xref: /qemu/block/qcow2-cluster.c (revision 4f2d31fb)
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     BDRVQcow2State *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     /* Do a sanity check on min_size before trying to calculate new_l1_size
46      * (this prevents overflows during the while loop for the calculation of
47      * new_l1_size) */
48     if (min_size > INT_MAX / sizeof(uint64_t)) {
49         return -EFBIG;
50     }
51 
52     if (exact_size) {
53         new_l1_size = min_size;
54     } else {
55         /* Bump size up to reduce the number of times we have to grow */
56         new_l1_size = s->l1_size;
57         if (new_l1_size == 0) {
58             new_l1_size = 1;
59         }
60         while (min_size > new_l1_size) {
61             new_l1_size = (new_l1_size * 3 + 1) / 2;
62         }
63     }
64 
65     if (new_l1_size > INT_MAX / sizeof(uint64_t)) {
66         return -EFBIG;
67     }
68 
69 #ifdef DEBUG_ALLOC2
70     fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n",
71             s->l1_size, new_l1_size);
72 #endif
73 
74     new_l1_size2 = sizeof(uint64_t) * new_l1_size;
75     new_l1_table = qemu_try_blockalign(bs->file->bs,
76                                        align_offset(new_l1_size2, 512));
77     if (new_l1_table == NULL) {
78         return -ENOMEM;
79     }
80     memset(new_l1_table, 0, align_offset(new_l1_size2, 512));
81 
82     memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
83 
84     /* write new table (align to cluster) */
85     BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE);
86     new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
87     if (new_l1_table_offset < 0) {
88         qemu_vfree(new_l1_table);
89         return new_l1_table_offset;
90     }
91 
92     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
93     if (ret < 0) {
94         goto fail;
95     }
96 
97     /* the L1 position has not yet been updated, so these clusters must
98      * indeed be completely free */
99     ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset,
100                                         new_l1_size2);
101     if (ret < 0) {
102         goto fail;
103     }
104 
105     BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE);
106     for(i = 0; i < s->l1_size; i++)
107         new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
108     ret = bdrv_pwrite_sync(bs->file->bs, new_l1_table_offset,
109                            new_l1_table, new_l1_size2);
110     if (ret < 0)
111         goto fail;
112     for(i = 0; i < s->l1_size; i++)
113         new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
114 
115     /* set new table */
116     BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE);
117     cpu_to_be32w((uint32_t*)data, new_l1_size);
118     stq_be_p(data + 4, new_l1_table_offset);
119     ret = bdrv_pwrite_sync(bs->file->bs, offsetof(QCowHeader, l1_size),
120                            data, sizeof(data));
121     if (ret < 0) {
122         goto fail;
123     }
124     qemu_vfree(s->l1_table);
125     old_l1_table_offset = s->l1_table_offset;
126     s->l1_table_offset = new_l1_table_offset;
127     s->l1_table = new_l1_table;
128     old_l1_size = s->l1_size;
129     s->l1_size = new_l1_size;
130     qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * sizeof(uint64_t),
131                         QCOW2_DISCARD_OTHER);
132     return 0;
133  fail:
134     qemu_vfree(new_l1_table);
135     qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2,
136                         QCOW2_DISCARD_OTHER);
137     return ret;
138 }
139 
140 /*
141  * l2_load
142  *
143  * Loads a L2 table into memory. If the table is in the cache, the cache
144  * is used; otherwise the L2 table is loaded from the image file.
145  *
146  * Returns a pointer to the L2 table on success, or NULL if the read from
147  * the image file failed.
148  */
149 
150 static int l2_load(BlockDriverState *bs, uint64_t l2_offset,
151     uint64_t **l2_table)
152 {
153     BDRVQcow2State *s = bs->opaque;
154     int ret;
155 
156     ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, (void**) l2_table);
157 
158     return ret;
159 }
160 
161 /*
162  * Writes one sector of the L1 table to the disk (can't update single entries
163  * and we really don't want bdrv_pread to perform a read-modify-write)
164  */
165 #define L1_ENTRIES_PER_SECTOR (512 / 8)
166 int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index)
167 {
168     BDRVQcow2State *s = bs->opaque;
169     uint64_t buf[L1_ENTRIES_PER_SECTOR] = { 0 };
170     int l1_start_index;
171     int i, ret;
172 
173     l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
174     for (i = 0; i < L1_ENTRIES_PER_SECTOR && l1_start_index + i < s->l1_size;
175          i++)
176     {
177         buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
178     }
179 
180     ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1,
181             s->l1_table_offset + 8 * l1_start_index, sizeof(buf));
182     if (ret < 0) {
183         return ret;
184     }
185 
186     BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
187     ret = bdrv_pwrite_sync(bs->file->bs,
188                            s->l1_table_offset + 8 * l1_start_index,
189                            buf, sizeof(buf));
190     if (ret < 0) {
191         return ret;
192     }
193 
194     return 0;
195 }
196 
197 /*
198  * l2_allocate
199  *
200  * Allocate a new l2 entry in the file. If l1_index points to an already
201  * used entry in the L2 table (i.e. we are doing a copy on write for the L2
202  * table) copy the contents of the old L2 table into the newly allocated one.
203  * Otherwise the new table is initialized with zeros.
204  *
205  */
206 
207 static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
208 {
209     BDRVQcow2State *s = bs->opaque;
210     uint64_t old_l2_offset;
211     uint64_t *l2_table = NULL;
212     int64_t l2_offset;
213     int ret;
214 
215     old_l2_offset = s->l1_table[l1_index];
216 
217     trace_qcow2_l2_allocate(bs, l1_index);
218 
219     /* allocate a new l2 entry */
220 
221     l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
222     if (l2_offset < 0) {
223         ret = l2_offset;
224         goto fail;
225     }
226 
227     ret = qcow2_cache_flush(bs, s->refcount_block_cache);
228     if (ret < 0) {
229         goto fail;
230     }
231 
232     /* allocate a new entry in the l2 cache */
233 
234     trace_qcow2_l2_allocate_get_empty(bs, l1_index);
235     ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table);
236     if (ret < 0) {
237         goto fail;
238     }
239 
240     l2_table = *table;
241 
242     if ((old_l2_offset & L1E_OFFSET_MASK) == 0) {
243         /* if there was no old l2 table, clear the new table */
244         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
245     } else {
246         uint64_t* old_table;
247 
248         /* if there was an old l2 table, read it from the disk */
249         BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ);
250         ret = qcow2_cache_get(bs, s->l2_table_cache,
251             old_l2_offset & L1E_OFFSET_MASK,
252             (void**) &old_table);
