We return 0 unconditionally in most of the error handling paths of
'btrfs_mark_extent_written()'.
However, 'ret' is set to some error codes in several error handling paths.
Return 'ret' instead to propagate the error code.
Fixes: 9c8e63db1de9 ("Btrfs: kill BUG_ON()'s in btrfs_mark_extent_written")
Signed-off-by: Christophe JAILLET <christophe.jaillet@xxxxxxxxxx>
---
This patch proposal is purely speculative.
I'm not sure at all that returning 'ret' is correct (but it looks like it
is :) )
What puzzles me is when 'ret' is set, 'btrfs_abort_transaction()' is also
called.
However, the only caller of 'btrfs_mark_extent_written()' (i.e.
'btrfs_finish_ordered_io()') also calls 'btrfs_abort_transaction()' if an
error is returned.
So returning an error code here, would lead to a double call to this abort
function.
I'm usure of if it is correct and/or intented.
If returning 'ret' is correct, should we also axe the 'btrfs_abort_transaction()'
calls here, and leave the caller do the clean-up?
Before the commit in the Fixes tag, we were BUGing_ON in case of errror. So
propagating the error was pointless.
---
fs/btrfs/file.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 15b925142793..cac0bd744de3 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -1374,7 +1374,7 @@ int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
}
out:
btrfs_free_path(path);
- return 0;
+ return ret;
}
/*
--
2.17.1
