3 perlipc - Perl interprocess communication (signals, fifos, pipes, safe subprocesses, sockets, and semaphores)
7 The basic IPC facilities of Perl are built out of the good old Unix
8 signals, named pipes, pipe opens, the Berkeley socket routines, and SysV
9 IPC calls. Each is used in slightly different situations.
13 Perl uses a simple signal handling model: the %SIG hash contains names
14 or references of user-installed signal handlers. These handlers will
15 be called with an argument which is the name of the signal that
16 triggered it. A signal may be generated intentionally from a
17 particular keyboard sequence like control-C or control-Z, sent to you
18 from another process, or triggered automatically by the kernel when
19 special events transpire, like a child process exiting, your own process
20 running out of stack space, or hitting a process file-size limit.
22 For example, to trap an interrupt signal, set up a handler like this:
29 die "Somebody sent me a SIG$signame";
31 $SIG{INT} = __PACKAGE__ . "::catch_zap";
32 $SIG{INT} = \&catch_zap; # best strategy
34 Prior to Perl 5.8.0 it was necessary to do as little as you possibly
35 could in your handler; notice how all we do is set a global variable
36 and then raise an exception. That's because on most systems,
37 libraries are not re-entrant; particularly, memory allocation and I/O
38 routines are not. That meant that doing nearly I<anything> in your
39 handler could in theory trigger a memory fault and subsequent core
40 dump - see L</Deferred Signals (Safe Signals)> below.
42 The names of the signals are the ones listed out by C<kill -l> on your
43 system, or you can retrieve them using the CPAN module L<IPC::Signal>.
45 You may also choose to assign the strings C<"IGNORE"> or C<"DEFAULT"> as
46 the handler, in which case Perl will try to discard the signal or do the
49 On most Unix platforms, the C<CHLD> (sometimes also known as C<CLD>) signal
50 has special behavior with respect to a value of C<"IGNORE">.
51 Setting C<$SIG{CHLD}> to C<"IGNORE"> on such a platform has the effect of
52 not creating zombie processes when the parent process fails to C<wait()>
53 on its child processes (i.e., child processes are automatically reaped).
54 Calling C<wait()> with C<$SIG{CHLD}> set to C<"IGNORE"> usually returns
55 C<-1> on such platforms.
57 Some signals can be neither trapped nor ignored, such as the KILL and STOP
58 (but not the TSTP) signals. Note that ignoring signals makes them disappear.
59 If you only want them blocked temporarily without them getting lost you'll
60 have to use the C<POSIX> module's L<sigprocmask|POSIX/sigprocmask>.
62 Sending a signal to a negative process ID means that you send the signal
63 to the entire Unix process group. This code sends a hang-up signal to all
64 processes in the current process group, and also sets $SIG{HUP} to C<"IGNORE">
65 so it doesn't kill itself:
67 # block scope for local
69 local $SIG{HUP} = "IGNORE";
70 kill HUP => -getpgrp();
71 # snazzy writing of: kill("HUP", -getpgrp())
74 Another interesting signal to send is signal number zero. This doesn't
75 actually affect a child process, but instead checks whether it's alive
76 or has changed its UIDs.
78 unless (kill 0 => $kid_pid) {
79 warn "something wicked happened to $kid_pid";
82 Signal number zero may fail because you lack permission to send the
83 signal when directed at a process whose real or saved UID is not
84 identical to the real or effective UID of the sending process, even
85 though the process is alive. You may be able to determine the cause of
86 failure using C<$!> or C<%!>.
88 unless (kill(0 => $pid) || $!{EPERM}) {
89 warn "$pid looks dead";
92 You might also want to employ anonymous functions for simple signal
95 $SIG{INT} = sub { die "\nOutta here!\n" };
97 SIGCHLD handlers require some special care. If a second child dies
98 while in the signal handler caused by the first death, we won't get
99 another signal. So must loop here else we will leave the unreaped child
100 as a zombie. And the next time two children die we get another zombie.
103 use POSIX ":sys_wait_h";
105 while ((my $child = waitpid(-1, WNOHANG)) > 0) {
106 $Kid_Status{$child} = $?;
109 # do something that forks...
111 Be careful: qx(), system(), and some modules for calling external commands
112 do a fork(), then wait() for the result. Thus, your signal handler
113 will be called. Because wait() was already called by system() or qx(),
114 the wait() in the signal handler will see no more zombies and will
117 The best way to prevent this issue is to use waitpid(), as in the following
120 use POSIX ":sys_wait_h"; # for nonblocking read
125 # don't change $! and $? outside handler
127 while ( (my $pid = waitpid(-1, WNOHANG)) > 0 ) {
128 delete $children{$pid};
129 cleanup_child($pid, $?);
135 die "cannot fork" unless defined $pid;
147 Signal handling is also used for timeouts in Unix. While safely
148 protected within an C<eval{}> block, you set a signal handler to trap
149 alarm signals and then schedule to have one delivered to you in some
150 number of seconds. Then try your blocking operation, clearing the alarm
151 when it's done but not before you've exited your C<eval{}> block. If it
152 goes off, you'll use die() to jump out of the block.
156 my $ALARM_EXCEPTION = "alarm clock restart";
158 local $SIG{ALRM} = sub { die $ALARM_EXCEPTION };
160 flock($fh, 2) # blocking write lock
161 || die "cannot flock: $!";
164 if ($@ && $@ !~ quotemeta($ALARM_EXCEPTION)) { die }
166 If the operation being timed out is system() or qx(), this technique
167 is liable to generate zombies. If this matters to you, you'll
168 need to do your own fork() and exec(), and kill the errant child process.
170 For more complex signal handling, you might see the standard POSIX
171 module. Lamentably, this is almost entirely undocumented, but the
172 F<ext/POSIX/t/sigaction.t> file from the Perl source distribution has
175 =head2 Handling the SIGHUP Signal in Daemons
177 A process that usually starts when the system boots and shuts down
178 when the system is shut down is called a daemon (Disk And Execution
179 MONitor). If a daemon process has a configuration file which is
180 modified after the process has been started, there should be a way to
181 tell that process to reread its configuration file without stopping
182 the process. Many daemons provide this mechanism using a C<SIGHUP>
183 signal handler. When you want to tell the daemon to reread the file,
184 simply send it the C<SIGHUP> signal.
186 The following example implements a simple daemon, which restarts
187 itself every time the C<SIGHUP> signal is received. The actual code is
188 located in the subroutine C<code()>, which just prints some debugging
189 info to show that it works; it should be replaced with the real code.
198 use File::Basename ();
199 use File::Spec::Functions qw(catfile);
203 # make the daemon cross-platform, so exec always calls the script
204 # itself with the right path, no matter how the script was invoked.
205 my $script = File::Basename::basename($0);
206 my $SELF = catfile($FindBin::Bin, $script);
208 # POSIX unmasks the sigprocmask properly
210 print "got SIGHUP\n";
211 exec($SELF, @ARGV) || die "$0: couldn't restart: $!";
218 print "ARGV: @ARGV\n";
222 print ++$count, "\n";
227 =head2 Deferred Signals (Safe Signals)
229 Before Perl 5.8.0, installing Perl code to deal with signals exposed you to
230 danger from two things. First, few system library functions are
231 re-entrant. If the signal interrupts while Perl is executing one function
232 (like malloc(3) or printf(3)), and your signal handler then calls the same
233 function again, you could get unpredictable behavior--often, a core dump.
234 Second, Perl isn't itself re-entrant at the lowest levels. If the signal
235 interrupts Perl while Perl is changing its own internal data structures,
236 similarly unpredictable behavior may result.
238 There were two things you could do, knowing this: be paranoid or be
239 pragmatic. The paranoid approach was to do as little as possible in your
240 signal handler. Set an existing integer variable that already has a
241 value, and return. This doesn't help you if you're in a slow system call,
242 which will just restart. That means you have to C<die> to longjmp(3) out
243 of the handler. Even this is a little cavalier for the true paranoiac,
244 who avoids C<die> in a handler because the system I<is> out to get you.
