Skip to main content

Security intelligence and review

Netra now offers native DNS QTYPE telemetry, managed threat-feed history, flat deny-predicate review and temporal security correlation. These are metadata-based features inspired by enterprise firewall workflows. They are not a PAN-OS API integration or a claim of full NGFW/IDS/DLP parity.

DNS query type​

The matched UDP/53 response event carries dnsQType. The parser requires one complete, uncompressed first question and IN class, bounded to 96 question-name bytes. Truncated, compressed, oversized or unsupported questions return zero (unknown). This enables the detector's existing txt-heavy signal when QTYPE=16; a TXT query alone is not an attack. No answer records or application payloads are retained. Encrypted DNS, TCP DNS and stream reassembly are outside this parser's scope.

QTYPE consumes two bytes of existing event padding at offset 194. The packed event remains 196 bytes, all previous offsets stay unchanged, and old agents simply leave QTYPE zero. No pinned-map ABI changes.

Feed lifecycle​

PUT /api/v1/intel/feed accepts the existing JSON, CSV or line list. IP literals, masked CIDRs and lowercased hostnames are canonicalized before matching and deduplication. Manual imports return the accepted preview and skipped entries. Optional ?ttl=1h sets overall feed expiry (0..720h). A missing TTL preserves the existing non-expiring operator workflow. Use X-Netra-Intel-Revision: N on PUT, DELETE or rollback for compare-and-set updates. A revision conflict returns 409 without changing state.

  • GET /api/v1/intel/history: last 16 revisions, including entry snapshots.
  • POST /api/v1/intel/rollback/{revision}: append a revision restoring a retained snapshot and its original expiry. Expired revisions cannot be restored. Rollback never imports entries into deny maps.
  • DELETE /api/v1/intel/feed: append an empty revision.
  • netractl intel history / netractl intel rollback REVISION.

Expiry removes the feed from live matching and new applications. It does not revoke already imported deny entries; those retain Netra's existing leased enforcement behavior. Applying a feed still requires an active lease and X-Netra-Confirm-Risk: high. Feed writes include auditRecorded so operators can see if the separate controller audit write failed.

Set NETRA_INTEL_STATE_PATH to an operator-provisioned writable file path for persistence. The parent directory must already exist. Writes use a private 0600 temporary file, fsync and atomic rename before publishing in memory. Invalid/unreadable journals report journalUnavailable, leave the feed empty and refuse updates; repair the journal before writing again. Memory-only operation remains the default. In HA, put the journal on the same leader-managed shared persistent storage as controller state. The journal holds metadata and indicator snapshots, not packet payloads.

Optional HTTPS refresh​

Environment variableDefaultMeaning
NETRA_INTEL_SOURCE_URLunsetExplicit operator-configured HTTPS feed; unset disables polling
NETRA_INTEL_REFRESH_INTERVAL15mPoll interval, 1m..24h
NETRA_INTEL_TTL1hExpiry after a successful refresh, at least twice the interval, at most 720h

Refresh runs immediately then on the interval, only during a controller leader stint in HA. Demotion cancels and waits for the poller. TLS uses normal certificate validation. Redirects, URL userinfo and fragments are rejected. Fetches have a 15s timeout and 1 MiB body limit. Empty, malformed, partially parsed, duplicate, truncated and over-cap feeds retain the last good revision. Revision compare-and-set prevents a download from replacing an intervening operator update. Query strings are stripped from persisted provenance; public errors do not echo URLs or remote bodies. Only configure trusted feed URLs; this downloader does not discover or crawl sources. lastRefresh / refreshError show the most recent attempt. Refresh never applies firewall rules, changes mode, or extends an enforcement lease.

Deny-predicate review​

GET /api/v1/security/optimizer / netractl security optimizer reports IP, CIDR and port predicates with equivalent, covered or overlapping predicate suggestions. Addresses/families, directions, protocols and TCP SYN restrictions are considered. Suggestions are capped at 200 and the review at the first 1,000 predicates; truncated makes limits visible. Current configuration is not changed.

matchingEvents counts fresh sampled agent events within ten minutes, excluding stale reports. Port predicates use the packet destination port, as the datapath does. Counts are not winning-rule hits or evidence that a rule is unused. Allow exceptions, native NetPol, workload scope, conntrack, hook order and the enforcement lease can change actual outcomes. TCP SYN restrictions vary by hook; the sampler accounts for their cgroup/TC check. There is no automatic deletion or policy-precedence proof.

Security correlation​

GET /api/v1/security/incidents / netractl security incidents joins recent DNS findings, scan findings and dated agent events matching active IP/CIDR feed entries. It requires exact node/namespace/pod attribution, a non-stale reporting node, timestamps in the past ten minutes and at least two different evidence kinds. Unknown attribution, future timestamps and stale evidence are excluded. This is temporal correlation, not proof of causality or an attack. It does not invent timestamps for aggregate counters. DNS/SNI feed matches remain available through their existing APIs.

IDs and ordering are deterministic. Output is capped at 200 incidents and 50 evidence items per incident while retaining each independent kind; truncated is explicit. Detector-enabled flags distinguish disabled sensors from an empty result. No quarantine or deny is automatically applied.

The dashboard exposes Security → Security Review, with feed history, rule review, independent evidence and refresh/error states. MCP tools netra_security_optimizer, netra_security_incidents and netra_intel_history are always read-only. No mutation tools were added.