]>
Commit | Line | Data |
---|---|---|
390802ab DM |
1 | package PVE::Daemon; |
2 | ||
3 | # Abstract class to implement Daemons | |
4 | # | |
5 | # Features: | |
6 | # * lock and write PID file /var/run/$name.pid to make sure onyl | |
7 | # one instance is running. | |
8 | # * correctly daemonize (redirect STDIN/STDOUT) | |
4fc69176 DM |
9 | # * restart by stop/start, exec, or signal HUP |
10 | # * daemon restart on error (option 'restart_on_error') | |
b2132054 | 11 | # * handle worker processes (option 'max_workers') |
390802ab DM |
12 | |
13 | use strict; | |
14 | use warnings; | |
15 | use PVE::SafeSyslog; | |
16 | use PVE::INotify; | |
17 | ||
18 | use POSIX ":sys_wait_h"; | |
19 | use Fcntl ':flock'; | |
20 | use Getopt::Long; | |
21 | use Time::HiRes qw (gettimeofday); | |
22 | ||
23 | use base qw(PVE::CLIHandler); | |
24 | ||
25 | $SIG{'__WARN__'} = sub { | |
26 | my $err = $@; | |
27 | my $t = $_[0]; | |
28 | chomp $t; | |
29 | print "$t\n"; | |
30 | syslog('warning', "WARNING: %s", $t); | |
31 | $@ = $err; | |
32 | }; | |
33 | ||
34 | $ENV{'PATH'} = '/sbin:/bin:/usr/sbin:/usr/bin'; | |
35 | ||
36 | my $daemon_initialized = 0; # we only allow one instance | |
37 | ||
3206cdf6 DM |
38 | my $close_daemon_lock = sub { |
39 | my ($self) = @_; | |
40 | ||
41 | close $self->{daemon_lock_fh}; | |
42 | delete $self->{daemon_lock_fh}; | |
43 | }; | |
44 | ||
390802ab DM |
45 | my $lockpidfile = sub { |
46 | my ($self) = @_; | |
47 | ||
48 | my $lkfn = $self->{pidfile} . ".lock"; | |
49 | ||
3206cdf6 DM |
50 | $self->{daemon_lock_fh} = IO::File->new(">>$lkfn"); |
51 | if (!$self->{daemon_lock_fh}) { | |
390802ab DM |
52 | my $msg = "can't aquire lock on file '$lkfn' - $!"; |
53 | syslog ('err', $msg); | |
54 | die "ERROR: $msg\n"; | |
55 | } | |
56 | ||
3206cdf6 DM |
57 | for (my $i = 0; $i < 5; $i ++) { |
58 | return if flock ($self->{daemon_lock_fh}, LOCK_EX|LOCK_NB); | |
59 | sleep(1); | |
60 | } | |
61 | ||
62 | if (!flock ($self->{daemon_lock_fh}, LOCK_EX|LOCK_NB)) { | |
63 | &$close_daemon_lock($self); | |
390802ab DM |
64 | my $msg = "can't aquire lock '$lkfn' - $!"; |
65 | syslog ('err', $msg); | |
66 | die "ERROR: $msg\n"; | |
67 | } | |
68 | }; | |
69 | ||
70 | my $writepidfile = sub { | |
71 | my ($self) = @_; | |
72 | ||
73 | my $pidfile = $self->{pidfile}; | |
74 | ||
75 | if (!open (PIDFH, ">$pidfile")) { | |
76 | my $msg = "can't open pid file '$pidfile' - $!"; | |
77 | syslog ('err', $msg); | |
78 | die "ERROR: $msg\n"; | |
79 | } | |
80 | print PIDFH "$$\n"; | |
81 | close (PIDFH); | |
82 | }; | |
83 | ||
84 | my $server_cleanup = sub { | |
85 | my ($self) = @_; | |
86 | ||
87 | unlink $self->{pidfile} . ".lock"; | |
88 | unlink $self->{pidfile}; | |
89 | }; | |
90 | ||
b2132054 DM |
91 | my $finish_workers = sub { |
92 | my ($self) = @_; | |
93 | ||
94 | foreach my $cpid (keys %{$self->{workers}}) { | |
95 | my $waitpid = waitpid($cpid, WNOHANG); | |
96 | if (defined($waitpid) && ($waitpid == $cpid)) { | |
97 | delete ($self->{workers}->{$cpid}); | |
98 | syslog('info', "worker $cpid finished"); | |
99 | } | |
100 | } | |
101 | }; | |
102 | ||
103 | my $start_workers = sub { | |
104 | my ($self) = @_; | |