245 The pragmatic approach was to say "I know the risks, but prefer the
246 convenience", and to do anything you wanted in your signal handler,
247 and be prepared to clean up core dumps now and again.
249 Perl 5.8.0 and later avoid these problems by "deferring" signals. That is,
250 when the signal is delivered to the process by the system (to the C code
251 that implements Perl) a flag is set, and the handler returns immediately.
252 Then at strategic "safe" points in the Perl interpreter (e.g. when it is
253 about to execute a new opcode) the flags are checked and the Perl level
254 handler from %SIG is executed. The "deferred" scheme allows much more
255 flexibility in the coding of signal handlers as we know the Perl
256 interpreter is in a safe state, and that we are not in a system library
257 function when the handler is called. However the implementation does
258 differ from previous Perls in the following ways:
262 =item Long-running opcodes
264 As the Perl interpreter looks at signal flags only when it is about
265 to execute a new opcode, a signal that arrives during a long-running
266 opcode (e.g. a regular expression operation on a very large string) will
267 not be seen until the current opcode completes.
269 If a signal of any given type fires multiple times during an opcode
270 (such as from a fine-grained timer), the handler for that signal will
271 be called only once, after the opcode completes; all other
272 instances will be discarded. Furthermore, if your system's signal queue
273 gets flooded to the point that there are signals that have been raised
274 but not yet caught (and thus not deferred) at the time an opcode
275 completes, those signals may well be caught and deferred during
276 subsequent opcodes, with sometimes surprising results. For example, you
277 may see alarms delivered even after calling C<alarm(0)> as the latter
278 stops the raising of alarms but does not cancel the delivery of alarms
279 raised but not yet caught. Do not depend on the behaviors described in
280 this paragraph as they are side effects of the current implementation and
281 may change in future versions of Perl.
283 =item Interrupting IO
285 When a signal is delivered (e.g., SIGINT from a control-C) the operating
286 system breaks into IO operations like I<read>(2), which is used to
287 implement Perl's readline() function, the C<< <> >> operator. On older
288 Perls the handler was called immediately (and as C<read> is not "unsafe",
289 this worked well). With the "deferred" scheme the handler is I<not> called
290 immediately, and if Perl is using the system's C<stdio> library that
291 library may restart the C<read> without returning to Perl to give it a
292 chance to call the %SIG handler. If this happens on your system the
293 solution is to use the C<:perlio> layer to do IO--at least on those handles
294 that you want to be able to break into with signals. (The C<:perlio> layer
295 checks the signal flags and calls %SIG handlers before resuming IO
298 The default in Perl 5.8.0 and later is to automatically use
299 the C<:perlio> layer.
301 Note that it is not advisable to access a file handle within a signal
302 handler where that signal has interrupted an I/O operation on that same
303 handle. While perl will at least try hard not to crash, there are no
304 guarantees of data integrity; for example, some data might get dropped or
307 Some networking library functions like gethostbyname() are known to have
308 their own implementations of timeouts which may conflict with your
309 timeouts. If you have problems with such functions, try using the POSIX
310 sigaction() function, which bypasses Perl safe signals. Be warned that
311 this does subject you to possible memory corruption, as described above.
313 Instead of setting C<$SIG{ALRM}>:
315 local $SIG{ALRM} = sub { die "alarm" };
317 try something like the following:
319 use POSIX qw(SIGALRM);
320 POSIX::sigaction(SIGALRM,
321 POSIX::SigAction->new(sub { die "alarm" }))
322 || die "Error setting SIGALRM handler: $!\n";
324 Another way to disable the safe signal behavior locally is to use
325 the C<Perl::Unsafe::Signals> module from CPAN, which affects
328 =item Restartable system calls
330 On systems that supported it, older versions of Perl used the
331 SA_RESTART flag when installing %SIG handlers. This meant that
332 restartable system calls would continue rather than returning when
333 a signal arrived. In order to deliver deferred signals promptly,
334 Perl 5.8.0 and later do I<not> use SA_RESTART. Consequently,
335 restartable system calls can fail (with $! set to C<EINTR>) in places
336 where they previously would have succeeded.
338 The default C<:perlio> layer retries C<read>, C<write>
339 and C<close> as described above; interrupted C<wait> and
340 C<waitpid> calls will always be retried.
342 =item Signals as "faults"
344 Certain signals like SEGV, ILL, BUS and FPE are generated by virtual memory
345 addressing errors and similar "faults". These are normally fatal: there is
346 little a Perl-level handler can do with them. So Perl delivers them
347 immediately rather than attempting to defer them.
349 It is possible to catch these with a C<%SIG> handler (see L<perlvar>),
350 but on top of the usual problems of "unsafe" signals the signal is likely
351 to get rethrown immediately on return from the signal handler, so such
352 a handler should C<die> or C<exit> instead.
354 =item Signals triggered by operating system state
356 On some operating systems certain signal handlers are supposed to "do
357 something" before returning. One example can be CHLD or CLD, which
358 indicates a child process has completed. On some operating systems the
359 signal handler is expected to C<wait> for the completed child
360 process. On such systems the deferred signal scheme will not work for
361 those signals: it does not do the C<wait>. Again the failure will
362 look like a loop as the operating system will reissue the signal because
363 there are completed child processes that have not yet been C<wait>ed for.
367 If you want the old signal behavior back despite possible
368 memory corruption, set the environment variable C<PERL_SIGNALS> to
369 C<"unsafe">. This feature first appeared in Perl 5.8.1.
373 A named pipe (often referred to as a FIFO) is an old Unix IPC
374 mechanism for processes communicating on the same machine. It works
375 just like regular anonymous pipes, except that the
376 processes rendezvous using a filename and need not be related.
378 To create a named pipe, use the C<POSIX::mkfifo()> function.
380 use POSIX qw(mkfifo);
381 mkfifo($path, 0700) || die "mkfifo $path failed: $!";
383 You can also use the Unix command mknod(1), or on some
384 systems, mkfifo(1). These may not be in your normal path, though.
386 # system return val is backwards, so && not ||
388 $ENV{PATH} .= ":/etc:/usr/etc";
389 if ( system("mknod", $path, "p")
390 && system("mkfifo", $path) )
392 die "mk{nod,fifo} $path failed";
396 A fifo is convenient when you want to connect a process to an unrelated
397 one. When you open a fifo, the program will block until there's something
400 For example, let's say you'd like to have your F<.signature> file be a
401 named pipe that has a Perl program on the other end. Now every time any
402 program (like a mailer, news reader, finger program, etc.) tries to read
403 from that file, the reading program will read the new signature from your
404 program. We'll use the pipe-checking file-test operator, B<-p>, to find
405 out whether anyone (or anything) has accidentally removed our fifo.
408 my $FIFO = ".signature";
412 unlink $FIFO; # discard any failure, will catch later
413 require POSIX; # delayed loading of heavy module
414 POSIX::mkfifo($FIFO, 0700)
415 || die "can't mkfifo $FIFO: $!";
418 # next line blocks till there's a reader
419 open (my $fh, ">", $FIFO) || die "can't open $FIFO: $!";
420 print $fh "John Smith (smith\@host.org)\n", `fortune -s`;
421 close($fh) || die "can't close $FIFO: $!";
422 sleep 2; # to avoid dup signals
425 =head1 Using open() for IPC
427 Perl's basic open() statement can also be used for unidirectional
428 interprocess communication by specifying the open mode as C<|-> or C<-|>.
430 something up in a child process you intend to write to:
432 open(my $spooler, "|-", "cat -v | lpr -h 2>/dev/null")
433 || die "can't fork: $!";
434 local $SIG{PIPE} = sub { die "spooler pipe broke" };
435 print $spooler "stuff\n";
436 close $spooler || die "bad spool: $! $?";
438 And here's how to start up a child process you intend to read from:
440 open(my $status, "-|", "netstat -an 2>&1")
441 || die "can't fork: $!";
443 next if /^(tcp|udp)/;
446 close $status || die "bad netstat: $! $?";
448 Be aware that these operations are full Unix forks, which means they may
449 not be correctly implemented on all alien systems. See L<perlport/open>
450 for portability details.
