xref: /openbmc/qemu/block/qcow2-cluster.c (revision ad6ef0a42e314a8c6ac6c96d5f6e607a1e5644b5)
1  /*
2   * Block driver for the QCOW version 2 format
3   *
4   * Copyright (c) 2004-2006 Fabrice Bellard
5   *
6   * Permission is hereby granted, free of charge, to any person obtaining a copy
7   * of this software and associated documentation files (the "Software"), to deal
8   * in the Software without restriction, including without limitation the rights
9   * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10   * copies of the Software, and to permit persons to whom the Software is
11   * furnished to do so, subject to the following conditions:
12   *
13   * The above copyright notice and this permission notice shall be included in
14   * all copies or substantial portions of the Software.
15   *
16   * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17   * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18   * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19   * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20   * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21   * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22   * THE SOFTWARE.
23   */
24  
25  #include "qemu/osdep.h"
26  #include <zlib.h>
27  
28  #include "block/block-io.h"
29  #include "qapi/error.h"
30  #include "qcow2.h"
31  #include "qemu/bswap.h"
32  #include "qemu/memalign.h"
33  #include "trace.h"
34  
qcow2_shrink_l1_table(BlockDriverState * bs,uint64_t exact_size)35  int coroutine_fn qcow2_shrink_l1_table(BlockDriverState *bs,
36                                         uint64_t exact_size)
37  {
38      BDRVQcow2State *s = bs->opaque;
39      int new_l1_size, i, ret;
40  
41      if (exact_size >= s->l1_size) {
42          return 0;
43      }
44  
45      new_l1_size = exact_size;
46  
47  #ifdef DEBUG_ALLOC2
48      fprintf(stderr, "shrink l1_table from %d to %d\n", s->l1_size, new_l1_size);
49  #endif
50  
51      BLKDBG_CO_EVENT(bs->file, BLKDBG_L1_SHRINK_WRITE_TABLE);
52      ret = bdrv_co_pwrite_zeroes(bs->file,
53                                  s->l1_table_offset + new_l1_size * L1E_SIZE,
54                                  (s->l1_size - new_l1_size) * L1E_SIZE, 0);
55      if (ret < 0) {
56          goto fail;
57      }
58  
59      ret = bdrv_co_flush(bs->file->bs);
60      if (ret < 0) {
61          goto fail;
62      }
63  
64      BLKDBG_CO_EVENT(bs->file, BLKDBG_L1_SHRINK_FREE_L2_CLUSTERS);
65      for (i = s->l1_size - 1; i > new_l1_size - 1; i--) {
66          if ((s->l1_table[i] & L1E_OFFSET_MASK) == 0) {
67              continue;
68          }
69          qcow2_free_clusters(bs, s->l1_table[i] & L1E_OFFSET_MASK,
70                              s->cluster_size, QCOW2_DISCARD_ALWAYS);
71          s->l1_table[i] = 0;
72      }
73      return 0;
74  
75  fail:
76      /*
77       * If the write in the l1_table failed the image may contain a partially
78       * overwritten l1_table. In this case it would be better to clear the
79       * l1_table in memory to avoid possible image corruption.
80       */
81      memset(s->l1_table + new_l1_size, 0,
82             (s->l1_size - new_l1_size) * L1E_SIZE);
83      return ret;
84  }
85  
qcow2_grow_l1_table(BlockDriverState * bs,uint64_t min_size,bool exact_size)86  int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
87                          bool exact_size)
88  {
89      BDRVQcow2State *s = bs->opaque;
90      int new_l1_size2, ret, i;
91      uint64_t *new_l1_table;
92      int64_t old_l1_table_offset, old_l1_size;
93      int64_t new_l1_table_offset, new_l1_size;
94      uint8_t data[12];
95  
96      if (min_size <= s->l1_size)
97          return 0;
98  
99      /* Do a sanity check on min_size before trying to calculate new_l1_size
100       * (this prevents overflows during the while loop for the calculation of
101       * new_l1_size) */
102      if (min_size > INT_MAX / L1E_SIZE) {
103          return -EFBIG;
104      }
105  
106      if (exact_size) {
107          new_l1_size = min_size;
108      } else {
109          /* Bump size up to reduce the number of times we have to grow */
110          new_l1_size = s->l1_size;
111          if (new_l1_size == 0) {
112              new_l1_size = 1;
113          }
114          while (min_size > new_l1_size) {
115              new_l1_size = DIV_ROUND_UP(new_l1_size * 3, 2);
116          }
117      }
118  
119      QEMU_BUILD_BUG_ON(QCOW_MAX_L1_SIZE > INT_MAX);
120      if (new_l1_size > QCOW_MAX_L1_SIZE / L1E_SIZE) {
121          return -EFBIG;
122      }
123  
124  #ifdef DEBUG_ALLOC2
125      fprintf(stderr, "grow l1_table from %d to %" PRId64 "\n",
126              s->l1_size, new_l1_size);
127  #endif
128  
129      new_l1_size2 = L1E_SIZE * new_l1_size;
130      new_l1_table = qemu_try_blockalign(bs->file->bs, new_l1_size2);
131      if (new_l1_table == NULL) {
132          return -ENOMEM;
133      }
134      memset(new_l1_table, 0, new_l1_size2);
135  
136      if (s->l1_size) {
137          memcpy(new_l1_table, s->l1_table, s->l1_size * L1E_SIZE);
138      }
139  
140      /* write new table (align to cluster) */
141      BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE);
142      new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
143      if (new_l1_table_offset < 0) {
144          qemu_vfree(new_l1_table);
145          return new_l1_table_offset;
146      }
147  
148      ret = qcow2_cache_flush(bs, s->refcount_block_cache);
149      if (ret < 0) {
150          goto fail;
151      }
152  
153      /* the L1 position has not yet been updated, so these clusters must
154       * indeed be completely free */
155      ret = qcow2_pre_write_overlap_check(bs, 0, new_l1_table_offset,
156                                          new_l1_size2, false);
157      if (ret < 0) {
158          goto fail;
159      }
160  
161      BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE);
162      for(i = 0; i < s->l1_size; i++)
163          new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
164      ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, new_l1_size2,
165                             new_l1_table, 0);
166      if (ret < 0)
167          goto fail;
168      for(i = 0; i < s->l1_size; i++)
169          new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
170  
171      /* set new table */
172      BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE);
173      stl_be_p(data, new_l1_size);
174      stq_be_p(data + 4, new_l1_table_offset);
175      ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size),
176                             sizeof(data), data, 0);
177      if (ret < 0) {
178          goto fail;
179      }
180      qemu_vfree(s->l1_table);
181      old_l1_table_offset = s->l1_table_offset;
182      s->l1_table_offset = new_l1_table_offset;
183      s->l1_table = new_l1_table;
184      old_l1_size = s->l1_size;
185      s->l1_size = new_l1_size;
186      qcow2_free_clusters(bs, old_l1_table_offset, old_l1_size * L1E_SIZE,
187                          QCOW2_DISCARD_OTHER);
188      return 0;
189   fail:
190      qemu_vfree(new_l1_table);
191      qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2,
192                          QCOW2_DISCARD_OTHER);
193      return ret;
194  }
195  
196  /*
197   * l2_load
198   *
199   * @bs: The BlockDriverState
200   * @offset: A guest offset, used to calculate what slice of the L2
201   *          table to load.
202   * @l2_offset: Offset to the L2 table in the image file.
203   * @l2_slice: Location to store the pointer to the L2 slice.
204   *
205   * Loads a L2 slice into memory (L2 slices are the parts of L2 tables
206   * that are loaded by the qcow2 cache). If the slice is in the cache,
207   * the cache is used; otherwise the L2 slice is loaded from the image
208   * file.
209   */
210  static int GRAPH_RDLOCK
l2_load(BlockDriverState * bs,uint64_t offset,uint64_t l2_offset,uint64_t ** l2_slice)211  l2_load(BlockDriverState *bs, uint64_t offset,
212          uint64_t l2_offset, uint64_t **l2_slice)
213  {
214      BDRVQcow2State *s = bs->opaque;
215      int start_of_slice = l2_entry_size(s) *
216          (offset_to_l2_index(s, offset) - offset_to_l2_slice_index(s, offset));
217  
218      return qcow2_cache_get(bs, s->l2_table_cache, l2_offset + start_of_slice,
219                             (void **)l2_slice);
220  }
221  
222  /*
223   * Writes an L1 entry to disk (note that depending on the alignment
224   * requirements this function may write more that just one entry in
225   * order to prevent bdrv_pwrite from performing a read-modify-write)
226   */
qcow2_write_l1_entry(BlockDriverState * bs,int l1_index)227  int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index)
228  {
229      BDRVQcow2State *s = bs->opaque;
230      int l1_start_index;
231      int i, ret;
232      int bufsize = MAX(L1E_SIZE,
233                        MIN(bs->file->bs->bl.request_alignment, s->cluster_size));
234      int nentries = bufsize / L1E_SIZE;
235      g_autofree uint64_t *buf = g_try_new0(uint64_t, nentries);
236  
237      if (buf == NULL) {
238          return -ENOMEM;
239      }
240  
241      l1_start_index = QEMU_ALIGN_DOWN(l1_index, nentries);
242      for (i = 0; i < MIN(nentries, s->l1_size - l1_start_index); i++) {
243          buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
244      }
245  
246      ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1,
247              s->l1_table_offset + L1E_SIZE * l1_start_index, bufsize, false);
248      if (ret < 0) {
249          return ret;
250      }
251  
252      BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
253      ret = bdrv_pwrite_sync(bs->file,
254                             s->l1_table_offset + L1E_SIZE * l1_start_index,
255                             bufsize, buf, 0);
256      if (ret < 0) {
257          return ret;
258      }
259  
260      return 0;
261  }
262  
263  /*
264   * l2_allocate
265   *
266   * Allocate a new l2 entry in the file. If l1_index points to an already
267   * used entry in the L2 table (i.e. we are doing a copy on write for the L2
268   * table) copy the contents of the old L2 table into the newly allocated one.
269   * Otherwise the new table is initialized with zeros.
270   *
271   */
272  
l2_allocate(BlockDriverState * bs,int l1_index)273  static int GRAPH_RDLOCK l2_allocate(BlockDriverState *bs, int l1_index)
274  {
275      BDRVQcow2State *s = bs->opaque;
276      uint64_t old_l2_offset;
277      uint64_t *l2_slice = NULL;
278      unsigned slice, slice_size2, n_slices;
279      int64_t l2_offset;
280      int ret;
281  
282      old_l2_offset = s->l1_table[l1_index];
283  
284      trace_qcow2_l2_allocate(bs, l1_index);
285  
286      /* allocate a new l2 entry */
287  
288      l2_offset = qcow2_alloc_clusters(bs, s->l2_size * l2_entry_size(s));
289      if (l2_offset < 0) {
290          ret = l2_offset;
291          goto fail;
292      }
293  
294      /* The offset must fit in the offset field of the L1 table entry */
295      assert((l2_offset & L1E_OFFSET_MASK) == l2_offset);
296  
297      /* If we're allocating the table at offset 0 then something is wrong */
298      if (l2_offset == 0) {
299          qcow2_signal_corruption(bs, true, -1, -1, "Preventing invalid "
300                                  "allocation of L2 table at offset 0");
301          ret = -EIO;
302          goto fail;
303      }
304  
305      ret = qcow2_cache_flush(bs, s->refcount_block_cache);
306      if (ret < 0) {
307          goto fail;
308      }
309  
310      /* allocate a new entry in the l2 cache */
311  
312      slice_size2 = s->l2_slice_size * l2_entry_size(s);
313      n_slices = s->cluster_size / slice_size2;
314  
315      trace_qcow2_l2_allocate_get_empty(bs, l1_index);
316      for (slice = 0; slice < n_slices; slice++) {
317          ret = qcow2_cache_get_empty(bs, s->l2_table_cache,
318                                      l2_offset + slice * slice_size2,
319                                      (void **) &l2_slice);
320          if (ret < 0) {
321              goto fail;
322          }
323  
324          if ((old_l2_offset & L1E_OFFSET_MASK) == 0) {
325              /* if there was no old l2 table, clear the new slice */
326              memset(l2_slice, 0, slice_size2);
327          } else {
328              uint64_t *old_slice;
329              uint64_t old_l2_slice_offset =
330                  (old_l2_offset & L1E_OFFSET_MASK) + slice * slice_size2;
331  
332              /* if there was an old l2 table, read a slice from the disk */
333              BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ);
334              ret = qcow2_cache_get(bs, s->l2_table_cache, old_l2_slice_offset,
335                                    (void **) &old_slice);
336              if (ret < 0) {
337                  goto fail;
338              }
339  
340              memcpy(l2_slice, old_slice, slice_size2);
341  
342              qcow2_cache_put(s->l2_table_cache, (void **) &old_slice);
343          }
344  
345          /* write the l2 slice to the file */
346          BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
347  
348          trace_qcow2_l2_allocate_write_l2(bs, l1_index);
349          qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
350          qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
351      }
352  
353      ret = qcow2_cache_flush(bs, s->l2_table_cache);
354      if (ret < 0) {
355          goto fail;
356      }
357  
358      /* update the L1 entry */
359      trace_qcow2_l2_allocate_write_l1(bs, l1_index);
360      s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
361      ret = qcow2_write_l1_entry(bs, l1_index);
362      if (ret < 0) {
363          goto fail;
364      }
365  
366      trace_qcow2_l2_allocate_done(bs, l1_index, 0);
367      return 0;
368  
369  fail:
370      trace_qcow2_l2_allocate_done(bs, l1_index, ret);
371      if (l2_slice != NULL) {
372          qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
373      }
374      s->l1_table[l1_index] = old_l2_offset;
375      if (l2_offset > 0) {
376          qcow2_free_clusters(bs, l2_offset, s->l2_size * l2_entry_size(s),
377                              QCOW2_DISCARD_ALWAYS);
378      }
379      return ret;
380  }
381  
382  /*
383   * For a given L2 entry, count the number of contiguous subclusters of
384   * the same type starting from @sc_from. Compressed clusters are
385   * treated as if they were divided into subclusters of size
386   * s->subcluster_size.
