| -rw-r--r-- | src/configuration.cc | 5 | ||||
| -rw-r--r-- | src/dudki.cc | 7 | ||||
| -rw-r--r-- | src/process.cc | 118 | ||||
| -rw-r--r-- | src/process.h | 6 |
4 files changed, 124 insertions, 12 deletions
diff --git a/src/configuration.cc b/src/configuration.cc index eb010c1..edc8c04 100644 --- a/src/configuration.cc +++ b/src/configuration.cc @@ -1,149 +1,154 @@ #include <stdexcept> using namespace std; #include <dotconf.h> #include "configuration.h" #ifndef DEFAULT_PID_FILE # define DEFAULT_PID_FILE "/var/run/dudki.pid" #endif configuration::configuration() : check_interval(60), pidfile(DEFAULT_PID_FILE), daemonize(true) { } enum dc_ctx { DCC_ROOT = 1, DCC_PROCESS = 2 }; struct dc_context { dc_ctx ctx; configuration* cf; process* ps; dc_context() : ctx(DCC_ROOT), cf(NULL), ps(NULL) { } }; static DOTCONF_CB(dco_check_interval) { dc_context *dcc = (dc_context*)ctx; dcc->cf->check_interval = cmd->data.value; return NULL; } static DOTCONF_CB(dco_daemonize) { dc_context *dcc = (dc_context*)ctx; dcc->cf->daemonize = cmd->data.value; return NULL; } static DOTCONF_CB(dco_pid_file) { dc_context *dcc = (dc_context*)ctx; switch(dcc->ctx) { case DCC_ROOT: dcc->cf->pidfile = cmd->data.str; break; case DCC_PROCESS: dcc->ps->pidfile = cmd->data.str; break; default: return "Unexpected PidFile"; } return NULL; } static DOTCONF_CB(dco_mailto_header) { dc_context *dcc = (dc_context*)ctx; if(cmd->arg_count!=2) return "Invalid number of arguments"; string h = cmd->data.list[0]; string v = cmd->data.list[1]; switch(dcc->ctx) { case DCC_ROOT: dcc->cf->mailto_headers[h] = v; break; case DCC_PROCESS: dcc->ps->mailto_headers[h] = v; break; default: return "Unexpected MailtoHeader"; } return NULL; } static DOTCONF_CB(dco_notify) { dc_context *dcc = (dc_context*)ctx; switch(dcc->ctx) { case DCC_ROOT: dcc->cf->notify = cmd->data.str; break; case DCC_PROCESS: dcc->ps->notify = cmd->data.str; break; default: return "Unexpected Notify"; } return NULL; } static DOTCONF_CB(dco_process) { dc_context *dcc = (dc_context*)ctx; string id = cmd->data.str; if(id[id.length()-1]=='>') id.erase(id.length()-1); dcc->ps = &(dcc->cf->processes[id]); dcc->ctx = DCC_PROCESS; return NULL; } static DOTCONF_CB(dco__process) { dc_context *dcc = (dc_context*)ctx; dcc->ps = NULL; dcc->ctx = DCC_ROOT; return NULL; } +static DOTCONF_CB(dco_process_name) { dc_context *dcc = (dc_context*)ctx; + dcc->ps->process_name = cmd->data.str; + return NULL; +} static DOTCONF_CB(dco_restart_command) { dc_context *dcc = (dc_context*)ctx; dcc->ps->restart_cmd = cmd->data.str; return NULL; } static DOTCONF_CB(dco_user) { dc_context *dcc = (dc_context*)ctx; dcc->ps->user = cmd->data.str; return NULL; } static DOTCONF_CB(dco_group) { dc_context *dcc = (dc_context*)ctx; dcc->ps->group = cmd->data.str; return NULL; } static DOTCONF_CB(dco_chroot) { dc_context *dcc = (dc_context*)ctx; dcc->ps->chroot = cmd->data.str; return NULL; } static const configoption_t dc_options[] = { { "CheckInterval", ARG_INT, dco_check_interval, NULL, DCC_ROOT }, { "Daemonize", ARG_TOGGLE, dco_daemonize, NULL, DCC_ROOT }, { "PidFile", ARG_STR, dco_pid_file, NULL, DCC_ROOT|DCC_PROCESS }, { "MailtoHeader", ARG_STR, dco_mailto_header, NULL, DCC_ROOT|DCC_PROCESS }, { "Notify", ARG_STR, dco_notify, NULL, DCC_ROOT|DCC_PROCESS }, { "<Process", ARG_STR, dco_process, NULL, DCC_ROOT }, + { "ProcessName", ARG_STR, dco_process_name, NULL, DCC_PROCESS }, { "RestartCommand", ARG_STR, dco_restart_command, NULL, DCC_PROCESS }, { "User", ARG_STR, dco_user, NULL, DCC_PROCESS }, { "Group", ARG_STR, dco_group, NULL, DCC_PROCESS }, { "Chroot", ARG_STR, dco_chroot, NULL, DCC_PROCESS }, { "</Process>", ARG_NONE, dco__process, NULL, DCC_PROCESS }, LAST_OPTION }; static const char *dc_context_checker(command_t *cmd,unsigned long mask) { dc_context *dcc = (dc_context*)cmd->context; if( (mask==CTX_ALL) || ((mask&dcc->ctx)==dcc->ctx) ) return NULL; return "misplaced option"; } static FUNC_ERRORHANDLER(dc_error_handler) { throw runtime_error(string("error parsing config file: ")+msg); } void configuration::parse(const string& cfile) { struct dc_context dcc; dcc.cf = this; dcc.ctx = DCC_ROOT; configfile_t *cf = dotconf_create((char*)cfile.c_str(),dc_options,(context_t*)&dcc,CASE_INSENSITIVE); if(!cf) throw runtime_error(string(__PRETTY_FUNCTION__)+": failed to dotconf_create()"); cf->errorhandler = (dotconf_errorhandler_t) dc_error_handler; cf->contextchecker = (dotconf_contextchecker_t) dc_context_checker; if(!dotconf_command_loop(cf)) throw runtime_error(string(__PRETTY_FUNCTION__)+": failed to dotconf_command_loop()"); dotconf_cleanup(cf); } diff --git a/src/dudki.cc b/src/dudki.cc index 9562079..c966695 100644 --- a/src/dudki.cc +++ b/src/dudki.cc @@ -1,287 +1,292 @@ #include <unistd.h> #include <signal.h> #include <syslog.h> #include <errno.h> #include <iostream> #include <fstream> #include <stdexcept> using namespace std; #include "configuration.h" #include "util.h" #include "config.h" #ifdef HAVE_GETOPT_H # include <getopt.h> #endif #ifndef DEFAULT_CONF_FILE # define DEFAULT_CONF_FILE "/etc/dudki.conf" #endif #define PHEADER PACKAGE " Version " VERSION #define PCOPY "Copyright (c) 2004 Klever Group" bool finishing = false; bool restarting = false; static char **_argv = NULL; static void lethal_signal_handler(int signum) { syslog(LOG_NOTICE,"Lethal signal received. Terminating."); finishing = true; } static void sighup_handler(int signum) { syslog(LOG_NOTICE,"SUGHUP received, reloading."); restarting = finishing = true; } void check_herd(configuration& config) { + process::prepare_herd(); for(processes_t::iterator i=config.processes.begin();i!=config.processes.end();++i) i->second.check(i->first,config); + process::unprepare_herd(); } void signal_self(const configuration& config,int signum) { ifstream pids(config.pidfile.c_str(),ios::in); if(!pids) throw runtime_error("Can't detect running instance"); pid_t pid = 0; pids >> pid; if(!pid) throw runtime_error("Can't detect running instance"); if(pid==getpid()) throw 0; if(kill(pid,signum)) throw runtime_error("Failed to signal running instance"); } int main(int argc,char **argv) { try { _argv = new char*[argc+1]; if(!_argv) throw runtime_error("memory allocation problem at the very start"); memmove(_argv,argv,sizeof(*_argv)*(argc+1)); string config_file = DEFAULT_CONF_FILE; enum { op_default, op_work, op_signal, op_ensure, op_test } op = op_default; int op_signum = 0; while(true) { #define SHORTOPTSTRING "f:hVLrkcets:" #ifdef HAVE_GETOPT_LONG static struct option opts[] = { { "help", no_argument, 0, 'h' }, { "usage", no_argument, 0, 'h' }, { "version", no_argument, 0, 'V' }, { "license", no_argument, 0, 'L' }, { "config", required_argument, 0, 'f' }, { "kill", no_argument, 0, 'k' }, { "reload", no_argument, 0, 'r' }, { "signal", required_argument, 0, 's' }, { "check", no_argument, 0, 'c' }, { "ensure", no_argument, 