about summary refs log tree commit homepage
path: root/lib/PublicInbox/DS.pm
diff options
context:
space:
mode:
Diffstat (limited to 'lib/PublicInbox/DS.pm')
-rw-r--r--lib/PublicInbox/DS.pm1052
1 files changed, 1052 insertions, 0 deletions
diff --git a/lib/PublicInbox/DS.pm b/lib/PublicInbox/DS.pm
new file mode 100644
index 00000000..779215c9
--- /dev/null
+++ b/lib/PublicInbox/DS.pm
@@ -0,0 +1,1052 @@
+# This library is free software; you can redistribute it and/or modify
+# it under the same terms as Perl itself.
+#
+# This license differs from the rest of public-inbox
+#
+# This is a fork of the (for now) unmaintained Danga::Socket 1.61.
+# Unused features will be removed, and updates will be made to take
+# advantage of newer kernels
+
+package PublicInbox::DS;
+use strict;
+use bytes;
+use POSIX ();
+use Time::HiRes ();
+
+use vars qw{$VERSION};
+$VERSION = "1.61";
+
+use warnings;
+no  warnings qw(deprecated);
+
+use PublicInbox::Syscall qw(:epoll);
+
+use fields ('sock',              # underlying socket
+            'fd',                # numeric file descriptor
+            'write_buf',         # arrayref of scalars, scalarrefs, or coderefs to write
+            'write_buf_offset',  # offset into first array of write_buf to start writing at
+            'write_buf_size',    # total length of data in all write_buf items
+            'write_set_watch',   # bool: true if we internally set watch_write rather than by a subclass
+            'closed',            # bool: socket is closed
+            'event_watch',       # bitmask of events the client is interested in (POLLIN,OUT,etc.)
+            'writer_func',       # subref which does writing.  must return bytes written (or undef) and set $! on errors
+            );
+
+use Errno  qw(EINPROGRESS EWOULDBLOCK EISCONN ENOTSOCK
+              EPIPE EAGAIN EBADF ECONNRESET ENOPROTOOPT);
+use Socket qw(IPPROTO_TCP);
+use Carp   qw(croak confess);
+
+use constant DebugLevel => 0;
+
+use constant POLLIN        => 1;
+use constant POLLOUT       => 4;
+use constant POLLERR       => 8;
+use constant POLLHUP       => 16;
+use constant POLLNVAL      => 32;
+
+our $HAVE_KQUEUE = eval { require IO::KQueue; 1 };
+
+our (
+     $HaveEpoll,                 # Flag -- is epoll available?  initially undefined.
+     $HaveKQueue,
+     %DescriptorMap,             # fd (num) -> PublicInbox::DS object
+     $Epoll,                     # Global epoll fd (for epoll mode only)
+     $KQueue,                    # Global kqueue fd (for kqueue mode only)
+     @ToClose,                   # sockets to close when event loop is done
+     %OtherFds,                  # A hash of "other" (non-PublicInbox::DS) file
+                                 # descriptors for the event loop to track.
+
+     $PostLoopCallback,          # subref to call at the end of each loop, if defined (global)
+     %PLCMap,                    # fd (num) -> PostLoopCallback (per-object)
+
+     $LoopTimeout,               # timeout of event loop in milliseconds
+     $DoneInit,                  # if we've done the one-time module init yet
+     @Timers,                    # timers
+     );
+
+# this may be set to zero with old kernels
+our $EPOLLEXCLUSIVE = EPOLLEXCLUSIVE;
+Reset();
+
+#####################################################################
+### C L A S S   M E T H O D S
+#####################################################################
+
+=head2 C<< CLASS->Reset() >>
+
+Reset all state
+
+=cut
+sub Reset {
+    %DescriptorMap = ();
+    @ToClose = ();
+    %OtherFds = ();
+    $LoopTimeout = -1;  # no timeout by default
+    @Timers = ();
+
+    $PostLoopCallback = undef;
+    %PLCMap = ();
+    $DoneInit = 0;
+
+    POSIX::close($Epoll)  if defined $Epoll  && $Epoll  >= 0;
+    POSIX::close($KQueue) if defined $KQueue && $KQueue >= 0;
+
+    *EventLoop = *FirstTimeEventLoop;
+}
+
+=head2 C<< CLASS->HaveEpoll() >>
+
+Returns a true value if this class will use IO::Epoll for async IO.
+
+=cut
+sub HaveEpoll {
+    _InitPoller();
+    return $HaveEpoll;
+}
+
+=head2 C<< CLASS->WatchedSockets() >>
+
+Returns the number of file descriptors which are registered with the global
+poll object.
