Do not accept pattern_type from user input in push rules

changelog.d/15088.bugfix

@@ -0,0 +1 @@
+Fix a long-standing bug where Synapse handled an unspecced field on push rules.

rust/benches/evaluator.rs

@@ -60,8 +60,7 @@ fn bench_match_exact(b: &mut Bencher) {
     let condition = Condition::Known(synapse::push::KnownCondition::EventMatch(
         EventMatchCondition {
             key: "room_id".into(),
-            pattern: Some("!room:server".into()),
-            pattern_type: None,
+            pattern: "!room:server".into(),
         },
     ));
 
@@ -109,8 +108,7 @@ fn bench_match_word(b: &mut Bencher) {
     let condition = Condition::Known(synapse::push::KnownCondition::EventMatch(
         EventMatchCondition {
             key: "content.body".into(),
-            pattern: Some("test".into()),
-            pattern_type: None,
+            pattern: "test".into(),
         },
     ));
 
@@ -158,8 +156,7 @@ fn bench_match_word_miss(b: &mut Bencher) {
     let condition = Condition::Known(synapse::push::KnownCondition::EventMatch(
         EventMatchCondition {
             key: "content.body".into(),
-            pattern: Some("foobar".into()),
-            pattern_type: None,
+            pattern: "foobar".into(),
         },
     ));
 

rust/src/push/evaluator.rs

@@ -12,9 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use std::borrow::Cow;
 use std::collections::{BTreeMap, BTreeSet};
 
-use crate::push::JsonValue;
+use crate::push::{EventMatchPatternType, JsonValue};
 use anyhow::{Context, Error};
 use lazy_static::lazy_static;
 use log::warn;
@@ -23,8 +24,8 @@ use regex::Regex;
 
 use super::{
     utils::{get_glob_matcher, get_localpart_from_id, GlobMatchType},
-    Action, Condition, EventMatchCondition, ExactEventMatchCondition, FilteredPushRules,
-    KnownCondition, RelatedEventMatchCondition, SimpleJsonValue,
+    Action, Condition, ExactEventMatchCondition, FilteredPushRules, KnownCondition,
+    SimpleJsonValue,
 };
 
 lazy_static! {
@@ -256,14 +257,58 @@ impl PushRuleEvaluator {
         };
 
         let result = match known_condition {
-            KnownCondition::EventMatch(event_match) => {
-                self.match_event_match(event_match, user_id)?
+            KnownCondition::EventMatch(event_match) => self.match_event_match(
+                &self.flattened_keys,
+                &event_match.key,
+                &event_match.pattern,
+            )?,
+            KnownCondition::EventMatchType(event_match) => {
+                // The `pattern_type` can either be "user_id" or "user_localpart",
+                // either way if we don't have a `user_id` then the condition can't
+                // match.
+                let user_id = if let Some(user_id) = user_id {
+                    user_id
+                } else {
+                    return Ok(false);
+                };
+
+                let pattern = match &*event_match.pattern_type {
+                    EventMatchPatternType::UserId => user_id,
+                    EventMatchPatternType::UserLocalpart => get_localpart_from_id(user_id)?,
+                };
+
+                self.match_event_match(&self.flattened_keys, &event_match.key, pattern)?
             }
             KnownCondition::ExactEventMatch(exact_event_match) => {
                 self.match_exact_event_match(exact_event_match)?
             }
-            KnownCondition::RelatedEventMatch(event_match) => {
-                self.match_related_event_match(event_match, user_id)?
+            KnownCondition::RelatedEventMatch(event_match) => self.match_related_event_match(
+                &event_match.rel_type.clone(),
+                event_match.include_fallbacks,
+                event_match.key.clone(),
+                event_match.pattern.clone(),
+            )?,
+            KnownCondition::RelatedEventMatchType(event_match) => {
+                // The `pattern_type` can either be "user_id" or "user_localpart",
+                // either way if we don't have a `user_id` then the condition can't
+                // match.
+                let user_id = if let Some(user_id) = user_id {
+                    user_id
+                } else {
+                    return Ok(false);
+                };
+
+                let pattern = match &*event_match.pattern_type {
+                    EventMatchPatternType::UserId => user_id,
+                    EventMatchPatternType::UserLocalpart => get_localpart_from_id(user_id)?,
+                };
+
+                self.match_related_event_match(
+                    &event_match.rel_type.clone(),
+                    event_match.include_fallbacks,
+                    Some(event_match.key.clone()),
+                    Some(Cow::Borrowed(pattern)),
+                )?
             }
             KnownCondition::ExactEventPropertyContains(exact_event_match) => {
                 self.match_exact_event_property_contains(exact_event_match)?
@@ -325,32 +370,12 @@ impl PushRuleEvaluator {
     /// Evaluates a `event_match` condition.
     fn match_event_match(
         &self,
-        event_match: &EventMatchCondition,
-        user_id: Option<&str>,
+        flattened_event: &BTreeMap<String, JsonValue>,
+        key: &str,
+        pattern: &str,
     ) -> Result<bool, Error> {
-        let pattern = if let Some(pattern) = &event_match.pattern {
-            pattern
-        } else if let Some(pattern_type) = &event_match.pattern_type {
-            // The `pattern_type` can either be "user_id" or "user_localpart",
-            // either way if we don't have a `user_id` then the condition can't
-            // match.
-            let user_id = if let Some(user_id) = user_id {
-                user_id
-            } else {
-                return Ok(false);
-            };
-
-            match &**pattern_type {
-                "user_id" => user_id,
-                "user_localpart" => get_localpart_from_id(user_id)?,
-                _ => return Ok(false),
-            }
-        } else {
-            return Ok(false);
-        };
-
         let haystack = if let Some(JsonValue::Value(SimpleJsonValue::Str(haystack))) =
-            self.flattened_keys.get(&*event_match.key)
+            flattened_event.get(key)
         {
             haystack
         } else {
@@ -359,7 +384,7 @@ impl PushRuleEvaluator {
 
