Skip to content

Protocol behaviour

What an attacker sees on each protocol, and what ReD-Pot logs.

Protocol Behaviour Logged as
SMB SMB 2.0.2 / 2.1 only (SMB1 refused like a hardened Server 2019); persona OS version, names and a stable ServerGUID; a Windows-like share list (ADMIN$, C$, IPC$) for a logged-on bait user. A null session cannot list shares but can open the read-only bait shares by name. IPC$ is pipe-only, and ADMIN$ / C$ refuse access. A named logon succeeds only with a bait password valid on this decoy (NTLMv2/v1 verified); anything else gets STATUS_LOGON_FAILURE. Uploads to Public go to an inert upload store; other shares deny access user, domain, workstation name, shares, files read and written. Failed logons feed spraying detection; bait name with a wrong password = L4, correct bait password = L5; any use of IPC$, ADMIN$ or C$ as a disk = smb.admin_share (L3)
SSH Ed25519 / ECDSA / RSA host keys, password and key authentication, a simulated bash with quoting, $VAR, ; && \|\|, pipes (grep, head, tail, wc, sort) and redirections. Nothing is executed every command; download URLs are recorded, never fetched
HTTP / HTTPS webmail, helpdesk or intranet front page, /admin login, .env, Apache 2.4.52 (Ubuntu) error pages and header order, Apache's default SSL site on 443 path, user agent, scanner verdict; exploit payloads (Log4Shell, path traversal ...) and unusual methods raise http.scanner
FTP vsftpd banner, listing and download (root /srv/ftp, also visible over SSH). Uploads are stored inert (mode 0400, named by SHA-256) within a per-sensor quota shared with SMB logins, files; uploads over the quota are logged with dropped
RDP X.224 + TLS + CredSSP/NLA like Windows Server 2019 with NLA required; persona NTLM challenge (what rdp-ntlm-info shows). The NTLM AUTHENTICATE message gives user, domain and workstation; the logon never completes (correct bait password: "logon type not granted", else logon failure) rdp.auth (L3) with user, domain and workstation; L5 for a correct bait password, L4 for a bait name with a wrong one
LDAP / LDAPS / GC real rootDSE with Windows BER encoding, AD-accurate errors, simple, Sicily NTLM and SASL GSS-SPNEGO NTLM binds verified against the bait accounts, real filter evaluation. LDAPS and GC-over-SSL present a certificate from a per-install fake enterprise CA following the "Domain Controller" template binds (with workstation) and searches, for example servicePrincipalName=*
Kerberos a real KDC for the bait accounts: the correct password gives a TGT; the TGS for a honey SPN is encrypted with that account's key, so the Kerberoast hash cracks to the bait password. Everyone else gets PREAUTH_REQUIRED, C_PRINCIPAL_UNKNOWN or PREAUTH_FAILED user enumeration, password guesses (auth.failed, feeds spraying), TGT issued, Kerberoasting
CLDAP / NetBIOS / NTP (DC, UDP) netlogon "LDAP ping" answered from the persona; NetBIOS name and node status with the decoy NIC's real MAC; NTP server (stratum 3) DC locator pings, nbtstat -A / nbtscan, time queries
DNS authoritative for your zone only, never an open resolver. Out of zone: REFUSED (BIND on the DNS role) or SERVFAIL (Microsoft DNS on the DC). version.bind: BIND 9 on the DNS role, NOTIMP on the DC. UDP is rate-limited, replies are at most 4 times the request size. Zone transfers are refused lookups, zone-transfer (AXFR) attempts, version.bind fingerprinting

Resource limits

Limit Value
Concurrent connections per decoy 250
Connections per source IP 20
Thread ceiling for new clients 320 (container limit 384)
SMB session without named logon, idle closed after 60 s
SMB named pipes 8 per session, 16 per source IP, 128 per decoy
SSH channels per connection 10
Upload quota per sensor (SMB + FTP) RPOT_UPLOAD_MAX_MB 200, RPOT_UPLOAD_MAX_FILES 500, RPOT_UPLOAD_MAX_FILE_MB 20

A source that hits a cap raises a level-3 conn.flood event (once a minute per source). Hitting its own pipe or channel cap raises a level-4 event. When the decoy-wide cap refuses a client, the manager opens an L4 "Decoy NAME saturated - possible DoS" incident naming the top source addresses.