+
+=cut
+sub WatchedSockets {
+    return scalar keys %DescriptorMap;
+}
+*watched_sockets = *WatchedSockets;
+
+=head2 C<< CLASS->ToClose() >>
+
+Return the list of sockets that are awaiting close() at the end of the
+current event loop.
+
+=cut
+sub ToClose { return @ToClose; }
+
+=head2 C<< CLASS->OtherFds( [%fdmap] ) >>
+
+Get/set the hash of file descriptors that need processing in parallel with
+the registered PublicInbox::DS objects.
+
+=cut
+sub OtherFds {
+    my $class = shift;
+    if ( @_ ) { %OtherFds = @_ }
+    return wantarray ? %OtherFds : \%OtherFds;
+}
+
+=head2 C<< CLASS->AddOtherFds( [%fdmap] ) >>
+
+Add fds to the OtherFds hash for processing.
+
+=cut
+sub AddOtherFds {
+    my $class = shift;
+    %OtherFds = ( %OtherFds, @_ ); # FIXME investigate what happens on dupe fds
+    return wantarray ? %OtherFds : \%OtherFds;
+}
+
+=head2 C<< CLASS->SetLoopTimeout( $timeout ) >>
+
+Set the loop timeout for the event loop to some value in milliseconds.
+
+A timeout of 0 (zero) means poll forever. A timeout of -1 means poll and return
+immediately.
+
+=cut
+sub SetLoopTimeout {
+    return $LoopTimeout = $_[1] + 0;
+}
+
+=head2 C<< CLASS->DebugMsg( $format, @args ) >>
+
+Print the debugging message specified by the C<sprintf>-style I<format> and
+I<args>
+
+=cut
+sub DebugMsg {
+    my ( $class, $fmt, @args ) = @_;
+    chomp $fmt;
+    printf STDERR ">>> $fmt\n", @args;
+}
+
+=head2 C<< CLASS->AddTimer( $seconds, $coderef ) >>
+
+Add a timer to occur $seconds from now. $seconds may be fractional, but timers
+are not guaranteed to fire at the exact time you ask for.
+
+Returns a timer object which you can call C<< $timer->cancel >> on if you need to.
+
+=cut
+sub AddTimer {
+    my $class = shift;
+    my ($secs, $coderef) = @_;
+
+    my $fire_time = Time::HiRes::time() + $secs;
+
+    my $timer = bless [$fire_time, $coderef], "PublicInbox::DS::Timer";
+
+    if (!@Timers || $fire_time >= $Timers[-1][0]) {
+        push @Timers, $timer;
+        return $timer;
+    }
+
+    # Now, where do we insert?  (NOTE: this appears slow, algorithm-wise,
+    # but it was compared against calendar queues, heaps, naive push/sort,
+    # and a bunch of other versions, and found to be fastest with a large
+    # variety of datasets.)
+    for (my $i = 0; $i < @Timers; $i++) {
+        if ($Timers[$i][0] > $fire_time) {
+            splice(@Timers, $i, 0, $timer);
+            return $timer;
+        }
+    }
+
+    die "Shouldn't get here.";
+}
+
+=head2 C<< CLASS->DescriptorMap() >>
+
+Get the hash of PublicInbox::DS objects keyed by the file descriptor (fileno) they
+are wrapping.
+
+Returns a hash in list context or a hashref in scalar context.
+
+=cut
+sub DescriptorMap {
+    return wantarray ? %DescriptorMap : \%DescriptorMap;
+}
+*descriptor_map = *DescriptorMap;
+*get_sock_ref = *DescriptorMap;
+
+sub _InitPoller
+{
+    return if $DoneInit;
+    $DoneInit = 1;
+
+    if ($HAVE_KQUEUE) {
+        $KQueue = IO::KQueue->new();
+        $HaveKQueue = $KQueue >= 0;
+        if ($HaveKQueue) {
+            *EventLoop = *KQueueEventLoop;
+        }
+    }
+    elsif (PublicInbox::Syscall::epoll_defined()) {
+        $Epoll = eval { epoll_create(1024); };
+        $HaveEpoll = defined $Epoll && $Epoll >= 0;
+        if ($HaveEpoll) {
+            *EventLoop = *EpollEventLoop;
+        }
+    }
+
+    if (!$HaveEpoll && !$HaveKQueue) {
+        require IO::Poll;
+        *EventLoop = *PollEventLoop;
+    }
+}
+
+=head2 C<< CLASS->EventLoop() >>
+
+Start processing IO events. In most daemon programs this never exits. See
+C<PostLoopCallback> below for how to exit the loop.
+
+=cut
+sub FirstTimeEventLoop {
+    my $class = shift;
+
+    _InitPoller();
+
+    if ($HaveEpoll) {
+        EpollEventLoop($class);
+    } elsif ($HaveKQueue) {
+        KQueueEventLoop($class);
+    } else {
+        PollEventLoop($class);
+    }
+}
+
+# runs timers and returns milliseconds for next one, or next event loop
+sub RunTimers {
+    return $LoopTimeout unless @Timers;
+
+    my $now = Time::HiRes::time();
+
+    # Run expired timers
+    while (@Timers && $Timers[0][0] <= $now) {
+        my $to_run = shift(@Timers);
+        $to_run->[1]->($now) if $to_run->[1];
+    }
+
+    return $LoopTimeout unless @Timers;
+
+    # convert time to an even number of milliseconds, adding 1
+    # extra, otherwise floating point fun can occur and we'll
+    # call RunTimers like 20-30 times, each returning a timeout
+    # of 0.0000212 seconds
+    my $timeout = int(($Timers[0][0] - $now) * 1000) + 1;
+
+    # -1 is an infinite timeout, so prefer a real timeout
+    return $timeout     if $LoopTimeout == -1;
+
+    # otherwise pick the lower of our regular timeout and time until
+    # the next timer
+    return $LoopTimeout if $LoopTimeout < $timeout;
+    return $timeout;
+}
+
+### The epoll-based event loop. Gets installed as EventLoop if IO::Epoll loads
+### okay.
+sub EpollEventLoop {
+    my $class = shift;
+
+    foreach my $fd ( keys %OtherFds ) {
+        if (epoll_ctl($Epoll, EPOLL_CTL_ADD, $fd, EPOLLIN) == -1) {
+            warn "epoll_ctl(): failure adding fd=$fd; $! (", $!+0, ")\n";
+        }
+    }
+
+    while (1) {
+        my @events;
+        my $i;
+        my $timeout = RunTimers();
+
+        # get up to 1000 events
+        my $evcount = epoll_wait($Epoll, 1000, $timeout, \@events);
+      EVENT:
+        for ($i=0; $i<$evcount; $i++) {
+            my $ev = $events[$i];
+
+            # it's possible epoll_wait returned many events, including some at the end
+            # that ones in the front triggered unregister-interest actions.  if we
+            # can't find the %sock entry, it's because we're no longer interested
+            # in that event.
+            my PublicInbox::DS $pob = $DescriptorMap{$ev->[0]};
+            my $code;
+            my $state = $ev->[1];
+
+            # if we didn't find a Perlbal::Socket subclass for that fd, try other
+            # pseudo-registered (above) fds.
+            if (! $pob) {
+                if (my $code = $OtherFds{$ev->[0]}) {
+                    $code->($state);
+                } else {
+                    my $fd = $ev->[0];
+                    warn "epoll() returned fd $fd w/ state $state for which we have no mapping.  removing.\n";
+                    epoll_ctl($Epoll, EPOLL_CTL_DEL, $fd, 0);
+                    POSIX::close($fd);
+                }
+                next;
+            }
+
+            DebugLevel >= 1 && $class->DebugMsg("Event: fd=%d (%s), state=%d \@ %s\n",
+                                                $ev->[0], ref($pob), $ev->[1], time);
+
+            # standard non-profiling codepat
+            $pob->event_read   if $state & EPOLLIN && ! $pob->{closed};
+            $pob->event_write  if $state & EPOLLOUT && ! $pob->{closed};
+            if ($state & (EPOLLERR|EPOLLHUP)) {
+                $pob->event_err    if $state & EPOLLERR && ! $pob->{closed};
+                $pob->event_hup    if $state & EPOLLHUP && ! $pob->{closed};
+            }
+        }
+        return unless PostEventLoop();
+    }
+    exit 0;
+}
+
+### The fallback IO::Poll-based event loop. Gets installed as EventLoop if
+### IO::Epoll fails to load.
+sub PollEventLoop {
+    my $class = shift;
+
+    my PublicInbox::DS $pob;
+
+    while (1) {
+        my $timeout = RunTimers();
+
+        # the following sets up @poll as a series of ($poll,$event_mask)
+        # items, then uses IO::Poll::_poll, implemented in XS, which
+        # modifies the array in place with the even elements being
+        # replaced with the event masks that occured.
