In the earlier commit, I added a Dovecot userdb lookup. Without a userdb lookup, Dovecot would use the password db for user lookups. With a userdb lookup we can support iterating over users.
But I forgot the WHERE clause in the query, resulting in every incoming message being accepted if the user database contained any users at all. Since the mailbox path template is the same for all users, mail was delivered correctly except that mail that should have been rejected was delivered too.
On some machines localhost is defined as something other than 127.0.0.1, and if we mix "127.0.0.1" and "localhost" then some connections won't be to to the address a service is actually running on.
This was the case with DKIM: It was running on "localhost" but Postfix was connecting to it at 127.0.0.1. (https://discourse.mailinabox.email/t/opendkim-is-not-running-port-8891/1188/12.)
I suppose "localhost" could be an alias to an IPv6 address? We don't really want local services binding on IPv6, so use "127.0.0.1" to be explicit and don't use "localhost" to be sure we get an IPv4 address.
Fixes#797
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJW/mJqAAoJELkgQfTBC92B/F8H/2s6wKhzzeoqkhLU2nvYJh0B
Q1d0SbtdQWIWrTQbcjIR3aGYwJzJ+HC7rylrwS4lB2ugpJBA0MnfD+ktwbe/EyDa
pN6WLlmnXyAw28//ubq0FQqC8Gawsj4WMfmSEw/XuDShik8XJmU7QUEnewClJ7So
ko4eVp9KL8MU3Rj/DebhyoW0EjpB/qrJvLSqtj4KCxKYES9J8nUVBFVRDL48yNx4
2KTIjqreGZmtW0/wxPnganMeV6DZn3B6vBmqOYYvw7bf6r/cY0ZkNK/ENlo+ntJD
3jFKki4TJChhGVWH5T4Tw2bys4Cua1+SA3cleNRH1rYSvRWyOCwK+LS4YBJHYp4=
=umMp
-----END PGP SIGNATURE-----
merge hotfix release tag 'v0.17c' into master
The hotfixes were all already applied to master in original PRs. This merge merely brings over the CHANGELOG and the updated install instructions (v0.17b=>v0.17c), including to bootstrap.sh which is what triggers v0.17c being the latest release.
----------------------
ownCloud moved their source code to a new location, breaking our installation script.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJW1YnLAAoJELkgQfTBC92BZg8H/RVfJRrhkM6B0UocYOmp2LQB
u6b/KxSM5JbBq9BTyQTAkQ0Dj6caWpBwAATGgxvyAYt0Du/QTx/ycfgenkP2DWcI
UYm+AkV2Eg9BiXpFqG33vJ/aWjATB+S3XM+77jMbeMb4kK+PfOHguvmxnsS4x8HY
h+unkXPy8fr1rATaOGg32fTZrKkVWAriHFfJqy3kkoTHS+9fj7MVGbTBPntPaC46
/42IsCFiQsNdz6QeHmXzFnP6AuqmkbcN9dpsFe36qfSfwVhNqC776lxxP+r1nK3p
AWs/q5VQ4V1ght3MzDch9dOlmGuqBVB1WkGdQVV7RemYrdNi5S1h+dXXUDiNXY8=
=C7LW
-----END PGP SIGNATURE-----
merge point release v0.17b
ownCloud moved their source code to a new location, breaking our installation script.
https://discourse.mailinabox.email/t/status-check-emails-empty-after-upgrading-to-v0-16/1082/3
A user on that thread suggests an alternate solution, adding `PYTHONIOENCODING=utf-8` to `/etc/environment`. Python docs say that affects stdin/out/err. But we also use these environment variables elsewhere to ensure that config files we read/write are opened with UTF8 too. Maybe all that can be simplified too.
The default timeout for Postfix's maximal_queue_lifetime and bounce_queue_lifetime is 5 days. This is way too long if you expect someone to have an answer and after 5 days you'll get the message that it's not delivered. This disrupts communication. It would be more responsive if the user got the 'can't deliver' error after 24 hours.
Sending mail through Exchange/ActiveSync (Z-Push) had been broken since v0.14. This is now fixed.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJWkQ9rAAoJELkgQfTBC92BPjQH/ibFBZHgma7C53Q4X9iPfUk2
dPK75rNBx06d6yW4LNAYuWVnNO1Mb0khb2k2UPwg4noImYWqsNS9hXih7C6oPHiK
Szz4ubCc5MEqnmhxiNkzLdIBvsyKmz8IfmCl+LCXu8uk0Fb+pB6zbSdAGxjtaSPL
itCwz8+ApTC4bl1CoNYPn2zudHmHNeC7L6INYdb+xbtc/Tz6mO/xMaBVPDiKeq9P
LqLTOXiJNENz7vKSZytlWGOTdtSTZqwd7JBXuBg0QFz5C9yg8EV4LWB9wOm5aTIf
Fol3WK5ZHA5YeihZKmZSjz9+p4iwv5hqR5osKL2n46LeVHyafESl+QnZSXRlDjE=
=i3ei
-----END PGP SIGNATURE-----
Merge release branch v0.15a
v0.15a
Sending mail through Exchange/ActiveSync (Z-Push) had been broken since v0.14. This is now fixed.
-----------------------
Mail:
* Updated Roundcube to version 1.1.3.
* Auto-create aliases for abuse@, as required by RFC2142.
* The DANE TLSA record is changed to use the certificate subject public key rather than the whole certificate, which means the record remains valid after certificate changes (so long as the private key remains the same, which it does for us).
Control panel:
* When IPv6 is enabled, check that system services are accessible over IPv6 too, that the box's hostname resolves over IPv6, and that reverse DNS is setup correctly for IPv6.
* Explanatory text for setting up secondary nameserver is added/fixed.
* DNS checks now have a timeout in case a DNS server is not responding, so the checks don't stall indefinitely.
* Better messages if external DNS is used and, weirdly, custom secondary nameservers are set.
* Add POP to the mail client settings documentation.
* The box's IP address is added to the fail2ban whitelist so that the status checks don't trigger the machine banning itself, which results in the status checks showing services down even though they are running.
* For SSL certificates, rather than asking you what country you are in during setup, ask at the time a CSR is generated. The default system self-signed certificate now omits a country in the subject (it was never needed). The CSR_COUNTRY Mail-in-a-Box setting is dropped entirely.
System:
* Nightly backups and system status checks are now moved to 3am in the system's timezone.
* fail2ban's recidive jail is now active, which guards against persistent brute force login attacks over long periods of time.
* Setup (first run only) now asks for your timezone to set the system time.
* The Exchange/ActiveSync server is now taken offline during nightly backups (along with SMTP and IMAP).
* The machine's random number generator (/dev/urandom) is now seeded with Ubuntu Pollinate and a blocking read on /dev/random.
* DNSSEC key generation during install now uses /dev/urandom (instead of /dev/random), which is faster.
* The $STORAGE_ROOT/ssl directory is flattened by a migration script and the system SSL certificate path is now a symlink to the actual certificate.
* If ownCloud sends out email, it will use the box's administrative address now (admin@yourboxname).
* Z-Push (Exchange/ActiveSync) logs now exclude warnings and are now rotated to save disk space.
* Fix pip command that might have not installed all necessary Python packages.
* The control panel and backup would not work on Google Compute Engine because GCE installs a conflicting boto package.
* Added a new command `management/backup.py --restore` to restore files from a backup to a target directory (command line arguments are passed to `duplicity restore`).
v0.14 (November 4, 2015)
------------------------
Mail:
* Spamassassin's network-based tests (Pyzor, others) and DKIM tests are now enabled. (Pyzor had always been installed but was not active due to a misconfiguration.)
* Moving spam out of the Spam folder and into Trash would incorrectly train Spamassassin that those messages were not spam.
* Automatically create the Sent and Archive folders for new users.
* The HTML5_Notifier plugin for Roundcube is now included, which when turned on in Roundcube settings provides desktop notifications for new mail.
* The Exchange/ActiveSync backend Z-Push has been updated to fix a problem with CC'd emails not being sent to the CC recipients.
Calender/Contacts:
* CalDAV/CardDAV and Exchange/ActiveSync for calendar/contacts wasn't working in some network configurations.
Web:
* When a new domain is added to the box, rather than applying a new self-signed certificate for that domain, the SSL certificate for the box's primary hostname will be used instead.
* If a custom DNS record is set on a domain or 'www'+domain, web would not be served for that domain. If the custom DNS record is just the box's IP address, that's a configuration mistake, but allow it and let web continue to be served.
* Accommodate really long domain names by increasing an nginx setting.
Control panel:
* Added an option to check for new Mail-in-a-Box versions within status checks. It is off by default so that boxes don't "phone home" without permission.
* Added a random password generator on the users page to simplify creating new accounts.
* When S3 backup credentials are set, the credentials are now no longer ever sent back from the box to the client, for better security.
* Fixed the jumpiness when a modal is displayed.
* Focus is put into the login form fields when the login form is displayed.
* Status checks now include a warning if a custom DNS record has been set on a domain that would normally serve web and as a result that domain no longer is serving web.
* Status checks now check that secondary nameservers, if specified, are actually serving the domains.
* Some errors in the control panel when there is invalid data in the database or an improperly named archived user account have been suppressed.
* Added subresource integrity attributes to all remotely-sourced resources (i.e. via CDNs) to guard against CDNs being used as an attack vector.
System:
* Tweaks to fail2ban settings.
