[Devel] Re: [PATCH 2/5] c/r: let entire thread group in sys_restart before restoring a thread

[Serge E. Hallyn]

Quoting Oren Laadan (orenl at librato.com):
> Ensure that all members of a thread group are in sys_restart before
> restoring any of them. Otherwise, restore may modify shared state and
> crash or fault a thread still in userspace,
> 
> For thread groups, each thread scans the entire group and tests for
> PF_RESTARTING on every member. If not all are set, then we wait, and
> when woken up try again (unless signaled). If all are set, then we're
> done and wakeup all threads.
> 
> Signed-off-by: Oren Laadan <orenl at cs.columbia.edu>
> ---
>  checkpoint/restart.c |   52 ++++++++++++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 52 insertions(+), 0 deletions(-)
> 
> diff --git a/checkpoint/restart.c b/checkpoint/restart.c
> index 5d936cf..37454c5 100644
> --- a/checkpoint/restart.c
> +++ b/checkpoint/restart.c
> @@ -695,6 +695,54 @@ static int do_ghost_task(void)
>  	/* NOT REACHED */
>  }
> 
> +/*
> + * Ensure that all members of a thread group are in sys_restart before
> + * restoring any of them. Otherwise, restore may modify shared state
> + * and crash or fault a thread still in userspace,
> + */
> +static int wait_sync_threads(void)
> +{
> +	struct task_struct *p, *leader;
> +
> +	if (thread_group_empty(current))
> +		return 0;
> +
> +	p = leader = current->group_leader;
> +
> +	/*
> +	 * Our PF_RESTARTING is already set. Each thread loops through
> +	 * the group testing everyone's PF_RESTARTING. If not set on
> +	 * all members, it sleeps to retry later. Otherwise it wakes
> +	 * up all sleepers and returns.
> +	 */
> + retry:
> +	__set_current_state(TASK_INTERRUPTIBLE);
> +
> +	read_lock(&tasklist_lock);
> +	do {
> +		if (!(p->flags & PF_RESTARTING))
> +			break;
> +		p = next_thread(p);
> +	} while (p != leader);
> +
> +	if (p != leader) {
> +		read_unlock(&tasklist_lock);
> +		if (signal_pending(current))

Not sure...  but do you need to get back to TASK_RUNNING
in this case?  (the schedule() below does it automatically,
but not this failure case)

> +			return -EINTR;
> +		schedule();
> +		goto retry;
> +	}
> +
> +	do {
> +		wake_up_process(p);
> +		p = next_thread(p);
> +	} while (p != leader);
> +	read_unlock(&tasklist_lock);
> +
> +	__set_current_state(TASK_RUNNING);
> +	return 0;
> +}
> +
>  static int do_restore_task(void)
>  {
>  	struct ckpt_ctx *ctx;
> @@ -706,6 +754,10 @@ static int do_restore_task(void)
> 
>  	current->flags |= PF_RESTARTING;
> 
> +	ret = wait_sync_threads();
> +	if (ret < 0)
> +		return ret;
> +
>  	/* wait for our turn, do the restore, and tell next task in line */
>  	ret = wait_task_active(ctx);
>  	if (ret < 0)
> -- 
> 1.6.0.4
> 
> _______________________________________________
> Containers mailing list
> Containers at lists.linux-foundation.org
> https://lists.linux-foundation.org/mailman/listinfo/containers
_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers