btrfs: add missing path cache update during fiemap

When looking the stored result for a cached path node, if the stored
result is valid and has a value of true, we must update all the nodes for
all levels below it with a result of true as well. This is necessary when
moving from one leaf in the fs tree to the next one, as well as when
moving from a node at any level to the next node at the same level.

Currently this logic is missing as it was somehow forgotten by a recent
patch with the subject: "btrfs: speedup checking for extent sharedness
during fiemap".

This adds the missing logic, which is the counter part to what we do
when adding a shared node to the cache at store_backref_shared_cache().

Fixes: 12a824dc67a6 ("btrfs: speedup checking for extent sharedness during fiemap")
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

diff --git a/fs/btrfs/backref.c b/fs/btrfs/backref.c
index dce3a16996b951d3aed5ebe789f727d4d7289149..3c0c1f626c75dfd0e951c95a2d36e48a5a755c3d 100644 (file)
--- a/fs/btrfs/backref.c
+++ b/fs/btrfs/backref.c
@@ -1557,6 +1557,19 @@ static bool lookup_backref_shared_cache(struct btrfs_backref_shared_cache *cache
                return false;
 
        *is_shared = entry->is_shared;
+       /*
+        * If the node at this level is shared, than all nodes below are also
+        * shared. Currently some of the nodes below may be marked as not shared
+        * because we have just switched from one leaf to another, and switched
+        * also other nodes above the leaf and below the current level, so mark
+        * them as shared.
+        */
+       if (*is_shared) {
+               for (int i = 0; i < level; i++) {
+                       cache->entries[i].is_shared = true;
+                       cache->entries[i].gen = entry->gen;
+               }
+       }
 
        return true;
 }