253         if (ret < 0) {
254             goto fail;
255         }
256 
257         memcpy(l2_table, old_table, s->cluster_size);
258 
259         qcow2_cache_put(bs, s->l2_table_cache, (void **) &old_table);
260     }
261 
262     /* write the l2 table to the file */
263     BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
264 
265     trace_qcow2_l2_allocate_write_l2(bs, l1_index);
266     qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
267     ret = qcow2_cache_flush(bs, s->l2_table_cache);
268     if (ret < 0) {
269         goto fail;
270     }
271 
272     /* update the L1 entry */
273     trace_qcow2_l2_allocate_write_l1(bs, l1_index);
274     s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
275     ret = qcow2_write_l1_entry(bs, l1_index);
276     if (ret < 0) {
277         goto fail;
278     }
279 
280     *table = l2_table;
281     trace_qcow2_l2_allocate_done(bs, l1_index, 0);
282     return 0;
283 
284 fail:
285     trace_qcow2_l2_allocate_done(bs, l1_index, ret);
286     if (l2_table != NULL) {
287         qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
288     }
289     s->l1_table[l1_index] = old_l2_offset;
290     if (l2_offset > 0) {
291         qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
292                             QCOW2_DISCARD_ALWAYS);
293     }
294     return ret;
295 }
296 
297 /*
298  * Checks how many clusters in a given L2 table are contiguous in the image
299  * file. As soon as one of the flags in the bitmask stop_flags changes compared
300  * to the first cluster, the search is stopped and the cluster is not counted
301  * as contiguous. (This allows it, for example, to stop at the first compressed
302  * cluster which may require a different handling)
303  */
304 static int count_contiguous_clusters(int nb_clusters, int cluster_size,
305         uint64_t *l2_table, uint64_t stop_flags)
306 {
307     int i;
308     uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED;
309     uint64_t first_entry = be64_to_cpu(l2_table[0]);
310     uint64_t offset = first_entry & mask;
311 
312     if (!offset)
313         return 0;
314 
315     assert(qcow2_get_cluster_type(first_entry) == QCOW2_CLUSTER_NORMAL);
316 
317     for (i = 0; i < nb_clusters; i++) {
318         uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask;
319         if (offset + (uint64_t) i * cluster_size != l2_entry) {
320             break;
321         }
322     }
323 
324 	return i;
325 }
326 
327 static int count_contiguous_clusters_by_type(int nb_clusters,
328                                              uint64_t *l2_table,
329                                              int wanted_type)
330 {
331     int i;
332 
333     for (i = 0; i < nb_clusters; i++) {
334         int type = qcow2_get_cluster_type(be64_to_cpu(l2_table[i]));
335 
336         if (type != wanted_type) {
337             break;
338         }
339     }
340 
341     return i;
342 }
343 
344 /* The crypt function is compatible with the linux cryptoloop
345    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
346    supported */
347 int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num,
348                           uint8_t *out_buf, const uint8_t *in_buf,
349                           int nb_sectors, bool enc,
350                           Error **errp)
351 {
352     union {
353         uint64_t ll[2];
354         uint8_t b[16];
355     } ivec;
356     int i;
357     int ret;
358 
359     for(i = 0; i < nb_sectors; i++) {
360         ivec.ll[0] = cpu_to_le64(sector_num);
361         ivec.ll[1] = 0;
362         if (qcrypto_cipher_setiv(s->cipher,
363                                  ivec.b, G_N_ELEMENTS(ivec.b),
364                                  errp) < 0) {
365             return -1;
366         }
367         if (enc) {
368             ret = qcrypto_cipher_encrypt(s->cipher,
369                                          in_buf,
370                                          out_buf,
371                                          512,
372                                          errp);
373         } else {
374             ret = qcrypto_cipher_decrypt(s->cipher,
375                                          in_buf,
376                                          out_buf,
377                                          512,
378                                          errp);
379         }
380         if (ret < 0) {
381             return -1;
382         }
383         sector_num++;
384         in_buf += 512;
385         out_buf += 512;
386     }
387     return 0;
388 }
389 
390 static int coroutine_fn copy_sectors(BlockDriverState *bs,
391                                      uint64_t start_sect,
392                                      uint64_t cluster_offset,
393                                      int n_start, int n_end)
394 {
395     BDRVQcow2State *s = bs->opaque;
396     QEMUIOVector qiov;
397     struct iovec iov;
398     int n, ret;
399 
400     n = n_end - n_start;
401     if (n <= 0) {
402         return 0;
403     }
404 
405     iov.iov_len = n * BDRV_SECTOR_SIZE;
406     iov.iov_base = qemu_try_blockalign(bs, iov.iov_len);
407     if (iov.iov_base == NULL) {
408         return -ENOMEM;
409     }
410 
411     qemu_iovec_init_external(&qiov, &iov, 1);
412 
413     BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
414 
415     if (!bs->drv) {
416         ret = -ENOMEDIUM;
417         goto out;
418     }
419 
420     /* Call .bdrv_co_readv() directly instead of using the public block-layer
421      * interface.  This avoids double I/O throttling and request tracking,
422      * which can lead to deadlock when block layer copy-on-read is enabled.
423      */
424     ret = bs->drv->bdrv_co_readv(bs, start_sect + n_start, n, &qiov);
425     if (ret < 0) {
426         goto out;
427     }
428 
429     if (bs->encrypted) {
430         Error *err = NULL;
431         assert(s->cipher);
432         if (qcow2_encrypt_sectors(s, start_sect + n_start,
433                                   iov.iov_base, iov.iov_base, n,
434                                   true, &err) < 0) {
435             ret = -EIO;
436             error_free(err);
437             goto out;
438         }
439     }
440 
441     ret = qcow2_pre_write_overlap_check(bs, 0,
442             cluster_offset + n_start * BDRV_SECTOR_SIZE, n * BDRV_SECTOR_SIZE);
443     if (ret < 0) {
444         goto out;
445     }
446 
447     BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
448     ret = bdrv_co_writev(bs->file->bs, (cluster_offset >> 9) + n_start, n,
449                          &qiov);
450     if (ret < 0) {
451         goto out;
452     }
453 
454     ret = 0;
455 out:
456     qemu_vfree(iov.iov_base);
457     return ret;
458 }
459 
460 
461 /*
462  * get_cluster_offset
463  *
464  * For a given offset of the disk image, find the cluster offset in
465  * qcow2 file. The offset is stored in *cluster_offset.
466  *
467  * on entry, *num is the number of contiguous sectors we'd like to
468  * access following offset.
469  *
470  * on exit, *num is the number of contiguous sectors we can read.
471  *
472  * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error
473  * cases.
