
471 
472     /* load the l2 table in memory */
473 
474     ret = l2_load(bs, l2_offset, &l2_table);
475     if (ret < 0) {
476         return ret;
477     }
478 
479     /* find the cluster offset for the given disk offset */
480 
481     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
482     *cluster_offset = be64_to_cpu(l2_table[l2_index]);
483     nb_clusters = size_to_clusters(s, nb_needed << 9);
484 
485     ret = qcow2_get_cluster_type(*cluster_offset);
486     switch (ret) {
487     case QCOW2_CLUSTER_COMPRESSED:
488         /* Compressed clusters can only be processed one by one */
489         c = 1;
490         *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK;
491         break;
492     case QCOW2_CLUSTER_ZERO:
493         if (s->qcow_version < 3) {
494             qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
495             return -EIO;
496         }
497         c = count_contiguous_clusters(nb_clusters, s->cluster_size,
498                 &l2_table[l2_index], QCOW_OFLAG_ZERO);
499         *cluster_offset = 0;
500         break;
501     case QCOW2_CLUSTER_UNALLOCATED:
502         /* how many empty clusters ? */
503         c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
504         *cluster_offset = 0;
505         break;
506     case QCOW2_CLUSTER_NORMAL:
507         /* how many allocated clusters ? */
508         c = count_contiguous_clusters(nb_clusters, s->cluster_size,
509                 &l2_table[l2_index], QCOW_OFLAG_ZERO);
510         *cluster_offset &= L2E_OFFSET_MASK;
511         break;
512     default:
513         abort();
514     }
515 
516     qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
517 
518     nb_available = (c * s->cluster_sectors);
519 
520 out:
521     if (nb_available > nb_needed)
522         nb_available = nb_needed;
523 
524     *num = nb_available - index_in_cluster;
525 
526     return ret;
527 }
528 
529 /*
530  * get_cluster_table
531  *
532  * for a given disk offset, load (and allocate if needed)
533  * the l2 table.
534  *
535  * the l2 table offset in the qcow2 file and the cluster index
536  * in the l2 table are given to the caller.
537  *
538  * Returns 0 on success, -errno in failure case
539  */
540 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
541                              uint64_t **new_l2_table,
542                              int *new_l2_index)
543 {
544     BDRVQcowState *s = bs->opaque;
545     unsigned int l2_index;
546     uint64_t l1_index, l2_offset;
547     uint64_t *l2_table = NULL;
548     int ret;
549 
550     /* seek the the l2 offset in the l1 table */
551 
552     l1_index = offset >> (s->l2_bits + s->cluster_bits);
553     if (l1_index >= s->l1_size) {
554         ret = qcow2_grow_l1_table(bs, l1_index + 1, false);
555         if (ret < 0) {
556             return ret;
557         }
558     }
559 
560     assert(l1_index < s->l1_size);
561     l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
562 
563     /* seek the l2 table of the given l2 offset */
564 
565     if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) {
566         /* load the l2 table in memory */
567         ret = l2_load(bs, l2_offset, &l2_table);
568         if (ret < 0) {
569             return ret;
570         }
571     } else {
572         /* First allocate a new L2 table (and do COW if needed) */
573         ret = l2_allocate(bs, l1_index, &l2_table);
574         if (ret < 0) {
575             return ret;
576         }
577 
578         /* Then decrease the refcount of the old table */
579         if (l2_offset) {
580             qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
581                                 QCOW2_DISCARD_OTHER);
582         }
583     }
584 
585     /* find the cluster offset for the given disk offset */
586 
587     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
588 
589     *new_l2_table = l2_table;
590     *new_l2_index = l2_index;
591 
592     return 0;
593 }
594 
595 /*
596  * alloc_compressed_cluster_offset
597  *
598  * For a given offset of the disk image, return cluster offset in
599  * qcow2 file.
600  *
601  * If the offset is not found, allocate a new compressed cluster.
602  *
603  * Return the cluster offset if successful,
604  * Return 0, otherwise.
605  *
606  */
607 
608 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
609                                                uint64_t offset,
610                                                int compressed_size)
611 {
612     BDRVQcowState *s = bs->opaque;
613     int l2_index, ret;
614     uint64_t *l2_table;
615     int64_t cluster_offset;
616     int nb_csectors;
617 
618     ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
619     if (ret < 0) {
620         return 0;
621     }
622 
623     /* Compression can't overwrite anything. Fail if the cluster was already
624      * allocated. */
625     cluster_offset = be64_to_cpu(l2_table[l2_index]);
626     if (cluster_offset & L2E_OFFSET_MASK) {
627         qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
628         return 0;
629     }
630 
631     cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
632     if (cluster_offset < 0) {
633         qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
634         return 0;
635     }
636 
637     nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
638                   (cluster_offset >> 9);
639 
640     cluster_offset |= QCOW_OFLAG_COMPRESSED |
641                       ((uint64_t)nb_csectors << s->csize_shift);
642 
643     /* update L2 table */
644 
645     /* compressed clusters never have the copied flag */
646 
647     BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
648     qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
649     l2_table[l2_index] = cpu_to_be64(cluster_offset);
650     ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
651     if (ret < 0) {
652         return 0;
653     }
654 
655     return cluster_offset;
656 }
657 
658 static int perform_cow(BlockDriverState *bs, QCowL2Meta *m, Qcow2COWRegion *r)
659 {
660     BDRVQcowState *s = bs->opaque;
661     int ret;
662 
663     if (r->nb_sectors == 0) {
664         return 0;
665     }
666 
667     qemu_co_mutex_unlock(&s->lock);
668     ret = copy_sectors(bs, m->offset / BDRV_SECTOR_SIZE, m->alloc_offset,
669                        r->offset / BDRV_SECTOR_SIZE,
670                        r->offset / BDRV_SECTOR_SIZE + r->nb_sectors);
671     qemu_co_mutex_lock(&s->lock);
672 
673     if (ret < 0) {
674         return ret;
675     }
676 
677     /*
678      * Before we update the L2 table to actually point to the new cluster, we
679      * need to be sure that the refcounts have been increased and COW was
680      * handled.
