1 # This library is free software; you can redistribute it and/or modify
2 # it under the same terms as Perl itself.
4 # This license differs from the rest of public-inbox
6 # This is a fork of the unmaintained Danga::Socket (1.61) with
7 # significant changes. See Documentation/technical/ds.txt in our
10 # Do not expect this to be a stable API like Danga::Socket,
11 # but it will evolve to suite our needs and to take advantage of
12 # newer Linux and *BSD features.
13 # Bugs encountered were reported to bug-Danga-Socket@rt.cpan.org,
14 # fixed in Danga::Socket 1.62 and visible at:
15 # https://rt.cpan.org/Public/Dist/Display.html?Name=Danga-Socket
18 # sock: underlying socket
19 # rbuf: scalarref, usually undef
20 # wbuf: arrayref of coderefs or tmpio (autovivified))
21 # (tmpio = [ GLOB, offset, [ length ] ])
22 package PublicInbox::DS;
25 use parent qw(Exporter);
26 use bytes qw(length substr); # FIXME(?): needed for PublicInbox::NNTP
27 use POSIX qw(WNOHANG sigprocmask SIG_SETMASK);
28 use Fcntl qw(SEEK_SET :DEFAULT O_APPEND);
29 use Time::HiRes qw(clock_gettime CLOCK_MONOTONIC);
30 use Scalar::Util qw(blessed);
31 use PublicInbox::Syscall qw(:epoll);
32 use PublicInbox::Tmpfile;
33 use Errno qw(EAGAIN EINVAL);
34 use Carp qw(carp croak);
35 our @EXPORT_OK = qw(now msg_more dwaitpid awaitpid add_timer add_uniq_timer);
38 my $nextq; # queue for next_tick
39 my $wait_pids; # list of [ pid, callback, callback_arg ]
40 my $AWAIT_PIDS; # pid => [ $callback, @args ]
42 my $ToClose; # sockets to close when event loop is done
44 %DescriptorMap, # fd (num) -> PublicInbox::DS object
45 $Epoll, # Global epoll fd (or DSKQXS ref)
46 $ep_io, # IO::Handle for Epoll
48 $PostLoopCallback, # subref to call at the end of each loop, if defined (global)
50 $LoopTimeout, # timeout of event loop in milliseconds
58 #####################################################################
59 ### C L A S S M E T H O D S
60 #####################################################################
62 =head2 C<< CLASS->Reset() >>
69 $in_loop = undef; # first in case DESTROY callbacks use this
73 $PostLoopCallback = undef;
75 # we may be iterating inside one of these on our stack
76 my @q = delete @Stack{keys %Stack};
77 for my $q (@q) { @$q = () }
78 $AWAIT_PIDS = $wait_pids = $nextq = $ToClose = undef;
79 $ep_io = undef; # closes real $Epoll FD
80 $Epoll = undef; # may call DSKQXS::DESTROY
81 } while (@Timers || keys(%Stack) || $nextq || $wait_pids ||
82 $ToClose || keys(%DescriptorMap) || $AWAIT_PIDS ||
83 $PostLoopCallback || keys(%UniqTimer));
86 $LoopTimeout = -1; # no timeout by default
89 =head2 C<< CLASS->SetLoopTimeout( $timeout ) >>
91 Set the loop timeout for the event loop to some value in milliseconds.
93 A timeout of 0 (zero) means poll forever. A timeout of -1 means poll and return
97 sub SetLoopTimeout { $LoopTimeout = $_[1] + 0 }
99 sub _add_named_timer {
100 my ($name, $secs, $coderef, @args) = @_;
101 my $fire_time = now() + $secs;
102 my $timer = [$fire_time, $name, $coderef, @args];
104 if (!@Timers || $fire_time >= $Timers[-1][0]) {
105 push @Timers, $timer;
109 # Now, where do we insert? (NOTE: this appears slow, algorithm-wise,
110 # but it was compared against calendar queues, heaps, naive push/sort,
111 # and a bunch of other versions, and found to be fastest with a large
112 # variety of datasets.)
