ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/AnyEvent/lib/AnyEvent/Handle.pm
Revision: 1.150
Committed: Thu Jul 16 04:16:25 2009 UTC (14 years, 10 months ago) by root
Branch: MAIN
Changes since 1.149: +18 -15 lines
Log Message:
*** empty log message ***

File Contents

# Content
1 package AnyEvent::Handle;
2
3 no warnings;
4 use strict qw(subs vars);
5
6 use AnyEvent ();
7 use AnyEvent::Util qw(WSAEWOULDBLOCK);
8 use Scalar::Util ();
9 use Carp ();
10 use Fcntl ();
11 use Errno qw(EAGAIN EINTR);
12
13 =head1 NAME
14
15 AnyEvent::Handle - non-blocking I/O on file handles via AnyEvent
16
17 =cut
18
19 our $VERSION = 4.82;
20
21 =head1 SYNOPSIS
22
23 use AnyEvent;
24 use AnyEvent::Handle;
25
26 my $cv = AnyEvent->condvar;
27
28 my $hdl; $hdl = new AnyEvent::Handle
29 fh => \*STDIN,
30 on_error => sub {
31 warn "got error $_[2]\n";
32 $cv->send;
33 );
34
35 # send some request line
36 $hdl->push_write ("getinfo\015\012");
37
38 # read the response line
39 $hdl->push_read (line => sub {
40 my ($hdl, $line) = @_;
41 warn "got line <$line>\n";
42 $cv->send;
43 });
44
45 $cv->recv;
46
47 =head1 DESCRIPTION
48
49 This module is a helper module to make it easier to do event-based I/O on
50 filehandles. For utility functions for doing non-blocking connects and accepts
51 on sockets see L<AnyEvent::Util>.
52
53 The L<AnyEvent::Intro> tutorial contains some well-documented
54 AnyEvent::Handle examples.
55
56 In the following, when the documentation refers to of "bytes" then this
57 means characters. As sysread and syswrite are used for all I/O, their
58 treatment of characters applies to this module as well.
59
60 All callbacks will be invoked with the handle object as their first
61 argument.
62
63 =head1 METHODS
64
65 =over 4
66
67 =item $handle = B<new> AnyEvent::TLS fh => $filehandle, key => value...
68
69 The constructor supports these arguments (all as C<< key => value >> pairs).
70
71 =over 4
72
73 =item fh => $filehandle [MANDATORY]
74
75 The filehandle this L<AnyEvent::Handle> object will operate on.
76
77 NOTE: The filehandle will be set to non-blocking mode (using
78 C<AnyEvent::Util::fh_nonblocking>) by the constructor and needs to stay in
79 that mode.
80
81 =item on_eof => $cb->($handle)
82
83 Set the callback to be called when an end-of-file condition is detected,
84 i.e. in the case of a socket, when the other side has closed the
85 connection cleanly, and there are no outstanding read requests in the
86 queue (if there are read requests, then an EOF counts as an unexpected
87 connection close and will be flagged as an error).
88
89 For sockets, this just means that the other side has stopped sending data,
90 you can still try to write data, and, in fact, one can return from the EOF
91 callback and continue writing data, as only the read part has been shut
92 down.
93
94 If an EOF condition has been detected but no C<on_eof> callback has been
95 set, then a fatal error will be raised with C<$!> set to <0>.
96
97 =item on_error => $cb->($handle, $fatal, $message)
98
99 This is the error callback, which is called when, well, some error
100 occured, such as not being able to resolve the hostname, failure to
101 connect or a read error.
102
103 Some errors are fatal (which is indicated by C<$fatal> being true). On
104 fatal errors the handle object will be destroyed (by a call to C<< ->
105 destroy >>) after invoking the error callback (which means you are free to
106 examine the handle object). Examples of fatal errors are an EOF condition
107 with active (but unsatisifable) read watchers (C<EPIPE>) or I/O errors.
108
109 AnyEvent::Handle tries to find an appropriate error code for you to check
110 against, but in some cases (TLS errors), this does not work well. It is
111 recommended to always output the C<$message> argument in human-readable
112 error messages (it's usually the same as C<"$!">).
113
114 Non-fatal errors can be retried by simply returning, but it is recommended
115 to simply ignore this parameter and instead abondon the handle object
116 when this callback is invoked. Examples of non-fatal errors are timeouts
117 C<ETIMEDOUT>) or badly-formatted data (C<EBADMSG>).
118
119 On callback entrance, the value of C<$!> contains the operating system
120 error code (or C<ENOSPC>, C<EPIPE>, C<ETIMEDOUT>, C<EBADMSG> or
121 C<EPROTO>).
122
123 While not mandatory, it is I<highly> recommended to set this callback, as
124 you will not be notified of errors otherwise. The default simply calls
125 C<croak>.
126
127 =item on_read => $cb->($handle)
128
129 This sets the default read callback, which is called when data arrives
130 and no read request is in the queue (unlike read queue callbacks, this
131 callback will only be called when at least one octet of data is in the
132 read buffer).
133
134 To access (and remove data from) the read buffer, use the C<< ->rbuf >>
135 method or access the C<< $handle->{rbuf} >> member directly. Note that you
136 must not enlarge or modify the read buffer, you can only remove data at
137 the beginning from it.
138
139 When an EOF condition is detected then AnyEvent::Handle will first try to
140 feed all the remaining data to the queued callbacks and C<on_read> before
141 calling the C<on_eof> callback. If no progress can be made, then a fatal
142 error will be raised (with C<$!> set to C<EPIPE>).
143
144 Note that, unlike requests in the read queue, an C<on_read> callback
145 doesn't mean you I<require> some data: if there is an EOF and there
146 are outstanding read requests then an error will be flagged. With an
147 C<on_read> callback, the C<on_eof> callback will be invoked.
148
149 =item on_drain => $cb->($handle)
150
151 This sets the callback that is called when the write buffer becomes empty
152 (or when the callback is set and the buffer is empty already).
153
154 To append to the write buffer, use the C<< ->push_write >> method.
155
156 This callback is useful when you don't want to put all of your write data
157 into the queue at once, for example, when you want to write the contents
158 of some file to the socket you might not want to read the whole file into
159 memory and push it into the queue, but instead only read more data from
160 the file when the write queue becomes empty.
161
162 =item timeout => $fractional_seconds
163
164 If non-zero, then this enables an "inactivity" timeout: whenever this many
165 seconds pass without a successful read or write on the underlying file
166 handle, the C<on_timeout> callback will be invoked (and if that one is
167 missing, a non-fatal C<ETIMEDOUT> error will be raised).
168
169 Note that timeout processing is also active when you currently do not have
170 any outstanding read or write requests: If you plan to keep the connection
171 idle then you should disable the timout temporarily or ignore the timeout
172 in the C<on_timeout> callback, in which case AnyEvent::Handle will simply
173 restart the timeout.
174
175 Zero (the default) disables this timeout.
176
177 =item on_timeout => $cb->($handle)
178
179 Called whenever the inactivity timeout passes. If you return from this
180 callback, then the timeout will be reset as if some activity had happened,
181 so this condition is not fatal in any way.
182
183 =item rbuf_max => <bytes>
184
185 If defined, then a fatal error will be raised (with C<$!> set to C<ENOSPC>)
186 when the read buffer ever (strictly) exceeds this size. This is useful to
187 avoid some forms of denial-of-service attacks.
188
189 For example, a server accepting connections from untrusted sources should
190 be configured to accept only so-and-so much data that it cannot act on
191 (for example, when expecting a line, an attacker could send an unlimited
192 amount of data without a callback ever being called as long as the line
193 isn't finished).
194
195 =item autocork => <boolean>
196
197 When disabled (the default), then C<push_write> will try to immediately
198 write the data to the handle, if possible. This avoids having to register
199 a write watcher and wait for the next event loop iteration, but can
200 be inefficient if you write multiple small chunks (on the wire, this
201 disadvantage is usually avoided by your kernel's nagle algorithm, see
202 C<no_delay>, but this option can save costly syscalls).
203
204 When enabled, then writes will always be queued till the next event loop
205 iteration. This is efficient when you do many small writes per iteration,
206 but less efficient when you do a single write only per iteration (or when
207 the write buffer often is full). It also increases write latency.