105 | ||
106 | return if $self->{terminate}; | |
107 | ||
108 | my $count = 0; | |
109 | foreach my $cpid (keys %{$self->{workers}}) { | |
110 | $count++; | |
111 | } | |
112 | ||
113 | my $need = $self->{max_workers} - $count; | |
114 | ||
115 | return if $need <= 0; | |
116 | ||
117 | syslog('info', "starting $need worker(s)"); | |
118 | ||
119 | while ($need > 0) { | |
120 | my $pid = fork; | |
121 | ||
122 | if (!defined ($pid)) { | |
123 | syslog('err', "can't fork worker"); | |
124 | sleep (1); | |
125 | } elsif ($pid) { # parent | |
126 | $self->{workers}->{$pid} = 1; | |
127 | syslog('info', "worker $pid started"); | |
128 | $need--; | |
129 | } else { | |
130 | $0 = "$self->{name} worker"; | |
131 | ||
3206cdf6 | 132 | &$close_daemon_lock($self); |
b2132054 DM |
133 | |
134 | PVE::INotify::inotify_close(); | |
135 | ||
136 | for my $sig (qw(CHLD HUP INT TERM QUIT)) { | |
137 | $SIG{$sig} = 'DEFAULT'; # restore default handler | |
138 | # AnyEvent signals only works if $SIG{XX} is | |
139 | # undefined (perl event loop) | |
140 | delete $SIG{$sig}; # so that we can handle events with AnyEvent | |
141 | } | |
142 | ||
143 | eval { $self->run(); }; | |
144 | if (my $err = $@) { | |
145 | syslog('err', $err); | |
146 | sleep(5); # avoid fast restarts | |
147 | } | |
148 | ||
149 | syslog('info', "worker exit"); | |
150 | exit (0); | |
151 | } | |
152 | } | |
153 | }; | |
154 | ||
155 | my $terminate_server = sub { | |
156 | my ($self) = @_; | |
157 | ||
158 | $self->{terminate} = 1; # set flag to avoid worker restart | |
159 | ||
160 | if (!$self->{max_workers}) { | |
161 | eval { $self->shutdown(); }; | |
162 | warn $@ if $@; | |
163 | return; | |
164 | } | |
165 | ||
166 | eval { $self->shutdown(); }; | |
167 | warn $@ if $@; | |
168 | ||
169 | # we have workers - terminate them | |
170 | ||
171 | foreach my $cpid (keys %{$self->{workers}}) { | |
172 | kill(15, $cpid); # TERM childs | |
173 | } | |
174 | ||
175 | # nicely shutdown childs (give them max 10 seconds to shut down) | |
176 | my $previous_alarm = alarm(10); | |
177 | eval { | |
178 | local $SIG{ALRM} = sub { die "timeout\n" }; | |
179 | ||
180 | while ((my $pid = waitpid (-1, 0)) > 0) { | |
181 | if (defined($self->{workers}->{$pid})) { | |
182 | delete($self->{workers}->{$pid}); | |
183 | syslog('info', "worker $pid finished"); | |
184 | } | |
185 | } | |
186 | alarm(0); # avoid race condition | |
187 | }; | |
188 | my $err = $@; | |
189 | ||
190 | alarm ($previous_alarm); | |
191 | ||
192 | if ($err) { | |
193 | syslog('err', "error stopping workers (will kill them now) - $err"); | |
194 | foreach my $cpid (keys %{$self->{workers}}) { | |
195 | # KILL childs still alive! | |
196 | if (kill (0, $cpid)) { | |
197 | delete($self->{workers}->{$cpid}); | |
198 | syslog("err", "kill worker $cpid"); | |
199 | kill(9, $cpid); | |
200 | # fixme: waitpid? | |
201 | } | |
202 | } | |
203 | } | |
204 | }; | |
205 | ||
390802ab DM |
206 | my $server_run = sub { |
207 | my ($self, $debug) = @_; | |
208 | ||
209 | &$lockpidfile($self); | |
210 | ||
211 | # run in background | |
212 | my $spid; | |
213 | ||
214 | my $restart = $ENV{RESTART_PVE_DAEMON}; | |