113 for (my $i = 0; $i < @Timers; $i++) {
114 if ($Timers[$i][0] > $fire_time) {
115 splice(@Timers, $i, 0, $timer);
119 die "Shouldn't get here.";
122 sub add_timer { _add_named_timer(undef, @_) }
124 sub add_uniq_timer { # ($name, $secs, $coderef, @args) = @_;
125 $UniqTimer{$_[0]} //= _add_named_timer(@_);
128 # caller sets return value to $Epoll
130 if (defined $PublicInbox::Syscall::SYS_epoll_create) {
131 my $fd = epoll_create();
132 die "epoll_create: $!" if $fd < 0;
133 open($ep_io, '+<&=', $fd) or return;
134 fcntl($ep_io, F_SETFD, FD_CLOEXEC);
138 for (qw(DSKQXS DSPoll)) {
139 $cls = "PublicInbox::$_";
140 last if eval "require $cls";
142 $cls->import(qw(epoll_ctl epoll_wait));
147 sub now () { clock_gettime(CLOCK_MONOTONIC) }
150 my $q = $nextq or return;
152 $Stack{cur_runq} = $q;
154 # avoid "ref" on blessed refs to workaround a Perl 5.16.3 leak:
155 # https://rt.perl.org/Public/Bug/Display.html?id=114340
162 delete $Stack{cur_runq};
165 # runs timers and returns milliseconds for next one, or next event loop
169 return (($nextq || $ToClose) ? 0 : $LoopTimeout) unless @Timers;
174 while (@Timers && $Timers[0][0] <= $now) {
175 my $to_run = shift(@Timers);
176 delete $UniqTimer{$to_run->[1] // ''};
177 $to_run->[2]->(@$to_run[3..$#$to_run]);
180 # timers may enqueue into nextq:
181 return 0 if ($nextq || $ToClose);
183 return $LoopTimeout unless @Timers;
185 # convert time to an even number of milliseconds, adding 1
186 # extra, otherwise floating point fun can occur and we'll
187 # call RunTimers like 20-30 times, each returning a timeout
188 # of 0.0000212 seconds
189 my $timeout = int(($Timers[0][0] - $now) * 1000) + 1;
191 # -1 is an infinite timeout, so prefer a real timeout
192 ($LoopTimeout < 0 || $LoopTimeout >= $timeout) ? $timeout : $LoopTimeout
195 sub sig_setmask { sigprocmask(SIG_SETMASK, @_) or die "sigprocmask: $!" }
197 sub block_signals () {
198 my $oldset = POSIX::SigSet->new;
199 my $newset = POSIX::SigSet->new;
200 $newset->fillset or die "fillset: $!";
201 sig_setmask($newset, $oldset);
206 my ($pid, @cb_args) = @_;
207 my $cb = shift @cb_args or return;
208 eval { $cb->($pid, @cb_args) };
209 warn "E: awaitpid($pid): $@" if $@;
212 # We can't use waitpid(-1) safely here since it can hit ``, system(),
213 # and other things. So we scan the $wait_pids list, which is hopefully
214 # not too big. We keep $wait_pids small by not calling dwaitpid()
215 # until we've hit EOF when reading the stdout of the child.
219 my $tmp = $wait_pids // [];
221 $Stack{reap_runq} = $tmp;
222 my $oldset = block_signals();
225 foreach my $ary (@$tmp) {
226 my ($pid, $cb, $arg) = @$ary;
227 my $ret = waitpid($pid, WNOHANG);
229 push @$wait_pids, $ary; # autovivifies @$wait_pids
230 } elsif ($ret == $pid) {
232 eval { $cb->($arg, $pid) };
233 warn "E: dwaitpid($pid) in_loop: $@" if $@;
236 warn "waitpid($pid, WNOHANG) = $ret, \$!=$!, \$?=$?";
240 # new API TODO: convert to waitpid(-1) in the future as long
241 # as we don't use threads
242 for my $pid (keys %$AWAIT_PIDS) {
243 my $wpid = waitpid($pid, WNOHANG) // next;
244 my $cb_args = delete $AWAIT_PIDS->{$wpid} or next;
245 await_cb($pid, @$cb_args);
247 sig_setmask($oldset);
248 delete $Stack{reap_runq};
251 # reentrant SIGCHLD handler (since reap_pids is not reentrant)
252 sub enqueue_reap () { $reap_armed //= requeue(\&reap_pids) }
254 sub in_loop () { $in_loop }
256 # Internal function: run the post-event callback, send read events
257 # for pushed-back data, and close pending connections. returns 1
258 # if event loop should continue, or 0 to shut it all down.