681      */
682     qcow2_cache_depends_on_flush(s->l2_table_cache);
683 
684     return 0;
685 }
686 
687 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
688 {
689     BDRVQcowState *s = bs->opaque;
690     int i, j = 0, l2_index, ret;
691     uint64_t *old_cluster, *l2_table;
692     uint64_t cluster_offset = m->alloc_offset;
693 
694     trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters);
695     assert(m->nb_clusters > 0);
696 
697     old_cluster = g_malloc(m->nb_clusters * sizeof(uint64_t));
698 
699     /* copy content of unmodified sectors */
700     ret = perform_cow(bs, m, &m->cow_start);
701     if (ret < 0) {
702         goto err;
703     }
704 
705     ret = perform_cow(bs, m, &m->cow_end);
706     if (ret < 0) {
707         goto err;
708     }
709 
710     /* Update L2 table. */
711     if (s->use_lazy_refcounts) {
712         qcow2_mark_dirty(bs);
713     }
714     if (qcow2_need_accurate_refcounts(s)) {
715         qcow2_cache_set_dependency(bs, s->l2_table_cache,
716                                    s->refcount_block_cache);
717     }
718 
719     ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index);
720     if (ret < 0) {
721         goto err;
722     }
723     qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
724 
725     assert(l2_index + m->nb_clusters <= s->l2_size);
726     for (i = 0; i < m->nb_clusters; i++) {
727         /* if two concurrent writes happen to the same unallocated cluster
728 	 * each write allocates separate cluster and writes data concurrently.
729 	 * The first one to complete updates l2 table with pointer to its
730 	 * cluster the second one has to do RMW (which is done above by
731 	 * copy_sectors()), update l2 table with its cluster pointer and free
732 	 * old cluster. This is what this loop does */
733         if(l2_table[l2_index + i] != 0)
734             old_cluster[j++] = l2_table[l2_index + i];
735 
736         l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
737                     (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
738      }
739 
740 
741     ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
742     if (ret < 0) {
743         goto err;
744     }
745 
746     /*
747      * If this was a COW, we need to decrease the refcount of the old cluster.
748      * Also flush bs->file to get the right order for L2 and refcount update.
749      *
750      * Don't discard clusters that reach a refcount of 0 (e.g. compressed
751      * clusters), the next write will reuse them anyway.
752      */
753     if (j != 0) {
754         for (i = 0; i < j; i++) {
755             qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1,
756                                     QCOW2_DISCARD_NEVER);
757         }
758     }
759 
760     ret = 0;
761 err:
762     g_free(old_cluster);
763     return ret;
764  }
765 
766 /*
767  * Returns the number of contiguous clusters that can be used for an allocating
768  * write, but require COW to be performed (this includes yet unallocated space,
769  * which must copy from the backing file)
770  */
771 static int count_cow_clusters(BDRVQcowState *s, int nb_clusters,
772     uint64_t *l2_table, int l2_index)
773 {
774     int i;
775 
776     for (i = 0; i < nb_clusters; i++) {
777         uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]);
778         int cluster_type = qcow2_get_cluster_type(l2_entry);
779 
780         switch(cluster_type) {
781         case QCOW2_CLUSTER_NORMAL:
782             if (l2_entry & QCOW_OFLAG_COPIED) {
783                 goto out;
784             }
785             break;
786         case QCOW2_CLUSTER_UNALLOCATED:
787         case QCOW2_CLUSTER_COMPRESSED:
788         case QCOW2_CLUSTER_ZERO:
789             break;
790         default:
791             abort();
792         }
793     }
794 
795 out:
796     assert(i <= nb_clusters);
797     return i;
798 }
799 
800 /*
801  * Check if there already is an AIO write request in flight which allocates
802  * the same cluster. In this case we need to wait until the previous
803  * request has completed and updated the L2 table accordingly.
804  *
805  * Returns:
806  *   0       if there was no dependency. *cur_bytes indicates the number of
807  *           bytes from guest_offset that can be read before the next
808  *           dependency must be processed (or the request is complete)
809  *
810  *   -EAGAIN if we had to wait for another request, previously gathered
811  *           information on cluster allocation may be invalid now. The caller
812  *           must start over anyway, so consider *cur_bytes undefined.