+        my @poll;
+        foreach my $fd ( keys %OtherFds ) {
+            push @poll, $fd, POLLIN;
+        }
+        while ( my ($fd, $sock) = each %DescriptorMap ) {
+            push @poll, $fd, $sock->{event_watch};
+        }
+
+        # if nothing to poll, either end immediately (if no timeout)
+        # or just keep calling the callback
+        unless (@poll) {
+            select undef, undef, undef, ($timeout / 1000);
+            return unless PostEventLoop();
+            next;
+        }
+
+        my $count = IO::Poll::_poll($timeout, @poll);
+        unless ($count >= 0) {
+            return unless PostEventLoop();
+            next;
+        }
+
+        # Fetch handles with read events
+        while (@poll) {
+            my ($fd, $state) = splice(@poll, 0, 2);
+            next unless $state;
+
+            $pob = $DescriptorMap{$fd};
+
+            if (!$pob) {
+                if (my $code = $OtherFds{$fd}) {
+                    $code->($state);
+                }
+                next;
+            }
+
+            $pob->event_read   if $state & POLLIN && ! $pob->{closed};
+            $pob->event_write  if $state & POLLOUT && ! $pob->{closed};
+            $pob->event_err    if $state & POLLERR && ! $pob->{closed};
+            $pob->event_hup    if $state & POLLHUP && ! $pob->{closed};
+        }
+
+        return unless PostEventLoop();
+    }
+
+    exit 0;
+}
+
+### The kqueue-based event loop. Gets installed as EventLoop if IO::KQueue works
+### okay.
+sub KQueueEventLoop {
+    my $class = shift;
+
+    foreach my $fd (keys %OtherFds) {
+        $KQueue->EV_SET($fd, IO::KQueue::EVFILT_READ(), IO::KQueue::EV_ADD());
+    }
+
+    while (1) {
+        my $timeout = RunTimers();
+        my @ret = eval { $KQueue->kevent($timeout) };
+        if (my $err = $@) {
+            # workaround https://rt.cpan.org/Ticket/Display.html?id=116615
+            if ($err =~ /Interrupted system call/) {
+                @ret = ();
+            } else {
+                die $err;
+            }
+        }
+
+        foreach my $kev (@ret) {
+            my ($fd, $filter, $flags, $fflags) = @$kev;
+            my PublicInbox::DS $pob = $DescriptorMap{$fd};
+            if (!$pob) {
+                if (my $code = $OtherFds{$fd}) {
+                    $code->($filter);
+                }  else {
+                    warn "kevent() returned fd $fd for which we have no mapping.  removing.\n";
+                    POSIX::close($fd); # close deletes the kevent entry
+                }
+                next;
+            }
+
+            DebugLevel >= 1 && $class->DebugMsg("Event: fd=%d (%s), flags=%d \@ %s\n",
+                                                        $fd, ref($pob), $flags, time);
+
+            $pob->event_read  if $filter == IO::KQueue::EVFILT_READ()  && !$pob->{closed};
+            $pob->event_write if $filter == IO::KQueue::EVFILT_WRITE() && !$pob->{closed};
+            if ($flags ==  IO::KQueue::EV_EOF() && !$pob->{closed}) {
+                if ($fflags) {
+                    $pob->event_err;
+                } else {
+                    $pob->event_hup;
+                }
+            }
+        }
+        return unless PostEventLoop();
+    }
+
+    exit(0);
+}
+
+=head2 C<< CLASS->SetPostLoopCallback( CODEREF ) >>
+
+Sets post loop callback function.  Pass a subref and it will be
+called every time the event loop finishes.
+
+Return 1 (or any true value) from the sub to make the loop continue, 0 or false
+and it will exit.
+
+The callback function will be passed two parameters: \%DescriptorMap, \%OtherFds.
+
+=cut
+sub SetPostLoopCallback {
+    my ($class, $ref) = @_;
+
+    if (ref $class) {
+        # per-object callback
+        my PublicInbox::DS $self = $class;
+        if (defined $ref && ref $ref eq 'CODE') {
+            $PLCMap{$self->{fd}} = $ref;
+        } else {
+            delete $PLCMap{$self->{fd}};
+        }
+    } else {
+        # global callback
+        $PostLoopCallback = (defined $ref && ref $ref eq 'CODE') ? $ref : undef;
+    }
+}
+
+# Internal function: run the post-event callback, send read events
+# for pushed-back data, and close pending connections.  returns 1
+# if event loop should continue, or 0 to shut it all down.
+sub PostEventLoop {
+    # now we can close sockets that wanted to close during our event processing.
+    # (we didn't want to close them during the loop, as we didn't want fd numbers
+    #  being reused and confused during the event loop)
+    while (my $sock = shift @ToClose) {
+        my $fd = fileno($sock);
+
+        # close the socket.  (not a PublicInbox::DS close)
+        $sock->close;
+
+        # and now we can finally remove the fd from the map.  see
+        # comment above in _cleanup.
