Bundle the spam filter rules with the server

The server fetched upstream's latest published rules from GitHub at run
time: a version nobody here tested, code-like expressions from an account
we don't control, and the upstream name as a default in the admin form.

The published rules of spam-filter v3.0.2 are now embedded
(resources/spam-filter/, MIT, in THIRD-PARTY.md) and used whenever no other
source is configured. An empty setting and upstream's old default both mean
the bundled rules, so existing installs switch without a settings change;
the URL stays an operator override (https:// or file://). The schema default
is dropped and its description says what empty means, and the strip's
rename pass does the same to each import.

Rules load on first boot as before, and again whenever the bundled version
differs from the last one loaded, which only adds missing rules and tags.
That brings the AI classifier's LLM_* scores to installs that predate them:
production has none today.

upstream-watch now also opens an issue when spam-filter publishes a newer
release; resources/spam-filter/README.md says how to take it.

The antispam test now runs on the bundled rules, the path production
takes; SPAM_RULES_URL tests another set. Unit tests cover the URL handling
and that the bundled rules parse and score the AI tags as the AI spec says.
This commit is contained in:
2026-09-22 22:01:30 -07:00
parent 0d8caaa514
commit 17426f6d60
16 changed files with 234 additions and 42 deletions
@@ -243,7 +243,8 @@ impl SpamFilterConfig {
spam_threshold: spam.score_spam.into_inner() as f32,
},
grey_list_expiry: spam.greylist_for.map(|d| d.into_inner().as_secs()),
spam_rules_url: spam.spam_filter_rules_url,
// inbuxa: unset, empty or upstream's old default means the bundled rules
spam_rules_url: crate::manager::spam_rules::rules_url(spam.spam_filter_rules_url),
url_client: utils::http::http_client_builder(true)
.pool_max_idle_per_host(0)
.redirect(reqwest::redirect::Policy::none())
+14 -5
View File
@@ -530,13 +530,22 @@ async fn insert_safe_defaults(bp: &mut Bootstrap) -> trc::Result<()> {
use store::write::BatchBuilder;
use types::id::Id;
if bp.registry.count_object(ObjectType::SpamRule).await? == 0
&& bp
.registry
// inbuxa: rules are always to hand, since a copy ships with the server
// (spam_rules). They load on first boot, and again when the bundled
// version differs from the one last loaded, which only adds what's
// missing: new tags and rules, never a changed score.
let rules_url = super::spam_rules::rules_url(
bp.registry
.object::<SpamSettings>(Id::singleton())
.await?
.is_none_or(|spam| spam.spam_filter_rules_url.is_some())
{
.and_then(|spam| spam.spam_filter_rules_url),
);
let bundled_is_new = rules_url.is_none()
&& super::spam_rules::applied_version(&bp.data_store)
.await?
.as_deref()
!= Some(super::spam_rules::BUNDLED_SPAM_RULES_VERSION);
if bp.registry.count_object(ObjectType::SpamRule).await? == 0 || bundled_is_new {
let mut batch = BatchBuilder::new();
batch.schedule_task(Task::SpamFilterMaintenance(TaskSpamFilterMaintenance {
maintenance_type: TaskSpamFilterMaintenanceType::UpdateRules,
+1
View File
@@ -22,6 +22,7 @@ pub mod console;
pub mod defaults;
pub mod first_party;
pub mod restore;
pub mod spam_rules; // inbuxa: rules bundled with the server
pub const SPAM_TRAINER_KEY: &[u8] = "INBUXA_SPAM_TRAIN_DATA.lz4".as_bytes();
pub const SPAM_CLASSIFIER_KEY: &[u8] = "INBUXA_SPAM_CLASSIFIER_MODEL.lz4".as_bytes();
+103
View File
@@ -0,0 +1,103 @@
/*
* SPDX-FileCopyrightText: 2026 Coffey Labs
*
* SPDX-License-Identifier: AGPL-3.0-only
*/
//! inbuxa: the spam filter rules that ship with the server.
//!
//! Upstream fetches its latest published rules from GitHub at run time, so
//! scoring changes with a release nobody here tested and depends on reaching
//! it. The fork embeds a pinned copy (resources/spam-filter/, with its version
//! and license) and uses it whenever no other source is configured. The rules
//! URL remains an operator override (`https://` or `file://`).
//!
//! Loading rules only ever adds what's missing, never changes an existing rule
//! or score. They load on first boot, and again whenever the bundled version
//! differs from the one last applied, so an upgrade brings new tags (the AI
//! classifier's `LLM_*` scores, say) to an install that already had rules.
use std::io::Read;
use store::{
SUBSPACE_INBUXA, Store, ValueKey,
write::{AnyClass, BatchBuilder, ValueClass},
};
use trc::AddContext;
/// The version of spam-filter the embedded rules come from.
pub const BUNDLED_SPAM_RULES_VERSION: &str = "3.0.2";
static BUNDLED_SPAM_RULES: &[u8] =
include_bytes!("../../../../resources/spam-filter/spam-filter-rules.json.gz");
/// Upstream's default rules source, the value every install created before
/// the rules were bundled has saved. Read only to treat it as unset.
const LEGACY_DEFAULT_URL: &str =
"https://github.com/stalwartlabs/spam-filter/releases/latest/download/spam-filter-rules.json.gz";
