ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/IO-AIO/AIO.pm

Comparing IO-AIO/AIO.pm (file contents):
Revision 1.6 by root, Sun Jul 10 22:19:48 2005 UTC vs.
Revision 1.237 by root, Thu Oct 11 03:20:52 2012 UTC

…		…
4		4
5	=head1 SYNOPSIS	5	=head1 SYNOPSIS
6		6
7	use IO::AIO;	7	use IO::AIO;
8		8
9	aio_open "/etc/passwd", O_RDONLY, 0, sub {	9	aio_open "/etc/passwd", IO::AIO::O_RDONLY, 0, sub {
10	my ($fh) = @_;	10	my $fh = shift
		11	or die "/etc/passwd: $!";
11	...	12	...
12	};	13	};
13		14
14	aio_unlink "/tmp/file", sub { };	15	aio_unlink "/tmp/file", sub { };
15		16
16	aio_read $fh, 30000, 1024, $buffer, 0, sub {	17	aio_read $fh, 30000, 1024, $buffer, 0, sub {
17	$_[0] >= 0 or die "read error: $!";	18	$_[0] > 0 or die "read error: $!";
18	};	19	};
19		20
20	# Event	21	# version 2+ has request and group objects
21	Event->io (fd => IO::AIO::poll_fileno,	22	use IO::AIO 2;
22	poll => 'r', async => 1,
23	cb => \&IO::AIO::poll_cb);
24		23
25	# Glib/Gtk2	24	aioreq_pri 4; # give next request a very high priority
26	add_watch Glib::IO IO::AIO::poll_fileno,	25	my $req = aio_unlink "/tmp/file", sub { };
27	\&IO::AIO::poll_cb;	26	$req->cancel; # cancel request if still in queue
28		27
29	# Tk	28	my $grp = aio_group sub { print "all stats done\n" };
30	Tk::Event::IO->fileevent (IO::AIO::poll_fileno, "",	29	add $grp aio_stat "..." for ...;
31	readable => \&IO::AIO::poll_cb);
32		30
33	=head1 DESCRIPTION	31	=head1 DESCRIPTION
34		32
35	This module implements asynchronous I/O using whatever means your	33	This module implements asynchronous I/O using whatever means your
36	operating system supports.	34	operating system supports. It is implemented as an interface to C<libeio>
		35	(L<http://software.schmorp.de/pkg/libeio.html>).
37		36
		37	Asynchronous means that operations that can normally block your program
		38	(e.g. reading from disk) will be done asynchronously: the operation
		39	will still block, but you can do something else in the meantime. This
		40	is extremely useful for programs that need to stay interactive even
		41	when doing heavy I/O (GUI programs, high performance network servers
		42	etc.), but can also be used to easily do operations in parallel that are
		43	normally done sequentially, e.g. stat'ing many files, which is much faster
		44	on a RAID volume or over NFS when you do a number of stat operations
		45	concurrently.
		46
		47	While most of this works on all types of file descriptors (for
		48	example sockets), using these functions on file descriptors that
		49	support nonblocking operation (again, sockets, pipes etc.) is
		50	very inefficient. Use an event loop for that (such as the L<EV>
		51	module): IO::AIO will naturally fit into such an event loop itself.
		52
38	Currently, a number of threads are started that execute your read/writes	53	In this version, a number of threads are started that execute your
39	and signal their completion. You don't need thread support in your libc or	54	requests and signal their completion. You don't need thread support
40	perl, and the threads created by this module will not be visible to the	55	in perl, and the threads created by this module will not be visible
41	pthreads library. In the future, this module might make use of the native	56	to perl. In the future, this module might make use of the native aio
42	aio functions available on many operating systems. However, they are often	57	functions available on many operating systems. However, they are often
43	not well-supported (Linux doesn't allow them on normal files currently,	58	not well-supported or restricted (GNU/Linux doesn't allow them on normal
44	for example), and they would only support aio_read and aio_write, so the	59	files currently, for example), and they would only support aio_read and
45	remaining functionality would have to be implemented using threads anyway.	60	aio_write, so the remaining functionality would have to be implemented
		61	using threads anyway.
46		62
47	Although the module will work with in the presence of other threads, it is	63	Although the module will work in the presence of other (Perl-) threads,
48	currently not reentrant, so use appropriate locking yourself.	64	it is currently not reentrant in any way, so use appropriate locking
		65	yourself, always call C<poll_cb> from within the same thread, or never
		66	call C<poll_cb> (or other C<aio_> functions) recursively.
		67
		68	=head2 EXAMPLE
		69
		70	This is a simple example that uses the EV module and loads
		71	F</etc/passwd> asynchronously:
		72
		73	use Fcntl;
		74	use EV;
		75	use IO::AIO;
		76
		77	# register the IO::AIO callback with EV
		78	my $aio_w = EV::io IO::AIO::poll_fileno, EV::READ, \&IO::AIO::poll_cb;
		79
		80	# queue the request to open /etc/passwd
		81	aio_open "/etc/passwd", IO::AIO::O_RDONLY, 0, sub {
		82	my $fh = shift
		83	or die "error while opening: $!";
		84
		85	# stat'ing filehandles is generally non-blocking
		86	my $size = -s $fh;
		87
		88	# queue a request to read the file
		89	my $contents;
		90	aio_read $fh, 0, $size, $contents, 0, sub {
		91	$_[0] == $size
		92	or die "short read: $!";
		93
		94	close $fh;
		95
		96	# file contents now in $contents
		97	print $contents;
		98
		99	# exit event loop and program
		100	EV::unloop;
		101	};
		102	};
		103
		104	# possibly queue up other requests, or open GUI windows,
		105	# check for sockets etc. etc.
		106
		107	# process events as long as there are some:
		108	EV::loop;
		109
		110	=head1 REQUEST ANATOMY AND LIFETIME
		111
		112	Every C<aio_*> function creates a request. which is a C data structure not
		113	directly visible to Perl.
		114
		115	If called in non-void context, every request function returns a Perl
		116	object representing the request. In void context, nothing is returned,
		117	which saves a bit of memory.
		118
		119	The perl object is a fairly standard ref-to-hash object. The hash contents
		120	are not used by IO::AIO so you are free to store anything you like in it.
		121
		122	During their existance, aio requests travel through the following states,
		123	in order:
		124
		125	=over 4
		126
		127	=item ready
		128
		129	Immediately after a request is created it is put into the ready state,
		130	waiting for a thread to execute it.
		131
		132	=item execute
		133
		134	A thread has accepted the request for processing and is currently
		135	executing it (e.g. blocking in read).
		136
		137	=item pending
		138
		139	The request has been executed and is waiting for result processing.
		140
		141	While request submission and execution is fully asynchronous, result
		142	processing is not and relies on the perl interpreter calling C<poll_cb>
		143	(or another function with the same effect).
		144
		145	=item result
		146
		147	The request results are processed synchronously by C<poll_cb>.
		148
		149	The C<poll_cb> function will process all outstanding aio requests by
		150	calling their callbacks, freeing memory associated with them and managing
		151	any groups they are contained in.
		152
		153	=item done
		154
		155	Request has reached the end of its lifetime and holds no resources anymore
		156	(except possibly for the Perl object, but its connection to the actual
		157	aio request is severed and calling its methods will either do nothing or
		158	result in a runtime error).
		159
		160	=back
49		161
50	=cut	162	=cut
51		163
52	package IO::AIO;	164	package IO::AIO;
53		165
		166	use Carp ();
		167
		168	use common::sense;
		169
54	use base 'Exporter';	170	use base 'Exporter';
55		171
56	use Fcntl ();
57
58	BEGIN {	172	BEGIN {
59	$VERSION = 0.2;	173	our $VERSION = '4.17';
60		174
61	@EXPORT = qw(aio_read aio_write aio_open aio_close aio_stat aio_lstat aio_unlink	175	our @AIO_REQ = qw(aio_sendfile aio_seek aio_read aio_write aio_open aio_close
62	aio_fsync aio_fdatasync aio_readahead);	176	aio_stat aio_lstat aio_unlink aio_rmdir aio_readdir aio_readdirx
63	@EXPORT_OK = qw(poll_fileno poll_cb min_parallel max_parallel max_outstanding nreqs);	177	aio_scandir aio_symlink aio_readlink aio_realpath aio_sync
		178	aio_fsync aio_syncfs aio_fdatasync aio_sync_file_range aio_allocate
		179	aio_pathsync aio_readahead aio_fiemap
		180	aio_rename aio_link aio_move aio_copy aio_group
		181	aio_nop aio_mknod aio_load aio_rmtree aio_mkdir aio_chown
		182	aio_chmod aio_utime aio_truncate
		183	aio_msync aio_mtouch aio_mlock aio_mlockall
		184	aio_statvfs
		185	aio_wd);
		186
		187	our @EXPORT = (@AIO_REQ, qw(aioreq_pri aioreq_nice));
		188	our @EXPORT_OK = qw(poll_fileno poll_cb poll_wait flush
		189	min_parallel max_parallel max_idle idle_timeout
		190	nreqs nready npending nthreads
		191	max_poll_time max_poll_reqs
		192	sendfile fadvise madvise
		193	mmap munmap munlock munlockall);
		194
		195	push @AIO_REQ, qw(aio_busy); # not exported
		196
		197	@IO::AIO::GRP::ISA = 'IO::AIO::REQ';
64		198
65	require XSLoader;	199	require XSLoader;
66	XSLoader::load IO::AIO, $VERSION;	200	XSLoader::load ("IO::AIO", $VERSION);
67	}	201	}
68		202
69	=head1 FUNCTIONS	203	=head1 FUNCTIONS
70		204
71	=head2 AIO FUNCTIONS	205	=head2 QUICK OVERVIEW
		206
		207	This section simply lists the prototypes most of the functions for
		208	quick reference. See the following sections for function-by-function
		209	documentation.
		210
		211	aio_wd $pathname, $callback->($wd)
		212	aio_open $pathname, $flags, $mode, $callback->($fh)
		213	aio_close $fh, $callback->($status)
		214	aio_seek $fh,$offset,$whence, $callback->($offs)
		215	aio_read $fh,$offset,$length, $data,$dataoffset, $callback->($retval)
		216	aio_write $fh,$offset,$length, $data,$dataoffset, $callback->($retval)
		217	aio_sendfile $out_fh, $in_fh, $in_offset, $length, $callback->($retval)
		218	aio_readahead $fh,$offset,$length, $callback->($retval)
		219	aio_stat $fh_or_path, $callback->($status)
		220	aio_lstat $fh, $callback->($status)
		221	aio_statvfs $fh_or_path, $callback->($statvfs)
		222	aio_utime $fh_or_path, $atime, $mtime, $callback->($status)
		223	aio_chown $fh_or_path, $uid, $gid, $callback->($status)
		224	aio_chmod $fh_or_path, $mode, $callback->($status)
		225	aio_truncate $fh_or_path, $offset, $callback->($status)
		226	aio_allocate $fh, $mode, $offset, $len, $callback->($status)
		227	aio_fiemap $fh, $start, $length, $flags, $count, $cb->(\@extents)
		228	aio_unlink $pathname, $callback->($status)
		229	aio_mknod $pathname, $mode, $dev, $callback->($status)
		230	aio_link $srcpath, $dstpath, $callback->($status)
		231	aio_symlink $srcpath, $dstpath, $callback->($status)
		232	aio_readlink $pathname, $callback->($link)
		233	aio_realpath $pathname, $callback->($link)
		234	aio_rename $srcpath, $dstpath, $callback->($status)
		235	aio_mkdir $pathname, $mode, $callback->($status)
		236	aio_rmdir $pathname, $callback->($status)
		237	aio_readdir $pathname, $callback->($entries)
		238	aio_readdirx $pathname, $flags, $callback->($entries, $flags)
		239	IO::AIO::READDIR_DENTS IO::AIO::READDIR_DIRS_FIRST
		240	IO::AIO::READDIR_STAT_ORDER IO::AIO::READDIR_FOUND_UNKNOWN
		241	aio_scandir $pathname, $maxreq, $callback->($dirs, $nondirs)
		242	aio_load $pathname, $data, $callback->($status)
		243	aio_copy $srcpath, $dstpath, $callback->($status)
		244	aio_move $srcpath, $dstpath, $callback->($status)
		245	aio_rmtree $pathname, $callback->($status)
		246	aio_sync $callback->($status)
		247	aio_syncfs $fh, $callback->($status)
		248	aio_fsync $fh, $callback->($status)
		249	aio_fdatasync $fh, $callback->($status)
		250	aio_sync_file_range $fh, $offset, $nbytes, $flags, $callback->($status)
		251	aio_pathsync $pathname, $callback->($status)
		252	aio_msync $scalar, $offset = 0, $length = undef, flags = 0, $callback->($status)
		253	aio_mtouch $scalar, $offset = 0, $length = undef, flags = 0, $callback->($status)
		254	aio_mlock $scalar, $offset = 0, $length = undef, $callback->($status)
		255	aio_mlockall $flags, $callback->($status)
		256	aio_group $callback->(...)