208
209 =item no_delay => <boolean>
210
211 When doing small writes on sockets, your operating system kernel might
212 wait a bit for more data before actually sending it out. This is called
213 the Nagle algorithm, and usually it is beneficial.
214
215 In some situations you want as low a delay as possible, which can be
216 accomplishd by setting this option to a true value.
217
218 The default is your opertaing system's default behaviour (most likely
219 enabled), this option explicitly enables or disables it, if possible.
220
221 =item read_size => <bytes>
222
223 The default read block size (the amount of bytes this module will
224 try to read during each loop iteration, which affects memory
225 requirements). Default: C<8192>.
226
227 =item low_water_mark => <bytes>
228
229 Sets the amount of bytes (default: C<0>) that make up an "empty" write
230 buffer: If the write reaches this size or gets even samller it is
231 considered empty.
232
233 Sometimes it can be beneficial (for performance reasons) to add data to
234 the write buffer before it is fully drained, but this is a rare case, as
235 the operating system kernel usually buffers data as well, so the default
236 is good in almost all cases.
237
238 =item linger => <seconds>
239
240 If non-zero (default: C<3600>), then the destructor of the
241 AnyEvent::Handle object will check whether there is still outstanding
242 write data and will install a watcher that will write this data to the
243 socket. No errors will be reported (this mostly matches how the operating
244 system treats outstanding data at socket close time).
245
246 This will not work for partial TLS data that could not be encoded
247 yet. This data will be lost. Calling the C<stoptls> method in time might
248 help.
249
250 =item peername => $string
251
252 A string used to identify the remote site - usually the DNS hostname
253 (I<not> IDN!) used to create the connection, rarely the IP address.
254
255 Apart from being useful in error messages, this string is also used in TLS
256 peername verification (see C<verify_peername> in L<AnyEvent::TLS>). This
257 verification will be skipped when C<peername> is not specified or
258 C<undef>.
259
260 =item tls => "accept" | "connect" | Net::SSLeay::SSL object
261
262 When this parameter is given, it enables TLS (SSL) mode, that means
263 AnyEvent will start a TLS handshake as soon as the conenction has been
264 established and will transparently encrypt/decrypt data afterwards.
265
266 All TLS protocol errors will be signalled as C<EPROTO>, with an
267 appropriate error message.
268
269 TLS mode requires Net::SSLeay to be installed (it will be loaded
270 automatically when you try to create a TLS handle): this module doesn't
271 have a dependency on that module, so if your module requires it, you have
272 to add the dependency yourself.
273
274 Unlike TCP, TLS has a server and client side: for the TLS server side, use
275 C<accept>, and for the TLS client side of a connection, use C<connect>
276 mode.
277
278 You can also provide your own TLS connection object, but you have
279 to make sure that you call either C<Net::SSLeay::set_connect_state>
280 or C<Net::SSLeay::set_accept_state> on it before you pass it to
281 AnyEvent::Handle. Also, this module will take ownership of this connection
282 object.
283
284 At some future point, AnyEvent::Handle might switch to another TLS
285 implementation, then the option to use your own session object will go
286 away.
287
288 B<IMPORTANT:> since Net::SSLeay "objects" are really only integers,
289 passing in the wrong integer will lead to certain crash. This most often
290 happens when one uses a stylish C<< tls => 1 >> and is surprised about the
291 segmentation fault.
292
293 See the C<< ->starttls >> method for when need to start TLS negotiation later.
294
295 =item tls_ctx => $anyevent_tls
296
297 Use the given C<AnyEvent::TLS> object to create the new TLS connection
298 (unless a connection object was specified directly). If this parameter is
299 missing, then AnyEvent::Handle will use C<AnyEvent::Handle::TLS_CTX>.
300
301 Instead of an object, you can also specify a hash reference with C<< key
302 => value >> pairs. Those will be passed to L<AnyEvent::TLS> to create a
303 new TLS context object.
304
305 =item on_starttls => $cb->($handle, $success[, $error_message])
306
307 This callback will be invoked when the TLS/SSL handshake has finished. If
308 C<$success> is true, then the TLS handshake succeeded, otherwise it failed
309 (C<on_stoptls> will not be called in this case).
310
311 The session in C<< $handle->{tls} >> can still be examined in this
312 callback, even when the handshake was not successful.
313
314 TLS handshake failures will not cause C<on_error> to be invoked when this
315 callback is in effect, instead, the error message will be passed to C<on_starttls>.
316
317 Without this callback, handshake failures lead to C<on_error> being
318 called, as normal.
319
320 Note that you cannot call C<starttls> right again in this callback. If you
321 need to do that, start an zero-second timer instead whose callback can
322 then call C<< ->starttls >> again.
323
324 =item on_stoptls => $cb->($handle)
325
326 When a SSLv3/TLS shutdown/close notify/EOF is detected and this callback is
327 set, then it will be invoked after freeing the TLS session. If it is not,
328 then a TLS shutdown condition will be treated like a normal EOF condition
329 on the handle.
330
331 The session in C<< $handle->{tls} >> can still be examined in this
332 callback.
333
334 This callback will only be called on TLS shutdowns, not when the
335 underlying handle signals EOF.
336
337 =item json => JSON or JSON::XS object
338
339 This is the json coder object used by the C<json> read and write types.
340
341 If you don't supply it, then AnyEvent::Handle will create and use a
342 suitable one (on demand), which will write and expect UTF-8 encoded JSON
343 texts.
344
345 Note that you are responsible to depend on the JSON module if you want to
346 use this functionality, as AnyEvent does not have a dependency itself.
347
348 =back
349
350 =cut
351
352 sub new {
353 my $class = shift;
354 my $self = bless { @_ }, $class;
355
356 $self->{fh} or Carp::croak "mandatory argument fh is missing";
357
358 AnyEvent::Util::fh_nonblocking $self->{fh}, 1;
359
360 $self->{_activity} = AnyEvent->now;
361 $self->_timeout;
362
363 $self->no_delay (delete $self->{no_delay}) if exists $self->{no_delay};
364
365 $self->starttls (delete $self->{tls}, delete $self->{tls_ctx})
366 if $self->{tls};
367
368 $self->on_drain (delete $self->{on_drain}) if $self->{on_drain};
369
370 $self->start_read
371 if $self->{on_read};
372
373 $self->{fh} && $self
374 }
375
376 #sub _shutdown {
377 # my ($self) = @_;
378 #
379 # delete @$self{qw(_tw _rw _ww fh wbuf on_read _queue)};
380 # $self->{_eof} = 1; # tell starttls et. al to stop trying
381 #
382 # &_freetls;
383 #}
384
385 sub _error {
386 my ($self, $errno, $fatal, $message) = @_;
387
388 $! = $errno;
389 $message ||= "$!";
390
391 if ($self->{on_error}) {
392 $self->{on_error}($self, $fatal, $message);
393 $self->destroy;
394 } elsif ($self->{fh}) {
395 $self->destroy;
396 Carp::croak "AnyEvent::Handle uncaught error: $message";
397 }
398 }
399
400 =item $fh = $handle->fh
401
402 This method returns the file handle used to create the L<AnyEvent::Handle> object.
403
404 =cut
405
406 sub fh { $_[0]{fh} }
407
408 =item $handle->on_error ($cb)
409
410 Replace the current C<on_error> callback (see the C<on_error> constructor argument).
411
412 =cut
413
414 sub on_error {
415 $_[0]{on_error} = $_[1];
416 }
417
418 =item $handle->on_eof ($cb)
419
420 Replace the current C<on_eof> callback (see the C<on_eof> constructor argument).
421
422 =cut
423
424 sub on_eof {
425 $_[0]{on_eof} = $_[1];
426 }
427
428 =item $handle->on_timeout ($cb)
429
430 Replace the current C<on_timeout> callback, or disables the callback (but
431 not the timeout) if C<$cb> = C<undef>. See the C<timeout> constructor
432 argument and method.
433
434 =cut
435
436 sub on_timeout {
437 $_[0]{on_timeout} = $_[1];
438 }
439
440 =item $handle->autocork ($boolean)
441
442 Enables or disables the current autocork behaviour (see C<autocork>
443 constructor argument). Changes will only take effect on the next write.