259 sub PostEventLoop () {
260 # now we can close sockets that wanted to close during our event
261 # processing. (we didn't want to close them during the loop, as we
262 # didn't want fd numbers being reused and confused during the event
264 if (my $close_now = $ToClose) {
265 $ToClose = undef; # will be autovivified on push
266 @$close_now = map { fileno($_) } @$close_now;
268 # ->DESTROY methods may populate ToClose
269 delete @DescriptorMap{@$close_now};
272 # by default we keep running, unless a postloop callback cancels it
273 $PostLoopCallback ? $PostLoopCallback->(\%DescriptorMap) : 1;
276 # Start processing IO events. In most daemon programs this never exits. See
277 # C<PostLoopCallback> for how to exit the loop.
278 sub event_loop (;$$) {
279 my ($sig, $oldset) = @_;
280 $Epoll //= _InitPoller();
281 require PublicInbox::Sigfd if $sig;
282 my $sigfd = PublicInbox::Sigfd->new($sig, 1) if $sig;
283 local @SIG{keys %$sig} = values(%$sig) if $sig && !$sigfd;
284 local $SIG{PIPE} = 'IGNORE';
285 if (!$sigfd && $sig) {
286 # wake up every second to accept signals if we don't
287 # have signalfd or IO::KQueue:
288 sig_setmask($oldset);
289 PublicInbox::DS->SetLoopTimeout(1000);
291 $_[0] = $sigfd = $sig = undef; # $_[0] == sig
295 my $timeout = RunTimers();
297 # get up to 1000 events
298 epoll_wait($Epoll, 1000, $timeout, \@events);
299 for my $fd (@events) {
300 # it's possible epoll_wait returned many events,
301 # including some at the end that ones in the front
302 # triggered unregister-interest actions. if we can't
303 # find the %sock entry, it's because we're no longer
304 # interested in that event.
306 # guard stack-not-refcounted w/ Carp + @DB::args
307 my $obj = $DescriptorMap{$fd};
310 } while (PostEventLoop());
313 =head2 C<< CLASS->SetPostLoopCallback( CODEREF ) >>
315 Sets post loop callback function. Pass a subref and it will be
316 called every time the event loop finishes.
318 Return 1 (or any true value) from the sub to make the loop continue, 0 or false
321 The callback function will be passed two parameters: \%DescriptorMap
324 sub SetPostLoopCallback {
325 my ($class, $ref) = @_;
328 $PostLoopCallback = (defined $ref && ref $ref eq 'CODE') ? $ref : undef;
331 #####################################################################
332 ### PublicInbox::DS-the-object code
333 #####################################################################
335 =head2 OBJECT METHODS
337 =head2 C<< CLASS->new( $socket ) >>
339 Create a new PublicInbox::DS subclass object for the given I<socket> which will
340 react to events on it during the C<event_loop>.