813  */
814 static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset,
815     uint64_t *cur_bytes, QCowL2Meta **m)
816 {
817     BDRVQcowState *s = bs->opaque;
818     QCowL2Meta *old_alloc;
819     uint64_t bytes = *cur_bytes;
820 
821     QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
822 
823         uint64_t start = guest_offset;
824         uint64_t end = start + bytes;
825         uint64_t old_start = l2meta_cow_start(old_alloc);
826         uint64_t old_end = l2meta_cow_end(old_alloc);
827 
828         if (end <= old_start || start >= old_end) {
829             /* No intersection */
830         } else {
831             if (start < old_start) {
832                 /* Stop at the start of a running allocation */
833                 bytes = old_start - start;
834             } else {
835                 bytes = 0;
836             }
837 
838             /* Stop if already an l2meta exists. After yielding, it wouldn't
839              * be valid any more, so we'd have to clean up the old L2Metas
840              * and deal with requests depending on them before starting to
841              * gather new ones. Not worth the trouble. */
842             if (bytes == 0 && *m) {
843                 *cur_bytes = 0;
844                 return 0;
845             }
846 
847             if (bytes == 0) {
848                 /* Wait for the dependency to complete. We need to recheck
849                  * the free/allocated clusters when we continue. */
850                 qemu_co_mutex_unlock(&s->lock);
851                 qemu_co_queue_wait(&old_alloc->dependent_requests);
852                 qemu_co_mutex_lock(&s->lock);
853                 return -EAGAIN;
854             }
855         }
856     }
857 
858     /* Make sure that existing clusters and new allocations are only used up to
859      * the next dependency if we shortened the request above */
860     *cur_bytes = bytes;
861 
862     return 0;
863 }
864 
865 /*
866  * Checks how many already allocated clusters that don't require a copy on
867  * write there are at the given guest_offset (up to *bytes). If
868  * *host_offset is not zero, only physically contiguous clusters beginning at
869  * this host offset are counted.
870  *
871  * Note that guest_offset may not be cluster aligned. In this case, the
872  * returned *host_offset points to exact byte referenced by guest_offset and
873  * therefore isn't cluster aligned as well.
874  *
875  * Returns:
876  *   0:     if no allocated clusters are available at the given offset.
877  *          *bytes is normally unchanged. It is set to 0 if the cluster
878  *          is allocated and doesn't need COW, but doesn't have the right
879  *          physical offset.
880  *
881  *   1:     if allocated clusters that don't require a COW are available at
882  *          the requested offset. *bytes may have decreased and describes
883  *          the length of the area that can be written to.
884  *
885  *  -errno: in error cases
886  */
887 static int handle_copied(BlockDriverState *bs, uint64_t guest_offset,
888     uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
889 {
890     BDRVQcowState *s = bs->opaque;
891     int l2_index;
892     uint64_t cluster_offset;
893     uint64_t *l2_table;
894     unsigned int nb_clusters;
895     unsigned int keep_clusters;
896     int ret, pret;
897 
898     trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset,
899                               *bytes);
900 
901     assert(*host_offset == 0 ||    offset_into_cluster(s, guest_offset)
902                                 == offset_into_cluster(s, *host_offset));
903 
904     /*
905      * Calculate the number of clusters to look for. We stop at L2 table
906      * boundaries to keep things simple.
907      */
908     nb_clusters =
909         size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
910 
911     l2_index = offset_to_l2_index(s, guest_offset);
912     nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
913 
914     /* Find L2 entry for the first involved cluster */
915     ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index);
916     if (ret < 0) {
917         return ret;
918     }
919 
920     cluster_offset = be64_to_cpu(l2_table[l2_index]);
921 
922     /* Check how many clusters are already allocated and don't need COW */
923     if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL
924         && (cluster_offset & QCOW_OFLAG_COPIED))
925     {
926         /* If a specific host_offset is required, check it */
927         bool offset_matches =
928             (cluster_offset & L2E_OFFSET_MASK) == *host_offset;
929 
930         if (*host_offset != 0 && !offset_matches) {
931             *bytes = 0;
932             ret = 0;
933             goto out;
934         }
935 
936         /* We keep all QCOW_OFLAG_COPIED clusters */
937         keep_clusters =
938             count_contiguous_clusters(nb_clusters, s->cluster_size,
939                                       &l2_table[l2_index],
940                                       QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO);
941         assert(keep_clusters <= nb_clusters);
942 
943         *bytes = MIN(*bytes,
944                  keep_clusters * s->cluster_size
945                  - offset_into_cluster(s, guest_offset));
946 
947         ret = 1;
948     } else {
949         ret = 0;
950     }
951 
952     /* Cleanup */
953 out:
954     pret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
955     if (pret < 0) {
956         return pret;
957     }
958 
959     /* Only return a host offset if we actually made progress. Otherwise we
960      * would make requirements for handle_alloc() that it can't fulfill */
961     if (ret) {
962         *host_offset = (cluster_offset & L2E_OFFSET_MASK)
963                      + offset_into_cluster(s, guest_offset);
964     }
965 
966     return ret;
967 }
968 
969 /*
970  * Allocates new clusters for the given guest_offset.
971  *
972  * At most *nb_clusters are allocated, and on return *nb_clusters is updated to
973  * contain the number of clusters that have been allocated and are contiguous
974  * in the image file.
975  *
976  * If *host_offset is non-zero, it specifies the offset in the image file at
977  * which the new clusters must start. *nb_clusters can be 0 on return in this
978  * case if the cluster at host_offset is already in use. If *host_offset is
979  * zero, the clusters can be allocated anywhere in the image file.
980  *
981  * *host_offset is updated to contain the offset into the image file at which
982  * the first allocated cluster starts.
983  *
984  * Return 0 on success and -errno in error cases. -EAGAIN means that the
985  * function has been waiting for another request and the allocation must be
986  * restarted, but the whole request should not be failed.