387   *
388   * Return the number of contiguous subclusters and set @type to the
389   * subcluster type.
390   *
391   * If the L2 entry is invalid return -errno and set @type to
392   * QCOW2_SUBCLUSTER_INVALID.
393   */
394  static int GRAPH_RDLOCK
qcow2_get_subcluster_range_type(BlockDriverState * bs,uint64_t l2_entry,uint64_t l2_bitmap,unsigned sc_from,QCow2SubclusterType * type)395  qcow2_get_subcluster_range_type(BlockDriverState *bs, uint64_t l2_entry,
396                                  uint64_t l2_bitmap, unsigned sc_from,
397                                  QCow2SubclusterType *type)
398  {
399      BDRVQcow2State *s = bs->opaque;
400      uint32_t val;
401  
402      *type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc_from);
403  
404      if (*type == QCOW2_SUBCLUSTER_INVALID) {
405          return -EINVAL;
406      } else if (!has_subclusters(s) || *type == QCOW2_SUBCLUSTER_COMPRESSED) {
407          return s->subclusters_per_cluster - sc_from;
408      }
409  
410      switch (*type) {
411      case QCOW2_SUBCLUSTER_NORMAL:
412          val = l2_bitmap | QCOW_OFLAG_SUB_ALLOC_RANGE(0, sc_from);
413          return cto32(val) - sc_from;
414  
415      case QCOW2_SUBCLUSTER_ZERO_PLAIN:
416      case QCOW2_SUBCLUSTER_ZERO_ALLOC:
417          val = (l2_bitmap | QCOW_OFLAG_SUB_ZERO_RANGE(0, sc_from)) >> 32;
418          return cto32(val) - sc_from;
419  
420      case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
421      case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
422          val = ((l2_bitmap >> 32) | l2_bitmap)
423              & ~QCOW_OFLAG_SUB_ALLOC_RANGE(0, sc_from);
424          return ctz32(val) - sc_from;
425  
426      default:
427          g_assert_not_reached();
428      }
429  }
430  
431  /*
432   * Return the number of contiguous subclusters of the exact same type
433   * in a given L2 slice, starting from cluster @l2_index, subcluster
434   * @sc_index. Allocated subclusters are required to be contiguous in
435   * the image file.
436   * At most @nb_clusters are checked (note that this means clusters,
437   * not subclusters).
438   * Compressed clusters are always processed one by one but for the
439   * purpose of this count they are treated as if they were divided into
440   * subclusters of size s->subcluster_size.
441   * On failure return -errno and update @l2_index to point to the
442   * invalid entry.
443   */
444  static int GRAPH_RDLOCK
count_contiguous_subclusters(BlockDriverState * bs,int nb_clusters,unsigned sc_index,uint64_t * l2_slice,unsigned * l2_index)445  count_contiguous_subclusters(BlockDriverState *bs, int nb_clusters,
446                               unsigned sc_index, uint64_t *l2_slice,
447                               unsigned *l2_index)
448  {
449      BDRVQcow2State *s = bs->opaque;
450      int i, count = 0;
451      bool check_offset = false;
452      uint64_t expected_offset = 0;
453      QCow2SubclusterType expected_type = QCOW2_SUBCLUSTER_NORMAL, type;
454  
455      assert(*l2_index + nb_clusters <= s->l2_slice_size);
456  
457      for (i = 0; i < nb_clusters; i++) {
458          unsigned first_sc = (i == 0) ? sc_index : 0;
459          uint64_t l2_entry = get_l2_entry(s, l2_slice, *l2_index + i);
460          uint64_t l2_bitmap = get_l2_bitmap(s, l2_slice, *l2_index + i);
461          int ret = qcow2_get_subcluster_range_type(bs, l2_entry, l2_bitmap,
462                                                    first_sc, &type);
463          if (ret < 0) {
464              *l2_index += i; /* Point to the invalid entry */
465              return -EIO;
466          }
467          if (i == 0) {
468              if (type == QCOW2_SUBCLUSTER_COMPRESSED) {
469                  /* Compressed clusters are always processed one by one */
470                  return ret;
471              }
472              expected_type = type;
473              expected_offset = l2_entry & L2E_OFFSET_MASK;
474              check_offset = (type == QCOW2_SUBCLUSTER_NORMAL ||
475                              type == QCOW2_SUBCLUSTER_ZERO_ALLOC ||
476                              type == QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC);
477          } else if (type != expected_type) {
478              break;
479          } else if (check_offset) {
480              expected_offset += s->cluster_size;
481              if (expected_offset != (l2_entry & L2E_OFFSET_MASK)) {
482                  break;
483              }
484          }
485          count += ret;
486          /* Stop if there are type changes before the end of the cluster */
487          if (first_sc + ret < s->subclusters_per_cluster) {
488              break;
489          }
490      }
491  
492      return count;
493  }
494  
495  static int coroutine_fn GRAPH_RDLOCK
do_perform_cow_read(BlockDriverState * bs,uint64_t src_cluster_offset,unsigned offset_in_cluster,QEMUIOVector * qiov)496  do_perform_cow_read(BlockDriverState *bs, uint64_t src_cluster_offset,
497                      unsigned offset_in_cluster, QEMUIOVector *qiov)
498  {
499      int ret;
500  
501      if (qiov->size == 0) {
502          return 0;
503      }
504  
505      BLKDBG_CO_EVENT(bs->file, BLKDBG_COW_READ);
506  
507      if (!bs->drv) {
508          return -ENOMEDIUM;
509      }
510  
511      /*
512       * We never deal with requests that don't satisfy
513       * bdrv_check_qiov_request(), and aligning requests to clusters never
514       * breaks this condition. So, do some assertions before calling
515       * bs->drv->bdrv_co_preadv_part() which has int64_t arguments.
516       */
517      assert(src_cluster_offset <= INT64_MAX);
518      assert(src_cluster_offset + offset_in_cluster <= INT64_MAX);
519      /* Cast qiov->size to uint64_t to silence a compiler warning on -m32 */
520      assert((uint64_t)qiov->size <= INT64_MAX);
521      bdrv_check_qiov_request(src_cluster_offset + offset_in_cluster, qiov->size,
522                              qiov, 0, &error_abort);
523      /*
524       * Call .bdrv_co_readv() directly instead of using the public block-layer
525       * interface.  This avoids double I/O throttling and request tracking,
526       * which can lead to deadlock when block layer copy-on-read is enabled.
527       */
528      ret = bs->drv->bdrv_co_preadv_part(bs,
529                                         src_cluster_offset + offset_in_cluster,
530                                         qiov->size, qiov, 0, 0);
531      if (ret < 0) {
532          return ret;
533      }
534  
535      return 0;
536  }
537  
538  static int coroutine_fn GRAPH_RDLOCK
do_perform_cow_write(BlockDriverState * bs,uint64_t cluster_offset,unsigned offset_in_cluster,QEMUIOVector * qiov)539  do_perform_cow_write(BlockDriverState *bs, uint64_t cluster_offset,
540                       unsigned offset_in_cluster, QEMUIOVector *qiov)
541  {
542      BDRVQcow2State *s = bs->opaque;
543      int ret;
544  
545      if (qiov->size == 0) {
546          return 0;
547      }
548  
549      ret = qcow2_pre_write_overlap_check(bs, 0,
550              cluster_offset + offset_in_cluster, qiov->size, true);
551      if (ret < 0) {
552          return ret;
553      }
554  
555      BLKDBG_CO_EVENT(bs->file, BLKDBG_COW_WRITE);
556      ret = bdrv_co_pwritev(s->data_file, cluster_offset + offset_in_cluster,
557                            qiov->size, qiov, 0);
558      if (ret < 0) {
559          return ret;
560      }
561  
562      return 0;
563  }
564  
565  
566  /*
567   * get_host_offset
568   *
569   * For a given offset of the virtual disk find the equivalent host
570   * offset in the qcow2 file and store it in *host_offset. Neither
571   * offset needs to be aligned to a cluster boundary.
572   *
573   * If the cluster is unallocated then *host_offset will be 0.
574   * If the cluster is compressed then *host_offset will contain the l2 entry.
575   *
576   * On entry, *bytes is the maximum number of contiguous bytes starting at
577   * offset that we are interested in.
578   *
579   * On exit, *bytes is the number of bytes starting at offset that have the same
580   * subcluster type and (if applicable) are stored contiguously in the image
581   * file. The subcluster type is stored in *subcluster_type.
582   * Compressed clusters are always processed one by one.
583   *
584   * Returns 0 on success, -errno in error cases.