342 This is normally (always?) called from your subclass via:
344 $class->SUPER::new($socket);
348 my ($self, $sock, $ev) = @_;
349 $self->{sock} = $sock;
350 my $fd = fileno($sock);
352 $Epoll //= _InitPoller();
354 if (epoll_ctl($Epoll, EPOLL_CTL_ADD, $fd, $ev)) {
355 if ($! == EINVAL && ($ev & EPOLLEXCLUSIVE)) {
356 $ev &= ~EPOLLEXCLUSIVE;
359 die "EPOLL_CTL_ADD $self/$sock/$fd: $!";
361 croak("FD:$fd in use by $DescriptorMap{$fd} (for $self/$sock)")
362 if defined($DescriptorMap{$fd});
364 $DescriptorMap{$fd} = $self;
367 # for IMAP, NNTP, and POP3 which greet clients upon connect
369 my ($self, $sock) = @_;
372 if ($sock->can('accept_SSL') && !$sock->accept_SSL) {
373 return CORE::close($sock) if $! != EAGAIN;
374 $ev = PublicInbox::TLS::epollbit() or return CORE::close($sock);
375 $wbuf = [ \&accept_tls_step, $self->can('do_greet')];
377 new($self, $sock, $ev | EPOLLONESHOT);
379 $self->{wbuf} = $wbuf;
386 #####################################################################
387 ### I N S T A N C E M E T H O D S
388 #####################################################################
390 sub requeue ($) { push @$nextq, $_[0] } # autovivifies
392 =head2 C<< $obj->close >>
399 my $sock = delete $self->{sock} or return;
401 # we need to flush our write buffer, as there may
402 # be self-referential closures (sub { $client->close })
403 # preventing the object from being destroyed
404 delete $self->{wbuf};
406 # if we're using epoll, we have to remove this from our epoll fd so we stop getting
407 # notifications about it
408 my $fd = fileno($sock);
409 epoll_ctl($Epoll, EPOLL_CTL_DEL, $fd, 0) and
410 croak("EPOLL_CTL_DEL($self/$sock): $!");
412 # we explicitly don't delete from DescriptorMap here until we
413 # actually close the socket, as we might be in the middle of
414 # processing an epoll_wait/etc that returned hundreds of fds, one
415 # of which is not yet processed and is what we're closing. if we
416 # keep it in DescriptorMap, then the event harnesses can just
417 # looked at $pob->{sock} == undef and ignore it. but if it's an
418 # un-accounted for fd, then it (understandably) freak out a bit
419 # and emit warnings, thinking their state got off.
421 # defer closing the actual socket until the event loop is done
422 # processing this round of events. (otherwise we might reuse fds)
423 push @$ToClose, $sock; # autovivifies $ToClose
428 # portable, non-thread-safe sendfile emulation (no pread, yet)
429 sub send_tmpio ($$) {
430 my ($sock, $tmpio) = @_;
432 sysseek($tmpio->[0], $tmpio->[1], SEEK_SET) or return;
433 my $n = $tmpio->[2] // 65536;
434 $n = 65536 if $n > 65536;
435 defined(my $to_write = sysread($tmpio->[0], my $buf, $n)) or return;
437 while ($to_write > 0) {
438 if (defined(my $w = syswrite($sock, $buf, $to_write, $written))) {
442 return if $written == 0;
446 $tmpio->[1] += $written; # offset
447 $tmpio->[2] -= $written if defined($tmpio->[2]); # length
451 sub epbit ($$) { # (sock, default)
452 $_[0]->can('stop_SSL') ? PublicInbox::TLS::epollbit() : $_[1];
455 # returns 1 if done, 0 if incomplete
456 sub flush_write ($) {
458 my $sock = $self->{sock} or return;
459 my $wbuf = $self->{wbuf} or return 1;
462 while (my $bref = $wbuf->[0]) {
463 if (ref($bref) ne 'CODE') {
465 my $w = send_tmpio($sock, $bref); # bref is tmpio
471 } elsif ($! == EAGAIN) {
472 my $ev = epbit($sock, EPOLLOUT) or return $self->close;
473 epwait($sock, $ev | EPOLLONESHOT);
479 } else { #(ref($bref) eq 'CODE') {
481 my $before = scalar(@$wbuf);
484 # bref may be enqueueing more CODE to call (see accept_tls_step)
485 return 0 if (scalar(@$wbuf) > $before);
489 delete $self->{wbuf};
494 my ($self, $rbuf) = @_;
495 if ($$rbuf eq '') { # who knows how long till we can read again
496 delete $self->{rbuf};
498 $self->{rbuf} = $rbuf;
502 sub do_read ($$$;$) {
503 my ($self, $rbuf, $len, $off) = @_;
504 my $r = sysread(my $sock = $self->{sock}, $$rbuf, $len, $off // 0);
505 return ($r == 0 ? $self->close : $r) if defined $r;
506 # common for clients to break connections without warning,
507 # would be too noisy to log here:
509 my $ev = epbit($sock, EPOLLIN) or return $self->close;
510 epwait($sock, $ev | EPOLLONESHOT);
511 rbuf_idle($self, $rbuf);
518 # drop the socket if we hit unrecoverable errors on our system which
519 # require BOFH attention: ENOSPC, EFBIG, EIO, EMFILE, ENFILE...