474  */
475 int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
476     int *num, uint64_t *cluster_offset)
477 {
478     BDRVQcow2State *s = bs->opaque;
479     unsigned int l2_index;
480     uint64_t l1_index, l2_offset, *l2_table;
481     int l1_bits, c;
482     unsigned int index_in_cluster, nb_clusters;
483     uint64_t nb_available, nb_needed;
484     int ret;
485 
486     index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
487     nb_needed = *num + index_in_cluster;
488 
489     l1_bits = s->l2_bits + s->cluster_bits;
490 
491     /* compute how many bytes there are between the offset and
492      * the end of the l1 entry
493      */
494 
495     nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
496 
497     /* compute the number of available sectors */
498 
499     nb_available = (nb_available >> 9) + index_in_cluster;
500 
501     if (nb_needed > nb_available) {
502         nb_needed = nb_available;
503     }
504     assert(nb_needed <= INT_MAX);
505 
506     *cluster_offset = 0;
507 
508     /* seek to the l2 offset in the l1 table */
509 
510     l1_index = offset >> l1_bits;
511     if (l1_index >= s->l1_size) {
512         ret = QCOW2_CLUSTER_UNALLOCATED;
513         goto out;
514     }
515 
516     l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
517     if (!l2_offset) {
518         ret = QCOW2_CLUSTER_UNALLOCATED;
519         goto out;
520     }
521 
522     if (offset_into_cluster(s, l2_offset)) {
523         qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64
524                                 " unaligned (L1 index: %#" PRIx64 ")",
525                                 l2_offset, l1_index);
526         return -EIO;
527     }
528 
529     /* load the l2 table in memory */
530 
531     ret = l2_load(bs, l2_offset, &l2_table);
532     if (ret < 0) {
533         return ret;
534     }
535 
536     /* find the cluster offset for the given disk offset */
537 
538     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
539     *cluster_offset = be64_to_cpu(l2_table[l2_index]);
540 
541     /* nb_needed <= INT_MAX, thus nb_clusters <= INT_MAX, too */
542     nb_clusters = size_to_clusters(s, nb_needed << 9);
543 
544     ret = qcow2_get_cluster_type(*cluster_offset);
545     switch (ret) {
546     case QCOW2_CLUSTER_COMPRESSED:
547         /* Compressed clusters can only be processed one by one */
548         c = 1;
549         *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK;
550         break;
551     case QCOW2_CLUSTER_ZERO:
552         if (s->qcow_version < 3) {
553             qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found"
554                                     " in pre-v3 image (L2 offset: %#" PRIx64
555                                     ", L2 index: %#x)", l2_offset, l2_index);
556             ret = -EIO;
557             goto fail;
558         }
559         c = count_contiguous_clusters_by_type(nb_clusters, &l2_table[l2_index],
560                                               QCOW2_CLUSTER_ZERO);
561         *cluster_offset = 0;
562         break;
563     case QCOW2_CLUSTER_UNALLOCATED:
564         /* how many empty clusters ? */
565         c = count_contiguous_clusters_by_type(nb_clusters, &l2_table[l2_index],
566                                               QCOW2_CLUSTER_UNALLOCATED);
567         *cluster_offset = 0;
568         break;
569     case QCOW2_CLUSTER_NORMAL:
570         /* how many allocated clusters ? */
571         c = count_contiguous_clusters(nb_clusters, s->cluster_size,
572                 &l2_table[l2_index], QCOW_OFLAG_ZERO);
573         *cluster_offset &= L2E_OFFSET_MASK;
574         if (offset_into_cluster(s, *cluster_offset)) {
575             qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset %#"
576                                     PRIx64 " unaligned (L2 offset: %#" PRIx64
577                                     ", L2 index: %#x)", *cluster_offset,
578                                     l2_offset, l2_index);
579             ret = -EIO;
580             goto fail;
581         }
582         break;
583     default:
584         abort();
585     }
586 
587     qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
588 
589     nb_available = (c * s->cluster_sectors);
590 
591 out:
592     if (nb_available > nb_needed)
593         nb_available = nb_needed;
594 
595     *num = nb_available - index_in_cluster;
596 
597     return ret;
598 
599 fail:
600     qcow2_cache_put(bs, s->l2_table_cache, (void **)&l2_table);
601     return ret;
602 }
603 
604 /*
605  * get_cluster_table
606  *
607  * for a given disk offset, load (and allocate if needed)
608  * the l2 table.
609  *
610  * the l2 table offset in the qcow2 file and the cluster index
611  * in the l2 table are given to the caller.
612  *
613  * Returns 0 on success, -errno in failure case
614  */
615 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
616                              uint64_t **new_l2_table,
617                              int *new_l2_index)
618 {
619     BDRVQcow2State *s = bs->opaque;
620     unsigned int l2_index;
621     uint64_t l1_index, l2_offset;
622     uint64_t *l2_table = NULL;
623     int ret;
624 
625     /* seek to the l2 offset in the l1 table */
626 
627     l1_index = offset >> (s->l2_bits + s->cluster_bits);
628     if (l1_index >= s->l1_size) {
629         ret = qcow2_grow_l1_table(bs, l1_index + 1, false);
630         if (ret < 0) {
631             return ret;
632         }
633     }
634 
635     assert(l1_index < s->l1_size);
636     l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
637     if (offset_into_cluster(s, l2_offset)) {
638         qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64
639                                 " unaligned (L1 index: %#" PRIx64 ")",
640                                 l2_offset, l1_index);
641         return -EIO;
642     }
643 
644     /* seek the l2 table of the given l2 offset */
645 
646     if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) {
647         /* load the l2 table in memory */
648         ret = l2_load(bs, l2_offset, &l2_table);
649         if (ret < 0) {
650             return ret;
651         }
652     } else {
653         /* First allocate a new L2 table (and do COW if needed) */
654         ret = l2_allocate(bs, l1_index, &l2_table);
655         if (ret < 0) {
656             return ret;
657         }
658 
659         /* Then decrease the refcount of the old table */
660         if (l2_offset) {
661             qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
662                                 QCOW2_DISCARD_OTHER);
663         }
664     }
665 
666     /* find the cluster offset for the given disk offset */
667 
668     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
669 
670     *new_l2_table = l2_table;
671     *new_l2_index = l2_index;
672 
673     return 0;
674 }
675 
676 /*
677  * alloc_compressed_cluster_offset
678  *
679  * For a given offset of the disk image, return cluster offset in
680  * qcow2 file.
681  *
682  * If the offset is not found, allocate a new compressed cluster.
683  *
684  * Return the cluster offset if successful,
685  * Return 0, otherwise.
686  *
687  */
688 
689 uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
690                                                uint64_t offset,
691                                                int compressed_size)
692 {
693     BDRVQcow2State *s = bs->opaque;
694     int l2_index, ret;
695     uint64_t *l2_table;
696     int64_t cluster_offset;
697     int nb_csectors;
698 
699     ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
700     if (ret < 0) {
701         return 0;
702     }
703 
704     /* Compression can't overwrite anything. Fail if the cluster was already
705      * allocated. */
706     cluster_offset = be64_to_cpu(l2_table[l2_index]);
707     if (cluster_offset & L2E_OFFSET_MASK) {
708         qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
709         return 0;
710     }
711 
712     cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
713     if (cluster_offset < 0) {
714         qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
715         return 0;
716     }
717 
718     nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
719                   (cluster_offset >> 9);
720 
721     cluster_offset |= QCOW_OFLAG_COMPRESSED |
722                       ((uint64_t)nb_csectors << s->csize_shift);
723 
724     /* update L2 table */
725 
726     /* compressed clusters never have the copied flag */
727 
728     BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
729     qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
730     l2_table[l2_index] = cpu_to_be64(cluster_offset);
731     qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
732 
733     return cluster_offset;
734 }
735 
736 static int perform_cow(BlockDriverState *bs, QCowL2Meta *m, Qcow2COWRegion *r)
737 {
738     BDRVQcow2State *s = bs->opaque;
739     int ret;
740 
741     if (r->nb_sectors == 0) {
742         return 0;
743     }
744 
745     qemu_co_mutex_unlock(&s->lock);
746     ret = copy_sectors(bs, m->offset / BDRV_SECTOR_SIZE, m->alloc_offset,
747                        r->offset / BDRV_SECTOR_SIZE,
748                        r->offset / BDRV_SECTOR_SIZE + r->nb_sectors);
749     qemu_co_mutex_lock(&s->lock);
750 
751     if (ret < 0) {
752         return ret;
753     }
754 
755     /*
756      * Before we update the L2 table to actually point to the new cluster, we
757      * need to be sure that the refcounts have been increased and COW was
758      * handled.
759      */
760     qcow2_cache_depends_on_flush(s->l2_table_cache);
761 
762     return 0;
763 }
764 
765 int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
766 {
767     BDRVQcow2State *s = bs->opaque;
768     int i, j = 0, l2_index, ret;
769     uint64_t *old_cluster, *l2_table;
770     uint64_t cluster_offset = m->alloc_offset;
771 
772     trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters);
773     assert(m->nb_clusters > 0);
774 
775     old_cluster = g_try_new(uint64_t, m->nb_clusters);
776     if (old_cluster == NULL) {
777         ret = -ENOMEM;
778         goto err;
779     }
780 
781     /* copy content of unmodified sectors */
782     ret = perform_cow(bs, m, &m->cow_start);
783     if (ret < 0) {
784         goto err;
785     }
786 
787     ret = perform_cow(bs, m, &m->cow_end);
788     if (ret < 0) {
789         goto err;
790     }
791 
792     /* Update L2 table. */
793     if (s->use_lazy_refcounts) {
794         qcow2_mark_dirty(bs);
795     }
796     if (qcow2_need_accurate_refcounts(s)) {
797         qcow2_cache_set_dependency(bs, s->l2_table_cache,
798                                    s->refcount_block_cache);
799     }
800 
801     ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index);
802     if (ret < 0) {
803         goto err;
804     }
805     qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
806 
807     assert(l2_index + m->nb_clusters <= s->l2_size);
808     for (i = 0; i < m->nb_clusters; i++) {
809         /* if two concurrent writes happen to the same unallocated cluster
810 	 * each write allocates separate cluster and writes data concurrently.
