control

name

control

description

Use when controlling a running Synchronet instance — recycling/reloading, shutdown, pause, clearing the failed-login list, or forcing a timed event. Covers cross-platform **semaphore files** in `ctrl/`/`data/` (`recycle`/`shutdown`/`pause`/`clear` with per-server `.<service>` and per-host `.<hostname>` suffix variants for shared-`ctrl/` setups), POSIX signals to `sbbscon` (SIGTERM/etc. = quit, SIGHUP = recycle-all), OS service managers (systemctl, service, launchctl, sc.exe), Windows front-ends (sbbsctrl, sbbsNTsvcs), the `node rerun` utility, the "won't recycle while in use" gotcha, and the `NO_RECYCLE` flag. For control via MQTT — often preferred even on the same host — see the `mqtt` skill. Trigger on "recycle the web server", "restart sbbs", "pause connections", "clear failed-login list", "send a signal to sbbs", "force a timed event", or "remove an IP ban".

Synchronet — controlling a running instance

Reference (authoritative):

• Semaphore files inventory: https://wiki.synchro.net/config:semfiles
• sbbscon command-line / signal flags: https://wiki.synchro.net/monitor:sbbscon
• sbbsctrl (Windows GUI): https://wiki.synchro.net/monitor:sbbsctrl
• sbbsNTsvcs (Windows NT Services): https://wiki.synchro.net/monitor:ntsvcs
• "Why doesn't my recycle take effect?" FAQ: https://wiki.synchro.net/faq:nix#recycle
• node control utility: https://wiki.synchro.net/util:node
• For remote / cross-host control via MQTT: see the mqtt skill.

The skill captures which lever is right for which job. Synchronet's own semaphore-file mechanism is the cross-platform native primitive — it works the same on Linux, BSD, macOS, and Windows, with or without systemd / sbbsctrl / NT Services in front. OS-level service managers and signals are the escalation path when the semfile isn't enough.

Choose the gentlest lever that does the job

   Less disruptive ─────────────────────────────────────────────► More disruptive
   ┌────────────┬────────────┬────────────┬────────────┬────────────┬────────────┐
   │ touch      │ touch      │ touch      │ kill -HUP  │ systemctl  │ kill -9 /  │
   │ ctrl/clear │ ctrl/pause │ ctrl/recyc │ <pid>      │ stop+start │ TerminateP │
   │ (or .S)    │ (or .S)    │ le (or .S) │ (recycle-  │ / sc stop  │ rocess     │
   │            │            │ - .ini     │  all)      │  / sc      │            │
   │            │            │   reload   │            │  start     │            │
   │            │            │ - threads  │            │            │            │
   │            │            │   restart  │            │            │            │
   └────────────┴────────────┴────────────┴────────────┴────────────┴────────────┘
   (in-process    (existing    (full server  (every server  (process     (data-loss
    list clear)   sessions     reinit;       in one         restart;     risk; only
                  remain)      no process    handler;       any open     for hung
                               restart)      no process     sessions     processes)
                                             restart)      drop)

The skill below documents each rung. Default to the gentlest one that achieves the goal, and never reach for kill -9 while a less destructive option is untried.

Synchronet semaphore files (cross-platform primitive)

Synchronet servers poll for semaphore files in ctrl/ (mostly) or data/ (events). When the mtime of a known semfile path advances, the corresponding action fires on the next poll tick (SemaphoreCheckFrequency, typically a couple of seconds). Contents are mostly ignored — touching the file is the signal. The exception is clear, whose first line is read as an optional IP filter.

(Each operation below has an MQTT equivalent that's preferable in several situations — see When MQTT is preferred below. The semfile path is documented first because it's the always-available, cross-platform primitive; choose between them based on the criteria in that section.)