527 my ($self, $bref, $off) = @_;
528 my $fh = tmpfile('wbuf', $self->{sock}, O_APPEND) or
529 return drop($self, "tmpfile $!");
531 my $len = length($$bref) - $off;
532 my $n = syswrite($fh, $$bref, $len, $off) //
533 return drop($self, "write ($len): $!");
534 $n == $len or return drop($self, "wrote $n < $len bytes");
535 [ $fh, 0 ] # [1] = offset, [2] = length, not set by us
538 =head2 C<< $obj->write( $data ) >>
540 Write the specified data to the underlying handle. I<data> may be scalar,
541 scalar ref, code ref (to run when there).
542 Returns 1 if writes all went through, or 0 if there are writes in queue. If
543 it returns 1, caller should stop waiting for 'writable' events)
547 my ($self, $data) = @_;
549 # nobody should be writing to closed sockets, but caller code can
550 # do two writes within an event, have the first fail and
551 # disconnect the other side (whose destructor then closes the
552 # calling object, but it's still in a method), and then the
553 # now-dead object does its second write. that is this case. we
554 # just lie and say it worked. it'll be dead soon and won't be
556 my $sock = $self->{sock} or return 1;
558 my $bref = $ref ? $data : \$data;
559 my $wbuf = $self->{wbuf};
560 if ($wbuf && scalar(@$wbuf)) { # already buffering, can't write more...
561 if ($ref eq 'CODE') {
564 my $tmpio = $wbuf->[-1];
565 if ($tmpio && !defined($tmpio->[2])) { # append to tmp file buffer
566 $tmpio->[0]->print($$bref) or return drop($self, "print: $!");
568 my $tmpio = tmpio($self, $bref, 0) or return 0;
573 } elsif ($ref eq 'CODE') {
577 my $to_write = length($$bref);
578 my $written = syswrite($sock, $$bref, $to_write);
580 if (defined $written) {
581 return 1 if $written == $to_write;
582 requeue($self); # runs: event_step -> flush_write
583 } elsif ($! == EAGAIN) {
584 my $ev = epbit($sock, EPOLLOUT) or return $self->close;
585 epwait($sock, $ev | EPOLLONESHOT);
591 # deal with EAGAIN or partial write:
592 my $tmpio = tmpio($self, $bref, $written) or return 0;
594 # wbuf may be an empty array if we're being called inside
595 # ->flush_write via CODE bref:
596 push @{$self->{wbuf}}, $tmpio; # autovivifies
601 use constant MSG_MORE => ($^O eq 'linux') ? 0x8000 : 0;
605 my $sock = $self->{sock} or return 1;
606 my $wbuf = $self->{wbuf};
608 if (MSG_MORE && (!defined($wbuf) || !scalar(@$wbuf)) &&
609 !$sock->can('stop_SSL')) {
610 my $n = send($sock, $_[1], MSG_MORE);
612 my $nlen = length($_[1]) - $n;
613 return 1 if $nlen == 0; # all done!
614 # queue up the unwritten substring:
615 my $tmpio = tmpio($self, \($_[1]), $n) or return 0;
616 push @{$self->{wbuf}}, $tmpio; # autovivifies
617 epwait($sock, EPOLLOUT|EPOLLONESHOT);
622 # don't redispatch into NNTPdeflate::write
623 PublicInbox::DS::write($self, \($_[1]));
627 my ($sock, $ev) = @_;
628 epoll_ctl($Epoll, EPOLL_CTL_MOD, fileno($sock), $ev) and
629 croak("EPOLL_CTL_MOD($sock): $!");
632 # return true if complete, false if incomplete (or failure)
633 sub accept_tls_step ($) {
635 my $sock = $self->{sock} or return;
636 return 1 if $sock->accept_SSL;
637 return $self->close if $! != EAGAIN;
638 my $ev = PublicInbox::TLS::epollbit() or return $self->close;
639 epwait($sock, $ev | EPOLLONESHOT);
640 unshift(@{$self->{wbuf}}, \&accept_tls_step); # autovivifies
644 # return true if complete, false if incomplete (or failure)
645 sub shutdn_tls_step ($) {
647 my $sock = $self->{sock} or return;
648 return $self->close if $sock->stop_SSL(SSL_fast_shutdown => 1);
649 return $self->close if $! != EAGAIN;
650 my $ev = PublicInbox::TLS::epollbit() or return $self->close;
651 epwait($sock, $ev | EPOLLONESHOT);
652 unshift(@{$self->{wbuf}}, \&shutdn_tls_step); # autovivifies
656 # don't bother with shutdown($sock, 2), we don't fork+exec w/o CLOEXEC
657 # or fork w/o exec, so no inadvertent socket sharing
660 my $sock = $self->{sock} or return;
661 if ($sock->can('stop_SSL')) {
662 shutdn_tls_step($self);
668 sub dflush {} # overridden by DSdeflate
669 sub compressed {} # overridden by DSdeflate
670 sub long_response_done {} # overridden by Net::NNTP
674 # wbuf is unset or empty, here; {long} may add to it
675 my ($fd, $cb, $t0, @args) = @{$self->{long_cb}};
676 my $more = eval { $cb->($self, @args) };
677 if ($@ || !$self->{sock}) { # something bad happened...