585   */
qcow2_get_host_offset(BlockDriverState * bs,uint64_t offset,unsigned int * bytes,uint64_t * host_offset,QCow2SubclusterType * subcluster_type)586  int qcow2_get_host_offset(BlockDriverState *bs, uint64_t offset,
587                            unsigned int *bytes, uint64_t *host_offset,
588                            QCow2SubclusterType *subcluster_type)
589  {
590      BDRVQcow2State *s = bs->opaque;
591      unsigned int l2_index, sc_index;
592      uint64_t l1_index, l2_offset, *l2_slice, l2_entry, l2_bitmap;
593      int sc;
594      unsigned int offset_in_cluster;
595      uint64_t bytes_available, bytes_needed, nb_clusters;
596      QCow2SubclusterType type;
597      int ret;
598  
599      offset_in_cluster = offset_into_cluster(s, offset);
600      bytes_needed = (uint64_t) *bytes + offset_in_cluster;
601  
602      /* compute how many bytes there are between the start of the cluster
603       * containing offset and the end of the l2 slice that contains
604       * the entry pointing to it */
605      bytes_available =
606          ((uint64_t) (s->l2_slice_size - offset_to_l2_slice_index(s, offset)))
607          << s->cluster_bits;
608  
609      if (bytes_needed > bytes_available) {
610          bytes_needed = bytes_available;
611      }
612  
613      *host_offset = 0;
614  
615      /* seek to the l2 offset in the l1 table */
616  
617      l1_index = offset_to_l1_index(s, offset);
618      if (l1_index >= s->l1_size) {
619          type = QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN;
620          goto out;
621      }
622  
623      l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
624      if (!l2_offset) {
625          type = QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN;
626          goto out;
627      }
628  
629      if (offset_into_cluster(s, l2_offset)) {
630          qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64
631                                  " unaligned (L1 index: %#" PRIx64 ")",
632                                  l2_offset, l1_index);
633          return -EIO;
634      }
635  
636      /* load the l2 slice in memory */
637  
638      ret = l2_load(bs, offset, l2_offset, &l2_slice);
639      if (ret < 0) {
640          return ret;
641      }
642  
643      /* find the cluster offset for the given disk offset */
644  
645      l2_index = offset_to_l2_slice_index(s, offset);
646      sc_index = offset_to_sc_index(s, offset);
647      l2_entry = get_l2_entry(s, l2_slice, l2_index);
648      l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index);
649  
650      nb_clusters = size_to_clusters(s, bytes_needed);
651      /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned
652       * integers; the minimum cluster size is 512, so this assertion is always
653       * true */
654      assert(nb_clusters <= INT_MAX);
655  
656      type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc_index);
657      if (s->qcow_version < 3 && (type == QCOW2_SUBCLUSTER_ZERO_PLAIN ||
658                                  type == QCOW2_SUBCLUSTER_ZERO_ALLOC)) {
659          qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found"
660                                  " in pre-v3 image (L2 offset: %#" PRIx64
661                                  ", L2 index: %#x)", l2_offset, l2_index);
662          ret = -EIO;
663          goto fail;
664      }
665      switch (type) {
666      case QCOW2_SUBCLUSTER_INVALID:
667          break; /* This is handled by count_contiguous_subclusters() below */
668      case QCOW2_SUBCLUSTER_COMPRESSED:
669          if (has_data_file(bs)) {
670              qcow2_signal_corruption(bs, true, -1, -1, "Compressed cluster "
671                                      "entry found in image with external data "
672                                      "file (L2 offset: %#" PRIx64 ", L2 index: "
673                                      "%#x)", l2_offset, l2_index);
674              ret = -EIO;
675              goto fail;
676          }
677          *host_offset = l2_entry;
678          break;
679      case QCOW2_SUBCLUSTER_ZERO_PLAIN:
680      case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
681          break;
682      case QCOW2_SUBCLUSTER_ZERO_ALLOC:
683      case QCOW2_SUBCLUSTER_NORMAL:
684      case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC: {
685          uint64_t host_cluster_offset = l2_entry & L2E_OFFSET_MASK;
686          *host_offset = host_cluster_offset + offset_in_cluster;
687          if (offset_into_cluster(s, host_cluster_offset)) {
688              qcow2_signal_corruption(bs, true, -1, -1,
689                                      "Cluster allocation offset %#"
690                                      PRIx64 " unaligned (L2 offset: %#" PRIx64
691                                      ", L2 index: %#x)", host_cluster_offset,
692                                      l2_offset, l2_index);
693              ret = -EIO;
694              goto fail;
695          }
696          if (has_data_file(bs) && *host_offset != offset) {
697              qcow2_signal_corruption(bs, true, -1, -1,
698                                      "External data file host cluster offset %#"
699                                      PRIx64 " does not match guest cluster "
700                                      "offset: %#" PRIx64
701                                      ", L2 index: %#x)", host_cluster_offset,
702                                      offset - offset_in_cluster, l2_index);
703              ret = -EIO;
704              goto fail;
705          }
706          break;
707      }
708      default:
709          abort();
710      }
711  
712      sc = count_contiguous_subclusters(bs, nb_clusters, sc_index,
713                                        l2_slice, &l2_index);
714      if (sc < 0) {
715          qcow2_signal_corruption(bs, true, -1, -1, "Invalid cluster entry found "
716                                  " (L2 offset: %#" PRIx64 ", L2 index: %#x)",
717                                  l2_offset, l2_index);
718          ret = -EIO;
719          goto fail;
720      }
721      qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
722  
723      bytes_available = ((int64_t)sc + sc_index) << s->subcluster_bits;
724  
725  out:
726      if (bytes_available > bytes_needed) {
727          bytes_available = bytes_needed;
728      }
729  
730      /* bytes_available <= bytes_needed <= *bytes + offset_in_cluster;
731       * subtracting offset_in_cluster will therefore definitely yield something
732       * not exceeding UINT_MAX */
733      assert(bytes_available - offset_in_cluster <= UINT_MAX);
734      *bytes = bytes_available - offset_in_cluster;
735  
736      *subcluster_type = type;
737  
738      return 0;
739  
740  fail:
741      qcow2_cache_put(s->l2_table_cache, (void **)&l2_slice);
742      return ret;
743  }
744  
745  /*
746   * get_cluster_table
747   *
748   * for a given disk offset, load (and allocate if needed)
749   * the appropriate slice of its l2 table.
750   *
751   * the cluster index in the l2 slice is given to the caller.
752   *
753   * Returns 0 on success, -errno in failure case
754   */
755  static int GRAPH_RDLOCK
get_cluster_table(BlockDriverState * bs,uint64_t offset,uint64_t ** new_l2_slice,int * new_l2_index)756  get_cluster_table(BlockDriverState *bs, uint64_t offset,
757                    uint64_t **new_l2_slice, int *new_l2_index)
758  {
759      BDRVQcow2State *s = bs->opaque;
760      unsigned int l2_index;
761      uint64_t l1_index, l2_offset;
762      uint64_t *l2_slice = NULL;
763      int ret;
764  
765      /* seek to the l2 offset in the l1 table */
766  
767      l1_index = offset_to_l1_index(s, offset);
768      if (l1_index >= s->l1_size) {
769          ret = qcow2_grow_l1_table(bs, l1_index + 1, false);
770          if (ret < 0) {
771              return ret;
772          }
773      }
774  
775      assert(l1_index < s->l1_size);
776      l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
777      if (offset_into_cluster(s, l2_offset)) {
778          qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#" PRIx64
779                                  " unaligned (L1 index: %#" PRIx64 ")",
780                                  l2_offset, l1_index);
781          return -EIO;
782      }
783  
784      if (!(s->l1_table[l1_index] & QCOW_OFLAG_COPIED)) {
785          /* First allocate a new L2 table (and do COW if needed) */
786          ret = l2_allocate(bs, l1_index);
787          if (ret < 0) {
788              return ret;
789          }
790  
791          /* Then decrease the refcount of the old table */
792          if (l2_offset) {
793              qcow2_free_clusters(bs, l2_offset, s->l2_size * l2_entry_size(s),
794                                  QCOW2_DISCARD_OTHER);
795          }
796  
797          /* Get the offset of the newly-allocated l2 table */
798          l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
799          assert(offset_into_cluster(s, l2_offset) == 0);
800      }
801  
802      /* load the l2 slice in memory */
803      ret = l2_load(bs, offset, l2_offset, &l2_slice);
804      if (ret < 0) {
805          return ret;
806      }
807  
808      /* find the cluster offset for the given disk offset */
809  
810      l2_index = offset_to_l2_slice_index(s, offset);
811  
812      *new_l2_slice = l2_slice;
813      *new_l2_index = l2_index;
814  
815      return 0;
816  }
817  
818  /*
819   * alloc_compressed_cluster_offset
820   *
821   * For a given offset on the virtual disk, allocate a new compressed cluster
822   * and put the host offset of the cluster into *host_offset. If a cluster is
823   * already allocated at the offset, return an error.
824   *
825   * Return 0 on success and -errno in error cases
826   */
827  int coroutine_fn GRAPH_RDLOCK
qcow2_alloc_compressed_cluster_offset(BlockDriverState * bs,uint64_t offset,int compressed_size,uint64_t * host_offset)828  qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs, uint64_t offset,
829                                        int compressed_size, uint64_t *host_offset)
830  {
831      BDRVQcow2State *s = bs->opaque;
832      int l2_index, ret;
833      uint64_t *l2_slice;
834      int64_t cluster_offset;
835      int nb_csectors;
836  
837      if (has_data_file(bs)) {
838          return 0;
839      }
840  
841      ret = get_cluster_table(bs, offset, &l2_slice, &l2_index);
842      if (ret < 0) {
843          return ret;
844      }
845  
846      /* Compression can't overwrite anything. Fail if the cluster was already
847       * allocated. */
848      cluster_offset = get_l2_entry(s, l2_slice, l2_index);
849      if (cluster_offset & L2E_OFFSET_MASK) {
850          qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
851          return -EIO;
852      }
853  
854      cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
855      if (cluster_offset < 0) {
856          qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
857          return cluster_offset;
858      }
859  
860      nb_csectors =
861          (cluster_offset + compressed_size - 1) / QCOW2_COMPRESSED_SECTOR_SIZE -
862          (cluster_offset / QCOW2_COMPRESSED_SECTOR_SIZE);
863  
864      /* The offset and size must fit in their fields of the L2 table entry */
865      assert((cluster_offset & s->cluster_offset_mask) == cluster_offset);
866      assert((nb_csectors & s->csize_mask) == nb_csectors);
867  
868      cluster_offset |= QCOW_OFLAG_COMPRESSED |
869                        ((uint64_t)nb_csectors << s->csize_shift);
870  
871      /* update L2 table */
872  
873      /* compressed clusters never have the copied flag */
874  
875      BLKDBG_CO_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
876      qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
877      set_l2_entry(s, l2_slice, l2_index, cluster_offset);
878      if (has_subclusters(s)) {
879          set_l2_bitmap(s, l2_slice, l2_index, 0);
880      }
881      qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
882  
883      *host_offset = cluster_offset & s->cluster_offset_mask;
884      return 0;
885  }
886  
887  static int coroutine_fn GRAPH_RDLOCK
perform_cow(BlockDriverState * bs,QCowL2Meta * m)888  perform_cow(BlockDriverState *bs, QCowL2Meta *m)
889  {
890      BDRVQcow2State *s = bs->opaque;
891      Qcow2COWRegion *start = &m->cow_start;
892      Qcow2COWRegion *end = &m->cow_end;
893      unsigned buffer_size;
894      unsigned data_bytes = end->offset - (start->offset + start->nb_bytes);
895      bool merge_reads;
896      uint8_t *start_buffer, *end_buffer;
897      QEMUIOVector qiov;
898      int ret;
899  
900      assert(start->nb_bytes <= UINT_MAX - end->nb_bytes);
901      assert(start->nb_bytes + end->nb_bytes <= UINT_MAX - data_bytes);
902      assert(start->offset + start->nb_bytes <= end->offset);
903  
904      if ((start->nb_bytes == 0 && end->nb_bytes == 0) || m->skip_cow) {
905          return 0;
906      }
907  
908      /* If we have to read both the start and end COW regions and the
909       * middle region is not too large then perform just one read
910       * operation */
911      merge_reads = start->nb_bytes && end->nb_bytes && data_bytes <= 16384;
912      if (merge_reads) {
913          buffer_size = start->nb_bytes + data_bytes + end->nb_bytes;
914      } else {
915          /* If we have to do two reads, add some padding in the middle
916           * if necessary to make sure that the end region is optimally
917           * aligned. */
918          size_t align = bdrv_opt_mem_align(bs);
919          assert(align > 0 && align <= UINT_MAX);
920          assert(QEMU_ALIGN_UP(start->nb_bytes, align) <=
921                 UINT_MAX - end->nb_bytes);
922          buffer_size = QEMU_ALIGN_UP(start->nb_bytes, align) + end->nb_bytes;
923      }
924  
925      /* Reserve a buffer large enough to store all the data that we're
926       * going to read */
927      start_buffer = qemu_try_blockalign(bs, buffer_size);
928      if (start_buffer == NULL) {
929          return -ENOMEM;
930      }
931      /* The part of the buffer where the end region is located */
932      end_buffer = start_buffer + buffer_size - end->nb_bytes;
933  
934      qemu_iovec_init(&qiov, 2 + (m->data_qiov ?
935                                  qemu_iovec_subvec_niov(m->data_qiov,
936                                                         m->data_qiov_offset,
937                                                         data_bytes)
938                                  : 0));
939  
940      qemu_co_mutex_unlock(&s->lock);
941      /* First we read the existing data from both COW regions. We
942       * either read the whole region in one go, or the start and end
943       * regions separately. */
944      if (merge_reads) {
945          qemu_iovec_add(&qiov, start_buffer, buffer_size);
946          ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov);
947      } else {
948          qemu_iovec_add(&qiov, start_buffer, start->nb_bytes);
949          ret = do_perform_cow_read(bs, m->offset, start->offset, &qiov);
950          if (ret < 0) {
951              goto fail;
952          }
953  
954          qemu_iovec_reset(&qiov);
955          qemu_iovec_add(&qiov, end_buffer, end->nb_bytes);
956          ret = do_perform_cow_read(bs, m->offset, end->offset, &qiov);
957      }
958      if (ret < 0) {
959          goto fail;
960      }
961  
962      /* Encrypt the data if necessary before writing it */
963      if (bs->encrypted) {
964          ret = qcow2_co_encrypt(bs,
965                                 m->alloc_offset + start->offset,
966                                 m->offset + start->offset,
967                                 start_buffer, start->nb_bytes);
968          if (ret < 0) {
969              goto fail;
970          }
971  
972          ret = qcow2_co_encrypt(bs,
973                                 m->alloc_offset + end->offset,
974                                 m->offset + end->offset,
975                                 end_buffer, end->nb_bytes);
976          if (ret < 0) {
977              goto fail;
978          }
979      }
980  
981      /* And now we can write everything. If we have the guest data we
982       * can write everything in one single operation */
983      if (m->data_qiov) {
984          qemu_iovec_reset(&qiov);
985          if (start->nb_bytes) {
986              qemu_iovec_add(&qiov, start_buffer, start->nb_bytes);
987          }
988          qemu_iovec_concat(&qiov, m->data_qiov, m->data_qiov_offset, data_bytes);
989          if (end->nb_bytes) {
990              qemu_iovec_add(&qiov, end_buffer, end->nb_bytes);
991          }
992          /* NOTE: we have a write_aio blkdebug event here followed by
993           * a cow_write one in do_perform_cow_write(), but there's only
994           * one single I/O operation */
995          BLKDBG_CO_EVENT(bs->file, BLKDBG_WRITE_AIO);
996          ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov);
997      } else {
998          /* If there's no guest data then write both COW regions separately */
999          qemu_iovec_reset(&qiov);
1000          qemu_iovec_add(&qiov, start_buffer, start->nb_bytes);
1001          ret = do_perform_cow_write(bs, m->alloc_offset, start->offset, &qiov);
1002          if (ret < 0) {
1003              goto fail;
1004          }
1005  
1006          qemu_iovec_reset(&qiov);
1007          qemu_iovec_add(&qiov, end_buffer, end->nb_bytes);
1008          ret = do_perform_cow_write(bs, m->alloc_offset, end->offset, &qiov);
1009      }
1010  
1011  fail:
1012      qemu_co_mutex_lock(&s->lock);
1013  
1014      /*
1015       * Before we update the L2 table to actually point to the new cluster, we
1016       * need to be sure that the refcounts have been increased and COW was
1017       * handled.
