[CRIU] [PATCH 3/3] restore: use breakpoints instead of tracing syscalls

[Pavel Emelyanov]

On 09/17/2014 07:12 PM, Andrey Vagin wrote:
> Currently CRIU traces syscalls to catch a moment, when sigreturn() is
> called. Now we trace recv(cmd), close(logfd), close(cmdfd), sigreturn().
> 
> We can reduce a number of steps by using hw breakpoints. A breakpoint is
> set before sigreturn, so we will need to trace only it.
> 
> Signed-off-by: Andrey Vagin <avagin at openvz.org>
> ---
>  cr-restore.c       | 10 ++++++++--
>  include/restorer.h |  1 +
>  include/rst_info.h |  2 ++
>  pie/restorer.c     | 10 ++++++++--
>  4 files changed, 19 insertions(+), 4 deletions(-)
> 
> diff --git a/cr-restore.c b/cr-restore.c
> index 4d5ccd5..3be32cf 100644
> --- a/cr-restore.c
> +++ b/cr-restore.c
> @@ -1532,7 +1532,7 @@ static int attach_to_tasks(bool root_seized)
>  
>  	for_each_pstree_item(item) {
>  		pid_t pid = item->pid.real;
> -		int status, i;
> +		int status, i, ret;
>  
>  		if (!task_alive(item))
>  			continue;
> @@ -1565,7 +1565,12 @@ static int attach_to_tasks(bool root_seized)
>  				return -1;
>  			}
>  
> -			if (ptrace(PTRACE_SYSCALL, pid, NULL, NULL)) {
> +			ret = ptrace_set_breakpoint(pid, item->rst->breakpoint);
> +			if (ret < 0)
> +				return -1;
> +
> +			/* A breakpoint was not set */
> +			if (ret == 0 && ptrace(PTRACE_SYSCALL, pid, NULL, NULL)) {

This is not symmetrical to parasite unload.

For parasite:

	wake_up_socket()	

	if (set_breakpoint())
		wait()

	PTRACE_SYSCALL
	while (1) {
		wait()
		PTRACE_SYSCALL
	}

For restorer

	if (set_breakpoint())
		/* nothing */
	else
		PTRACE_SYSCALL

	wake_up_futex()

	while (1) {
		wait()
		PTRACE_SYSCALL
	}

Why?

>  				pr_perror("Unable to start %d", pid);
>  				return -1;
>  			}
> @@ -2674,6 +2679,7 @@ static int sigreturn_restore(pid_t pid, CoreEntry *core)
>  	if (rst_mem_remap(mem))
>  		goto err;
>  
> +	task_args->breakpoint = &current->rst->breakpoint;
>  	task_args->task_entries = rst_mem_remap_ptr(task_entries_pos, RM_SHREMAP);
>  
>  	task_args->rst_mem = mem;
> diff --git a/include/restorer.h b/include/restorer.h
> index a690341..1dd35f9 100644
> --- a/include/restorer.h
> +++ b/include/restorer.h
> @@ -155,6 +155,7 @@ struct task_restore_args {
>  	struct vdso_symtable		vdso_sym_rt;		/* runtime vdso symbols */
>  	unsigned long			vdso_rt_parked_at;	/* safe place to keep vdso */
>  #endif
> +	void				**breakpoint;
>  } __aligned(64);
>  
>  #define RESTORE_ALIGN_STACK(start, size)	\
> diff --git a/include/rst_info.h b/include/rst_info.h
> index e89be02..6b00c50 100644
> --- a/include/rst_info.h
> +++ b/include/rst_info.h
> @@ -54,6 +54,8 @@ struct rst_info {
>  	struct file_desc	*root;
>  	bool			has_umask;
>  	u32			umask;
> +
> +	void			*breakpoint;
>  };
>  
>  #endif /* __CR_RST_INFO_H__ */
> diff --git a/pie/restorer.c b/pie/restorer.c
> index 11ad574..65bbf01 100644
> --- a/pie/restorer.c
> +++ b/pie/restorer.c
> @@ -321,6 +321,11 @@ static int restore_thread_common(struct rt_sigframe *sigframe,
>  	return 0;
>  }
>  
> +static void noinline rst_sigreturn(unsigned long new_sp)
> +{
> +	ARCH_RT_SIGRETURN(new_sp);
> +}
> +
>  /*
>   * Threads restoration via sigreturn. Note it's locked
>   * routine and calls for unlock at the end.
> @@ -372,7 +377,7 @@ long __export_restore_thread(struct thread_restore_args *args)
>  	futex_dec_and_wake(&thread_inprogress);
>  
>  	new_sp = (long)rt_sigframe + SIGFRAME_OFFSET;
> -	ARCH_RT_SIGRETURN(new_sp);
> +	rst_sigreturn(new_sp);
>  
>  core_restore_end:
>  	pr_err("Restorer abnormal termination for %ld\n", sys_getpid());
> @@ -729,6 +734,7 @@ long __export_restore_task(struct task_restore_args *args)
>  #endif
>  
>  	task_entries = args->task_entries;
> +	*args->breakpoint = rst_sigreturn;
>  
>  	ksigfillset(&act.rt_sa_mask);
>  	act.rt_sa_handler = sigchld_handler;
> @@ -1092,7 +1098,7 @@ long __export_restore_task(struct task_restore_args *args)
>  	 * pure assembly since we don't need any additional
>  	 * code insns from gcc.
>  	 */
> -	ARCH_RT_SIGRETURN(new_sp);
> +	rst_sigreturn(new_sp);
>  
>  core_restore_end:
>  	futex_abort_and_wake(&task_entries->nr_in_progress);
>