0, 'e' }, { "test", no_argument, 0, 't' }, { NULL, 0, 0, 0 } }; int c = getopt_long(argc,argv,SHORTOPTSTRING,opts,NULL); #else /* !HAVE_GETOPT_LONG */ int c = getopt(argc,argv,SHORTOPTSTRING); #endif /* /HAVE_GETOPT_LONG */ if(c==-1) break; switch(c) { case 'h': cerr << PHEADER << endl << PCOPY << endl << endl << " " << argv[0] << " [options] [processes]" << endl << endl << #ifdef HAVE_GETOPT_LONG " -h, --help\n" " --usage display this text\n" " -V, --version display version number\n" " -L, --license show license\n" " -f filename, --config=filename\n" " specify the configuration file to use\n" "\n" " -k, --kill stop running instance (send SIGTERM)\n" " -r, --reload reload running instance (send SIGHUP)\n" " -s signum, --signal=signum\n" " send the specified signal to the running process\n" " -c, --check check if the process is running\n" " (the above commands operate on dudki itself if no\n" " process name has been specified)\n" " -e, --ensure ensure that dudki is running\n" " -t, --test test configuration file and exit" #else /* !HAVE_GETOPT_LONG */ " -h display this text\n" " -V display version number\n" " -L show license\n" " -f filename specify the configuration file to use\n" "\n" " -k stop running instance (send SIGTERM)\n" " -r reload running instance (send SIGHUP)\n" " -s signum send the specified signal to the running process\n" " -c check if the process is running\n" " (the above commands operate on dudki itself if no\n" " process name has been specified)\n" " -e ensure that dudki is running\n" " -t test configuration file and exit" #endif /* /HAVE_GETOPT_LONG */ << endl; exit(0); break; case 'V': cerr << VERSION << endl; exit(0); break; case 'L': extern const char *COPYING; cerr << COPYING << endl; exit(0); break; case 'f': config_file = optarg; break; case 'k': if(op!=op_default) { cerr << "Can't obey two or more orders at once" << endl; exit(1); } op = op_signal; op_signum = SIGTERM; break; case 'r': if(op!=op_default) { cerr << "Can't obey two or more orders at once" << endl; exit(1); } op = op_signal; op_signum = SIGHUP; break; case 'c': if(op!=op_default) { cerr << "Can't obey two or more orders at once" << endl; exit(1); } op = op_signal; op_signum = 0; break; case 'e': if(op!=op_default) { cerr << "Can't obey two or more orders at once" << endl; exit(1); } op = op_ensure; break; case 't': if(op!=op_default) { cerr << "Can't obey two or more orders at once" << endl; exit(1); } op = op_test; break; case 's': if(op!=op_default) { cerr << "Can't obey two or more orders at once" << endl; exit(1); } op = op_signal; errno = 0; op_signum = strtol(optarg,NULL,0); if(errno) { cerr << "Can't obtain the signal value" << endl; exit(1); } break; default: cerr << "Huh??" << endl; exit(1); break; } } const char *sid = *argv; const char *t; while(t = index(sid,'/')) { sid = t; sid++; } openlog(sid,LOG_CONS|LOG_PERROR|LOG_PID,LOG_DAEMON); configuration config; config.parse(config_file); switch(op) { case op_test: cerr << "Configuration OK" << endl; break; case op_signal: try { if(optind>=argc) { signal_self(config,op_signum); }else{ int failures = 0; for(int narg=optind;narg<argc;narg++) { try { processes_t::const_iterator i = config.processes.find(argv[narg]); if(i==config.processes.end()) throw runtime_error("no such process configured"); - i->second.signal(op_signum); + if(op_signum) + i->second.signal(op_signum); + else + i->second.check(); }catch(exception& e) { cerr << "dudki(" << argv[narg] << "): " << e.what() << endl; failures++; } } if(failures) throw runtime_error("not all processes have been successfully signaled"); } if(!op_signum) exit(0); }catch(exception& e) { if(!op_signum) exit(1); } case op_ensure: try { signal_self(config,0); break; }catch(exception& e) { syslog(LOG_NOTICE,"The dudki process is down, taking its place"); config.daemonize = true; }catch(int zero) { // we throw zero in case we're ensuring that this very process is running. // we don't have to daemonize if we're daemonic. config.daemonize = false; } case op_default: case op_work: { if(config.daemonize) { pid_t pf = fork(); if(pf<0) throw runtime_error(string(__PRETTY_FUNCTION__)+": failed to fork()"); if(pf) { _exit(0); } } pid_file pidfile; pidfile.set(config.pidfile); signal(SIGINT,lethal_signal_handler); signal(SIGABRT,lethal_signal_handler); signal(SIGTERM,lethal_signal_handler); signal(SIGHUP,sighup_handler); sigset_t sset; sigemptyset(&sset); sigaddset(&sset,SIGINT); sigaddset(&sset,SIGABRT); sigaddset(&sset,SIGTERM); sigaddset(&sset,SIGHUP); sigprocmask(SIG_UNBLOCK,&sset,NULL); while(!finishing) { check_herd(config); sleep(config.check_interval); } if(restarting) execvp(_argv[0],_argv); } break; default: throw runtime_error(string(__PRETTY_FUNCTION__)+": internal error"); } }catch(exception& e) { cerr << "Oops: " << e.what() << endl; return 1; } } diff --git a/src/process.cc b/src/process.cc index 1ffac9f..8a5b5d2 100644 --- a/src/process.cc +++ b/src/process.cc @@ -1,189 +1,285 @@ #include <stdio.h> #include <sys/types.h> #include <unistd.h> #include <signal.h> #include <pwd.h> #include <grp.h> +#include <dirent.h> #include <sys/wait.h> #include <syslog.h> #include <errno.h> #include <iostream> #include <fstream> +#include <sstream> #include <stdexcept> +#include <string> +#include <vector> using namespace std; #include "process.h" #include "configuration.h" +static multimap<string,pid_t> procpids; + +void process::check() const { + if(!pidfile.empty()) { + signal(0); + }else if(!process_name.empty()) { + if(procpids.empty()) + gather_proc_info(); + if(procpids.find(process_name)==procpids.end()) + throw runtime_error("no such process"); + } // XXX: or else? +} void process::check(const string& id,configuration& config) { try { - signal(0); + check(); patience = 0; }catch(exception& e) { if(patience>60) { // TODO: configurable patience = 0; }else{ if(patience<10) { // TODO: configurable syslog(LOG_NOTICE,"The process '%s' is down, trying to launch.",id.c_str()); do_notify(id,"Starting up", "The named process seems to be down. Dudki will try\n" "to revive it by running the specified command.\n", config); try { launch(id,config); }catch(exception& e) { syslog(LOG_ERR,"Error trying to launch process '%s': %s",id.c_str(),e.what()); } }else if(patience==10){ // TODO: configurable like the above syslog(LOG_NOTICE,"Giving up on process '%s' for a while",id.c_str()); do_notify(id,"Giving up", "After a number of attempts to relaunch the named process\n" "It still seems to be down. Dudki is giving up attempts\n" "to revive the process for a while.