		257	aio_nop $callback->()
		258
		259	$prev_pri = aioreq_pri [$pri]
		260	aioreq_nice $pri_adjust
		261
		262	IO::AIO::poll_wait
		263	IO::AIO::poll_cb
		264	IO::AIO::poll
		265	IO::AIO::flush
		266	IO::AIO::max_poll_reqs $nreqs
		267	IO::AIO::max_poll_time $seconds
		268	IO::AIO::min_parallel $nthreads
		269	IO::AIO::max_parallel $nthreads
		270	IO::AIO::max_idle $nthreads
		271	IO::AIO::idle_timeout $seconds
		272	IO::AIO::max_outstanding $maxreqs
		273	IO::AIO::nreqs
		274	IO::AIO::nready
		275	IO::AIO::npending
		276
		277	IO::AIO::sendfile $ofh, $ifh, $offset, $count
		278	IO::AIO::fadvise $fh, $offset, $len, $advice
		279	IO::AIO::mmap $scalar, $length, $prot, $flags[, $fh[, $offset]]
		280	IO::AIO::munmap $scalar
		281	IO::AIO::madvise $scalar, $offset, $length, $advice
		282	IO::AIO::mprotect $scalar, $offset, $length, $protect
		283	IO::AIO::munlock $scalar, $offset = 0, $length = undef
		284	IO::AIO::munlockall
		285
		286	=head2 API NOTES
72		287
73	All the C<aio_*> calls are more or less thin wrappers around the syscall	288	All the C<aio_*> calls are more or less thin wrappers around the syscall
74	with the same name (sans C<aio_>). The arguments are similar or identical,	289	with the same name (sans C<aio_>). The arguments are similar or identical,
75	and they all accept an additional C<$callback> argument which must be	290	and they all accept an additional (and optional) C<$callback> argument
76	a code reference. This code reference will get called with the syscall	291	which must be a code reference. This code reference will be called after
		292	the syscall has been executed in an asynchronous fashion. The results
		293	of the request will be passed as arguments to the callback (and, if an
		294	error occured, in C<$!>) - for most requests the syscall return code (e.g.
77	return code (e.g. most syscalls return C<-1> on error, unlike perl, which	295	most syscalls return C<-1> on error, unlike perl, which usually delivers
78	usually delivers "false") as it's sole argument when the given syscall has	296	"false").
79	been executed asynchronously.
80		297
81	All functions that expect a filehandle will also accept a file descriptor.	298	Some requests (such as C<aio_readdir>) pass the actual results and
		299	communicate failures by passing C<undef>.
82		300
		301	All functions expecting a filehandle keep a copy of the filehandle
		302	internally until the request has finished.
		303
		304	All functions return request objects of type L<IO::AIO::REQ> that allow
		305	further manipulation of those requests while they are in-flight.
		306
83	The filenames you pass to these routines I<must> be absolute. The reason	307	The pathnames you pass to these routines I<should> be absolute. The
84	is that at the time the request is being executed, the current working	308	reason for this is that at the time the request is being executed, the
85	directory could have changed. Alternatively, you can make sure that you	309	current working directory could have changed. Alternatively, you can
86	never change the current working directory.	310	make sure that you never change the current working directory anywhere
		311	in the program and then use relative paths. You can also take advantage
		312	of IO::AIOs working directory abstraction, that lets you specify paths
		313	relative to some previously-opened "working directory object" - see the
		314	description of the C<IO::AIO::WD> class later in this document.
		315
		316	To encode pathnames as octets, either make sure you either: a) always pass
		317	in filenames you got from outside (command line, readdir etc.) without
		318	tinkering, b) are in your native filesystem encoding, c) use the Encode
		319	module and encode your pathnames to the locale (or other) encoding in
		320	effect in the user environment, d) use Glib::filename_from_unicode on
		321	unicode filenames or e) use something else to ensure your scalar has the
		322	correct contents.
		323
		324	This works, btw. independent of the internal UTF-8 bit, which IO::AIO
		325	handles correctly whether it is set or not.
		326
		327	=head2 AIO REQUEST FUNCTIONS
87		328
88	=over 4	329	=over 4
89		330
		331	=item $prev_pri = aioreq_pri [$pri]
		332
		333	Returns the priority value that would be used for the next request and, if
		334	C<$pri> is given, sets the priority for the next aio request.
		335
		336	The default priority is C<0>, the minimum and maximum priorities are C<-4>
		337	and C<4>, respectively. Requests with higher priority will be serviced
		338	first.
		339
		340	The priority will be reset to C<0> after each call to one of the C<aio_*>
		341	functions.
		342
		343	Example: open a file with low priority, then read something from it with
		344	higher priority so the read request is serviced before other low priority
		345	open requests (potentially spamming the cache):
		346
		347	aioreq_pri -3;
		348	aio_open ..., sub {
		349	return unless $_[0];
		350
		351	aioreq_pri -2;
		352	aio_read $_[0], ..., sub {
		353	...
		354	};
		355	};
		356
		357
		358	=item aioreq_nice $pri_adjust
		359
		360	Similar to C<aioreq_pri>, but subtracts the given value from the current
		361	priority, so the effect is cumulative.
		362
		363
90	=item aio_open $pathname, $flags, $mode, $callback	364	=item aio_open $pathname, $flags, $mode, $callback->($fh)
91		365
92	Asynchronously open or create a file and call the callback with a newly	366	Asynchronously open or create a file and call the callback with a newly
93	created filehandle for the file.	367	created filehandle for the file (or C<undef> in case of an error).
94		368
95	The pathname passed to C<aio_open> must be absolute. See API NOTES, above,	369	The pathname passed to C<aio_open> must be absolute. See API NOTES, above,
96	for an explanation.	370	for an explanation.
97		371
98	The C<$mode> argument is a bitmask. See the C<Fcntl> module for a	372	The C<$flags> argument is a bitmask. See the C<Fcntl> module for a
99	list. They are the same as used in C<sysopen>.	373	list. They are the same as used by C<sysopen>.
		374
		375	Likewise, C<$mode> specifies the mode of the newly created file, if it
		376	didn't exist and C<O_CREAT> has been given, just like perl's C<sysopen>,
		377	except that it is mandatory (i.e. use C<0> if you don't create new files,
		378	and C<0666> or C<0777> if you do). Note that the C<$mode> will be modified
		379	by the umask in effect then the request is being executed, so better never
		380	change the umask.
100		381
101	Example:	382	Example:
102		383
103	aio_open "/etc/passwd", O_RDONLY, 0, sub {	384	aio_open "/etc/passwd", IO::AIO::O_RDONLY, 0, sub {
104	if ($_[0]) {	385	if ($_[0]) {
105	print "open successful, fh is $_[0]\n";	386	print "open successful, fh is $_[0]\n";
106	...	387	...
107	} else {	388	} else {
108	die "open failed: $!\n";	389	die "open failed: $!\n";
109	}	390	}
110	};	391	};
111		392
		393	In addition to all the common open modes/flags (C<O_RDONLY>, C<O_WRONLY>,
		394	C<O_RDWR>, C<O_CREAT>, C<O_TRUNC>, C<O_EXCL> and C<O_APPEND>), the
		395	following POSIX and non-POSIX constants are available (missing ones on
		396	your system are, as usual, C<0>):
		397
		398	C<O_ASYNC>, C<O_DIRECT>, C<O_NOATIME>, C<O_CLOEXEC>, C<O_NOCTTY>, C<O_NOFOLLOW>,
		399	C<O_NONBLOCK>, C<O_EXEC>, C<O_SEARCH>, C<O_DIRECTORY>, C<O_DSYNC>,
		400	C<O_RSYNC>, C<O_SYNC> and C<O_TTY_INIT>.
		401
		402
112	=item aio_close $fh, $callback	403	=item aio_close $fh, $callback->($status)
113		404
114	Asynchronously close a file and call the callback with the result	405	Asynchronously close a file and call the callback with the result
115	code. I<WARNING:> although accepted, you should not pass in a perl	406	code.
116	filehandle here, as perl will likely close the file descriptor itself when
117	the filehandle is destroyed. Normally, you can safely call perls C<close>
118	or just let filehandles go out of scope.
119		407
		408	Unfortunately, you can't do this to perl. Perl I<insists> very strongly on
		409	closing the file descriptor associated with the filehandle itself.
		410
		411	Therefore, C<aio_close> will not close the filehandle - instead it will
		412	use dup2 to overwrite the file descriptor with the write-end of a pipe
		413	(the pipe fd will be created on demand and will be cached).
		414
		415	Or in other words: the file descriptor will be closed, but it will not be
		416	free for reuse until the perl filehandle is closed.
		417
		418	=cut
		419
		420	=item aio_seek $fh, $offset, $whence, $callback->($offs)
		421
		422	Seeks the filehandle to the new C<$offset>, similarly to perl's
		423	C<sysseek>. The C<$whence> can use the traditional values (C<0> for
		424	C<IO::AIO::SEEK_SET>, C<1> for C<IO::AIO::SEEK_CUR> or C<2> for
		425	C<IO::AIO::SEEK_END>).
		426
		427	The resulting absolute offset will be passed to the callback, or C<-1> in
		428	case of an error.
		429
		430	In theory, the C<$whence> constants could be different than the
		431	corresponding values from L<Fcntl>, but perl guarantees they are the same,
		432	so don't panic.
		433
		434	As a GNU/Linux (and maybe Solaris) extension, also the constants
		435	C<IO::AIO::SEEK_DATA> and C<IO::AIO::SEEK_HOLE> are available, if they
		436	could be found. No guarantees about suitability for use in C<aio_seek> or
		437	Perl's C<sysseek> can be made though, although I would naively assume they
		438	"just work".
		439
120	=item aio_read $fh,$offset,$length, $data,$dataoffset,$callback	440	=item aio_read $fh,$offset,$length, $data,$dataoffset, $callback->($retval)
121		441
122	=item aio_write $fh,$offset,$length, $data,$dataoffset,$callback	442	=item aio_write $fh,$offset,$length, $data,$dataoffset, $callback->($retval)
123		443
124	Reads or writes C<length> bytes from the specified C<fh> and C<offset>	444	Reads or writes C<$length> bytes from or to the specified C<$fh> and
125	into the scalar given by C<data> and offset C<dataoffset> and calls the	445	C<$offset> into the scalar given by C<$data> and offset C<$dataoffset>
126	callback without the actual number of bytes read (or -1 on error, just	446	and calls the callback without the actual number of bytes read (or -1 on
127	like the syscall).	447	error, just like the syscall).
128		448
		449	C<aio_read> will, like C<sysread>, shrink or grow the C<$data> scalar to
		450	offset plus the actual number of bytes read.
		451
		452	If C<$offset> is undefined, then the current file descriptor offset will
		453	be used (and updated), otherwise the file descriptor offset will not be
		454	changed by these calls.
		455
		456	If C<$length> is undefined in C<aio_write>, use the remaining length of
		457	C<$data>.
		458
		459	If C<$dataoffset> is less than zero, it will be counted from the end of
		460	C<$data>.
		461
		462	The C<$data> scalar I<MUST NOT> be modified in any way while the request
		463	is outstanding. Modifying it can result in segfaults or World War III (if
		464	the necessary/optional hardware is installed).
		465
129	Example: Read 15 bytes at offset 7 into scalar C<$buffer>, strating at	466	Example: Read 15 bytes at offset 7 into scalar C<$buffer>, starting at
130	offset C<0> within the scalar:	467	offset C<0> within the scalar:
131		468
132	aio_read $fh, 7, 15, $buffer, 0, sub {	469	aio_read $fh, 7, 15, $buffer, 0, sub {
133	$_[0] >= 0 or die "read error: $!";	470	$_[0] > 0 or die "read error: $!";
134	print "read <$buffer>\n";	471	print "read $_[0] bytes: <$buffer>\n";
135	};	472	};
136		473
		474
		475	=item aio_sendfile $out_fh, $in_fh, $in_offset, $length, $callback->($retval)
		476
		477	Tries to copy C<$length> bytes from C<$in_fh> to C<$out_fh>. It starts
		478	reading at byte offset C<$in_offset>, and starts writing at the current
		479	file offset of C<$out_fh>. Because of that, it is not safe to issue more
		480	than one C<aio_sendfile> per C<$out_fh>, as they will interfere with each
		481	other. The same C<$in_fh> works fine though, as this function does not
		482	move or use the file offset of C<$in_fh>.