215 | ||
216 | delete $ENV{RESTART_PVE_DAEMON}; | |
217 | ||
218 | $self->{debug} = 1 if $debug; | |
219 | ||
220 | $self->init(); | |
221 | ||
222 | if (!$debug) { | |
223 | open STDIN, '</dev/null' || die "can't read /dev/null"; | |
224 | open STDOUT, '>/dev/null' || die "can't write /dev/null"; | |
225 | } | |
226 | ||
227 | if (!$restart && !$debug) { | |
228 | PVE::INotify::inotify_close(); | |
229 | $spid = fork(); | |
230 | if (!defined ($spid)) { | |
231 | my $msg = "can't put server into background - fork failed"; | |
232 | syslog('err', $msg); | |
233 | die "ERROR: $msg\n"; | |
234 | } elsif ($spid) { # parent | |
235 | exit (0); | |
236 | } | |
237 | PVE::INotify::inotify_init(); | |
238 | } | |
239 | ||
240 | &$writepidfile($self); | |
241 | ||
891b9097 DM |
242 | POSIX::setsid(); |
243 | ||
390802ab DM |
244 | if ($restart) { |
245 | syslog('info' , "restarting server"); | |
246 | } else { | |
247 | syslog('info' , "starting server"); | |
248 | } | |
249 | ||
250 | open STDERR, '>&STDOUT' || die "can't close STDERR\n"; | |
251 | ||
b2132054 DM |
252 | my $old_sig_term = $SIG{TERM}; |
253 | local $SIG{TERM} = sub { | |
254 | local ($@, $!, $?); # do not overwrite error vars | |
255 | syslog('info', "received signal TERM"); | |
256 | &$terminate_server($self); | |
257 | &$server_cleanup($self); | |
258 | &$old_sig_term(@_) if $old_sig_term; | |
259 | }; | |
390802ab | 260 | |
b2132054 DM |
261 | my $old_sig_quit = $SIG{QUIT}; |
262 | local $SIG{QUIT} = sub { | |
263 | local ($@, $!, $?); # do not overwrite error vars | |
264 | syslog('info', "received signal QUIT"); | |
265 | &$terminate_server($self); | |
266 | &$server_cleanup($self); | |
267 | &$old_sig_quit(@_) if $old_sig_quit; | |
268 | }; | |
390802ab | 269 | |
b2132054 DM |
270 | my $old_sig_int = $SIG{INT}; |
271 | local $SIG{INT} = sub { | |
272 | local ($@, $!, $?); # do not overwrite error vars | |
273 | syslog('info', "received signal INT"); | |
274 | $SIG{INT} = 'DEFAULT'; # allow to terminate now | |
275 | &$terminate_server($self); | |
390802ab | 276 | &$server_cleanup($self); |
b2132054 | 277 | &$old_sig_int(@_) if $old_sig_int; |
390802ab DM |
278 | }; |
279 | ||
b2132054 DM |
280 | $SIG{HUP} = sub { |
281 | local ($@, $!, $?); # do not overwrite error vars | |
282 | syslog('info', "received signal HUP"); | |
283 | if ($self->{max_workers}) { | |
284 | &$terminate_server($self); | |
285 | $self->{got_hup_signal} = 1; | |
286 | } elsif ($self->can('hup')) { | |
bdb5acce DM |
287 | eval { $self->hup() }; |
288 | warn $@ if $@; | |
b2132054 DM |
289 | } |
290 | }; | |
291 | ||
292 | eval { | |
293 | if ($self->{max_workers}) { | |
294 | my $old_sig_chld = $SIG{CHLD}; | |
295 | local $SIG{CHLD} = sub { | |
296 | local ($@, $!, $?); # do not overwrite error vars | |
297 | &$finish_workers($self); | |
298 | &$old_sig_chld(@_) if $old_sig_chld; | |
299 | }; | |
300 | ||
301 | for (;;) { # forever | |
302 | &$start_workers($self); | |
303 | sleep(5); | |
304 | &$finish_workers($self); | |
305 | last if $self->{terminate}; | |
306 | } | |
bdb5acce | 307 | |
b2132054 DM |
308 | } else { |
309 | $self->run(); | |
310 | } | |
311 | }; | |
390802ab DM |