452 In the two-argument form of open(), a pipe open can be achieved by
453 either appending or prepending a pipe symbol to the second argument:
455 open(my $spooler, "| cat -v | lpr -h 2>/dev/null")
456 || die "can't fork: $!";
457 open(my $status, "netstat -an 2>&1 |")
458 || die "can't fork: $!";
460 This can be used even on systems that do not support forking, but this
461 possibly allows code intended to read files to unexpectedly execute
462 programs. If one can be sure that a particular program is a Perl script
463 expecting filenames in @ARGV using the two-argument form of open() or the
464 C<< <> >> operator, the clever programmer can write something like this:
466 % program f1 "cmd1|" - f2 "cmd2|" f3 < tmpfile
468 and no matter which sort of shell it's called from, the Perl program will
469 read from the file F<f1>, the process F<cmd1>, standard input (F<tmpfile>
470 in this case), the F<f2> file, the F<cmd2> command, and finally the F<f3>
471 file. Pretty nifty, eh?
473 You might notice that you could use backticks for much the
474 same effect as opening a pipe for reading:
476 print grep { !/^(tcp|udp)/ } `netstat -an 2>&1`;
477 die "bad netstatus ($?)" if $?;
479 While this is true on the surface, it's much more efficient to process the
480 file one line or record at a time because then you don't have to read the
481 whole thing into memory at once. It also gives you finer control of the
482 whole process, letting you kill off the child process early if you'd like.
484 Be careful to check the return values from both open() and close(). If
485 you're I<writing> to a pipe, you should also trap SIGPIPE. Otherwise,
486 think of what happens when you start up a pipe to a command that doesn't
487 exist: the open() will in all likelihood succeed (it only reflects the
488 fork()'s success), but then your output will fail--spectacularly. Perl
489 can't know whether the command worked, because your command is actually
490 running in a separate process whose exec() might have failed. Therefore,
491 while readers of bogus commands return just a quick EOF, writers
492 to bogus commands will get hit with a signal, which they'd best be prepared
495 open(my $fh, "|-", "bogus") || die "can't fork: $!";
496 print $fh "bang\n"; # neither necessary nor sufficient
497 # to check print retval!
498 close($fh) || die "can't close: $!";
500 The reason for not checking the return value from print() is because of
501 pipe buffering; physical writes are delayed. That won't blow up until the
502 close, and it will blow up with a SIGPIPE. To catch it, you could use
505 $SIG{PIPE} = "IGNORE";
506 open(my $fh, "|-", "bogus") || die "can't fork: $!";
508 close($fh) || die "can't close: status=$?";
512 Both the main process and any child processes it forks share the same
513 STDIN, STDOUT, and STDERR filehandles. If both processes try to access
514 them at once, strange things can happen. You may also want to close
515 or reopen the filehandles for the child. You can get around this by
516 opening your pipe with open(), but on some systems this means that the
517 child process cannot outlive the parent.
519 =head2 Background Processes
521 You can run a command in the background with:
525 The command's STDOUT and STDERR (and possibly STDIN, depending on your
526 shell) will be the same as the parent's. You won't need to catch
527 SIGCHLD because of the double-fork taking place; see below for details.
529 =head2 Complete Dissociation of Child from Parent
531 In some cases (starting server processes, for instance) you'll want to
532 completely dissociate the child process from the parent. This is
533 often called daemonization. A well-behaved daemon will also chdir()
534 to the root directory so it doesn't prevent unmounting the filesystem
535 containing the directory from which it was launched, and redirect its
536 standard file descriptors from and to F</dev/null> so that random
537 output doesn't wind up on the user's terminal.
542 chdir("/") || die "can't chdir to /: $!";
543 open(STDIN, "<", "/dev/null") || die "can't read /dev/null: $!";
544 open(STDOUT, ">", "/dev/null") || die "can't write /dev/null: $!";
545 defined(my $pid = fork()) || die "can't fork: $!";
546 exit if $pid; # non-zero now means I am the parent
547 (setsid() != -1) || die "Can't start a new session: $!";
548 open(STDERR, ">&", STDOUT) || die "can't dup stdout: $!";
551 The fork() has to come before the setsid() to ensure you aren't a
552 process group leader; the setsid() will fail if you are. If your
553 system doesn't have the setsid() function, open F</dev/tty> and use the
554 C<TIOCNOTTY> ioctl() on it instead. See tty(4) for details.
556 Non-Unix users should check their C<< I<Your_OS>::Process >> module for
557 other possible solutions.
559 =head2 Safe Pipe Opens
561 Another interesting approach to IPC is making your single program go
562 multiprocess and communicate between--or even amongst--yourselves. The
563 two-argument form of the
564 open() function will accept a file argument of either C<"-|"> or C<"|-">
565 to do a very interesting thing: it forks a child connected to the
566 filehandle you've opened. The child is running the same program as the
567 parent. This is useful for safely opening a file when running under an
568 assumed UID or GID, for example. If you open a pipe I<to> minus, you can
569 write to the filehandle you opened and your kid will find it in I<his>
570 STDIN. If you open a pipe I<from> minus, you can read from the filehandle
571 you opened whatever your kid writes to I<his> STDOUT.
573 my $PRECIOUS = "/path/to/some/safe/file";
579 $pid = open($kid_to_write, "|-");
580 unless (defined $pid) {
581 warn "cannot fork: $!";
582 die "bailing out" if $sleep_count++ > 6;
585 } until defined $pid;
587 if ($pid) { # I am the parent
588 print $kid_to_write @some_data;
589 close($kid_to_write) || warn "kid exited $?";
590 } else { # I am the child
591 # drop permissions in setuid and/or setgid programs:
593 open (my $outfile, ">", $PRECIOUS)
594 || die "can't open $PRECIOUS: $!";
596 print $outfile; # child STDIN is parent $kid_to_write
598 close($outfile) || die "can't close $PRECIOUS: $!";
599 exit(0); # don't forget this!!
602 Another common use for this construct is when you need to execute
603 something without the shell's interference. With system(), it's
604 straightforward, but you can't use a pipe open or backticks safely.
605 That's because there's no way to stop the shell from getting its hands on
606 your arguments. Instead, use lower-level control to call exec() directly.
608 Here's a safe backtick or pipe open for read:
610 my $pid = open(my $kid_to_read, "-|");
611 defined($pid) || die "can't fork: $!";
614 while (<$kid_to_read>) {
615 # do something interesting
617 close($kid_to_read) || warn "kid exited $?";
620 ($>, $)) = ($<, $(); # suid only
621 exec($program, @options, @args)
622 || die "can't exec program: $!";
626 And here's a safe pipe open for writing:
628 my $pid = open(my $kid_to_write, "|-");
629 defined($pid) || die "can't fork: $!";
631 $SIG{PIPE} = sub { die "whoops, $program pipe broke" };
634 print $kid_to_write @data;
635 close($kid_to_write) || warn "kid exited $?";
639 exec($program, @options, @args)
640 || die "can't exec program: $!";
644 It is very easy to dead-lock a process using this form of open(), or
645 indeed with any use of pipe() with multiple subprocesses. The
646 example above is "safe" because it is simple and calls exec(). See
647 L</"Avoiding Pipe Deadlocks"> for general safety principles, but there
648 are extra gotchas with Safe Pipe Opens.
650 In particular, if you opened the pipe using C<open $fh, "|-">, then you
651 cannot simply use close() in the parent process to close an unwanted
652 writer. Consider this code:
654 my $pid = open(my $writer, "|-"); # fork open a kid
655 defined($pid) || die "first fork failed: $!";
657 if (my $sub_pid = fork()) {
658 defined($sub_pid) || die "second fork failed: $!";
659 close($writer) || die "couldn't close writer: $!";
660 # now do something else...
663 # first write to $writer
666 close($writer) || die "couldn't close writer: $!";
671 # first do something with STDIN, then
675 In the example above, the true parent does not want to write to the $writer
676 filehandle, so it closes it. However, because $writer was opened using
677 C<open $fh, "|-">, it has a special behavior: closing it calls
678 waitpid() (see L<perlfunc/waitpid>), which waits for the subprocess
679 to exit. If the child process ends up waiting for something happening
680 in the section marked "do something else", you have deadlock.