1018       */
1019      if (ret == 0) {
1020          qcow2_cache_depends_on_flush(s->l2_table_cache);
1021      }
1022  
1023      qemu_vfree(start_buffer);
1024      qemu_iovec_destroy(&qiov);
1025      return ret;
1026  }
1027  
qcow2_alloc_cluster_link_l2(BlockDriverState * bs,QCowL2Meta * m)1028  int coroutine_fn qcow2_alloc_cluster_link_l2(BlockDriverState *bs,
1029                                               QCowL2Meta *m)
1030  {
1031      BDRVQcow2State *s = bs->opaque;
1032      int i, j = 0, l2_index, ret;
1033      uint64_t *old_cluster, *l2_slice;
1034      uint64_t cluster_offset = m->alloc_offset;
1035  
1036      trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters);
1037      assert(m->nb_clusters > 0);
1038  
1039      old_cluster = g_try_new(uint64_t, m->nb_clusters);
1040      if (old_cluster == NULL) {
1041          ret = -ENOMEM;
1042          goto err;
1043      }
1044  
1045      /* copy content of unmodified sectors */
1046      ret = perform_cow(bs, m);
1047      if (ret < 0) {
1048          goto err;
1049      }
1050  
1051      /* Update L2 table. */
1052      if (s->use_lazy_refcounts) {
1053          qcow2_mark_dirty(bs);
1054      }
1055      if (qcow2_need_accurate_refcounts(s)) {
1056          qcow2_cache_set_dependency(bs, s->l2_table_cache,
1057                                     s->refcount_block_cache);
1058      }
1059  
1060      ret = get_cluster_table(bs, m->offset, &l2_slice, &l2_index);
1061      if (ret < 0) {
1062          goto err;
1063      }
1064      qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
1065  
1066      assert(l2_index + m->nb_clusters <= s->l2_slice_size);
1067      assert(m->cow_end.offset + m->cow_end.nb_bytes <=
1068             m->nb_clusters << s->cluster_bits);
1069      for (i = 0; i < m->nb_clusters; i++) {
1070          uint64_t offset = cluster_offset + ((uint64_t)i << s->cluster_bits);
1071          /* if two concurrent writes happen to the same unallocated cluster
1072           * each write allocates separate cluster and writes data concurrently.
1073           * The first one to complete updates l2 table with pointer to its
1074           * cluster the second one has to do RMW (which is done above by
1075           * perform_cow()), update l2 table with its cluster pointer and free
1076           * old cluster. This is what this loop does */
1077          if (get_l2_entry(s, l2_slice, l2_index + i) != 0) {
1078              old_cluster[j++] = get_l2_entry(s, l2_slice, l2_index + i);
1079          }
1080  
1081          /* The offset must fit in the offset field of the L2 table entry */
1082          assert((offset & L2E_OFFSET_MASK) == offset);
1083  
1084          set_l2_entry(s, l2_slice, l2_index + i, offset | QCOW_OFLAG_COPIED);
1085  
1086          /* Update bitmap with the subclusters that were just written */
1087          if (has_subclusters(s) && !m->prealloc) {
1088              uint64_t l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i);
1089              unsigned written_from = m->cow_start.offset;
1090              unsigned written_to = m->cow_end.offset + m->cow_end.nb_bytes;
1091              int first_sc, last_sc;
1092              /* Narrow written_from and written_to down to the current cluster */
1093              written_from = MAX(written_from, i << s->cluster_bits);
1094              written_to   = MIN(written_to, (i + 1) << s->cluster_bits);
1095              assert(written_from < written_to);
1096              first_sc = offset_to_sc_index(s, written_from);
1097              last_sc  = offset_to_sc_index(s, written_to - 1);
1098              l2_bitmap |= QCOW_OFLAG_SUB_ALLOC_RANGE(first_sc, last_sc + 1);
1099              l2_bitmap &= ~QCOW_OFLAG_SUB_ZERO_RANGE(first_sc, last_sc + 1);
1100              set_l2_bitmap(s, l2_slice, l2_index + i, l2_bitmap);
1101          }
1102       }
1103  
1104  
1105      qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
1106  
1107      /*
1108       * If this was a COW, we need to decrease the refcount of the old cluster.
1109       *
1110       * Don't discard clusters that reach a refcount of 0 (e.g. compressed
1111       * clusters), the next write will reuse them anyway.
1112       */
1113      if (!m->keep_old_clusters && j != 0) {
1114          for (i = 0; i < j; i++) {
1115              qcow2_free_any_cluster(bs, old_cluster[i], QCOW2_DISCARD_NEVER);
1116          }
1117      }
1118  
1119      ret = 0;
1120  err:
1121      g_free(old_cluster);
1122      return ret;
1123   }
1124  
1125  /**
1126   * Frees the allocated clusters because the request failed and they won't
1127   * actually be linked.
1128   */
qcow2_alloc_cluster_abort(BlockDriverState * bs,QCowL2Meta * m)1129  void coroutine_fn qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m)
1130  {
1131      BDRVQcow2State *s = bs->opaque;
1132      if (!has_data_file(bs) && !m->keep_old_clusters) {
1133          qcow2_free_clusters(bs, m->alloc_offset,
1134                              m->nb_clusters << s->cluster_bits,
1135                              QCOW2_DISCARD_NEVER);
1136      }
1137  }
1138  
1139  /*
1140   * For a given write request, create a new QCowL2Meta structure, add
1141   * it to @m and the BDRVQcow2State.cluster_allocs list. If the write
1142   * request does not need copy-on-write or changes to the L2 metadata
1143   * then this function does nothing.
1144   *
1145   * @host_cluster_offset points to the beginning of the first cluster.
1146   *
1147   * @guest_offset and @bytes indicate the offset and length of the
1148   * request.
1149   *
1150   * @l2_slice contains the L2 entries of all clusters involved in this
1151   * write request.
1152   *
1153   * If @keep_old is true it means that the clusters were already
1154   * allocated and will be overwritten. If false then the clusters are
1155   * new and we have to decrease the reference count of the old ones.
1156   *
1157   * Returns 0 on success, -errno on failure.
1158   */
1159  static int coroutine_fn GRAPH_RDLOCK
calculate_l2_meta(BlockDriverState * bs,uint64_t host_cluster_offset,uint64_t guest_offset,unsigned bytes,uint64_t * l2_slice,QCowL2Meta ** m,bool keep_old)1160  calculate_l2_meta(BlockDriverState *bs, uint64_t host_cluster_offset,
1161                    uint64_t guest_offset, unsigned bytes, uint64_t *l2_slice,
1162                    QCowL2Meta **m, bool keep_old)
1163  {
1164      BDRVQcow2State *s = bs->opaque;
1165      int sc_index, l2_index = offset_to_l2_slice_index(s, guest_offset);
1166      uint64_t l2_entry, l2_bitmap;
1167      unsigned cow_start_from, cow_end_to;
1168      unsigned cow_start_to = offset_into_cluster(s, guest_offset);
1169      unsigned cow_end_from = cow_start_to + bytes;
1170      unsigned nb_clusters = size_to_clusters(s, cow_end_from);
1171      QCowL2Meta *old_m = *m;
1172      QCow2SubclusterType type;
1173      int i;
1174      bool skip_cow = keep_old;
1175  
1176      assert(nb_clusters <= s->l2_slice_size - l2_index);
1177  
1178      /* Check the type of all affected subclusters */
1179      for (i = 0; i < nb_clusters; i++) {
1180          l2_entry = get_l2_entry(s, l2_slice, l2_index + i);
1181          l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i);
1182          if (skip_cow) {
1183              unsigned write_from = MAX(cow_start_to, i << s->cluster_bits);
1184              unsigned write_to = MIN(cow_end_from, (i + 1) << s->cluster_bits);
1185              int first_sc = offset_to_sc_index(s, write_from);
1186              int last_sc = offset_to_sc_index(s, write_to - 1);
1187              int cnt = qcow2_get_subcluster_range_type(bs, l2_entry, l2_bitmap,
1188                                                        first_sc, &type);
1189              /* Is any of the subclusters of type != QCOW2_SUBCLUSTER_NORMAL ? */
1190              if (type != QCOW2_SUBCLUSTER_NORMAL || first_sc + cnt <= last_sc) {
1191                  skip_cow = false;
1192              }
1193          } else {
1194              /* If we can't skip the cow we can still look for invalid entries */
1195              type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, 0);
1196          }
1197          if (type == QCOW2_SUBCLUSTER_INVALID) {
1198              int l1_index = offset_to_l1_index(s, guest_offset);
1199              uint64_t l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
1200              qcow2_signal_corruption(bs, true, -1, -1, "Invalid cluster "
1201                                      "entry found (L2 offset: %#" PRIx64
1202                                      ", L2 index: %#x)",
1203                                      l2_offset, l2_index + i);
1204              return -EIO;
1205          }
1206      }
1207  
1208      if (skip_cow) {
1209          return 0;
1210      }
1211  
1212      /* Get the L2 entry of the first cluster */
1213      l2_entry = get_l2_entry(s, l2_slice, l2_index);
1214      l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index);
1215      sc_index = offset_to_sc_index(s, guest_offset);
1216      type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc_index);
1217  
1218      if (!keep_old) {
1219          switch (type) {
1220          case QCOW2_SUBCLUSTER_COMPRESSED:
1221              cow_start_from = 0;
1222              break;
1223          case QCOW2_SUBCLUSTER_NORMAL:
1224          case QCOW2_SUBCLUSTER_ZERO_ALLOC:
1225          case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
1226              if (has_subclusters(s)) {
1227                  /* Skip all leading zero and unallocated subclusters */
1228                  uint32_t alloc_bitmap = l2_bitmap & QCOW_L2_BITMAP_ALL_ALLOC;
1229                  cow_start_from =
1230                      MIN(sc_index, ctz32(alloc_bitmap)) << s->subcluster_bits;
1231              } else {
1232                  cow_start_from = 0;
1233              }
1234              break;
1235          case QCOW2_SUBCLUSTER_ZERO_PLAIN:
1236          case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
1237              cow_start_from = sc_index << s->subcluster_bits;
1238              break;
1239          default:
1240              g_assert_not_reached();
1241          }
1242      } else {
1243          switch (type) {
1244          case QCOW2_SUBCLUSTER_NORMAL:
1245              cow_start_from = cow_start_to;
1246              break;
1247          case QCOW2_SUBCLUSTER_ZERO_ALLOC:
1248          case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
1249              cow_start_from = sc_index << s->subcluster_bits;
1250              break;
1251          default:
1252              g_assert_not_reached();
1253          }
1254      }
1255  
1256      /* Get the L2 entry of the last cluster */
1257      l2_index += nb_clusters - 1;
1258      l2_entry = get_l2_entry(s, l2_slice, l2_index);
1259      l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index);
1260      sc_index = offset_to_sc_index(s, guest_offset + bytes - 1);
1261      type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc_index);
1262  
1263      if (!keep_old) {
1264          switch (type) {
1265          case QCOW2_SUBCLUSTER_COMPRESSED:
1266              cow_end_to = ROUND_UP(cow_end_from, s->cluster_size);
1267              break;
1268          case QCOW2_SUBCLUSTER_NORMAL:
1269          case QCOW2_SUBCLUSTER_ZERO_ALLOC:
1270          case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
1271              cow_end_to = ROUND_UP(cow_end_from, s->cluster_size);
1272              if (has_subclusters(s)) {
1273                  /* Skip all trailing zero and unallocated subclusters */
1274                  uint32_t alloc_bitmap = l2_bitmap & QCOW_L2_BITMAP_ALL_ALLOC;
1275                  cow_end_to -=
1276                      MIN(s->subclusters_per_cluster - sc_index - 1,
1277                          clz32(alloc_bitmap)) << s->subcluster_bits;
1278              }
1279              break;
1280          case QCOW2_SUBCLUSTER_ZERO_PLAIN:
1281          case QCOW2_SUBCLUSTER_UNALLOCATED_PLAIN:
1282              cow_end_to = ROUND_UP(cow_end_from, s->subcluster_size);
1283              break;
1284          default:
1285              g_assert_not_reached();
1286          }
1287      } else {
1288          switch (type) {
1289          case QCOW2_SUBCLUSTER_NORMAL:
1290              cow_end_to = cow_end_from;
1291              break;
1292          case QCOW2_SUBCLUSTER_ZERO_ALLOC:
1293          case QCOW2_SUBCLUSTER_UNALLOCATED_ALLOC:
1294              cow_end_to = ROUND_UP(cow_end_from, s->subcluster_size);
1295              break;
1296          default:
1297              g_assert_not_reached();
1298          }
1299      }
1300  
1301      *m = g_malloc0(sizeof(**m));
1302      **m = (QCowL2Meta) {
1303          .next           = old_m,
1304  
1305          .alloc_offset   = host_cluster_offset,
1306          .offset         = start_of_cluster(s, guest_offset),
1307          .nb_clusters    = nb_clusters,
1308  
1309          .keep_old_clusters = keep_old,
1310  
1311          .cow_start = {
1312              .offset     = cow_start_from,
1313              .nb_bytes   = cow_start_to - cow_start_from,
1314          },
1315          .cow_end = {
1316              .offset     = cow_end_from,
1317              .nb_bytes   = cow_end_to - cow_end_from,
1318          },
1319      };
1320  
1321      qemu_co_queue_init(&(*m)->dependent_requests);
1322      QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
1323  
1324      return 0;
1325  }
1326  
1327  /*
1328   * Returns true if writing to the cluster pointed to by @l2_entry
1329   * requires a new allocation (that is, if the cluster is unallocated
1330   * or has refcount > 1 and therefore cannot be written in-place).