		483
		484	Please note that C<aio_sendfile> can read more bytes from C<$in_fh> than
		485	are written, and there is no way to find out how many more bytes have been
		486	read from C<aio_sendfile> alone, as C<aio_sendfile> only provides the
		487	number of bytes written to C<$out_fh>. Only if the result value equals
		488	C<$length> one can assume that C<$length> bytes have been read.
		489
		490	Unlike with other C<aio_> functions, it makes a lot of sense to use
		491	C<aio_sendfile> on non-blocking sockets, as long as one end (typically
		492	the C<$in_fh>) is a file - the file I/O will then be asynchronous, while
		493	the socket I/O will be non-blocking. Note, however, that you can run
		494	into a trap where C<aio_sendfile> reads some data with readahead, then
		495	fails to write all data, and when the socket is ready the next time, the
		496	data in the cache is already lost, forcing C<aio_sendfile> to again hit
		497	the disk. Explicit C<aio_read> + C<aio_write> let's you better control
		498	resource usage.
		499
		500	This call tries to make use of a native C<sendfile>-like syscall to
		501	provide zero-copy operation. For this to work, C<$out_fh> should refer to
		502	a socket, and C<$in_fh> should refer to an mmap'able file.
		503
		504	If a native sendfile cannot be found or it fails with C<ENOSYS>,
		505	C<EINVAL>, C<ENOTSUP>, C<EOPNOTSUPP>, C<EAFNOSUPPORT>, C<EPROTOTYPE> or
		506	C<ENOTSOCK>, it will be emulated, so you can call C<aio_sendfile> on any
		507	type of filehandle regardless of the limitations of the operating system.
		508
		509	As native sendfile syscalls (as practically any non-POSIX interface hacked
		510	together in a hurry to improve benchmark numbers) tend to be rather buggy
		511	on many systems, this implementation tries to work around some known bugs
		512	in Linux and FreeBSD kernels (probably others, too), but that might fail,
		513	so you really really should check the return value of C<aio_sendfile> -
		514	fewre bytes than expected might have been transferred.
		515
		516
137	=item aio_readahead $fh,$offset,$length, $callback	517	=item aio_readahead $fh,$offset,$length, $callback->($retval)
138		518
139	Asynchronously reads the specified byte range into the page cache, using
140	the C<readahead> syscall. If that syscall doesn't exist the status will be
141	C<-1> and C<$!> is set to ENOSYS.
142
143	readahead() populates the page cache with data from a file so that	519	C<aio_readahead> populates the page cache with data from a file so that
144	subsequent reads from that file will not block on disk I/O. The C<$offset>	520	subsequent reads from that file will not block on disk I/O. The C<$offset>
145	argument specifies the starting point from which data is to be read and	521	argument specifies the starting point from which data is to be read and
146	C<$length> specifies the number of bytes to be read. I/O is performed in	522	C<$length> specifies the number of bytes to be read. I/O is performed in
147	whole pages, so that offset is effectively rounded down to a page boundary	523	whole pages, so that offset is effectively rounded down to a page boundary
148	and bytes are read up to the next page boundary greater than or equal to	524	and bytes are read up to the next page boundary greater than or equal to
149	(off-set+length). aio_readahead() does not read beyond the end of the	525	(off-set+length). C<aio_readahead> does not read beyond the end of the
150	file. The current file offset of the file is left unchanged.	526	file. The current file offset of the file is left unchanged.
151		527
		528	If that syscall doesn't exist (likely if your OS isn't Linux) it will be
		529	emulated by simply reading the data, which would have a similar effect.
		530
		531
152	=item aio_stat $fh_or_path, $callback	532	=item aio_stat $fh_or_path, $callback->($status)
153		533
154	=item aio_lstat $fh, $callback	534	=item aio_lstat $fh, $callback->($status)
155		535
156	Works like perl's C<stat> or C<lstat> in void context. The callback will	536	Works like perl's C<stat> or C<lstat> in void context. The callback will
157	be called after the stat and the results will be available using C<stat _>	537	be called after the stat and the results will be available using C<stat _>
158	or C<-s _> etc...	538	or C<-s _> etc...
159		539
…		…
161	for an explanation.	541	for an explanation.
162		542
163	Currently, the stats are always 64-bit-stats, i.e. instead of returning an	543	Currently, the stats are always 64-bit-stats, i.e. instead of returning an
164	error when stat'ing a large file, the results will be silently truncated	544	error when stat'ing a large file, the results will be silently truncated
165	unless perl itself is compiled with large file support.	545	unless perl itself is compiled with large file support.
		546
		547	To help interpret the mode and dev/rdev stat values, IO::AIO offers the
		548	following constants and functions (if not implemented, the constants will
		549	be C<0> and the functions will either C<croak> or fall back on traditional
		550	behaviour).
		551
		552	C<S_IFMT>, C<S_IFIFO>, C<S_IFCHR>, C<S_IFBLK>, C<S_IFLNK>, C<S_IFREG>,
		553	C<S_IFDIR>, C<S_IFWHT>, C<S_IFSOCK>, C<IO::AIO::major $dev_t>,
		554	C<IO::AIO::minor $dev_t>, C<IO::AIO::makedev $major, $minor>.
166		555
167	Example: Print the length of F</etc/passwd>:	556	Example: Print the length of F</etc/passwd>:
168		557
169	aio_stat "/etc/passwd", sub {	558	aio_stat "/etc/passwd", sub {
170	$_[0] and die "stat failed: $!";	559	$_[0] and die "stat failed: $!";
171	print "size is ", -s _, "\n";	560	print "size is ", -s _, "\n";
172	};	561	};
173		562
		563
		564	=item aio_statvfs $fh_or_path, $callback->($statvfs)
		565
		566	Works like the POSIX C<statvfs> or C<fstatvfs> syscalls, depending on
		567	whether a file handle or path was passed.
		568
		569	On success, the callback is passed a hash reference with the following
		570	members: C<bsize>, C<frsize>, C<blocks>, C<bfree>, C<bavail>, C<files>,
		571	C<ffree>, C<favail>, C<fsid>, C<flag> and C<namemax>. On failure, C<undef>
		572	is passed.
		573
		574	The following POSIX IO::AIO::ST_* constants are defined: C<ST_RDONLY> and
		575	C<ST_NOSUID>.
		576
		577	The following non-POSIX IO::AIO::ST_* flag masks are defined to
		578	their correct value when available, or to C<0> on systems that do
		579	not support them: C<ST_NODEV>, C<ST_NOEXEC>, C<ST_SYNCHRONOUS>,
		580	C<ST_MANDLOCK>, C<ST_WRITE>, C<ST_APPEND>, C<ST_IMMUTABLE>, C<ST_NOATIME>,
		581	C<ST_NODIRATIME> and C<ST_RELATIME>.
		582
		583	Example: stat C</wd> and dump out the data if successful.
		584
		585	aio_statvfs "/wd", sub {
		586	my $f = $_[0]
		587	or die "statvfs: $!";
		588
		589	use Data::Dumper;
		590	say Dumper $f;
		591	};
		592
		593	# result:
		594	{
		595	bsize => 1024,
		596	bfree => 4333064312,
		597	blocks => 10253828096,
		598	files => 2050765568,
		599	flag => 4096,
		600	favail => 2042092649,
		601	bavail => 4333064312,
		602	ffree => 2042092649,
		603	namemax => 255,
		604	frsize => 1024,
		605	fsid => 1810
		606	}
		607
		608	Here is a (likely partial) list of fsid values used by Linux - it is safe
		609	to hardcode these when the $^O is C<linux>:
		610
		611	0x0000adf5 adfs
		612	0x0000adff affs
		613	0x5346414f afs
		614	0x09041934 anon-inode filesystem
		615	0x00000187 autofs
		616	0x42465331 befs
		617	0x1badface bfs
		618	0x42494e4d binfmt_misc
		619	0x9123683e btrfs
		620	0x0027e0eb cgroupfs
		621	0xff534d42 cifs
		622	0x73757245 coda
		623	0x012ff7b7 coh
		624	0x28cd3d45 cramfs
		625	0x453dcd28 cramfs-wend (wrong endianness)
		626	0x64626720 debugfs
		627	0x00001373 devfs
		628	0x00001cd1 devpts
		629	0x0000f15f ecryptfs
		630	0x00414a53 efs
		631	0x0000137d ext
		632	0x0000ef53 ext2/ext3
		633	0x0000ef51 ext2
		634	0x00004006 fat
		635	0x65735546 fuseblk
		636	0x65735543 fusectl
		637	0x0bad1dea futexfs
		638	0x01161970 gfs2
		639	0x47504653 gpfs
		640	0x00004244 hfs
		641	0xf995e849 hpfs
		642	0x958458f6 hugetlbfs
		643	0x2bad1dea inotifyfs
		644	0x00009660 isofs
		645	0x000072b6 jffs2
		646	0x3153464a jfs
		647	0x6b414653 k-afs
		648	0x0bd00bd0 lustre
		649	0x0000137f minix
		650	0x0000138f minix 30 char names
		651	0x00002468 minix v2
		652	0x00002478 minix v2 30 char names
		653	0x00004d5a minix v3
		654	0x19800202 mqueue
		655	0x00004d44 msdos
		656	0x0000564c novell
		657	0x00006969 nfs
		658	0x6e667364 nfsd
		659	0x00003434 nilfs
		660	0x5346544e ntfs
		661	0x00009fa1 openprom
		662	0x7461636F ocfs2
		663	0x00009fa0 proc
		664	0x6165676c pstorefs
		665	0x0000002f qnx4
		666	0x858458f6 ramfs
		667	0x52654973 reiserfs
		668	0x00007275 romfs
		669	0x67596969 rpc_pipefs
		670	0x73636673 securityfs
		671	0xf97cff8c selinux
		672	0x0000517b smb
		673	0x534f434b sockfs
		674	0x73717368 squashfs
		675	0x62656572 sysfs
		676	0x012ff7b6 sysv2
		677	0x012ff7b5 sysv4
		678	0x01021994 tmpfs
		679	0x15013346 udf
		680	0x00011954 ufs
		681	0x54190100 ufs byteswapped
		682	0x00009fa2 usbdevfs
		683	0x01021997 v9fs
		684	0xa501fcf5 vxfs
		685	0xabba1974 xenfs
		686	0x012ff7b4 xenix
		687	0x58465342 xfs
		688	0x012fd16d xia
		689
		690	=item aio_utime $fh_or_path, $atime, $mtime, $callback->($status)
		691
		692	Works like perl's C<utime> function (including the special case of $atime
		693	and $mtime being undef). Fractional times are supported if the underlying
		694	syscalls support them.
		695
		696	When called with a pathname, uses utimes(2) if available, otherwise
		697	utime(2). If called on a file descriptor, uses futimes(2) if available,
		698	otherwise returns ENOSYS, so this is not portable.
		699
		700	Examples:
		701
		702	# set atime and mtime to current time (basically touch(1)):
		703	aio_utime "path", undef, undef;
		704	# set atime to current time and mtime to beginning of the epoch:
		705	aio_utime "path", time, undef; # undef==0
		706
		707
		708	=item aio_chown $fh_or_path, $uid, $gid, $callback->($status)
		709
		710	Works like perl's C<chown> function, except that C<undef> for either $uid
		711	or $gid is being interpreted as "do not change" (but -1 can also be used).
		712
		713	Examples:
		714
		715	# same as "chown root path" in the shell:
		716	aio_chown "path", 0, -1;
		717	# same as above:
		718	aio_chown "path", 0, undef;
		719
		720
		721	=item aio_truncate $fh_or_path, $offset, $callback->($status)
		722
		723	Works like truncate(2) or ftruncate(2).
		724
		725
		726	=item aio_allocate $fh, $mode, $offset, $len, $callback->($status)
		727
		728	Allocates or freed disk space according to the C<$mode> argument. See the
		729	linux C<fallocate> docuemntation for details.
		730
		731	C<$mode> can currently be C<0> or C<IO::AIO::FALLOC_FL_KEEP_SIZE>
		732	to allocate space, or C<IO::AIO::FALLOC_FL_PUNCH_HOLE |
		733	IO::AIO::FALLOC_FL_KEEP_SIZE>, to deallocate a file range.
		734
		735	The file system block size used by C<fallocate> is presumably the
		736	C<f_bsize> returned by C<statvfs>.
		737
		738	If C<fallocate> isn't available or cannot be emulated (currently no
		739	emulation will be attempted), passes C<-1> and sets C<$!> to C<ENOSYS>.
		740
		741
		742	=item aio_chmod $fh_or_path, $mode, $callback->($status)
		743
		744	Works like perl's C<chmod> function.
		745
		746
174	=item aio_unlink $pathname, $callback	747	=item aio_unlink $pathname, $callback->($status)
175		748
176	Asynchronously unlink (delete) a file and call the callback with the	749	Asynchronously unlink (delete) a file and call the callback with the
177	result code.	750	result code.