         // For the content.body we match against "words", but for everything
         // else we match against the entire value.
-        let match_type = if event_match.key == "content.body" {
+        let match_type = if key == "content.body" {
             GlobMatchType::Word
         } else {
             GlobMatchType::Whole
@@ -395,75 +420,37 @@ impl PushRuleEvaluator {
     /// Evaluates a `related_event_match` condition. (MSC3664)
     fn match_related_event_match(
         &self,
-        event_match: &RelatedEventMatchCondition,
-        user_id: Option<&str>,
+        rel_type: &str,
+        include_fallbacks: Option<bool>,
+        key: Option<Cow<str>>,
+        pattern: Option<Cow<str>>,
     ) -> Result<bool, Error> {
         // First check if related event matching is enabled...
         if !self.related_event_match_enabled {
             return Ok(false);
         }
 
         // get the related event, fail if there is none.
-        let event = if let Some(event) = self.related_events_flattened.get(&*event_match.rel_type) {
+        let event = if let Some(event) = self.related_events_flattened.get(rel_type) {
             event
         } else {
             return Ok(false);
         };
 
         // If we are not matching fallbacks, don't match if our special key indicating this is a
         // fallback relation is not present.
-        if !event_match.include_fallbacks.unwrap_or(false)
-            && event.contains_key("im.vector.is_falling_back")
-        {
+        if !include_fallbacks.unwrap_or(false) && event.contains_key("im.vector.is_falling_back") {
             return Ok(false);
         }
 
-        // if we have no key, accept the event as matching, if it existed without matching any
-        // fields.
-        let key = if let Some(key) = &event_match.key {
-            key
-        } else {
-            return Ok(true);
-        };
-
-        let pattern = if let Some(pattern) = &event_match.pattern {
-            pattern
-        } else if let Some(pattern_type) = &event_match.pattern_type {
-            // The `pattern_type` can either be "user_id" or "user_localpart",
-            // either way if we don't have a `user_id` then the condition can't
-            // match.
-            let user_id = if let Some(user_id) = user_id {
-                user_id
-            } else {
-                return Ok(false);
-            };
-
-            match &**pattern_type {
-                "user_id" => user_id,
-                "user_localpart" => get_localpart_from_id(user_id)?,
-                _ => return Ok(false),
-            }
-        } else {
-            return Ok(false);
-        };
-
-        let haystack =
-            if let Some(JsonValue::Value(SimpleJsonValue::Str(haystack))) = event.get(&**key) {
-                haystack
-            } else {
-                return Ok(false);
-            };
-
-        // For the content.body we match against "words", but for everything
-        // else we match against the entire value.
-        let match_type = if key == "content.body" {
-            GlobMatchType::Word
-        } else {
-            GlobMatchType::Whole
-        };
-
-        let mut compiled_pattern = get_glob_matcher(pattern, match_type)?;
-        compiled_pattern.is_match(haystack)
+        match (key, pattern) {
+            // if we have no key, accept the event as matching.
+            (None, _) => Ok(true),
+            // There was a key, so we *must* have a pattern to go with it.
+            (Some(_), None) => Ok(false),
+            // If there is a key & pattern, check if they're in the flattened event (given by rel_type).
+            (Some(key), Some(pattern)) => self.match_event_match(event, &key, &pattern),
+        }
     }
 