987  */
988 static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset,
989     uint64_t *host_offset, unsigned int *nb_clusters)
990 {
991     BDRVQcowState *s = bs->opaque;
992 
993     trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset,
994                                          *host_offset, *nb_clusters);
995 
996     /* Allocate new clusters */
997     trace_qcow2_cluster_alloc_phys(qemu_coroutine_self());
998     if (*host_offset == 0) {
999         int64_t cluster_offset =
1000             qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size);
1001         if (cluster_offset < 0) {
1002             return cluster_offset;
1003         }
1004         *host_offset = cluster_offset;
1005         return 0;
1006     } else {
1007         int ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters);
1008         if (ret < 0) {
1009             return ret;
1010         }
1011         *nb_clusters = ret;
1012         return 0;
1013     }
1014 }
1015 
1016 /*
1017  * Allocates new clusters for an area that either is yet unallocated or needs a
1018  * copy on write. If *host_offset is non-zero, clusters are only allocated if
1019  * the new allocation can match the specified host offset.
1020  *
1021  * Note that guest_offset may not be cluster aligned. In this case, the
1022  * returned *host_offset points to exact byte referenced by guest_offset and
1023  * therefore isn't cluster aligned as well.
1024  *
1025  * Returns:
1026  *   0:     if no clusters could be allocated. *bytes is set to 0,
1027  *          *host_offset is left unchanged.
1028  *
1029  *   1:     if new clusters were allocated. *bytes may be decreased if the
1030  *          new allocation doesn't cover all of the requested area.
1031  *          *host_offset is updated to contain the host offset of the first
1032  *          newly allocated cluster.
1033  *
1034  *  -errno: in error cases
1035  */
1036 static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
1037     uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
1038 {
1039     BDRVQcowState *s = bs->opaque;
1040     int l2_index;
1041     uint64_t *l2_table;
1042     uint64_t entry;
1043     unsigned int nb_clusters;
1044     int ret;
1045 
1046     uint64_t alloc_cluster_offset;
1047 
1048     trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset,
1049                              *bytes);
1050     assert(*bytes > 0);
1051 
1052     /*
1053      * Calculate the number of clusters to look for. We stop at L2 table
1054      * boundaries to keep things simple.
1055      */
1056     nb_clusters =
1057         size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
1058 
1059     l2_index = offset_to_l2_index(s, guest_offset);
1060     nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1061 
1062     /* Find L2 entry for the first involved cluster */
1063     ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index);
1064     if (ret < 0) {
1065         return ret;
1066     }
1067 
1068     entry = be64_to_cpu(l2_table[l2_index]);
1069 
1070     /* For the moment, overwrite compressed clusters one by one */
1071     if (entry & QCOW_OFLAG_COMPRESSED) {
1072         nb_clusters = 1;
1073     } else {
1074         nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index);
1075     }
1076 
1077     /* This function is only called when there were no non-COW clusters, so if
1078      * we can't find any unallocated or COW clusters either, something is
1079      * wrong with our code. */
1080     assert(nb_clusters > 0);
1081 
1082     ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1083     if (ret < 0) {
1084         return ret;
1085     }
1086 
1087     /* Allocate, if necessary at a given offset in the image file */
1088     alloc_cluster_offset = start_of_cluster(s, *host_offset);
1089     ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset,
1090                                   &nb_clusters);
1091     if (ret < 0) {
1092         goto fail;
1093     }
1094 
1095     /* Can't extend contiguous allocation */
1096     if (nb_clusters == 0) {
1097         *bytes = 0;
1098         return 0;
1099     }
1100 
1101     /*
1102      * Save info needed for meta data update.
1103      *
1104      * requested_sectors: Number of sectors from the start of the first
1105      * newly allocated cluster to the end of the (possibly shortened
1106      * before) write request.
1107      *
1108      * avail_sectors: Number of sectors from the start of the first
1109      * newly allocated to the end of the last newly allocated cluster.
1110      *
1111      * nb_sectors: The number of sectors from the start of the first
1112      * newly allocated cluster to the end of the area that the write
1113      * request actually writes to (excluding COW at the end)
1114      */
1115     int requested_sectors =
1116         (*bytes + offset_into_cluster(s, guest_offset))
1117         >> BDRV_SECTOR_BITS;
1118     int avail_sectors = nb_clusters
1119                         << (s->cluster_bits - BDRV_SECTOR_BITS);
1120     int alloc_n_start = offset_into_cluster(s, guest_offset)
1121                         >> BDRV_SECTOR_BITS;
1122     int nb_sectors = MIN(requested_sectors, avail_sectors);
1123     QCowL2Meta *old_m = *m;
1124 
1125     *m = g_malloc0(sizeof(**m));
1126 
1127     **m = (QCowL2Meta) {
1128         .next           = old_m,
1129 
1130         .alloc_offset   = alloc_cluster_offset,
1131         .offset         = start_of_cluster(s, guest_offset),
1132         .nb_clusters    = nb_clusters,
1133         .nb_available   = nb_sectors,
1134 
1135         .cow_start = {
1136             .offset     = 0,
1137             .nb_sectors = alloc_n_start,
1138         },
1139         .cow_end = {
1140             .offset     = nb_sectors * BDRV_SECTOR_SIZE,
1141             .nb_sectors = avail_sectors - nb_sectors,
1142         },
1143     };
1144     qemu_co_queue_init(&(*m)->dependent_requests);
1145     QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
1146 
1147     *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset);
1148     *bytes = MIN(*bytes, (nb_sectors * BDRV_SECTOR_SIZE)
1149                          - offset_into_cluster(s, guest_offset));
1150     assert(*bytes != 0);
1151 
1152     return 1;
1153 
1154 fail:
1155     if (*m && (*m)->nb_clusters > 0) {
1156         QLIST_REMOVE(*m, next_in_flight);
1157     }
1158     return ret;
1159 }
1160 
1161 /*
1162  * alloc_cluster_offset
1163  *
1164  * For a given offset on the virtual disk, find the cluster offset in qcow2
1165  * file. If the offset is not found, allocate a new cluster.
