* fix: Source `VERSION` from image ENV
Now CI builds triggered from tagged releases will always have the correct version. No need for manually updating a separate file.
* fix: Query latest GH release tag
Compare to the remote GH release tag published, rather than contents of a `VERSION` file.
`VERSION` file remains in source for now as prior releases still rely on it for an update notification.
* chore: Switch from `yq` to `jaq`
- Can more easily express a string subslice.
- Lighter weight: 9.3M vs 1.7M.
- Drawback, no YAML input/output support.
If `yq` is preferred, the `v` prefix could be removed via BASH easily enough.
* chore: Add entry to `CHANGELOG.md`
* ci: `VERSION` has no relevance to `:edge`
* docs: Update build guide + simplify `make build`
---------
Co-authored-by: Georg Lauterbach <44545919+georglauterbach@users.noreply.github.com>
* remove PostSRSD wrapper
The setup is now completely done during _actual_ setup phase. The
wrapper did not even catch signals (SIGINT, etc.), which I think is
strange.
I also added all the ENVs the wrapper relied on (which previously could
have been unset) to the variables script.
* forgot adjusting the `Dockerfile`
* only add Amavis configuration to Postfix when enabled
Since I am running Rspamd nowadays, I noticed there still are ports open
that belong to Amavis. This is because the Amavis configuration is a
fixed part of Postfix's `master.cf`. I changed that. Now, the Amavis
section is added when Amavis really is enabled.
I took the chance and added proper indentation to `master.cf`; hence the
diff is a bit fuzzy. **But**, only the Amavis part was adjusted, the
rest is just styling.
* chore: Remove the wrapper script for `fail2ban`
- This does not appear necessary. The server can be run with foreground mode.
- `daemons-stack.sh` removal of the socket can be handled by the fail2ban server when using the `-x` option.
* chore: Remove `touch /var/log/auth.log`
These were both added as supposed fixes in 2016 for the then Ubuntu 2014 base image.
Removing them causes no failures in tests.
* fix: Install optional python packages for `fail2ban`
These have barely any overhead in layer weight. The DNS package may provide some QoL improvements, while the `pyinotify` is a better alternative than polling logs to check for updates.
We have `gamin` package installed but `fail2ban` would complain in the log that it was not able to initialize the module for it. There only appears to be a `python-gamin` dependent on EOL python 2, no longer available from Debian Bullseye.
* fix: RSPAM ENV should only add to array if ENV enabled
* fix: Correctly match ownership for Postfix content
- `/var/lib/postfix` dir and content is `postfix:postfix`, not `postfix:root`.
- `/var/spool/postfix` is `root:root` not `postfix:root` like it's content.
- Add additional comments, including ownership changes by Postfix to `/var/spool/postfix` when process starts / restarts.
* fix: Ensure correct `chown -R` user and groups applied
These were all fine except for clamav not using the correct clamav group. `fetchmail` group is `nogroup` as per the group set by the debian package.
Additionally formatted the `-eq 1 ]]` content to align on the same columns, and added additional comment about the purpose of this `chown -R` usage so that it's clear what bug / breakage it's attempting to prevent / fix.
* refactor: `misc-stack.sh` conditional handling
The last condition doesn't get triggered at all AFAIK. Nor does it make sense to make a folder path with `mkdir -p` to symlink to when the container does not have anything to copy over?
- If that was for files, the `mkdir -p` approach seems invalid?
- If it was for a directory that could come up later, it should instead be created in advance? None of the current values for `FILES` seem to hit this path.
Removing as it doesn't seem relevant to current support.
Symlinking was done for each case, I've opted to just perform that after the conditional instead.
Additional inline docs added for additional context.
* chore: Move amavis `chown -R` fix into `misc-stack.sh`
This was handled separately for some reason. It belongs with the other services handling this fix in `misc-stack.sh`.
The `-h` option isn't relevant, when paired with `-R` it has no effect.
* fix: Dockerfile should preserve `clamav` ownership with `COPY --link`
The UID and GID were copied over but would not match `clamav` user and group due to numeric ID mismatch between containers. `--chown=clamav` fixes that.
* chore: Workaround `buildx` bug with separate `chown -R`
Avoids increasing the image weight from this change by leveraging `COPY` in the final stage.
* chore: `COPY --link` from a separate stage instead of relying on scratch
The `scratch` approach wasn't great. A single layer invalidation in the previous stage would result in a new 600MB layer to store.
