2010/9/15 KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx>:
>
> Briefly says, to introduce new limit has bad benefit/risk balance. Sadly.
Well, I mostly agree. That said, I do think we could extend the
limiter some ways.
For example, I think the "stack limit / 4" is perfectly sane, but it
would make total sense to perhaps also take into account the AS and
RSS limits.
And I do think that your attempt to use __vm_enough_memory() was good.
It happens to be coded in a way that makes it useless for a one-pass
model, and some of what it does would be too expensive to do up-front
when you can't short-circuit it, but I do think that it would probably
be appropriate to at least try to take the _rough_ code there and use
it as a limit for maximum stack size too.
For example, we could have a function somewhat like
unsigned long max_stack_size(void)
{
unsigned long allowed, used, limit;
switch (sysctl_overcommit_memory) {
case OVERCOMMIT_ALWAYS:
allowed = ULONG_MAX;
break;
case OVERCOMMIT_GUESS:
.. maybe we can come up with some upper bound here too ..
break;
default:
allowed = (totalram_pages - hugetlb_total_pages())
* sysctl_overcommit_ratio / 100;
if (!cap_sys_admin)
allowed -= allowed / 32;
allowed += total_swap_pages;
/* Don't let a single process grow too big:
leave 3% of the size of this process for other processes */
if (mm)
allowed -= mm->total_vm / 32;
/* What is already committed to? */
used = percpu_counter_read_positive(&vm_committed_as);
if (used > allowed)
return 0;
allowed -= used;
break;
}
limit = ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4;
if (allowed > limit)
allowed = limit;
return allowed;
}
which we'd call once at the beginning of the execve(), and then
remember that result and use it instead of the current 'rlimit/4'
value.
Now, admittedly the OVERCOMMIT_GUESS case is the interesting one, and
the one that is hard to write efficiently. But maybe we could make
'nr_free_pages()' cheap enough that doin that whole OVERCOMMIT_GUESS
"approximate free pages" thing from __vm_enough_memory would work out
too?
I dunno. It doesn't look hopeless.
Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