444
445 =cut
446
447 sub autocork {
448 $_[0]{autocork} = $_[1];
449 }
450
451 =item $handle->no_delay ($boolean)
452
453 Enables or disables the C<no_delay> setting (see constructor argument of
454 the same name for details).
455
456 =cut
457
458 sub no_delay {
459 $_[0]{no_delay} = $_[1];
460
461 eval {
462 local $SIG{__DIE__};
463 setsockopt $_[0]{fh}, &Socket::IPPROTO_TCP, &Socket::TCP_NODELAY, int $_[1];
464 };
465 }
466
467 =item $handle->on_starttls ($cb)
468
469 Replace the current C<on_starttls> callback (see the C<on_starttls> constructor argument).
470
471 =cut
472
473 sub on_starttls {
474 $_[0]{on_starttls} = $_[1];
475 }
476
477 =item $handle->on_stoptls ($cb)
478
479 Replace the current C<on_stoptls> callback (see the C<on_stoptls> constructor argument).
480
481 =cut
482
483 sub on_starttls {
484 $_[0]{on_stoptls} = $_[1];
485 }
486
487 #############################################################################
488
489 =item $handle->timeout ($seconds)
490
491 Configures (or disables) the inactivity timeout.
492
493 =cut
494
495 sub timeout {
496 my ($self, $timeout) = @_;
497
498 $self->{timeout} = $timeout;
499 $self->_timeout;
500 }
501
502 # reset the timeout watcher, as neccessary
503 # also check for time-outs
504 sub _timeout {
505 my ($self) = @_;
506
507 if ($self->{timeout}) {
508 my $NOW = AnyEvent->now;
509
510 # when would the timeout trigger?
511 my $after = $self->{_activity} + $self->{timeout} - $NOW;
512
513 # now or in the past already?
514 if ($after <= 0) {
515 $self->{_activity} = $NOW;
516
517 if ($self->{on_timeout}) {
518 $self->{on_timeout}($self);
519 } else {
520 $self->_error (Errno::ETIMEDOUT);
521 }
522
523 # callback could have changed timeout value, optimise
524 return unless $self->{timeout};
525
526 # calculate new after
527 $after = $self->{timeout};
528 }
529
530 Scalar::Util::weaken $self;
531 return unless $self; # ->error could have destroyed $self
532
533 $self->{_tw} ||= AnyEvent->timer (after => $after, cb => sub {
534 delete $self->{_tw};
535 $self->_timeout;
536 });
537 } else {
538 delete $self->{_tw};
539 }
540 }
541
542 #############################################################################
543
544 =back
545
546 =head2 WRITE QUEUE
547
548 AnyEvent::Handle manages two queues per handle, one for writing and one
549 for reading.
550
551 The write queue is very simple: you can add data to its end, and
552 AnyEvent::Handle will automatically try to get rid of it for you.
553
554 When data could be written and the write buffer is shorter then the low
555 water mark, the C<on_drain> callback will be invoked.
556
557 =over 4
558
559 =item $handle->on_drain ($cb)
560
561 Sets the C<on_drain> callback or clears it (see the description of
562 C<on_drain> in the constructor).
563
564 =cut
565
566 sub on_drain {
567 my ($self, $cb) = @_;
568
569 $self->{on_drain} = $cb;
570
571 $cb->($self)
572 if $cb && $self->{low_water_mark} >= (length $self->{wbuf}) + (length $self->{_tls_wbuf});
573 }
574
575 =item $handle->push_write ($data)
576
577 Queues the given scalar to be written. You can push as much data as you
578 want (only limited by the available memory), as C<AnyEvent::Handle>
579 buffers it independently of the kernel.
580
581 =cut
582
583 sub _drain_wbuf {
584 my ($self) = @_;
585
586 if (!$self->{_ww} && length $self->{wbuf}) {
587
588 Scalar::Util::weaken $self;
589
590 my $cb = sub {
591 my $len = syswrite $self->{fh}, $self->{wbuf};
592
593 if (defined $len) {
594 substr $self->{wbuf}, 0, $len, "";
595
596 $self->{_activity} = AnyEvent->now;
597
598 $self->{on_drain}($self)
599 if $self->{low_water_mark} >= (length $self->{wbuf}) + (length $self->{_tls_wbuf})
600 && $self->{on_drain};
601
602 delete $self->{_ww} unless length $self->{wbuf};
603 } elsif ($! != EAGAIN && $! != EINTR && $! != WSAEWOULDBLOCK) {
604 $self->_error ($!, 1);
605 }
606 };
607
608 # try to write data immediately
609 $cb->() unless $self->{autocork};
610
611 # if still data left in wbuf, we need to poll
612 $self->{_ww} = AnyEvent->io (fh => $self->{fh}, poll => "w", cb => $cb)
613 if length $self->{wbuf};
614 };
615 }
616
617 our %WH;
618
619 sub register_write_type($$) {
620 $WH{$_[0]} = $_[1];
621 }
622
623 sub push_write {
624 my $self = shift;
625
626 if (@_ > 1) {
627 my $type = shift;
628
629 @_ = ($WH{$type} or Carp::croak "unsupported type passed to AnyEvent::Handle::push_write")
630 ->($self, @_);
631 }
632
633 if ($self->{tls}) {
634 $self->{_tls_wbuf} .= $_[0];
635
636 &_dotls ($self);
637 } else {
638 $self->{wbuf} .= $_[0];
639 $self->_drain_wbuf;
640 }
641 }
642
643 =item $handle->push_write (type => @args)
644
645 Instead of formatting your data yourself, you can also let this module do
646 the job by specifying a type and type-specific arguments.
647
648 Predefined types are (if you have ideas for additional types, feel free to
649 drop by and tell us):
650
651 =over 4
652
653 =item netstring => $string
654
655 Formats the given value as netstring
656 (http://cr.yp.to/proto/netstrings.txt, this is not a recommendation to use them).
657
658 =cut
659
660 register_write_type netstring => sub {
661 my ($self, $string) = @_;
662
663 (length $string) . ":$string,"
664 };
665
666 =item packstring => $format, $data
667
668 An octet string prefixed with an encoded length. The encoding C<$format>
669 uses the same format as a Perl C<pack> format, but must specify a single
670 integer only (only one of C<cCsSlLqQiInNvVjJw> is allowed, plus an
671 optional C<!>, C<< < >> or C<< > >> modifier).
672
673 =cut
674
675 register_write_type packstring => sub {
676 my ($self, $format, $string) = @_;
677
678 pack "$format/a*", $string
679 };
680
681 =item json => $array_or_hashref
682
683 Encodes the given hash or array reference into a JSON object. Unless you
684 provide your own JSON object, this means it will be encoded to JSON text
685 in UTF-8.
686
687 JSON objects (and arrays) are self-delimiting, so you can write JSON at
688 one end of a handle and read them at the other end without using any
689 additional framing.
690
691 The generated JSON text is guaranteed not to contain any newlines: While
692 this module doesn't need delimiters after or between JSON texts to be
693 able to read them, many other languages depend on that.
694
695 A simple RPC protocol that interoperates easily with others is to send
696 JSON arrays (or objects, although arrays are usually the better choice as
697 they mimic how function argument passing works) and a newline after each
698 JSON text:
699
700 $handle->push_write (json => ["method", "arg1", "arg2"]); # whatever
701 $handle->push_write ("\012");
702
703 An AnyEvent::Handle receiver would simply use the C<json> read type and
704 rely on the fact that the newline will be skipped as leading whitespace:
705
706 $handle->push_read (json => sub { my $array = $_[1]; ... });
707
708 Other languages could read single lines terminated by a newline and pass
709 this line into their JSON decoder of choice.
710
711 =cut
712
713 register_write_type json => sub {
714 my ($self, $ref) = @_;
715
716 require JSON;
717
718 $self->{json} ? $self->{json}->encode ($ref)
719 : JSON::encode_json ($ref)
720 };
721
722 =item storable => $reference
723
724 Freezes the given reference using L<Storable> and writes it to the
725 handle. Uses the C<nfreeze> format.
