[Devel] Re: [PATCH 1/3] restart: coordinator in new pidns to always report status via pipe

[Serge E. Hallyn]

Quoting Oren Laadan (orenl at librato.com):
> Serge Hallyn reports:
>   "another question: if i run 'restart < out' and sys_restart returns
>   due to a -EPERM on some object, then restart.c returns 1.  but if i
>   'restart --pids', then it reports the error and returns 0.  unless i
>   add --copy-status to the flags.  that seems inconsistent?"
> 
> It was with a subtree checkpoint in a child pidns, root-task is not
> pid 1, So, the restarts calls ckpt_coordinator_pidns() execution.
> 
> In commit 2000bbb4b9... "restart: fix race in ckpt_coordinator_pidns
> and --no-wait" adds a pipe for a coordinator in a new pids to report
> success/failure of the restart operation back to the parent when the
> parent does not wish to wait.
> 
> IOW, the coordinator's exit value is overloaded - used once to report
> success/failure and once (optionally) to report root-tasks exit status.
> 
> This patch fixes this by extending the previous commit to make the
> coordinator-pidns always report the restart status via the pipe, and
> only use the exit status for --wait --copy-status case.
> 
> Signed-off-by: Oren Laadan <orenl at cs.columbia.edu>

Seems to be working fine for me.

Tested-by: Serge Hallyn <serue at us.ibm.com>

to all 3.

thanks,
-serge

> ---
>  restart.c |   25 ++++++++++++-------------
>  1 files changed, 12 insertions(+), 13 deletions(-)
> 
> diff --git a/restart.c b/restart.c
> index 35c54ea..5871bbf 100644
> --- a/restart.c
> +++ b/restart.c
> @@ -942,10 +942,12 @@ static int ckpt_coordinator_pidns(struct ckpt_ctx *ctx)
>  	ckpt_dbg("forking coordinator in new pidns\n");
> 
>  	/*
> -	 * We won't wait for (collect) the coordinator, so we use a
> -	 * pipe instead for the coordinator to report success/failure.
> +	 * The coordinator report restart susccess/failure via pipe.
> +	 * (It cannot use return value, because the in the default
> +	 * --wait --copy-status case it is already used to report the
> +	 * root-task's return value).
>  	 */
> -	if (!ctx->args->wait && pipe(ctx->pipe_coord)) {
> +	if (pipe(ctx->pipe_coord) < 0) {
>  		perror("pipe");
>  		return -1;
>  	}
> @@ -981,10 +983,7 @@ static int ckpt_coordinator_pidns(struct ckpt_ctx *ctx)
>  		return -1;
> 
>  	ctx->args->copy_status = copy;
> -	if (ctx->args->wait)
> -		return ckpt_collect_child(ctx);
> -	else
> -		return ckpt_coordinator_status(ctx);
> +	return ckpt_coordinator_status(ctx);
>  }
>  #else
>  static int ckpt_coordinator_pidns(struct ckpt_ctx *ctx)
> @@ -1040,13 +1039,13 @@ static int ckpt_coordinator(struct ckpt_ctx *ctx)
>  		 * around and be reaper until all tasks are gone.
>  		 * Otherwise, container will die as soon as we exit.
>  		 */
> -		if (!ctx->args->wait) {
> -			/* report status because parent won't wait for us */
> -			if (write(ctx->pipe_coord[1], &ret, sizeof(ret)) < 0) {
> -				perror("failed to report status");
> -				exit(1);
> -			}
> +
> +		/* Report success/failure to the parent */
> +		if (write(ctx->pipe_coord[1], &ret, sizeof(ret)) < 0) {
> +			perror("failed to report status");
> +			exit(1);
>  		}
> +
>  		ret = ckpt_pretend_reaper(ctx);
>  	} else if (ctx->args->wait) {
>  		ret = ckpt_collect_child(ctx);
> -- 
> 1.6.0.4
_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers