summaryrefslogtreecommitdiffstats
Commit message (Collapse)AuthorAgeFiles
* wibbleGuilhem Moulin2015-06-071
|
* Install dovecot from backports (for imapc).Guilhem Moulin2015-06-075
| | | | | Interesting features include caching of mail headers (v2.2.8+) as well as new IMAP capabilities.
* Install Rouncube from backports.Guilhem Moulin2015-06-072
| | | | | | Recent versions have a whole bunch of bugfixes and nice new features: http://trac.roundcube.net/wiki/Changelog
* Don't require a PKI for IPSec.Guilhem Moulin2015-06-075
| | | | | | | | | | | Instead, generate a server certificate for each host (on the machine itself). Then fetch all these certs locally, and copy them over to each IPSec peer. That requires more certs to be stored on each machines (n vs 2), but it can be done automatically, and is easier to deploy. Note: When adding a new machine to the inventory, one needs to run the playbook on that machine (to generate the cert and fetch it locally) first, then on all other machines.
* Don't try to start smart on VMs.Guilhem Moulin2015-06-071
|
* Support non-free firmwares. (Can be required :-()Guilhem Moulin2015-06-072
| | | | Also, always install contrib's intel-microcode on Intel CPUs.
* wibbleGuilhem Moulin2015-06-076
|
* Assume a DNS entry for each role.Guilhem Moulin2015-06-0718
| | | | | | E.g., ldap.fripost.org, ntp.fripost.org, etc. (Ideally the DNS zone would be provisioned by ansible, too.) It's a bit unclear how to index the subdomains (mx{1,2,3}, etc), though.
* Decongestion potential bottlenecks on trivial_rewrite(8).Guilhem Moulin2015-06-0710
| | | | | | | | | | | | | Which might be caused by slow LDAP lookups in transport_maps. Instead, we alias each addresses for which we want a custom transport to a dedicated "dummy" domain, and use a static (CDB) transport_maps to map said domains to their transport; the receiver can then use canonical(8) to restore the original envelope recipient. Since the alias resolution is performed by cleanup(8), which can run in parallel with other instances, it should decongestion bottlenecks under heavy loads. So far only the MX:es have been decongestioned. The list manager and the MDA should be treated as well.
* typoGuilhem Moulin2015-06-071
|
* Add XXX comments for ad hoc fixes for some known bugs.Guilhem Moulin2015-06-072
| | | | (To be removed when the fix enters stable.)
* Follow Qualys's SSL labs recommendation for HTTPS.Guilhem Moulin2015-06-071
| | | | | (Disable SSLv3 and extend STS' max age to 180 days.) See https://www.ssllabs.com/ssltest/ .
* Replace mktemp's deprecated -t option by --tmpdir.Guilhem Moulin2015-06-071
| | | | | But not in the installer, as busybox's implementation of mktemp didn't deprecate -t/-p.
* Make use of Ansible 1.5 new features.Guilhem Moulin2015-06-076
| | | | Most notably pipelining=True and sysctl_set=yes.
* wibbleGuilhem Moulin2015-06-071
|
* Ansible automatically creates parent directories.Guilhem Moulin2015-06-072
|
* Use Debian's usual location for static web content.Guilhem Moulin2015-06-072
| | | | Hence put the CSS and fonts under /usr/share/.
* Make the *_maps file names uniform.Guilhem Moulin2015-06-077
| | | | That is, don't put a leading virtual_ or a trailing _maps in file names.
* wibbleGuilhem Moulin2015-06-073
|
* Fix the catch-all resolution again.Guilhem Moulin2015-06-0718
| | | | | | | | | | | | | | | | | | | | We introduce a limitation on the domain-aliases: they can't have children (e.g., lists or users) any longer. The whole alias resolution, including catch-alls and domain aliases, is now done in 'virtual_alias_maps'. We stop the resolution by returning a dummy alias A -> A for mailboxes, before trying the catch-all maps. We're still using transport_maps for lists. If it turns out to be a bottleneck due to the high-latency coming from LDAP maps, (and the fact that there is a single qmgr(8) daemon), we could rewrite lists to a dummy subdomain and use a static transport_maps instead: virtual_alias_maps: mylist@example.org -> mylist#example.org@mlmmj.localhost.localdomain transport_maps: mlmmj.localhost.localdomain mlmmj:
* Mailing lists (using mlmmj).Guilhem Moulin2015-06-0718
| | | | | | | | | Right now the list server cannot be hosted with a MX, due to bug 51: http://mlmmj.org/bugs/bug.php?id=51 Web archive can be compiled with MHonArc, but the web server configuration is not there yet.
* Remove list commands.Guilhem Moulin2015-06-072
| | | | | | They were only a dirty hack for list commands à la Mailman such as mylist-request. If we are to use another list manager such as mlmmj, which uses a VERP delimiter instead, the problem disappears.