1331   */
1332  static bool GRAPH_RDLOCK
cluster_needs_new_alloc(BlockDriverState * bs,uint64_t l2_entry)1333  cluster_needs_new_alloc(BlockDriverState *bs, uint64_t l2_entry)
1334  {
1335      switch (qcow2_get_cluster_type(bs, l2_entry)) {
1336      case QCOW2_CLUSTER_NORMAL:
1337      case QCOW2_CLUSTER_ZERO_ALLOC:
1338          if (l2_entry & QCOW_OFLAG_COPIED) {
1339              return false;
1340          }
1341          /* fallthrough */
1342      case QCOW2_CLUSTER_UNALLOCATED:
1343      case QCOW2_CLUSTER_COMPRESSED:
1344      case QCOW2_CLUSTER_ZERO_PLAIN:
1345          return true;
1346      default:
1347          abort();
1348      }
1349  }
1350  
1351  /*
1352   * Returns the number of contiguous clusters that can be written to
1353   * using one single write request, starting from @l2_index.
1354   * At most @nb_clusters are checked.
1355   *
1356   * If @new_alloc is true this counts clusters that are either
1357   * unallocated, or allocated but with refcount > 1 (so they need to be
1358   * newly allocated and COWed).
1359   *
1360   * If @new_alloc is false this counts clusters that are already
1361   * allocated and can be overwritten in-place (this includes clusters
1362   * of type QCOW2_CLUSTER_ZERO_ALLOC).
1363   */
1364  static int GRAPH_RDLOCK
count_single_write_clusters(BlockDriverState * bs,int nb_clusters,uint64_t * l2_slice,int l2_index,bool new_alloc)1365  count_single_write_clusters(BlockDriverState *bs, int nb_clusters,
1366                              uint64_t *l2_slice, int l2_index, bool new_alloc)
1367  {
1368      BDRVQcow2State *s = bs->opaque;
1369      uint64_t l2_entry = get_l2_entry(s, l2_slice, l2_index);
1370      uint64_t expected_offset = l2_entry & L2E_OFFSET_MASK;
1371      int i;
1372  
1373      for (i = 0; i < nb_clusters; i++) {
1374          l2_entry = get_l2_entry(s, l2_slice, l2_index + i);
1375          if (cluster_needs_new_alloc(bs, l2_entry) != new_alloc) {
1376              break;
1377          }
1378          if (!new_alloc) {
1379              if (expected_offset != (l2_entry & L2E_OFFSET_MASK)) {
1380                  break;
1381              }
1382              expected_offset += s->cluster_size;
1383          }
1384      }
1385  
1386      assert(i <= nb_clusters);
1387      return i;
1388  }
1389  
1390  /*
1391   * Check if there already is an AIO write request in flight which allocates
1392   * the same cluster. In this case we need to wait until the previous
1393   * request has completed and updated the L2 table accordingly.
1394   *
1395   * Returns:
1396   *   0       if there was no dependency. *cur_bytes indicates the number of
1397   *           bytes from guest_offset that can be read before the next
1398   *           dependency must be processed (or the request is complete)
1399   *
1400   *   -EAGAIN if we had to wait for another request, previously gathered
1401   *           information on cluster allocation may be invalid now. The caller
1402   *           must start over anyway, so consider *cur_bytes undefined.
1403   */
handle_dependencies(BlockDriverState * bs,uint64_t guest_offset,uint64_t * cur_bytes,QCowL2Meta ** m)1404  static int coroutine_fn handle_dependencies(BlockDriverState *bs,
1405                                              uint64_t guest_offset,
1406                                              uint64_t *cur_bytes, QCowL2Meta **m)
1407  {
1408      BDRVQcow2State *s = bs->opaque;
1409      QCowL2Meta *old_alloc;
1410      uint64_t bytes = *cur_bytes;
1411  
1412      QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
1413  
1414          uint64_t start = guest_offset;
1415          uint64_t end = start + bytes;
1416          uint64_t old_start = start_of_cluster(s, l2meta_cow_start(old_alloc));
1417          uint64_t old_end = ROUND_UP(l2meta_cow_end(old_alloc), s->cluster_size);
1418  
1419          if (end <= old_start || start >= old_end) {
1420              /* No intersection */
1421              continue;
1422          }
1423  
1424          if (old_alloc->keep_old_clusters &&
1425              (end <= l2meta_cow_start(old_alloc) ||
1426               start >= l2meta_cow_end(old_alloc)))
1427          {
1428              /*
1429               * Clusters intersect but COW areas don't. And cluster itself is
1430               * already allocated. So, there is no actual conflict.
1431               */
1432              continue;
1433          }
1434  
1435          /* Conflict */
1436  
1437          if (start < old_start) {
1438              /* Stop at the start of a running allocation */
1439              bytes = old_start - start;
1440          } else {
1441              bytes = 0;
1442          }
1443  
1444          /*
1445           * Stop if an l2meta already exists. After yielding, it wouldn't
1446           * be valid any more, so we'd have to clean up the old L2Metas
1447           * and deal with requests depending on them before starting to
1448           * gather new ones. Not worth the trouble.
1449           */
1450          if (bytes == 0 && *m) {
1451              *cur_bytes = 0;
1452              return 0;
1453          }
1454  
1455          if (bytes == 0) {
1456              /*
1457               * Wait for the dependency to complete. We need to recheck
1458               * the free/allocated clusters when we continue.
1459               */
1460              qemu_co_queue_wait(&old_alloc->dependent_requests, &s->lock);
1461              return -EAGAIN;
1462          }
1463      }
1464  
1465      /* Make sure that existing clusters and new allocations are only used up to
1466       * the next dependency if we shortened the request above */
1467      *cur_bytes = bytes;
1468  
1469      return 0;
1470  }
1471  
1472  /*
1473   * Checks how many already allocated clusters that don't require a new
1474   * allocation there are at the given guest_offset (up to *bytes).
1475   * If *host_offset is not INV_OFFSET, only physically contiguous clusters
1476   * beginning at this host offset are counted.
1477   *
1478   * Note that guest_offset may not be cluster aligned. In this case, the
1479   * returned *host_offset points to exact byte referenced by guest_offset and
1480   * therefore isn't cluster aligned as well.
1481   *
1482   * Returns:
1483   *   0:     if no allocated clusters are available at the given offset.
1484   *          *bytes is normally unchanged. It is set to 0 if the cluster
1485   *          is allocated and can be overwritten in-place but doesn't have
1486   *          the right physical offset.
1487   *
1488   *   1:     if allocated clusters that can be overwritten in place are
1489   *          available at the requested offset. *bytes may have decreased
1490   *          and describes the length of the area that can be written to.
1491   *
1492   *  -errno: in error cases
1493   */
1494  static int coroutine_fn GRAPH_RDLOCK
handle_copied(BlockDriverState * bs,uint64_t guest_offset,uint64_t * host_offset,uint64_t * bytes,QCowL2Meta ** m)1495  handle_copied(BlockDriverState *bs, uint64_t guest_offset,
1496                uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
1497  {
1498      BDRVQcow2State *s = bs->opaque;
1499      int l2_index;
1500      uint64_t l2_entry, cluster_offset;
1501      uint64_t *l2_slice;
1502      uint64_t nb_clusters;
1503      unsigned int keep_clusters;
1504      int ret;
1505  
1506      trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset,
1507                                *bytes);
1508  
1509      assert(*host_offset == INV_OFFSET || offset_into_cluster(s, guest_offset)
1510                                        == offset_into_cluster(s, *host_offset));
1511  
1512      /*
1513       * Calculate the number of clusters to look for. We stop at L2 slice
1514       * boundaries to keep things simple.
1515       */
1516      nb_clusters =
1517          size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
1518  
1519      l2_index = offset_to_l2_slice_index(s, guest_offset);
1520      nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index);
1521      /* Limit total byte count to BDRV_REQUEST_MAX_BYTES */
1522      nb_clusters = MIN(nb_clusters, BDRV_REQUEST_MAX_BYTES >> s->cluster_bits);
1523  
1524      /* Find L2 entry for the first involved cluster */
1525      ret = get_cluster_table(bs, guest_offset, &l2_slice, &l2_index);
1526      if (ret < 0) {
1527          return ret;
1528      }
1529  
1530      l2_entry = get_l2_entry(s, l2_slice, l2_index);
1531      cluster_offset = l2_entry & L2E_OFFSET_MASK;
1532  
1533      if (!cluster_needs_new_alloc(bs, l2_entry)) {
1534          if (offset_into_cluster(s, cluster_offset)) {
1535              qcow2_signal_corruption(bs, true, -1, -1, "%s cluster offset "
1536                                      "%#" PRIx64 " unaligned (guest offset: %#"
1537                                      PRIx64 ")", l2_entry & QCOW_OFLAG_ZERO ?
1538                                      "Preallocated zero" : "Data",
1539                                      cluster_offset, guest_offset);
1540              ret = -EIO;
1541              goto out;
1542          }
1543  
1544          /* If a specific host_offset is required, check it */
1545          if (*host_offset != INV_OFFSET && cluster_offset != *host_offset) {
1546              *bytes = 0;
1547              ret = 0;
1548              goto out;
1549          }
1550  
1551          /* We keep all QCOW_OFLAG_COPIED clusters */
1552          keep_clusters = count_single_write_clusters(bs, nb_clusters, l2_slice,
1553                                                      l2_index, false);
1554          assert(keep_clusters <= nb_clusters);
1555  
1556          *bytes = MIN(*bytes,
1557                   keep_clusters * s->cluster_size
1558                   - offset_into_cluster(s, guest_offset));
1559          assert(*bytes != 0);
1560  
1561          ret = calculate_l2_meta(bs, cluster_offset, guest_offset,
1562                                  *bytes, l2_slice, m, true);
1563          if (ret < 0) {
1564              goto out;
1565          }
1566  
1567          ret = 1;
1568      } else {
1569          ret = 0;
1570      }
1571  
1572      /* Cleanup */
1573  out:
1574      qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
1575  
1576      /* Only return a host offset if we actually made progress. Otherwise we
1577       * would make requirements for handle_alloc() that it can't fulfill */
1578      if (ret > 0) {
1579          *host_offset = cluster_offset + offset_into_cluster(s, guest_offset);
1580      }
1581  
1582      return ret;
1583  }
1584  
1585  /*
1586   * Allocates new clusters for the given guest_offset.
1587   *
1588   * At most *nb_clusters are allocated, and on return *nb_clusters is updated to
1589   * contain the number of clusters that have been allocated and are contiguous
1590   * in the image file.
1591   *
1592   * If *host_offset is not INV_OFFSET, it specifies the offset in the image file
1593   * at which the new clusters must start. *nb_clusters can be 0 on return in
1594   * this case if the cluster at host_offset is already in use. If *host_offset
1595   * is INV_OFFSET, the clusters can be allocated anywhere in the image file.