+        delete $DescriptorMap{$fd};
+    }
+
+
+    # by default we keep running, unless a postloop callback (either per-object
+    # or global) cancels it
+    my $keep_running = 1;
+
+    # per-object post-loop-callbacks
+    for my $plc (values %PLCMap) {
+        $keep_running &&= $plc->(\%DescriptorMap, \%OtherFds);
+    }
+
+    # now we're at the very end, call callback if defined
+    if (defined $PostLoopCallback) {
+        $keep_running &&= $PostLoopCallback->(\%DescriptorMap, \%OtherFds);
+    }
+
+    return $keep_running;
+}
+
+#####################################################################
+### PublicInbox::DS-the-object code
+#####################################################################
+
+=head2 OBJECT METHODS
+
+=head2 C<< CLASS->new( $socket ) >>
+
+Create a new PublicInbox::DS subclass object for the given I<socket> which will
+react to events on it during the C<EventLoop>.
+
+This is normally (always?) called from your subclass via:
+
+  $class->SUPER::new($socket);
+
+=cut
+sub new {
+    my ($self, $sock, $exclusive) = @_;
+    $self = fields::new($self) unless ref $self;
+
+    $self->{sock}        = $sock;
+    my $fd = fileno($sock);
+
+    Carp::cluck("undef sock and/or fd in PublicInbox::DS->new.  sock=" . ($sock || "") . ", fd=" . ($fd || ""))
+        unless $sock && $fd;
+
+    $self->{fd}          = $fd;
+    $self->{write_buf}      = [];
+    $self->{write_buf_offset} = 0;
+    $self->{write_buf_size} = 0;
+    $self->{closed} = 0;
+
+    my $ev = $self->{event_watch} = POLLERR|POLLHUP|POLLNVAL;
+
+    _InitPoller();
+
+    if ($HaveEpoll) {
+        if ($exclusive) {
+            $ev = $self->{event_watch} = EPOLLIN|EPOLLERR|EPOLLHUP|$EPOLLEXCLUSIVE;
+        }
+retry:
+        if (epoll_ctl($Epoll, EPOLL_CTL_ADD, $fd, $ev)) {
+            if ($!{EINVAL} && ($ev & $EPOLLEXCLUSIVE)) {
+                $EPOLLEXCLUSIVE = 0; # old kernel
+                $ev = $self->{event_watch} = EPOLLIN|EPOLLERR|EPOLLHUP;
+                goto retry;
+            }
+            die "couldn't add epoll watch for $fd: $!\n";
+        }
+    }
+    elsif ($HaveKQueue) {
+        # Add them to the queue but disabled for now
+        $KQueue->EV_SET($fd, IO::KQueue::EVFILT_READ(),
+                        IO::KQueue::EV_ADD() | IO::KQueue::EV_DISABLE());
+        $KQueue->EV_SET($fd, IO::KQueue::EVFILT_WRITE(),
+                        IO::KQueue::EV_ADD() | IO::KQueue::EV_DISABLE());
+    }
+
+    Carp::cluck("PublicInbox::DS::new blowing away existing descriptor map for fd=$fd ($DescriptorMap{$fd})")
+        if $DescriptorMap{$fd};
+
+    $DescriptorMap{$fd} = $self;
+    return $self;
+}
+
+
+#####################################################################
+### I N S T A N C E   M E T H O D S
+#####################################################################
+
+=head2 C<< $obj->steal_socket() >>
+
+Basically returns our socket and makes it so that we don't try to close it,
+but we do remove it from epoll handlers.  THIS CLOSES $self.  It is the same
+thing as calling close, except it gives you the socket to use.
+
+=cut
+sub steal_socket {
+    my PublicInbox::DS $self = $_[0];
+    return if $self->{closed};
+
+    # cleanup does most of the work of closing this socket
+    $self->_cleanup();
+
+    # now undef our internal sock and fd structures so we don't use them
+    my $sock = $self->{sock};
+    $self->{sock} = undef;
+    return $sock;
+}
+
+=head2 C<< $obj->close( [$reason] ) >>
+
+Close the socket. The I<reason> argument will be used in debugging messages.
+
+=cut
+sub close {
+    my PublicInbox::DS $self = $_[0];
+    return if $self->{closed};
+
+    # print out debugging info for this close
+    if (DebugLevel) {
+        my ($pkg, $filename, $line) = caller;
+        my $reason = $_[1] || "";
+        warn "Closing \#$self->{fd} due to $pkg/$filename/$line ($reason)\n";
+    }
+
+    # this does most of the work of closing us
+    $self->_cleanup();
+
+    # defer closing the actual socket until the event loop is done
+    # processing this round of events.  (otherwise we might reuse fds)
+    if ($self->{sock}) {
+        push @ToClose, $self->{sock};
+        $self->{sock} = undef;
+    }
+
+    return 0;
+}
+
+### METHOD: _cleanup()
+### Called by our closers so we can clean internal data structures.
