Re: [fs/bcachefs] Question about memory allocation in btree_paths_realloc

[Zijie Zhao]

On 3/20/24 4:48 PM, Kent Overstreet wrote:

On Wed, Mar 20, 2024 at 04:37:02PM -0500, Zijie Zhao wrote:

Dear BCACHEFS maintainers,

We encountered an unusual usage of kzalloc in btree_paths_realloc while
performing a static analysis for kernel code.


https://elixir.bootlin.com/linux/latest/source/fs/bcachefs/btree_iter.c#L1523

```
static noinline void btree_paths_realloc(struct btree_trans *trans)
{
        unsigned nr = trans->nr_paths * 2;

        void *p = kzalloc(BITS_TO_LONGS(nr) * sizeof(unsigned long) +
                          sizeof(struct btree_trans_paths) +
                          nr * sizeof(struct btree_path) +
                          nr * sizeof(btree_path_idx_t) + 8 +
                          nr * sizeof(struct btree_insert_entry), 
GFP_KERNEL|__GFP_NOFAIL);

        unsigned long *paths_allocated = p;
        memcpy(paths_allocated, trans->paths_allocated,
BITS_TO_LONGS(trans->nr_paths) * sizeof(unsigned long));
        ...
}
```

Here kzalloc might return NULL in case of out-of-memory, making
memcpy(NULL,...) to have undefined behavior. Would a NULL check be needed
here?

See that __GFP_NOFAIL?

Thanks! Just realized that!