This support is based on github.com/foxcpp/go-msgauth fork until
emerison/go-msgauth#13 gets merged.
Further extensions are required to make sure only messages we can
actually "take responsibility for" are signed.
RSA-2048 is used as a default algorithm when generating new keys.
RSA-4096 can cause trouble with UDP-only DNS due to responses being
bigger than 512 octets. RSA-1024 is too weak and explicitly
disallowed in maddy for new keys. It could be possible to use Ed25519
but support is not widely deployed yet (according to warning in rspamd
docs dated 2019-09). Users concerned about security of RSA-2048 can
switch to RSA-4096 or Ed25519, keeping relevant problems in mind.
Ed25519 key format uses PKCS#8, this seems to be different from other
implementations that just dump key material into a file without any
wrapping. Interoperability is not considered to encourage key
rotation when migration, which is a good thing to do anyway.
There is no option to use "body limit", since it is dangerous
and go-msgauth/dkim does not support it for signing.
The default set of signed header fields is the list used by rspamd.
Most "core" fields are oversigned to provide strict integrity.
"Conditional oversigning" similar to rspamd is not implemented, though
it may be useful, further research is required.
Multi-tentant configuration with DKIM and DMARC is much more verbose,
configuration example is added to config.d/multitentant-dkim.conf to
explain how to make it work.
To support unusual configuration syntax, endpoint modules (imap, smtp,
etc) relied on rather awkward code using modName+instName+aliases as
arguments. This commit replaces old handling with use of special
signature similar to inlineArgs introduced in 1edd031.
Endpoint modules are placed in a separate 'registry' and use
different initialization callback signature for simplicity. This makes
them inaccessible for other modules, though they are not supposed to be
anyway.
Endpoint modules are initialized before other modules. This allows
detecting unused configuration blocks by checking for modules
that were not lazily initalized after endpoint initialization.
This relies on endpoint modules being essentially "roots" of
instances dependency tree.
Idea of "semantical module names" is completely dropped now and so
HACKING.md is updated to not mention it.
Now they are represented by implementations of the log.Output interface
instead of just callbacks. This allows to add additional methods to it.
Mainly, it was done for purpose of Close method for flushing and closing
log files. This is important to make runtime logger reinitialization
(e.g. log rotation) possible.
Additionally, documentation now explains goroutine-safety of logger
objects.
It makes sense to allow having default of bool flag to true and
allow writing 'name no' to disable it. This allows getting rid of
"dont", "no" etc in names or certain directives which are meant to be
enabled by default. IMO 'do_x no' looks better than 'dont_do_x'
Allows faster and simpler implementation for cases when maddy
is running as a privileged user.
Automatic fallback to helper binary is not implemented since otherwise
config directive 'use_helper no' would make less sense and config.Map
does not support "enums" (true/false/not-specified) yet.
Closes#127.
Root package now contains only initialization code and 'dummy' module.
Each module now got its own package. Module packages are grouped by
their main purpose (storage/, target/, auth/, etc). Shared code is
placed in these "group" packages.
Parser for module references in config is moved into config/module.
Code shared by tests (mock modules, etc) is placed in testutils.
This allows more readable configuration files without additional
explanations in cases where a single module is used for multiple
purposes.
Also cleans up certain problems with modules that rely on block
names having certain semantics (e.g. endpoint modules).
* Drop most of the implicit defaults in favor of explicit configuration
We no longer follow caddy's "zero-configuration" approach. Mail is much
more complex than HTTP and we want to be explicit about things, always.
* Remove commented out directives from maddy.conf
This allows modules to reference each other during initialization
independently of configuration blocks order.
Also this allows us to initialize default modules in a more clean way.
All configuration directives now use underscores instead of dashes for
consistency with Caddy conventions.
Disallow defining multiple module instances with implicit name.
Remove global module.WaitGroup and add as field where it is necessary
(endpoint modules).
buf.Reset() and use rewind Reader in SMTP pipeline.
Rename several entities in code (NewModule => FuncNewModule, CfgTreeNode
=> Node, etc).
Also fix several warnings from linters.
Currently default pipeline contains only two delivery steps, but in
future it will include more complex dispatching logic for spam filters,
DKIM and features.
