Re: [PATCH -tip v3 0/2] kprobes/x86: Another way to make insn buffer RO and cleanup

[Masami Hiramatsu]

Hi Ingo,

What would you think about fixing this way?
This makes the instruction buffer always RO and
poke it via text_poke.

Thank you,

On Fri, 18 Aug 2017 17:22:54 +0900
Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:

> Hi,
> 
> This series modifies how to handle RO insn buffer and
> cleans up addressof operators.
> 
> The 1st patch changes the RO insn buffer handling: instead
> of using set_memory_ro/rw to modify the buffer, it prepares
> new instructions in another buffer and write it with
> text_poke() as suggested by Ingo Molnar (Thanks!).
> Since the text_poke() is safely modifying code by
> mapping alias pages, it can write RO pages.
> This also override alloc_insn_page() so that it returns
> ROX page directly.
> 
> The 2nd one is not changed. It is a cleanup patch
> to remove addressof operators ("&") since
> it is meaningless anymore.
> 
> V3 has just a following update:
>  - [1/2] Not to just add set_memory_ro(), introduce new
>    patch to change the way to handle RO pages.
> 
> Thanks,
> 
> ---
> 
> Masami Hiramatsu (2):
>       kprobes/x86: Make insn buffer always ROX and use text_poke
>       kprobes/x86: Remove addressof operators
> 
> 
>  arch/x86/include/asm/kprobes.h   |    4 +-
>  arch/x86/kernel/kprobes/common.h |    6 ++-
>  arch/x86/kernel/kprobes/core.c   |   61 +++++++++++++++++++++------------
>  arch/x86/kernel/kprobes/opt.c    |   71 +++++++++++++++++++++-----------------
>  kernel/kprobes.c                 |    2 +
>  5 files changed, 86 insertions(+), 58 deletions(-)
> 
> --
> Masami Hiramatsu <mhiramat@xxxxxxxxxx>


-- 
Masami Hiramatsu <mhiramat@xxxxxxxxxx>