From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-2.5 required=3.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, MAILING_LIST_MULTI,SPF_PASS,USER_AGENT_MUTT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id ADCA6C43381 for ; Mon, 18 Mar 2019 17:43:11 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id 8126921473 for ; Mon, 18 Mar 2019 17:43:11 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1728029AbfCRRnK (ORCPT ); Mon, 18 Mar 2019 13:43:10 -0400 Received: from orbyte.nwl.cc ([151.80.46.58]:60922 "EHLO orbyte.nwl.cc" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1727722AbfCRRmx (ORCPT ); Mon, 18 Mar 2019 13:42:53 -0400 Received: from n0-1 by orbyte.nwl.cc with local (Exim 4.91) (envelope-from ) id 1h5wHb-0005Rd-2A; Mon, 18 Mar 2019 18:42:43 +0100 Date: Mon, 18 Mar 2019 18:42:43 +0100 From: Phil Sutter To: Pablo Neira Ayuso Cc: Fernando Fernandez Mancera , netfilter-devel@vger.kernel.org Subject: Re: [PATCH nft v2 1/6] osf: add version fingerprint support Message-ID: <20190318174243.GT4851@orbyte.nwl.cc> Mail-Followup-To: Phil Sutter , Pablo Neira Ayuso , Fernando Fernandez Mancera , netfilter-devel@vger.kernel.org References: <20190314201309.iqdyb7icreyyhhke@salvia> <20190314200737.erhjrhoaciclapsn@salvia> <27D82259-F921-48E0-A718-A08E2BCCAACD@riseup.net> <20190315100333.GD3511@orbyte.nwl.cc> <20190315171328.c5xgazye2hattjxr@salvia> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20190315171328.c5xgazye2hattjxr@salvia> User-Agent: Mutt/1.10.1 (2018-07-13) Sender: netfilter-devel-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: netfilter-devel@vger.kernel.org Hi, On Fri, Mar 15, 2019 at 06:13:28PM +0100, Pablo Neira Ayuso wrote: > On Fri, Mar 15, 2019 at 11:03:33AM +0100, Phil Sutter wrote: > [...] > > On Thu, Mar 14, 2019 at 09:07:37PM +0100, Pablo Neira Ayuso wrote: > > [...] > > > The osf expression returns a string with the OS genre, and if thev > > > version flag is set on, it appends the version to this string, ie. > > > genre + version. > > > > > > This allows us to build maps, ie. > > > > > > meta mark set osf genre { "linux" : 0x10, "windows" : 0x20, "macos" : 0x40 } > > > > > > But, with this new version, you could also do: > > > > > > meta mark set osf genre { "linux::4.0" : 0x11, "linux::3.0" : 0x12, ...} > > > > > > and so on. > > > > > > So I see this version thing as a extended matching. > > > > > > The osf engine actually _already_ finds a precise matching, ie. genre > > > + version, since the fingerprint is per genre + version. But you can > > > just decide to match on the genre (eg. linux). > > > > The problem we're facing IMO is that nft_cmp is limited to a simple > > memcmp(). This demands LHS to know what RHS contains. I'm not implying > > it would be a good idea, but imagine nft_cmp could handle wildcards, we > > could use "linux:*" to match on genre only, "linux:4.0:*" to match on > > genre and version and even "linux:4.*" to match genre and major version > > number. > > > > Actually we might be able to implement the above by setting 'len' field > > correctly. > > The wildcard at the end of the string already works out of the box > via: > > iifname eth\* > > The wildcard matching is generic, so it can be used from any string > datatype, including the osf expression. For osf expression, I guess we would want to add it implicitly in userspace. > > > > Applying the same logic to osf expression, we would implement 'osf name > > > > foo osf version 3.141' and add 'osf_try_merge()' routine to > > > > 'rule_postprocess()' which tries to combine the two statements. > > > > Obviously, this is quite a bit of extra work, not sure if feasible. > > > > > > I think the discussion here is the syntax, ie. > > > > > > osf genre "Linux::4.10" > > > > > > vs. > > > > > > osf genre "Linux" version "4.10" > > > > > > This only requires changes to the userspace nftables side, if you > > > prefer this syntax, which is what I understand you would like to see, > > > right? > > > > Not quite. I like how osf is an expression, not a statement. This makes > > things like 'osf name != "Linux"' possible. What I didn't like was how > > the proposed extension requires users to input redundant info: > > > > | osf name version "Linux:4.20" > > > > RHS contains the version number, so LHS should not need to have > > "version" explicitly stated. > > I see, then part of your discussion is focused on