`make build` with this change seems to barely be affected by such if a change came before copying over the linked stage, although with `buildx` and the `docker-container` driver with `--load` it would take much longer to import and seemed to keep adding storage. Possibly because I was testing with a minimal `buildx` command, that wasn't leveraging proper cache options?
* lint: Appease the linting gods
* chore: Align `misc-stack.sh` paths for `chown -R` operations
Review feedback
Co-authored-by: Casper <casperklein@users.noreply.github.com>
* fix: Reduce one extra cache layer copy
No apparent advantage of a `COPY --link` initially in separate stage.
Just `COPY --chown` in the separate stage and `COPY --link` the stage content. 230MB less in build cache used.
* fix: Remove separate ClamAV stage by adding `clamav` user explicitly
Creating the user before the package is installed allows to ensure a fixed numeric ID that we can provide to `--chown` that is compatible with `--link`.
This keeps the build cache minimal for CI, without being anymore complex as a workaround than the separate stage was for the most part.
* chore: Add reference link regarding users to `misc-stack.sh`
This new script is a clean way of handling the installation of packages.
I think the huge `RUN` command in `Dockerfile` was hard to read and
maintain.
Using a script is a non-issue, as the image is rebuilt whenever the
script is touched.
Co-authored-by: Brennan Kinney <5098581+polarathene@users.noreply.github.com>
The build arguments `VCS_REF` and `VCS_VER` were renamed and given
proper values according to their names.
1. `VCS_REVISION` holds the current SHA sum of the (git) HEAD pointer
2. `VCS_VERSION` now holds the contents of the `VERSION` file, i.e. a
semver version tag (one can now inspect the image and find a proper
version tag in the `org.opencontainers.image.version` label)
The build arguments were given defaults in order to allow the
`generic_build` and `generic_test` workflows to omit them (as they are
not need there anyways). When publishing images, this is fina as the
cache will rebuild almost all of the image except the last few layers
which are `LABEL`s anyways.
* ci: Cache builds by splitting into two jobs
For the cache to work properly, we need to derive a cache key from the build context (files that affect the Dockerfile build) instead of the cache key changing by commit SHA.
We also need to avoid a test suite failure from preventing the caching of a build, thus splitting into separate jobs.
This first attempt used `upload-artifact` and `download-artifact` to transfer the built image, but it has quite a bit of overhead and prevented multi-platform build (without complicating the workflow further).
* ci: Transfer to dependent job via cache only
While `download-artifact` + `docker load` is a little faster than rebuilding the image from cached layers, `upload-artifact` takes about 2 minutes to upload the AMD64 (330MB) tar image export (likely due to compression during upload?).
The `actions/cache` approach however does not incur that hit and is very quick (<10 secs) to complete it's post upload work. The dependent job still gets a cache-hit, and the build job is able to properly support multi-platform builds.
Added additional notes about timing and size of including ARM builds.
* ci: Move Dockerfile ARG to end of build
When the ARG changes due to commit SHA, it invalidates all cache due to the LABEL layers at the start. Then any RUN layers implicitly invalidate, even when the ARG is not used.
Introduced basic multi-stage build, and relocated the container config / metadata to the end of the build. This avoids invalidating expensive caching layers (size and build time) needlessly.
Dovecot master accounts can now be configured in DMS via `setup.sh`.
A master account is useful for administration purposes, or to perform mailbox backups of every user account over IMAP.
Upstream Docs: https://doc.dovecot.org/configuration_manual/authentication/master_users/
Co-authored-by: Casper <casperklein@users.noreply.github.com>
Co-authored-by: Brennan Kinney <5098581+polarathene@users.noreply.github.com>
* first adjustments to use Fail2Ban with nftables
* replace `iptables` -> `nftables` and adjust tests
nftables lists IPs a bit differently , so the order was adjusted for the
tests to be more flexible.
* line correction in mailserver.env
* change from `.conf` -> `.local` and remove redundant config
* revert HEREDOC to `echo`
Co-authored-by: Casper <casperklein@users.noreply.github.com>
Removes duplicate logic from `check-for-changes.sh` that is used/maintained elsewhere to avoid risk of problems, as this code is already starting to diverge / rot.
---
Previously the change detection support has had code added for rebuilding config upon change detection which is the same as code run during startup scripts. Unfortunately over time this has fallen out of sync. Mostly the startup scripts would get maintenance and the contributor and reviewers may not have been aware of the duplicate code handled by `check-for-changes.sh`.
That code was starting to diverge in addition to some changes in structure (_eg: relay host logic seems interleaved here vs separated out in startup scripts_). I wanted to address this before it risks becoming a much bigger headache.