* Fixed a spurrious warning while installing munin.
Seems like ownCloud 8.1.1 now doesn't play nice with trusted_domains. Whatever is put in ahead of time gets reset to an array containing 'localhost' only, probably because we invoke autoconfiguration from the command line where it doesn't know the hostname it's being accessed from. We now set this value after running autoconfig.
This has the added benefit of also fixing the problem that if PRIMARY_HOSTNAME changes, trusted_domains wasn't updated. Now it is. Fixes#503.
See #514.
Work-around for ownCloud 8.1.1 upgrade bug and tweaking munin's setup.
v0.13a (August 23, 2015)
------------------------
Note: v0.13 (no 'a', August 19, 2015) was pulled immediately due to an ownCloud bug that prevented upgrades. v0.13a works around that problem.
Mail:
* Outbound mail headers (the Recieved: header) are tweaked to possibly improve deliverability.
* Some MIME messages would hang Roundcube due to a missing package.
* The users permitted to send as an alias can now be different from where an alias forwards to.
DNS:
* The secondary nameservers option in the control panel now accepts more than one nameserver and a special xfr:IP format to specify zone-transfer-only IP addresses.
* A TLSA record is added for HTTPS for DNSSEC-aware clients that support it.
System:
* Backups can now be turned off, or stored in Amazon S3, through new control panel options.
* Munin was not working on machines confused about their hostname and had lots of errors related to PANGO, NTP peers and network interfaces that were not up.
* ownCloud updated to version 8.1.1 (with upgrade work-around), its memcached caching enabled.
* When upgrading, network checks like blocked port 25 are now skipped.
* Tweaks to the intrusion detection rules for IMAP.
* Mail-in-a-Box's setup is a lot quieter, hiding lots of irrelevant messages.
Control panel:
* SSL certificate checks were failing on OVH/OpenVZ servers due to missing /dev/stdin.
* Improve the sort order of the domains in the status checks.
* Some links in the control panel were only working in Chrome.
* let munin choose what it wants to monitor, it may need an update periodically
* deactivate monitoring of NTP peers (this makes no sense)
* deactivate monitoring of network interfaces that aren't up (also makes no sense)
* we get errors in munin-node.log if we don't create /var/lib/munin-node/plugin-state/
The antispam-plugin configuration is not ignoring deleted email (mail sent to a trash folder). This causes a big problem because if someone decides to "clean up" their Spam folder by deleting all of the emails, then sa-learn --ham is run on what is usually correctly caught spam messages. This causes big problems with the accuracy of the bayes scoring! It should really only be learning ham if someone drags the email to a non-Trash folder.
I ran "sa-learn --dbpath /home/user-data/mail/spamassassin --dump magic" before these changes and saw the nham counter increment with every message deleted from Spam. With this new config sa-learn --ham is not run when email is deleted from Spam so the bayes database is not incorrectly trained.
v0.13 (August 19, 2015)
-----------------------
Mail:
* Outbound mail headers (the Recieved: header) are tweaked to possibly improve deliverability.
* Some MIME messages would hang Roundcube due to a missing package.
* The users permitted to send as an alias can now be different from where an alias forwards to.
DNS:
* The secondary nameservers option in the control panel now accepts more than one nameserver and a special xfr:IP format to specify zone-transfer-only IP addresses.
* A TLSA record is added for HTTPS for DNSSEC-aware clients that support it.
System:
* Backups can now be turned off, or stored in Amazon S3, through new control panel options.
* Munin was not working on machines confused about their hostname.
* ownCloud updated to version 8.1.1, its memcached caching enabled.
* When upgrading, network checks like blocked port 25 are now skipped.
* Tweaks to the intrusion detection rules for IMAP.
* Mail-in-a-Box's setup is a lot quieter, hiding lots of irrelevant messages.
Control panel:
* SSL certificate checks were failing on OVH/OpenVZ servers due to missing /dev/stdin.
* Improve the sort order of the domains in the status checks.
* Some links in the control panel were only working in Chrome.
Contacts and Calendar are now versioned separately from ownCloud core.
Contacts appears to be in a v0.4.x.x series. This is the hash of v0.4.0.0.
Per https://github.com/owncloud/calendar/issues/892, Calendar is in a 0.7.x series. This is the hash for 0.7.2.
Backup location and maximum age can now be configured in the admin panel.
For now only S3 is supported, but adding other duplicity supported backends should be straightforward.
Some users report munin is broken because munin and munin-node disagree about the name of the machine. I think this occurs if hostname (used by munin-node) reports a different name than PRIMARY_HOSTNAME (which we put in the munin config).
Hard-code PRIMARY_HOSTNAME in munin-node.conf.
Fixes#474.
See https://discourse.mailinabox.email/t/404-not-found-on-admin-munin/623/24.