178		751
		752
		753	=item aio_mknod $pathname, $mode, $dev, $callback->($status)
		754
		755	[EXPERIMENTAL]
		756
		757	Asynchronously create a device node (or fifo). See mknod(2).
		758
		759	The only (POSIX-) portable way of calling this function is:
		760
		761	aio_mknod $pathname, IO::AIO::S_IFIFO | $mode, 0, sub { ...
		762
		763	See C<aio_stat> for info about some potentially helpful extra constants
		764	and functions.
		765
		766	=item aio_link $srcpath, $dstpath, $callback->($status)
		767
		768	Asynchronously create a new link to the existing object at C<$srcpath> at
		769	the path C<$dstpath> and call the callback with the result code.
		770
		771
		772	=item aio_symlink $srcpath, $dstpath, $callback->($status)
		773
		774	Asynchronously create a new symbolic link to the existing object at C<$srcpath> at
		775	the path C<$dstpath> and call the callback with the result code.
		776
		777
		778	=item aio_readlink $pathname, $callback->($link)
		779
		780	Asynchronously read the symlink specified by C<$path> and pass it to
		781	the callback. If an error occurs, nothing or undef gets passed to the
		782	callback.
		783
		784
		785	=item aio_realpath $pathname, $callback->($path)
		786
		787	Asynchronously make the path absolute and resolve any symlinks in
		788	C<$path>. The resulting path only consists of directories (Same as
		789	L<Cwd::realpath>).
		790
		791	This request can be used to get the absolute path of the current working
		792	directory by passing it a path of F<.> (a single dot).
		793
		794
		795	=item aio_rename $srcpath, $dstpath, $callback->($status)
		796
		797	Asynchronously rename the object at C<$srcpath> to C<$dstpath>, just as
		798	rename(2) and call the callback with the result code.
		799
		800
		801	=item aio_mkdir $pathname, $mode, $callback->($status)
		802
		803	Asynchronously mkdir (create) a directory and call the callback with
		804	the result code. C<$mode> will be modified by the umask at the time the
		805	request is executed, so do not change your umask.
		806
		807
		808	=item aio_rmdir $pathname, $callback->($status)
		809
		810	Asynchronously rmdir (delete) a directory and call the callback with the
		811	result code.
		812
		813
		814	=item aio_readdir $pathname, $callback->($entries)
		815
		816	Unlike the POSIX call of the same name, C<aio_readdir> reads an entire
		817	directory (i.e. opendir + readdir + closedir). The entries will not be
		818	sorted, and will B<NOT> include the C<.> and C<..> entries.
		819
		820	The callback is passed a single argument which is either C<undef> or an
		821	array-ref with the filenames.
		822
		823
		824	=item aio_readdirx $pathname, $flags, $callback->($entries, $flags)
		825
		826	Quite similar to C<aio_readdir>, but the C<$flags> argument allows one to
		827	tune behaviour and output format. In case of an error, C<$entries> will be
		828	C<undef>.
		829
		830	The flags are a combination of the following constants, ORed together (the
		831	flags will also be passed to the callback, possibly modified):
		832
		833	=over 4
		834
		835	=item IO::AIO::READDIR_DENTS
		836
		837	When this flag is off, then the callback gets an arrayref consisting of
		838	names only (as with C<aio_readdir>), otherwise it gets an arrayref with
		839	C<[$name, $type, $inode]> arrayrefs, each describing a single directory
		840	entry in more detail.
		841
		842	C<$name> is the name of the entry.
		843
		844	C<$type> is one of the C<IO::AIO::DT_xxx> constants:
		845
		846	C<IO::AIO::DT_UNKNOWN>, C<IO::AIO::DT_FIFO>, C<IO::AIO::DT_CHR>, C<IO::AIO::DT_DIR>,
		847	C<IO::AIO::DT_BLK>, C<IO::AIO::DT_REG>, C<IO::AIO::DT_LNK>, C<IO::AIO::DT_SOCK>,
		848	C<IO::AIO::DT_WHT>.
		849
		850	C<IO::AIO::DT_UNKNOWN> means just that: readdir does not know. If you need to
		851	know, you have to run stat yourself. Also, for speed reasons, the C<$type>
		852	scalars are read-only: you can not modify them.
		853
		854	C<$inode> is the inode number (which might not be exact on systems with 64
		855	bit inode numbers and 32 bit perls). This field has unspecified content on
		856	systems that do not deliver the inode information.
		857
		858	=item IO::AIO::READDIR_DIRS_FIRST
		859
		860	When this flag is set, then the names will be returned in an order where
		861	likely directories come first, in optimal stat order. This is useful when
		862	you need to quickly find directories, or you want to find all directories
		863	while avoiding to stat() each entry.
		864
		865	If the system returns type information in readdir, then this is used
		866	to find directories directly. Otherwise, likely directories are names
		867	beginning with ".", or otherwise names with no dots, of which names with
		868	short names are tried first.
		869
		870	=item IO::AIO::READDIR_STAT_ORDER
		871
		872	When this flag is set, then the names will be returned in an order
		873	suitable for stat()'ing each one. That is, when you plan to stat()
		874	all files in the given directory, then the returned order will likely
		875	be fastest.
		876
		877	If both this flag and C<IO::AIO::READDIR_DIRS_FIRST> are specified, then
		878	the likely dirs come first, resulting in a less optimal stat order.
		879
		880	=item IO::AIO::READDIR_FOUND_UNKNOWN
		881
		882	This flag should not be set when calling C<aio_readdirx>. Instead, it
		883	is being set by C<aio_readdirx>, when any of the C<$type>'s found were
		884	C<IO::AIO::DT_UNKNOWN>. The absence of this flag therefore indicates that all
		885	C<$type>'s are known, which can be used to speed up some algorithms.
		886
		887	=back
		888
		889
		890	=item aio_load $pathname, $data, $callback->($status)
		891
		892	This is a composite request that tries to fully load the given file into
		893	memory. Status is the same as with aio_read.
		894
		895	=cut
		896
		897	sub aio_load($$;$) {
		898	my ($path, undef, $cb) = @_;
		899	my $data = \$_[1];
		900
		901	my $pri = aioreq_pri;
		902	my $grp = aio_group $cb;
		903
		904	aioreq_pri $pri;
		905	add $grp aio_open $path, O_RDONLY, 0, sub {
		906	my $fh = shift
		907	or return $grp->result (-1);
		908
		909	aioreq_pri $pri;
		910	add $grp aio_read $fh, 0, (-s $fh), $$data, 0, sub {
		911	$grp->result ($_[0]);
		912	};
		913	};
		914
		915	$grp
		916	}
		917
		918	=item aio_copy $srcpath, $dstpath, $callback->($status)
		919
		920	Try to copy the I<file> (directories not supported as either source or
		921	destination) from C<$srcpath> to C<$dstpath> and call the callback with
		922	a status of C<0> (ok) or C<-1> (error, see C<$!>).
		923
		924	This is a composite request that creates the destination file with
		925	mode 0200 and copies the contents of the source file into it using
		926	C<aio_sendfile>, followed by restoring atime, mtime, access mode and
		927	uid/gid, in that order.
		928
		929	If an error occurs, the partial destination file will be unlinked, if
		930	possible, except when setting atime, mtime, access mode and uid/gid, where
		931	errors are being ignored.
		932
		933	=cut
		934
		935	sub aio_copy($$;$) {
		936	my ($src, $dst, $cb) = @_;
		937
		938	my $pri = aioreq_pri;
		939	my $grp = aio_group $cb;
		940
		941	aioreq_pri $pri;
		942	add $grp aio_open $src, O_RDONLY, 0, sub {
		943	if (my $src_fh = $_[0]) {
		944	my @stat = stat $src_fh; # hmm, might block over nfs?
		945
		946	aioreq_pri $pri;
		947	add $grp aio_open $dst, O_CREAT | O_WRONLY | O_TRUNC, 0200, sub {
		948	if (my $dst_fh = $_[0]) {
		949	aioreq_pri $pri;
		950	add $grp aio_sendfile $dst_fh, $src_fh, 0, $stat[7], sub {
		951	if ($_[0] == $stat[7]) {
		952	$grp->result (0);
		953	close $src_fh;
		954
		955	my $ch = sub {
		956	aioreq_pri $pri;
		957	add $grp aio_chmod $dst_fh, $stat[2] & 07777, sub {
		958	aioreq_pri $pri;
		959	add $grp aio_chown $dst_fh, $stat[4], $stat[5], sub {
		960	aioreq_pri $pri;
		961	add $grp aio_close $dst_fh;
		962	}
		963	};
		964	};
		965
		966	aioreq_pri $pri;
		967	add $grp aio_utime $dst_fh, $stat[8], $stat[9], sub {
		968	if ($_[0] < 0 && $! == ENOSYS) {
		969	aioreq_pri $pri;
		970	add $grp aio_utime $dst, $stat[8], $stat[9], $ch;
		971	} else {
		972	$ch->();
		973	}
		974	};
		975	} else {
		976	$grp->result (-1);
		977	close $src_fh;
		978	close $dst_fh;
		979
		980	aioreq $pri;
		981	add $grp aio_unlink $dst;
		982	}
		983	};
		984	} else {
		985	$grp->result (-1);
		986	}
		987	},
		988
		989	} else {
		990	$grp->result (-1);
		991	}
		992	};
		993
		994	$grp
		995	}
		996
		997	=item aio_move $srcpath, $dstpath, $callback->($status)
		998
		999	Try to move the I<file> (directories not supported as either source or
		1000	destination) from C<$srcpath> to C<$dstpath> and call the callback with
		1001	a status of C<0> (ok) or C<-1> (error, see C<$!>).
		1002
		1003	This is a composite request that tries to rename(2) the file first; if
		1004	rename fails with C<EXDEV>, it copies the file with C<aio_copy> and, if
		1005	that is successful, unlinks the C<$srcpath>.
		1006
		1007	=cut
		1008
		1009	sub aio_move($$;$) {
		1010	my ($src, $dst, $cb) = @_;
		1011
		1012	my $pri = aioreq_pri;
		1013	my $grp = aio_group $cb;
		1014
		1015	aioreq_pri $pri;
		1016	add $grp aio_rename $src, $dst, sub {
		1017	if ($_[0] && $! == EXDEV) {
		1018	aioreq_pri $pri;
		1019	add $grp aio_copy $src, $dst, sub {
		1020	$grp->result ($_[0]);
		1021
		1022	unless ($_[0]) {
		1023	aioreq_pri $pri;
		1024	add $grp aio_unlink $src;
		1025	}
		1026	};
		1027	} else {
		1028	$grp->result ($_[0]);
		1029	}
		1030	};
		1031
		1032	$grp
		1033	}
		1034
		1035	=item aio_scandir $pathname, $maxreq, $callback->($dirs, $nondirs)
		1036
		1037	Scans a directory (similar to C<aio_readdir>) but additionally tries to
		1038	efficiently separate the entries of directory C<$path> into two sets of
		1039	names, directories you can recurse into (directories), and ones you cannot
		1040	recurse into (everything else, including symlinks to directories).
		1041
		1042	C<aio_scandir> is a composite request that creates of many sub requests_
		1043	C<$maxreq> specifies the maximum number of outstanding aio requests that
		1044	this function generates. If it is C<< <= 0 >>, then a suitable default
		1045	will be chosen (currently 4).
		1046
		1047	On error, the callback is called without arguments, otherwise it receives
		1048	two array-refs with path-relative entry names.
		1049
		1050	Example:
		1051
		1052	aio_scandir $dir, 0, sub {
		1053	my ($dirs, $nondirs) = @_;
		1054	print "real directories: @$dirs\n";
		1055	print "everything else: @$nondirs\n";
		1056	};
		1057
		1058	Implementation notes.
		1059
		1060	The C<aio_readdir> cannot be avoided, but C<stat()>'ing every entry can.
		1061
		1062	If readdir returns file type information, then this is used directly to
		1063	find directories.
		1064
		1065	Otherwise, after reading the directory, the modification time, size etc.
		1066	of the directory before and after the readdir is checked, and if they
		1067	match (and isn't the current time), the link count will be used to decide
		1068	how many entries are directories (if >= 2). Otherwise, no knowledge of the
		1069	number of subdirectories will be assumed.
		1070
		1071	Then entries will be sorted into likely directories a non-initial dot
		1072	currently) and likely non-directories (see C<aio_readdirx>). Then every
		1073	entry plus an appended C</.> will be C<stat>'ed, likely directories first,
		1074	in order of their inode numbers. If that succeeds, it assumes that the
		1075	entry is a directory or a symlink to directory (which will be checked
		1076	separately). This is often faster than stat'ing the entry itself because
		1077	filesystems might detect the type of the entry without reading the inode
		1078	data (e.g. ext2fs filetype feature), even on systems that cannot return
		1079	the filetype information on readdir.
		1080
		1081	If the known number of directories (link count - 2) has been reached, the
		1082	rest of the entries is assumed to be non-directories.
		1083
		1084	This only works with certainty on POSIX (= UNIX) filesystems, which
		1085	fortunately are the vast majority of filesystems around.
		1086
		1087	It will also likely work on non-POSIX filesystems with reduced efficiency
		1088	as those tend to return 0 or 1 as link counts, which disables the
		1089	directory counting heuristic.
		1090
		1091	=cut
		1092
		1093	sub aio_scandir($$;$) {
		1094	my ($path, $maxreq, $cb) = @_;
		1095
		1096	my $pri = aioreq_pri;
		1097
		1098	my $grp = aio_group $cb;
		1099
		1100	$maxreq = 4 if $maxreq <= 0;
		1101
		1102	# get a wd object
		1103	aioreq_pri $pri;
		1104	add $grp aio_wd $path, sub {
		1105	$_[0]
		1106	or return $grp->result ();
		1107
		1108	my $wd = [shift, "."];
		1109
		1110	# stat once
		1111	aioreq_pri $pri;
		1112	add $grp aio_stat $wd, sub {
		1113	return $grp->result () if $_[0];
		1114	my $now = time;
		1115	my $hash1 = join ":", (stat _)[0,1,3,7,9];
		1116
		1117	# read the directory entries
		1118	aioreq_pri $pri;
		1119	add $grp aio_readdirx $wd, READDIR_DIRS_FIRST, sub {
		1120	my $entries = shift
		1121	or return $grp->result ();
		1122
		1123	# stat the dir another time
		1124	aioreq_pri $pri;
		1125	add $grp aio_stat $wd, sub {
		1126	my $hash2 = join ":", (stat _)[0,1,3,7,9];
		1127
		1128	my $ndirs;
		1129
		1130	# take the slow route if anything looks fishy
		1131	if ($hash1 ne $hash2 or (stat _)[9] == $now) {
		1132	$ndirs = -1;
		1133	} else {
		1134	# if nlink == 2, we are finished
		1135	# for non-posix-fs's, we rely on nlink < 2
		1136	$ndirs = (stat _)[3] - 2
		1137	or return $grp->result ([], $entries);
		1138	}
		1139
		1140	my (@dirs, @nondirs);
		1141
		1142	my $statgrp = add $grp aio_group sub {
		1143	$grp->result (\@dirs, \@nondirs);
		1144	};
		1145
		1146	limit $statgrp $maxreq;
		1147	feed $statgrp sub {
		1148	return unless @$entries;
		1149	my $entry = shift @$entries;
		1150
		1151	aioreq_pri $pri;
		1152	$wd->[1] = "$entry/.";
		1153	add $statgrp aio_stat $wd, sub {
		1154	if ($_[0] < 0) {
		1155	push @nondirs, $entry;
		1156	} else {
		1157	# need to check for real directory
		1158	aioreq_pri $pri;
		1159	$wd->[1] = $entry;
		1160	add $statgrp aio_lstat $wd, sub {
		1161	if (-d _) {
		1162	push @dirs, $entry;
		1163
		1164	unless (--$ndirs) {
		1165	push @nondirs, @$entries;
		1166	feed $statgrp;
		1167	}
		1168	} else {
		1169	push @nondirs, $entry;
		1170	}
		1171	}
		1172	}
		1173	};
		1174	};
		1175	};
		1176	};
		1177	};
		1178	};
		1179
		1180	$grp
		1181	}
		1182
		1183	=item aio_rmtree $pathname, $callback->($status)
		1184
		1185	Delete a directory tree starting (and including) C<$path>, return the
		1186	status of the final C<rmdir> only. This is a composite request that
		1187	uses C<aio_scandir> to recurse into and rmdir directories, and unlink
		1188	everything else.
		1189
		1190	=cut
		1191
		1192	sub aio_rmtree;
		1193	sub aio_rmtree($;$) {
		1194	my ($path, $cb) = @_;
		1195
		1196	my $pri = aioreq_pri;
		1197	my $grp = aio_group $cb;
		1198
		1199	aioreq_pri $pri;
		1200	add $grp aio_scandir $path, 0, sub {
		1201	my ($dirs, $nondirs) = @_;
		1202
		1203	my $dirgrp = aio_group sub {
		1204	add $grp aio_rmdir $path, sub {
		1205	$grp->result ($_[0]);
		1206	};
		1207	};
		1208
		1209	(aioreq_pri $pri), add $dirgrp aio_rmtree "$path/$_" for @$dirs;
		1210	(aioreq_pri $pri), add $dirgrp aio_unlink "$path/$_" for @$nondirs;
		1211
		1212	add $grp $dirgrp;
		1213	};
		1214
		1215	$grp
		1216	}
		1217
		1218	=item aio_sync $callback->($status)
		1219
		1220	Asynchronously call sync and call the callback when finished.
		1221
179	=item aio_fsync $fh, $callback	1222	=item aio_fsync $fh, $callback->($status)
180		1223
181	Asynchronously call fsync on the given filehandle and call the callback	1224	Asynchronously call fsync on the given filehandle and call the callback
182	with the fsync result code.	1225	with the fsync result code.
183		1226
184	=item aio_fdatasync $fh, $callback	1227	=item aio_fdatasync $fh, $callback->($status)
185		1228
186	Asynchronously call fdatasync on the given filehandle and call the	1229	Asynchronously call fdatasync on the given filehandle and call the
187	callback with the fdatasync result code.	1230	callback with the fdatasync result code.
188		1231
		1232	If this call isn't available because your OS lacks it or it couldn't be
		1233	detected, it will be emulated by calling C<fsync> instead.
		1234
		1235	=item aio_syncfs $fh, $callback->($status)
		1236
		1237	Asynchronously call the syncfs syscall to sync the filesystem associated
		1238	to the given filehandle and call the callback with the syncfs result
		1239	code. If syncfs is not available, calls sync(), but returns C<-1> and sets
		1240	errno to C<ENOSYS> nevertheless.
		1241
		1242	=item aio_sync_file_range $fh, $offset, $nbytes, $flags, $callback->($status)
		1243
		1244	Sync the data portion of the file specified by C<$offset> and C<$length>
		1245	to disk (but NOT the metadata), by calling the Linux-specific
		1246	sync_file_range call. If sync_file_range is not available or it returns
		1247	ENOSYS, then fdatasync or fsync is being substituted.
		1248
		1249	C<$flags> can be a combination of C<IO::AIO::SYNC_FILE_RANGE_WAIT_BEFORE>,
		1250	C<IO::AIO::SYNC_FILE_RANGE_WRITE> and
		1251	C<IO::AIO::SYNC_FILE_RANGE_WAIT_AFTER>: refer to the sync_file_range
		1252	manpage for details.
		1253
		1254	=item aio_pathsync $pathname, $callback->($status)
		1255
		1256	This request tries to open, fsync and close the given path. This is a
		1257	composite request intended to sync directories after directory operations
		1258	(E.g. rename). This might not work on all operating systems or have any
		1259	specific effect, but usually it makes sure that directory changes get
		1260	written to disc. It works for anything that can be opened for read-only,
		1261	not just directories.
		1262
		1263	Future versions of this function might fall back to other methods when
		1264	C<fsync> on the directory fails (such as calling C<sync>).
		1265
		1266	Passes C<0> when everything went ok, and C<-1> on error.
		1267
		1268	=cut
		1269
		1270	sub aio_pathsync($;$) {
		1271	my ($path, $cb) = @_;
		1272
		1273	my $pri = aioreq_pri;
		1274	my $grp = aio_group $cb;
		1275
		1276	aioreq_pri $pri;
		1277	add $grp aio_open $path, O_RDONLY, 0, sub {
		1278	my ($fh) = @_;
		1279	if ($fh) {
		1280	aioreq_pri $pri;
		1281	add $grp aio_fsync $fh, sub {
		1282	$grp->result ($_[0]);
		1283
		1284	aioreq_pri $pri;
		1285	add $grp aio_close $fh;
		1286	};
		1287	} else {
		1288	$grp->result (-1);
		1289	}
		1290	};
		1291
		1292	$grp
		1293	}
		1294
		1295	=item aio_msync $scalar, $offset = 0, $length = undef, flags = 0, $callback->($status)
		1296
		1297	This is a rather advanced IO::AIO call, which only works on mmap(2)ed
		1298	scalars (see the C<IO::AIO::mmap> function, although it also works on data
		1299	scalars managed by the L<Sys::Mmap> or L<Mmap> modules, note that the
		1300	scalar must only be modified in-place while an aio operation is pending on
		1301	it).
		1302
		1303	It calls the C<msync> function of your OS, if available, with the memory
		1304	area starting at C<$offset> in the string and ending C<$length> bytes
		1305	later. If C<$length> is negative, counts from the end, and if C<$length>
		1306	is C<undef>, then it goes till the end of the string. The flags can be
		1307	a combination of C<IO::AIO::MS_ASYNC>, C<IO::AIO::MS_INVALIDATE> and
		1308	C<IO::AIO::MS_SYNC>.
		1309
		1310	=item aio_mtouch $scalar, $offset = 0, $length = undef, flags = 0, $callback->($status)
		1311
		1312	This is a rather advanced IO::AIO call, which works best on mmap(2)ed
		1313	scalars.
		1314
		1315	It touches (reads or writes) all memory pages in the specified
		1316	range inside the scalar. All caveats and parameters are the same
		1317	as for C<aio_msync>, above, except for flags, which must be either
		1318	C<0> (which reads all pages and ensures they are instantiated) or
		1319	C<IO::AIO::MT_MODIFY>, which modifies the memory page s(by reading and
		1320	writing an octet from it, which dirties the page).
		1321
		1322	=item aio_mlock $scalar, $offset = 0, $length = undef, $callback->($status)
		1323
		1324	This is a rather advanced IO::AIO call, which works best on mmap(2)ed
		1325	scalars.
		1326
		1327	It reads in all the pages of the underlying storage into memory (if any)
		1328	and locks them, so they are not getting swapped/paged out or removed.
		1329
		1330	If C<$length> is undefined, then the scalar will be locked till the end.
		1331
		1332	On systems that do not implement C<mlock>, this function returns C<-1>
		1333	and sets errno to C<ENOSYS>.
		1334
		1335	Note that the corresponding C<munlock> is synchronous and is
		1336	documented under L<MISCELLANEOUS FUNCTIONS>.
		1337
		1338	Example: open a file, mmap and mlock it - both will be undone when
		1339	C<$data> gets destroyed.
		1340
		1341	open my $fh, "<", $path or die "$path: $!";
		1342	my $data;
		1343	IO::AIO::mmap $data, -s $fh, IO::AIO::PROT_READ, IO::AIO::MAP_SHARED, $fh;
		1344	aio_mlock $data; # mlock in background
		1345
		1346	=item aio_mlockall $flags, $callback->($status)
		1347
		1348	Calls the C<mlockall> function with the given C<$flags> (a combination of
		1349	C<IO::AIO::MCL_CURRENT> and C<IO::AIO::MCL_FUTURE>).
		1350
		1351	On systems that do not implement C<mlockall>, this function returns C<-1>
		1352	and sets errno to C<ENOSYS>.
		1353
		1354	Note that the corresponding C<munlockall> is synchronous and is
		1355	documented under L<MISCELLANEOUS FUNCTIONS>.
		1356
		1357	Example: asynchronously lock all current and future pages into memory.
		1358
		1359	aio_mlockall IO::AIO::MCL_FUTURE;
		1360
		1361	=item aio_fiemap $fh, $start, $length, $flags, $count, $cb->(\@extents)
		1362
		1363	Queries the extents of the given file (by calling the Linux C<FIEMAP>
		1364	ioctl, see L<http://cvs.schmorp.de/IO-AIO/doc/fiemap.txt> for details). If
		1365	the ioctl is not available on your OS, then this request will fail with
		1366	C<ENOSYS>.
		1367
		1368	C<$start> is the starting offset to query extents for, C<$length> is the
		1369	size of the range to query - if it is C<undef>, then the whole file will
		1370	be queried.
		1371
		1372	C<$flags> is a combination of flags (C<IO::AIO::FIEMAP_FLAG_SYNC> or
		1373	C<IO::AIO::FIEMAP_FLAG_XATTR> - C<IO::AIO::FIEMAP_FLAGS_COMPAT> is also
		1374	exported), and is normally C<0> or C<IO::AIO::FIEMAP_FLAG_SYNC> to query
		1375	the data portion.
		1376
		1377	C<$count> is the maximum number of extent records to return. If it is
		1378	C<undef>, then IO::AIO queries all extents of the range. As a very special
		1379	case, if it is C<0>, then the callback receives the number of extents
		1380	instead of the extents themselves (which is unreliable, see below).