+sub _cleanup {
+    my PublicInbox::DS $self = $_[0];
+
+    # we're effectively closed; we have no fd and sock when we leave here
+    $self->{closed} = 1;
+
+    # we need to flush our write buffer, as there may
+    # be self-referential closures (sub { $client->close })
+    # preventing the object from being destroyed
+    $self->{write_buf} = [];
+
+    # if we're using epoll, we have to remove this from our epoll fd so we stop getting
+    # notifications about it
+    if ($HaveEpoll && $self->{fd}) {
+        if (epoll_ctl($Epoll, EPOLL_CTL_DEL, $self->{fd}, $self->{event_watch}) != 0) {
+            # dump_error prints a backtrace so we can try to figure out why this happened
+            $self->dump_error("epoll_ctl(): failure deleting fd=$self->{fd} during _cleanup(); $! (" . ($!+0) . ")");
+        }
+    }
+
+    # now delete from mappings.  this fd no longer belongs to us, so we don't want
+    # to get alerts for it if it becomes writable/readable/etc.
+    delete $PLCMap{$self->{fd}};
+
+    # we explicitly don't delete from DescriptorMap here until we
+    # actually close the socket, as we might be in the middle of
+    # processing an epoll_wait/etc that returned hundreds of fds, one
+    # of which is not yet processed and is what we're closing.  if we
+    # keep it in DescriptorMap, then the event harnesses can just
+    # looked at $pob->{closed} and ignore it.  but if it's an
+    # un-accounted for fd, then it (understandably) freak out a bit
+    # and emit warnings, thinking their state got off.
+
+    # and finally get rid of our fd so we can't use it anywhere else
+    $self->{fd} = undef;
+}
+
+=head2 C<< $obj->sock() >>
+
+Returns the underlying IO::Handle for the object.
+
+=cut
+sub sock {
+    my PublicInbox::DS $self = shift;
+    return $self->{sock};
+}
+
+=head2 C<< $obj->set_writer_func( CODEREF ) >>
+
+Sets a function to use instead of C<syswrite()> when writing data to the socket.
+
+=cut
+sub set_writer_func {
+   my PublicInbox::DS $self = shift;
+   my $wtr = shift;
+   Carp::croak("Not a subref") unless !defined $wtr || UNIVERSAL::isa($wtr, "CODE");
+   $self->{writer_func} = $wtr;
+}
+
+=head2 C<< $obj->write( $data ) >>
+
+Write the specified data to the underlying handle.  I<data> may be scalar,
+scalar ref, code ref (to run when there), or undef just to kick-start.
+Returns 1 if writes all went through, or 0 if there are writes in queue. If
+it returns 1, caller should stop waiting for 'writable' events)
+
+=cut
+sub write {
+    my PublicInbox::DS $self;
+    my $data;
+    ($self, $data) = @_;
+
+    # nobody should be writing to closed sockets, but caller code can
+    # do two writes within an event, have the first fail and
+    # disconnect the other side (whose destructor then closes the
+    # calling object, but it's still in a method), and then the
+    # now-dead object does its second write.  that is this case.  we
+    # just lie and say it worked.  it'll be dead soon and won't be
+    # hurt by this lie.
+    return 1 if $self->{closed};
+
+    my $bref;
+
+    # just queue data if there's already a wait
+    my $need_queue;
+
+    if (defined $data) {
+        $bref = ref $data ? $data : \$data;
+        if ($self->{write_buf_size}) {
+            push @{$self->{write_buf}}, $bref;
+            $self->{write_buf_size} += ref $bref eq "SCALAR" ? length($$bref) : 1;
+            return 0;
+        }
+
+        # this flag says we're bypassing the queue system, knowing we're the
+        # only outstanding write, and hoping we don't ever need to use it.
+        # if so later, though, we'll need to queue
+        $need_queue = 1;
+    }
+
+  WRITE:
+    while (1) {
+        return 1 unless $bref ||= $self->{write_buf}[0];
+
+        my $len;
+        eval {
+            $len = length($$bref); # this will die if $bref is a code ref, caught below
+        };
+        if ($@) {
+            if (UNIVERSAL::isa($bref, "CODE")) {
+                unless ($need_queue) {
+                    $self->{write_buf_size}--; # code refs are worth 1
+                    shift @{$self->{write_buf}};
+                }
+                $bref->();
+
+                # code refs are just run and never get reenqueued
+                # (they're one-shot), so turn off the flag indicating the
+                # outstanding data needs queueing.
