Re: [RFC PATCH v11 for 4.15 01/24] Restartable sequences system call

[Mathieu Desnoyers]

----- On Nov 16, 2017, at 2:06 PM, Thomas Gleixner tglx@xxxxxxxxxxxxx wrote:

> On Thu, 16 Nov 2017, Mathieu Desnoyers wrote:
>> ----- On Nov 16, 2017, at 1:43 PM, Peter Zijlstra peterz@xxxxxxxxxxxxx wrote:
>> 
>> > On Tue, Nov 14, 2017 at 03:03:51PM -0500, Mathieu Desnoyers wrote:
>> >> +/*
>> >> + * If parent process has a registered restartable sequences area, the
>> >> + * child inherits. Only applies when forking a process, not a thread. In
>> >> + * case a parent fork() in the middle of a restartable sequence, set the
>> >> + * resume notifier to force the child to retry.
>> >> + */
>> >> +static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
>> >> +{
>> >> +	if (clone_flags & CLONE_THREAD) {
>> >> +		t->rseq = NULL;
>> >> +		t->rseq_len = 0;
>> >> +		t->rseq_sig = 0;
>> >> +	} else {
>> >> +		t->rseq = current->rseq;
>> >> +		t->rseq_len = current->rseq_len;
>> >> +		t->rseq_sig = current->rseq_sig;
>> >> +		rseq_set_notify_resume(t);
>> >> +	}
>> >> +}
>> > 
>> > This hurts my brain... what happens if you fork a multi-threaded
>> > process?
>> > 
>> > Do we fully inherit the TLS state of the calling thread?
>> 
>> Yes, exactly. The user-space TLS should be inherited from that of
>> the calling thread.
>> 
>> At kernel-level, the only thing that's not inherited here is the
>> task struct rseq_event_mask, which tracks whether a restart is
>> needed. But this would only be relevant if fork() can be invoked
>> from a signal handler, or if fork() could be invoked from a
>> rseq critical section (which really makes little sense).
> 
> Whether it makes sense or not does not matter much, especially in context
> of user space. You cannot make assumptions like that. When something can be
> done, then it's bound to happen sooner than later because somebody thinks
> he is extra clever.
> 
> The first priority is robustness in any aspect which has to do with user
> space.
> 
>> Should I copy the current->rseq_event_mask on process fork just to
>> be on the safe side though ?
> 
> I think so, unless you let fork() fail when invoked from a rseq critical
> section.

Allright, I'll set the rseq_event_mask to 0 explicitly on exec() and
thread-fork, and copy it from the parent on process-fork.

Thanks,

Mathieu

> 
> Thanks,
> 
> 	tglx

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com