* Don't pass the client information unless necessary.Guilhem Moulin2015-06-072
|
* Don't use IPSec to relay messages to localhost.Guilhem Moulin2015-06-074
|
* Excplicitely make local services run on localhost.Guilhem Moulin2015-06-074
|
* typoGuilhem Moulin2015-06-076
|
* Fix catchall resolution.Guilhem Moulin2015-06-0717
| | | | | | | | It has to be performed last, to give a chance to be accepted as a regular mailbox. We introduce a new, dedicated, smtpd daemon whose only purpose is to resolve catch-alls.
* Install haveged.Guilhem Moulin2015-06-072
| | | | | | To avoid low-entropy conditions, see http://www.issihosts.com/haveged/
* Install ClamAV.Guilhem Moulin2015-06-072
|
* Remove the 'fripostLocalAlias' attribute.Guilhem Moulin2015-06-076
| | | | | | | | | | | | | Instead, we pretend that lists are valid users (via a match in the mailbox_transport_maps) but choose a different transport (with the same request in transport_maps). The advantage is that we get rid of the ugly hack for list transport… A minor drawback is that we now have two LDAP lookups instead of one for non local addresses (ie, everything but reserved addresses). Hopefully the requests are cached; but even if they aren't, querying a local LDAP server is supposed to be cheap.
* Configure Sieve and ManageSieve.Guilhem Moulin2015-06-076
| | | | | Also, add the 'managesieve' RoundCube plugin to communicate with our server.
* Use a local IMAP caching proxy under the webmail.Guilhem Moulin2015-06-0712
| | | | | | | | | | | | (Unless the webmail is itself a full IMAP server.) It replaces RoundCube's own IMAP and message caches. Dovecot's IMAPC storage backend is not very documented, but provides smart IMAP proxying. References include: http://dovecot.org/pipermail/dovecot/2011-January/056975.html http://wiki2.dovecot.org/HowTo/ImapcProxy http://wiki2.dovecot.org/Migration/Dsync
* wibbleGuilhem Moulin2015-06-072
|
* Make the virtual mailboxes visible under RoundCube.Guilhem Moulin2015-06-077
| | | | | | RoundCubes lists subscribed mailboxes only (determined using LIST-EXTENDED by default); also, it seems to ignore new subscriptions to mailboxes not listed by the LIST command.
* Configure the webmail.Guilhem Moulin2015-06-0719
|
* Common web configuration.Guilhem Moulin2015-06-077
|
* Load relevant MySQL authplugins.Guilhem Moulin2015-06-073
| | | | Also, turn off all TCP/IP listener ports.
* Use postmulti to run postconf per instance.Guilhem Moulin2015-06-072
|
* Force expansion of escape sequences.Guilhem Moulin2015-06-073
| | | | | By using double quoted scalars, cf. https://groups.google.com/forum/#!topic/ansible-project/ZaB6o-eqDzw
* Compile Spamassassin rules.Guilhem Moulin2015-06-073
| | | | See /usr/share/doc/spamassassin/README.Debian.gz
* Auto-update Spamassassin's ruleset.Guilhem Moulin2015-06-071
|
* wibbleGuilhem Moulin2015-06-076
|
* Configure dovecot's antispam filter.Guilhem Moulin2015-06-076
| | | | | | | | | | | | | Mails to be retrained are stored in the spooldir /home/mail/spamspool; later a daemon catches them up and feed them to sa-learn(1p). (On busy systems batch-process the learning should be much more efficient.) The folder transisition matrix along with the corresponding actions can be found there: http://hg.dovecot.org/dovecot-antispam-plugin/raw-file/5ebc6aae4d7c/doc/dovecot-antispam.7.txt See also dovecot-antispam(7).
* Enable IMAP virtual mailboxes.Guilhem Moulin2015-06-077
| | | | | | | | | | | | | | Using dovecot's 'virtual' plugin, cf. http://wiki2.dovecot.org/Plugins/Virtual The 'virtual/' namespace is visible in the NAMESPACE command (hidden=no), but not in LIST (list=no). This should ensure that the namespace isn't automatically synced by offlineimap, but nevertheless visible by roundcube, cf. http://trac.roundcube.net/ticket/1486796 http://mailman2.u.washington.edu/pipermail/imap-protocol/2010-May/001076.html
* wibbleGuilhem Moulin2015-06-0711
|
* Include amavisd-new's LDAP schema.Guilhem Moulin2015-06-071
| | | | | | It'd certainly be nicer if we didn't have to deploy amavis' schema everywhere, but we need the 'objectClass' in our replicates, hence they need to be aware of the 'amavisAccount' class.
* Configure the content filter.Guilhem Moulin2015-06-0714
| | | | | | | | | | | Antispam & antivirus, using ClamAV and SpamAssassin through Amavisd-new. Each user has his/her amavis preferences, and own Bayes filter (to maximize privacy). One question remains, though: how to set spamassassin's trusted_networks / internal_networks / msa_networks? It seems not obivious to get it write with IPSec and dynamic IPs. (Cf. https://wiki.apache.org/spamassassin/AwlWrongWay)
* bugfixGuilhem Moulin2015-06-071
|
* Convert legacy *.schema into *.ldif.Guilhem Moulin2015-06-071
|
* wibbleGuilhem Moulin2015-06-072
|