+                $need_queue = 0;
+
+                undef $bref;
+                next WRITE;
+            }
+            die "Write error: $@ <$bref>";
+        }
+
+        my $to_write = $len - $self->{write_buf_offset};
+        my $written;
+        if (my $wtr = $self->{writer_func}) {
+            $written = $wtr->($bref, $to_write, $self->{write_buf_offset});
+        } else {
+            $written = syswrite($self->{sock}, $$bref, $to_write, $self->{write_buf_offset});
+        }
+
+        if (! defined $written) {
+            if ($! == EPIPE) {
+                return $self->close("EPIPE");
+            } elsif ($! == EAGAIN) {
+                # since connection has stuff to write, it should now be
+                # interested in pending writes:
+                if ($need_queue) {
+                    push @{$self->{write_buf}}, $bref;
+                    $self->{write_buf_size} += $len;
+                }
+                $self->{write_set_watch} = 1 unless $self->{event_watch} & POLLOUT;
+                $self->watch_write(1);
+                return 0;
+            } elsif ($! == ECONNRESET) {
+                return $self->close("ECONNRESET");
+            }
+
+            DebugLevel >= 1 && $self->debugmsg("Closing connection ($self) due to write error: $!\n");
+
+            return $self->close("write_error");
+        } elsif ($written != $to_write) {
+            DebugLevel >= 2 && $self->debugmsg("Wrote PARTIAL %d bytes to %d",
+                                               $written, $self->{fd});
+            if ($need_queue) {
+                push @{$self->{write_buf}}, $bref;
+                $self->{write_buf_size} += $len;
+            }
+            # since connection has stuff to write, it should now be
+            # interested in pending writes:
+            $self->{write_buf_offset} += $written;
+            $self->{write_buf_size} -= $written;
+            $self->on_incomplete_write;
+            return 0;
+        } elsif ($written == $to_write) {
+            DebugLevel >= 2 && $self->debugmsg("Wrote ALL %d bytes to %d (nq=%d)",
+                                               $written, $self->{fd}, $need_queue);
+            $self->{write_buf_offset} = 0;
+
+            if ($self->{write_set_watch}) {
+                $self->watch_write(0);
+                $self->{write_set_watch} = 0;
+            }
+
+            # this was our only write, so we can return immediately
+            # since we avoided incrementing the buffer size or
+            # putting it in the buffer.  we also know there
+            # can't be anything else to write.
+            return 1 if $need_queue;
+
+            $self->{write_buf_size} -= $written;
+            shift @{$self->{write_buf}};
+            undef $bref;
+            next WRITE;
+        }
+    }
+}
+
+sub on_incomplete_write {
+    my PublicInbox::DS $self = shift;
+    $self->{write_set_watch} = 1 unless $self->{event_watch} & POLLOUT;
+    $self->watch_write(1);
+}
+
+=head2 C<< $obj->read( $bytecount ) >>
+
+Read at most I<bytecount> bytes from the underlying handle; returns scalar
+ref on read, or undef on connection closed.
+
+=cut
+sub read {
+    my PublicInbox::DS $self = shift;
+    return if $self->{closed};
+    my $bytes = shift;
+    my $buf;
+    my $sock = $self->{sock};
+
+    # if this is too high, perl quits(!!).  reports on mailing lists
+    # don't seem to point to a universal answer.  5MB worked for some,
+    # crashed for others.  1MB works for more people.  let's go with 1MB
+    # for now.  :/
+    my $req_bytes = $bytes > 1048576 ? 1048576 : $bytes;
+
+    my $res = sysread($sock, $buf, $req_bytes, 0);
+    DebugLevel >= 2 && $self->debugmsg("sysread = %d; \$! = %d", $res, $!);
+
+    if (! $res && $! != EWOULDBLOCK) {
+        # catches 0=conn closed or undef=error
+        DebugLevel >= 2 && $self->debugmsg("Fd \#%d read hit the end of the road.", $self->{fd});
+        return undef;
+    }
+
+    return \$buf;
+}
+
+=head2 (VIRTUAL) C<< $obj->event_read() >>
+
+Readable event handler. Concrete deriviatives of PublicInbox::DS should
+provide an implementation of this. The default implementation will die if
+called.
+
+=cut
+sub event_read  { die "Base class event_read called for $_[0]\n"; }
+
+=head2 (VIRTUAL) C<< $obj->event_err() >>
+
+Error event handler. Concrete deriviatives of PublicInbox::DS should
+provide an implementation of this. The default implementation will die if
+called.
+
+=cut
+sub event_err   { die "Base class event_err called for $_[0]\n"; }
+
+=head2 (VIRTUAL) C<< $obj->event_hup() >>
+
+'Hangup' event handler. Concrete deriviatives of PublicInbox::DS should
+provide an implementation of this. The default implementation will die if
+called.
+
+=cut
+sub event_hup   { die "Base class event_hup called for $_[0]\n"; }
+
+=head2 C<< $obj->event_write() >>
+
+Writable event handler. Concrete deriviatives of PublicInbox::DS may wish to
+provide an implementation of this. The default implementation calls
+C<write()> with an C<undef>.