811 	 * The first one to complete updates l2 table with pointer to its
812 	 * cluster the second one has to do RMW (which is done above by
813 	 * copy_sectors()), update l2 table with its cluster pointer and free
814 	 * old cluster. This is what this loop does */
815         if(l2_table[l2_index + i] != 0)
816             old_cluster[j++] = l2_table[l2_index + i];
817 
818         l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
819                     (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
820      }
821 
822 
823     qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
824 
825     /*
826      * If this was a COW, we need to decrease the refcount of the old cluster.
827      *
828      * Don't discard clusters that reach a refcount of 0 (e.g. compressed
829      * clusters), the next write will reuse them anyway.
830      */
831     if (j != 0) {
832         for (i = 0; i < j; i++) {
833             qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1,
834                                     QCOW2_DISCARD_NEVER);
835         }
836     }
837 
838     ret = 0;
839 err:
840     g_free(old_cluster);
841     return ret;
842  }
843 
844 /*
845  * Returns the number of contiguous clusters that can be used for an allocating
846  * write, but require COW to be performed (this includes yet unallocated space,
847  * which must copy from the backing file)
848  */
849 static int count_cow_clusters(BDRVQcow2State *s, int nb_clusters,
850     uint64_t *l2_table, int l2_index)
851 {
852     int i;
853 
854     for (i = 0; i < nb_clusters; i++) {
855         uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]);
856         int cluster_type = qcow2_get_cluster_type(l2_entry);
857 
858         switch(cluster_type) {
859         case QCOW2_CLUSTER_NORMAL:
860             if (l2_entry & QCOW_OFLAG_COPIED) {
861                 goto out;
862             }
863             break;
864         case QCOW2_CLUSTER_UNALLOCATED:
865         case QCOW2_CLUSTER_COMPRESSED:
866         case QCOW2_CLUSTER_ZERO:
867             break;
868         default:
869             abort();
870         }
871     }
872 
873 out:
874     assert(i <= nb_clusters);
875     return i;
876 }
877 
878 /*
879  * Check if there already is an AIO write request in flight which allocates
880  * the same cluster. In this case we need to wait until the previous
881  * request has completed and updated the L2 table accordingly.
882  *
883  * Returns:
884  *   0       if there was no dependency. *cur_bytes indicates the number of
885  *           bytes from guest_offset that can be read before the next
886  *           dependency must be processed (or the request is complete)
887  *
888  *   -EAGAIN if we had to wait for another request, previously gathered
889  *           information on cluster allocation may be invalid now. The caller
890  *           must start over anyway, so consider *cur_bytes undefined.
891  */
892 static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset,
893     uint64_t *cur_bytes, QCowL2Meta **m)
894 {
895     BDRVQcow2State *s = bs->opaque;
896     QCowL2Meta *old_alloc;
897     uint64_t bytes = *cur_bytes;
898 
899     QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
900 
901         uint64_t start = guest_offset;
902         uint64_t end = start + bytes;
903         uint64_t old_start = l2meta_cow_start(old_alloc);
904         uint64_t old_end = l2meta_cow_end(old_alloc);
905 
906         if (end <= old_start || start >= old_end) {
907             /* No intersection */
908         } else {
909             if (start < old_start) {
910                 /* Stop at the start of a running allocation */
911                 bytes = old_start - start;
912             } else {
913                 bytes = 0;
914             }
915 
916             /* Stop if already an l2meta exists. After yielding, it wouldn't
917              * be valid any more, so we'd have to clean up the old L2Metas
918              * and deal with requests depending on them before starting to
919              * gather new ones. Not worth the trouble. */
920             if (bytes == 0 && *m) {
921                 *cur_bytes = 0;
922                 return 0;
923             }
924 
925             if (bytes == 0) {
926                 /* Wait for the dependency to complete. We need to recheck
927                  * the free/allocated clusters when we continue. */
928                 qemu_co_mutex_unlock(&s->lock);
929                 qemu_co_queue_wait(&old_alloc->dependent_requests);
930                 qemu_co_mutex_lock(&s->lock);
931                 return -EAGAIN;
932             }
933         }
934     }
935 
936     /* Make sure that existing clusters and new allocations are only used up to
937      * the next dependency if we shortened the request above */
938     *cur_bytes = bytes;
939 
940     return 0;
941 }
942 
943 /*
944  * Checks how many already allocated clusters that don't require a copy on
945  * write there are at the given guest_offset (up to *bytes). If
946  * *host_offset is not zero, only physically contiguous clusters beginning at
947  * this host offset are counted.
948  *
949  * Note that guest_offset may not be cluster aligned. In this case, the
950  * returned *host_offset points to exact byte referenced by guest_offset and
951  * therefore isn't cluster aligned as well.
952  *
953  * Returns:
954  *   0:     if no allocated clusters are available at the given offset.
955  *          *bytes is normally unchanged. It is set to 0 if the cluster
956  *          is allocated and doesn't need COW, but doesn't have the right
957  *          physical offset.
958  *
959  *   1:     if allocated clusters that don't require a COW are available at
960  *          the requested offset. *bytes may have decreased and describes
961  *          the length of the area that can be written to.
962  *
963  *  -errno: in error cases
964  */
965 static int handle_copied(BlockDriverState *bs, uint64_t guest_offset,
966     uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
967 {
968     BDRVQcow2State *s = bs->opaque;
969     int l2_index;
970     uint64_t cluster_offset;
971     uint64_t *l2_table;
972     uint64_t nb_clusters;
973     unsigned int keep_clusters;
974     int ret;
975 
976     trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset,
977                               *bytes);
978 
979     assert(*host_offset == 0 ||    offset_into_cluster(s, guest_offset)
980                                 == offset_into_cluster(s, *host_offset));
981 
982     /*
983      * Calculate the number of clusters to look for. We stop at L2 table
984      * boundaries to keep things simple.