726
727 =cut
728
729 register_write_type storable => sub {
730 my ($self, $ref) = @_;
731
732 require Storable;
733
734 pack "w/a*", Storable::nfreeze ($ref)
735 };
736
737 =back
738
739 =item $handle->push_shutdown
740
741 Sometimes you know you want to close the socket after writing your data
742 before it was actually written. One way to do that is to replace your
743 C<on_drain> handler by a callback that shuts down the socket (and set
744 C<low_water_mark> to C<0>). This method is a shorthand for just that, and
745 replaces the C<on_drain> callback with:
746
747 sub { shutdown $_[0]{fh}, 1 } # for push_shutdown
748
749 This simply shuts down the write side and signals an EOF condition to the
750 the peer.
751
752 You can rely on the normal read queue and C<on_eof> handling
753 afterwards. This is the cleanest way to close a connection.
754
755 =cut
756
757 sub push_shutdown {
758 my ($self) = @_;
759
760 delete $self->{low_water_mark};
761 $self->on_drain (sub { shutdown $_[0]{fh}, 1 });
762 }
763
764 =item AnyEvent::Handle::register_write_type type => $coderef->($handle, @args)
765
766 This function (not method) lets you add your own types to C<push_write>.
767 Whenever the given C<type> is used, C<push_write> will invoke the code
768 reference with the handle object and the remaining arguments.
769
770 The code reference is supposed to return a single octet string that will
771 be appended to the write buffer.
772
773 Note that this is a function, and all types registered this way will be
774 global, so try to use unique names.
775
776 =cut
777
778 #############################################################################
779
780 =back
781
782 =head2 READ QUEUE
783
784 AnyEvent::Handle manages two queues per handle, one for writing and one
785 for reading.
786
787 The read queue is more complex than the write queue. It can be used in two
788 ways, the "simple" way, using only C<on_read> and the "complex" way, using
789 a queue.
790
791 In the simple case, you just install an C<on_read> callback and whenever
792 new data arrives, it will be called. You can then remove some data (if
793 enough is there) from the read buffer (C<< $handle->rbuf >>). Or you cna
794 leave the data there if you want to accumulate more (e.g. when only a
795 partial message has been received so far).
796
797 In the more complex case, you want to queue multiple callbacks. In this
798 case, AnyEvent::Handle will call the first queued callback each time new
799 data arrives (also the first time it is queued) and removes it when it has
800 done its job (see C<push_read>, below).
801
802 This way you can, for example, push three line-reads, followed by reading
803 a chunk of data, and AnyEvent::Handle will execute them in order.
804
805 Example 1: EPP protocol parser. EPP sends 4 byte length info, followed by
806 the specified number of bytes which give an XML datagram.
807
808 # in the default state, expect some header bytes
809 $handle->on_read (sub {
810 # some data is here, now queue the length-header-read (4 octets)
811 shift->unshift_read (chunk => 4, sub {
812 # header arrived, decode
813 my $len = unpack "N", $_[1];
814
815 # now read the payload
816 shift->unshift_read (chunk => $len, sub {
817 my $xml = $_[1];
818 # handle xml
819 });
820 });
821 });
822
823 Example 2: Implement a client for a protocol that replies either with "OK"
824 and another line or "ERROR" for the first request that is sent, and 64
825 bytes for the second request. Due to the availability of a queue, we can
826 just pipeline sending both requests and manipulate the queue as necessary
827 in the callbacks.
828
829 When the first callback is called and sees an "OK" response, it will
830 C<unshift> another line-read. This line-read will be queued I<before> the
831 64-byte chunk callback.
832
833 # request one, returns either "OK + extra line" or "ERROR"
834 $handle->push_write ("request 1\015\012");
835
836 # we expect "ERROR" or "OK" as response, so push a line read
837 $handle->push_read (line => sub {
838 # if we got an "OK", we have to _prepend_ another line,
839 # so it will be read before the second request reads its 64 bytes
840 # which are already in the queue when this callback is called
841 # we don't do this in case we got an error
842 if ($_[1] eq "OK") {
843 $_[0]->unshift_read (line => sub {
844 my $response = $_[1];
845 ...
846 });
847 }
848 });
849
850 # request two, simply returns 64 octets
851 $handle->push_write ("request 2\015\012");
852
853 # simply read 64 bytes, always
854 $handle->push_read (chunk => 64, sub {
855 my $response = $_[1];
856 ...
857 });
858
859 =over 4
860
861 =cut
862
863 sub _drain_rbuf {
864 my ($self) = @_;
865
866 local $self->{_in_drain} = 1;
867
868 if (
869 defined $self->{rbuf_max}
870 && $self->{rbuf_max} < length $self->{rbuf}
871 ) {
872 $self->_error (Errno::ENOSPC, 1), return;
873 }
874
875 while () {
876 # we need to use a separate tls read buffer, as we must not receive data while
877 # we are draining the buffer, and this can only happen with TLS.
878 $self->{rbuf} .= delete $self->{_tls_rbuf} if exists $self->{_tls_rbuf};
879
880 my $len = length $self->{rbuf};
881
882 if (my $cb = shift @{ $self->{_queue} }) {
883 unless ($cb->($self)) {
884 if ($self->{_eof}) {
885 # no progress can be made (not enough data and no data forthcoming)
886 $self->_error (Errno::EPIPE, 1), return;
887 }
888
889 unshift @{ $self->{_queue} }, $cb;
890 last;
891 }
892 } elsif ($self->{on_read}) {
893 last unless $len;
894
895 $self->{on_read}($self);
896
897 if (
898 $len == length $self->{rbuf} # if no data has been consumed
899 && !@{ $self->{_queue} } # and the queue is still empty
900 && $self->{on_read} # but we still have on_read
901 ) {
902 # no further data will arrive
903 # so no progress can be made
904 $self->_error (Errno::EPIPE, 1), return
905 if $self->{_eof};
906
907 last; # more data might arrive
908 }
909 } else {
910 # read side becomes idle
911 delete $self->{_rw} unless $self->{tls};
912 last;
913 }
914 }
915
916 if ($self->{_eof}) {
917 if ($self->{on_eof}) {
918 $self->{on_eof}($self)
919 } else {
920 $self->_error (0, 1, "Unexpected end-of-file");
921 }
922 }
923
924 # may need to restart read watcher
925 unless ($self->{_rw}) {
926 $self->start_read
927 if $self->{on_read} || @{ $self->{_queue} };
928 }
929 }
930
931 =item $handle->on_read ($cb)
932
933 This replaces the currently set C<on_read> callback, or clears it (when
934 the new callback is C<undef>). See the description of C<on_read> in the
935 constructor.
936
937 =cut
938
939 sub on_read {
940 my ($self, $cb) = @_;
941
942 $self->{on_read} = $cb;
943 $self->_drain_rbuf if $cb && !$self->{_in_drain};
944 }
945
946 =item $handle->rbuf
947
948 Returns the read buffer (as a modifiable lvalue).
949
950 You can access the read buffer directly as the C<< ->{rbuf} >>
951 member, if you want. However, the only operation allowed on the
952 read buffer (apart from looking at it) is removing data from its
953 beginning. Otherwise modifying or appending to it is not allowed and will
954 lead to hard-to-track-down bugs.
955
956 NOTE: The read buffer should only be used or modified if the C<on_read>,
957 C<push_read> or C<unshift_read> methods are used. The other read methods
958 automatically manage the read buffer.
959
960 =cut
961
962 sub rbuf : lvalue {
963 $_[0]{rbuf}
964 }
965
966 =item $handle->push_read ($cb)
967
968 =item $handle->unshift_read ($cb)
969
970 Append the given callback to the end of the queue (C<push_read>) or
971 prepend it (C<unshift_read>).
972
973 The callback is called each time some additional read data arrives.
974
975 It must check whether enough data is in the read buffer already.
976
977 If not enough data is available, it must return the empty list or a false
978 value, in which case it will be called repeatedly until enough data is
979 available (or an error condition is detected).
980
981 If enough data was available, then the callback must remove all data it is
982 interested in (which can be none at all) and return a true value. After returning
983 true, it will be removed from the queue.