678 delete $self->{long_cb};
679 my $elapsed = now() - $t0;
680 $@ and warn("$@ during long response[$fd] - ",
681 sprintf('%0.6f', $elapsed),"\n");
682 $self->out(" deferred[$fd] aborted - %0.6f", $elapsed);
684 } elsif ($more) { # $self->{wbuf}:
685 # control passed to ibx_async_cat if $more == \undef
686 requeue_once($self) if !ref($more);
688 delete $self->{long_cb};
689 $self->long_response_done;
690 my $elapsed = now() - $t0;
691 my $fd = fileno($self->{sock});
692 $self->out(" deferred[$fd] done - %0.6f", $elapsed);
693 my $wbuf = $self->{wbuf}; # do NOT autovivify
694 requeue($self) unless $wbuf && @$wbuf;
700 # COMPRESS users all share the same DEFLATE context.
701 # Flush it here to ensure clients don't see each other's data
704 # no recursion, schedule another call ASAP,
705 # but only after all pending writes are done.
706 # autovivify wbuf. wbuf may be populated by $cb,
707 # no need to rearm if so: (push returns new size of array)
708 $self->requeue if push(@{$self->{wbuf}}, \&long_step) == 1;
711 sub long_response ($$;@) {
712 my ($self, $cb, @args) = @_; # cb returns true if more, false if done
713 my $sock = $self->{sock} or return;
714 # make sure we disable reading during a long response,
715 # clients should not be sending us stuff and making us do more
716 # work while we are stream a response to them
717 $self->{long_cb} = [ fileno($sock), $cb, now(), @args ];
718 long_step($self); # kick off!
722 sub dwaitpid ($;$$) {
723 my ($pid, $cb, $arg) = @_;
725 push @$wait_pids, [ $pid, $cb, $arg ];
726 # We could've just missed our SIGCHLD, cover it, here:
729 my $ret = waitpid($pid, 0);
732 eval { $cb->($arg, $pid) };
733 carp "E: dwaitpid($pid) !in_loop: $@" if $@;
736 carp "waitpid($pid, 0) = $ret, \$!=$!, \$?=$?";
742 my ($pid, @cb_args) = @_;
743 $AWAIT_PIDS->{$pid} //= @cb_args ? \@cb_args : 0;
744 # provide synchronous API
745 if (defined(wantarray) || (!$in_loop && !@cb_args)) {
746 my $ret = waitpid($pid, 0) // -2;
748 my $cb_args = delete $AWAIT_PIDS->{$pid};
749 @cb_args = @$cb_args if !@cb_args && $cb_args;
750 await_cb($pid, @cb_args);
754 # We could've just missed our SIGCHLD, cover it, here:
755 enqueue_reap() if $in_loop;
760 =head1 AUTHORS (Danga::Socket)
762 Brad Fitzpatrick <brad@danga.com> - author
764 Michael Granger <ged@danga.com> - docs, testing
766 Mark Smith <junior@danga.com> - contributor, heavy user, testing
768 Matt Sergeant <matt@sergeant.org> - kqueue support, docs, timers, other bits