The four lifecycle actions

Action	File (global)	What happens
recycle	ctrl/recycle	Server terminates its threads, closes sockets, re-reads its INI(s), and restarts — same process, new state. No client of that server stays connected. Other servers untouched (unless the global file fires them all).
shutdown	ctrl/shutdown	Server exits permanently — same process keeps running for other servers but this one is gone until the process restarts.
pause	ctrl/pause	Server stops accepting new connections; existing sessions continue. State becomes paused. Removing the file resumes. (Note: pause uses file existence, not mtime — delete to resume.)
clear	ctrl/clear	Clears the in-memory failed-login attempt list (and, for the term server, the max-concurrent-connection strike list). First non-empty line of the file is read as an optional IP — present → clear only that IP, absent → clear all.

Targeting: scope by suffix

For any action <A>, Synchronet polls the following paths in ctrl/ (verified against xpdev/semfile.c::semfile_list_init):

Path	Scope
ctrl/<A>	every server on every host
ctrl/<A>.<service>	one server: <service> ∈ term, web, mail, ftp, services. (The services server's own polled abbrev is the long form services, not srvc; per-server config uses srvc internally. Use .services for the semfile.)
ctrl/<A>.<hostname>	every server on this host only (matched against gethostname()'s FQDN)
ctrl/<A>.<hostname>.<service>	one server on this host only
ctrl/<A>.<short-hostname>	as above but matched against the short hostname (first dot-separated component)
ctrl/<A>.<short-hostname>.<service>	short hostname + one server

The hostname variants are essential when ctrl/ is on shared storage (NFS/SMB/CIFS/Syncthing) and you only want one host to react. Plain ctrl/recycle would otherwise fire on every host that mounts the share. (See the logs skill's "Multi-instance traps" section for the symptom on the log side.)

Examples:

touch  ctrl/recycle                          # recycle every server on every host
touch  ctrl/recycle.web                      # recycle just the web server, all hosts
touch  ctrl/recycle.web1.example.com         # recycle every server only on host web1.example.com
touch  ctrl/recycle.web1.example.com.web     # recycle just the web server on host web1
echo 203.0.113.5 > ctrl/clear.term           # clear that one IP from the term server's fail list
echo                > ctrl/clear             # (empty) clear every server's fail list everywhere

Special recycle triggers (config-file watches)

Each server's polling loop also includes its INI file in the recycle list — so editing and saving the config implicitly recycles the server:

Server	Watched INI(s)	Source
Terminal	ctrl/sbbs.ini, ctrl/text.dat, ctrl/text.ini, ctrl/attr.ini	main.cpp
Web	ctrl/sbbs.ini	websrvr.cpp
Mail	ctrl/sbbs.ini	mailsrvr.cpp
FTP	ctrl/sbbs.ini	ftpsrvr.cpp
Services	ctrl/sbbs.ini, ctrl/services.ini	services.cpp

Touching ctrl/sbbs.ini therefore recycles every server. That's intentional — sysop edits (SCFG saves, text.dat regen via textgen, etc.) become live without an explicit touch ctrl/recycle. Side effect: a backup tool, rsync without -t/-a, or any utility that touches mtime can accidentally trigger a recycle; tell backups to preserve mtime, or filter ctrl/sbbs.ini/ctrl/text.dat/ctrl/*.ini out of any mtime-modifying sync.

Event-trigger semfiles in data/

A separate category — these don't restart servers, they fire timed events:

File	Effect
data/<event>.now	Force the event with internal code <event> (SCFG → External Programs → Timed Events) to run on the term server's event thread
data/prepack.now	Force the daily QWK pre-pack event now
data/pack<NNNN>.now	Pack a QWK message packet for user #<NNNN> (4-digit, zero-padded); result lands in data/file/<NNNN>.qwk
data/qnet/<hub-id>.now	Force a QWKnet call-out to that hub
ctrl/sysavail.chat	Existence indicates sysop is available for chat (users get the page-sysop option)

"Won't recycle while in use" gotcha

If a recycle is requested but a server's clients/nodes are still active, the recycle is deferred until they all disconnect — it doesn't drop sessions. The node utility shows which nodes are flagged [R] (recycle pending):

node                # show node states (look for [R] = recycle-pending)
node rerun          # toggle the "rerun on next disconnect" flag for ALL nodes
node 3 rerun        # just node 3

Servers with the NO_RECYCLE option flag in sbbs.ini ([BBS] Options, [Web] Options, etc.) ignore the recycle semfile. If touch-recycle isn't working, check that flag first. See faq:nix#recycle for the longer Linux-flavoured discussion.

