Da

✏️ Editing: hc-ftpd-auth-proxy.service

[Unit]
Description=pure-ftpd / pure-authd authentication proxy (per-rule IP restrictions)
# Wants: order the proxy after pure-authd without binding to its lifecycle.
After=pure-authd.service
Wants=pure-authd.service

[Service]
Type=simple
# --check-config gates startup: a broken config fails here (ExecStartPre) so a
# bad deploy never brings FTP auth down with the daemon half-up.
ExecStartPre=/usr/src/hc-ftpd-auth-proxy/hc-ftpd-auth-proxy --config /etc/hc-ftpd-auth-proxy/config.yaml --check-config
ExecStart=/usr/src/hc-ftpd-auth-proxy/hc-ftpd-auth-proxy --config /etc/hc-ftpd-auth-proxy/config.yaml
# Runs as root (no User=) so it can bind /var/run/ftpd.sock — pure-authd's old
# path — which lets pure-ftpd reach the proxy with no reconfiguration. /run is
# root:root 0755, so an unprivileged user could not create that socket. The
# sandbox below contains the root process.
NoNewPrivileges=true
ProtectSystem=strict
# ProtectSystem=strict mounts the whole tree read-only; carve out the only paths
# the daemon must write: /run to bind /var/run/ftpd.sock, and /var/log for the
# optional log_file (omit /var/log if you only log to journald via stderr).
ReadWritePaths=/run /var/log
ProtectHome=true
PrivateTmp=true
PrivateDevices=true
ProtectKernelTunables=true
ProtectKernelModules=true
ProtectControlGroups=true
RestrictAddressFamilies=AF_UNIX
SystemCallFilter=@system-service
SystemCallErrorNumber=EPERM
LockPersonality=true
MemoryDenyWriteExecute=true
Restart=always
RestartSec=3
# CPython exits 120 when flushing stdout/stderr fails during interpreter
# finalization (e.g. a long-lived process whose journald stream was severed).
# The drain still completed; don't let that artifact register as a failure.
SuccessExitStatus=120
# Allow the graceful drain (request + backend timeout + slack) to finish before
# systemd escalates to SIGKILL. The Python side computes its grace as
# `request_timeout_seconds + backend_timeout_seconds + 2s`; with the shipped
# defaults that's 17s. The 60s cap leaves headroom for operators who tune the
# timeouts upward (e.g. backend_timeout=30 + request_timeout=10 -> 42s grace).
# Keep this comfortably above the worst configured grace, or systemd will
# SIGKILL mid-drain and drop in-flight auth requests.
KillMode=mixed
TimeoutStopSec=60

[Install]
WantedBy=multi-user.target

📁 Sid Gifari File Manager