1166  *
1167  * If the cluster was already allocated, m->nb_clusters is set to 0 and
1168  * other fields in m are meaningless.
1169  *
1170  * If the cluster is newly allocated, m->nb_clusters is set to the number of
1171  * contiguous clusters that have been allocated. In this case, the other
1172  * fields of m are valid and contain information about the first allocated
1173  * cluster.
1174  *
1175  * If the request conflicts with another write request in flight, the coroutine
1176  * is queued and will be reentered when the dependency has completed.
1177  *
1178  * Return 0 on success and -errno in error cases
1179  */
1180 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
1181     int *num, uint64_t *host_offset, QCowL2Meta **m)
1182 {
1183     BDRVQcowState *s = bs->opaque;
1184     uint64_t start, remaining;
1185     uint64_t cluster_offset;
1186     uint64_t cur_bytes;
1187     int ret;
1188 
1189     trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *num);
1190 
1191     assert((offset & ~BDRV_SECTOR_MASK) == 0);
1192 
1193 again:
1194     start = offset;
1195     remaining = *num << BDRV_SECTOR_BITS;
1196     cluster_offset = 0;
1197     *host_offset = 0;
1198     cur_bytes = 0;
1199     *m = NULL;
1200 
1201     while (true) {
1202 
1203         if (!*host_offset) {
1204             *host_offset = start_of_cluster(s, cluster_offset);
1205         }
1206 
1207         assert(remaining >= cur_bytes);
1208 
1209         start           += cur_bytes;
1210         remaining       -= cur_bytes;
1211         cluster_offset  += cur_bytes;
1212 
1213         if (remaining == 0) {
1214             break;
1215         }
1216 
1217         cur_bytes = remaining;
1218 
1219         /*
1220          * Now start gathering as many contiguous clusters as possible:
1221          *
1222          * 1. Check for overlaps with in-flight allocations
1223          *
1224          *      a) Overlap not in the first cluster -> shorten this request and
1225          *         let the caller handle the rest in its next loop iteration.
1226          *
1227          *      b) Real overlaps of two requests. Yield and restart the search
1228          *         for contiguous clusters (the situation could have changed
1229          *         while we were sleeping)
1230          *
1231          *      c) TODO: Request starts in the same cluster as the in-flight
1232          *         allocation ends. Shorten the COW of the in-fight allocation,
1233          *         set cluster_offset to write to the same cluster and set up
1234          *         the right synchronisation between the in-flight request and
1235          *         the new one.
1236          */
1237         ret = handle_dependencies(bs, start, &cur_bytes, m);
1238         if (ret == -EAGAIN) {
1239             /* Currently handle_dependencies() doesn't yield if we already had
1240              * an allocation. If it did, we would have to clean up the L2Meta
1241              * structs before starting over. */
1242             assert(*m == NULL);
1243             goto again;
1244         } else if (ret < 0) {
1245             return ret;
1246         } else if (cur_bytes == 0) {
1247             break;
1248         } else {
1249             /* handle_dependencies() may have decreased cur_bytes (shortened
1250              * the allocations below) so that the next dependency is processed
1251              * correctly during the next loop iteration. */
1252         }
1253 
1254         /*
1255          * 2. Count contiguous COPIED clusters.
1256          */
1257         ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m);
1258         if (ret < 0) {
1259             return ret;
1260         } else if (ret) {
1261             continue;
1262         } else if (cur_bytes == 0) {
1263             break;
1264         }
1265 
1266         /*
1267          * 3. If the request still hasn't completed, allocate new clusters,
1268          *    considering any cluster_offset of steps 1c or 2.
1269          */
1270         ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m);
1271         if (ret < 0) {
1272             return ret;
1273         } else if (ret) {
1274             continue;
1275         } else {
1276             assert(cur_bytes == 0);
1277             break;
1278         }
1279     }
1280 
1281     *num -= remaining >> BDRV_SECTOR_BITS;
1282     assert(*num > 0);
1283     assert(*host_offset != 0);
1284 
1285     return 0;
1286 }
1287 
1288 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1289                              const uint8_t *buf, int buf_size)
1290 {
1291     z_stream strm1, *strm = &strm1;
1292     int ret, out_len;
1293 
1294     memset(strm, 0, sizeof(*strm));
1295 
1296     strm->next_in = (uint8_t *)buf;
1297     strm->avail_in = buf_size;
1298     strm->next_out = out_buf;
1299     strm->avail_out = out_buf_size;
1300 
1301     ret = inflateInit2(strm, -12);
1302     if (ret != Z_OK)
1303         return -1;
1304     ret = inflate(strm, Z_FINISH);
1305     out_len = strm->next_out - out_buf;
1306     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1307         out_len != out_buf_size) {
1308         inflateEnd(strm);
1309         return -1;
1310     }
1311     inflateEnd(strm);
1312     return 0;
1313 }
1314 
1315 int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
1316 {
1317     BDRVQcowState *s = bs->opaque;
1318     int ret, csize, nb_csectors, sector_offset;
1319     uint64_t coffset;
1320 
1321     coffset = cluster_offset & s->cluster_offset_mask;
1322     if (s->cluster_cache_offset != coffset) {
1323         nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1324         sector_offset = coffset & 511;
1325         csize = nb_csectors * 512 - sector_offset;
1326         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
1327         ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, nb_csectors);
1328         if (ret < 0) {
1329             return ret;
1330         }
1331         if (decompress_buffer(s->cluster_cache, s->cluster_size,
1332                               s->cluster_data + sector_offset, csize) < 0) {
1333             return -EIO;
1334         }
1335         s->cluster_cache_offset = coffset;
1336     }
1337     return 0;
1338 }
1339 
1340 /*
1341  * This discards as many clusters of nb_clusters as possible at once (i.e.
