* [Conf] Add `one_shot` to some specific multimap rules
* [Conf] Add language detection configuration
* [Conf] Add missing attributes for the language detection configuration
* [Conf] Remove outdated composite rules
* [Feature] Add `sentinel_password` option
* [Feature] Add ability to deny specific fuzzy flags by default
* [Feature] Add controller endpoint to get fuzzy hashes from messages
* [Feature] Add extra symbol when URL redirector reaches nested limit
* [Feature] Add function to transliterate utf8 to ascii with some normalisation
* [Feature] Add html parsing limit
* [Feature] Add order to urls structure
* [Feature] Add some missing functions to `lua_rsa` library
* [Feature] Allow fuzzy workers to exchange blocked information
* [Feature] Allow to have weak flags in fuzzy storage
* [Feature] Allow to read options from maps in the multimap plugin
* [Feature] Allow to use other methods when fasttext detection is enabled
* [Feature] Count stats per key per flag
* [Feature] Finish all features of dkim_keygen in Lua
* [Feature] Khash: Allow static initialisation
* [Feature] Maps: Add on_load support
* [Feature] Preliminary implementation of dynamic composites
* [Feature] Process HTML parts before text ones
* [Feature] Reorganise struct rspamd_url to be 64 bytes size
* [Feature] Save fuzzy ratelimit buckets
* [Feature] Use in-place deflation for strings
* [Feature] external_relay: add ip_map strategy
* [Fix] Avoid race between config new/free by using a counter
* [Fix] Do not use `rspamadm.dkim_keygen`
* [Fix] Feed fasttext language model with the pre-tokenized words
* [Fix] Fix `rspamd_has_only_html_part`
* [Fix] Fix an old issue with order of destruction race between redis pool and lua
* [Fix] Fix format string usage
* [Fix] Fix parsing due to old bug revealed
* [Fix] Fix parsing of the mask values that are invalid
* [Fix] Ignore non-unique stop words
* [Fix] Include the last character when parsing the last header with no value
* [Fix] More fixes to fuzzystat
* [Fix] Set proper counter
* [Fix] Try harder to clean pending bucket
* [Fix] Try harder to remove bad hyperscan files
* [Fix] Update stats before encryption...
* [Fix] dmarc gramar - allow spaces before ";"
* [Fix] rbl: fix dependency registration when symbols_prefixes is used
* [Fix] remove obsolete rspamd-redirector files
* [Fix] test external_relay: count should always be the last rule, as it have no matching condition
* [Project] Allow to register multimap symbols dynamically
* [Project] Implement fasttext language detection
* [Rework] Default max shots must not influence options
* [Rework] Write dkim keygen tool in lua
* [Rules] Add thread hijacking composite rule
[Minor] Align scores of Spamhaus DBL, SURBL, URIBL DNSBL symbols
Given that they have about the same false positive rate, it makes sense
to treat them equal in terms of scoring:
- Particular threats (phishing, malware) are scored a bit higher than
mere spam domain listings
- "Abused legitimate" listings are scored lower for some DNSBLs already,
this has now been aligned.
- For SURBL, cracked and abused sites are treated with the same score.
Given that CSS is an automated component of SBL, this should not receive
the same scoring as manually conducted SBL listings. Particularly for
shared hosting environments, CSS hits on IP addresses derived from FQDNs
sometimes were found to be scored a bit too high.