From mboxrd@z Thu Jan 1 00:00:00 1970 From: "David S. Miller" Subject: Re: [RFC] tcp ephemeral port selection Date: Mon, 25 Oct 2004 16:34:18 -0700 Sender: netdev-bounce@oss.sgi.com Message-ID: <20041025163418.30c703e2.davem@redhat.com> References: <20041011171422.65f88472@zqx3.pdx.osdl.net> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Cc: netdev@oss.sgi.com Return-path: To: Stephen Hemminger In-Reply-To: <20041011171422.65f88472@zqx3.pdx.osdl.net> Errors-to: netdev-bounce@oss.sgi.com List-Id: netdev.vger.kernel.org On Mon, 11 Oct 2004 17:14:22 -0700 Stephen Hemminger wrote: > Here is a test patch which changes how TCP ephemeral ports are > selected from a simple roving pointer to: > * net_random() as starting point when doing pure local > searchs. > * use md hash (like isn) when doing connection based > assignments with a rover > > This is *not* for 2.6.9 but wanted to get it out for comment for > future versions. I'm not totally against this kind of idea at all. I hope we don't run into situations with the new code where we don't actually check the whole port range due to some logic error or similar. Another thing we need to eventually do is allow allocation into the local port space more completely. Even if every local port has one user, we could still allocate some ports if the daddr/dport is known and is different from the existing user.