		1381
		1382	If an error occurs, the callback receives no arguments. The special
		1383	C<errno> value C<IO::AIO::EBADR> is available to test for flag errors.
		1384
		1385	Otherwise, the callback receives an array reference with extent
		1386	structures. Each extent structure is an array reference itself, with the
		1387	following members:
		1388
		1389	[$logical, $physical, $length, $flags]
		1390
		1391	Flags is any combination of the following flag values (typically either C<0>
		1392	or C<IO::AIO::FIEMAP_EXTENT_LAST> (1)):
		1393
		1394	C<IO::AIO::FIEMAP_EXTENT_LAST>, C<IO::AIO::FIEMAP_EXTENT_UNKNOWN>,
		1395	C<IO::AIO::FIEMAP_EXTENT_DELALLOC>, C<IO::AIO::FIEMAP_EXTENT_ENCODED>,
		1396	C<IO::AIO::FIEMAP_EXTENT_DATA_ENCRYPTED>, C<IO::AIO::FIEMAP_EXTENT_NOT_ALIGNED>,
		1397	C<IO::AIO::FIEMAP_EXTENT_DATA_INLINE>, C<IO::AIO::FIEMAP_EXTENT_DATA_TAIL>,
		1398	C<IO::AIO::FIEMAP_EXTENT_UNWRITTEN>, C<IO::AIO::FIEMAP_EXTENT_MERGED> or
		1399	C<IO::AIO::FIEMAP_EXTENT_SHARED>.
		1400
		1401	At the time of this writing (Linux 3.2), this requets is unreliable unless
		1402	C<$count> is C<undef>, as the kernel has all sorts of bugs preventing
		1403	it to return all extents of a range for files with large number of
		1404	extents. The code works around all these issues if C<$count> is undef.
		1405
		1406	=item aio_group $callback->(...)
		1407
		1408	This is a very special aio request: Instead of doing something, it is a
		1409	container for other aio requests, which is useful if you want to bundle
		1410	many requests into a single, composite, request with a definite callback
		1411	and the ability to cancel the whole request with its subrequests.
		1412
		1413	Returns an object of class L<IO::AIO::GRP>. See its documentation below
		1414	for more info.
		1415
		1416	Example:
		1417
		1418	my $grp = aio_group sub {
		1419	print "all stats done\n";
		1420	};
		1421
		1422	add $grp
		1423	(aio_stat ...),
		1424	(aio_stat ...),
		1425	...;
		1426
		1427	=item aio_nop $callback->()
		1428
		1429	This is a special request - it does nothing in itself and is only used for
		1430	side effects, such as when you want to add a dummy request to a group so
		1431	that finishing the requests in the group depends on executing the given
		1432	code.
		1433
		1434	While this request does nothing, it still goes through the execution
		1435	phase and still requires a worker thread. Thus, the callback will not
		1436	be executed immediately but only after other requests in the queue have
		1437	entered their execution phase. This can be used to measure request
		1438	latency.
		1439
		1440	=item IO::AIO::aio_busy $fractional_seconds, $callback->() *NOT EXPORTED*
		1441
		1442	Mainly used for debugging and benchmarking, this aio request puts one of
		1443	the request workers to sleep for the given time.
		1444
		1445	While it is theoretically handy to have simple I/O scheduling requests
		1446	like sleep and file handle readable/writable, the overhead this creates is
		1447	immense (it blocks a thread for a long time) so do not use this function
		1448	except to put your application under artificial I/O pressure.
		1449
189	=back	1450	=back
190		1451
		1452
		1453	=head2 IO::AIO::WD - multiple working directories
		1454
		1455	Your process only has one current working directory, which is used by all
		1456	threads. This makes it hard to use relative paths (some other component
		1457	could call C<chdir> at any time, and it is hard to control when the path
		1458	will be used by IO::AIO).
		1459
		1460	One solution for this is to always use absolute paths. This usually works,
		1461	but can be quite slow (the kernel has to walk the whole path on every
		1462	access), and can also be a hassle to implement.
		1463
		1464	Newer POSIX systems have a number of functions (openat, fdopendir,
		1465	futimensat and so on) that make it possible to specify working directories
		1466	per operation.
		1467
		1468	For portability, and because the clowns who "designed", or shall I write,
		1469	perpetrated this new interface were obviously half-drunk, this abstraction
		1470	cannot be perfect, though.
		1471
		1472	IO::AIO allows you to convert directory paths into a so-called IO::AIO::WD
		1473	object. This object stores the canonicalised, absolute version of the
		1474	path, and on systems that allow it, also a directory file descriptor.
		1475
		1476	Everywhere where a pathname is accepted by IO::AIO (e.g. in C<aio_stat>
		1477	or C<aio_unlink>), one can specify an array reference with an IO::AIO::WD
		1478	object and a pathname instead (or the IO::AIO::WD object alone, which
		1479	gets interpreted as C<[$wd, "."]>). If the pathname is absolute, the
		1480	IO::AIO::WD object is ignored, otherwise the pathname is resolved relative
		1481	to that IO::AIO::WD object.
		1482
		1483	For example, to get a wd object for F</etc> and then stat F<passwd>
		1484	inside, you would write:
		1485
		1486	aio_wd "/etc", sub {
		1487	my $etcdir = shift;
		1488
		1489	# although $etcdir can be undef on error, there is generally no reason
		1490	# to check for errors here, as aio_stat will fail with ENOENT
		1491	# when $etcdir is undef.
		1492
		1493	aio_stat [$etcdir, "passwd"], sub {
		1494	# yay
		1495	};
		1496	};
		1497
		1498	That C<aio_wd> is a request and not a normal function shows that creating
		1499	an IO::AIO::WD object is itself a potentially blocking operation, which is
		1500	why it is done asynchronously.
		1501
		1502	To stat the directory obtained with C<aio_wd> above, one could write
		1503	either of the following three request calls:
		1504
		1505	aio_lstat "/etc" , sub { ... # pathname as normal string
		1506	aio_lstat [$wd, "."], sub { ... # "." relative to $wd (i.e. $wd itself)
		1507	aio_lstat $wd , sub { ... # shorthand for the previous
		1508
		1509	As with normal pathnames, IO::AIO keeps a copy of the working directory
		1510	object and the pathname string, so you could write the following without
		1511	causing any issues due to C<$path> getting reused:
		1512
		1513	my $path = [$wd, undef];
		1514
		1515	for my $name (qw(abc def ghi)) {
		1516	$path->[1] = $name;
		1517	aio_stat $path, sub {
		1518	# ...
		1519	};
		1520	}
		1521
		1522	There are some caveats: when directories get renamed (or deleted), the
		1523	pathname string doesn't change, so will point to the new directory (or
		1524	nowhere at all), while the directory fd, if available on the system,
		1525	will still point to the original directory. Most functions accepting a
		1526	pathname will use the directory fd on newer systems, and the string on
		1527	older systems. Some functions (such as realpath) will always rely on the
		1528	string form of the pathname.
		1529
		1530	So this fucntionality is mainly useful to get some protection against
		1531	C<chdir>, to easily get an absolute path out of a relative path for future
		1532	reference, and to speed up doing many operations in the same directory
		1533	(e.g. when stat'ing all files in a directory).
		1534
		1535	The following functions implement this working directory abstraction:
		1536
		1537	=over 4
		1538
		1539	=item aio_wd $pathname, $callback->($wd)
		1540
		1541	Asynchonously canonicalise the given pathname and convert it to an
		1542	IO::AIO::WD object representing it. If possible and supported on the
		1543	system, also open a directory fd to speed up pathname resolution relative
		1544	to this working directory.
		1545
		1546	If something goes wrong, then C<undef> is passwd to the callback instead
		1547	of a working directory object and C<$!> is set appropriately. Since
		1548	passing C<undef> as working directory component of a pathname fails the
		1549	request with C<ENOENT>, there is often no need for error checking in the
		1550	C<aio_wd> callback, as future requests using the value will fail in the
		1551	expected way.
		1552
		1553	If this call isn't available because your OS lacks it or it couldn't be
		1554	detected, it will be emulated by calling C<fsync> instead.
		1555
		1556	=item IO::AIO::CWD
		1557
		1558	This is a compiletime constant (object) that represents the process
		1559	current working directory.
		1560
		1561	Specifying this object as working directory object for a pathname is as
		1562	if the pathname would be specified directly, without a directory object,
		1563	e.g., these calls are functionally identical:
		1564
		1565	aio_stat "somefile", sub { ... };
		1566	aio_stat [IO::AIO::CWD, "somefile"], sub { ... };
		1567
		1568	=back
		1569
		1570
		1571	=head2 IO::AIO::REQ CLASS
		1572
		1573	All non-aggregate C<aio_*> functions return an object of this class when
		1574	called in non-void context.
		1575
		1576	=over 4
		1577
		1578	=item cancel $req
		1579
		1580	Cancels the request, if possible. Has the effect of skipping execution
		1581	when entering the B<execute> state and skipping calling the callback when
		1582	entering the the B<result> state, but will leave the request otherwise
		1583	untouched (with the exception of readdir). That means that requests that
		1584	currently execute will not be stopped and resources held by the request
		1585	will not be freed prematurely.
		1586
		1587	=item cb $req $callback->(...)
		1588
		1589	Replace (or simply set) the callback registered to the request.
		1590
		1591	=back
		1592
		1593	=head2 IO::AIO::GRP CLASS
		1594
		1595	This class is a subclass of L<IO::AIO::REQ>, so all its methods apply to
		1596	objects of this class, too.
		1597
		1598	A IO::AIO::GRP object is a special request that can contain multiple other
		1599	aio requests.
		1600
		1601	You create one by calling the C<aio_group> constructing function with a
		1602	callback that will be called when all contained requests have entered the
		1603	C<done> state:
		1604
		1605	my $grp = aio_group sub {
		1606	print "all requests are done\n";
		1607	};
		1608
		1609	You add requests by calling the C<add> method with one or more
		1610	C<IO::AIO::REQ> objects:
		1611
		1612	$grp->add (aio_unlink "...");
		1613
		1614	add $grp aio_stat "...", sub {
		1615	$_[0] or return $grp->result ("error");
		1616
		1617	# add another request dynamically, if first succeeded
		1618	add $grp aio_open "...", sub {
		1619	$grp->result ("ok");
		1620	};
		1621	};
		1622
		1623	This makes it very easy to create composite requests (see the source of
		1624	C<aio_move> for an application) that work and feel like simple requests.
		1625
		1626	=over 4
		1627
		1628	=item * The IO::AIO::GRP objects will be cleaned up during calls to
		1629	C<IO::AIO::poll_cb>, just like any other request.
		1630
		1631	=item * They can be canceled like any other request. Canceling will cancel not
		1632	only the request itself, but also all requests it contains.
		1633
		1634	=item * They can also can also be added to other IO::AIO::GRP objects.
		1635
		1636	=item * You must not add requests to a group from within the group callback (or
		1637	any later time).
		1638
		1639	=back
		1640
		1641	Their lifetime, simplified, looks like this: when they are empty, they
		1642	will finish very quickly. If they contain only requests that are in the
		1643	C<done> state, they will also finish. Otherwise they will continue to
		1644	exist.
		1645
		1646	That means after creating a group you have some time to add requests
		1647	(precisely before the callback has been invoked, which is only done within
		1648	the C<poll_cb>). And in the callbacks of those requests, you can add
		1649	further requests to the group. And only when all those requests have
		1650	finished will the the group itself finish.
		1651
		1652	=over 4
		1653
		1654	=item add $grp ...
		1655
		1656	=item $grp->add (...)
		1657
		1658	Add one or more requests to the group. Any type of L<IO::AIO::REQ> can
		1659	be added, including other groups, as long as you do not create circular
		1660	dependencies.
		1661
		1662	Returns all its arguments.
		1663
		1664	=item $grp->cancel_subs
		1665
		1666	Cancel all subrequests and clears any feeder, but not the group request
		1667	itself. Useful when you queued a lot of events but got a result early.
		1668
		1669	The group request will finish normally (you cannot add requests to the
		1670	group).
		1671
		1672	=item $grp->result (...)
		1673
		1674	Set the result value(s) that will be passed to the group callback when all
		1675	subrequests have finished and set the groups errno to the current value
		1676	of errno (just like calling C<errno> without an error number). By default,
		1677	no argument will be passed and errno is zero.
		1678
		1679	=item $grp->errno ([$errno])
		1680
		1681	Sets the group errno value to C<$errno>, or the current value of errno
		1682	when the argument is missing.
		1683
		1684	Every aio request has an associated errno value that is restored when
		1685	the callback is invoked. This method lets you change this value from its
		1686	default (0).
		1687
		1688	Calling C<result> will also set errno, so make sure you either set C<$!>
		1689	before the call to C<result>, or call c<errno> after it.