\n", config); } patience++; } } } void process::launch(const string& id,configuration& config) { uid_t uid = (uid_t)-1; gid_t gid = (gid_t)-1; if(!user.empty()) { struct passwd *ptmp = getpwnam(user.c_str()); if(ptmp) { uid = ptmp->pw_uid; gid = ptmp->pw_gid; }else{ errno=0; uid = strtol(user.c_str(),NULL,0); if(errno) throw runtime_error("Failed to resolve User value to uid"); } } if(!group.empty()) { struct group *gtmp = getgrnam(group.c_str()); if(gtmp) { gid = gtmp->gr_gid; }else{ errno = 0; gid = strtol(group.c_str(),NULL,0); if(errno) throw runtime_error("Failed to reslove Group value to gid"); } } pid_t p = fork(); if(p<0) throw runtime_error(string(__PRETTY_FUNCTION__)+": failed to fork()"); if(!p) { // child try { setsid(); if(user.empty()) { if((getgid()!=gid) && setgid(gid)) throw runtime_error(string(__PRETTY_FUNCTION__)+": failed to setgid()"); }else{ if(initgroups(user.c_str(),gid)) throw runtime_error(string(__PRETTY_FUNCTION__)+": failed to initgroups()"); } if(!chroot.empty()) { if(::chroot(chroot.c_str())) throw runtime_error(string(__PRETTY_FUNCTION__)+": failed to chroot()"); } if(!user.empty()) { if((getuid()!=uid) && setuid(uid)) throw runtime_error(string(__PRETTY_FUNCTION__)+": failed to setuid()"); } char *argv[] = { "/bin/sh", "-c", (char*)restart_cmd.c_str(), NULL }; close(0); close(1); close(2); execv("/bin/sh",argv); }catch(exception& e) { syslog(LOG_ERR,"Error trying to launch process '%s': %s",id.c_str(),e.what()); } _exit(-1); } // parent int rv; if(waitpid(p,&rv,0)<0) throw runtime_error(string(__PRETTY_FUNCTION__)+": failed to waitpid()"); } void process::do_notify(const string& id,const string& event,const string& description,configuration& config) { string the_notify; if(!notify.empty()) the_notify=notify; else if(!config.notify.empty()) the_notify=config.notify; else return; try { string::size_type colon = the_notify.find(':'); if(colon==string::npos) throw runtime_error("invalid notify action specification"); string nschema = the_notify.substr(0,colon); string ntarget = the_notify.substr(colon+1); if(nschema=="mailto") { notify_mailto(ntarget,id,event,description,config); }else throw runtime_error("unrecognized notification schema"); }catch(exception& e) { syslog(LOG_ERR,"Notification error: %s",e.what()); } } void process::notify_mailto(const string& email,const string& id,const string& event,const string& description,configuration& config) { int files[2]; if(pipe(files)) throw runtime_error("Failed to pipe()"); pid_t pid = vfork(); if(pid==-1) { close(files[0]); close(files[1]); throw runtime_error("Failed to vfork()"); } if(!pid) { // child if(dup2(files[0],0)!=0) _exit(-1); close(1); close(files[0]); close(files[1]); execl("/usr/sbin/sendmail","usr/sbin/sendmail","-i",email.c_str(),NULL); _exit(-1); } // parent close(files[0]); FILE *mta = fdopen(files[1],"w"); for(headers_t::const_iterator i=mailto_headers.begin();i!=mailto_headers.end();++i) { fprintf(mta,"%s: %s\n",i->first.c_str(),i->second.c_str()); } for(headers_t::const_iterator i=config.mailto_headers.begin();i!=config.mailto_headers.end();++i) { if(mailto_headers.find(i->first)!=mailto_headers.end()) continue; fprintf(mta,"%s: %s\n",i->first.c_str(),i->second.c_str()); } fprintf(mta, "Subject: [%s] %s\n\n" "%s\n" "---\n" "This message was sent automatically by the 'dudki' daemon\n", id.c_str(), event.c_str(), description.c_str() ); fclose(mta); int status; waitpid(pid,&status,0); // TODO: check the return code } void process::signal(int signum) const { - ifstream pids(pidfile.c_str(),ios::in); - if(!pids) - throw runtime_error("no pidfile found"); - pid_t pid = 0; - pids >> pid; - pids.close(); - if(!pid) - throw runtime_error("no pid in pidfile"); - if(kill(pid,signum)) - throw