Rather than bloat `helper-functions.sh` further, I've added a `helpers/` folder extracting relevant common logic between startup scripts and `changedetector`. If you want to follow that process I've kept scoped commits to make those diffs easier. Some minor changes/improvements were added but nothing significant.
---
- chore: Extract relay host logic to new `relay.sh` helper
- chore: Extract `/etc/postfix/sasl_passwd` logic to new `sasl.sh` helper
- chore: Extract `postfix-accounts.cf` logic to new `accounts.sh` helper
- chore: Extract `/etc/aliases` logic to new `aliases.sh` helper
- chore: Extract `/etc/postfix/vhost` logic to new `postfix.sh` helper
- chore: Add inline docs for Postfix configs
> These are possibly more verbose than needed and can be reduced at a later stage.
> They are helpful during this refactor process while investigating that everything is handled correctly.
`accounts.sh`:
- Add note regarding potential bug for bare domain setups with `/etc/postfix/vhost` and `mydestination` sharing same domain value.
`relay.sh`:
- Remove the tabs for a single space delimiter, revised associated comment.
- Add PR reference for original `_populate_relayhost_map` implementation which has some useful details.
Co-authored-by: Georg Lauterbach <44545919+georglauterbach@users.noreply.github.com>
Co-authored-by: Casper <casperklein@users.noreply.github.com>
Recent `sedfile` addition moved all scripts section earlier into the Dockerfile so that `sedfile` could be used within the Dockerfile.
However whenever a change is made to scripts which is most of the time for this project, building the Docker image for tests results in all layers after the scripts being invalidated, notably ClamAV, wasting storage of previous instances and increasing build time unnecessarily.
This isn't as noticeable of an issue via the CI as we don't leverage any caching at present there, but for iterating on a local branch and testing, it can be quite the drawback.
- `sedfile` is handled early in the Dockerfile still, while the scripts have been moved as far down as it made sense to.
- `chmod` was split out into it's own RUN command as again it's unnecessary for the rest of it's prior RUN command group to be invalidated.
* chore(refactor): DRY up the `_setup_ssl` method
- `/etc/postfix/ssl` was a bit misleading in usage here. As a maintainer (of my own contribution!) I was confused why only `/etc/postfix/ssl` was referenced and not `/etc/dovecot/ssl`.
- The postfix specific path is unnecessary, dovecot was referencing it via it's config, the same can be done from postfix to a generic DMS specific config location instead.
- This location is defined and created early as `/etc/dms/tls` (with var `DMS_TLS_PATH`). All usage of `/etc/postfix/ssl` has been replaced, making it easier to grok. Several `mkdir` commands related to this have been dropped as a result.
- Likewise, a related `TMP_DMS_TLS_PATH` var provides a reference to the config volume path `/tmp/docker-mailserver` which is used for conditions on presently hard-coded paths.
- Other values that benefit from being DRY have been lifted up into vars. Definitely easier to follow now and makes some further opportunities clearer to tackle in a future refactor.
- `chmod` has been updated where appropriate. Public key/cert is acceptable to have as readable by non-root users (644). The custom type with single fullchain file was not root accessible only, but should as it contains a private key.
- That said, the security benefit can be a bit moot due to source files that were copied remain present, the user would be responsible to ensure similar permissions on their source files.
- I've not touched LetsEncrypt section as I don't have time to investigate into that yet (not familiar with that portion).
---
* chore: Remove mkcert logic and dovecot cert
- No longer serving a purpose.
- Our own TLS startup script handles a variety of cert scenarios, while the dropped code was always generating a self-signed cert and persisting an unused cert regardless with `ONE_DIR=1`.
- To avoid similar issues that DH params had with doveadm validating filepath values in the SSL config, the default dummy values match postfix pointing to "snakeoil" cert. That serves the same purpose as mkcert was covering in the image.
- Bonus, no more hassle with differing mkcert target paths for users replacing our supplied Dovecot with the latest community edition.
---
* Error handling for SSL_TYPE
- Added a panic utility to exit early when SSL_TYPE conditions are misconfigured.
- Some info text had order of key/cert occurrence swapped to be consistent with key then cert.
- Some existing comments moved and rephrased.
- Additional comments added.
- `-f` test for cert files instead of `-e` (true also for directories/devices/symlinks).
- _notify messages lifted out of conditionals so that they always output when the case is hit.
- ~~Empty SSL_TYPE collapsed into catch all panic, while it's contents is now mapped to a new 'disabled' value.~~
---
* Use sedfile + improve sed expressions + update case style
- Uses sedfile when appropriate (file change intentional, not optional match/check).
