On Sun, Dec 15, 2019 at 11:12:36AM -0600, Aditya Pakki wrote:
> alloc_extent_state_atomic() allocates extents via GFP_ATOMIC flag
> and cannot fail.
Here you say it cannot fail (without a proof)
> There are multiple invocations of BUG_ON on the
> return value to check for failure. The patch replaces certain
> invocations of BUG_ON by returning the error upstream.
>
> Signed-off-by: Aditya Pakki <pakki001@xxxxxxx>
> ---
> fs/btrfs/extent_io.c | 20 ++++++++++++++++----
> 1 file changed, 16 insertions(+), 4 deletions(-)
>
> diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
> index eb8bd0258360..e72e5a333e71 100644
> --- a/fs/btrfs/extent_io.c
> +++ b/fs/btrfs/extent_io.c
> @@ -989,7 +989,10 @@ __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
> node = tree_search_for_insert(tree, start, &p, &parent);
> if (!node) {
> prealloc = alloc_extent_state_atomic(prealloc);
> - BUG_ON(!prealloc);
> + if (!prealloc) {
> + err = -ENOMEM;
> + goto out;
And add error handling when it fails, which is in contradiction but
anyway, replacing BUG_ON with error checks requires auditing the whole
call chain up. We've had attempts to do this change already, none of them
was correct so unfortunatelly the BUG_ON needs to stay to avoid errors
being silently ignored.