985      */
986     nb_clusters =
987         size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
988 
989     l2_index = offset_to_l2_index(s, guest_offset);
990     nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
991     assert(nb_clusters <= INT_MAX);
992 
993     /* Find L2 entry for the first involved cluster */
994     ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index);
995     if (ret < 0) {
996         return ret;
997     }
998 
999     cluster_offset = be64_to_cpu(l2_table[l2_index]);
1000 
1001     /* Check how many clusters are already allocated and don't need COW */
1002     if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL
1003         && (cluster_offset & QCOW_OFLAG_COPIED))
1004     {
1005         /* If a specific host_offset is required, check it */
1006         bool offset_matches =
1007             (cluster_offset & L2E_OFFSET_MASK) == *host_offset;
1008 
1009         if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) {
1010             qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset "
1011                                     "%#llx unaligned (guest offset: %#" PRIx64
1012                                     ")", cluster_offset & L2E_OFFSET_MASK,
1013                                     guest_offset);
1014             ret = -EIO;
1015             goto out;
1016         }
1017 
1018         if (*host_offset != 0 && !offset_matches) {
1019             *bytes = 0;
1020             ret = 0;
1021             goto out;
1022         }
1023 
1024         /* We keep all QCOW_OFLAG_COPIED clusters */
1025         keep_clusters =
1026             count_contiguous_clusters(nb_clusters, s->cluster_size,
1027                                       &l2_table[l2_index],
1028                                       QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO);
1029         assert(keep_clusters <= nb_clusters);
1030 
1031         *bytes = MIN(*bytes,
1032                  keep_clusters * s->cluster_size
1033                  - offset_into_cluster(s, guest_offset));
1034 
1035         ret = 1;
1036     } else {
1037         ret = 0;
1038     }
1039 
1040     /* Cleanup */
1041 out:
1042     qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1043 
1044     /* Only return a host offset if we actually made progress. Otherwise we
1045      * would make requirements for handle_alloc() that it can't fulfill */
1046     if (ret > 0) {
1047         *host_offset = (cluster_offset & L2E_OFFSET_MASK)
1048                      + offset_into_cluster(s, guest_offset);
1049     }
1050 
1051     return ret;
1052 }
1053 
1054 /*
1055  * Allocates new clusters for the given guest_offset.
1056  *
1057  * At most *nb_clusters are allocated, and on return *nb_clusters is updated to
1058  * contain the number of clusters that have been allocated and are contiguous
1059  * in the image file.
1060  *
1061  * If *host_offset is non-zero, it specifies the offset in the image file at
1062  * which the new clusters must start. *nb_clusters can be 0 on return in this
1063  * case if the cluster at host_offset is already in use. If *host_offset is
1064  * zero, the clusters can be allocated anywhere in the image file.
1065  *
1066  * *host_offset is updated to contain the offset into the image file at which
1067  * the first allocated cluster starts.
1068  *
1069  * Return 0 on success and -errno in error cases. -EAGAIN means that the
1070  * function has been waiting for another request and the allocation must be
1071  * restarted, but the whole request should not be failed.
1072  */
1073 static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset,
1074                                    uint64_t *host_offset, uint64_t *nb_clusters)
1075 {
1076     BDRVQcow2State *s = bs->opaque;
1077 
1078     trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset,
1079                                          *host_offset, *nb_clusters);
1080 
1081     /* Allocate new clusters */
1082     trace_qcow2_cluster_alloc_phys(qemu_coroutine_self());
1083     if (*host_offset == 0) {
1084         int64_t cluster_offset =
1085             qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size);
1086         if (cluster_offset < 0) {
1087             return cluster_offset;
1088         }
1089         *host_offset = cluster_offset;
1090         return 0;
1091     } else {
1092         int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters);
1093         if (ret < 0) {
1094             return ret;
1095         }
1096         *nb_clusters = ret;
1097         return 0;
1098     }
1099 }
1100 
1101 /*
1102  * Allocates new clusters for an area that either is yet unallocated or needs a
1103  * copy on write. If *host_offset is non-zero, clusters are only allocated if
1104  * the new allocation can match the specified host offset.
1105  *
1106  * Note that guest_offset may not be cluster aligned. In this case, the
1107  * returned *host_offset points to exact byte referenced by guest_offset and
1108  * therefore isn't cluster aligned as well.
1109  *
1110  * Returns:
1111  *   0:     if no clusters could be allocated. *bytes is set to 0,
1112  *          *host_offset is left unchanged.
1113  *
1114  *   1:     if new clusters were allocated. *bytes may be decreased if the
1115  *          new allocation doesn't cover all of the requested area.
1116  *          *host_offset is updated to contain the host offset of the first
1117  *          newly allocated cluster.
1118  *
1119  *  -errno: in error cases
1120  */
1121 static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
1122     uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
1123 {
1124     BDRVQcow2State *s = bs->opaque;
1125     int l2_index;
1126     uint64_t *l2_table;
1127     uint64_t entry;
1128     uint64_t nb_clusters;
1129     int ret;
1130 
1131     uint64_t alloc_cluster_offset;
1132 
1133     trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset,
1134                              *bytes);
1135     assert(*bytes > 0);
1136 
1137     /*
1138      * Calculate the number of clusters to look for. We stop at L2 table
1139      * boundaries to keep things simple.
1140      */
1141     nb_clusters =
1142         size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
1143 
1144     l2_index = offset_to_l2_index(s, guest_offset);
1145     nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1146     assert(nb_clusters <= INT_MAX);
1147 
1148     /* Find L2 entry for the first involved cluster */
1149     ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index);
1150     if (ret < 0) {
1151         return ret;
1152     }
1153 
1154     entry = be64_to_cpu(l2_table[l2_index]);
1155 
1156     /* For the moment, overwrite compressed clusters one by one */
1157     if (entry & QCOW_OFLAG_COMPRESSED) {
1158         nb_clusters = 1;
1159     } else {
1160         nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index);
1161     }
1162 
1163     /* This function is only called when there were no non-COW clusters, so if
1164      * we can't find any unallocated or COW clusters either, something is
1165      * wrong with our code. */
1166     assert(nb_clusters > 0);
1167 
1168     qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1169 
1170     /* Allocate, if necessary at a given offset in the image file */
1171     alloc_cluster_offset = start_of_cluster(s, *host_offset);
1172     ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset,
1173                                   &nb_clusters);
1174     if (ret < 0) {
1175         goto fail;
1176     }
1177 
1178     /* Can't extend contiguous allocation */
1179     if (nb_clusters == 0) {
1180         *bytes = 0;
1181         return 0;
1182     }
1183 
1184     /* !*host_offset would overwrite the image header and is reserved for "no
1185      * host offset preferred". If 0 was a valid host offset, it'd trigger the
1186      * following overlap check; do that now to avoid having an invalid value in
1187      * *host_offset. */
1188     if (!alloc_cluster_offset) {
1189         ret = qcow2_pre_write_overlap_check(bs, 0, alloc_cluster_offset,
1190                                             nb_clusters * s->cluster_size);
1191         assert(ret < 0);
1192         goto fail;
1193     }
1194 
1195     /*
1196      * Save info needed for meta data update.
1197      *
1198      * requested_sectors: Number of sectors from the start of the first
1199      * newly allocated cluster to the end of the (possibly shortened
1200      * before) write request.
1201      *
1202      * avail_sectors: Number of sectors from the start of the first
1203      * newly allocated to the end of the last newly allocated cluster.