1596   *
1597   * *host_offset is updated to contain the offset into the image file at which
1598   * the first allocated cluster starts.
1599   *
1600   * Return 0 on success and -errno in error cases. -EAGAIN means that the
1601   * function has been waiting for another request and the allocation must be
1602   * restarted, but the whole request should not be failed.
1603   */
1604  static int coroutine_fn GRAPH_RDLOCK
do_alloc_cluster_offset(BlockDriverState * bs,uint64_t guest_offset,uint64_t * host_offset,uint64_t * nb_clusters)1605  do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset,
1606                          uint64_t *host_offset, uint64_t *nb_clusters)
1607  {
1608      BDRVQcow2State *s = bs->opaque;
1609  
1610      trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset,
1611                                           *host_offset, *nb_clusters);
1612  
1613      if (has_data_file(bs)) {
1614          assert(*host_offset == INV_OFFSET ||
1615                 *host_offset == start_of_cluster(s, guest_offset));
1616          *host_offset = start_of_cluster(s, guest_offset);
1617          return 0;
1618      }
1619  
1620      /* Allocate new clusters */
1621      trace_qcow2_cluster_alloc_phys(qemu_coroutine_self());
1622      if (*host_offset == INV_OFFSET) {
1623          int64_t cluster_offset =
1624              qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size);
1625          if (cluster_offset < 0) {
1626              return cluster_offset;
1627          }
1628          *host_offset = cluster_offset;
1629          return 0;
1630      } else {
1631          int64_t ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters);
1632          if (ret < 0) {
1633              return ret;
1634          }
1635          *nb_clusters = ret;
1636          return 0;
1637      }
1638  }
1639  
1640  /*
1641   * Allocates new clusters for an area that is either still unallocated or
1642   * cannot be overwritten in-place. If *host_offset is not INV_OFFSET,
1643   * clusters are only allocated if the new allocation can match the specified
1644   * host offset.
1645   *
1646   * Note that guest_offset may not be cluster aligned. In this case, the
1647   * returned *host_offset points to exact byte referenced by guest_offset and
1648   * therefore isn't cluster aligned as well.
1649   *
1650   * Returns:
1651   *   0:     if no clusters could be allocated. *bytes is set to 0,
1652   *          *host_offset is left unchanged.
1653   *
1654   *   1:     if new clusters were allocated. *bytes may be decreased if the
1655   *          new allocation doesn't cover all of the requested area.
1656   *          *host_offset is updated to contain the host offset of the first
1657   *          newly allocated cluster.
1658   *
1659   *  -errno: in error cases
1660   */
1661  static int coroutine_fn GRAPH_RDLOCK
handle_alloc(BlockDriverState * bs,uint64_t guest_offset,uint64_t * host_offset,uint64_t * bytes,QCowL2Meta ** m)1662  handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
1663               uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m)
1664  {
1665      BDRVQcow2State *s = bs->opaque;
1666      int l2_index;
1667      uint64_t *l2_slice;
1668      uint64_t nb_clusters;
1669      int ret;
1670  
1671      uint64_t alloc_cluster_offset;
1672  
1673      trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset,
1674                               *bytes);
1675      assert(*bytes > 0);
1676  
1677      /*
1678       * Calculate the number of clusters to look for. We stop at L2 slice
1679       * boundaries to keep things simple.
1680       */
1681      nb_clusters =
1682          size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes);
1683  
1684      l2_index = offset_to_l2_slice_index(s, guest_offset);
1685      nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index);
1686      /* Limit total allocation byte count to BDRV_REQUEST_MAX_BYTES */
1687      nb_clusters = MIN(nb_clusters, BDRV_REQUEST_MAX_BYTES >> s->cluster_bits);
1688  
1689      /* Find L2 entry for the first involved cluster */
1690      ret = get_cluster_table(bs, guest_offset, &l2_slice, &l2_index);
1691      if (ret < 0) {
1692          return ret;
1693      }
1694  
1695      nb_clusters = count_single_write_clusters(bs, nb_clusters,
1696                                                l2_slice, l2_index, true);
1697  
1698      /* This function is only called when there were no non-COW clusters, so if
1699       * we can't find any unallocated or COW clusters either, something is
1700       * wrong with our code. */
1701      assert(nb_clusters > 0);
1702  
1703      /* Allocate at a given offset in the image file */
1704      alloc_cluster_offset = *host_offset == INV_OFFSET ? INV_OFFSET :
1705          start_of_cluster(s, *host_offset);
1706      ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset,
1707                                    &nb_clusters);
1708      if (ret < 0) {
1709          goto out;
1710      }
1711  
1712      /* Can't extend contiguous allocation */
1713      if (nb_clusters == 0) {
1714          *bytes = 0;
1715          ret = 0;
1716          goto out;
1717      }
1718  
1719      assert(alloc_cluster_offset != INV_OFFSET);
1720  
1721      /*
1722       * Save info needed for meta data update.
1723       *
1724       * requested_bytes: Number of bytes from the start of the first
1725       * newly allocated cluster to the end of the (possibly shortened
1726       * before) write request.
1727       *
1728       * avail_bytes: Number of bytes from the start of the first
1729       * newly allocated to the end of the last newly allocated cluster.
1730       *
1731       * nb_bytes: The number of bytes from the start of the first
1732       * newly allocated cluster to the end of the area that the write
1733       * request actually writes to (excluding COW at the end)
1734       */
1735      uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset);
1736      int avail_bytes = nb_clusters << s->cluster_bits;
1737      int nb_bytes = MIN(requested_bytes, avail_bytes);
1738  
1739      *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset);
1740      *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset));
1741      assert(*bytes != 0);
1742  
1743      ret = calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes,
1744                              l2_slice, m, false);
1745      if (ret < 0) {
1746          goto out;
1747      }
1748  
1749      ret = 1;
1750  
1751  out:
1752      qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
1753      return ret;
1754  }
1755  
1756  /*
1757   * For a given area on the virtual disk defined by @offset and @bytes,
1758   * find the corresponding area on the qcow2 image, allocating new
1759   * clusters (or subclusters) if necessary. The result can span a
1760   * combination of allocated and previously unallocated clusters.
1761   *
1762   * Note that offset may not be cluster aligned. In this case, the returned
1763   * *host_offset points to exact byte referenced by offset and therefore
1764   * isn't cluster aligned as well.
1765   *
1766   * On return, @host_offset is set to the beginning of the requested
1767   * area. This area is guaranteed to be contiguous on the qcow2 file
1768   * but it can be smaller than initially requested. In this case @bytes
1769   * is updated with the actual size.
1770   *
1771   * If any clusters or subclusters were allocated then @m contains a
1772   * list with the information of all the affected regions. Note that
1773   * this can happen regardless of whether this function succeeds or
1774   * not. The caller is responsible for updating the L2 metadata of the
1775   * allocated clusters (on success) or freeing them (on failure), and
1776   * for clearing the contents of @m afterwards in both cases.
1777   *
1778   * If the request conflicts with another write request in flight, the coroutine
1779   * is queued and will be reentered when the dependency has completed.
1780   *
1781   * Return 0 on success and -errno in error cases
1782   */
qcow2_alloc_host_offset(BlockDriverState * bs,uint64_t offset,unsigned int * bytes,uint64_t * host_offset,QCowL2Meta ** m)1783  int coroutine_fn qcow2_alloc_host_offset(BlockDriverState *bs, uint64_t offset,
1784                                           unsigned int *bytes,
1785                                           uint64_t *host_offset,
1786                                           QCowL2Meta **m)
1787  {
1788      BDRVQcow2State *s = bs->opaque;
1789      uint64_t start, remaining;
1790      uint64_t cluster_offset;
1791      uint64_t cur_bytes;
1792      int ret;
1793  
1794      trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, *bytes);
1795  
1796  again:
1797      start = offset;
1798      remaining = *bytes;
1799      cluster_offset = INV_OFFSET;
1800      *host_offset = INV_OFFSET;
1801      cur_bytes = 0;
1802      *m = NULL;
1803  
1804      while (true) {
1805  
1806          if (*host_offset == INV_OFFSET && cluster_offset != INV_OFFSET) {
1807              *host_offset = cluster_offset;
1808          }
1809  
1810          assert(remaining >= cur_bytes);
1811  
1812          start           += cur_bytes;
1813          remaining       -= cur_bytes;
1814  
1815          if (cluster_offset != INV_OFFSET) {
1816              cluster_offset += cur_bytes;
1817          }
1818  
1819          if (remaining == 0) {
1820              break;
1821          }
1822  
1823          cur_bytes = remaining;
1824  
1825          /*
1826           * Now start gathering as many contiguous clusters as possible:
1827           *
1828           * 1. Check for overlaps with in-flight allocations
1829           *
1830           *      a) Overlap not in the first cluster -> shorten this request and
1831           *         let the caller handle the rest in its next loop iteration.
1832           *
1833           *      b) Real overlaps of two requests. Yield and restart the search
1834           *         for contiguous clusters (the situation could have changed
1835           *         while we were sleeping)
1836           *
1837           *      c) TODO: Request starts in the same cluster as the in-flight
1838           *         allocation ends. Shorten the COW of the in-fight allocation,
1839           *         set cluster_offset to write to the same cluster and set up
1840           *         the right synchronisation between the in-flight request and
1841           *         the new one.
1842           */
1843          ret = handle_dependencies(bs, start, &cur_bytes, m);
1844          if (ret == -EAGAIN) {
1845              /* Currently handle_dependencies() doesn't yield if we already had
1846               * an allocation. If it did, we would have to clean up the L2Meta
1847               * structs before starting over. */
1848              assert(*m == NULL);
1849              goto again;
1850          } else if (ret < 0) {
1851              return ret;
1852          } else if (cur_bytes == 0) {
1853              break;
1854          } else {
1855              /* handle_dependencies() may have decreased cur_bytes (shortened
1856               * the allocations below) so that the next dependency is processed
1857               * correctly during the next loop iteration. */
1858          }
1859  
1860          /*
1861           * 2. Count contiguous COPIED clusters.
1862           */
1863          ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m);
1864          if (ret < 0) {
1865              return ret;
1866          } else if (ret) {
1867              continue;
1868          } else if (cur_bytes == 0) {
1869              break;
1870          }
1871  
1872          /*
1873           * 3. If the request still hasn't completed, allocate new clusters,
1874           *    considering any cluster_offset of steps 1c or 2.
1875           */
1876          ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m);
1877          if (ret < 0) {
1878              return ret;
1879          } else if (ret) {
1880              continue;
1881          } else {
1882              assert(cur_bytes == 0);
1883              break;
1884          }
1885      }
1886  
1887      *bytes -= remaining;
1888      assert(*bytes > 0);
1889      assert(*host_offset != INV_OFFSET);
1890      assert(offset_into_cluster(s, *host_offset) ==
1891             offset_into_cluster(s, offset));
1892  
1893      return 0;
1894  }
1895  
1896  /*
1897   * This discards as many clusters of nb_clusters as possible at once (i.e.
1898   * all clusters in the same L2 slice) and returns the number of discarded
1899   * clusters.
1900   */
1901  static int GRAPH_RDLOCK
discard_in_l2_slice(BlockDriverState * bs,uint64_t offset,uint64_t nb_clusters,enum qcow2_discard_type type,bool full_discard)1902  discard_in_l2_slice(BlockDriverState *bs, uint64_t offset, uint64_t nb_clusters,
1903                      enum qcow2_discard_type type, bool full_discard)
1904  {
1905      BDRVQcow2State *s = bs->opaque;
1906      uint64_t *l2_slice;
1907      int l2_index;
1908      int ret;
1909      int i;
1910  
1911      ret = get_cluster_table(bs, offset, &l2_slice, &l2_index);
1912      if (ret < 0) {
1913          return ret;
1914      }
1915  
1916      /* Limit nb_clusters to one L2 slice */
1917      nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index);
1918      assert(nb_clusters <= INT_MAX);
1919  
1920      for (i = 0; i < nb_clusters; i++) {
1921          uint64_t old_l2_entry = get_l2_entry(s, l2_slice, l2_index + i);
1922          uint64_t old_l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i);
1923          uint64_t new_l2_entry = old_l2_entry;
1924          uint64_t new_l2_bitmap = old_l2_bitmap;
1925          QCow2ClusterType cluster_type =
1926              qcow2_get_cluster_type(bs, old_l2_entry);
1927          bool keep_reference = (cluster_type != QCOW2_CLUSTER_COMPRESSED) &&
1928                                !full_discard &&
1929                                (s->discard_no_unref &&
1930                                 type == QCOW2_DISCARD_REQUEST);
1931  
1932          /*
1933           * If full_discard is true, the cluster should not read back as zeroes,
1934           * but rather fall through to the backing file.