1342  * all clusters in the same L2 table) and returns the number of discarded
1343  * clusters.
1344  */
1345 static int discard_single_l2(BlockDriverState *bs, uint64_t offset,
1346     unsigned int nb_clusters, enum qcow2_discard_type type)
1347 {
1348     BDRVQcowState *s = bs->opaque;
1349     uint64_t *l2_table;
1350     int l2_index;
1351     int ret;
1352     int i;
1353 
1354     ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1355     if (ret < 0) {
1356         return ret;
1357     }
1358 
1359     /* Limit nb_clusters to one L2 table */
1360     nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1361 
1362     for (i = 0; i < nb_clusters; i++) {
1363         uint64_t old_offset;
1364 
1365         old_offset = be64_to_cpu(l2_table[l2_index + i]);
1366 
1367         /*
1368          * Make sure that a discarded area reads back as zeroes for v3 images
1369          * (we cannot do it for v2 without actually writing a zero-filled
1370          * buffer). We can skip the operation if the cluster is already marked
1371          * as zero, or if it's unallocated and we don't have a backing file.
1372          *
1373          * TODO We might want to use bdrv_get_block_status(bs) here, but we're
1374          * holding s->lock, so that doesn't work today.
1375          */
1376         if (old_offset & QCOW_OFLAG_ZERO) {
1377             continue;
1378         }
1379 
1380         if ((old_offset & L2E_OFFSET_MASK) == 0 && !bs->backing_hd) {
1381             continue;
1382         }
1383 
1384         /* First remove L2 entries */
1385         qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
1386         if (s->qcow_version >= 3) {
1387             l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
1388         } else {
1389             l2_table[l2_index + i] = cpu_to_be64(0);
1390         }
1391 
1392         /* Then decrease the refcount */
1393         qcow2_free_any_clusters(bs, old_offset, 1, type);
1394     }
1395 
1396     ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1397     if (ret < 0) {
1398         return ret;
1399     }
1400 
1401     return nb_clusters;
1402 }
1403 
1404 int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
1405     int nb_sectors, enum qcow2_discard_type type)
1406 {
1407     BDRVQcowState *s = bs->opaque;
1408     uint64_t end_offset;
1409     unsigned int nb_clusters;
1410     int ret;
1411 
1412     end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS);
1413 
1414     /* Round start up and end down */
1415     offset = align_offset(offset, s->cluster_size);
1416     end_offset = start_of_cluster(s, end_offset);
1417 
1418     if (offset > end_offset) {
1419         return 0;
1420     }
1421 
1422     nb_clusters = size_to_clusters(s, end_offset - offset);
1423 
1424     s->cache_discards = true;
1425 
1426     /* Each L2 table is handled by its own loop iteration */
1427     while (nb_clusters > 0) {
1428         ret = discard_single_l2(bs, offset, nb_clusters, type);
1429         if (ret < 0) {
1430             goto fail;
1431         }
1432 
1433         nb_clusters -= ret;
1434         offset += (ret * s->cluster_size);
1435     }
1436 
1437     ret = 0;
1438 fail:
1439     s->cache_discards = false;
1440     qcow2_process_discards(bs, ret);
1441 
1442     return ret;
1443 }
1444 
1445 /*
1446  * This zeroes as many clusters of nb_clusters as possible at once (i.e.
1447  * all clusters in the same L2 table) and returns the number of zeroed
1448  * clusters.