984
985 =cut
986
987 our %RH;
988
989 sub register_read_type($$) {
990 $RH{$_[0]} = $_[1];
991 }
992
993 sub push_read {
994 my $self = shift;
995 my $cb = pop;
996
997 if (@_) {
998 my $type = shift;
999
1000 $cb = ($RH{$type} or Carp::croak "unsupported type passed to AnyEvent::Handle::push_read")
1001 ->($self, $cb, @_);
1002 }
1003
1004 push @{ $self->{_queue} }, $cb;
1005 $self->_drain_rbuf unless $self->{_in_drain};
1006 }
1007
1008 sub unshift_read {
1009 my $self = shift;
1010 my $cb = pop;
1011
1012 if (@_) {
1013 my $type = shift;
1014
1015 $cb = ($RH{$type} or Carp::croak "unsupported type passed to AnyEvent::Handle::unshift_read")
1016 ->($self, $cb, @_);
1017 }
1018
1019
1020 unshift @{ $self->{_queue} }, $cb;
1021 $self->_drain_rbuf unless $self->{_in_drain};
1022 }
1023
1024 =item $handle->push_read (type => @args, $cb)
1025
1026 =item $handle->unshift_read (type => @args, $cb)
1027
1028 Instead of providing a callback that parses the data itself you can chose
1029 between a number of predefined parsing formats, for chunks of data, lines
1030 etc.
1031
1032 Predefined types are (if you have ideas for additional types, feel free to
1033 drop by and tell us):
1034
1035 =over 4
1036
1037 =item chunk => $octets, $cb->($handle, $data)
1038
1039 Invoke the callback only once C<$octets> bytes have been read. Pass the
1040 data read to the callback. The callback will never be called with less
1041 data.
1042
1043 Example: read 2 bytes.
1044
1045 $handle->push_read (chunk => 2, sub {
1046 warn "yay ", unpack "H*", $_[1];
1047 });
1048
1049 =cut
1050
1051 register_read_type chunk => sub {
1052 my ($self, $cb, $len) = @_;
1053
1054 sub {
1055 $len <= length $_[0]{rbuf} or return;
1056 $cb->($_[0], substr $_[0]{rbuf}, 0, $len, "");
1057 1
1058 }
1059 };
1060
1061 =item line => [$eol, ]$cb->($handle, $line, $eol)
1062
1063 The callback will be called only once a full line (including the end of
1064 line marker, C<$eol>) has been read. This line (excluding the end of line
1065 marker) will be passed to the callback as second argument (C<$line>), and
1066 the end of line marker as the third argument (C<$eol>).
1067
1068 The end of line marker, C<$eol>, can be either a string, in which case it
1069 will be interpreted as a fixed record end marker, or it can be a regex
1070 object (e.g. created by C<qr>), in which case it is interpreted as a
1071 regular expression.
1072
1073 The end of line marker argument C<$eol> is optional, if it is missing (NOT
1074 undef), then C<qr|\015?\012|> is used (which is good for most internet
1075 protocols).
1076
1077 Partial lines at the end of the stream will never be returned, as they are
1078 not marked by the end of line marker.
1079
1080 =cut
1081
1082 register_read_type line => sub {
1083 my ($self, $cb, $eol) = @_;
1084
1085 if (@_ < 3) {
1086 # this is more than twice as fast as the generic code below
1087 sub {
1088 $_[0]{rbuf} =~ s/^([^\015\012]*)(\015?\012)// or return;
1089
1090 $cb->($_[0], $1, $2);
1091 1
1092 }
1093 } else {
1094 $eol = quotemeta $eol unless ref $eol;
1095 $eol = qr|^(.*?)($eol)|s;
1096
1097 sub {
1098 $_[0]{rbuf} =~ s/$eol// or return;
1099
1100 $cb->($_[0], $1, $2);
1101 1
1102 }
1103 }
1104 };
1105
1106 =item regex => $accept[, $reject[, $skip], $cb->($handle, $data)
1107
1108 Makes a regex match against the regex object C<$accept> and returns
1109 everything up to and including the match.
1110
1111 Example: read a single line terminated by '\n'.
1112
1113 $handle->push_read (regex => qr<\n>, sub { ... });
1114
1115 If C<$reject> is given and not undef, then it determines when the data is
1116 to be rejected: it is matched against the data when the C<$accept> regex
1117 does not match and generates an C<EBADMSG> error when it matches. This is
1118 useful to quickly reject wrong data (to avoid waiting for a timeout or a
1119 receive buffer overflow).
1120
1121 Example: expect a single decimal number followed by whitespace, reject
1122 anything else (not the use of an anchor).
1123
1124 $handle->push_read (regex => qr<^[0-9]+\s>, qr<[^0-9]>, sub { ... });
1125
1126 If C<$skip> is given and not C<undef>, then it will be matched against
1127 the receive buffer when neither C<$accept> nor C<$reject> match,
1128 and everything preceding and including the match will be accepted
1129 unconditionally. This is useful to skip large amounts of data that you
1130 know cannot be matched, so that the C<$accept> or C<$reject> regex do not
1131 have to start matching from the beginning. This is purely an optimisation
1132 and is usually worth only when you expect more than a few kilobytes.
1133
1134 Example: expect a http header, which ends at C<\015\012\015\012>. Since we
1135 expect the header to be very large (it isn't in practise, but...), we use
1136 a skip regex to skip initial portions. The skip regex is tricky in that
1137 it only accepts something not ending in either \015 or \012, as these are
1138 required for the accept regex.
1139
1140 $handle->push_read (regex =>
1141 qr<\015\012\015\012>,
1142 undef, # no reject
1143 qr<^.*[^\015\012]>,
1144 sub { ... });
1145
1146 =cut
1147
1148 register_read_type regex => sub {
1149 my ($self, $cb, $accept, $reject, $skip) = @_;
1150
1151 my $data;
1152 my $rbuf = \$self->{rbuf};
1153
1154 sub {
1155 # accept
1156 if ($$rbuf =~ $accept) {
1157 $data .= substr $$rbuf, 0, $+[0], "";
1158 $cb->($self, $data);
1159 return 1;
1160 }
1161
1162 # reject
1163 if ($reject && $$rbuf =~ $reject) {
1164 $self->_error (Errno::EBADMSG);
1165 }
1166
1167 # skip
1168 if ($skip && $$rbuf =~ $skip) {
1169 $data .= substr $$rbuf, 0, $+[0], "";
1170 }
1171
1172 ()
1173 }
1174 };
1175
1176 =item netstring => $cb->($handle, $string)
1177
1178 A netstring (http://cr.yp.to/proto/netstrings.txt, this is not an endorsement).
1179
1180 Throws an error with C<$!> set to EBADMSG on format violations.
1181
1182 =cut
1183
1184 register_read_type netstring => sub {
1185 my ($self, $cb) = @_;
1186
1187 sub {
1188 unless ($_[0]{rbuf} =~ s/^(0|[1-9][0-9]*)://) {
1189 if ($_[0]{rbuf} =~ /[^0-9]/) {
1190 $self->_error (Errno::EBADMSG);
1191 }
1192 return;
1193 }
1194
1195 my $len = $1;
1196
1197 $self->unshift_read (chunk => $len, sub {
1198 my $string = $_[1];
1199 $_[0]->unshift_read (chunk => 1, sub {
1200 if ($_[1] eq ",") {
1201 $cb->($_[0], $string);
1202 } else {
1203 $self->_error (Errno::EBADMSG);
1204 }
1205 });
1206 });
1207
1208 1
1209 }
1210 };
1211
1212 =item packstring => $format, $cb->($handle, $string)
1213
1214 An octet string prefixed with an encoded length. The encoding C<$format>
1215 uses the same format as a Perl C<pack> format, but must specify a single
1216 integer only (only one of C<cCsSlLqQiInNvVjJw> is allowed, plus an
1217 optional C<!>, C<< < >> or C<< > >> modifier).
1218
1219 For example, DNS over TCP uses a prefix of C<n> (2 octet network order),
1220 EPP uses a prefix of C<N> (4 octtes).
1221
1222 Example: read a block of data prefixed by its length in BER-encoded
1223 format (very efficient).
