Re: [PATCH 2/6] binfmt_elf: open code copy_siginfo_to_user to kernelspace buffer

[Arnd Bergmann]

On Mon, Apr 6, 2020 at 2:03 PM Christoph Hellwig <hch@xxxxxx> wrote:
>
> Instead of messing with the address limit just open code the trivial
> memcpy + memset logic.
>
> Signed-off-by: Christoph Hellwig <hch@xxxxxx>
> ---
>  fs/binfmt_elf.c | 7 +++----
>  1 file changed, 3 insertions(+), 4 deletions(-)
>
> diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
> index f4713ea76e82..d744ce9a4b52 100644
> --- a/fs/binfmt_elf.c
> +++ b/fs/binfmt_elf.c
> @@ -1556,10 +1556,9 @@ static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
>  static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
>                 const kernel_siginfo_t *siginfo)
>  {
> -       mm_segment_t old_fs = get_fs();
> -       set_fs(KERNEL_DS);
> -       copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
> -       set_fs(old_fs);
> +       memcpy(csigdata, siginfo, sizeof(struct kernel_siginfo));
> +       memset((char *)csigdata + sizeof(struct kernel_siginfo), 0,
> +               SI_EXPANSION_SIZE);
>         fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
>  }

I think this breaks compat binfmt-elf mode, which relies on this trick:

fs/compat_binfmt_elf.c:#define copy_siginfo_to_user     copy_siginfo_to_user32
fs/compat_binfmt_elf.c#include "binfmt_elf.c"

At least we seem to only have one remaining implementation of
__copy_siginfo_to_user32(), so fixing this won't require touching all
architectures, but I don't see an obvious way to do it right. Maybe
compat-binfmt-elf.c should just override fill_siginfo_note() itself
rather than overriding copy_siginfo_to_user().

       Arnd