1449  */
1450 static int zero_single_l2(BlockDriverState *bs, uint64_t offset,
1451     unsigned int nb_clusters)
1452 {
1453     BDRVQcowState *s = bs->opaque;
1454     uint64_t *l2_table;
1455     int l2_index;
1456     int ret;
1457     int i;
1458 
1459     ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1460     if (ret < 0) {
1461         return ret;
1462     }
1463 
1464     /* Limit nb_clusters to one L2 table */
1465     nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1466 
1467     for (i = 0; i < nb_clusters; i++) {
1468         uint64_t old_offset;
1469 
1470         old_offset = be64_to_cpu(l2_table[l2_index + i]);
1471 
1472         /* Update L2 entries */
1473         qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
1474         if (old_offset & QCOW_OFLAG_COMPRESSED) {
1475             l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
1476             qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST);
1477         } else {
1478             l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO);
1479         }
1480     }
1481 
1482     ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1483     if (ret < 0) {
1484         return ret;
1485     }
1486 
1487     return nb_clusters;
1488 }
1489 
1490 int qcow2_zero_clusters(BlockDriverState *bs, uint64_t offset, int nb_sectors)
1491 {
1492     BDRVQcowState *s = bs->opaque;
1493     unsigned int nb_clusters;
1494     int ret;
1495 
1496     /* The zero flag is only supported by version 3 and newer */
1497     if (s->qcow_version < 3) {
1498         return -ENOTSUP;
1499     }
1500 
1501     /* Each L2 table is handled by its own loop iteration */
1502     nb_clusters = size_to_clusters(s, nb_sectors << BDRV_SECTOR_BITS);
1503 
1504     s->cache_discards = true;
1505 
1506     while (nb_clusters > 0) {
1507         ret = zero_single_l2(bs, offset, nb_clusters);
1508         if (ret < 0) {
1509             goto fail;
1510         }
1511 
1512         nb_clusters -= ret;
1513         offset += (ret * s->cluster_size);
1514     }
1515 
1516     ret = 0;
1517 fail:
1518     s->cache_discards = false;
1519     qcow2_process_discards(bs, ret);
1520 
1521     return ret;
1522 }
1523 
1524 /*
1525  * Expands all zero clusters in a specific L1 table (or deallocates them, for
1526  * non-backed non-pre-allocated zero clusters).
1527  *
1528  * expanded_clusters is a bitmap where every bit corresponds to one cluster in
1529  * the image file; a bit gets set if the corresponding cluster has been used for
1530  * zero expansion (i.e., has been filled with zeroes and is referenced from an
1531  * L2 table). nb_clusters contains the total cluster count of the image file,
1532  * i.e., the number of bits in expanded_clusters.
1533  */
1534 static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
1535                                       int l1_size, uint8_t **expanded_clusters,
1536                                       uint64_t *nb_clusters)
1537 {
1538     BDRVQcowState *s = bs->opaque;
1539     bool is_active_l1 = (l1_table == s->l1_table);
1540     uint64_t *l2_table = NULL;
1541     int ret;
1542     int i, j;
1543 
1544     if (!is_active_l1) {
1545         /* inactive L2 tables require a buffer to be stored in when loading
1546          * them from disk */
1547         l2_table = qemu_blockalign(bs, s->cluster_size);
1548     }
1549 
1550     for (i = 0; i < l1_size; i++) {
1551         uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK;
1552         bool l2_dirty = false;
1553 
1554         if (!l2_offset) {
1555             /* unallocated */
1556             continue;
1557         }
1558 
1559         if (is_active_l1) {
1560             /* get active L2 tables from cache */
1561             ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
1562                     (void **)&l2_table);
1563         } else {
1564             /* load inactive L2 tables from disk */
1565             ret = bdrv_read(bs->file, l2_offset / BDRV_SECTOR_SIZE,
1566                     (void *)l2_table, s->cluster_sectors);
1567         }
1568         if (ret < 0) {
1569             goto fail;
1570         }
1571 
1572         for (j = 0; j < s->l2_size; j++) {
1573             uint64_t l2_entry = be64_to_cpu(l2_table[j]);
1574             int64_t offset = l2_entry & L2E_OFFSET_MASK, cluster_index;
1575             int cluster_type = qcow2_get_cluster_type(l2_entry);
1576             bool preallocated = offset != 0;
1577 
1578             if (cluster_type == QCOW2_CLUSTER_NORMAL) {
1579                 cluster_index = offset >> s->cluster_bits;
1580                 assert((cluster_index >= 0) && (cluster_index < *nb_clusters));
1581                 if ((*expanded_clusters)[cluster_index / 8] &
1582                     (1 << (cluster_index % 8))) {
1583                     /* Probably a shared L2 table; this cluster was a zero
1584                      * cluster which has been expanded, its refcount
1585                      * therefore most likely requires an update. */
1586                     ret = qcow2_update_cluster_refcount(bs, cluster_index, 1,
1587                                                         QCOW2_DISCARD_NEVER);
1588                     if (ret < 0) {
1589                         goto fail;
1590                     }
1591                     /* Since we just increased the refcount, the COPIED flag may
1592                      * no longer be set. */
1593                     l2_table[j] = cpu_to_be64(l2_entry & ~QCOW_OFLAG_COPIED);
1594                     l2_dirty = true;
1595                 }
1596                 continue;
1597             }
1598             else if (qcow2_get_cluster_type(l2_entry) != QCOW2_CLUSTER_ZERO) {
1599                 continue;
1600             }
1601 
1602             if (!preallocated) {
1603                 if (!bs->backing_hd) {
1604                     /* not backed; therefore we can simply deallocate the
1605                      * cluster */
1606                     l2_table[j] = 0;
1607                     l2_dirty = true;
1608                     continue;
1609                 }
1610 
1611                 offset = qcow2_alloc_clusters(bs, s->cluster_size);
1612                 if (offset < 0) {
1613                     ret = offset;
1614                     goto fail;
1615                 }
1616             }
1617 
