* feat: Change Postfix smtpd_tls key and cert files to chain_files
Since Postfix 3.4, `smtpd_tls_cert_file` and `smtpd_tls_key_file` have been deprecated in favor of `smtpd_tls_chain_files` which supports a list of values where a single or sequence of file paths provide a private key followed by it's certificate chain.
* feat: Dual certificate support
`smtpd_tls_chain_files` allows for multiple key+cert bundles so that you can provide different key types, such as ECDSA and RSA.
To maintain compatibility with the current CERT/KEY ENV vars only a 2nd certificate is supported.
Since Dovecot 2.2.31 a related feature is also available, but it is limited to only providing one alternative certificate via separate cert and key settings.
---
This feature enables support for multiple certificates, eg for serving modern ECDSA certs with RSA as fallback.
* chore: Refactor variable names to meet style guide
Improved some comments too.
* chore: Have function definitions respect style guide
* chore: Minor edits to comments
* chore: Expand on comments for maintenance, alert of insecure config
When `SSL_TYPE` isn't properly setup, we're still offering SSL connections but not warning in logs about the insecurity of such, or why a misconfiguration may have occurred.
This commit more clearly communicates to the user that they should look into the issue before considering deploying to production.
The `TODO` comments communicate to any future maintainer to consider treating these improper configs as disabling TLS instead.
* fix: Use `snakeoil` cert
I mistakenly thought this was placeholder text, which broke some tests. This adds the two files in the correct order (private key followed by cert/chain), to fix that issue.
* fix: Disable alt cert for Dovecot if necessary
Certain scenarios may persist state of previously configured alt cert via ENV vars that are removed from a future run. If the config is not reset to original immutable state, this will correctly disable the config from using alt cert unintentionally.
* fix: Satisfy ShellCheck lint
By switching from string var to array / list expansion, this better stores the extracted result and applies it in a manner that ShellCheck linting approves, removing the need to disable the rule.
* feat: Support dual cert test
Few tweaks to the test script allows re-purposing it for covering dual cert support as well.
* chore: Rearranged cert and key lines
A little reorganization, mostly placing private key ahead of related cert lines.
* chore: Refactor `_set_certificate`
This should make the parameters a little less confusing.
Previously was 3 parameters, but the Postfix parameter (1st) may look like two variables if you don't pay attention to the surrounding quotes; while the Dovecot parameters (2nd + 3rd) would have an opposing order. There was also a variant where the `FULLKEYCHAIN` var was passed in three times.
Now it's two params, with the 2nd param as an optional one. If the 2nd param is provided, then the two params are in the order of private key then certificate, otherwise if only a single parameter it's a single PEM file with the full cert chain and private key bundled.
This avoids implying that Postfix and Dovecot might use different files.
* chore: Document current state of `SSL_TYPE` logic better
Inlined for the benefit of anyone else maintaining this section if I'm unable to address the concerns within my own time.
* docs: ENV vars
`TLS_LEVEL=old` isn't in the codebase anymore, not likely to be relevant to retain.
No point in documenting what is considered invalid / unsupported config value in the first place for `SSL_TYPE`.
`SSL_TYPE=manual` was missing documentation for both related file path ENV vars, they've been added along with their alt fallback variants.
* chore: Update Dovecot LMTP SSL test config
Not sure how relevant this is, the file isn't complete sync with the main dovecot `10-ssl.conf` config, adding the support just in case.
* chore: Rename `FULLKEYCHAIN` to avoid confusion
There doesn't appear to be a standardized name for this type of file bundle, and `keychain` may be misleading (fullkeychain often provides macOS keychain results on search engines).
Opting for a more explicit `KEY_WITH_FULLCHAIN` name instead.
* fix: Invalid var name
`_set_certificate` refactor commit accidentally changed a var name and committed that breaking the dual cert support (thanks tests!).
* test: Refactor `mail_ssl_manual.bats`
Proper test return values instead of `wc -l` based checking.
Tests with dual cert support active, tests that feature (to better detect failure case.
Third test case was unable to verify new self-signed certificate, added new certs signed with self-signed root CA.
Adjusted openssl `CApath` parameter to use `CAfile` instead as `letsencrypt` cert was replaced thus CA cert is missing from the system trust store.
* test: Properly check for files in `mail_ssl_manual.bats`
Fixes lint error.
Also realized I was accidentally asserting a file exists in the test environment, not within the container.
Resolved that and also added an additional test case to ensure the ENV var files are valid when passed in, in the event a change misconfigures them and that the issue is identified earlier.
* chore: Apply PR review feedback
Better format some strings that had mixed quotes when they weren't necessary.
Additionally DRYed up the config path for Postfix and Dovecot within the `_setup_ssl` method.
Co-authored-by: Georg Lauterbach <infrastructure@itbsd.com>
To prevent announcing software or version to malicious people or scripts, it is advised to hide such information.
This information is provided as part of the Lynis community project. It is related to Lynis control MAIL-8818 and should be considered as-is and without guarantees.
https://cisofy.com/lynis/controls/MAIL-8818/
[Postfix docs](http://www.postfix.org/postconf.5.html#tls_ssl_options):
> Disable SSL compression even if supported by the OpenSSL library. Compression is CPU-intensive, and compression before encryption does not always improve security.
[Postfix mailing list discussion](http://postfix.1071664.n5.nabble.com/patch-mitigate-CRIME-attack-td57978.html):
> The CRIME attack does not apply to SMTP, because unlike SMTP, there is no javascript in SMTP clients that makes them send thousands of email messages with chosen plaintext compressed together in the same packet with SASL credentials or other sensitive data.
> The auditor completely failed to take the context into account.
[Mailing list discussion of potential compression CRIME-like attack](https://lists.cert.at/pipermail/ach/2014-December/001660.html)
> keeping compression disabled is a good idea.
If you need a good test score, PCI compliance will likely flag compression despite not having any known risk with non-HTTP TLS.
* remove two ciphers according to https://www.htbridge.com/ssl/ (NIST, HIPAA)
* added a switch via an environment variable to choose between modern and intermediate ciphers
* new setup.sh function, new tests, new script and some minor updates to main.cf
* fix for missing files
* removed obsolete test-files
* restart postfix if neccessary.
* see pr #845
* fixed typo
* fixed branchmixup
* changed postfix reload command & changed to operate on container instead of image
* reload postfix only on adding new restriction
* main.cf is only changed when user is added.
- Postfix reload changed
- working on container instead of image now in setup.sh
- added cleanup after tests
* moved cleanup to makefile
* Introduced Postscreen
cheaper, earlier and simpler blocking of zombies/spambots.
From http://postfix.cs.utah.edu/POSTSCREEN_README.html :
As a first layer, postscreen(8) blocks connections from zombies and other spambots that are responsible for about 90% of all spam. It is implemented as a single process to make this defense as cheap as possible.
Things we need to consider:
- Do we need a whitelist/backlist file? (http://postfix.cs.utah.edu/postconf.5.html#postscreen_access_list)
- Via introducing an optional config/postfix-access.cidr
- The only permanent whitelisting I could imagine are monitoring services(which might (still?) behave weird/hastely) or blacklisting backup servers(since no traffic should be coming from them anyway)
- Do we need deep inspections? They are desireable, but these tests are expensive: a good client must disconnect after it passes the test, before it can talk to a real Postfix SMTP server. Considered tests are:
- postscreen_bare_newline_enable (http://postfix.cs.utah.edu/postconf.5.html#postscreen_bare_newline_action)
- postscreen_non_smtp_command_enable (http://postfix.cs.utah.edu/postconf.5.html#postscreen_non_smtp_command_action)
- postscreen_pipelining_enable (http://postfix.cs.utah.edu/postconf.5.html#postscreen_pipelining_action)
- Do we need to make the blacklisting via dnsblocking configurable? It's currently set and weighted as follows, where a score of 3 results in blocking, a score of -1 results in whitelisting:
(*: adds the specified weight to the SMTP client's DNSBL score. Specify a negative number for whitelisting.)
(http://postfix.cs.utah.edu/postconf.5.html#postscreen_dnsbl_sites)
- zen.spamhaus.org*3
- bl.mailspike.net
- b.barracudacentral.org*2
- bl.spameatingmonkey.net
- bl.spamcop.net
- dnsbl.sorbs.net
- psbl.surriel.com
- list.dnswl.org=127.0.[0..255].0*-2
- list.dnswl.org=127.0.[0..255].1*-3
- list.dnswl.org=127.0.[0..255].[2..3]*-4
- What to do when blacklisting? I currently set it to drop. We could
- ignore: Ignore the failure of this test. Allow other tests to complete. Repeat this test the next time the client connects. This option is useful for testing and collecting statistics without blocking mail.