682 This can also be a problem with intermediate subprocesses in more
683 complicated code, which will call waitpid() on all open filehandles
684 during global destruction--in no predictable order.
686 To solve this, you must manually use pipe(), fork(), and the form of
687 open() which sets one file descriptor to another, as shown below:
689 pipe(my $reader, my $writer) || die "pipe failed: $!";
691 defined($pid) || die "first fork failed: $!";
694 if (my $sub_pid = fork()) {
695 defined($sub_pid) || die "first fork failed: $!";
696 close($writer) || die "can't close writer: $!";
699 # write to $writer...
702 close($writer) || die "can't close writer: $!";
705 # write to $writer...
708 open(STDIN, "<&", $reader) || die "can't reopen STDIN: $!";
709 close($writer) || die "can't close writer: $!";
714 Since Perl 5.8.0, you can also use the list form of C<open> for pipes.
715 This is preferred when you wish to avoid having the shell interpret
716 metacharacters that may be in your command string.
718 So for example, instead of using:
720 open(my $ps_pipe, "-|", "ps aux") || die "can't open ps pipe: $!";
722 One would use either of these:
724 open(my $ps_pipe, "-|", "ps", "aux")
725 || die "can't open ps pipe: $!";
727 my @ps_args = qw[ ps aux ];
728 open(my $ps_pipe, "-|", @ps_args)
729 || die "can't open @ps_args|: $!";
731 Because there are more than three arguments to open(), it forks the ps(1)
732 command I<without> spawning a shell, and reads its standard output via the
733 C<$ps_pipe> filehandle. The corresponding syntax to I<write> to command
734 pipes is to use C<"|-"> in place of C<"-|">.
736 This was admittedly a rather silly example, because you're using string
737 literals whose content is perfectly safe. There is therefore no cause to
738 resort to the harder-to-read, multi-argument form of pipe open(). However,
739 whenever you cannot be assured that the program arguments are free of shell
740 metacharacters, the fancier form of open() should be used. For example:
742 my @grep_args = ("egrep", "-i", $some_pattern, @many_files);
743 open(my $grep_pipe, "-|", @grep_args)
744 || die "can't open @grep_args|: $!";
746 Here the multi-argument form of pipe open() is preferred because the
747 pattern and indeed even the filenames themselves might hold metacharacters.
749 =head2 Avoiding Pipe Deadlocks
751 Whenever you have more than one subprocess, you must be careful that each
752 closes whichever half of any pipes created for interprocess communication
753 it is not using. This is because any child process reading from the pipe
754 and expecting an EOF will never receive it, and therefore never exit. A
755 single process closing a pipe is not enough to close it; the last process
756 with the pipe open must close it for it to read EOF.
758 Certain built-in Unix features help prevent this most of the time. For
759 instance, filehandles have a "close on exec" flag, which is set I<en masse>
760 under control of the C<$^F> variable. This is so any filehandles you
761 didn't explicitly route to the STDIN, STDOUT or STDERR of a child
762 I<program> will be automatically closed.
764 Always explicitly and immediately call close() on the writable end of any
765 pipe, unless that process is actually writing to it. Even if you don't
766 explicitly call close(), Perl will still close() all filehandles during
767 global destruction. As previously discussed, if those filehandles have
768 been opened with Safe Pipe Open, this will result in calling waitpid(),
769 which may again deadlock.
771 =head2 Bidirectional Communication with Another Process
773 While this works reasonably well for unidirectional communication, what
774 about bidirectional communication? The most obvious approach doesn't work:
776 # THIS DOES NOT WORK!!
777 open(my $prog_for_reading_and_writing, "| some program |")
779 If you forget to C<use warnings>, you'll miss out entirely on the
780 helpful diagnostic message:
782 Can't do bidirectional pipe at -e line 1.
784 If you really want to, you can use the standard open2() from the
785 L<IPC::Open2> module to catch both ends. There's also an open3() in
786 L<IPC::Open3> for tridirectional I/O so you can also catch your child's
787 STDERR, but doing so would then require an awkward select() loop and
788 wouldn't allow you to use normal Perl input operations.
790 If you look at its source, you'll see that open2() uses low-level
791 primitives like the pipe() and exec() syscalls to create all the
792 connections. Although it might have been more efficient by using
793 socketpair(), this would have been even less portable than it already
794 is. The open2() and open3() functions are unlikely to work anywhere
795 except on a Unix system, or at least one purporting POSIX compliance.
798 Hold on, is this even true? First it says that socketpair() is avoided
799 for portability, but then it says it probably won't work except on
800 Unixy systems anyway. Which one of those is true?
802 Here's an example of using open2():
805 my $pid = open2(my $reader, my $writer, "cat -un");
806 print $writer "stuff\n";
810 The problem with this is that buffering is really going to ruin your
811 day. Even though your C<$writer> filehandle is auto-flushed so the process
812 on the other end gets your data in a timely manner, you can't usually do
813 anything to force that process to give its data to you in a similarly quick
814 fashion. In this special case, we could actually so, because we gave
815 I<cat> a B<-u> flag to make it unbuffered. But very few commands are
816 designed to operate over pipes, so this seldom works unless you yourself
817 wrote the program on the other end of the double-ended pipe.
819 A solution to this is to use a library which uses pseudottys to make your
820 program behave more reasonably. This way you don't have to have control
821 over the source code of the program you're using. The C<Expect> module
822 from CPAN also addresses this kind of thing. This module requires two
823 other modules from CPAN, C<IO::Pty> and C<IO::Stty>. It sets up a pseudo
824 terminal to interact with programs that insist on talking to the terminal
825 device driver. If your system is supported, this may be your best bet.
827 =head2 Bidirectional Communication with Yourself
829 If you want, you may make low-level pipe() and fork() syscalls to stitch
830 this together by hand. This example only talks to itself, but you could
831 reopen the appropriate handles to STDIN and STDOUT and call other processes.
832 (The following example lacks proper error checking.)
835 # pipe1 - bidirectional communication using two pipe pairs
836 # designed for the socketpair-challenged
839 use IO::Handle; # enable autoflush method before Perl 5.14
840 pipe(my $parent_rdr, my $child_wtr); # XXX: check failure?
841 pipe(my $child_rdr, my $parent_wtr); # XXX: check failure?
842 $child_wtr->autoflush(1);
843 $parent_wtr->autoflush(1);
848 print $child_wtr "Parent Pid $$ is sending this\n";
849 chomp(my $line = <$child_rdr>);
850 print "Parent Pid $$ just read this: '$line'\n";
851 close $child_rdr; close $child_wtr;
854 die "cannot fork: $!" unless defined $pid;
857 chomp(my $line = <$parent_rdr>);
858 print "Child Pid $$ just read this: '$line'\n";
859 print $parent_wtr "Child Pid $$ is sending this\n";
865 But you don't actually have to make two pipe calls. If you
866 have the socketpair() system call, it will do this all for you.
869 # pipe2 - bidirectional communication using socketpair
870 # "the best ones always go both ways"
875 use IO::Handle; # enable autoflush method before Perl 5.14
877 # We say AF_UNIX because although *_LOCAL is the
878 # POSIX 1003.1g form of the constant, many machines
879 # still don't have it.
880 socketpair(my $child, my $parent, AF_UNIX, SOCK_STREAM, PF_UNSPEC)
881 || die "socketpair: $!";
883 $child->autoflush(1);
884 $parent->autoflush(1);
888 print $child "Parent Pid $$ is sending this\n";
889 chomp(my $line = <$child>);
890 print "Parent Pid $$ just read this: '$line'\n";
894 die "cannot fork: $!" unless defined $pid;
896 chomp(my $line = <$parent>);
897 print "Child Pid $$ just read this: '$line'\n";
898 print $parent "Child Pid $$ is sending this\n";
903 =head1 Sockets: Client/Server Communication
905 While not entirely limited to Unix-derived operating systems (e.g., WinSock
906 on PCs provides socket support, as do some VMS libraries), you might not have
907 sockets on your system, in which case this section probably isn't going to
908 do you much good. With sockets, you can do both virtual circuits like TCP
909 streams and datagrams like UDP packets. You may be able to do even more
910 depending on your system.