312 | my $err = $@; |
313 | ||
314 | if ($err) { | |
315 | syslog ('err', "ERROR: $err"); | |
b2132054 DM |
316 | |
317 | # fixme: kill all workers | |
318 | ||
390802ab DM |
319 | if (my $wait_time = $self->{restart_on_error}) { |
320 | $self->restart_daemon($wait_time); | |
321 | } else { | |
322 | $self->exit_daemon(-1); | |
323 | } | |
324 | } | |
325 | ||
b2132054 DM |
326 | if ($self->{got_hup_signal}) { |
327 | $self->restart_daemon(); | |
328 | } else { | |
329 | $self->exit_daemon(0); | |
330 | } | |
390802ab DM |
331 | }; |
332 | ||
333 | sub new { | |
334 | my ($this, $name, $cmdline, %params) = @_; | |
335 | ||
b2132054 | 336 | die "please run as root\n" if !$ENV{RESTART_PVE_DAEMON} && ($> != 0); |
390802ab DM |
337 | |
338 | die "missing name" if !$name; | |
339 | ||
340 | die "can't create more that one PVE::Daemon" if $daemon_initialized; | |
341 | $daemon_initialized = 1; | |
342 | ||
343 | PVE::INotify::inotify_init(); | |
344 | ||
345 | initlog($name); | |
346 | ||
347 | my $class = ref($this) || $this; | |
348 | ||
bdb5acce DM |
349 | my $self = bless { |
350 | name => $name, | |
351 | run_dir => '/var/run', | |
b2132054 | 352 | workers => {}, |
bdb5acce | 353 | }, $class; |
390802ab DM |
354 | |
355 | foreach my $opt (keys %params) { | |
356 | my $value = $params{$opt}; | |
357 | if ($opt eq 'restart_on_error') { | |
358 | $self->{$opt} = $value; | |
359 | } elsif ($opt eq 'stop_wait_time') { | |
360 | $self->{$opt} = $value; | |
bdb5acce DM |
361 | } elsif ($opt eq 'run_dir') { |
362 | $self->{$opt} = $value; | |
b2132054 DM |
363 | } elsif ($opt eq 'max_workers') { |
364 | $self->{$opt} = $value; | |
390802ab DM |
365 | } else { |
366 | die "unknown option '$opt'"; | |
367 | } | |
368 | } | |
369 | ||
bdb5acce DM |
370 | $self->{pidfile} = "$self->{run_dir}/${name}.pid"; |
371 | ||
372 | $self->{nodename} = PVE::INotify::nodename(); | |
373 | ||
b2132054 | 374 | $self->{cmdline} = []; |
bdb5acce | 375 | |
b2132054 DM |
376 | foreach my $el (@$cmdline) { |
377 | $el =~ m/^(.*)$/; # untaint | |
378 | push @{$self->{cmdline}}, $1; | |
379 | } | |
bdb5acce | 380 | |
b2132054 | 381 | $0 = $name; |
bdb5acce | 382 | |
390802ab DM |
383 | return $self; |
384 | } | |
385 | ||
386 | sub exit_daemon { | |
387 | my ($self, $status) = @_; | |
388 | ||
389 | syslog("info", "server stopped"); | |
390 | ||
391 | &$server_cleanup($self); | |
392 | ||
393 | exit($status); | |
394 | } | |
395 | ||
396 | sub restart_daemon { | |
397 | my ($self, $waittime) = @_; | |
398 | ||
399 | syslog('info', "server shutdown (restart)"); | |
400 | ||
401 | $ENV{RESTART_PVE_DAEMON} = 1; | |
402 | ||
403 | sleep($waittime) if $waittime; # avoid high server load due to restarts | |
404 | ||
405 | PVE::INotify::inotify_close(); | |
406 | ||
407 | exec (@{$self->{cmdline}}); | |
408 | ||
409 | exit (-1); # never reached? | |
410 | } | |
411 | ||
412 | # please overwrite in subclass | |
413 | # this is called at startup - before forking | |
414 | sub init { | |
415 | my ($self) = @_; | |
416 | ||
417 | } | |
418 | ||
419 | # please overwrite in subclass | |
420 | sub shutdown { | |
421 | my ($self) = @_; | |
422 | ||