POSIX signals (sbbscon on *nix)

sbbscon writes its PID to a .pid file at startup so you can find it (typically /var/run/sbbs/sbbs.pid or wherever PidFile in [UNIX] points). It then runs a dedicated sigwait() thread:

Signal	Handler	Effect
SIGTERM, SIGINT, SIGQUIT, SIGABRT	_sighandler_quit	Graceful shutdown — every server stops, the process exits. Equivalent to touch ctrl/shutdown for every server at once, with the process going away too.
SIGHUP	_sighandler_rerun	recycle_all() — sets recycle_now=true on bbs/ftp/web/mail/services. Same effect as touch ctrl/recycle, but no file is created on disk. Useful when ctrl/ is read-only or you want to avoid leaving artefacts.
SIGPIPE, SIGALRM	ignored (SIG_IGN)	broken-socket / alarm fall through harmlessly

SIGKILL (and kill -9) cannot be caught — they terminate immediately, dropping every open session and skipping cleanup. Only for hung processes. Try SIGTERM first; if it doesn't exit within a few seconds, then escalate.

OS service managers

Detect which one this host uses, then prefer the semfile approach for anything granular; use the service manager for full-process restarts and start/stop.

Linux (systemd)

systemctl list-units --type=service | grep -iE 'sbbs|synchro|httpd|ircd'
systemctl status   <unit>
systemctl restart  <unit>     # full process restart — drops all sessions
systemctl reload   <unit>     # sends SIGHUP if the unit is configured for it = recycle-all
systemctl stop     <unit>
systemctl start    <unit>
systemctl kill -s HUP <unit>  # explicit SIGHUP if `reload` isn't wired in

systemctl reload is the systemd-native way to trigger recycle (via SIGHUP) without restarting the process. Whether systemctl reload works depends on whether the unit file declares ExecReload=/bin/kill -HUP $MAINPID — if not, systemctl kill -s HUP <unit> does the same thing without unit-file changes.

Linux (sysvinit / OpenRC / others)

service sbbs <start|stop|restart|reload>     # SysV-init wrapper
/etc/init.d/sbbs <start|stop|restart|reload> # direct
rc-service sbbs <start|stop|restart>         # OpenRC

# Always available regardless of init system:
kill -HUP  $(cat /var/run/sbbs/sbbs.pid)     # recycle-all via signal
kill -TERM $(cat /var/run/sbbs/sbbs.pid)     # graceful shutdown

macOS (launchd)

launchctl list | grep -i sbbs
sudo launchctl kickstart -k system/<label>    # stop + start
sudo launchctl unload  /Library/LaunchDaemons/<plist>
sudo launchctl load    /Library/LaunchDaemons/<plist>

The semfile and signal mechanisms work identically on macOS — kill -HUP $(cat <pidfile>) recycles without going through launchd.

Windows (sbbsNTsvcs)

sc query   SynchronetBBS                     # status (one per service: BBS/FTP/Mail/Web/Services)
sc stop    SynchronetBBS
sc start   SynchronetBBS
sc pause   SynchronetBBS                     # not all services support this
sbbsNTsvcs install / remove / start / stop   # the bundled wrapper from exec/
services.msc                                 # GUI

sbbsNTsvcs.exe reads the SBBSCTRL environment variable for the ctrl directory path. Semaphore files in that ctrl/ work the same as on Unix — echo. > C:\sbbs\ctrl\recycle does the right thing.