/// The URL to fetch rules from, or `None` for the bundled rules. An empty
/// setting and upstream's old default both mean the bundled rules.
pub fn rules_url(configured: Option<String>) -> Option<String> {
configured.filter(|url| !url.trim().is_empty() && url != LEGACY_DEFAULT_URL)
}
/// The bundled rules, uncompressed: the same JSON the rules URL serves.
pub fn bundled_rules() -> Result<Vec<u8>, String> {
let mut json = Vec::new();
mail_auth::flate2::read::GzDecoder::new(BUNDLED_SPAM_RULES)
.read_to_end(&mut json)
.map_err(|err| format!("Failed to decompress the bundled spam rules: {err}"))?;
Ok(json)
}
fn applied_key() -> ValueClass {
ValueClass::Any(AnyClass {
subspace: SUBSPACE_INBUXA,
key: b"Sr".to_vec(),
})
}
/// The bundled version last loaded into the registry, if any.
pub async fn applied_version(data: &Store) -> trc::Result<Option<String>> {
data.get_value::<String>(ValueKey::from(applied_key()))
.await
.caused_by(trc::location!())
}
/// Records that the bundled rules of this version have been loaded.
pub async fn set_applied_version(data: &Store, version: &str) -> trc::Result<()> {
let mut batch = BatchBuilder::new();
batch.set(applied_key(), version.as_bytes().to_vec());
data.write(batch.build_all())
.await
.caused_by(trc::location!())
.map(|_| ())
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn upstream_default_and_empty_mean_bundled() {
assert_eq!(rules_url(None), None);
assert_eq!(rules_url(Some(String::new())), None);
assert_eq!(rules_url(Some(" ".into())), None);
assert_eq!(rules_url(Some(LEGACY_DEFAULT_URL.into())), None);
assert_eq!(
rules_url(Some("file:///srv/rules.json.gz".into())).as_deref(),
Some("file:///srv/rules.json.gz")
);
}
#[test]
fn bundled_rules_parse_and_score_the_ai_tags() {
let rules: serde_json::Value = serde_json::from_slice(&bundled_rules().unwrap()).unwrap();
let tags = rules["SpamTag"].as_array().unwrap();
for (tag, score) in [("LLM_UNSOLICITED_HIGH", 3.0), ("LLM_LEGITIMATE_HIGH", -3.0)] {
let found = tags.iter().find(|t| t["tag"] == tag).unwrap();
assert_eq!(found["score"].as_f64(), Some(score), "{tag}");
}
assert!(!rules["SpamRule"].as_array().unwrap().is_empty());
}
}
+1 -1
View File
@@ -40215,7 +40215,7 @@ impl Default for SpamSettings {
score_reject: Float::new(0.0f64),
score_spam: Float::new(5.0f64),
trust_replies: true,
spam_filter_rules_url: Some("https://github.com/stalwartlabs/spam-filter/releases/latest/download/spam-filter-rules.json.gz".to_string()),
spam_filter_rules_url: None,
}
}
}
@@ -2,13 +2,15 @@
* SPDX-FileCopyrightText: 2020 Stalwart Labs LLC <[email protected]>
*
* SPDX-License-Identifier: AGPL-3.0-only OR LicenseRef-SEL
*
* Modified by Coffey Labs in 2026 for INBUXA.
*/
use crate::task_manager::{TaskFailureType, TaskResult};
use common::{
Server,
ipc::{BroadcastEvent, RegistryChange},
manager::{SPAM_CLASSIFIER_KEY, SPAM_TRAINER_KEY, fetch_resource},
manager::{SPAM_CLASSIFIER_KEY, SPAM_TRAINER_KEY, fetch_resource, spam_rules},
};
use registry::{
schema::{
@@ -106,6 +108,7 @@ struct RuleUpdateResult {
async fn update_spam_rules(server: &Server) -> trc::Result<TaskResult> {
let started = Instant::now();
let bundled = server.core.spam.spam_rules_url.is_none();
let rules = match fetch_spam_rules(server).await {
Ok(rules) => rules,
Err(err) => {
@@ -289,29 +292,36 @@ async fn update_spam_rules(server: &Server) -> trc::Result<TaskResult> {
Elapsed = started.elapsed(),
);
// inbuxa: so the next start knows these bundled rules are in
if bundled {
spam_rules::set_applied_version(server.store(), spam_rules::BUNDLED_SPAM_RULES_VERSION)
.await?;
}
Ok(TaskResult::Success(vec![]))
}
async fn fetch_spam_rules(server: &Server) -> Result<Rules, RuleUpdateError> {
let Some(rules_url) = server.core.spam.spam_rules_url.as_ref() else {
return Err(RuleUpdateError {
typ: TaskFailureType::Permanent,
reason: "Spam rules resource URL not configured".to_string(),
});
};
let rules_json: AHashMap<String, Vec<serde_json::Value>> =
fetch_resource(rules_url, None, Duration::from_secs(60), 1024 * 500)
// inbuxa: no URL means the rules bundled with the server
let bytes = match server.core.spam.spam_rules_url.as_ref() {
Some(rules_url) => fetch_resource(rules_url, None, Duration::from_secs(60), 1024 * 500)
.await
.map_err(|reason| RuleUpdateError {
typ: TaskFailureType::Temporary,
reason,
}),
None => spam_rules::bundled_rules().map_err(|reason| RuleUpdateError {
typ: TaskFailureType::Permanent,
reason,
}),
};
let rules_json: AHashMap<String, Vec<serde_json::Value>> =
bytes.and_then(|bytes| {
serde_json::from_slice(&bytes).map_err(|err| RuleUpdateError {
typ: TaskFailureType::Permanent,
reason: format!("Failed to parse spam rules JSON: {err}"),
})
.and_then(|bytes| {
serde_json::from_slice(&bytes).map_err(|err| RuleUpdateError {
typ: TaskFailureType::Permanent,
reason: format!("Failed to parse spam rules JSON: {err}"),
})
})?;
})?;
let mut rules = Rules::default();
for (object_type, values) in rules_json {