1224
1225 $handle->push_read (packstring => "w", sub {
1226 my ($handle, $data) = @_;
1227 });
1228
1229 =cut
1230
1231 register_read_type packstring => sub {
1232 my ($self, $cb, $format) = @_;
1233
1234 sub {
1235 # when we can use 5.10 we can use ".", but for 5.8 we use the re-pack method
1236 defined (my $len = eval { unpack $format, $_[0]{rbuf} })
1237 or return;
1238
1239 $format = length pack $format, $len;
1240
1241 # bypass unshift if we already have the remaining chunk
1242 if ($format + $len <= length $_[0]{rbuf}) {
1243 my $data = substr $_[0]{rbuf}, $format, $len;
1244 substr $_[0]{rbuf}, 0, $format + $len, "";
1245 $cb->($_[0], $data);
1246 } else {
1247 # remove prefix
1248 substr $_[0]{rbuf}, 0, $format, "";
1249
1250 # read remaining chunk
1251 $_[0]->unshift_read (chunk => $len, $cb);
1252 }
1253
1254 1
1255 }
1256 };
1257
1258 =item json => $cb->($handle, $hash_or_arrayref)
1259
1260 Reads a JSON object or array, decodes it and passes it to the
1261 callback. When a parse error occurs, an C<EBADMSG> error will be raised.
1262
1263 If a C<json> object was passed to the constructor, then that will be used
1264 for the final decode, otherwise it will create a JSON coder expecting UTF-8.
1265
1266 This read type uses the incremental parser available with JSON version
1267 2.09 (and JSON::XS version 2.2) and above. You have to provide a
1268 dependency on your own: this module will load the JSON module, but
1269 AnyEvent does not depend on it itself.
1270
1271 Since JSON texts are fully self-delimiting, the C<json> read and write
1272 types are an ideal simple RPC protocol: just exchange JSON datagrams. See
1273 the C<json> write type description, above, for an actual example.
1274
1275 =cut
1276
1277 register_read_type json => sub {
1278 my ($self, $cb) = @_;
1279
1280 my $json = $self->{json} ||=
1281 eval { require JSON::XS; JSON::XS->new->utf8 }
1282 || do { require JSON; JSON->new->utf8 };
1283
1284 my $data;
1285 my $rbuf = \$self->{rbuf};
1286
1287 sub {
1288 my $ref = eval { $json->incr_parse ($self->{rbuf}) };
1289
1290 if ($ref) {
1291 $self->{rbuf} = $json->incr_text;
1292 $json->incr_text = "";
1293 $cb->($self, $ref);
1294
1295 1
1296 } elsif ($@) {
1297 # error case
1298 $json->incr_skip;
1299
1300 $self->{rbuf} = $json->incr_text;
1301 $json->incr_text = "";
1302
1303 $self->_error (Errno::EBADMSG);
1304
1305 ()
1306 } else {
1307 $self->{rbuf} = "";
1308
1309 ()
1310 }
1311 }
1312 };
1313
1314 =item storable => $cb->($handle, $ref)
1315
1316 Deserialises a L<Storable> frozen representation as written by the
1317 C<storable> write type (BER-encoded length prefix followed by nfreeze'd
1318 data).
1319
1320 Raises C<EBADMSG> error if the data could not be decoded.
1321
1322 =cut
1323
1324 register_read_type storable => sub {
1325 my ($self, $cb) = @_;
1326
1327 require Storable;
1328
1329 sub {
1330 # when we can use 5.10 we can use ".", but for 5.8 we use the re-pack method
1331 defined (my $len = eval { unpack "w", $_[0]{rbuf} })
1332 or return;
1333
1334 my $format = length pack "w", $len;
1335
1336 # bypass unshift if we already have the remaining chunk
1337 if ($format + $len <= length $_[0]{rbuf}) {
1338 my $data = substr $_[0]{rbuf}, $format, $len;
1339 substr $_[0]{rbuf}, 0, $format + $len, "";
1340 $cb->($_[0], Storable::thaw ($data));
1341 } else {
1342 # remove prefix
1343 substr $_[0]{rbuf}, 0, $format, "";
1344
1345 # read remaining chunk
1346 $_[0]->unshift_read (chunk => $len, sub {
1347 if (my $ref = eval { Storable::thaw ($_[1]) }) {
1348 $cb->($_[0], $ref);
1349 } else {
1350 $self->_error (Errno::EBADMSG);
1351 }
1352 });
1353 }
1354
1355 1
1356 }
1357 };
1358
1359 =back
1360
1361 =item AnyEvent::Handle::register_read_type type => $coderef->($handle, $cb, @args)
1362
1363 This function (not method) lets you add your own types to C<push_read>.
1364
1365 Whenever the given C<type> is used, C<push_read> will invoke the code
1366 reference with the handle object, the callback and the remaining
1367 arguments.
1368
1369 The code reference is supposed to return a callback (usually a closure)
1370 that works as a plain read callback (see C<< ->push_read ($cb) >>).
1371
1372 It should invoke the passed callback when it is done reading (remember to
1373 pass C<$handle> as first argument as all other callbacks do that).
1374
1375 Note that this is a function, and all types registered this way will be
1376 global, so try to use unique names.
1377
1378 For examples, see the source of this module (F<perldoc -m AnyEvent::Handle>,
1379 search for C<register_read_type>)).
1380
1381 =item $handle->stop_read
1382
1383 =item $handle->start_read
1384
1385 In rare cases you actually do not want to read anything from the
1386 socket. In this case you can call C<stop_read>. Neither C<on_read> nor
1387 any queued callbacks will be executed then. To start reading again, call
1388 C<start_read>.
1389
1390 Note that AnyEvent::Handle will automatically C<start_read> for you when
1391 you change the C<on_read> callback or push/unshift a read callback, and it
1392 will automatically C<stop_read> for you when neither C<on_read> is set nor
1393 there are any read requests in the queue.
1394
1395 These methods will have no effect when in TLS mode (as TLS doesn't support
1396 half-duplex connections).
1397
1398 =cut
1399
1400 sub stop_read {
1401 my ($self) = @_;
1402
1403 delete $self->{_rw} unless $self->{tls};
1404 }
1405
1406 sub start_read {
1407 my ($self) = @_;
1408
1409 unless ($self->{_rw} || $self->{_eof}) {
1410 Scalar::Util::weaken $self;
1411
1412 $self->{_rw} = AnyEvent->io (fh => $self->{fh}, poll => "r", cb => sub {
1413 my $rbuf = \($self->{tls} ? my $buf : $self->{rbuf});
1414 my $len = sysread $self->{fh}, $$rbuf, $self->{read_size} || 8192, length $$rbuf;
1415
1416 if ($len > 0) {
1417 $self->{_activity} = AnyEvent->now;
1418
1419 if ($self->{tls}) {
1420 Net::SSLeay::BIO_write ($self->{_rbio}, $$rbuf);
1421
1422 &_dotls ($self);
1423 } else {
1424 $self->_drain_rbuf unless $self->{_in_drain};
1425 }
1426
1427 } elsif (defined $len) {
1428 delete $self->{_rw};
1429 $self->{_eof} = 1;
1430 $self->_drain_rbuf unless $self->{_in_drain};
1431
1432 } elsif ($! != EAGAIN && $! != EINTR && $! != WSAEWOULDBLOCK) {
1433 return $self->_error ($!, 1);
1434 }
1435 });
1436 }
1437 }
1438
1439 our $ERROR_SYSCALL;
1440 our $ERROR_WANT_READ;
1441
1442 sub _tls_error {
1443 my ($self, $err) = @_;
1444
1445 return $self->_error ($!, 1)
1446 if $err == Net::SSLeay::ERROR_SYSCALL ();
1447
1448 my $err =Net::SSLeay::ERR_error_string (Net::SSLeay::ERR_get_error ());
1449
1450 # reduce error string to look less scary
1451 $err =~ s/^error:[0-9a-fA-F]{8}:[^:]+:([^:]+):/\L$1: /;
1452
1453 if ($self->{_on_starttls}) {
1454 (delete $self->{_on_starttls})->($self, undef, $err);
1455 &_freetls;
1456 } else {
1457 &_freetls;
1458 $self->_error (Errno::EPROTO, 1, $err);
1459 }
1460 }
1461
1462 # poll the write BIO and send the data if applicable
1463 # also decode read data if possible
1464 # this is basiclaly our TLS state machine
1465 # more efficient implementations are possible with openssl,
1466 # but not with the buggy and incomplete Net::SSLeay.