Windows front-ends

sbbsctrl (GUI)

The Synchronet Control Panel hosts the servers in-process. Recycle / Pause / Shutdown live in each server's pane menu and in the top-level BBS menu. Closing sbbsctrl shuts down every server it was hosting.

sbbs.exe (Windows console mode)

Same servers, same code, no GUI; output goes to the console window. Use Ctrl+C for graceful shutdown (mapped to SIGINT-equivalent) and the semfile mechanism for everything else.

Choosing between them

Want	Use
Always-on, survives logout, no user logged in	sbbsNTsvcs
GUI monitoring + log panes (the WS<date>.LOG files come from here — see logs)	sbbsctrl
Foreground debugging, console output visible	sbbs.exe

When MQTT is preferred (even on the same host)

Every operation in this skill has an MQTT-published equivalent: host/+/recycle, host/+/pause, host/+/resume, host/+/clear, per-server server/+/recycle/clear, per-node node/+/set/*, etc. See the mqtt skill for broker connect/auth and the full control-plane topic list. MQTT and semfiles are not mutually exclusive — both work on the same install, and either is correct for most operations. Reach for MQTT (rather than a semfile or signal) when any of the following applies:

• You don't have shell access to the host, but you do have broker credentials (a dashboard, IRC bot, JS module, remote agent).
• You want positive acknowledgement. The retained …/server/+ status topic reflects the new state within a few seconds — stopping → initializing → ready after a recycle confirms it took effect. Semfiles give no feedback channel; you have to watch the journal yourself.
• You want to fan out across multiple hosts in one shot (a single publish to host/+/recycle reaches every host running this BBSID; semfiles on shared ctrl/ can do the same but require per-host hostname-suffix files if you want different targets).
• You're already in MQTT tooling. A script subscribed to status topics shouldn't bounce out to the filesystem for the action — same channel, immediate, auditable on the broker side.
• Faster reaction. Semfile polling happens on SemaphoreCheckFrequency (typically a couple of seconds); MQTT delivery is event-driven and effectively immediate.
• ctrl/ is on slow / laggy shared storage. NFS over WAN, intermittently-mounted CIFS, eventually-consistent object storage — mtime detection there can be late or inconsistent. MQTT bypasses the filesystem.
• Filesystem write access to ctrl/ is restricted but the BBS user can publish to its own broker.

Reach for semfiles (rather than MQTT) when:

• MQTT isn't enabled on the install ([MQTT] Enabled=false in main.ini), or you can't reach the broker.
• You want a durable on-disk record that the operation was requested (ctrl/recycle.<host> sticks around; an MQTT publish disappears once delivered, modulo broker logs).
• You're scripting from inside the BBS user's shell with no MQTT client installed and adding mosquitto-clients is more friction than touch.

One thing MQTT does not buy you: a bypass for the NO_RECYCLE option flag. A server with NO_RECYCLE set in sbbs.ini ignores every recycle path — semfile, SIGHUP, and MQTT — because the check is on the recycle_now flag itself, which all three mechanisms feed. The only way to apply a config change to a NO_RECYCLE server is a full process restart (service-manager stop+start). Shutdown still works via every mechanism regardless.

Equivalence reference

Operation	Semfile (any OS)	Signal (*nix)	systemd	sbbsNTsvcs	MQTT
Recycle all servers	touch ctrl/recycle	kill -HUP <pid>	systemctl reload <unit> or kill -s HUP	sbbsctrl menu / restart each service	host/+/recycle
Recycle one server	touch ctrl/recycle.<S>	(no per-server signal)	(restart the per-server unit if split)	sc stop/start Synchronet<S>	host/+/server/<S>/recycle
Shutdown everything	touch ctrl/shutdown + wait	kill -TERM <pid>	systemctl stop <unit>	sc stop each	(no single MQTT topic; combine pause+OS stop)
Stop accepting (keep existing)	touch ctrl/pause	(none)	(none)	sc pause (limited)	host/+/pause
Resume	rm ctrl/pause	(none)	(none)	sc continue	host/+/resume
Clear failed-login list	touch ctrl/clear	(none)	(none)	(none)	host/+/clear (empty payload)
Clear one IP	echo IP > ctrl/clear	(none)	(none)	(none)	host/+/clear (IP as payload)
Force a timed event	touch data/<event>.now	(none)	(none)	(none)	(none — local only)