912 The Perl functions for dealing with sockets have the same names as
913 the corresponding system calls in C, but their arguments tend to differ
914 for two reasons. First, Perl filehandles work differently than C file
915 descriptors. Second, Perl already knows the length of its strings, so you
916 don't need to pass that information.
918 One of the major problems with ancient, antemillennial socket code in Perl
919 was that it used hard-coded values for some of the constants, which
920 severely hurt portability. If you ever see code that does anything like
921 explicitly setting C<$AF_INET = 2>, you know you're in for big trouble.
922 An immeasurably superior approach is to use the L<Socket> module, which more
923 reliably grants access to the various constants and functions you'll need.
925 If you're not writing a server/client for an existing protocol like
926 NNTP or SMTP, you should give some thought to how your server will
927 know when the client has finished talking, and vice-versa. Most
928 protocols are based on one-line messages and responses (so one party
929 knows the other has finished when a "\n" is received) or multi-line
930 messages and responses that end with a period on an empty line
931 ("\n.\n" terminates a message/response).
933 =head2 Internet Line Terminators
935 The Internet line terminator is "\015\012". Under ASCII variants of
936 Unix, that could usually be written as "\r\n", but under other systems,
937 "\r\n" might at times be "\015\015\012", "\012\012\015", or something
938 completely different. The standards specify writing "\015\012" to be
939 conformant (be strict in what you provide), but they also recommend
940 accepting a lone "\012" on input (be lenient in what you require).
941 We haven't always been very good about that in the code in this manpage,
942 but unless you're on a Mac from way back in its pre-Unix dark ages, you'll
945 =head2 Internet TCP Clients and Servers
947 Use Internet-domain sockets when you want to do client-server
948 communication that might extend to machines outside of your own system.
950 Here's a sample TCP client using Internet-domain sockets:
957 my $remote = shift || "localhost";
958 my $port = shift || 2345; # random port
959 if ($port =~ /\D/) { $port = getservbyname($port, "tcp") }
960 die "No port" unless $port;
961 my $iaddr = inet_aton($remote) || die "no host: $remote";
962 my $paddr = sockaddr_in($port, $iaddr);
964 my $proto = getprotobyname("tcp");
965 socket(my $sock, PF_INET, SOCK_STREAM, $proto) || die "socket: $!";
966 connect($sock, $paddr) || die "connect: $!";
967 while (my $line = <$sock>) {
971 close ($sock) || die "close: $!";
974 And here's a corresponding server to go along with it. We'll
975 leave the address as C<INADDR_ANY> so that the kernel can choose
976 the appropriate interface on multihomed hosts. If you want sit
977 on a particular interface (like the external side of a gateway
978 or firewall machine), fill this in with your real address instead.
983 BEGIN { $ENV{PATH} = "/usr/bin:/bin" }
986 my $EOL = "\015\012";
988 sub logmsg { print "$0 $$: @_ at ", scalar localtime(), "\n" }
990 my $port = shift || 2345;
991 die "invalid port" unless $port =~ /^ \d+ $/x;
993 my $proto = getprotobyname("tcp");
995 socket(my $server, PF_INET, SOCK_STREAM, $proto) || die "socket: $!";
996 setsockopt($server, SOL_SOCKET, SO_REUSEADDR, pack("l", 1))
997 || die "setsockopt: $!";
998 bind($server, sockaddr_in($port, INADDR_ANY)) || die "bind: $!";
999 listen($server, SOMAXCONN) || die "listen: $!";
1001 logmsg "server started on port $port";
1003 for (my $paddr; $paddr = accept(my $client, $server); close $client) {
1004 my($port, $iaddr) = sockaddr_in($paddr);
1005 my $name = gethostbyaddr($iaddr, AF_INET);
1007 logmsg "connection from $name [",
1008 inet_ntoa($iaddr), "]
1011 print $client "Hello there, $name, it's now ",
1012 scalar localtime(), $EOL;
1015 And here's a multitasking version. It's multitasked in that
1016 like most typical servers, it spawns (fork()s) a child server to
1017 handle the client request so that the master server can quickly
1018 go back to service a new client.
1023 BEGIN { $ENV{PATH} = "/usr/bin:/bin" }
1026 my $EOL = "\015\012";
1028 sub spawn; # forward declaration
1029 sub logmsg { print "$0 $$: @_ at ", scalar localtime(), "\n" }
1031 my $port = shift || 2345;
1032 die "invalid port" unless $port =~ /^ \d+ $/x;
1034 my $proto = getprotobyname("tcp");
1036 socket(my $server, PF_INET, SOCK_STREAM, $proto) || die "socket: $!";
1037 setsockopt($server, SOL_SOCKET, SO_REUSEADDR, pack("l", 1))
1038 || die "setsockopt: $!";
1039 bind($server, sockaddr_in($port, INADDR_ANY)) || die "bind: $!";
1040 listen($server, SOMAXCONN) || die "listen: $!";
1042 logmsg "server started on port $port";
1046 use POSIX ":sys_wait_h";
1050 local $!; # don't let waitpid() overwrite current error
1051 while ((my $pid = waitpid(-1, WNOHANG)) > 0 && WIFEXITED($?)) {
1052 logmsg "reaped $waitedpid" . ($? ? " with exit $?" : "");
1054 $SIG{CHLD} = \&REAPER; # loathe SysV
1057 $SIG{CHLD} = \&REAPER;
1060 my $paddr = accept(my $client, $server) || do {
1061 # try again if accept() returned because got a signal
1065 my ($port, $iaddr) = sockaddr_in($paddr);
1066 my $name = gethostbyaddr($iaddr, AF_INET);
1068 logmsg "connection from $name [",
1072 spawn $client, sub {
1074 print "Hello there, $name, it's now ",
1077 exec "/usr/games/fortune" # XXX: "wrong" line terminators
1078 or confess "can't exec fortune: $!";
1085 my $coderef = shift;
1087 unless (@_ == 0 && $coderef && ref($coderef) eq "CODE") {
1088 confess "usage: spawn CLIENT CODEREF";
1092 unless (defined($pid = fork())) {
1093 logmsg "cannot fork: $!";
1097 logmsg "begat $pid";
1098 return; # I'm the parent
1100 # else I'm the child -- go spawn
1102 open(STDIN, "<&", $client) || die "can't dup client to stdin";
1103 open(STDOUT, ">&", $client) || die "can't dup client to stdout";
1104 ## open(STDERR, ">&", STDOUT) || die "can't dup stdout to stderr";
1108 This server takes the trouble to clone off a child version via fork()
1109 for each incoming request. That way it can handle many requests at
1110 once, which you might not always want. Even if you don't fork(), the
1111 listen() will allow that many pending connections. Forking servers
1112 have to be particularly careful about cleaning up their dead children
1113 (called "zombies" in Unix parlance), because otherwise you'll quickly
1114 fill up your process table. The REAPER subroutine is used here to
1115 call waitpid() for any child processes that have finished, thereby
1116 ensuring that they terminate cleanly and don't join the ranks of the
1119 Within the while loop we call accept() and check to see if it returns
1120 a false value. This would normally indicate a system error needs
1121 to be reported. However, the introduction of safe signals (see
1122 L</Deferred Signals (Safe Signals)> above) in Perl 5.8.0 means that
1123 accept() might also be interrupted when the process receives a signal.
1124 This typically happens when one of the forked subprocesses exits and
1125 notifies the parent process with a CHLD signal.
1127 If accept() is interrupted by a signal, $! will be set to EINTR.
1128 If this happens, we can safely continue to the next iteration of
1129 the loop and another call to accept(). It is important that your
1130 signal handling code not modify the value of $!, or else this test
1131 will likely fail. In the REAPER subroutine we create a local version
1132 of $! before calling waitpid(). When waitpid() sets $! to ECHILD as
1133 it inevitably does when it has no more children waiting, it
1134 updates the local copy and leaves the original unchanged.
