# * lock and write PID file /var/run/$name.pid to make sure onyl
# one instance is running.
# * correctly daemonize (redirect STDIN/STDOUT)
-# * daemon restart (option 'restart_on_error')
+# * restart by stop/start, exec, or signal HUP
+# * daemon restart on error (option 'restart_on_error')
+# * handle worker processes (option 'max_workers')
use strict;
use warnings;
my $daemon_initialized = 0; # we only allow one instance
+my $close_daemon_lock = sub {
+ my ($self) = @_;
+
+ return if !$self->{daemon_lock_fh};
+
+ close $self->{daemon_lock_fh};
+ delete $self->{daemon_lock_fh};
+};
+
+# call this if you fork() from child
+# Note: we already call this for workers, so it is only required
+# if you fork inside a simple daemon (max_workers == 0).
+sub after_fork_cleanup {
+ my ($self) = @_;
+
+ &$close_daemon_lock($self);
+
+ PVE::INotify::inotify_close();
+
+ for my $sig (qw(CHLD HUP INT TERM QUIT)) {
+ $SIG{$sig} = 'DEFAULT'; # restore default handler
+ # AnyEvent signals only works if $SIG{XX} is
+ # undefined (perl event loop)
+ delete $SIG{$sig}; # so that we can handle events with AnyEvent
+ }
+}
+
my $lockpidfile = sub {
my ($self) = @_;
my $lkfn = $self->{pidfile} . ".lock";
- if (!open (FLCK, ">>$lkfn")) {
- my $msg = "can't aquire lock on file '$lkfn' - $!";
- syslog ('err', $msg);
- die "ERROR: $msg\n";
+ if (my $fd = $self->{env_pve_lock_fd}) {
+
+ $self->{daemon_lock_fh} = IO::Handle->new_from_fd($fd, "a");
+
+ } else {
+
+ $self->{daemon_lock_fh} = IO::File->new(">>$lkfn");
}
- if (!flock (FLCK, LOCK_EX|LOCK_NB)) {
- close (FLCK);
- my $msg = "can't aquire lock '$lkfn' - $!";
- syslog ('err', $msg);
- die "ERROR: $msg\n";
+ if (!$self->{daemon_lock_fh}) {
+ die "can't open lock '$lkfn' - $!\n";
+ }
+
+ for (my $i = 0; $i < 5; $i ++) {
+ return if flock ($self->{daemon_lock_fh}, LOCK_EX|LOCK_NB);
+ sleep(1);
+ }
+
+ if (!flock ($self->{daemon_lock_fh}, LOCK_EX|LOCK_NB)) {
+ &$close_daemon_lock($self);
+ die "can't aquire lock '$lkfn' - $!\n";
}
};
unlink $self->{pidfile};
};
+my $finish_workers = sub {
+ my ($self) = @_;
+
+ foreach my $cpid (keys %{$self->{workers}}) {
+ my $waitpid = waitpid($cpid, WNOHANG);
+ if (defined($waitpid) && ($waitpid == $cpid)) {
+ delete ($self->{workers}->{$cpid});
+ syslog('info', "worker $cpid finished");
+ }
+ }
+};
+
+my $start_workers = sub {
+ my ($self) = @_;
+
+ return if $self->{terminate};
+
+ my $count = 0;
+ foreach my $cpid (keys %{$self->{workers}}) {
+ $count++;
+ }
+
+ my $need = $self->{max_workers} - $count;
+
+ return if $need <= 0;
+
+ syslog('info', "starting $need worker(s)");
+
+ while ($need > 0) {
+ my $pid = fork;
+
+ if (!defined ($pid)) {
+ syslog('err', "can't fork worker");
+ sleep (1);
+ } elsif ($pid) { # parent
+ $self->{workers}->{$pid} = 1;
+ syslog('info', "worker $pid started");
+ $need--;
+ } else {
+ $0 = "$self->{name} worker";
+
+ $self->after_fork_cleanup();
+
+ eval { $self->run(); };
+ if (my $err = $@) {
+ syslog('err', $err);
+ sleep(5); # avoid fast restarts
+ }
+
+ syslog('info', "worker exit");
+ exit (0);
+ }
+ }
+};
+
+my $terminate_server = sub {
+ my ($self) = @_;
+
+ $self->{terminate} = 1; # set flag to avoid worker restart
+
+ if (!$self->{max_workers}) {
+ eval { $self->shutdown(); };
+ warn $@ if $@;
+ return;
+ }
+
+ eval { $self->shutdown(); };
+ warn $@ if $@;
+
+ # we have workers - terminate them
+
+ foreach my $cpid (keys %{$self->{workers}}) {
+ kill(15, $cpid); # TERM childs
+ }
+
+ # nicely shutdown childs (give them max 10 seconds to shut down)
+ my $previous_alarm = alarm(10);
+ eval {
+ local $SIG{ALRM} = sub { die "timeout\n" };
+
+ while ((my $pid = waitpid (-1, 0)) > 0) {
+ if (defined($self->{workers}->{$pid})) {
+ delete($self->{workers}->{$pid});
+ syslog('info', "worker $pid finished");
+ }
+ }
+ alarm(0); # avoid race condition
+ };
+ my $err = $@;
+
+ alarm ($previous_alarm);
+
+ if ($err) {
+ syslog('err', "error stopping workers (will kill them now) - $err");
+ foreach my $cpid (keys %{$self->{workers}}) {
+ # KILL childs still alive!