		1690
		1691	=item feed $grp $callback->($grp)
		1692
		1693	Sets a feeder/generator on this group: every group can have an attached
		1694	generator that generates requests if idle. The idea behind this is that,
		1695	although you could just queue as many requests as you want in a group,
		1696	this might starve other requests for a potentially long time. For example,
		1697	C<aio_scandir> might generate hundreds of thousands of C<aio_stat>
		1698	requests, delaying any later requests for a long time.
		1699
		1700	To avoid this, and allow incremental generation of requests, you can
		1701	instead a group and set a feeder on it that generates those requests. The
		1702	feed callback will be called whenever there are few enough (see C<limit>,
		1703	below) requests active in the group itself and is expected to queue more
		1704	requests.
		1705
		1706	The feed callback can queue as many requests as it likes (i.e. C<add> does
		1707	not impose any limits).
		1708
		1709	If the feed does not queue more requests when called, it will be
		1710	automatically removed from the group.
		1711
		1712	If the feed limit is C<0> when this method is called, it will be set to
		1713	C<2> automatically.
		1714
		1715	Example:
		1716
		1717	# stat all files in @files, but only ever use four aio requests concurrently:
		1718
		1719	my $grp = aio_group sub { print "finished\n" };
		1720	limit $grp 4;
		1721	feed $grp sub {
		1722	my $file = pop @files
		1723	or return;
		1724
		1725	add $grp aio_stat $file, sub { ... };
		1726	};
		1727
		1728	=item limit $grp $num
		1729
		1730	Sets the feeder limit for the group: The feeder will be called whenever
		1731	the group contains less than this many requests.
		1732
		1733	Setting the limit to C<0> will pause the feeding process.
		1734
		1735	The default value for the limit is C<0>, but note that setting a feeder
		1736	automatically bumps it up to C<2>.
		1737
		1738	=back
		1739
191	=head2 SUPPORT FUNCTIONS	1740	=head2 SUPPORT FUNCTIONS
192		1741
		1742	=head3 EVENT PROCESSING AND EVENT LOOP INTEGRATION
		1743
193	=over 4	1744	=over 4
194		1745
195	=item $fileno = IO::AIO::poll_fileno	1746	=item $fileno = IO::AIO::poll_fileno
196		1747
197	Return the I<request result pipe filehandle>. This filehandle must be	1748	Return the I<request result pipe file descriptor>. This filehandle must be
198	polled for reading by some mechanism outside this module (e.g. Event	1749	polled for reading by some mechanism outside this module (e.g. EV, Glib,
199	or select, see below). If the pipe becomes readable you have to call	1750	select and so on, see below or the SYNOPSIS). If the pipe becomes readable
200	C<poll_cb> to check the results.	1751	you have to call C<poll_cb> to check the results.
201		1752
202	See C<poll_cb> for an example.	1753	See C<poll_cb> for an example.
203		1754
204	=item IO::AIO::poll_cb	1755	=item IO::AIO::poll_cb
205		1756
206	Process all outstanding events on the result pipe. You have to call this	1757	Process some outstanding events on the result pipe. You have to call
207	regularly. Returns the number of events processed. Returns immediately	1758	this regularly. Returns C<0> if all events could be processed (or there
208	when no events are outstanding.	1759	were no events to process), or C<-1> if it returned earlier for whatever
		1760	reason. Returns immediately when no events are outstanding. The amount of
		1761	events processed depends on the settings of C<IO::AIO::max_poll_req> and
		1762	C<IO::AIO::max_poll_time>.
209		1763
210	You can use Event to multiplex, e.g.:	1764	If not all requests were processed for whatever reason, the filehandle
		1765	will still be ready when C<poll_cb> returns, so normally you don't have to
		1766	do anything special to have it called later.
		1767
		1768	Apart from calling C<IO::AIO::poll_cb> when the event filehandle becomes
		1769	ready, it can be beneficial to call this function from loops which submit
		1770	a lot of requests, to make sure the results get processed when they become
		1771	available and not just when the loop is finished and the event loop takes
		1772	over again. This function returns very fast when there are no outstanding
		1773	requests.
		1774
		1775	Example: Install an Event watcher that automatically calls
		1776	IO::AIO::poll_cb with high priority (more examples can be found in the
		1777	SYNOPSIS section, at the top of this document):
211		1778
212	Event->io (fd => IO::AIO::poll_fileno,	1779	Event->io (fd => IO::AIO::poll_fileno,
213	poll => 'r', async => 1,	1780	poll => 'r', async => 1,
214	cb => \&IO::AIO::poll_cb);	1781	cb => \&IO::AIO::poll_cb);
215		1782
216	=item IO::AIO::poll_wait	1783	=item IO::AIO::poll_wait
217		1784
		1785	If there are any outstanding requests and none of them in the result
218	Wait till the result filehandle becomes ready for reading (simply does a	1786	phase, wait till the result filehandle becomes ready for reading (simply
219	select on the filehandle. This is useful if you want to synchronously wait	1787	does a C<select> on the filehandle. This is useful if you want to
220	for some requests to finish).	1788	synchronously wait for some requests to finish).
221		1789
222	See C<nreqs> for an example.	1790	See C<nreqs> for an example.
223		1791
		1792	=item IO::AIO::poll
		1793
		1794	Waits until some requests have been handled.
		1795
		1796	Returns the number of requests processed, but is otherwise strictly
		1797	equivalent to:
		1798
		1799	IO::AIO::poll_wait, IO::AIO::poll_cb
		1800
224	=item IO::AIO::nreqs	1801	=item IO::AIO::flush
225		1802
226	Returns the number of requests currently outstanding.	1803	Wait till all outstanding AIO requests have been handled.
227		1804
228	Example: wait till there are no outstanding requests anymore:	1805	Strictly equivalent to:
229		1806
230	IO::AIO::poll_wait, IO::AIO::poll_cb	1807	IO::AIO::poll_wait, IO::AIO::poll_cb
231	while IO::AIO::nreqs;	1808	while IO::AIO::nreqs;
232		1809
		1810	=item IO::AIO::max_poll_reqs $nreqs
		1811
		1812	=item IO::AIO::max_poll_time $seconds
		1813
		1814	These set the maximum number of requests (default C<0>, meaning infinity)
		1815	that are being processed by C<IO::AIO::poll_cb> in one call, respectively
		1816	the maximum amount of time (default C<0>, meaning infinity) spent in
		1817	C<IO::AIO::poll_cb> to process requests (more correctly the mininum amount
		1818	of time C<poll_cb> is allowed to use).
		1819
		1820	Setting C<max_poll_time> to a non-zero value creates an overhead of one
		1821	syscall per request processed, which is not normally a problem unless your
		1822	callbacks are really really fast or your OS is really really slow (I am
		1823	not mentioning Solaris here). Using C<max_poll_reqs> incurs no overhead.
		1824
		1825	Setting these is useful if you want to ensure some level of
		1826	interactiveness when perl is not fast enough to process all requests in
		1827	time.
		1828
		1829	For interactive programs, values such as C<0.01> to C<0.1> should be fine.
		1830
		1831	Example: Install an Event watcher that automatically calls
		1832	IO::AIO::poll_cb with low priority, to ensure that other parts of the
		1833	program get the CPU sometimes even under high AIO load.
		1834
		1835	# try not to spend much more than 0.1s in poll_cb
		1836	IO::AIO::max_poll_time 0.1;
		1837
		1838	# use a low priority so other tasks have priority
		1839	Event->io (fd => IO::AIO::poll_fileno,
		1840	poll => 'r', nice => 1,
		1841	cb => &IO::AIO::poll_cb);
		1842
		1843	=back
		1844
		1845	=head3 CONTROLLING THE NUMBER OF THREADS
		1846
		1847	=over
		1848
233	=item IO::AIO::min_parallel $nthreads	1849	=item IO::AIO::min_parallel $nthreads
234		1850
235	Set the minimum number of AIO threads to C<$nthreads>. The default is	1851	Set the minimum number of AIO threads to C<$nthreads>. The current
236	C<1>, which means a single asynchronous operation can be done at one time	1852	default is C<8>, which means eight asynchronous operations can execute
237	(the number of outstanding operations, however, is unlimited).	1853	concurrently at any one time (the number of outstanding requests,
		1854	however, is unlimited).
238		1855
		1856	IO::AIO starts threads only on demand, when an AIO request is queued and
		1857	no free thread exists. Please note that queueing up a hundred requests can
		1858	create demand for a hundred threads, even if it turns out that everything
		1859	is in the cache and could have been processed faster by a single thread.
		1860
239	It is recommended to keep the number of threads low, as some Linux	1861	It is recommended to keep the number of threads relatively low, as some
240	kernel versions will scale negatively with the number of threads (higher	1862	Linux kernel versions will scale negatively with the number of threads
241	parallelity => MUCH higher latency). With current Linux 2.6 versions, 4-32	1863	(higher parallelity => MUCH higher latency). With current Linux 2.6
242	threads should be fine.	1864	versions, 4-32 threads should be fine.
243		1865
244	Under normal circumstances you don't need to call this function, as this	1866	Under most circumstances you don't need to call this function, as the
245	module automatically starts some threads (the exact number might change,	1867	module selects a default that is suitable for low to moderate load.
246	and is currently 4).
247		1868
248	=item IO::AIO::max_parallel $nthreads	1869	=item IO::AIO::max_parallel $nthreads
249		1870
250	Sets the maximum number of AIO threads to C<$nthreads>. If more than	1871	Sets the maximum number of AIO threads to C<$nthreads>. If more than the
251	the specified number of threads are currently running, kill them. This	1872	specified number of threads are currently running, this function kills
252	function blocks until the limit is reached.	1873	them. This function blocks until the limit is reached.
		1874
		1875	While C<$nthreads> are zero, aio requests get queued but not executed
		1876	until the number of threads has been increased again.
253		1877
254	This module automatically runs C<max_parallel 0> at program end, to ensure	1878	This module automatically runs C<max_parallel 0> at program end, to ensure
255	that all threads are killed and that there are no outstanding requests.	1879	that all threads are killed and that there are no outstanding requests.
256		1880
257	Under normal circumstances you don't need to call this function.	1881	Under normal circumstances you don't need to call this function.
258		1882
		1883	=item IO::AIO::max_idle $nthreads
		1884
		1885	Limit the number of threads (default: 4) that are allowed to idle
		1886	(i.e., threads that did not get a request to process within the idle
		1887	timeout (default: 10 seconds). That means if a thread becomes idle while
		1888	C<$nthreads> other threads are also idle, it will free its resources and
		1889	exit.
		1890
		1891	This is useful when you allow a large number of threads (e.g. 100 or 1000)
		1892	to allow for extremely high load situations, but want to free resources
		1893	under normal circumstances (1000 threads can easily consume 30MB of RAM).
		1894
		1895	The default is probably ok in most situations, especially if thread
		1896	creation is fast. If thread creation is very slow on your system you might
		1897	want to use larger values.
		1898
		1899	=item IO::AIO::idle_timeout $seconds
		1900
		1901	Sets the minimum idle timeout (default 10) after which worker threads are
		1902	allowed to exit. SEe C<IO::AIO::max_idle>.
		1903
259	=item $oldnreqs = IO::AIO::max_outstanding $nreqs	1904	=item IO::AIO::max_outstanding $maxreqs
260		1905
261	Sets the maximum number of outstanding requests to C<$nreqs>. If you	1906	Sets the maximum number of outstanding requests to C<$nreqs>. If
262	try to queue up more than this number of requests, the caller will block until	1907	you do queue up more than this number of requests, the next call to
263	some requests have been handled.	1908	C<IO::AIO::poll_cb> (and other functions calling C<poll_cb>, such as
		1909	C<IO::AIO::flush> or C<IO::AIO::poll>) will block until the limit is no
		1910	longer exceeded.
264		1911
265	The default is very large, so normally there is no practical limit. If you	1912	In other words, this setting does not enforce a queue limit, but can be
266	queue up many requests in a loop it it often improves speed if you set	1913	used to make poll functions block if the limit is exceeded.
267	this to a relatively low number, such as C<100>.
268		1914
269	Under normal circumstances you don't need to call this function.	1915	This is a very bad function to use in interactive programs because it
		1916	blocks, and a bad way to reduce concurrency because it is inexact: Better
		1917	use an C<aio_group> together with a feed callback.
		1918
		1919	It's main use is in scripts without an event loop - when you want to stat
		1920	a lot of files, you can write somehting like this:
		1921
		1922	IO::AIO::max_outstanding 32;
		1923
		1924	for my $path (...) {
		1925	aio_stat $path , ...;
		1926	IO::AIO::poll_cb;
		1927	}
		1928
		1929	IO::AIO::flush;
		1930
		1931	The call to C<poll_cb> inside the loop will normally return instantly, but
		1932	as soon as more thna C<32> reqeusts are in-flight, it will block until
		1933	some requests have been handled. This keeps the loop from pushing a large
		1934	number of C<aio_stat> requests onto the queue.
