Re: [PATCH -mm 1/1] ptrace: PTRACE_GETFDPIC: fix the unsafe usage

[Oleg Nesterov <oleg@redhat.com>]

On 05/25, David Howells wrote:
>
> Roland McGrath <roland@redhat.com> wrote:
>
> > The mm pointer is only used by these uncommon ptrace operations
>
> Like PEEKTEXT and POKETEXT?

They use access_process_vm().

According to grep, mm is only use to read a couple of members.

Perhaps can even add the simple helper

	struct mm_xxx {
		unsigned long start_code, end_code, start_data, end_data;
		... some more ...
	};

	int get_mm_xxx(struct task_struct *tracee, struct mm_xxx *mm_xxx)
	{
		struct mm_struct *mm = get_task_mm(tracee);
		...
	}

Except:

	- arch/um/kernel/ptrace.c PTRACE_SWITCH_MM does something
	  really strange

	- arch/blackfin/kernel/ptrace.c:is_user_addr_valid()
	  needs mmap_sem around find_vma()

	  The lockless access to mm->context.sram_list doesn't look
	  safe to me.

	  If we add get_task_mm() - this protects us against
	  destroy_context() only. What is the tracee's sub-thread
	  does sys_sram_alloc() or sys_sram_free() in parallel?

Oleg.