Re: [PATCH] exec: log when wait_for_dump_helpers aborts due to a signal

[Oleg Nesterov <oleg@redhat.com>]

On 10/25, Mandeep Singh Baines wrote:
>
> To allow coredump pipe readers to look at /proc/<pid> of the crashing
> process, the kernel waits for the reader to exit. However, the wait
> is silently aborted if the crashing process is signalled.

Well, yes... But note that this signal_pending() is only used because
with TIF_SIGPENDING we have the problems anyway.

Oh. I promise myself I'll make do_coredump() killable and cleanup this
all every time I look at this code...

> This patch, logs whenever wait_for_dump_helpers is aborted or in order
> to assist in debugging cases where /proc/<pid> is gone.

I don't really understand why this is useful. The reader process
can complain if it can't collect the data (say, /proc/pid goes away
or EOF doesn't come).



As for the patch itself,

> -	while ((pipe->readers > 1) && (!signal_pending(current))) {
> +	while (pipe->readers > 1) {
> +		if (signal_pending(current)) {
> +			pr_info("wait_for_dump_helpers[%d]: "
> +				"aborted due to signal\n",
> +				task_pid_nr(current));
> +			break;
> +		}
> +
>  		wake_up_interruptible_sync(&pipe->wait);

This can't help in general. If signal_pending() == T, it is quite
possible that pipe_write() already failed before.

Oleg.