+
+=cut
+sub event_write {
+    my $self = shift;
+    $self->write(undef);
+}
+
+=head2 C<< $obj->watch_read( $boolean ) >>
+
+Turn 'readable' event notification on or off.
+
+=cut
+sub watch_read {
+    my PublicInbox::DS $self = shift;
+    return if $self->{closed} || !$self->{sock};
+
+    my $val = shift;
+    my $event = $self->{event_watch};
+
+    $event &= ~POLLIN if ! $val;
+    $event |=  POLLIN if   $val;
+
+    # If it changed, set it
+    if ($event != $self->{event_watch}) {
+        if ($HaveKQueue) {
+            $KQueue->EV_SET($self->{fd}, IO::KQueue::EVFILT_READ(),
+                            $val ? IO::KQueue::EV_ENABLE() : IO::KQueue::EV_DISABLE());
+        }
+        elsif ($HaveEpoll) {
+            epoll_ctl($Epoll, EPOLL_CTL_MOD, $self->{fd}, $event)
+                and $self->dump_error("couldn't modify epoll settings for $self->{fd} " .
+                                      "from $self->{event_watch} -> $event: $! (" . ($!+0) . ")");
+        }
+        $self->{event_watch} = $event;
+    }
+}
+
+=head2 C<< $obj->watch_write( $boolean ) >>
+
+Turn 'writable' event notification on or off.
+
+=cut
+sub watch_write {
+    my PublicInbox::DS $self = shift;
+    return if $self->{closed} || !$self->{sock};
+
+    my $val = shift;
+    my $event = $self->{event_watch};
+
+    $event &= ~POLLOUT if ! $val;
+    $event |=  POLLOUT if   $val;
+
+    if ($val && caller ne __PACKAGE__) {
+        # A subclass registered interest, it's now responsible for this.
+        $self->{write_set_watch} = 0;
+    }
+
+    # If it changed, set it
+    if ($event != $self->{event_watch}) {
+        if ($HaveKQueue) {
+            $KQueue->EV_SET($self->{fd}, IO::KQueue::EVFILT_WRITE(),
+                            $val ? IO::KQueue::EV_ENABLE() : IO::KQueue::EV_DISABLE());
+        }
+        elsif ($HaveEpoll) {
+            epoll_ctl($Epoll, EPOLL_CTL_MOD, $self->{fd}, $event)
+                and $self->dump_error("couldn't modify epoll settings for $self->{fd} " .
+                                      "from $self->{event_watch} -> $event: $! (" . ($!+0) . ")");
+        }
+        $self->{event_watch} = $event;
+    }
+}
+
+=head2 C<< $obj->dump_error( $message ) >>
+
+Prints to STDERR a backtrace with information about this socket and what lead
+up to the dump_error call.
+
+=cut
+sub dump_error {
+    my $i = 0;
+    my @list;
+    while (my ($file, $line, $sub) = (caller($i++))[1..3]) {
+        push @list, "\t$file:$line called $sub\n";
+    }
+
+    warn "ERROR: $_[1]\n" .
+        "\t$_[0] = " . $_[0]->as_string . "\n" .
+        join('', @list);
+}
+
+=head2 C<< $obj->debugmsg( $format, @args ) >>
+
+Print the debugging message specified by the C<sprintf>-style I<format> and
+I<args>.
+
+=cut
+sub debugmsg {
+    my ( $self, $fmt, @args ) = @_;
+    confess "Not an object" unless ref $self;
+
+    chomp $fmt;
+    printf STDERR ">>> $fmt\n", @args;
+}
+
+=head2 C<< $obj->as_string() >>
+
+Returns a string describing this socket.
+
+=cut
+sub as_string {
+    my PublicInbox::DS $self = shift;
+    my $rw = "(" . ($self->{event_watch} & POLLIN ? 'R' : '') .
+                   ($self->{event_watch} & POLLOUT ? 'W' : '') . ")";
+    my $ret = ref($self) . "$rw: " . ($self->{closed} ? "closed" : "open");
+    return $ret;
+}
+
+package PublicInbox::DS::Timer;
+# [$abs_float_firetime, $coderef];
+sub cancel {
+    $_[0][1] = undef;
+}
+
+1;
+
+=head1 AUTHORS (Danga::Socket)
+
+Brad Fitzpatrick <brad@danga.com> - author
+
+Michael Granger <ged@danga.com> - docs, testing
+
+Mark Smith <junior@danga.com> - contributor, heavy user, testing
+
+Matt Sergeant <matt@sergeant.org> - kqueue support, docs, timers, other bits