1136 You should use the B<-T> flag to enable taint checking (see L<perlsec>)
1137 even if we aren't running setuid or setgid. This is always a good idea
1138 for servers or any program run on behalf of someone else (like CGI
1139 scripts), because it lessens the chances that people from the outside will
1140 be able to compromise your system.
1142 Let's look at another TCP client. This one connects to the TCP "time"
1143 service on a number of different machines and shows how far their clocks
1144 differ from the system on which it's being run:
1151 my $SECS_OF_70_YEARS = 2208988800;
1152 sub ctime { scalar localtime(shift() || time()) }
1154 my $iaddr = gethostbyname("localhost");
1155 my $proto = getprotobyname("tcp");
1156 my $port = getservbyname("time", "tcp");
1157 my $paddr = sockaddr_in(0, $iaddr);
1160 printf "%-24s %8s %s\n", "localhost", 0, ctime();
1162 foreach my $host (@ARGV) {
1163 printf "%-24s ", $host;
1164 my $hisiaddr = inet_aton($host) || die "unknown host";
1165 my $hispaddr = sockaddr_in($port, $hisiaddr);
1166 socket(my $socket, PF_INET, SOCK_STREAM, $proto)
1167 || die "socket: $!";
1168 connect($socket, $hispaddr) || die "connect: $!";
1169 my $rtime = pack("C4", ());
1170 read($socket, $rtime, 4);
1172 my $histime = unpack("N", $rtime) - $SECS_OF_70_YEARS;
1173 printf "%8d %s\n", $histime - time(), ctime($histime);
1176 =head2 Unix-Domain TCP Clients and Servers
1178 That's fine for Internet-domain clients and servers, but what about local
1179 communications? While you can use the same setup, sometimes you don't
1180 want to. Unix-domain sockets are local to the current host, and are often
1181 used internally to implement pipes. Unlike Internet domain sockets, Unix
1182 domain sockets can show up in the file system with an ls(1) listing.
1185 srw-rw-rw- 1 root 0 Oct 31 07:23 /dev/log
1187 You can test for these with Perl's B<-S> file test:
1189 unless (-S "/dev/log") {
1190 die "something's wicked with the log system";
1193 Here's a sample Unix-domain client:
1200 my $rendezvous = shift || "catsock";
1201 socket(my $sock, PF_UNIX, SOCK_STREAM, 0) || die "socket: $!";
1202 connect($sock, sockaddr_un($rendezvous)) || die "connect: $!";
1203 while (defined(my $line = <$sock>)) {
1208 And here's a corresponding server. You don't have to worry about silly
1209 network terminators here because Unix domain sockets are guaranteed
1210 to be on the localhost, and thus everything works right.
1218 BEGIN { $ENV{PATH} = "/usr/bin:/bin" }
1219 sub spawn; # forward declaration
1220 sub logmsg { print "$0 $$: @_ at ", scalar localtime(), "\n" }
1222 my $NAME = "catsock";
1223 my $uaddr = sockaddr_un($NAME);
1224 my $proto = getprotobyname("tcp");
1226 socket(my $server, PF_UNIX, SOCK_STREAM, 0) || die "socket: $!";
1228 bind ($server, $uaddr) || die "bind: $!";
1229 listen($server, SOMAXCONN) || die "listen: $!";
1231 logmsg "server started on $NAME";
1235 use POSIX ":sys_wait_h";
1238 while (($waitedpid = waitpid(-1, WNOHANG)) > 0) {
1239 logmsg "reaped $waitedpid" . ($? ? " with exit $?" : "");
1241 $SIG{CHLD} = \&REAPER; # loathe SysV
1244 $SIG{CHLD} = \&REAPER;
1247 for ( $waitedpid = 0;
1248 accept(my $client, $server) || $waitedpid;
1249 $waitedpid = 0, close $client)
1252 logmsg "connection on $NAME";
1253 spawn $client, sub {
1254 print "Hello there, it's now ", scalar localtime(), "\n";
1255 exec("/usr/games/fortune") || die "can't exec fortune: $!";
1260 my $client = shift();
1261 my $coderef = shift();
1263 unless (@_ == 0 && $coderef && ref($coderef) eq "CODE") {
1264 confess "usage: spawn CLIENT CODEREF";
1268 unless (defined($pid = fork())) {
1269 logmsg "cannot fork: $!";
1273 logmsg "begat $pid";
1274 return; # I'm the parent
1277 # I'm the child -- go spawn
1280 open(STDIN, "<&", $client)
1281 || die "can't dup client to stdin";
1282 open(STDOUT, ">&", $client)
1283 || die "can't dup client to stdout";
1284 ## open(STDERR, ">&", STDOUT)
1285 ## || die "can't dup stdout to stderr";
1289 As you see, it's remarkably similar to the Internet domain TCP server, so
1290 much so, in fact, that we've omitted several duplicate functions--spawn(),
1291 logmsg(), ctime(), and REAPER()--which are the same as in the other server.
1293 So why would you ever want to use a Unix domain socket instead of a
1294 simpler named pipe? Because a named pipe doesn't give you sessions. You
1295 can't tell one process's data from another's. With socket programming,
1296 you get a separate session for each client; that's why accept() takes two
1299 For example, let's say that you have a long-running database server daemon
1300 that you want folks to be able to access from the Web, but only
1301 if they go through a CGI interface. You'd have a small, simple CGI
1302 program that does whatever checks and logging you feel like, and then acts
1303 as a Unix-domain client and connects to your private server.
1305 =head1 TCP Clients with IO::Socket
1307 For those preferring a higher-level interface to socket programming, the
1308 IO::Socket module provides an object-oriented approach. If for some reason
1309 you lack this module, you can just fetch IO::Socket from CPAN, where you'll also
1310 find modules providing easy interfaces to the following systems: DNS, FTP,
1311 Ident (RFC 931), NIS and NISPlus, NNTP, Ping, POP3, SMTP, SNMP, SSLeay,
1312 Telnet, and Time--to name just a few.
1314 =head2 A Simple Client
1316 Here's a client that creates a TCP connection to the "daytime"
1317 service at port 13 of the host name "localhost" and prints out everything
1318 that the server there cares to provide.
1324 my $remote = IO::Socket::INET->new(
1326 PeerAddr => "localhost",
1327 PeerPort => "daytime(13)",
1329 || die "can't connect to daytime service on localhost";
1330 while (<$remote>) { print }
1332 When you run this program, you should get something back that
1335 Wed May 14 08:40:46 MDT 1997
1337 Here are what those parameters to the new() constructor mean:
1343 This is which protocol to use. In this case, the socket handle returned
1344 will be connected to a TCP socket, because we want a stream-oriented
1345 connection, that is, one that acts pretty much like a plain old file.
1346 Not all sockets are this of this type. For example, the UDP protocol
1347 can be used to make a datagram socket, used for message-passing.
1351 This is the name or Internet address of the remote host the server is
1352 running on. We could have specified a longer name like C<"www.perl.com">,
1353 or an address like C<"207.171.7.72">. For demonstration purposes, we've
1354 used the special hostname C<"localhost">, which should always mean the
1355 current machine you're running on. The corresponding Internet address
1356 for localhost is C<"127.0.0.1">, if you'd rather use that.
1360 This is the service name or port number we'd like to connect to.
1361 We could have gotten away with using just C<"daytime"> on systems with a
1362 well-configured system services file,[FOOTNOTE: The system services file
1363 is found in I</etc/services> under Unixy systems.] but here we've specified the
1364 port number (13) in parentheses. Using just the number would have also
1365 worked, but numeric literals make careful programmers nervous.
1369 =head2 A Webget Client
1371 Here's a simple client that takes a remote host to fetch a document
1372 from, and then a list of files to get from that host. This is a
1373 more interesting client than the previous one because it first sends
1374 something to the server before fetching the server's response.
