User Tools

Site Tools


server:running_the_server:signals

Kill Signals

The crossfire main binary server handles several signals sent to it :

  1. 1 SIGHUP
  2. 2 SIGINT
  3. 3 SIGQUT
  4. 7 SIGBUS
  5. 11 SIGSEGV
  6. 13 SIGPIPE
  7. 15 SIGTERM

1 Hang Up [SIG]HUP

This signal forces the crossfire server unfortunately not to reload the configuration files,
but reopens the log file in append mode, if the logging output is not send to stderr ( of some terminal ) .

Syntax : kill -1 `pidof crossfire-server`

The whole path to the logfile is defined in server/include/config.h as absolute path-name :

  • Windows : #define LOGFILE “var\\crossfire.log”
  • #else : #define LOGFILE “/var/log/crossfire/logfile”

The configure script inside the source does not alter that path, even when the –prefix option had been applied.
The server may choke at start up, if the directory var or /var/log/crossfire does not exist,
or has not set proper writable permissions.

The source code file server/common.logger.c opens the existing logfile
in append mode inside the function void LOG (LogLevel logLevel, const char *format, …) :

if ((logfile = fopen(settings.logfilename, "a")) == NULL)

“a” means append :

a Open for appending (writing at end of file). The file is
created if it does not exist. The stream is positioned at the
end of the file.

From manual page http://man7.org/linux/man-pages/man3/fopen.3.html

FIXME It makes no sense to append to a logfile; I have a 55 MB logfile and would like the idea, to completely overwrite it, without the need to shut down the server entirely.
In the case of a long-time running server like the metalforge one , this log file could crash the Operating System because of it's sheer size ( filling up the partition ).

2 Interrupt [SIG}INT

This signal actually shutdowns the crossfire server, with leaving a message about it in the logfile, having received that signal. Calls fatal_signal(0, 1) .

Syntax : kill -2 `ps -o pid,args | grep -e 'crossfire-server' | grep -v 'grep' | awk '{print $1}'`
Of course, pidof is much easier to use … :-)

3 Quit [SIG]QUIT

Writes log at level info and calls fatal_signal(1, 1) .

7 [SIG]BUS

Only handled if #ifdef SIGBUS was defined at server compiling time; writes to log level error and calls fatal_signal(1, 1)

11 Segmentation Fault [SIG]SEGV

Should write to log level error and calls fatal_signal(1, 1) .

13 [SIG]PIPE

On Windows OS writes at log level error and calls fatal_signal(1, 1), on other Operating Systems writes at log level info and is otherwise ignored.

15 Termination [SIG]TERM

Writes at log level info and calls fatal_signal(0, 1) .

Technical

The file that contains the signal handling code is server/server/init.c
Functions are named rec_sigNAME like

void rec_sigint(int i) {
  LOG(llevInfo,"\nSIGINT received.\n");
  fatal_signal(0, 1);
}

Except for rec_sighup , these functions all call the function fatal_signal with the parameters to make_core and close_sockets .

void fatal_signal(int make_core, int close_sockets) {
  if(init_done) {
    emergency_save(0);
    clean_tmp_files();
  }
  if(make_core)
    abort();
  exit(0);
}
server/running_the_server/signals.txt · Last modified: 2018/03/03 15:03 by karl