Security: disable debug mode, sanitize query input, validate Qdrant filters, add size/offset bounds

search_app.py

@@ -17,6 +17,7 @@ from __future__ import annotations
 import copy
 import json
 import logging
+import os
 import re
 from pathlib import Path
 from typing import Any, Dict, Iterable, List, Optional, Sequence, Set, Tuple
@@ -46,6 +47,46 @@ LOGGER = logging.getLogger(__name__)
 _EMBED_MODEL = None
 _EMBED_MODEL_NAME: Optional[str] = None
 
+# Security constants
+MAX_QUERY_SIZE = 100
+MAX_OFFSET = 10000
+ALLOWED_QDRANT_FILTER_FIELDS = {"channel_id", "date", "video_status", "external_reference"}
+
+
+def sanitize_query_string(query: str) -> str:
+    """
+    Sanitize user input for Elasticsearch query_string queries.
+    Removes dangerous field targeting and script injection patterns.
+    """
+    if not query:
+        return "*"
+    sanitized = query.strip()
+    # Remove field targeting patterns like "_id:", "_source:", "script:"
+    dangerous_field_patterns = [
+        r'\b_[a-z_]+\s*:',  # Internal fields like _id:, _source:
+        r'\bscript\s*:',    # Script injection
+    ]
+    for pattern in dangerous_field_patterns:
+        sanitized = re.sub(pattern, '', sanitized, flags=re.IGNORECASE)
+    # Remove excessive wildcards that could cause ReDoS
+    sanitized = re.sub(r'\*{2,}', '*', sanitized)
+    sanitized = re.sub(r'\?{2,}', '?', sanitized)
+    return sanitized.strip() or "*"
+
+
+def validate_qdrant_filter(filters: Any) -> Dict[str, Any]:
+    """
+    Validate and sanitize Qdrant filter objects.
+    Only allows whitelisted fields to prevent filter injection.
+    """
+    if not isinstance(filters, dict):
+        return {}
+    validated: Dict[str, Any] = {}
+    for key, value in filters.items():
+        if key in ALLOWED_QDRANT_FILTER_FIELDS:
+            validated[key] = value
+    return validated
+
 
 def _ensure_embedder(model_name: str) -> "SentenceTransformer":
     global _EMBED_MODEL, _EMBED_MODEL_NAME
@@ -415,7 +456,7 @@ def build_query_payload(
 
     if use_query_string:
         base_fields = ["title^3", "description^2", "transcript_full", "transcript_secondary_full"]
-        qs_query = (query or "").strip() or "*"
+        qs_query = sanitize_query_string(query or "")
         query_body: Dict[str, Any] = {
             "query_string": {
                 "query": qs_query,
@@ -1496,9 +1537,9 @@ def create_app(config: AppConfig = CONFIG) -> Flask:
     def api_vector_search():
         payload = request.get_json(silent=True) or {}
         query_text = (payload.get("query") or "").strip()
-        filters = payload.get("filters") or {}
-        limit = max(int(payload.get("size", 10)), 1)
-        offset = max(int(payload.get("offset", 0)), 0)
+        filters = validate_qdrant_filter(payload.get("filters"))
+        limit = min(max(int(payload.get("size", 10)), 1), MAX_QUERY_SIZE)
+        offset = min(max(int(payload.get("offset", 0)), 0), MAX_OFFSET)
 
         if not query_text:
             return jsonify(
@@ -1667,7 +1708,8 @@ def create_app(config: AppConfig = CONFIG) -> Flask:
 def main() -> None:  # pragma: no cover
     logging.basicConfig(level=logging.INFO, format="%(levelname)s %(message)s")
     app = create_app()
-    app.run(host="0.0.0.0", port=8080, debug=True)
+    debug_mode = os.environ.get("FLASK_DEBUG", "0").lower() in ("1", "true")
+    app.run(host="0.0.0.0", port=8080, debug=debug_mode)
 
 
 if __name__ == "__main__":  # pragma: no cover