Look for unusual HTTP/2 sessions.
Xint observed a long-lived connection receiving much more data than it sent, alongside many short connections from the same source. Inspect network counters for an affected listener and compare them over time.
sudo ss -tinp state established '( sport = :443 )'Large uploads and legitimate load tests can create similar patterns. A CDN may hide the original client at the origin; use its connection telemetry as well.
Add a connection identifier.
Where safe for your logging policy, include $connection, $connection_requests, protocol and request length in a temporary incident log format. HTTP/2 control frames still will not appear as requests.
log_format h2_watch '$time_iso8601 src=$remote_addr conn=$connection '
'reqs=$connection_requests proto=$server_protocol '
'status=$status request_length=$request_length';Compare logged request lengths against received connection bytes. A large gap is a lead, not proof of this exploit.
Investigate abnormal execution.
An unexpected child process of an NGINX worker is a stronger signal than a worker crash. Monitor process creation, egress and file writes by the worker account. The account can also be shared with other services, so preserve PID and parent PID.
ps -eo uid,user,pid,ppid,args | grep '[n]ginx: worker process'
sudo grep -Ei 'worker process.*exited|signal 11|segfault' /var/log/nginx/error.logCore dumps, if enabled, can aid analysis. A crash alone can have benign causes or reflect a failed attempt.
Correlate before declaring compromise.
Preserve edge and origin logs, NGINX error logs, connection statistics, process trees and deployment records. Inspect whether suspicious traffic reached the actual NGINX listener. A quiet access log does not rule out control-frame traffic.
Understand the demonstrated exploit