1380 unless (@ARGV > 1) { die "usage: $0 host url ..." }
1381 my $host = shift(@ARGV);
1382 my $EOL = "\015\012";
1383 my $BLANK = $EOL x 2;
1384 for my $document (@ARGV) {
1385 my $remote = IO::Socket::INET->new( Proto => "tcp",
1387 PeerPort => "http(80)",
1388 ) || die "cannot connect to httpd on $host";
1389 $remote->autoflush(1);
1390 print $remote "GET $document HTTP/1.0" . $BLANK;
1391 while ( <$remote> ) { print }
1395 The web server handling the HTTP service is assumed to be at
1396 its standard port, number 80. If the server you're trying to
1397 connect to is at a different port, like 1080 or 8080, you should specify it
1398 as the named-parameter pair, C<< PeerPort => 8080 >>. The C<autoflush>
1399 method is used on the socket because otherwise the system would buffer
1400 up the output we sent it. (If you're on a prehistoric Mac, you'll also
1401 need to change every C<"\n"> in your code that sends data over the network
1402 to be a C<"\015\012"> instead.)
1404 Connecting to the server is only the first part of the process: once you
1405 have the connection, you have to use the server's language. Each server
1406 on the network has its own little command language that it expects as
1407 input. The string that we send to the server starting with "GET" is in
1408 HTTP syntax. In this case, we simply request each specified document.
1409 Yes, we really are making a new connection for each document, even though
1410 it's the same host. That's the way you always used to have to speak HTTP.
1411 Recent versions of web browsers may request that the remote server leave
1412 the connection open a little while, but the server doesn't have to honor
1415 Here's an example of running that program, which we'll call I<webget>:
1417 % webget www.perl.com /guanaco.html
1418 HTTP/1.1 404 File Not Found
1419 Date: Thu, 08 May 1997 18:02:32 GMT
1420 Server: Apache/1.2b6
1422 Content-type: text/html
1424 <HEAD><TITLE>404 File Not Found</TITLE></HEAD>
1425 <BODY><H1>File Not Found</H1>
1426 The requested URL /guanaco.html was not found on this server.<P>
1429 Ok, so that's not very interesting, because it didn't find that
1430 particular document. But a long response wouldn't have fit on this page.
1432 For a more featureful version of this program, you should look to
1433 the I<lwp-request> program included with the LWP modules from CPAN.
1435 =head2 Interactive Client with IO::Socket
1437 Well, that's all fine if you want to send one command and get one answer,
1438 but what about setting up something fully interactive, somewhat like
1439 the way I<telnet> works? That way you can type a line, get the answer,
1440 type a line, get the answer, etc.
1442 This client is more complicated than the two we've done so far, but if
1443 you're on a system that supports the powerful C<fork> call, the solution
1444 isn't that rough. Once you've made the connection to whatever service
1445 you'd like to chat with, call C<fork> to clone your process. Each of
1446 these two identical process has a very simple job to do: the parent
1447 copies everything from the socket to standard output, while the child
1448 simultaneously copies everything from standard input to the socket.
1449 To accomplish the same thing using just one process would be I<much>
1450 harder, because it's easier to code two processes to do one thing than it
1451 is to code one process to do two things. (This keep-it-simple principle
1452 a cornerstones of the Unix philosophy, and good software engineering as
1453 well, which is probably why it's spread to other systems.)
1462 unless (@ARGV == 2) { die "usage: $0 host port" }
1463 my ($host, $port) = @ARGV;
1465 # create a tcp connection to the specified host and port
1466 my $handle = IO::Socket::INET->new(Proto => "tcp",
1469 || die "can't connect to port $port on $host: $!";
1471 $handle->autoflush(1); # so output gets there right away
1472 print STDERR "[Connected to $host:$port]\n";
1474 # split the program into two processes, identical twins
1475 die "can't fork: $!" unless defined(my $kidpid = fork());
1477 # the if{} block runs only in the parent process
1479 # copy the socket to standard output
1480 while (defined (my $line = <$handle>)) {
1483 kill("TERM", $kidpid); # send SIGTERM to child
1485 # the else{} block runs only in the child process
1487 # copy standard input to the socket
1488 while (defined (my $line = <STDIN>)) {
1489 print $handle $line;
1491 exit(0); # just in case
1494 The C<kill> function in the parent's C<if> block is there to send a
1495 signal to our child process, currently running in the C<else> block,
1496 as soon as the remote server has closed its end of the connection.
1498 If the remote server sends data a byte at time, and you need that
1499 data immediately without waiting for a newline (which might not happen),
1500 you may wish to replace the C<while> loop in the parent with the
1504 while (sysread($handle, $byte, 1) == 1) {
1508 Making a system call for each byte you want to read is not very efficient
1509 (to put it mildly) but is the simplest to explain and works reasonably
1512 =head1 TCP Servers with IO::Socket
1514 As always, setting up a server is little bit more involved than running a client.
1515 The model is that the server creates a special kind of socket that
1516 does nothing but listen on a particular port for incoming connections.
1517 It does this by calling the C<< IO::Socket::INET->new() >> method with
1518 slightly different arguments than the client did.
1524 This is which protocol to use. Like our clients, we'll
1525 still specify C<"tcp"> here.
1530 port in the C<LocalPort> argument, which we didn't do for the client.
1531 This is service name or port number for which you want to be the
1532 server. (Under Unix, ports under 1024 are restricted to the
1533 superuser.) In our sample, we'll use port 9000, but you can use
1534 any port that's not currently in use on your system. If you try
1535 to use one already in used, you'll get an "Address already in use"
1536 message. Under Unix, the C<netstat -a> command will show
1537 which services current have servers.
1541 The C<Listen> parameter is set to the maximum number of
1542 pending connections we can accept until we turn away incoming clients.
1543 Think of it as a call-waiting queue for your telephone.
1544 The low-level Socket module has a special symbol for the system maximum, which
1549 The C<Reuse> parameter is needed so that we restart our server
1550 manually without waiting a few minutes to allow system buffers to
1555 Once the generic server socket has been created using the parameters
1556 listed above, the server then waits for a new client to connect
1557 to it. The server blocks in the C<accept> method, which eventually accepts a
1558 bidirectional connection from the remote client. (Make sure to autoflush
1559 this handle to circumvent buffering.)
1561 To add to user-friendliness, our server prompts the user for commands.
1562 Most servers don't do this. Because of the prompt without a newline,
1563 you'll have to use the C<sysread> variant of the interactive client above.
1565 This server accepts one of five different commands, sending output back to
1566 the client. Unlike most network servers, this one handles only one
1567 incoming client at a time. Multitasking servers are covered in
1568 Chapter 16 of the Camel.
1576 use Net::hostent; # for OOish version of gethostbyaddr
1578 my $PORT = 9000; # pick something not in use
1580 my $server = IO::Socket::INET->new( Proto => "tcp",
1582 Listen => SOMAXCONN,
1585 die "can't setup server" unless $server;
1586 print "[Server $0 accepting clients]\n";
1588 while (my $client = $server->accept()) {
1589 $client->autoflush(1);
1590 print $client "Welcome to $0; type help for command list.\n";
1591 my $hostinfo = gethostbyaddr($client->peeraddr);
1592 printf "[Connect from %s]\n",
1593 $hostinfo ? $hostinfo->name : $client->peerhost;
1594 print $client "Command? ";
1595 while ( <$client>) {
1596 next unless /\S/; # blank line
1597 if (/quit|exit/i) { last }
1598 elsif (/date|time/i) { printf $client "%s\n", scalar localtime() }
1599 elsif (/who/i ) { print $client `who 2>&1` }
1600 elsif (/cookie/i ) { print $client `/usr/games/fortune 2>&1` }
1601 elsif (/motd/i ) { print $client `cat /etc/motd 2>&1` }
1603 print $client "Commands: quit date who cookie motd\n";
1606 print $client "Command? ";
1611 =head1 UDP: Message Passing
1613 Another kind of client-server setup is one that uses not connections, but
1614 messages. UDP communications involve much lower overhead but also provide
1615 less reliability, as there are no promises that messages will arrive at
1616 all, let alone in order and unmangled. Still, UDP offers some advantages
1617 over TCP, including being able to "broadcast" or "multicast" to a whole
1618 bunch of destination hosts at once (usually on your local subnet). If you
1619 find yourself overly concerned about reliability and start building checks
1620 into your message system, then you probably should use just TCP to start
1623 UDP datagrams are I<not> a bytestream and should not be treated as such.
1624 This makes using I/O mechanisms with internal buffering like stdio (i.e.
1625 print() and friends) especially cumbersome. Use syswrite(), or better
1626 send(), like in the example below.