423 | syslog('info' , "server closing"); | |
424 | ||
b2132054 DM |
425 | if (!$self->{max_workers}) { |
426 | # wait for children | |
427 | 1 while (waitpid(-1, POSIX::WNOHANG()) > 0); | |
428 | } | |
390802ab DM |
429 | } |
430 | ||
bdb5acce DM |
431 | # please define in subclass |
432 | #sub hup { | |
433 | # my ($self) = @_; | |
434 | # | |
435 | # syslog('info' , "received signal HUP (restart)"); | |
436 | #} | |
390802ab DM |
437 | |
438 | # please overwrite in subclass | |
439 | sub run { | |
440 | my ($self) = @_; | |
441 | ||
442 | for (;;) { # forever | |
443 | syslog('info' , "server is running"); | |
444 | sleep(5); | |
445 | } | |
446 | } | |
447 | ||
448 | sub start { | |
449 | my ($self, $debug) = @_; | |
450 | ||
451 | &$server_run($self, $debug); | |
452 | } | |
453 | ||
bdb5acce DM |
454 | my $read_pid = sub { |
455 | my ($self) = @_; | |
456 | ||
457 | my $pid_str = PVE::Tools::file_read_firstline($self->{pidfile}); | |
458 | ||
459 | return 0 if !$pid_str; | |
460 | ||
461 | return 0 if $pid_str !~ m/^(\d+)$/; # untaint | |
462 | ||
463 | my $pid = int($1); | |
464 | ||
465 | return $pid; | |
466 | }; | |
467 | ||
390802ab DM |
468 | sub running { |
469 | my ($self) = @_; | |
470 | ||
bdb5acce | 471 | my $pid = &$read_pid($self); |
390802ab DM |
472 | |
473 | if ($pid) { | |
474 | my $res = PVE::ProcFSTools::check_process_running($pid) ? 1 : 0; | |
475 | return wantarray ? ($res, $pid) : $res; | |
476 | } | |
477 | ||
478 | return wantarray ? (0, 0) : 0; | |
479 | } | |
480 | ||
481 | sub stop { | |
482 | my ($self) = @_; | |
483 | ||
bdb5acce DM |
484 | my $pid = &$read_pid($self); |
485 | ||
390802ab DM |
486 | return if !$pid; |
487 | ||
488 | if (PVE::ProcFSTools::check_process_running($pid)) { | |
489 | kill(15, $pid); # send TERM signal | |
490 | # give some time | |
491 | my $wait_time = $self->{stop_wait_time} || 5; | |
492 | my $running = 1; | |
493 | for (my $i = 0; $i < $wait_time; $i++) { | |
494 | $running = PVE::ProcFSTools::check_process_running($pid); | |
495 | last if !$running; | |
496 | sleep (1); | |
497 | } | |
498 | ||
499 | syslog('err', "server still running - send KILL") if $running; | |
500 | ||
501 | # to be sure | |
502 | kill(9, $pid); | |
503 | waitpid($pid, 0); | |
504 | } | |
505 | ||
506 | if (-f $self->{pidfile}) { | |
507 | # try to get the lock | |
508 | &$lockpidfile($self); | |
509 | &$server_cleanup($self); | |
510 | } | |
511 | } | |
512 | ||
513 | sub register_start_command { | |
514 | my ($self, $class, $description) = @_; | |
515 | ||
516 | $class->register_method({ | |
517 | name => 'start', | |
518 | path => 'start', | |
519 | method => 'POST', | |
520 | description => $description || "Start the daemon.", | |
521 | parameters => { | |
522 | additionalProperties => 0, | |
523 | properties => { | |
524 | debug => { | |
525 | description => "Debug mode - stay in foreground", | |
526 | type => "boolean", | |
527 | optional => 1, | |
528 | default => 0, | |
529 | }, | |
530 | }, | |
531 | }, | |
532 | returns => { type => 'null' }, | |
533 | ||
534 | code => sub { | |
535 | my ($param) = @_; | |
536 | ||
537 | $self->start($param->{debug}); | |
538 | ||
539 | return undef; | |
