[CRIU] [PATCH 20/23] ptrace.c: don't seize the task that doesn't run in the userspace

[Pavel Emelyanov]

On 01/14/2013 11:26 AM, Alexander Kartashov wrote:
> If a task executes a non-userspace code (that sometimes takes place on ARM)
> while being seized the subsequent infection with the syscall blob fails
> because the VMA the PC is in is surely shouldn't be written.

I don't buy this explanation. Infection happens only after task gets stopped
by tracer. Can we stop kernel code execution on ARM???

> The patch modifies the function seize_task() to check the PC of the task
> being seized is below the TASK_SIZE boundary. If the check fails
> another seizure attempt is made in 2 sec.
> 
> Signed-off-by: Alexander Kartashov <alekskartashov at parallels.com>
> ---
>  ptrace.c |   18 ++++++++++++++++++
>  1 file changed, 18 insertions(+)
> 
> diff --git a/ptrace.c b/ptrace.c
> index 0140918..67b499a 100644
> --- a/ptrace.c
> +++ b/ptrace.c
> @@ -50,6 +50,7 @@ int seize_task(pid_t pid, pid_t ppid, pid_t *pgid, pid_t *sid)
>  	int status;
>  	int ret, ret2;
>  	struct proc_pid_stat_small ps;
> +	user_regs_struct_t regs;
>  
>  	ret = ptrace(PTRACE_SEIZE, pid, NULL, 0);
>  
> @@ -129,6 +130,23 @@ try_again:
>  		goto try_again;
>  	}
>  
> +	if (ptrace(PTRACE_GETREGS, pid, NULL, &regs)) {
> +		pr_perror("SEIZE %d: failed to get the task registers.", pid);
> +		goto err;
> +	}
> +
> +	if (REG_IP(regs) >= TASK_SIZE) {
> +		pr_info("SEIZE %d: the current PC is outside the user virtual address space; retrying to seize in 2 sec.\n", pid);
> +
> +		if (ptrace(PTRACE_CONT, pid, NULL, NULL)) {
> +			pr_perror("SEIZE %d: can't continue on a seize retry.", pid);
> +			goto err;
> +		}
> +
> +		sleep(2);
> +		goto try_again;
> +	}
> +
>  	if (si.si_signo == SIGTRAP)
>  		return TASK_ALIVE;
>  	else if (si.si_signo == SIGSTOP)
>