Re: [RFC Patch] net: reserve ports for applications using fixed port numbers

[Cong Wang <amwang@redhat.com>]

Octavian Purdila wrote:
> On Thursday 04 February 2010 19:41:10 you wrote:
> 
>> From: Octavian Purdila <opurdila@ixiacom.com>
>> Date: Thu, 4 Feb 2010 14:44:01 +0200
>>
>>> My concern is that we can have multiple applications that require a
>>> fixed port and if those ports are significantly apart we will
>>> decrease the port range available for connect. And that will hurt
>>> the rate of which new connections can be opened.
>> I'm already uneasy about adding the simple check every time
>> we loop around in the bind port allocator.
>>
>> Adding an LSM hook to this spot?  I absolutely refuse to allow
>> that, it will completely kill bind performance.
>>
> 
> I think Tetsuo was proposing the LSM hook, so I'll leave him the daunting task 
> of convincing you of the benefit of that :) - I have no opinion on this due to 
> massive lack of knowledge.
> 
> I was just proposing to use a discrete set of ports instead of a range. The 
> check in the current patch:
> 
> int inet_is_reserved_local_port(int port)
> {
>        int min, max;
> 
>        inet_get_local_reserved_ports(&min, &max);
>        if (min && max)
>                return (port >= min && port <= max);
>        return 0;
> }
> 
> would become:
> 
> int inet_is_reserved_local_port(int port)
> {
> 	if (test_bit(port, reserved_ports))
> 		return 1;
> 	return 0;
> }
> 
> In theory it might be slower because of the reserved_ports bitmap will have a 
> larger memory footprint than just a min/max, especially with random port 
> allocation. But is this an issue in practice?

Again, using bitmap algorithm is not a problem and it's better, the
problem is sysctl interface, how would you plan to interact with users
via sysctl/proc if you use bitmap to handle this? I would like to hear
more details about this.

Thanks!