Re: [PATCH] bus: fsl-mc: fix improper free of mc_dev

[Nick Desaulniers <ndesaulniers@google.com>]

On Tue, Apr 27, 2021 at 11:36 AM <trix@redhat.com> wrote:
>
> From: Tom Rix <trix@redhat.com>
>
> Clang static analysis reports this error
>
> fsl-mc-bus.c:891:2: warning: Attempt to free released memory
>         kfree(mc_dev);
>         ^~~~~~~~~~~~~
>
> In this block of code
>
> if (strcmp(obj_desc->type, "dprc") == 0) {
> ..
>   mc_bus = kzalloc(..)
>   mc_dev = &mc_bus->mc_dev;

Thanks for the patch.

Aren't the allocations for mc_bus and mc_dev mutually exclusive based
on that conditional? If so...

>
> mc_dev is not alloc-ed, so it should not be freed.
> Old handler triggers a false positive from checkpatch, so add a
> comment and change logic a bit.
>
> Fixes: a042fbed0290 ("staging: fsl-mc: simplify couple of deallocations")
> Signed-off-by: Tom Rix <trix@redhat.com>
> ---
>  drivers/bus/fsl-mc/fsl-mc-bus.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/bus/fsl-mc/fsl-mc-bus.c b/drivers/bus/fsl-mc/fsl-mc-bus.c
> index 380ad1fdb745..fb3e1d8a7f63 100644
> --- a/drivers/bus/fsl-mc/fsl-mc-bus.c
> +++ b/drivers/bus/fsl-mc/fsl-mc-bus.c
> @@ -887,8 +887,10 @@ int fsl_mc_device_add(struct fsl_mc_obj_desc *obj_desc,
>
>  error_cleanup_dev:
>         kfree(mc_dev->regions);
> +       /* mc_dev is only allocated when it is not part of mc_bus */
> +       if (!mc_bus)
> +               kfree(mc_dev);
>         kfree(mc_bus);
> -       kfree(mc_dev);

The error handling here seems quite wrong (regardless of your patch).
mc_dev->regions is allocated by fsl_mc_device_get_mmio_regions() IIUC.
Wouldn't the first `goto error_cleanup_dev;` taken end up passing an
uninitialized pointer to kfree()?

what if `strcmp(obj_desc->type, "dprc") == 0` is false? We allocate
`mc_dev`, but then call kfree on `mc_bus`?

I think it would be safer to locally save the result of
`strcmp(obj_desc->type, "dprc") == 0`, then check that throughout this
function, including the error handling at the end, or use multiple
labels to unwind the allocations correctly.

>
>         return error;
>  }

> --


--
Thanks,
~Nick Desaulniers