- sed expressions modified to be DRY and reduce escaping via `-r` flag (acceptable if actual text content contains no `?`,`+`,`()` or `{}` characters, [otherwise they must be escaped](https://www.gnu.org/software/sed/manual/html_node/Extended-regexps.html)).
- sed captures anything matched between the parenthesis`()` and inserts it via `\1` as part of the replacement.
- case statements adopt the `(` prefix, adopting recent shell style for consistency.
---
* Refactor SSL_TYPE=disabled
- Postfix is also disabled now.
- Included heavy inline documentation reference for maintainers.
- Dropped an obsolete postfix config option 'use_tls' on the relayhost function, it was replaced by 'security_level'.
---
* I'm a friggin' sed wizard now
- The `modern` TLS_LEVEL is the default values for the configs they modify. As such, `sedfile` outputs an "Error" which isn't an actual concern, back to regular `sed`.
- I realized that multiple edits for the same file can all be done at once via `-e` (assuming other sed options are the same for each operation), and that `g` suffix is global scope for single line match, not whole file (default as sed iterates through individual lines).
- Some postfix replacements have `smtp` and `smtpd` lines, collapsed into a single `smtpd?` instead now that I know sed better.
---
* tests(fix): Tests that require SSL/TLS to pass
- SSL_TYPE=snakeoil added as temporary workaround.
- nmap tests are being dropped. These were added about 4-5 years ago, I have since made these redundant with the `testssl.sh` tests.
- Additionally the `--link` option is deprecated and IIRC these grades were a bit misleading when I initially used nmap in my own TLS cipher suite update PRs in the past.
- The removed SSL test is already handled in mail_ssl_manual.bats
ldap test:
- Replace `--link` alias option with `--network` and alias assignment.
- Parameterized some values and added the `SSL_TYPE` to resolve the starttls test failure.
privacy test:
- Also needed `SSL_TYPE` to pass the starttls test.
`tests.bats` had another starttls test for imap:
- Workaround for now is to give the main test container `SSL_TYPE=snakeoil`.
---
* Remove the expired lets-encrypt cert
This expired in March 2021. It was originally required when first added back in 2016 as LetsEncrypt was fairly new and not as broadly accepted into OS trust stores.
No longer the case today.
---
* chore: Housekeeping
Not required for this PR branch, little bit of tidying up while working on these two test files.
- privacy test copied over content when extracted from `tests.bats` that isn't relevant.
- ldap test was not as easy to identify the source of DOVECOT_TLS. Added comment to make the prefix connection to `configomat.sh` and `.ext` files more easier to find.
- Additionally converted the two localhost FQDN to vars.
---
* Default SSL_TYPE becomes `''` (aka equivalent to desired `disabled` case)
- This is to prevent other tests from failing by hitting the panic catchall case.
- More ideal would be adjusting tests to default to `disabled`, rather than treating `disabled` as an empty / unset SSL_TYPE value.
---
* Add inline documentation for `dms_panic`
- This could later be better formatted and placed into contributor docs.
Panic with kill (shutdown) not exit (errex):
- `kill 1` from `_shutdown` will send SIGTERM signal to PID 1 (init process).
- `exit 1` within the `start-mailserver.sh` init scripts context, will just exit the initialization script leaving the container running when it shouldn't.
The two previous `_shutdown` methods can benefit from using `dms_panic` wrapper instead to standardize on panic messages.
This feature was originally introduced by the PR: https://github.com/docker-mailserver/docker-mailserver/pull/1463
- Assign default DH params to use via Dockerfile build instead of copy and update at runtime.
- Parameterized service names and paths.
- Refactor postfix and dovecot dh methods to wrap shared dh logic
- I don't see any value in checking the alternative service for dh params file to copy over, so that's now dropped too.
- Another conditional check is dropped and the default fallback message for existing DH params file is no longer relevant.
- Improved the remaining `_notify` messages. Collapsing the warning into a single logged message also seemed relevant.
- There is no apparent need for special handling with `ONE_DIR=1`. Dropped it.
- Refactor DH params tests
- Combine custom and default DH param tests into single test file
- docs: Add instructions to use custom DH params
There is no official documented support for custom DH parameters. As no guarantee is provided, this is considered an internal change, not a breaking one.
* changed the locking function to better support multiple servers running at once and sharing the same config
* helper function testing now runs inside of container
Co-authored-by: Brennan Kinney <5098581+polarathene@users.noreply.github.com>