runtime_error("failed to signal process"); + if(!pidfile.empty()) { + ifstream pids(pidfile.c_str(),ios::in); + if(!pids) + throw runtime_error("no pidfile found"); + pid_t pid = 0; + pids >> pid; + pids.close(); + if(!pid) + throw runtime_error("no pid in pidfile"); + if(kill(pid,signum)) + throw runtime_error("failed to signal process"); + }else if(!process_name.empty()) { + if(procpids.empty()) + gather_proc_info(); + pair<multimap<string,pid_t>::const_iterator,multimap<string,pid_t>::const_iterator> range = procpids.equal_range(process_name); + int count = 0; + for(multimap<string,pid_t>::const_iterator i=range.first;i!=range.second;++i) { + pid_t pid = i->second; + if(kill(i->second,signum)) + throw runtime_error("failed to signal process"); + count++; + } + if(!count) + throw runtime_error("no running instance detected"); + }else + throw runtime_error("nothing is known about the process"); +} + +void process::prepare_herd() { + procpids.clear(); +} +void process::unprepare_herd() { + procpids.clear(); +} +void process::gather_proc_info() { + vector<pid_t> allpids; + DIR *pd = opendir("/proc"); + if(!pd) + throw runtime_error("failed to open /proc"); + struct dirent *pde; + pid_t selfpid = getpid(); + while(pde=readdir(pd)) { + errno=0; + pid_t pid = atoi(pde->d_name); + if((!pid) || pid==selfpid) + continue; + allpids.push_back(pid); + } + closedir(pd); + char s[256]; + procpids.clear(); + for(vector<pid_t>::const_iterator i=allpids.begin();i!=allpids.end();++i) { + int r = snprintf(s,sizeof(s),"/proc/%d/stat",*i); + if(r>=sizeof(s) || r<1) + continue; + string cmd; + ifstream ss(s,ios::in); + if(!ss) + continue; + getline(ss,cmd); + string::size_type op = cmd.find('('); + if(op==string::npos) + continue; + cmd.erase(0,op+1); + string::size_type cp = cmd.find(')'); + if(cp==string::npos) + continue; + cmd.erase(cp); + r = snprintf(s,sizeof(s),"/proc/%d/cmdline",*i); + if(r>=sizeof(s) || r<1) + continue; + ifstream cs(s,ios::binary); + if(!cs) + continue; + string command; + while(cs) { + string cl; + getline(cs,cl,(char)0); + string::size_type lsl = cl.rfind('/'); + if(lsl!=string::npos) + cl.erase(0,lsl+1); + if(cl.substr(0,cmd.length())==cmd) { + command = cl; + break; + } + } + procpids.insert(pair<string,pid_t>(cmd,*i)); + if((!command.empty()) && cmd!=command) + procpids.insert(pair<string,pid_t>(command,*i)); + } } diff --git a/src/process.h b/src/process.h index 27ee049..90b12d9 100644 --- a/src/process.h +++ b/src/process.h @@ -1,38 +1,44 @@ #ifndef __PROCESS_H #define __PROCESS_H #include <string> #include <map> using namespace std; class configuration; typedef map<string,string> headers_t; class process { public: string pidfile; + string process_name; string restart_cmd; string notify; string user; string group; string chroot; headers_t mailto_headers; int patience; process() : patience(0) { } void check(const string& id,configuration& config); void launch(const string& id,configuration& config); void do_notify(const string& id,const string& event,const string& description,configuration& config); void notify_mailto(const string& email,const string& id,const string& event, const string& description,configuration& config); void signal(int signum) const; + void check() const; + + static void prepare_herd(); + static void gather_proc_info(); + static void unprepare_herd(); }; typedef map<string,process> processes_t; #endif /* __PROCESS_H */ |