On 06/04, Martin Ågren wrote:
> We allocate a `struct refspec_item` on the stack without initializing
> it. In particular, its `dst` and `src` members will contain some random
> data from the stack. When we later call `refspec_item_clear()`, it will
> call `free()` on those pointers. So if the call to `parse_refspec()` did
> not assign to them, we will be freeing some random "pointers". This is
> undefined behavior.
>
> To the best of my understanding, this cannot currently be triggered by
> user-provided data. And for what it's worth, the test-suite does not
> trigger this with SANITIZE=address. It can be provoked by calling
> `valid_fetch_refspec(":*")`.
>
> Zero the struct, as is done in other users of `struct refspec_item`.
>
> Signed-off-by: Martin Ågren <martin.ag...@gmail.com>
> ---
> I found some time to look into this. It does not seem to be a
> user-visible bug, so not particularly critical.
Thanks for fixing this. I don't think I noticed this because at some
point in developing this series I had a memset call in parse_refspec.
I don't remember why I ended up removing it, but maybe it would have
been better to leave it in there.
>
> refspec.c | 5 ++++-
> 1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/refspec.c b/refspec.c
> index ada7854f7a..7dd7e361e5 100644
> --- a/refspec.c
> +++ b/refspec.c
> @@ -189,7 +189,10 @@ void refspec_clear(struct refspec *rs)
> int valid_fetch_refspec(const char *fetch_refspec_str)
> {
> struct refspec_item refspec;
> - int ret = parse_refspec(&refspec, fetch_refspec_str, REFSPEC_FETCH);
> + int ret;
> +
> + memset(&refspec, 0, sizeof(refspec));
> + ret = parse_refspec(&refspec, fetch_refspec_str, REFSPEC_FETCH);
> refspec_item_clear(&refspec);
> return ret;
> }
> --
> 2.18.0.rc0.43.gb85e7bcbff
>
--
Brandon Williams