1204      *
1205      * nb_sectors: The number of sectors from the start of the first
1206      * newly allocated cluster to the end of the area that the write
1207      * request actually writes to (excluding COW at the end)
1208      */
1209     int requested_sectors =
1210         (*bytes + offset_into_cluster(s, guest_offset))
1211         >> BDRV_SECTOR_BITS;
1212     int avail_sectors = nb_clusters
1213                         << (s->cluster_bits - BDRV_SECTOR_BITS);
1214     int alloc_n_start = offset_into_cluster(s, guest_offset)
1215                         >> BDRV_SECTOR_BITS;
1216     int nb_sectors = MIN(requested_sectors, avail_sectors);
1217     QCowL2Meta *old_m = *m;
1218 
1219     *m = g_malloc0(sizeof(**m));
1220 
1221     **m = (QCowL2Meta) {
1222         .next           = old_m,
1223 
1224         .alloc_offset   = alloc_cluster_offset,
1225         .offset         = start_of_cluster(s, guest_offset),
1226         .nb_clusters    = nb_clusters,
1227         .nb_available   = nb_sectors,
1228 
1229         .cow_start = {
1230             .offset     = 0,
1231             .nb_sectors = alloc_n_start,
1232         },
1233         .cow_end = {
1234             .offset     = nb_sectors * BDRV_SECTOR_SIZE,
1235             .nb_sectors = avail_sectors - nb_sectors,
1236         },
1237     };
1238     qemu_co_queue_init(&(*m)->dependent_requests);
1239     QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
1240 
1241     *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset);
1242     *bytes = MIN(*bytes, (nb_sectors * BDRV_SECTOR_SIZE)
1243                          - offset_into_cluster(s, guest_offset));
1244     assert(*bytes != 0);
1245 
1246     return 1;
1247 
1248 fail:
1249     if (*m && (*m)->nb_clusters > 0) {
1250         QLIST_REMOVE(*m, next_in_flight);
1251     }
1252     return ret;
1253 }
1254 
1255 /*
1256  * alloc_cluster_offset
1257  *
1258  * For a given offset on the virtual disk, find the cluster offset in qcow2
1259  * file. If the offset is not found, allocate a new cluster.
1260  *
1261  * If the cluster was already allocated, m->nb_clusters is set to 0 and
1262  * other fields in m are meaningless.
1263  *
1264  * If the cluster is newly allocated, m->nb_clusters is set to the number of
1265  * contiguous clusters that have been allocated. In this case, the other
1266  * fields of m are valid and contain information about the first allocated
1267  * cluster.
1268  *
1269  * If the request conflicts with another write request in flight, the coroutine
1270  * is queued and will be reentered when the dependency has completed.
1271  *
1272  * Return 0 on success and -errno in error cases
1273  */
1274 int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
1275     int *num, uint64_t *host_offset, QCowL2Meta **m)
1276 {
1277     BDRVQcow2State *s = bs->opaque;
1278     uint64_t start, remaining;
1279     uint64_t cluster_offset;
1280     uint64_t cur_bytes;
1281     int ret;
1282 
1283     trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *num);
1284 
1285     assert((offset & ~BDRV_SECTOR_MASK) == 0);
1286 
1287 again:
1288     start = offset;
1289     remaining = (uint64_t)*num << BDRV_SECTOR_BITS;
1290     cluster_offset = 0;
1291     *host_offset = 0;
1292     cur_bytes = 0;
1293     *m = NULL;
1294 
1295     while (true) {
1296 
1297         if (!*host_offset) {
1298             *host_offset = start_of_cluster(s, cluster_offset);
1299         }
1300 
1301         assert(remaining >= cur_bytes);
1302 
1303         start           += cur_bytes;
1304         remaining       -= cur_bytes;
1305         cluster_offset  += cur_bytes;
1306 
1307         if (remaining == 0) {
1308             break;
1309         }
1310 
1311         cur_bytes = remaining;
1312 
1313         /*
1314          * Now start gathering as many contiguous clusters as possible:
1315          *
1316          * 1. Check for overlaps with in-flight allocations
1317          *
1318          *      a) Overlap not in the first cluster -> shorten this request and
1319          *         let the caller handle the rest in its next loop iteration.
1320          *
1321          *      b) Real overlaps of two requests. Yield and restart the search
1322          *         for contiguous clusters (the situation could have changed
1323          *         while we were sleeping)
1324          *
1325          *      c) TODO: Request starts in the same cluster as the in-flight
1326          *         allocation ends. Shorten the COW of the in-fight allocation,
1327          *         set cluster_offset to write to the same cluster and set up
1328          *         the right synchronisation between the in-flight request and
1329          *         the new one.
1330          */
1331         ret = handle_dependencies(bs, start, &cur_bytes, m);
1332         if (ret == -EAGAIN) {
1333             /* Currently handle_dependencies() doesn't yield if we already had
1334              * an allocation. If it did, we would have to clean up the L2Meta
1335              * structs before starting over. */
1336             assert(*m == NULL);
1337             goto again;
1338         } else if (ret < 0) {
1339             return ret;
1340         } else if (cur_bytes == 0) {
1341             break;
1342         } else {
1343             /* handle_dependencies() may have decreased cur_bytes (shortened
1344              * the allocations below) so that the next dependency is processed
1345              * correctly during the next loop iteration. */
1346         }
1347 
1348         /*
1349          * 2. Count contiguous COPIED clusters.
1350          */
1351         ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m);
1352         if (ret < 0) {
1353             return ret;
1354         } else if (ret) {
1355             continue;
1356         } else if (cur_bytes == 0) {
1357             break;
1358         }
1359 
1360         /*
1361          * 3. If the request still hasn't completed, allocate new clusters,
1362          *    considering any cluster_offset of steps 1c or 2.
1363          */
1364         ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m);
1365         if (ret < 0) {
1366             return ret;
1367         } else if (ret) {
1368             continue;
1369         } else {
1370             assert(cur_bytes == 0);
1371             break;
1372         }
1373     }
1374 
1375     *num -= remaining >> BDRV_SECTOR_BITS;
1376     assert(*num > 0);
1377     assert(*host_offset != 0);
1378 
1379     return 0;
1380 }
1381 
1382 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1383                              const uint8_t *buf, int buf_size)
1384 {
1385     z_stream strm1, *strm = &strm1;
1386     int ret, out_len;
1387 
1388     memset(strm, 0, sizeof(*strm));
1389 
1390     strm->next_in = (uint8_t *)buf;
1391     strm->avail_in = buf_size;
1392     strm->next_out = out_buf;
1393     strm->avail_out = out_buf_size;
1394 
1395     ret = inflateInit2(strm, -12);
1396     if (ret != Z_OK)
1397         return -1;
1398     ret = inflate(strm, Z_FINISH);
1399     out_len = strm->next_out - out_buf;
1400     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1401         out_len != out_buf_size) {
1402         inflateEnd(strm);
1403         return -1;
1404     }
1405     inflateEnd(strm);
1406     return 0;
1407 }
1408 
1409 int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
1410 {
1411     BDRVQcow2State *s = bs->opaque;
1412     int ret, csize, nb_csectors, sector_offset;
1413     uint64_t coffset;
1414 
1415     coffset = cluster_offset & s->cluster_offset_mask;
1416     if (s->cluster_cache_offset != coffset) {
1417         nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1418         sector_offset = coffset & 511;
1419         csize = nb_csectors * 512 - sector_offset;
1420         BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
1421         ret = bdrv_read(bs->file->bs, coffset >> 9, s->cluster_data,
1422                         nb_csectors);
1423         if (ret < 0) {
1424             return ret;
1425         }
1426         if (decompress_buffer(s->cluster_cache, s->cluster_size,
1427                               s->cluster_data + sector_offset, csize) < 0) {
1428             return -EIO;
1429         }
1430         s->cluster_cache_offset = coffset;
1431     }
1432     return 0;
1433 }
1434 
1435 /*
1436  * This discards as many clusters of nb_clusters as possible at once (i.e.
1437  * all clusters in the same L2 table) and returns the number of discarded
1438  * clusters.
