From: "Ian Latter" Subject: Re: [PATCH] RPC match and conntrack modules v2.1 Date: Sun, 12 Jan 2003 21:33:07 +1100 Sender: netfilter-devel-admin@lists.netfilter.org Message-ID: <200301121133.h0CBX9K10094@singularity.tronunltd.com> Reply-To: "Ian Latter" Content-Type: text/plain; charset="us-ascii" Cc: Return-path: To: "Harald Welte" Errors-To: netfilter-devel-admin@lists.netfilter.org List-Help: List-Post: List-Subscribe: , List-Unsubscribe: , List-Archive: List-Id: netfilter-devel.vger.kernel.org Hi Harald, I didn't get your gz attachment, but I reassembled your patch from the mailling list archive. I have done all bar the cache implementation. You may be right about its usage (and hence its performance) but I'm not sure -- if NFS only does one RPC lookup per procedure then the ls data will be a packet per file in the expectation, not the decision making process, which will put it out of my hands ... I don't have the gear to test that particular scenario on hand. I would like to apply the cache stuff, but it looked a little challenging for the time I had available. I am also interested in combining the tcp and udp modules into a single module. I have done a fair bit of work on its cleanliness since you last saw it. The file is on the ftp server -- it applies cleanly to 2.4.20. I think you'll be much happier with that implementation (v2.2). 11572 Jan 12 21:08 pom-rpc-030112-01.tgz afd50da3f913ba860b69c74030b0a056 pom-rpc-030112-01.tgz Let me know how you find it ... ----- Original Message ----- >From: "Harald Welte" >To: "Ian Latter" >Subject: Re: [PATCH] RPC match and conntrack modules v2.1 >Date: Sat, 11 Jan 2003 01:05:55 +0100 > > On Sat, Jan 11, 2003 at 09:47:01AM +1100, Ian Latter wrote: > > > Cool, thanks .. :-) I think there will be a lot of embedded firewall > > manufacturers that will enjoy this one ... > > agreed :) > > > That's correct - there was no NAT built into record-rpc, not even in an > > old API. I wanted to make the module set complete by writing the > > corresponding NAT module(s) but I didn't get time. I might be re-inspired > > later in January if no one else picks it up -- the code should be almost > > identical to the RSH work module that I wrote, because the connection > > handling is quite similar ..... actually ... maybe it doesn't need a > > nat helper; the connections are all initiated from the client to the > > server, and the internal protocols aren't carrying payloads that need > > to be rewritten ... so as long as the nat stuff handles "related" > > streams sanely, then its probably ok ... > > no problem, let's just ignore NAT for now. > > > > See attached patch. Mostly CodingStyle updates (you seem to like 8 spaces > > > instead of tab), but also a minor fix: > > Cool ... could you resent it as a "zip" attachment ... my mail software > > decodes text attachments which will make all the tabs in your patch > > spaces again ... > > well, here it is as .gz again. Strange mailclient you must be using... > > > > btw: I'd really love to see the large chunk of code in the switch > > > statement of the match() function in ipt_rpc.c be split up in seperate > > > functions (thus reducing indentation and improving readability). > > possibly could be ... the gumby way might be to do udp and tcp > > functions ... but a better way might be to pull out the rpc payload > > stuff .... I can take a look at that. > > thanks. > > > > I don't know, but I think esp. in cases where you have lots of rpc > > > activity (NFS?, I'm not an RPC expert) it might be wise to use a slab > > > cache for the 'struct request_p' list items. > > I have not done that before, is there an example some place in the > > existing modules? > > yes, it's done in ip_conntrack_core. see kmem_cache_create / > kmem_cache_alloc / and related functions. The idea is for the memory > management to keep a pool of memory chunks of the requested size around, > in case they are heavily allocated / deallocated. We do this with > struct sk_buff, struct ip_conntrack. > > I'm not sure whether it's worth the effort in the rpc case... But if I'm > not mistaken in NFSv2, every single file in an 'ls' is a seperate RPC > request/reply, isn't it? > > > > Please integrate my patch [and maybe consider my suggestions, that's of > > > course up to you], test it and resubmit it. > > hopefully be done by the end of the weekend/early next week. > > no problem, I'll be off over the wekend anyway. > > > > Thanks! > > Thanks for taking a look at it, and providing feed back ... > > You're welcome. > > -- > > Ian Latter > > -- > - Harald Welte / laforge@gnumonks.org http://www.gnumonks.org/ > =========================================================================== = > "If this were a dictatorship, it'd be a heck of a lot easier, just so long > as I'm the dictator." -- George W. Bush Dec 18, 2000 > -- Ian Latter Internet and Networking Security Officer Macquarie University