1467 sub _dotls {
1468 my ($self) = @_;
1469
1470 my $tmp;
1471
1472 if (length $self->{_tls_wbuf}) {
1473 while (($tmp = Net::SSLeay::write ($self->{tls}, $self->{_tls_wbuf})) > 0) {
1474 substr $self->{_tls_wbuf}, 0, $tmp, "";
1475 }
1476
1477 $tmp = Net::SSLeay::get_error ($self->{tls}, $tmp);
1478 return $self->_tls_error ($tmp)
1479 if $tmp != $ERROR_WANT_READ
1480 && ($tmp != $ERROR_SYSCALL || $!);
1481 }
1482
1483 while (defined ($tmp = Net::SSLeay::read ($self->{tls}))) {
1484 unless (length $tmp) {
1485 $self->{_on_starttls}
1486 and (delete $self->{_on_starttls})->($self, undef, "EOF during handshake"); # ???
1487 &_freetls;
1488
1489 if ($self->{on_stoptls}) {
1490 $self->{on_stoptls}($self);
1491 return;
1492 } else {
1493 # let's treat SSL-eof as we treat normal EOF
1494 delete $self->{_rw};
1495 $self->{_eof} = 1;
1496 }
1497 }
1498
1499 $self->{_tls_rbuf} .= $tmp;
1500 $self->_drain_rbuf unless $self->{_in_drain};
1501 $self->{tls} or return; # tls session might have gone away in callback
1502 }
1503
1504 $tmp = Net::SSLeay::get_error ($self->{tls}, -1);
1505 return $self->_tls_error ($tmp)
1506 if $tmp != $ERROR_WANT_READ
1507 && ($tmp != $ERROR_SYSCALL || $!);
1508
1509 while (length ($tmp = Net::SSLeay::BIO_read ($self->{_wbio}))) {
1510 $self->{wbuf} .= $tmp;
1511 $self->_drain_wbuf;
1512 }
1513
1514 $self->{_on_starttls}
1515 and Net::SSLeay::state ($self->{tls}) == Net::SSLeay::ST_OK ()
1516 and (delete $self->{_on_starttls})->($self, 1, "TLS/SSL connection established");
1517 }
1518
1519 =item $handle->starttls ($tls[, $tls_ctx])
1520
1521 Instead of starting TLS negotiation immediately when the AnyEvent::Handle
1522 object is created, you can also do that at a later time by calling
1523 C<starttls>.
1524
1525 The first argument is the same as the C<tls> constructor argument (either
1526 C<"connect">, C<"accept"> or an existing Net::SSLeay object).
1527
1528 The second argument is the optional C<AnyEvent::TLS> object that is used
1529 when AnyEvent::Handle has to create its own TLS connection object, or
1530 a hash reference with C<< key => value >> pairs that will be used to
1531 construct a new context.
1532
1533 The TLS connection object will end up in C<< $handle->{tls} >>, the TLS
1534 context in C<< $handle->{tls_ctx} >> after this call and can be used or
1535 changed to your liking. Note that the handshake might have already started
1536 when this function returns.
1537
1538 If it an error to start a TLS handshake more than once per
1539 AnyEvent::Handle object (this is due to bugs in OpenSSL).
1540
1541 =cut
1542
1543 our %TLS_CACHE; #TODO not yet documented, should we?
1544
1545 sub starttls {
1546 my ($self, $ssl, $ctx) = @_;
1547
1548 require Net::SSLeay;
1549
1550 Carp::croak "it is an error to call starttls more than once on an AnyEvent::Handle object"
1551 if $self->{tls};
1552
1553 $ERROR_SYSCALL = Net::SSLeay::ERROR_SYSCALL ();
1554 $ERROR_WANT_READ = Net::SSLeay::ERROR_WANT_READ ();
1555
1556 $ctx ||= $self->{tls_ctx};
1557
1558 if ("HASH" eq ref $ctx) {
1559 require AnyEvent::TLS;
1560
1561 local $Carp::CarpLevel = 1; # skip ourselves when creating a new context
1562
1563 if ($ctx->{cache}) {
1564 my $key = $ctx+0;
1565 $ctx = $TLS_CACHE{$key} ||= new AnyEvent::TLS %$ctx;
1566 } else {
1567 $ctx = new AnyEvent::TLS %$ctx;
1568 }
1569 }
1570
1571 $self->{tls_ctx} = $ctx || TLS_CTX ();
1572 $self->{tls} = $ssl = $self->{tls_ctx}->_get_session ($ssl, $self, $self->{peername});
1573
1574 # basically, this is deep magic (because SSL_read should have the same issues)
1575 # but the openssl maintainers basically said: "trust us, it just works".
1576 # (unfortunately, we have to hardcode constants because the abysmally misdesigned
1577 # and mismaintained ssleay-module doesn't even offer them).
1578 # http://www.mail-archive.com/openssl-dev@openssl.org/msg22420.html
1579 #
1580 # in short: this is a mess.
1581 #
1582 # note that we do not try to keep the length constant between writes as we are required to do.
1583 # we assume that most (but not all) of this insanity only applies to non-blocking cases,
1584 # and we drive openssl fully in blocking mode here. Or maybe we don't - openssl seems to
1585 # have identity issues in that area.
1586 # Net::SSLeay::CTX_set_mode ($ssl,
1587 # (eval { local $SIG{__DIE__}; Net::SSLeay::MODE_ENABLE_PARTIAL_WRITE () } || 1)
1588 # | (eval { local $SIG{__DIE__}; Net::SSLeay::MODE_ACCEPT_MOVING_WRITE_BUFFER () } || 2));
1589 Net::SSLeay::CTX_set_mode ($ssl, 1|2);
1590
1591 $self->{_rbio} = Net::SSLeay::BIO_new (Net::SSLeay::BIO_s_mem ());
1592 $self->{_wbio} = Net::SSLeay::BIO_new (Net::SSLeay::BIO_s_mem ());
1593
1594 Net::SSLeay::set_bio ($ssl, $self->{_rbio}, $self->{_wbio});
1595
1596 $self->{_on_starttls} = sub { $_[0]{on_starttls}(@_) }
1597 if $self->{on_starttls};
1598
1599 &_dotls; # need to trigger the initial handshake
1600 $self->start_read; # make sure we actually do read
1601 }
1602
1603 =item $handle->stoptls
1604
1605 Shuts down the SSL connection - this makes a proper EOF handshake by
1606 sending a close notify to the other side, but since OpenSSL doesn't
1607 support non-blocking shut downs, it is not possible to re-use the stream
1608 afterwards.
1609
1610 =cut
1611
1612 sub stoptls {
1613 my ($self) = @_;
1614
1615 if ($self->{tls}) {
1616 Net::SSLeay::shutdown ($self->{tls});
1617
1618 &_dotls;
1619
1620 # # we don't give a shit. no, we do, but we can't. no...#d#
1621 # # we, we... have to use openssl :/#d#
1622 # &_freetls;#d#
1623 }
1624 }
1625
1626 sub _freetls {
1627 my ($self) = @_;
1628
1629 return unless $self->{tls};
1630
1631 $self->{tls_ctx}->_put_session (delete $self->{tls});
1632
1633 delete @$self{qw(_rbio _wbio _tls_wbuf _on_starttls)};
1634 }
1635
1636 sub DESTROY {
1637 my ($self) = @_;
1638
1639 &_freetls;
1640
1641 my $linger = exists $self->{linger} ? $self->{linger} : 3600;
1642
1643 if ($linger && length $self->{wbuf}) {
1644 my $fh = delete $self->{fh};
1645 my $wbuf = delete $self->{wbuf};
1646
1647 my @linger;
1648
1649 push @linger, AnyEvent->io (fh => $fh, poll => "w", cb => sub {
1650 my $len = syswrite $fh, $wbuf, length $wbuf;
1651
1652 if ($len > 0) {
1653 substr $wbuf, 0, $len, "";
1654 } else {
1655 @linger = (); # end
1656 }
1657 });
1658 push @linger, AnyEvent->timer (after => $linger, cb => sub {
1659 @linger = ();
1660 });
1661 }
1662 }
1663
1664 =item $handle->destroy
1665
1666 Shuts down the handle object as much as possible - this call ensures that
1667 no further callbacks will be invoked and as many resources as possible
1668 will be freed. You must not call any methods on the object afterwards.
1669
1670 Normally, you can just "forget" any references to an AnyEvent::Handle
1671 object and it will simply shut down. This works in fatal error and EOF
1672 callbacks, as well as code outside. It does I<NOT> work in a read or write
1673 callback, so when you want to destroy the AnyEvent::Handle object from
1674 within such an callback. You I<MUST> call C<< ->destroy >> explicitly in
1675 that case.
1676
1677 Destroying the handle object in this way has the advantage that callbacks
1678 will be removed as well, so if those are the only reference holders (as
1679 is common), then one doesn't need to do anything special to break any
1680 reference cycles.
1681
1682 The handle might still linger in the background and write out remaining
1683 data, as specified by the C<linger> option, however.
1684
1685 =cut
1686
1687 sub destroy {
1688 my ($self) = @_;
1689
1690 $self->DESTROY;
1691 %$self = ();
1692 }
1693
1694 =item AnyEvent::Handle::TLS_CTX
1695
1696 This function creates and returns the AnyEvent::TLS object used by default
1697 for TLS mode.
1698
1699 The context is created by calling L<AnyEvent::TLS> without any arguments.
1700
1701 =cut
1702
1703 our $TLS_CTX;
1704
1705 sub TLS_CTX() {
1706 $TLS_CTX ||= do {
1707 require AnyEvent::TLS;
1708
1709 new AnyEvent::TLS
1710 }
1711 }
1712
1713 =back
1714
1715
1716 =head1 NONFREQUENTLY ASKED QUESTIONS
1717
1718 =over 4
1719
1720 =item I C<undef> the AnyEvent::Handle reference inside my callback and
1721 still get further invocations!
1722
1723 That's because AnyEvent::Handle keeps a reference to itself when handling
1724 read or write callbacks.
1725
1726 It is only safe to "forget" the reference inside EOF or error callbacks,
1727 from within all other callbacks, you need to explicitly call the C<<
1728 ->destroy >> method.
1729
1730 =item I get different callback invocations in TLS mode/Why can't I pause
1731 reading?
1732
1733 Unlike, say, TCP, TLS connections do not consist of two independent
1734 communication channels, one for each direction. Or put differently. The
1735 read and write directions are not independent of each other: you cannot
1736 write data unless you are also prepared to read, and vice versa.
1737
1738 This can mean than, in TLS mode, you might get C<on_error> or C<on_eof>
1739 callback invocations when you are not expecting any read data - the reason
1740 is that AnyEvent::Handle always reads in TLS mode.
1741
1742 During the connection, you have to make sure that you always have a
1743 non-empty read-queue, or an C<on_read> watcher. At the end of the
1744 connection (or when you no longer want to use it) you can call the
1745 C<destroy> method.
1746
1747 =item How do I read data until the other side closes the connection?
1748
1749 If you just want to read your data into a perl scalar, the easiest way
1750 to achieve this is by setting an C<on_read> callback that does nothing,
1751 clearing the C<on_eof> callback and in the C<on_error> callback, the data
1752 will be in C<$_[0]{rbuf}>:
1753
1754 $handle->on_read (sub { });
1755 $handle->on_eof (undef);
1756 $handle->on_error (sub {
1757 my $data = delete $_[0]{rbuf};
1758 });
1759
1760 The reason to use C<on_error> is that TCP connections, due to latencies
1761 and packets loss, might get closed quite violently with an error, when in
1762 fact, all data has been received.
1763
1764 It is usually better to use acknowledgements when transferring data,
1765 to make sure the other side hasn't just died and you got the data
1766 intact. This is also one reason why so many internet protocols have an
1767 explicit QUIT command.
1768
1769 =item I don't want to destroy the handle too early - how do I wait until
1770 all data has been written?
1771
1772 After writing your last bits of data, set the C<on_drain> callback
1773 and destroy the handle in there - with the default setting of
1774 C<low_water_mark> this will be called precisely when all data has been
1775 written to the socket:
1776
1777 $handle->push_write (...);
1778 $handle->on_drain (sub {
1779 warn "all data submitted to the kernel\n";
1780 undef $handle;
1781 });
1782
1783 If you just want to queue some data and then signal EOF to the other side,
1784 consider using C<< ->push_shutdown >> instead.
1785
1786 =item I want to contact a TLS/SSL server, I don't care about security.
1787
1788 If your TLS server is a pure TLS server (e.g. HTTPS) that only speaks TLS,
1789 simply connect to it and then create the AnyEvent::Handle with the C<tls>
1790 parameter:
1791
1792 tcp_connect $host, $port, sub {
1793 my ($fh) = @_;
1794
1795 my $handle = new AnyEvent::Handle
1796 fh => $fh,
1797 tls => "connect",
1798 on_error => sub { ... };
1799
1800 $handle->push_write (...);
1801 };
1802
1803 =item I want to contact a TLS/SSL server, I do care about security.
1804
1805 Then you should additionally enable certificate verification, including
1806 peername verification, if the protocol you use supports it (see
1807 L<AnyEvent::TLS>, C<verify_peername>).
1808
1809 E.g. for HTTPS:
1810
1811 tcp_connect $host, $port, sub {
1812 my ($fh) = @_;
1813
1814 my $handle = new AnyEvent::Handle
1815 fh => $fh,
1816 peername => $host,
1817 tls => "connect",
1818 tls_ctx => { verify => 1, verify_peername => "https" },
1819 ...
1820
1821 Note that you must specify the hostname you connected to (or whatever
1822 "peername" the protocol needs) as the C<peername> argument, otherwise no
1823 peername verification will be done.
1824
1825 The above will use the system-dependent default set of trusted CA
1826 certificates. If you want to check against a specific CA, add the
1827 C<ca_file> (or C<ca_cert>) arguments to C<tls_ctx>:
1828
1829 tls_ctx => {
1830 verify => 1,
1831 verify_peername => "https",
1832 ca_file => "my-ca-cert.pem",
1833 },
1834
1835 =item I want to create a TLS/SSL server, how do I do that?
1836
1837 Well, you first need to get a server certificate and key. You have
1838 three options: a) ask a CA (buy one, use cacert.org etc.) b) create a
1839 self-signed certificate (cheap. check the search engine of your choice,
1840 there are many tutorials on the net) or c) make your own CA (tinyca2 is a
1841 nice program for that purpose).
1842
1843 Then create a file with your private key (in PEM format, see
1844 L<AnyEvent::TLS>), followed by the certificate (also in PEM format). The
1845 file should then look like this:
1846
1847 -----BEGIN RSA PRIVATE KEY-----
1848 ...header data
1849 ... lots of base64'y-stuff
1850 -----END RSA PRIVATE KEY-----
1851
1852 -----BEGIN CERTIFICATE-----
1853 ... lots of base64'y-stuff
1854 -----END CERTIFICATE-----
1855
1856 The important bits are the "PRIVATE KEY" and "CERTIFICATE" parts. Then
1857 specify this file as C<cert_file>:
1858
1859 tcp_server undef, $port, sub {
1860 my ($fh) = @_;
1861
1862 my $handle = new AnyEvent::Handle
1863 fh => $fh,
1864 tls => "accept",
1865 tls_ctx => { cert_file => "my-server-keycert.pem" },
1866 ...
1867
1868 When you have intermediate CA certificates that your clients might not
1869 know about, just append them to the C<cert_file>.
1870
1871 =back
1872
1873
1874 =head1 SUBCLASSING AnyEvent::Handle
1875
1876 In many cases, you might want to subclass AnyEvent::Handle.
1877
1878 To make this easier, a given version of AnyEvent::Handle uses these
1879 conventions:
1880
1881 =over 4
1882
1883 =item * all constructor arguments become object members.
1884
1885 At least initially, when you pass a C<tls>-argument to the constructor it
1886 will end up in C<< $handle->{tls} >>. Those members might be changed or
1887 mutated later on (for example C<tls> will hold the TLS connection object).
1888
1889 =item * other object member names are prefixed with an C<_>.
1890
1891 All object members not explicitly documented (internal use) are prefixed
1892 with an underscore character, so the remaining non-C<_>-namespace is free
1893 for use for subclasses.
1894
1895 =item * all members not documented here and not prefixed with an underscore
1896 are free to use in subclasses.
1897
1898 Of course, new versions of AnyEvent::Handle may introduce more "public"
1899 member variables, but thats just life, at least it is documented.
1900
1901 =back
1902
1903 =head1 AUTHOR
1904
1905 Robin Redeker C<< <elmex at ta-sa.org> >>, Marc Lehmann <schmorp@schmorp.de>.
1906
1907 =cut
1908
1909 1; # End of AnyEvent::Handle