1439  */
1440 static int discard_single_l2(BlockDriverState *bs, uint64_t offset,
1441                              uint64_t nb_clusters, enum qcow2_discard_type type,
1442                              bool full_discard)
1443 {
1444     BDRVQcow2State *s = bs->opaque;
1445     uint64_t *l2_table;
1446     int l2_index;
1447     int ret;
1448     int i;
1449 
1450     ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1451     if (ret < 0) {
1452         return ret;
1453     }
1454 
1455     /* Limit nb_clusters to one L2 table */
1456     nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1457     assert(nb_clusters <= INT_MAX);
1458 
1459     for (i = 0; i < nb_clusters; i++) {
1460         uint64_t old_l2_entry;
1461 
1462         old_l2_entry = be64_to_cpu(l2_table[l2_index + i]);
1463 
1464         /*
1465          * If full_discard is false, make sure that a discarded area reads back
1466          * as zeroes for v3 images (we cannot do it for v2 without actually
1467          * writing a zero-filled buffer). We can skip the operation if the
1468          * cluster is already marked as zero, or if it's unallocated and we
1469          * don't have a backing file.
1470          *
1471          * TODO We might want to use bdrv_get_block_status(bs) here, but we're
1472          * holding s->lock, so that doesn't work today.
1473          *
1474          * If full_discard is true, the sector should not read back as zeroes,
1475          * but rather fall through to the backing file.
1476          */
1477         switch (qcow2_get_cluster_type(old_l2_entry)) {
1478             case QCOW2_CLUSTER_UNALLOCATED:
1479                 if (full_discard || !bs->backing) {
1480                     continue;
1481                 }
1482                 break;
1483 
1484             case QCOW2_CLUSTER_ZERO:
1485                 if (!full_discard) {
1486                     continue;
1487                 }
1488                 break;
1489 
1490             case QCOW2_CLUSTER_NORMAL:
1491             case QCOW2_CLUSTER_COMPRESSED:
1492                 break;
1493 
1494             default:
1495                 abort();
1496         }
1497 
1498         /* First remove L2 entries */
1499         qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
1500         if (!full_discard && s->qcow_version >= 3) {
1501             l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
1502         } else {
1503             l2_table[l2_index + i] = cpu_to_be64(0);
1504         }
1505 
1506         /* Then decrease the refcount */
1507         qcow2_free_any_clusters(bs, old_l2_entry, 1, type);
1508     }
1509 
1510     qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1511 
1512     return nb_clusters;
1513 }
1514 
1515 int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
1516     int nb_sectors, enum qcow2_discard_type type, bool full_discard)
1517 {
1518     BDRVQcow2State *s = bs->opaque;
1519     uint64_t end_offset;
1520     uint64_t nb_clusters;
1521     int ret;
1522 
1523     end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS);
1524 
1525     /* Round start up and end down */
1526     offset = align_offset(offset, s->cluster_size);
1527     end_offset = start_of_cluster(s, end_offset);
1528 
1529     if (offset > end_offset) {
1530         return 0;
1531     }
1532 
1533     nb_clusters = size_to_clusters(s, end_offset - offset);
1534 
1535     s->cache_discards = true;
1536 
1537     /* Each L2 table is handled by its own loop iteration */
1538     while (nb_clusters > 0) {
1539         ret = discard_single_l2(bs, offset, nb_clusters, type, full_discard);
1540         if (ret < 0) {
1541             goto fail;
1542         }
1543 
1544         nb_clusters -= ret;
1545         offset += (ret * s->cluster_size);
1546     }
1547 
1548     ret = 0;
1549 fail:
1550     s->cache_discards = false;
1551     qcow2_process_discards(bs, ret);
1552 
1553     return ret;
1554 }
1555 
1556 /*
1557  * This zeroes as many clusters of nb_clusters as possible at once (i.e.
1558  * all clusters in the same L2 table) and returns the number of zeroed
1559  * clusters.
1560  */
1561 static int zero_single_l2(BlockDriverState *bs, uint64_t offset,
1562                           uint64_t nb_clusters)
1563 {
1564     BDRVQcow2State *s = bs->opaque;
1565     uint64_t *l2_table;
1566     int l2_index;
1567     int ret;
1568     int i;
1569 
1570     ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1571     if (ret < 0) {
1572         return ret;
1573     }
1574 
1575     /* Limit nb_clusters to one L2 table */
1576     nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1577     assert(nb_clusters <= INT_MAX);
1578 
1579     for (i = 0; i < nb_clusters; i++) {
1580         uint64_t old_offset;
1581 
1582         old_offset = be64_to_cpu(l2_table[l2_index + i]);
1583 
1584         /* Update L2 entries */
1585         qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
1586         if (old_offset & QCOW_OFLAG_COMPRESSED) {
1587             l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
1588             qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST);
1589         } else {
1590             l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO);
1591         }
1592     }
1593 
1594     qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1595 
1596     return nb_clusters;
1597 }
1598 
1599 int qcow2_zero_clusters(BlockDriverState *bs, uint64_t offset, int nb_sectors)
1600 {
1601     BDRVQcow2State *s = bs->opaque;
1602     uint64_t nb_clusters;
1603     int ret;
1604 
1605     /* The zero flag is only supported by version 3 and newer */
1606     if (s->qcow_version < 3) {
1607         return -ENOTSUP;
1608     }
1609 
1610     /* Each L2 table is handled by its own loop iteration */
1611     nb_clusters = size_to_clusters(s, nb_sectors << BDRV_SECTOR_BITS);
1612 
1613     s->cache_discards = true;
1614 
1615     while (nb_clusters > 0) {
1616         ret = zero_single_l2(bs, offset, nb_clusters);
1617         if (ret < 0) {
1618             goto fail;
1619         }
1620 
1621         nb_clusters -= ret;
1622         offset += (ret * s->cluster_size);
1623     }
1624 
1625     ret = 0;
1626 fail:
1627     s->cache_discards = false;
1628     qcow2_process_discards(bs, ret);
1629 
1630     return ret;
1631 }
1632 
1633 /*
1634  * Expands all zero clusters in a specific L1 table (or deallocates them, for
1635  * non-backed non-pre-allocated zero clusters).
1636  *
1637  * l1_entries and *visited_l1_entries are used to keep track of progress for
1638  * status_cb(). l1_entries contains the total number of L1 entries and
1639  * *visited_l1_entries counts all visited L1 entries.