1628 Here's a UDP program similar to the sample Internet TCP client given
1629 earlier. However, instead of checking one host at a time, the UDP version
1630 will check many of them asynchronously by simulating a multicast and then
1631 using select() to do a timed-out wait for I/O. To do something similar
1632 with TCP, you'd have to use a different socket handle for each host.
1640 my $SECS_OF_70_YEARS = 2_208_988_800;
1642 my $iaddr = gethostbyname(hostname());
1643 my $proto = getprotobyname("udp");
1644 my $port = getservbyname("time", "udp");
1645 my $paddr = sockaddr_in(0, $iaddr); # 0 means let kernel pick
1647 socket(my $socket, PF_INET, SOCK_DGRAM, $proto) || die "socket: $!";
1648 bind($socket, $paddr) || die "bind: $!";
1651 printf "%-12s %8s %s\n", "localhost", 0, scalar localtime();
1653 for my $host (@ARGV) {
1655 my $hisiaddr = inet_aton($host) || die "unknown host";
1656 my $hispaddr = sockaddr_in($port, $hisiaddr);
1657 defined(send($socket, 0, 0, $hispaddr)) || die "send $host: $!";
1660 my $rout = my $rin = "";
1661 vec($rin, fileno($socket), 1) = 1;
1663 # timeout after 10.0 seconds
1664 while ($count && select($rout = $rin, undef, undef, 10.0)) {
1666 my $hispaddr = recv($socket, $rtime, 4, 0) || die "recv: $!";
1667 my ($port, $hisiaddr) = sockaddr_in($hispaddr);
1668 my $host = gethostbyaddr($hisiaddr, AF_INET);
1669 my $histime = unpack("N", $rtime) - $SECS_OF_70_YEARS;
1670 printf "%-12s ", $host;
1671 printf "%8d %s\n", $histime - time(), scalar localtime($histime);
1675 This example does not include any retries and may consequently fail to
1676 contact a reachable host. The most prominent reason for this is congestion
1677 of the queues on the sending host if the number of hosts to contact is
1682 While System V IPC isn't so widely used as sockets, it still has some
1683 interesting uses. However, you cannot use SysV IPC or Berkeley mmap() to
1684 have a variable shared amongst several processes. That's because Perl
1685 would reallocate your string when you weren't wanting it to. You might
1686 look into the C<IPC::Shareable> or C<threads::shared> modules for that.
1688 Here's a small example showing shared memory usage.
1690 use IPC::SysV qw(IPC_PRIVATE IPC_RMID S_IRUSR S_IWUSR);
1693 my $id = shmget(IPC_PRIVATE, $size, S_IRUSR | S_IWUSR);
1694 defined($id) || die "shmget: $!";
1695 print "shm key $id\n";
1697 my $message = "Message #1";
1698 shmwrite($id, $message, 0, 60) || die "shmwrite: $!";
1699 print "wrote: '$message'\n";
1700 shmread($id, my $buff, 0, 60) || die "shmread: $!";
1701 print "read : '$buff'\n";
1703 # the buffer of shmread is zero-character end-padded.
1704 substr($buff, index($buff, "\0")) = "";
1705 print "un" unless $buff eq $message;
1708 print "deleting shm $id\n";
1709 shmctl($id, IPC_RMID, 0) || die "shmctl: $!";
1711 Here's an example of a semaphore:
1713 use IPC::SysV qw(IPC_CREAT);
1716 my $id = semget($IPC_KEY, 10, 0666 | IPC_CREAT);
1717 defined($id) || die "semget: $!";
1718 print "sem id $id\n";
1720 Put this code in a separate file to be run in more than one process.
1721 Call the file F<take>:
1723 # create a semaphore
1726 my $id = semget($IPC_KEY, 0, 0);
1727 defined($id) || die "semget: $!";
1733 # wait for semaphore to be zero
1735 my $opstring1 = pack("s!s!s!", $semnum, $semop, $semflag);
1737 # Increment the semaphore count
1739 my $opstring2 = pack("s!s!s!", $semnum, $semop, $semflag);
1740 my $opstring = $opstring1 . $opstring2;
1742 semop($id, $opstring) || die "semop: $!";
1744 Put this code in a separate file to be run in more than one process.
1745 Call this file F<give>:
1747 # "give" the semaphore
1748 # run this in the original process and you will see
1749 # that the second process continues
1752 my $id = semget($IPC_KEY, 0, 0);
1753 die unless defined($id);
1758 # Decrement the semaphore count
1760 my $opstring = pack("s!s!s!", $semnum, $semop, $semflag);
1762 semop($id, $opstring) || die "semop: $!";
1764 The SysV IPC code above was written long ago, and it's definitely
1765 clunky looking. For a more modern look, see the IPC::SysV module.
1767 A small example demonstrating SysV message queues:
1769 use IPC::SysV qw(IPC_PRIVATE IPC_RMID IPC_CREAT S_IRUSR S_IWUSR);
1771 my $id = msgget(IPC_PRIVATE, IPC_CREAT | S_IRUSR | S_IWUSR);
1772 defined($id) || die "msgget failed: $!";
1774 my $sent = "message";
1775 my $type_sent = 1234;
1777 msgsnd($id, pack("l! a*", $type_sent, $sent), 0)
1778 || die "msgsnd failed: $!";
1780 msgrcv($id, my $rcvd_buf, 60, 0, 0)
1781 || die "msgrcv failed: $!";
1783 my($type_rcvd, $rcvd) = unpack("l! a*", $rcvd_buf);
1785 if ($rcvd eq $sent) {
1791 msgctl($id, IPC_RMID, 0) || die "msgctl failed: $!\n";
1795 Most of these routines quietly but politely return C<undef> when they
1796 fail instead of causing your program to die right then and there due to
1797 an uncaught exception. (Actually, some of the new I<Socket> conversion
1798 functions do croak() on bad arguments.) It is therefore essential to
1799 check return values from these functions. Always begin your socket
1800 programs this way for optimal success, and don't forget to add the B<-T>
1801 taint-checking flag to the C<#!> line for servers:
1811 These routines all create system-specific portability problems. As noted
1812 elsewhere, Perl is at the mercy of your C libraries for much of its system
1813 behavior. It's probably safest to assume broken SysV semantics for
1814 signals and to stick with simple TCP and UDP socket operations; e.g., don't
1815 try to pass open file descriptors over a local UDP datagram socket if you
1816 want your code to stand a chance of being portable.
1820 Tom Christiansen, with occasional vestiges of Larry Wall's original
1821 version and suggestions from the Perl Porters.
1825 There's a lot more to networking than this, but this should get you
1828 For intrepid programmers, the indispensable textbook is I<Unix Network
1829 Programming, 2nd Edition, Volume 1> by W. Richard Stevens (published by
1830 Prentice-Hall). Most books on networking address the subject from the
1831 perspective of a C programmer; translation to Perl is left as an exercise
1834 The IO::Socket(3) manpage describes the object library, and the Socket(3)
1835 manpage describes the low-level interface to sockets. Besides the obvious
1836 functions in L<perlfunc>, you should also check out the F<modules> file at
1837 your nearest CPAN site, especially
1838 L<http://www.cpan.org/modules/00modlist.long.html#ID5_Networking_>.
1839 See L<perlmodlib> or best yet, the F<Perl FAQ> for a description
1840 of what CPAN is and where to get it if the previous link doesn't work
1843 Section 5 of CPAN's F<modules> file is devoted to "Networking, Device
1844 Control (modems), and Interprocess Communication", and contains numerous
1845 unbundled modules numerous networking modules, Chat and Expect operations,
1846 CGI programming, DCE, FTP, IPC, NNTP, Proxy, Ptty, RPC, SNMP, SMTP, Telnet,
1847 Threads, and ToolTalk--to name just a few.