Re: [PATCH] fs: pipe, fix undefined behaviour

[Vegard Nossum]

On 14 October 2016 at 11:23, Jiri Slaby <jslaby@xxxxxxx> wrote:
> echo | ./program
> where ./program contains fcntl(0, F_SETPIPE_SZ, 0) this is triggered:
> UBSAN: Undefined behaviour in ../include/linux/log2.h:63:13
> shift exponent 64 is too large for 64-bit type 'long unsigned int'
> CPU: 3 PID: 4978 Comm: syz-executor Not tainted 4.8.1-0-syzkaller #1
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.9.1-0-gb3ef39f-prebuilt.qemu-project.org 04/01/2014
> ...
> Call Trace:
> ...
>  [<ffffffffa18ced53>] ? pipe_fcntl+0xa3/0x7a0
>  [<ffffffffa18f1940>] ? SyS_fcntl+0x930/0xf30
>  [<ffffffffa2d1fb00>] ? entry_SYSCALL_64_fastpath+0x23/0xc1
>
> Documentaion of roundup_pow_of_two explicitly mentions that passing
> size 0 is undefined. So return 0 from round_pipe_size prematurely to
> fix this.
>
> Signed-off-by: Jiri Slaby <jslaby@xxxxxxx>
> Cc: Alexander Viro <viro@xxxxxxxxxxxxxxxxxx>
> Cc: linux-fsdevel@xxxxxxxxxxxxxxx
> ---
>  fs/pipe.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/fs/pipe.c b/fs/pipe.c
> index 8e0d9f26dfad..23c0c06ffc33 100644
> --- a/fs/pipe.c
> +++ b/fs/pipe.c
> @@ -1024,6 +1024,9 @@ static inline unsigned int round_pipe_size(unsigned int size)
>  {
>         unsigned long nr_pages;
>
> +       if (!size)
> +               return 0;
> +
>         nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
>         return roundup_pow_of_two(nr_pages) << PAGE_SHIFT;
>  }

Just FYI I sent a patch for this earlier with a bit more analysis,
although it probably no longer applies cleanly:

https://lkml.org/lkml/2016/8/12/215

Personally I felt it was cleaner to limit the argument passed to
round_pipe_size() instead of relying on the implicit long -> int
truncation. Anyway, feel free to crib from the patch and/or changelog.


Vegard