540 | }}); | |
541 | } | |
542 | ||
bdb5acce DM |
543 | my $reload_daemon = sub { |
544 | my ($self, $use_hup) = @_; | |
545 | ||
546 | if (my $restart = $ENV{RESTART_PVE_DAEMON}) { | |
547 | $self->start(); | |
548 | } else { | |
549 | my ($running, $pid) = $self->running(); | |
550 | if (!$running) { | |
551 | $self->start(); | |
552 | } else { | |
553 | if ($use_hup) { | |
b2132054 DM |
554 | syslog('info', "send HUP to $pid"); |
555 | kill 1, $pid; | |
bdb5acce DM |
556 | } else { |
557 | $self->stop(); | |
558 | $self->start(); | |
559 | } | |
560 | } | |
561 | } | |
562 | }; | |
563 | ||
390802ab | 564 | sub register_restart_command { |
bdb5acce | 565 | my ($self, $class, $use_hup, $description) = @_; |
390802ab DM |
566 | |
567 | $class->register_method({ | |
568 | name => 'restart', | |
569 | path => 'restart', | |
570 | method => 'POST', | |
571 | description => $description || "Restart the daemon (or start if not running).", | |
572 | parameters => { | |
573 | additionalProperties => 0, | |
574 | properties => {}, | |
575 | }, | |
576 | returns => { type => 'null' }, | |
577 | ||
578 | code => sub { | |
579 | my ($param) = @_; | |
580 | ||
bdb5acce DM |
581 | &$reload_daemon($self, $use_hup); |
582 | ||
583 | return undef; | |
584 | }}); | |
585 | } | |
586 | ||
587 | sub register_reload_command { | |
588 | my ($self, $class, $description) = @_; | |
589 | ||
590 | $class->register_method({ | |
591 | name => 'reload', | |
592 | path => 'reload', | |
593 | method => 'POST', | |
594 | description => $description || "Reload daemon configuration (or start if not running).", | |
595 | parameters => { | |
596 | additionalProperties => 0, | |
597 | properties => {}, | |
598 | }, | |
599 | returns => { type => 'null' }, | |
600 | ||
601 | code => sub { | |
602 | my ($param) = @_; | |
603 | ||
604 | &$reload_daemon($self, 1); | |
390802ab DM |
605 | |
606 | return undef; | |
607 | }}); | |
608 | } | |
609 | ||
610 | sub register_stop_command { | |
611 | my ($self, $class, $description) = @_; | |
612 | ||
613 | $class->register_method({ | |
614 | name => 'stop', | |
615 | path => 'stop', | |
616 | method => 'POST', | |
617 | description => $description || "Stop the daemon.", | |
618 | parameters => { | |
619 | additionalProperties => 0, | |
620 | properties => {}, | |
621 | }, | |
622 | returns => { type => 'null' }, | |
623 | ||
624 | code => sub { | |
625 | my ($param) = @_; | |
626 | ||
627 | $self->stop(); | |
628 | ||
629 | return undef; | |
630 | }}); | |
631 | } | |
632 | ||
633 | sub register_status_command { | |
634 | my ($self, $class, $description) = @_; | |
635 | ||
636 | $class->register_method({ | |
637 | name => 'status', | |
638 | path => 'status', | |
639 | method => 'GET', | |
640 | description => "Get daemon status.", | |
641 | parameters => { | |
642 | additionalProperties => 0, | |
643 | properties => {}, | |
644 | }, | |
645 | returns => { | |
646 | type => 'string', | |
647 | enum => ['stopped', 'running'], | |
648 | }, | |
649 | code => sub { | |
650 | my ($param) = @_; | |
651 | ||
652 | return $self->running() ? 'running' : 'stopped'; | |
653 | }}); | |
654 | } | |
655 | ||
656 | 1; | |
657 |