1935           *
1936           * If full_discard is false, make sure that a discarded area reads back
1937           * as zeroes for v3 images (we cannot do it for v2 without actually
1938           * writing a zero-filled buffer). We can skip the operation if the
1939           * cluster is already marked as zero, or if it's unallocated and we
1940           * don't have a backing file.
1941           *
1942           * TODO We might want to use bdrv_block_status(bs) here, but we're
1943           * holding s->lock, so that doesn't work today.
1944           */
1945          if (full_discard) {
1946              new_l2_entry = new_l2_bitmap = 0;
1947          } else if (bs->backing || qcow2_cluster_is_allocated(cluster_type)) {
1948              if (has_subclusters(s)) {
1949                  if (keep_reference) {
1950                      new_l2_entry = old_l2_entry;
1951                  } else {
1952                      new_l2_entry = 0;
1953                  }
1954                  new_l2_bitmap = QCOW_L2_BITMAP_ALL_ZEROES;
1955              } else {
1956                  if (s->qcow_version >= 3) {
1957                      if (keep_reference) {
1958                          new_l2_entry |= QCOW_OFLAG_ZERO;
1959                      } else {
1960                          new_l2_entry = QCOW_OFLAG_ZERO;
1961                      }
1962                  } else {
1963                      new_l2_entry = 0;
1964                  }
1965              }
1966          }
1967  
1968          if (old_l2_entry == new_l2_entry && old_l2_bitmap == new_l2_bitmap) {
1969              continue;
1970          }
1971  
1972          /* First remove L2 entries */
1973          qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
1974          set_l2_entry(s, l2_slice, l2_index + i, new_l2_entry);
1975          if (has_subclusters(s)) {
1976              set_l2_bitmap(s, l2_slice, l2_index + i, new_l2_bitmap);
1977          }
1978          if (!keep_reference) {
1979              /* Then decrease the refcount */
1980              qcow2_free_any_cluster(bs, old_l2_entry, type);
1981          } else if (s->discard_passthrough[type] &&
1982                     (cluster_type == QCOW2_CLUSTER_NORMAL ||
1983                      cluster_type == QCOW2_CLUSTER_ZERO_ALLOC)) {
1984              /* If we keep the reference, pass on the discard still */
1985              bdrv_pdiscard(s->data_file, old_l2_entry & L2E_OFFSET_MASK,
1986                            s->cluster_size);
1987          }
1988      }
1989  
1990      qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
1991  
1992      return nb_clusters;
1993  }
1994  
qcow2_cluster_discard(BlockDriverState * bs,uint64_t offset,uint64_t bytes,enum qcow2_discard_type type,bool full_discard)1995  int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset,
1996                            uint64_t bytes, enum qcow2_discard_type type,
1997                            bool full_discard)
1998  {
1999      BDRVQcow2State *s = bs->opaque;
2000      uint64_t end_offset = offset + bytes;
2001      uint64_t nb_clusters;
2002      int64_t cleared;
2003      int ret;
2004  
2005      /* Caller must pass aligned values, except at image end */
2006      assert(QEMU_IS_ALIGNED(offset, s->cluster_size));
2007      assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) ||
2008             end_offset == bs->total_sectors << BDRV_SECTOR_BITS);
2009  
2010      nb_clusters = size_to_clusters(s, bytes);
2011  
2012      s->cache_discards = true;
2013  
2014      /* Each L2 slice is handled by its own loop iteration */
2015      while (nb_clusters > 0) {
2016          cleared = discard_in_l2_slice(bs, offset, nb_clusters, type,
2017                                        full_discard);
2018          if (cleared < 0) {
2019              ret = cleared;
2020              goto fail;
2021          }
2022  
2023          nb_clusters -= cleared;
2024          offset += (cleared * s->cluster_size);
2025      }
2026  
2027      ret = 0;
2028  fail:
2029      s->cache_discards = false;
2030      qcow2_process_discards(bs, ret);
2031  
2032      return ret;
2033  }
2034  
2035  /*
2036   * This zeroes as many clusters of nb_clusters as possible at once (i.e.
2037   * all clusters in the same L2 slice) and returns the number of zeroed
2038   * clusters.
2039   */
2040  static int coroutine_fn GRAPH_RDLOCK
zero_in_l2_slice(BlockDriverState * bs,uint64_t offset,uint64_t nb_clusters,int flags)2041  zero_in_l2_slice(BlockDriverState *bs, uint64_t offset,
2042                   uint64_t nb_clusters, int flags)
2043  {
2044      BDRVQcow2State *s = bs->opaque;
2045      uint64_t *l2_slice;
2046      int l2_index;
2047      int ret;
2048      int i;
2049  
2050      ret = get_cluster_table(bs, offset, &l2_slice, &l2_index);
2051      if (ret < 0) {
2052          return ret;
2053      }
2054  
2055      /* Limit nb_clusters to one L2 slice */
2056      nb_clusters = MIN(nb_clusters, s->l2_slice_size - l2_index);
2057      assert(nb_clusters <= INT_MAX);
2058  
2059      for (i = 0; i < nb_clusters; i++) {
2060          uint64_t old_l2_entry = get_l2_entry(s, l2_slice, l2_index + i);
2061          uint64_t old_l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i);
2062          QCow2ClusterType type = qcow2_get_cluster_type(bs, old_l2_entry);
2063          bool unmap = (type == QCOW2_CLUSTER_COMPRESSED) ||
2064              ((flags & BDRV_REQ_MAY_UNMAP) && qcow2_cluster_is_allocated(type));
2065          bool keep_reference =
2066              (s->discard_no_unref && type != QCOW2_CLUSTER_COMPRESSED);
2067          uint64_t new_l2_entry = old_l2_entry;
2068          uint64_t new_l2_bitmap = old_l2_bitmap;
2069  
2070          if (unmap && !keep_reference) {
2071              new_l2_entry = 0;
2072          }
2073  
2074          if (has_subclusters(s)) {
2075              new_l2_bitmap = QCOW_L2_BITMAP_ALL_ZEROES;
2076          } else {
2077              new_l2_entry |= QCOW_OFLAG_ZERO;
2078          }
2079  
2080          if (old_l2_entry == new_l2_entry && old_l2_bitmap == new_l2_bitmap) {
2081              continue;
2082          }
2083  
2084          /* First update L2 entries */
2085          qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
2086          set_l2_entry(s, l2_slice, l2_index + i, new_l2_entry);
2087          if (has_subclusters(s)) {
2088              set_l2_bitmap(s, l2_slice, l2_index + i, new_l2_bitmap);
2089          }
2090  
2091          if (unmap) {
2092              if (!keep_reference) {
2093                  /* Then decrease the refcount */
2094                  qcow2_free_any_cluster(bs, old_l2_entry, QCOW2_DISCARD_REQUEST);
2095              } else if (s->discard_passthrough[QCOW2_DISCARD_REQUEST] &&
2096                         (type == QCOW2_CLUSTER_NORMAL ||
2097                          type == QCOW2_CLUSTER_ZERO_ALLOC)) {
2098                  /* If we keep the reference, pass on the discard still */
2099                  bdrv_pdiscard(s->data_file, old_l2_entry & L2E_OFFSET_MASK,
2100                              s->cluster_size);
2101              }
2102          }
2103      }
2104  
2105      qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
2106  
2107      return nb_clusters;
2108  }
2109  
2110  static int coroutine_fn GRAPH_RDLOCK
zero_l2_subclusters(BlockDriverState * bs,uint64_t offset,unsigned nb_subclusters)2111  zero_l2_subclusters(BlockDriverState *bs, uint64_t offset,
2112                      unsigned nb_subclusters)
2113  {
2114      BDRVQcow2State *s = bs->opaque;
2115      uint64_t *l2_slice;
2116      uint64_t old_l2_bitmap, l2_bitmap;
2117      int l2_index, ret, sc = offset_to_sc_index(s, offset);
2118  
2119      /* For full clusters use zero_in_l2_slice() instead */
2120      assert(nb_subclusters > 0 && nb_subclusters < s->subclusters_per_cluster);
2121      assert(sc + nb_subclusters <= s->subclusters_per_cluster);
2122      assert(offset_into_subcluster(s, offset) == 0);
2123  
2124      ret = get_cluster_table(bs, offset, &l2_slice, &l2_index);
2125      if (ret < 0) {
2126          return ret;
2127      }
2128  
2129      switch (qcow2_get_cluster_type(bs, get_l2_entry(s, l2_slice, l2_index))) {
2130      case QCOW2_CLUSTER_COMPRESSED:
2131          ret = -ENOTSUP; /* We cannot partially zeroize compressed clusters */
2132          goto out;
2133      case QCOW2_CLUSTER_NORMAL:
2134      case QCOW2_CLUSTER_UNALLOCATED:
2135          break;
2136      default:
2137          g_assert_not_reached();
2138      }
2139  
2140      old_l2_bitmap = l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index);
2141  
2142      l2_bitmap |=  QCOW_OFLAG_SUB_ZERO_RANGE(sc, sc + nb_subclusters);
2143      l2_bitmap &= ~QCOW_OFLAG_SUB_ALLOC_RANGE(sc, sc + nb_subclusters);
2144  
2145      if (old_l2_bitmap != l2_bitmap) {
2146          set_l2_bitmap(s, l2_slice, l2_index, l2_bitmap);
2147          qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
2148      }
2149  
2150      ret = 0;
2151  out:
2152      qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
2153  
2154      return ret;
2155  }
2156  
qcow2_subcluster_zeroize(BlockDriverState * bs,uint64_t offset,uint64_t bytes,int flags)2157  int coroutine_fn qcow2_subcluster_zeroize(BlockDriverState *bs, uint64_t offset,
2158                                            uint64_t bytes, int flags)
2159  {
2160      BDRVQcow2State *s = bs->opaque;
2161      uint64_t end_offset = offset + bytes;
2162      uint64_t nb_clusters;
2163      unsigned head, tail;
2164      int64_t cleared;
2165      int ret;
2166  
2167      /* If we have to stay in sync with an external data file, zero out
2168       * s->data_file first. */
2169      if (data_file_is_raw(bs)) {
2170          assert(has_data_file(bs));
2171          ret = bdrv_co_pwrite_zeroes(s->data_file, offset, bytes, flags);
2172          if (ret < 0) {
2173              return ret;
2174          }
2175      }
2176  
2177      /* Caller must pass aligned values, except at image end */
2178      assert(offset_into_subcluster(s, offset) == 0);
2179      assert(offset_into_subcluster(s, end_offset) == 0 ||
2180             end_offset >= bs->total_sectors << BDRV_SECTOR_BITS);
2181  
2182      /*
2183       * The zero flag is only supported by version 3 and newer. However, if we
2184       * have no backing file, we can resort to discard in version 2.
2185       */
2186      if (s->qcow_version < 3) {
2187          if (!bs->backing) {
2188              return qcow2_cluster_discard(bs, offset, bytes,
2189                                           QCOW2_DISCARD_REQUEST, false);
2190          }
2191          return -ENOTSUP;
2192      }
2193  
2194      head = MIN(end_offset, ROUND_UP(offset, s->cluster_size)) - offset;
2195      offset += head;
2196  
2197      tail = (end_offset >= bs->total_sectors << BDRV_SECTOR_BITS) ? 0 :
2198          end_offset - MAX(offset, start_of_cluster(s, end_offset));
2199      end_offset -= tail;
2200  
2201      s->cache_discards = true;
2202  
2203      if (head) {
2204          ret = zero_l2_subclusters(bs, offset - head,
2205                                    size_to_subclusters(s, head));
2206          if (ret < 0) {
2207              goto fail;
2208          }
2209      }
2210  
2211      /* Each L2 slice is handled by its own loop iteration */
2212      nb_clusters = size_to_clusters(s, end_offset - offset);
2213  
2214      while (nb_clusters > 0) {
2215          cleared = zero_in_l2_slice(bs, offset, nb_clusters, flags);
2216          if (cleared < 0) {
2217              ret = cleared;
2218              goto fail;
2219          }
2220  
2221          nb_clusters -= cleared;
2222          offset += (cleared * s->cluster_size);
2223      }
2224  
2225      if (tail) {
2226          ret = zero_l2_subclusters(bs, end_offset, size_to_subclusters(s, tail));
2227          if (ret < 0) {
2228              goto fail;
2229          }
2230      }
2231  
2232      ret = 0;
2233  fail:
2234      s->cache_discards = false;
2235      qcow2_process_discards(bs, ret);
2236  
2237      return ret;
2238  }
2239  
2240  /*
2241   * Expands all zero clusters in a specific L1 table (or deallocates them, for
2242   * non-backed non-pre-allocated zero clusters).
2243   *
2244   * l1_entries and *visited_l1_entries are used to keep track of progress for
2245   * status_cb(). l1_entries contains the total number of L1 entries and
2246   * *visited_l1_entries counts all visited L1 entries.
