** Description changed: [ Current State of amavisd-new ] The latest merge of amavisd-new changed a perl dependency from libberkeleydb-perl (main) to libzmq-ffi-perl (universe), triggering a component mismatch that needs to be resolved. Per static reports, libzmq-ffi-perl has many transitive dependencies (ten binaries from seven sources) that would require MIRing. I don't think this is worth doing (or worth patching back to the previous library), and instead I think the package should be dropped from main, for the following reasons: 1) The original MIR [1] for this package was made in 2008, and upstream release cadence has slowed to roughly 1 per year, with the - primary maintainer stepping back in 2018 [2]. As such, maintainability might come into question when there are more active alternatives (see below). + primary maintainer stepping back in 2018 [2]. As such, maintainability might come into question when there are more active alternatives (see below). 2) The companion packages for amavisd, namely - - amavisd-milter "amavisd-new interface for milter-capable MTAs" - * Note: not to be confused with amavisd*-new-*milter, a long dead binary NACK'd in [1] which this package was meant to replace. - - amavisd-new-postfix (a binary built from amavisd-new), the packaged way to wire it into postfix - are both in universe, so the convenient methods of setting amavisd up are already in universe. + - amavisd-milter "amavisd-new interface for milter-capable MTAs" + * Note: not to be confused with amavisd*-new-*milter, a long dead binary NACK'd in [1] which this package was meant to replace. + - amavisd-new-postfix (a binary built from amavisd-new), the packaged way to wire it into postfix + are both in universe, so the convenient methods of setting amavisd up are already in universe. 3) The original rationale for the MIR [3] states thus: - > Amavisd-new is the upstream recommended approach for integrating anti-virus and spam filtering into Postfix - > and so bringing this into main gives a common, supported package for this (and also brings some helpful functionality - > of its own). + > Amavisd-new is the upstream recommended approach for integrating anti-virus and spam filtering into Postfix + > and so bringing this into main gives a common, supported package for this (and also brings some helpful functionality + > of its own). - This is no longer true. Postfix documents a number of different mechanisms for the filtering capabilities provided by amavisd: The milter protocol [4] and - after-queue content filtering [5]. The latter is amavisd's method and the page doesn't mention amavisd. The primary mention of amavisd comes in the general - documentation on postscreening [6] which mentions it alongside spamassassin (also in main) and the milter protocol generally. This, combined with the add-ons - page [7] mentioning many milter plugins to cover some of the functionality implies a strong ecosystem preference towards the milter protocol. + This is no longer true. Postfix documents a number of different mechanisms for the filtering capabilities provided by amavisd: The milter protocol [4] and + after-queue content filtering [5]. The latter is amavisd's method and the page doesn't mention amavisd. The primary mention of amavisd comes in the general + documentation on postscreening [6] which mentions it alongside spamassassin (also in main) and the milter protocol generally. This, combined with the add-ons + page [7] mentioning many milter plugins to cover some of the functionality implies a strong ecosystem preference towards the milter protocol. - It is also worth noting that amavisd itself does not support the + It is also worth noting that amavisd itself does not support the milter protocol, one needs the aforementioned amavisd-milter package (universe) to use it. + 4) There are no reverse dependencies that require addressing. The only + ones are the two aforementioned universe packages, amavisd-milter and + amavisd-new-postfix, both in universe. [ Proposed Replacement for amavisd-new ] The clearest candidate for promotion to replace amavisd-new to maintain officially supported features is rspamd. It has a very active upstream [8] and supports the milter protocol, meaning it also can work perfectly well with postfix. We currently have the package, with some small delta from Debian related to build flags (and some OpenSSL build fixes likely to get removed as upstream/Debian adopt them). MIRing this package would require the elevation of several dependencies, however, each requiring varying levels of changes to rspamd's packaging: - hyperscan - the most problematic of the dependencies - hyperscan is a regular expression matching library by Intel. In the past, this library was fully open source (BSD license), but after 5.4.2 Intel took