1618             ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size);
1619             if (ret < 0) {
1620                 if (!preallocated) {
1621                     qcow2_free_clusters(bs, offset, s->cluster_size,
1622                                         QCOW2_DISCARD_ALWAYS);
1623                 }
1624                 goto fail;
1625             }
1626 
1627             ret = bdrv_write_zeroes(bs->file, offset / BDRV_SECTOR_SIZE,
1628                                     s->cluster_sectors, 0);
1629             if (ret < 0) {
1630                 if (!preallocated) {
1631                     qcow2_free_clusters(bs, offset, s->cluster_size,
1632                                         QCOW2_DISCARD_ALWAYS);
1633                 }
1634                 goto fail;
1635             }
1636 
1637             l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED);
1638             l2_dirty = true;
1639 
1640             cluster_index = offset >> s->cluster_bits;
1641 
1642             if (cluster_index >= *nb_clusters) {
1643                 uint64_t old_bitmap_size = (*nb_clusters + 7) / 8;
1644                 uint64_t new_bitmap_size;
1645                 /* The offset may lie beyond the old end of the underlying image
1646                  * file for growable files only */
1647                 assert(bs->file->growable);
1648                 *nb_clusters = size_to_clusters(s, bs->file->total_sectors *
1649                                                 BDRV_SECTOR_SIZE);
1650                 new_bitmap_size = (*nb_clusters + 7) / 8;
1651                 *expanded_clusters = g_realloc(*expanded_clusters,
1652                                                new_bitmap_size);
1653                 /* clear the newly allocated space */
1654                 memset(&(*expanded_clusters)[old_bitmap_size], 0,
1655                        new_bitmap_size - old_bitmap_size);
1656             }
1657 
1658             assert((cluster_index >= 0) && (cluster_index < *nb_clusters));
1659             (*expanded_clusters)[cluster_index / 8] |= 1 << (cluster_index % 8);
1660         }
1661 
1662         if (is_active_l1) {
1663             if (l2_dirty) {
1664                 qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
1665                 qcow2_cache_depends_on_flush(s->l2_table_cache);
1666             }
1667             ret = qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table);
1668             if (ret < 0) {
1669                 l2_table = NULL;
1670                 goto fail;
1671             }
1672         } else {
1673             if (l2_dirty) {
1674                 ret = qcow2_pre_write_overlap_check(bs,
1675                         QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, l2_offset,
1676                         s->cluster_size);
1677                 if (ret < 0) {
1678                     goto fail;
1679                 }
1680 
1681                 ret = bdrv_write(bs->file, l2_offset / BDRV_SECTOR_SIZE,
1682                         (void *)l2_table, s->cluster_sectors);
1683                 if (ret < 0) {
1684                     goto fail;
1685                 }
1686             }
1687         }
1688     }
1689 
1690     ret = 0;
1691 
1692 fail:
1693     if (l2_table) {
1694         if (!is_active_l1) {
1695             qemu_vfree(l2_table);
1696         } else {
1697             if (ret < 0) {
1698                 qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table);
1699             } else {
1700                 ret = qcow2_cache_put(bs, s->l2_table_cache,
1701                         (void **)&l2_table);
1702             }
1703         }
1704     }
1705     return ret;
1706 }
1707 
1708 /*
1709  * For backed images, expands all zero clusters on the image. For non-backed
1710  * images, deallocates all non-pre-allocated zero clusters (and claims the
1711  * allocation for pre-allocated ones). This is important for downgrading to a
1712  * qcow2 version which doesn't yet support metadata zero clusters.
1713  */
1714 int qcow2_expand_zero_clusters(BlockDriverState *bs)
1715 {
1716     BDRVQcowState *s = bs->opaque;
1717     uint64_t *l1_table = NULL;
1718     uint64_t nb_clusters;
1719     uint8_t *expanded_clusters;
1720     int ret;
1721     int i, j;
1722 
1723     nb_clusters = size_to_clusters(s, bs->file->total_sectors *
1724                                    BDRV_SECTOR_SIZE);
1725     expanded_clusters = g_malloc0((nb_clusters + 7) / 8);
1726 
1727     ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size,
1728                                      &expanded_clusters, &nb_clusters);
1729     if (ret < 0) {
1730         goto fail;
1731     }
1732 
1733     /* Inactive L1 tables may point to active L2 tables - therefore it is
1734      * necessary to flush the L2 table cache before trying to access the L2
1735      * tables pointed to by inactive L1 entries (else we might try to expand
1736      * zero clusters that have already been expanded); furthermore, it is also
1737      * necessary to empty the L2 table cache, since it may contain tables which
1738      * are now going to be modified directly on disk, bypassing the cache.
1739      * qcow2_cache_empty() does both for us. */
1740     ret = qcow2_cache_empty(bs, s->l2_table_cache);
1741     if (ret < 0) {
1742         goto fail;
1743     }
1744 
1745     for (i = 0; i < s->nb_snapshots; i++) {
1746         int l1_sectors = (s->snapshots[i].l1_size * sizeof(uint64_t) +
1747                 BDRV_SECTOR_SIZE - 1) / BDRV_SECTOR_SIZE;
1748 
1749         l1_table = g_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE);
1750 
1751         ret = bdrv_read(bs->file, s->snapshots[i].l1_table_offset /
1752                 BDRV_SECTOR_SIZE, (void *)l1_table, l1_sectors);
1753         if (ret < 0) {
1754             goto fail;
1755         }
1756 
1757         for (j = 0; j < s->snapshots[i].l1_size; j++) {
1758             be64_to_cpus(&l1_table[j]);
1759         }
1760 
1761         ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size,
1762                                          &expanded_clusters, &nb_clusters);
1763         if (ret < 0) {
1764             goto fail;
1765         }
1766     }
1767 
1768     ret = 0;
1769 
1770 fail:
1771     g_free(expanded_clusters);
1772     g_free(l1_table);
1773     return ret;
1774 }
1775