     /// Evaluates a `exact_event_property_contains` condition. (MSC3758)

rust/src/push/mod.rs

@@ -328,10 +328,16 @@ pub enum Condition {
 #[serde(tag = "kind")]
 pub enum KnownCondition {
     EventMatch(EventMatchCondition),
+    // Identical to event_match but gives predefined patterns. Cannot be added by users.
+    #[serde(skip_deserializing, rename = "event_match")]
+    EventMatchType(EventMatchTypeCondition),
     #[serde(rename = "com.beeper.msc3758.exact_event_match")]
     ExactEventMatch(ExactEventMatchCondition),
     #[serde(rename = "im.nheko.msc3664.related_event_match")]
     RelatedEventMatch(RelatedEventMatchCondition),
+    // Identical to related_event_match but gives predefined patterns. Cannot be added by users.
+    #[serde(skip_deserializing, rename = "im.nheko.msc3664.related_event_match")]
+    RelatedEventMatchType(RelatedEventMatchTypeCondition),
     #[serde(rename = "org.matrix.msc3966.exact_event_property_contains")]
     ExactEventPropertyContains(ExactEventMatchCondition),
     #[serde(rename = "org.matrix.msc3952.is_user_mention")]
@@ -362,14 +368,27 @@ impl<'source> FromPyObject<'source> for Condition {
     }
 }
 
-/// The body of a [`Condition::EventMatch`]
+/// The body of a [`Condition::EventMatch`] with a pattern.
 #[derive(Serialize, Deserialize, Debug, Clone)]
 pub struct EventMatchCondition {
     pub key: Cow<'static, str>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    pub pattern: Option<Cow<'static, str>>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    pub pattern_type: Option<Cow<'static, str>>,
+    pub pattern: Cow<'static, str>,
+}
+
+#[derive(Serialize, Debug, Clone)]
+#[serde(rename_all = "snake_case")]
+pub enum EventMatchPatternType {
+    UserId,
+    UserLocalpart,
+}
+
+/// The body of a [`Condition::EventMatch`] that uses user_id or user_localpart as a pattern.
+#[derive(Serialize, Debug, Clone)]
+pub struct EventMatchTypeCondition {
+    pub key: Cow<'static, str>,
+    // During serialization, the pattern_type property gets replaced with a
+    // pattern property of the correct value in synapse.push.clientformat.format_push_rules_for_user.
+    pub pattern_type: Cow<'static, EventMatchPatternType>,
 }
 
 /// The body of a [`Condition::ExactEventMatch`]
@@ -386,8 +405,18 @@ pub struct RelatedEventMatchCondition {
     pub key: Option<Cow<'static, str>>,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub pattern: Option<Cow<'static, str>>,
+    pub rel_type: Cow<'static, str>,
     #[serde(skip_serializing_if = "Option::is_none")]
-    pub pattern_type: Option<Cow<'static, str>>,
+    pub include_fallbacks: Option<bool>,
+}
+
+/// The body of a [`Condition::RelatedEventMatch`] that uses user_id or user_localpart as a pattern.
+#[derive(Serialize, Debug, Clone)]
+pub struct RelatedEventMatchTypeCondition {
+    // This is only used if pattern_type exists (and thus key must exist), so is
+    // a bit simpler than RelatedEventMatchCondition.
+    pub key: Cow<'static, str>,
+    pub pattern_type: Cow<'static, EventMatchPatternType>,
     pub rel_type: Cow<'static, str>,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub include_fallbacks: Option<bool>,
@@ -571,8 +600,7 @@ impl FilteredPushRules {
 fn test_serialize_condition() {
     let condition = Condition::Known(KnownCondition::EventMatch(EventMatchCondition {
         key: "content.body".into(),
-        pattern: Some("coffee".into()),
-        pattern_type: None,
+        pattern: "coffee".into(),
     }));
 
     let json = serde_json::to_string(&condition).unwrap();

synapse/push/clientformat.py

@@ -42,6 +42,7 @@ def format_push_rules_for_user(
         rulearray.append(template_rule)
 
         pattern_type = template_rule.pop("pattern_type", None)
+        print(pattern_type)
         if pattern_type == "user_id":
             template_rule["pattern"] = user.to_string()
         elif pattern_type == "user_localpart":

tests/push/test_push_rule_evaluator.py

@@ -401,6 +401,33 @@ def test_event_match_non_body(self) -> None:
             "pattern should not match before a newline",
         )
 
+    def test_event_match_pattern(self) -> None:
+        """Check that event_match conditions do not use a "pattern_type" from user data."""
+
+        # The pattern_type should not be deserialized into anything valid.
+        condition = {
+            "kind": "event_match",
+            "key": "content.value",
+            "pattern_type": "user_id",
+        }
+        self._assert_not_matches(
+            condition,
+            {"value": "@user:test"},
+            "should not be possible to pass a pattern_type in",
+        )
+
+        # This is an internal-only condition which shouldn't get deserialized.
+        condition = {
+            "kind": "event_match_type",
+            "key": "content.value",
+            "pattern_type": "user_id",
+        }
+        self._assert_not_matches(
+            condition,
+            {"value": "@user:test"},
+            "should not be possible to pass a pattern_type in",
+        )
+
     def test_exact_event_match_string(self) -> None:
         """Check that exact_event_match conditions work as expected for strings."""