the project proprietary. - Intel still backports security fixes to the latest open source release of hyperscan at [9], but even then neither Ubuntu nor Debian is up to date on the security fixes that already - exist. + hyperscan is a regular expression matching library by Intel. In the past, this library was fully open source (BSD license), but after 5.4.2 Intel took the project proprietary. + Intel still backports security fixes to the latest open source release of hyperscan at [9], but even then neither Ubuntu nor Debian is up to date on the security fixes that already + exist. - vectorscan - fixes the problem of hyperscan - vectorscan is an actively maintained open source fork of hyperscan + vectorscan is an actively maintained open source fork of hyperscan located at [10]. Already conditionally present in the dependency closure of rspamd: - libhyperscan-dev [amd64 i386] | libvectorscan-dev [arm64], + libhyperscan-dev [amd64 i386] | libvectorscan-dev [arm64], - as a replacement for hyperscan. An rspamd MIR would likely involve + as a replacement for hyperscan. An rspamd MIR would likely involve replacing hyperscan with vectorscan on those architectures where we currently use hyperscan (amd64, i386). - vectorscan also has one transitive dependency, isa-support, that + vectorscan also has one transitive dependency, isa-support, that requires MIR, which is straightforward. - bootstrap-html - static files for rspamd's optional GUI. - This requires one dependency of its own to be promoted (node-popper2), and consists of only static files. - I think the sanest option here might be to just introduce a delta dropping it to Suggests. + This requires one dependency of its own to be promoted (node-popper2), and consists of only static files. + I think the sanest option here might be to just introduce a delta dropping it to Suggests. - - requirejs - also required for the optional GUI only, same recommendation as bootstrap-html - + - requirejs - also required for the optional GUI only, same + recommendation as bootstrap-html + - luajit - Being a JIT compiler, security will need to take a closer look at this one given the attack surface of executing code, which might be concerning given that - the original, legacy upstream is maintained by a single person. However, Ubuntu via Debian already ships a fork of luaJIT maintained by OpenResty [11], which should - mitigate those concerns. Furthermore, as far as rspamd is concerned, luajit does not run any end-user supplied untrusted code, only rspamd's own. + Being a JIT compiler, security will need to take a closer look at this one given the attack surface of executing code, which might be concerning given that + the original, legacy upstream is maintained by a single person. However, Ubuntu via Debian already ships a fork of luaJIT maintained by OpenResty [11], which should + mitigate those concerns. Furthermore, as far as rspamd is concerned, luajit does not run any end-user supplied untrusted code, only rspamd's own. No transitive dependencies require promotion, save for those aforementioned. Note also rspamd has a segfault bug (https://bugs.launchpad.net/ubuntu/+source/rspamd/+bug/1947495) on arm64 that will need re-triage and evaluation before the MIR can proceed. [ Suggested Next Steps ] The component mismatch needs to be resolved sooner rather than later in one direction or the other, and for the reasons above I think we should drop the package out of main. As for its replacement, I think rspamd should be elevated, but I think it is okay for there to be a gap between demotion of one and promotion of the other. We are far enough away from the next LTS that we should be able to promote rspamd in time. This requires dropping amavisd out of one seed, supported-misc-servers. The MP doing this is linked to this bug. - [1] https://bugs.launchpad.net/ubuntu/+source/amavisd-new/+bug/183418 [2] https://www.mail-archive.com/[email protected]/msg05361.html - ANNOUNCE: Amavis project handed over to new project leaders (Mark Martinec, Oct 2018) [3] https://web.archive.org/web/20220926220148/https://wiki.ubuntu.com/MainInclusionReportAmavisd-new [4] http://www.postfix.org/MILTER_README.html [5] http://www.postfix.org/FILTER_README.html [6] https://www.postfix.org/POSTSCREEN_README.html [7] https://www.postfix.org/addon.html [8] https://github.com/rspamd/rspamd, last commit 5 hours ago at time of writing [9] https://github.com/intel/hyperscan [10] https://github.com/VectorCamp/vectorscan [11] https://launchpad.net/ubuntu/+source/luajit/2.1.0+openresty20251030-1
-- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2175041 Title: demote amavisd-new from main to universe To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/amavisd-new/+bug/2175041/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