2247   */
2248  static int GRAPH_RDLOCK
expand_zero_clusters_in_l1(BlockDriverState * bs,uint64_t * l1_table,int l1_size,int64_t * visited_l1_entries,int64_t l1_entries,BlockDriverAmendStatusCB * status_cb,void * cb_opaque)2249  expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
2250                             int l1_size, int64_t *visited_l1_entries,
2251                             int64_t l1_entries,
2252                             BlockDriverAmendStatusCB *status_cb,
2253                             void *cb_opaque)
2254  {
2255      BDRVQcow2State *s = bs->opaque;
2256      bool is_active_l1 = (l1_table == s->l1_table);
2257      uint64_t *l2_slice = NULL;
2258      unsigned slice, slice_size2, n_slices;
2259      int ret;
2260      int i, j;
2261  
2262      /* qcow2_downgrade() is not allowed in images with subclusters */
2263      assert(!has_subclusters(s));
2264  
2265      slice_size2 = s->l2_slice_size * l2_entry_size(s);
2266      n_slices = s->cluster_size / slice_size2;
2267  
2268      if (!is_active_l1) {
2269          /* inactive L2 tables require a buffer to be stored in when loading
2270           * them from disk */
2271          l2_slice = qemu_try_blockalign(bs->file->bs, slice_size2);
2272          if (l2_slice == NULL) {
2273              return -ENOMEM;
2274          }
2275      }
2276  
2277      for (i = 0; i < l1_size; i++) {
2278          uint64_t l2_offset = l1_table[i] & L1E_OFFSET_MASK;
2279          uint64_t l2_refcount;
2280  
2281          if (!l2_offset) {
2282              /* unallocated */
2283              (*visited_l1_entries)++;
2284              if (status_cb) {
2285                  status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque);
2286              }
2287              continue;
2288          }
2289  
2290          if (offset_into_cluster(s, l2_offset)) {
2291              qcow2_signal_corruption(bs, true, -1, -1, "L2 table offset %#"
2292                                      PRIx64 " unaligned (L1 index: %#x)",
2293                                      l2_offset, i);
2294              ret = -EIO;
2295              goto fail;
2296          }
2297  
2298          ret = qcow2_get_refcount(bs, l2_offset >> s->cluster_bits,
2299                                   &l2_refcount);
2300          if (ret < 0) {
2301              goto fail;
2302          }
2303  
2304          for (slice = 0; slice < n_slices; slice++) {
2305              uint64_t slice_offset = l2_offset + slice * slice_size2;
2306              bool l2_dirty = false;
2307              if (is_active_l1) {
2308                  /* get active L2 tables from cache */
2309                  ret = qcow2_cache_get(bs, s->l2_table_cache, slice_offset,
2310                                        (void **)&l2_slice);
2311              } else {
2312                  /* load inactive L2 tables from disk */
2313                  ret = bdrv_pread(bs->file, slice_offset, slice_size2,
2314                                   l2_slice, 0);
2315              }
2316              if (ret < 0) {
2317                  goto fail;
2318              }
2319  
2320              for (j = 0; j < s->l2_slice_size; j++) {
2321                  uint64_t l2_entry = get_l2_entry(s, l2_slice, j);
2322                  int64_t offset = l2_entry & L2E_OFFSET_MASK;
2323                  QCow2ClusterType cluster_type =
2324                      qcow2_get_cluster_type(bs, l2_entry);
2325  
2326                  if (cluster_type != QCOW2_CLUSTER_ZERO_PLAIN &&
2327                      cluster_type != QCOW2_CLUSTER_ZERO_ALLOC) {
2328                      continue;
2329                  }
2330  
2331                  if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) {
2332                      if (!bs->backing) {
2333                          /*
2334                           * not backed; therefore we can simply deallocate the
2335                           * cluster. No need to call set_l2_bitmap(), this
2336                           * function doesn't support images with subclusters.
2337                           */
2338                          set_l2_entry(s, l2_slice, j, 0);
2339                          l2_dirty = true;
2340                          continue;
2341                      }
2342  
2343                      offset = qcow2_alloc_clusters(bs, s->cluster_size);
2344                      if (offset < 0) {
2345                          ret = offset;
2346                          goto fail;
2347                      }
2348  
2349                      /* The offset must fit in the offset field */
2350                      assert((offset & L2E_OFFSET_MASK) == offset);
2351  
2352                      if (l2_refcount > 1) {
2353                          /* For shared L2 tables, set the refcount accordingly
2354                           * (it is already 1 and needs to be l2_refcount) */
2355                          ret = qcow2_update_cluster_refcount(
2356                              bs, offset >> s->cluster_bits,
2357                              refcount_diff(1, l2_refcount), false,
2358                              QCOW2_DISCARD_OTHER);
2359                          if (ret < 0) {
2360                              qcow2_free_clusters(bs, offset, s->cluster_size,
2361                                                  QCOW2_DISCARD_OTHER);
2362                              goto fail;
2363                          }
2364                      }
2365                  }
2366  
2367                  if (offset_into_cluster(s, offset)) {
2368                      int l2_index = slice * s->l2_slice_size + j;
2369                      qcow2_signal_corruption(
2370                          bs, true, -1, -1,
2371                          "Cluster allocation offset "
2372                          "%#" PRIx64 " unaligned (L2 offset: %#"
2373                          PRIx64 ", L2 index: %#x)", offset,
2374                          l2_offset, l2_index);
2375                      if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) {
2376                          qcow2_free_clusters(bs, offset, s->cluster_size,
2377                                              QCOW2_DISCARD_ALWAYS);
2378                      }
2379                      ret = -EIO;
2380                      goto fail;
2381                  }
2382  
2383                  ret = qcow2_pre_write_overlap_check(bs, 0, offset,
2384                                                      s->cluster_size, true);
2385                  if (ret < 0) {
2386                      if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) {
2387                          qcow2_free_clusters(bs, offset, s->cluster_size,
2388                                              QCOW2_DISCARD_ALWAYS);
2389                      }
2390                      goto fail;
2391                  }
2392  
2393                  ret = bdrv_pwrite_zeroes(s->data_file, offset,
2394                                           s->cluster_size, 0);
2395                  if (ret < 0) {
2396                      if (cluster_type == QCOW2_CLUSTER_ZERO_PLAIN) {
2397                          qcow2_free_clusters(bs, offset, s->cluster_size,
2398                                              QCOW2_DISCARD_ALWAYS);
2399                      }
2400                      goto fail;
2401                  }
2402  
2403                  if (l2_refcount == 1) {
2404                      set_l2_entry(s, l2_slice, j, offset | QCOW_OFLAG_COPIED);
2405                  } else {
2406                      set_l2_entry(s, l2_slice, j, offset);
2407                  }
2408                  /*
2409                   * No need to call set_l2_bitmap() after set_l2_entry() because
2410                   * this function doesn't support images with subclusters.
2411                   */
2412                  l2_dirty = true;
2413              }
2414  
2415              if (is_active_l1) {
2416                  if (l2_dirty) {
2417                      qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
2418                      qcow2_cache_depends_on_flush(s->l2_table_cache);
2419                  }
2420                  qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
2421              } else {
2422                  if (l2_dirty) {
2423                      ret = qcow2_pre_write_overlap_check(
2424                          bs, QCOW2_OL_INACTIVE_L2 | QCOW2_OL_ACTIVE_L2,
2425                          slice_offset, slice_size2, false);
2426                      if (ret < 0) {
2427                          goto fail;
2428                      }
2429  
2430                      ret = bdrv_pwrite(bs->file, slice_offset, slice_size2,
2431                                        l2_slice, 0);
2432                      if (ret < 0) {
2433                          goto fail;
2434                      }
2435                  }
2436              }
2437          }
2438  
2439          (*visited_l1_entries)++;
2440          if (status_cb) {
2441              status_cb(bs, *visited_l1_entries, l1_entries, cb_opaque);
2442          }
2443      }
2444  
2445      ret = 0;
2446  
2447  fail:
2448      if (l2_slice) {
2449          if (!is_active_l1) {
2450              qemu_vfree(l2_slice);
2451          } else {
2452              qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
2453          }
2454      }
2455      return ret;
2456  }
2457  
2458  /*
2459   * For backed images, expands all zero clusters on the image. For non-backed
2460   * images, deallocates all non-pre-allocated zero clusters (and claims the
2461   * allocation for pre-allocated ones). This is important for downgrading to a
2462   * qcow2 version which doesn't yet support metadata zero clusters.
2463   */
qcow2_expand_zero_clusters(BlockDriverState * bs,BlockDriverAmendStatusCB * status_cb,void * cb_opaque)2464  int qcow2_expand_zero_clusters(BlockDriverState *bs,
2465                                 BlockDriverAmendStatusCB *status_cb,
2466                                 void *cb_opaque)
2467  {
2468      BDRVQcow2State *s = bs->opaque;
2469      uint64_t *l1_table = NULL;
2470      int64_t l1_entries = 0, visited_l1_entries = 0;
2471      int ret;
2472      int i, j;
2473  
2474      if (status_cb) {
2475          l1_entries = s->l1_size;
2476          for (i = 0; i < s->nb_snapshots; i++) {
2477              l1_entries += s->snapshots[i].l1_size;
2478          }
2479      }
2480  
2481      ret = expand_zero_clusters_in_l1(bs, s->l1_table, s->l1_size,
2482                                       &visited_l1_entries, l1_entries,
2483                                       status_cb, cb_opaque);
2484      if (ret < 0) {
2485          goto fail;
2486      }
2487  
2488      /* Inactive L1 tables may point to active L2 tables - therefore it is
2489       * necessary to flush the L2 table cache before trying to access the L2
2490       * tables pointed to by inactive L1 entries (else we might try to expand
2491       * zero clusters that have already been expanded); furthermore, it is also
2492       * necessary to empty the L2 table cache, since it may contain tables which
2493       * are now going to be modified directly on disk, bypassing the cache.
2494       * qcow2_cache_empty() does both for us. */
2495      ret = qcow2_cache_empty(bs, s->l2_table_cache);
2496      if (ret < 0) {
2497          goto fail;
2498      }
2499  
2500      for (i = 0; i < s->nb_snapshots; i++) {
2501          int l1_size2;
2502          uint64_t *new_l1_table;
2503          Error *local_err = NULL;
2504  
2505          ret = qcow2_validate_table(bs, s->snapshots[i].l1_table_offset,
2506                                     s->snapshots[i].l1_size, L1E_SIZE,
2507                                     QCOW_MAX_L1_SIZE, "Snapshot L1 table",
2508                                     &local_err);
2509          if (ret < 0) {
2510              error_report_err(local_err);
2511              goto fail;
2512          }
2513  
2514          l1_size2 = s->snapshots[i].l1_size * L1E_SIZE;
2515          new_l1_table = g_try_realloc(l1_table, l1_size2);
2516  
2517          if (!new_l1_table) {
2518              ret = -ENOMEM;
2519              goto fail;
2520          }
2521  
2522          l1_table = new_l1_table;
2523  
2524          ret = bdrv_pread(bs->file, s->snapshots[i].l1_table_offset, l1_size2,
2525                           l1_table, 0);
2526          if (ret < 0) {
2527              goto fail;
2528          }
2529  
2530          for (j = 0; j < s->snapshots[i].l1_size; j++) {
2531              be64_to_cpus(&l1_table[j]);
2532          }
2533  
2534          ret = expand_zero_clusters_in_l1(bs, l1_table, s->snapshots[i].l1_size,
2535                                           &visited_l1_entries, l1_entries,
2536                                           status_cb, cb_opaque);
2537          if (ret < 0) {
2538              goto fail;
2539          }
2540      }
2541  
2542      ret = 0;
2543  
2544  fail:
2545      g_free(l1_table);
2546      return ret;
2547  }
2548  
qcow2_parse_compressed_l2_entry(BlockDriverState * bs,uint64_t l2_entry,uint64_t * coffset,int * csize)2549  void qcow2_parse_compressed_l2_entry(BlockDriverState *bs, uint64_t l2_entry,
2550                                       uint64_t *coffset, int *csize)
2551  {
2552      BDRVQcow2State *s = bs->opaque;
2553      int nb_csectors;
2554  
2555      assert(qcow2_get_cluster_type(bs, l2_entry) == QCOW2_CLUSTER_COMPRESSED);
2556  
2557      *coffset = l2_entry & s->cluster_offset_mask;
2558  
2559      nb_csectors = ((l2_entry >> s->csize_shift) & s->csize_mask) + 1;
2560      *csize = nb_csectors * QCOW2_COMPRESSED_SECTOR_SIZE -
2561          (*coffset & (QCOW2_COMPRESSED_SECTOR_SIZE - 1));
2562  }
2563