+ if (kill (0, $cpid)) {
+ delete($self->{workers}->{$cpid});
+ syslog("err", "kill worker $cpid");
+ kill(9, $cpid);
+ # fixme: waitpid?
+ }
+ }
+ }
+};
+
my $server_run = sub {
my ($self, $debug) = @_;
+ # fixme: handle restart lockfd
&$lockpidfile($self);
- # run in background
- my $spid;
+ # remove FD_CLOEXEC bit to reuse on exec
+ $self->{daemon_lock_fh}->fcntl(Fcntl::F_SETFD(), 0);
- my $restart = $ENV{RESTART_PVE_DAEMON};
+ $ENV{PVE_DAEMON_LOCK_FD} = $self->{daemon_lock_fh}->fileno;
+
+# my $fd = POSIX::dup($self->{daemon_lock_fh}) ||
+# die "unable to duplicate daemon_lock_fh\n";
- delete $ENV{RESTART_PVE_DAEMON};
+
+ # run in background
+ my $spid;
$self->{debug} = 1 if $debug;
open STDOUT, '>/dev/null' || die "can't write /dev/null";
}
- if (!$restart && !$debug) {
+ if (!$self->{env_restart_pve_daemon} && !$debug) {
PVE::INotify::inotify_close();
$spid = fork();
if (!defined ($spid)) {
POSIX::setsid();
- if ($restart) {
+ if ($self->{env_restart_pve_daemon}) {
syslog('info' , "restarting server");
} else {
syslog('info' , "starting server");
open STDERR, '>&STDOUT' || die "can't close STDERR\n";
- $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = sub {
- $SIG{INT} = 'DEFAULT';
+ my $old_sig_term = $SIG{TERM};
+ local $SIG{TERM} = sub {
+ local ($@, $!, $?); # do not overwrite error vars
+ syslog('info', "received signal TERM");
+ &$terminate_server($self);
+ &$server_cleanup($self);
+ &$old_sig_term(@_) if $old_sig_term;
+ };
- eval { $self->shutdown(); };
- warn $@ if $@;
+ my $old_sig_quit = $SIG{QUIT};
+ local $SIG{QUIT} = sub {
+ local ($@, $!, $?); # do not overwrite error vars
+ syslog('info', "received signal QUIT");
+ &$terminate_server($self);
+ &$server_cleanup($self);
+ &$old_sig_quit(@_) if $old_sig_quit;
+ };
+ my $old_sig_int = $SIG{INT};
+ local $SIG{INT} = sub {
+ local ($@, $!, $?); # do not overwrite error vars
+ syslog('info', "received signal INT");
+ $SIG{INT} = 'DEFAULT'; # allow to terminate now
+ &$terminate_server($self);
&$server_cleanup($self);
+ &$old_sig_int(@_) if $old_sig_int;
};
- if ($self->can('hup')) {
- $SIG{HUP} = sub {
+ $SIG{HUP} = sub {
+ local ($@, $!, $?); # do not overwrite error vars
+ syslog('info', "received signal HUP");
+ if ($self->{max_workers}) {
+ &$terminate_server($self);
+ $self->{got_hup_signal} = 1;
+ } elsif ($self->can('hup')) {
eval { $self->hup() };
warn $@ if $@;
- };
- }
+ }
+ };
+
+ eval {
+ if ($self->{max_workers}) {
+ my $old_sig_chld = $SIG{CHLD};
+ local $SIG{CHLD} = sub {
+ local ($@, $!, $?); # do not overwrite error vars
+ &$finish_workers($self);
+ &$old_sig_chld(@_) if $old_sig_chld;
+ };
+
+ for (;;) { # forever
+ &$start_workers($self);
+ sleep(5);