- enforce: Allow other tests to complete. Reject attempts to deliver mail with a 550 SMTP reply, and log the helo/sender/recipient information. Repeat this test the next time the client connects.
- drop: Drop the connection immediately with a 521 SMTP reply. Repeat this test the next time the client connects.
In the end I think we could drop postgrey support. Postscreen replaces postgrey in its entirety, while being more selective and not delaying mail. Especially if we consider using the deep inspection options of postscreen.
Hope that wasn't too much to read! ;)
* main.cf got misformatted..
Don't know how, should be ok now.
* fixed malformatted main.cf & repaired master.cf
* reenabled rbl stuff.. It's cached, therefore doesn't hurt
* fixed tests
* added tests, repaired tests, added info, introduced new Variable POSTSCREEN_ACTION, fixes
* Switch to stretch-slim as base image.
- first step correct the testdata, as newer packages are more strict
about the mail-structure.
* Switch to stretch-slim: correcting the test-environment and the build
- add missing build-step to make
- clean the userdb aswell
- use timeout of netcat, as postgrey would not close the connection
- there is 2 extra mail-logs -> assert_output 5
- cosmetic: use "" instead of ''
* Switch to stretch-slim:
new image:
- smaller size
- 0 CVEs compared to 11 CVEs in ubuntu 16.04 Image
better backport situation
- postfix 3.1.6 vs 3.1.0
- fail2ban 0.9.6 vs 0.9.3
...
changes needed because of stretch-slim:
- add missing gnupg and iproute2 package
- remove non-free rar, unrar-free should do
- rsyslog does not add syslog user and has different conf-structure
- pyzor command discover was deprecated and is missing in the new
stretch package
- dovecot does not know SSLv2 anymore. removed because of warnings in
log
- iptables does not know imap3, IMAP working group chose imap2 in favor
of imap3
* Switch to debian stretch slim:
SSLv2 seems to be a not known protocol anymore - good!
* switch to debian stretch slim:
make this test more stable. there might be more than only one mail.log
(mail.info, mail.warn, ...)
* switch to debian stretc slim:
new openssl 1.1.0 needs stronger ciphers, removed some weekers ones.
Please, look through the new list of cipher! this needs to be done in
another commit for all other SSL/TLS-Endpoints aswell.
* Switch to debian stretch slim:
let our server pre-empt the cipher list.
Did a read through, wwwDOTpostfixDOTorg/FORWARD_SECRECY_READMEDOThtml
and
wwwDOTpostfixDOTorg/TLS_READMEDOThtml
* Switch to debian stretch slim: lets give this openssl-based test a new and independent but identical container. many other test on the main 'mail' container might interfere here.
* Switch to debian stretch slim: remove unused lines
* Adding the PERMIT_DOCKER option
See README.md for more informations
* Adding some test for PERMIT_DOCKER option
* Fix test cases
* Opendkim and Openmarc configuration
Fix docker network range
Adding opendkim and openmarc configuration
* Adding some options for tests
* Update log message
* Update tests
The dovecot-sieve plugin is installed and configured to apply sieve
as soon as a .dovecot.sieve file is encountered in the virtual user's
home directory (that is /var/mail/${domain}/${username}/.dovecot.sieve).
Transport has been changed in the postfix configuration to use
Dovecot LDA (see http://wiki.dovecot.org/LDA/Postfix) to actually
enable sieve filtering.
Tests have been added.
mail servers using an enhanced client, sender and helo
restriction configuration.
The configuration has been adapted using this blog post:
https://www.webstershome.co.uk/2014/04/07/postfix-blocking-spam-enters-server/
Basically mail servers having invalid configuration (as e.g. sending
from and dynamic IP or a misconfigured hostname) will have their
mails rejected.
Additionnally three RBL servers are used to detect spam sending
IPs: dnsbl.sorbs.net, zen.spamhaus.org and bl.spamcop.net.
The results of a 12h test drive using a 100+ daily spam
mail account (SpamAssasin was always enabled, just counting
delivered mails to inbox not counting what SA detected):
- Before: 34 incoming mails
- Afer change: 6 incoming mails (82% reduction)
Fixes#161.
Brennan Kinney
Georg Lauterbach
youtous
millerjason
n00dl3
Paul Adams
Marek Walczak
17Halbe
TechnicLab
Jurek Barth
Johan Smits
Daniel S. Reichenbach
Kai Ren
Thomas VIAL
Daniele Bellavista
Zehir
Christoph
André Stein