this syntax: > > osf name version "Linux:4.20" > > in order to remove the "version" keyword there and make it more > compact. I really don't think it's needed. The only reason for it I can see is supporting a use-case where users pass "Linux:4.20" on RHS but actually don't want to match on version, but they could just omit it in the first place then. > > On Thu, Mar 14, 2019 at 09:13:09PM +0100, Pablo Neira Ayuso wrote: > > [...] > > > I think we could even extend this later on to match things like: > > > > > > # Popular cluster config scripts disable timestamps and > > > # selective ACK: > > > S4:64:1:48:M1460,N,W0: Linux:2.4:cluster:Linux 2.4 in cluster > > > ^^^^^^^^^^^^^^^^^ > > > Then, do: > > > > > > os gente "Linux:2.4:cluster" > > > > > > by adding a new flag to match the "Subtype" field (according to the > > > file description in pf.os). > > > > In an ideal world, we could match on any (combination of) fields in the > > database. I am aware this is probably over-engineering. :) > > We can probably achieve this with a more advance nft_cmp expression, > that allows us to do some sort of limited regex matching. But I agree > in that adding this only for the osf expression is probably too much. > I don't like regex, they will use it for layer-7, and users do not > understand the computational complexity of the regular expressions > (they can easily ruin performance by adding a few expression that need > to be search all over the packet payload). > > Anyway, this is a different topic :-). Yes, no point in making this widely used expression more complicated just for a case nobody uses eventually. :) > > What we could do though with little effort is to make use of the OS info > > structure in database by making use of nft_cmp comparing only the first > > 'len' bytes of data in registers. My idea would be that: > > > > * 'osf' expression always returns "full" data, i.e.: "OS:VER:SUB" > > * nft_cmp compares that string to RHS up to RHS length > > > > So let's assume DB lookup returns "Windows:2003:AS:", then: > > > > osf name "Windows" -> match > > osf name "Windows:" -> match > > osf name "Windows:XP:" -> no match > > osf name "Windows:2000:" -> no match > > osf name "Windows:200" -> match > > > > So we have optional version match and even a poor-man's wildcard > > functionality. Specifying the trailing semi-colon implicitly causes an > > exact match on the last field. > > > > What do you think? > > Hm, if we follow this path, this would need a bit more work, note > that: > > * nft userspace currently compares 16 bytes for the string case, > according to what I see via --debug=mnl. This is because osf_expr_alloc() sets expr->len to NFT_OSF_MAXGENRELEN * BITS_PER_BYTE (with NFT_OSF_MAXGENRELEN being defined to 16 in include/linux/netfilter/nf_tables.h). > * When the string is less than 16 bytes, it assumes it is a wildcard > matching and the end of the string. I guess that's because RHS length is then smaller than LHS? > * Kernel would need to inconditionally build the OS:version string. Yes, exactly. nft_osf.c would return "OS:version:subtype:opt" and nft_cmp.c will do the right thing if nft_cmp_expr->len is set to RHS length. > * We may need to ask users to break existing osf ruleset so they > explicitly add the wildcard at the end, ie. > > osf name "Windows\*" > osf name "Linux:4.\*" Not if cmp expr len is set correctly. > And the kernel would have no notion of what userspace is willing to > match. Yes, it boils down to a string comparison. Do you see a downside in that? > If the problem is the syntax, not the NFT_OSF_F_VERSION flags, we > could explore this syntax: > > osf genre "Linux" > osf version "Linux:4.20" > > then, in the future (if ever needed) add subtypes: > > osf subtype "Linux:2.4:cluster" IMHO this is still redundant: RHS contains :, user wants to match on OS and version. RHS contains ::, user wants to match on OS, version and subtype. > With flags in place, we would have a bit of knowledge of what the user > is doing vs. matching part of a string. Yes, but I don't see where this would be beneficial. Maybe in optimizing redundant expressions? Like | osf name "Linux" osf version "Linux:4.20" but it's not a big deal. With my approach described above, userspace could compare both RHS strings: * if memcmp(str1, str2, min(len1, len2) != 0 rule will never match * else keep only the expression with longer RHS. Cheers, Phil