+ &$finish_workers($self);
+ last if $self->{terminate};
+ }
- eval { $self->run() };
+ } else {
+ $self->run();
+ }
+ };
my $err = $@;
if ($err) {
syslog ('err', "ERROR: $err");
+
+ # fixme: kill all workers
+
if (my $wait_time = $self->{restart_on_error}) {
$self->restart_daemon($wait_time);
} else {
}
}
- $self->exit_daemon(0);
+ if ($self->{got_hup_signal}) {
+ $self->restart_daemon();
+ } else {
+ $self->exit_daemon(0);
+ }
};
sub new {
my ($this, $name, $cmdline, %params) = @_;
- die "please run as root\n" if $> != 0;
-
die "missing name" if !$name;
+ initlog($name);
+
+ my $restart = $ENV{RESTART_PVE_DAEMON};
+ delete $ENV{RESTART_PVE_DAEMON};
+
+ my $lockfd = $ENV{PVE_DAEMON_LOCK_FD};
+ delete $ENV{PVE_DAEMON_LOCK_FD};
+
+ die "please run as root\n" if !$restart && ($> != 0);
+
die "can't create more that one PVE::Daemon" if $daemon_initialized;
$daemon_initialized = 1;
PVE::INotify::inotify_init();
- initlog($name);
-
my $class = ref($this) || $this;
my $self = bless {
name => $name,
run_dir => '/var/run',
+ env_restart_pve_daemon => $restart,
+ env_pve_lock_fd => $lockfd,
+ workers => {},
}, $class;
foreach my $opt (keys %params) {
$self->{$opt} = $value;
} elsif ($opt eq 'run_dir') {
$self->{$opt} = $value;
+ } elsif ($opt eq 'max_workers') {
+ $self->{$opt} = $value;
} else {
die "unknown option '$opt'";
}
$self->{nodename} = PVE::INotify::nodename();
- $self->{cmdline} = $cmdline;
+ $self->{cmdline} = [];
- $0 = $name;
+ foreach my $el (@$cmdline) {
+ $el =~ m/^(.*)$/; # untaint
+ push @{$self->{cmdline}}, $1;
+ }
+ $0 = $name;
return $self;
}
syslog('info' , "server closing");
- # wait for children
- 1 while (waitpid(-1, POSIX::WNOHANG()) > 0);
+ if (!$self->{max_workers}) {
+ # wait for children
+ 1 while (waitpid(-1, POSIX::WNOHANG()) > 0);
+ }
}
# please define in subclass
sub start {
my ($self, $debug) = @_;
- &$server_run($self, $debug);
+ eval { &$server_run($self, $debug); };
+ if (my $err = $@) {
+ syslog('err', "start failed - $err");
+ }
}
my $read_pid = sub {
}
if (-f $self->{pidfile}) {
- # try to get the lock
- &$lockpidfile($self);
- &$server_cleanup($self);
+ eval {
+ # try to get the lock
+ &$lockpidfile($self);
+ &$server_cleanup($self);
+ };
+ if (my $err = $@) {
+ syslog('err', "cleanup failed - $err");
+ }
}
}
my $reload_daemon = sub {
my ($self, $use_hup) = @_;
- if (my $restart = $ENV{RESTART_PVE_DAEMON}) {
+ if ($self->{env_restart_pve_daemon}) {
$self->start();
} else {
my ($running, $pid) = $self->running();
$self->start();
} else {
if ($use_hup) {
- kill(1, $pid);
+ syslog('info', "send HUP to $pid");
+ kill 1, $pid;
} else {
$self->stop();
$self->start();
projects / pve-common.git / blobdiff