1640  */
1641 static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
1642                                       int l1_size, int64_t *visited_l1_entries,
1643                                       int64_t l1_entries,
1644                                       BlockDriverAmendStatusCB *status_cb)
1645 {
1646     BDRVQcow2State *s = bs->opaque;
1647     bool is_active_l1 = (l1_table == s->l1_table);
1648     uint64_t *l2_table = NULL;
1649     int ret;
1650     int i, j;
1651 
1652     if (!is_active_l1) {
1653         /* inactive L2 tables require a buffer to be stored in when loading
1654          * them from disk */
1655         l2_table = qemu_try_blockalign(bs->file->bs, s->cluster_size);
1656         if (l2_table == NULL) {
1657             return -ENOMEM;
1658         }
1659     }
1660 
1661     for (i = 0; i < l1_size; i++) {
1662         uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK;
1663         bool l2_dirty = false;
1664         uint64_t l2_refcount;
1665 
1666         if (!l2_offset) {
1667             /* unallocated */
1668             (*visited_l1_entries)++;
1669             if (status_cb) {
1670                 status_cb(bs, *visited_l1_entries, l1_entries);
1671             }
1672             continue;
1673         }
1674 
1675         if (offset_into_cluster(s, l2_offset)) {
1676             qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#"
1677                                     PRIx64 " unaligned (L1 index: %#x)",
1678                                     l2_offset, i);
1679             ret = -EIO;
1680             goto fail;
1681         }
1682 
1683         if (is_active_l1) {
1684             /* get active L2 tables from cache */
1685             ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset,
1686                     (void **)&l2_table);
1687         } else {
1688             /* load inactive L2 tables from disk */
1689             ret = bdrv_read(bs->file->bs, l2_offset / BDRV_SECTOR_SIZE,
1690                             (void *)l2_table, s->cluster_sectors);
1691         }
1692         if (ret < 0) {
1693             goto fail;
1694         }
1695 
1696         ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
1697                                  &l2_refcount);
1698         if (ret < 0) {
1699             goto fail;
1700         }
1701 
1702         for (j = 0; j < s->l2_size; j++) {
1703             uint64_t l2_entry = be64_to_cpu(l2_table[j]);
1704             int64_t offset = l2_entry & L2E_OFFSET_MASK;
1705             int cluster_type = qcow2_get_cluster_type(l2_entry);
1706             bool preallocated = offset != 0;
1707 
1708             if (cluster_type != QCOW2_CLUSTER_ZERO) {
1709                 continue;
1710             }
1711 
1712             if (!preallocated) {
1713                 if (!bs->backing) {
1714                     /* not backed; therefore we can simply deallocate the
1715                      * cluster */
1716                     l2_table[j] = 0;
1717                     l2_dirty = true;
1718                     continue;
1719                 }
1720 
1721                 offset = qcow2_alloc_clusters(bs, s->cluster_size);
1722                 if (offset < 0) {
1723                     ret = offset;
1724                     goto fail;
1725                 }
1726 
1727                 if (l2_refcount > 1) {
1728                     /* For shared L2 tables, set the refcount accordingly (it is
1729                      * already 1 and needs to be l2_refcount) */
1730                     ret = qcow2_update_cluster_refcount(bs,
1731                             offset >> s->cluster_bits,
1732                             refcount_diff(1, l2_refcount), false,
1733                             QCOW2_DISCARD_OTHER);
1734                     if (ret < 0) {
1735                         qcow2_free_clusters(bs, offset, s->cluster_size,
1736                                             QCOW2_DISCARD_OTHER);
1737                         goto fail;
1738                     }
1739                 }
1740             }
1741 
1742             if (offset_into_cluster(s, offset)) {
1743                 qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset "
1744                                         "%#" PRIx64 " unaligned (L2 offset: %#"
1745                                         PRIx64 ", L2 index: %#x)", offset,
1746                                         l2_offset, j);
1747                 if (!preallocated) {
1748                     qcow2_free_clusters(bs, offset, s->cluster_size,
1749                                         QCOW2_DISCARD_ALWAYS);
1750                 }
1751                 ret = -EIO;
1752                 goto fail;
1753             }
1754 
1755             ret = qcow2_pre_write_overlap_check(bs, 0, offset, s->cluster_size);
1756             if (ret < 0) {
1757                 if (!preallocated) {
1758                     qcow2_free_clusters(bs, offset, s->cluster_size,
1759                                         QCOW2_DISCARD_ALWAYS);
1760                 }
1761                 goto fail;
1762             }
1763 
1764             ret = bdrv_write_zeroes(bs->file->bs, offset / BDRV_SECTOR_SIZE,
1765                                     s->cluster_sectors, 0);
1766             if (ret < 0) {
1767                 if (!preallocated) {
1768                     qcow2_free_clusters(bs, offset, s->cluster_size,
1769                                         QCOW2_DISCARD_ALWAYS);
1770                 }
1771                 goto fail;
1772             }
1773 
1774             if (l2_refcount == 1) {
1775                 l2_table[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED);
1776             } else {
1777                 l2_table[j] = cpu_to_be64(offset);
1778             }
1779             l2_dirty = true;
1780         }
1781 
1782         if (is_active_l1) {
1783             if (l2_dirty) {
1784                 qcow2_cache_entry_mark_dirty(bs, s->l2_table_cache, l2_table);
1785                 qcow2_cache_depends_on_flush(s->l2_table_cache);
1786             }
1787             qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1788         } else {
1789             if (l2_dirty) {
1790                 ret = qcow2_pre_write_overlap_check(bs,
1791                         QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2, l2_offset,
1792                         s->cluster_size);
1793                 if (ret < 0) {
1794                     goto fail;
1795                 }
1796 
1797                 ret = bdrv_write(bs->file->bs, l2_offset / BDRV_SECTOR_SIZE,
1798                                  (void *)l2_table, s->cluster_sectors);
1799                 if (ret < 0) {
1800                     goto fail;
1801                 }
1802             }
1803         }
1804 
1805         (*visited_l1_entries)++;
1806         if (status_cb) {
1807             status_cb(bs, *visited_l1_entries, l1_entries);
1808         }
1809     }
1810 
1811     ret = 0;
1812 
1813 fail:
1814     if (l2_table) {
1815         if (!is_active_l1) {
1816             qemu_vfree(l2_table);
1817         } else {
1818             qcow2_cache_put(bs, s->l2_table_cache, (void **) &l2_table);
1819         }
1820     }
1821     return ret;
1822 }
1823 
1824 /*
1825  * For backed images, expands all zero clusters on the image. For non-backed
1826  * images, deallocates all non-pre-allocated zero clusters (and claims the
1827  * allocation for pre-allocated ones). This is important for downgrading to a
1828  * qcow2 version which doesn't yet support metadata zero clusters.
1829  */
1830 int qcow2_expand_zero_clusters(BlockDriverState *bs,
1831                                BlockDriverAmendStatusCB *status_cb)
1832 {
1833     BDRVQcow2State *s = bs->opaque;
1834     uint64_t *l1_table = NULL;
1835     int64_t l1_entries = 0, visited_l1_entries = 0;
1836     int ret;
1837     int i, j;
1838 
1839     if (status_cb) {
1840         l1_entries = s->l1_size;
1841         for (i = 0; i < s->nb_snapshots; i++) {
1842             l1_entries += s->snapshots[i].l1_size;
1843         }
1844     }
1845 
1846     ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size,
1847                                      &visited_l1_entries, l1_entries,
1848                                      status_cb);
1849     if (ret < 0) {
1850         goto fail;
1851     }
1852 
1853     /* Inactive L1 tables may point to active L2 tables - therefore it is
1854      * necessary to flush the L2 table cache before trying to access the L2
1855      * tables pointed to by inactive L1 entries (else we might try to expand
1856      * zero clusters that have already been expanded); furthermore, it is also
1857      * necessary to empty the L2 table cache, since it may contain tables which
1858      * are now going to be modified directly on disk, bypassing the cache.
1859      * qcow2_cache_empty() does both for us. */
1860     ret = qcow2_cache_empty(bs, s->l2_table_cache);
1861     if (ret < 0) {
1862         goto fail;
1863     }
1864 
1865     for (i = 0; i < s->nb_snapshots; i++) {
1866         int l1_sectors = (s->snapshots[i].l1_size * sizeof(uint64_t) +
1867                 BDRV_SECTOR_SIZE - 1) / BDRV_SECTOR_SIZE;
1868 
1869         l1_table = g_realloc(l1_table, l1_sectors * BDRV_SECTOR_SIZE);
1870 
1871         ret = bdrv_read(bs->file->bs,
1872                         s->snapshots[i].l1_table_offset / BDRV_SECTOR_SIZE,
1873                         (void *)l1_table, l1_sectors);
1874         if (ret < 0) {
1875             goto fail;
1876         }
1877 
1878         for (j = 0; j < s->snapshots[i].l1_size; j++) {
1879             be64_to_cpus(&l1_table[j]);
1880         }
1881 
1882         ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size,
1883                                          &visited_l1_entries, l1_entries,
1884                                          status_cb);
1885         if (ret < 0) {
1886             goto fail;
1887         }
1888     }
1889 
1890     ret = 0;
1891 
1892 fail:
1893     g_free(l1_table);
1894     return ret;
1895 }
1896