Re: [v2 PATCH] mm: introduce arg_lock to protect arg_start|end and env_start|end in mm_struct

[Yang Shi]

On 3/26/18 11:29 PM, Michal Hocko wrote:
> On Tue 27-03-18 02:20:39, Yang Shi wrote:
> [...]
> The patch looks reasonable to me. Maybe it would be better to be more
> explicit about the purpose of the patch. As others noticed, this alone
> wouldn't solve the mmap_sem contention issues. I _think_ that if you
> were more explicit about the mmap_sem abuse it would trigger less
> questions.
>
> I have just one more question. Now that you are touching this area,
> would you be willing to remove the following ugliness?
>
> > diff --git a/kernel/sys.c b/kernel/sys.c
> > index f2289de..17bddd2 100644
> > --- a/kernel/sys.c
> > +++ b/kernel/sys.c
> > @@ -1959,7 +1959,7 @@ static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data
> >   			return error;
> >   	}
> >   
> > -	down_write(&mm->mmap_sem);
> > +	down_read(&mm->mmap_sem);
> Why do we need to hold mmap_sem here and call find_vma, when only
> PR_SET_MM_ENV_END: is consuming it? I guess we can replace it wit the
> new lock and take the mmap_sem only for PR_SET_MM_ENV_END.

Sorry for taking a little bit longer to get back since I was traveling. 
I think all the stuff can be protected by the new arg_lock except 
mm->brk since arg_lock can't prevent from concurrent writing from sys_brk().

We may use arg_lock to protect everything else other than mm->brk. The 
complexity sounds acceptable.

Of course, as Cyrill mentioned, he prefers to deprecating prctl_set_mm 
since C/R is the only user of it. We may wait until he is done?

Thanks,
Yang

> Thanks!