Safety patterns

• Verify the lever took effect. After a recycle, watch the journal (journalctl -fu <unit>) or the MQTT …/server/+ status topic. After a clear, the next deny-at-accept hit from the released IP confirms the list was cleared (or didn't get cleared, if you see fresh !CLIENT BLOCKED lines for the same IP straight away).
• Prefer per-server over global. ctrl/recycle.web only restarts web — five hosts × five servers becomes one host × one server instead of 25 restarts.
• Use hostname variants on shared ctrl/. Without them, every host mounting the share reacts to the same global semfile. (Diagnose the symptom from the log side via logs.)
• pause instead of shutdown for maintenance. Pause lets you watch existing sessions drain naturally; shutdown kills them. Resume by rm-ing the pause file (or publishing to the MQTT resume topic).
• Touch sbbs.ini deliberately or not at all. It's a recycle trigger via mtime. Backup/sync tools that preserve mtime won't trigger it; tools that don't (cp without -p, rsync without -t) will.
• Beware "stuck" recycles. If node shows nodes flagged [R] for a long time, those users are still connected — the recycle is waiting for them. Either let them disconnect or, in an emergency, force-drop via node <N> down or the MQTT node/<N>/set/intr topic.
• kill -9 is a last resort. It bypasses every cleanup path (sockets stuck in TIME_WAIT, lockfiles left behind, MQTT will-messages not sent). Try SIGTERM and a few seconds of patience first.

Common mistakes

• Reaching for systemctl restart for a config reload. That's a full process restart — drops every session on every server. systemctl reload (if wired) or touch ctrl/recycle does it without the process-restart penalty.
• Touching ctrl/recycle on a shared-storage ctrl/ directory. It fires on every host that mounts the share. Use ctrl/recycle.<hostname> to scope.
• Expecting ctrl/pause to drain by mtime update. It uses existence, not mtime — touch to pause, rm to resume. (ctrl/recycle/shutdown/clear use mtime; pause is the odd one out.)
• Setting clear to clear a specific IP but trailing-newline-truncating the IP. echo -n 203.0.113.5 > ctrl/clear and echo 203.0.113.5 > ctrl/clear both work — Synchronet trims trailing whitespace via truncsp(). But printf "203.0.113.5\0junk" and similar weird payloads might not.
• Treating kill -HUP as a per-server signal. It isn't — it recycles every server in the process (recycle_all()). For per-server, use the semfile.
• sc stop SynchronetBBS when meaning "recycle". That stops the service entirely; you then need a matching sc start. There's no native sc reload — use the semfile or the MQTT recycle topic.
• Editing sbbs.ini by hand and being surprised by an instant recycle. Every server watches it. If you're doing a bulk edit, copy it aside, edit the copy, and move it back atomically when ready.
• Looking for recycle_now flags in main.ini. They're runtime in-memory state set by recycle_all()/semfile detection, not config-file settings. The NO_RECYCLE option flag is a config setting (per server), and disables the recycle response.

Cross-references

• For the file/log side of these operations (the !Recycling, !Shutdown semaphore file detected, !Pause semaphore signaled console lines, and the data/error.log entries that follow): the logs skill.
• For the MQTT equivalents (remote control, cross-host fan-out, retained status feedback): the mqtt skill.
• For building Synchronet (not running it): the synchronet-build skill.
• For source-level verification of the semfile naming algorithm: src/xpdev/semfile.c::semfile_list_init. For the per-server polling loops: main.cpp, websrvr.cpp, mailsrvr.cpp, ftpsrvr.cpp, services.cpp — search for semfile_list_check. For signal handlers: src/sbbs3/sbbscon.c — search for _sighandler_quit and _sighandler_rerun.