		1935
		1936	The default value for C<max_outstanding> is very large, so there is no
		1937	practical limit on the number of outstanding requests.
270		1938
271	=back	1939	=back
272		1940
		1941	=head3 STATISTICAL INFORMATION
		1942
		1943	=over
		1944
		1945	=item IO::AIO::nreqs
		1946
		1947	Returns the number of requests currently in the ready, execute or pending
		1948	states (i.e. for which their callback has not been invoked yet).
		1949
		1950	Example: wait till there are no outstanding requests anymore:
		1951
		1952	IO::AIO::poll_wait, IO::AIO::poll_cb
		1953	while IO::AIO::nreqs;
		1954
		1955	=item IO::AIO::nready
		1956
		1957	Returns the number of requests currently in the ready state (not yet
		1958	executed).
		1959
		1960	=item IO::AIO::npending
		1961
		1962	Returns the number of requests currently in the pending state (executed,
		1963	but not yet processed by poll_cb).
		1964
		1965	=back
		1966
		1967	=head3 MISCELLANEOUS FUNCTIONS
		1968
		1969	IO::AIO implements some functions that might be useful, but are not
		1970	asynchronous.
		1971
		1972	=over 4
		1973
		1974	=item IO::AIO::sendfile $ofh, $ifh, $offset, $count
		1975
		1976	Calls the C<eio_sendfile_sync> function, which is like C<aio_sendfile>,
		1977	but is blocking (this makes most sense if you know the input data is
		1978	likely cached already and the output filehandle is set to non-blocking
		1979	operations).
		1980
		1981	Returns the number of bytes copied, or C<-1> on error.
		1982
		1983	=item IO::AIO::fadvise $fh, $offset, $len, $advice
		1984
		1985	Simply calls the C<posix_fadvise> function (see its
		1986	manpage for details). The following advice constants are
		1987	available: C<IO::AIO::FADV_NORMAL>, C<IO::AIO::FADV_SEQUENTIAL>,
		1988	C<IO::AIO::FADV_RANDOM>, C<IO::AIO::FADV_NOREUSE>,
		1989	C<IO::AIO::FADV_WILLNEED>, C<IO::AIO::FADV_DONTNEED>.
		1990
		1991	On systems that do not implement C<posix_fadvise>, this function returns
		1992	ENOSYS, otherwise the return value of C<posix_fadvise>.
		1993
		1994	=item IO::AIO::madvise $scalar, $offset, $len, $advice
		1995
		1996	Simply calls the C<posix_madvise> function (see its
		1997	manpage for details). The following advice constants are
		1998	available: C<IO::AIO::MADV_NORMAL>, C<IO::AIO::MADV_SEQUENTIAL>,
		1999	C<IO::AIO::MADV_RANDOM>, C<IO::AIO::MADV_WILLNEED>, C<IO::AIO::MADV_DONTNEED>.
		2000
		2001	On systems that do not implement C<posix_madvise>, this function returns
		2002	ENOSYS, otherwise the return value of C<posix_madvise>.
		2003
		2004	=item IO::AIO::mprotect $scalar, $offset, $len, $protect
		2005
		2006	Simply calls the C<mprotect> function on the preferably AIO::mmap'ed
		2007	$scalar (see its manpage for details). The following protect
		2008	constants are available: C<IO::AIO::PROT_NONE>, C<IO::AIO::PROT_READ>,
		2009	C<IO::AIO::PROT_WRITE>, C<IO::AIO::PROT_EXEC>.
		2010
		2011	On systems that do not implement C<mprotect>, this function returns
		2012	ENOSYS, otherwise the return value of C<mprotect>.
		2013
		2014	=item IO::AIO::mmap $scalar, $length, $prot, $flags, $fh[, $offset]
		2015
		2016	Memory-maps a file (or anonymous memory range) and attaches it to the
		2017	given C<$scalar>, which will act like a string scalar. Returns true on
		2018	success, and false otherwise.
		2019
		2020	The only operations allowed on the scalar are C<substr>/C<vec> that don't
		2021	change the string length, and most read-only operations such as copying it
		2022	or searching it with regexes and so on.
		2023
		2024	Anything else is unsafe and will, at best, result in memory leaks.
		2025
		2026	The memory map associated with the C<$scalar> is automatically removed
		2027	when the C<$scalar> is destroyed, or when the C<IO::AIO::mmap> or
		2028	C<IO::AIO::munmap> functions are called.
		2029
		2030	This calls the C<mmap>(2) function internally. See your system's manual
		2031	page for details on the C<$length>, C<$prot> and C<$flags> parameters.
		2032
		2033	The C<$length> must be larger than zero and smaller than the actual
		2034	filesize.
		2035
		2036	C<$prot> is a combination of C<IO::AIO::PROT_NONE>, C<IO::AIO::PROT_EXEC>,
		2037	C<IO::AIO::PROT_READ> and/or C<IO::AIO::PROT_WRITE>,
		2038
		2039	C<$flags> can be a combination of C<IO::AIO::MAP_SHARED> or
		2040	C<IO::AIO::MAP_PRIVATE>, or a number of system-specific flags (when
		2041	not available, the are defined as 0): C<IO::AIO::MAP_ANONYMOUS>
		2042	(which is set to C<MAP_ANON> if your system only provides this
		2043	constant), C<IO::AIO::MAP_HUGETLB>, C<IO::AIO::MAP_LOCKED>,
		2044	C<IO::AIO::MAP_NORESERVE>, C<IO::AIO::MAP_POPULATE> or
		2045	C<IO::AIO::MAP_NONBLOCK>
		2046
		2047	If C<$fh> is C<undef>, then a file descriptor of C<-1> is passed.
		2048
		2049	C<$offset> is the offset from the start of the file - it generally must be
		2050	a multiple of C<IO::AIO::PAGESIZE> and defaults to C<0>.
		2051
		2052	Example:
		2053
		2054	use Digest::MD5;
		2055	use IO::AIO;
		2056
		2057	open my $fh, "<verybigfile"
		2058	or die "$!";
		2059
		2060	IO::AIO::mmap my $data, -s $fh, IO::AIO::PROT_READ, IO::AIO::MAP_SHARED, $fh
		2061	or die "verybigfile: $!";
		2062
		2063	my $fast_md5 = md5 $data;
		2064
		2065	=item IO::AIO::munmap $scalar
		2066
		2067	Removes a previous mmap and undefines the C<$scalar>.
		2068
		2069	=item IO::AIO::munlock $scalar, $offset = 0, $length = undef
		2070
		2071	Calls the C<munlock> function, undoing the effects of a previous
		2072	C<aio_mlock> call (see its description for details).
		2073
		2074	=item IO::AIO::munlockall
		2075
		2076	Calls the C<munlockall> function.
		2077
		2078	On systems that do not implement C<munlockall>, this function returns
		2079	ENOSYS, otherwise the return value of C<munlockall>.
		2080
		2081	=item IO::AIO::splice $r_fh, $r_off, $w_fh, $w_off, $length, $flags
		2082
		2083	Calls the GNU/Linux C<splice(2)> syscall, if available. If C<$r_off> or
		2084	C<$w_off> are C<undef>, then C<NULL> is passed for these, otherwise they
		2085	should be the file offset.
		2086
		2087	C<$r_fh> and C<$w_fh> should not refer to the same file, as splice might
		2088	silently corrupt the data in this case.
		2089
		2090	The following symbol flag values are available: C<IO::AIO::SPLICE_F_MOVE>,
		2091	C<IO::AIO::SPLICE_F_NONBLOCK>, C<IO::AIO::SPLICE_F_MORE> and
		2092	C<IO::AIO::SPLICE_F_GIFT>.
		2093
		2094	See the C<splice(2)> manpage for details.
		2095
		2096	=item IO::AIO::tee $r_fh, $w_fh, $length, $flags
		2097
		2098	Calls the GNU/Linux C<tee(2)> syscall, see it's manpage and the
		2099	description for C<IO::AIO::splice> above for details.
		2100
		2101	=back
		2102
273	=cut	2103	=cut
274		2104
275	# support function to convert a fd into a perl filehandle
276	sub _fd2fh {
277	return undef if $_[0] < 0;
278
279	# try to be perl5.6-compatible
280	local *AIO_FH;
281	open AIO_FH, "+<&=$_[0]"
282	or return undef;
283
284	*AIO_FH
285	}
286
287	min_parallel 4;	2105	min_parallel 8;
288		2106
289	END {	2107	END { flush }
290	max_parallel 0;
291	}
292		2108
293	1;	2109	1;
294		2110
		2111	=head1 EVENT LOOP INTEGRATION
		2112
		2113	It is recommended to use L<AnyEvent::AIO> to integrate IO::AIO
		2114	automatically into many event loops:
		2115
		2116	# AnyEvent integration (EV, Event, Glib, Tk, POE, urxvt, pureperl...)
		2117	use AnyEvent::AIO;
		2118
		2119	You can also integrate IO::AIO manually into many event loops, here are
		2120	some examples of how to do this:
		2121
		2122	# EV integration
		2123	my $aio_w = EV::io IO::AIO::poll_fileno, EV::READ, \&IO::AIO::poll_cb;
		2124
		2125	# Event integration
		2126	Event->io (fd => IO::AIO::poll_fileno,
		2127	poll => 'r',
		2128	cb => \&IO::AIO::poll_cb);
		2129
		2130	# Glib/Gtk2 integration
		2131	add_watch Glib::IO IO::AIO::poll_fileno,
		2132	in => sub { IO::AIO::poll_cb; 1 };
		2133
		2134	# Tk integration
		2135	Tk::Event::IO->fileevent (IO::AIO::poll_fileno, "",
		2136	readable => \&IO::AIO::poll_cb);
		2137
		2138	# Danga::Socket integration
		2139	Danga::Socket->AddOtherFds (IO::AIO::poll_fileno =>
		2140	\&IO::AIO::poll_cb);
		2141
		2142	=head2 FORK BEHAVIOUR
		2143
		2144	Usage of pthreads in a program changes the semantics of fork
		2145	considerably. Specifically, only async-safe functions can be called after
		2146	fork. Perl doesn't know about this, so in general, you cannot call fork
		2147	with defined behaviour in perl if pthreads are involved. IO::AIO uses
		2148	pthreads, so this applies, but many other extensions and (for inexplicable
		2149	reasons) perl itself often is linked against pthreads, so this limitation
		2150	applies to quite a lot of perls.
		2151
		2152	This module no longer tries to fight your OS, or POSIX. That means IO::AIO
		2153	only works in the process that loaded it. Forking is fully supported, but
		2154	using IO::AIO in the child is not.
		2155
		2156	You might get around by not I<using> IO::AIO before (or after)
		2157	forking. You could also try to call the L<IO::AIO::reinit> function in the
		2158	child:
		2159
		2160	=over 4
		2161
		2162	=item IO::AIO::reinit
		2163
		2164	Abandons all current requests and I/O threads and simply reinitialises all
		2165	data structures. This is not an operation supported by any standards, but
		2166	happens to work on GNU/Linux and some newer BSD systems.
		2167
		2168	The only reasonable use for this function is to call it after forking, if
		2169	C<IO::AIO> was used in the parent. Calling it while IO::AIO is active in
		2170	the process will result in undefined behaviour. Calling it at any time
		2171	will also result in any undefined (by POSIX) behaviour.
		2172
		2173	=back
		2174
		2175	=head2 MEMORY USAGE
		2176
		2177	Per-request usage:
		2178
		2179	Each aio request uses - depending on your architecture - around 100-200
		2180	bytes of memory. In addition, stat requests need a stat buffer (possibly
		2181	a few hundred bytes), readdir requires a result buffer and so on. Perl
		2182	scalars and other data passed into aio requests will also be locked and
		2183	will consume memory till the request has entered the done state.
		2184
		2185	This is not awfully much, so queuing lots of requests is not usually a
		2186	problem.
		2187
		2188	Per-thread usage:
		2189
		2190	In the execution phase, some aio requests require more memory for
		2191	temporary buffers, and each thread requires a stack and other data
		2192	structures (usually around 16k-128k, depending on the OS).
		2193
		2194	=head1 KNOWN BUGS
		2195
		2196	Known bugs will be fixed in the next release.
		2197
295	=head1 SEE ALSO	2198	=head1 SEE ALSO
296		2199
297	L<Coro>, L<Linux::AIO>.	2200	L<AnyEvent::AIO> for easy integration into event loops, L<Coro::AIO> for a
		2201	more natural syntax.
298		2202
299	=head1 AUTHOR	2203	=head1 AUTHOR
300		2204
301	Marc Lehmann <schmorp@schmorp.de>	2205	Marc Lehmann <schmorp@schmorp.de>
302	http://home.schmorp.de/	2206	http://home.schmorp.de/

Diff Legend

–	Removed lines
+	Added lines
<	Changed lines
>	Changed lines