On 09/14/2018 07:43 AM, Qu Wenruo wrote:
On 2018/9/13 上午4:49, damenly.su@xxxxxxxxx wrote:
From: Su Yue <suy.fnst@xxxxxxxxxxxxxx>
After call of check_inode_item(), path may point to the last unchecked
slot of the leaf. The outer walk_up_tree() always treats the position
as checked item then skips to next item.
So check_inode_item() always set its path to *unchecked* slot, while
walk_up_tree() always think its path is set to *checked* slot.
Then this is indeed a problem.
Yep, that's the case.
If the last item was an inode item, yes, it was unchecked.
Then walk_up_tree() will think the leaf is checked and walk up to
upper node, process_one_leaf() in walk_down_tree() would skip to
check next inode item. Which means, the inode item won't be checked.
Solution:
After check_inode_item returns, if found path point to the last item
of a leaf,
Would you please explain more about why last item makes a difference here?
From previous statement, it looks like it's the difference in how
walk_up_tree() and check_inode_item() handles the path.
Not really related to last item.
Yes, the change is tricky. The core problem is walk_up_tree() will
skip to other nodes. Decreament of slot will let path still
point to the leaf after walk_up_tree() returns.
Or we must change logical of walk_up_tree() or check_inode_item().
BTW,now process_one_leaf() checks inode_item from first inode_item
or second item with different ino. Change it start from last slot
should be the right way.
Thanks,
Su
Thanks,
Su
Or did I miss something?
Thanks,
Qu
decrease path slot manually, so walk_up_tree() will stay
on the leaf.
Fixes: 5e2dc770471b ("btrfs-progs: check: skip shared node or leaf check for low_memory mode")
Signed-off-by: Su Yue <suy.fnst@xxxxxxxxxxxxxx>
---
check/mode-lowmem.c | 12 ++++++++++++
1 file changed, 12 insertions(+)
diff --git a/check/mode-lowmem.c b/check/mode-lowmem.c
index 8fc9edab1d66..b6b33786d02b 100644
--- a/check/mode-lowmem.c
+++ b/check/mode-lowmem.c
@@ -2612,6 +2612,18 @@ again:
if (cur->start == cur_bytenr)
goto again;
+ /*
+ * path may point at the last item(a inode item maybe) in a leaf.
+ * Without below lines, walk_up_tree() will skip the item which
+ * means all items related to the inode will never be checked.
+ * Decrease the slot manually, walk_up_tree won't skip to next node
+ * if it occurs.
+ */
+ if (path->slots[0] + 1 >= btrfs_header_nritems(path->nodes[0])) {
+ if (path->slots[0])
+ path->slots[0]--;
+ }
+
/*
* we have switched to another leaf, above nodes may
* have changed, here walk down the path, if a node
