xref: /freebsd-11-stable/lib/libc/sys/kqueue.2 (revision 6533f31aaaa8f03ffd6fe147c1ad69c3052cdc98)
1.\" Copyright (c) 2000 Jonathan Lemon
2.\" All rights reserved.
3.\"
4.\" Redistribution and use in source and binary forms, with or without
5.\" modification, are permitted provided that the following conditions
6.\" are met:
7.\" 1. Redistributions of source code must retain the above copyright
8.\"    notice, this list of conditions and the following disclaimer.
9.\" 2. Redistributions in binary form must reproduce the above copyright
10.\"    notice, this list of conditions and the following disclaimer in the
11.\"    documentation and/or other materials provided with the distribution.
12.\"
13.\" THIS SOFTWARE IS PROVIDED ``AS IS'' AND
14.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16.\" ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23.\" SUCH DAMAGE.
24.\"
25.\" $FreeBSD$
26.\"
27.Dd April 21, 2020
28.Dt KQUEUE 2
29.Os
30.Sh NAME
31.Nm kqueue ,
32.Nm kevent
33.Nd kernel event notification mechanism
34.Sh LIBRARY
35.Lb libc
36.Sh SYNOPSIS
37.In sys/types.h
38.In sys/event.h
39.In sys/time.h
40.Ft int
41.Fn kqueue "void"
42.Ft int
43.Fn kevent "int kq" "const struct kevent *changelist" "int nchanges" "struct kevent *eventlist" "int nevents" "const struct timespec *timeout"
44.Fn EV_SET "kev" ident filter flags fflags data udata
45.Sh DESCRIPTION
46The
47.Fn kqueue
48system call
49provides a generic method of notifying the user when an event
50happens or a condition holds, based on the results of small
51pieces of kernel code termed filters.
52A kevent is identified by the (ident, filter) pair; there may only
53be one unique kevent per kqueue.
54.Pp
55The filter is executed upon the initial registration of a kevent
56in order to detect whether a preexisting condition is present, and is also
57executed whenever an event is passed to the filter for evaluation.
58If the filter determines that the condition should be reported,
59then the kevent is placed on the kqueue for the user to retrieve.
60.Pp
61The filter is also run when the user attempts to retrieve the kevent
62from the kqueue.
63If the filter indicates that the condition that triggered
64the event no longer holds, the kevent is removed from the kqueue and
65is not returned.
66.Pp
67Multiple events which trigger the filter do not result in multiple
68kevents being placed on the kqueue; instead, the filter will aggregate
69the events into a single struct kevent.
70Calling
71.Fn close
72on a file descriptor will remove any kevents that reference the descriptor.
73.Pp
74The
75.Fn kqueue
76system call
77creates a new kernel event queue and returns a descriptor.
78The queue is not inherited by a child created with
79.Xr fork 2 .
80However, if
81.Xr rfork 2
82is called without the
83.Dv RFFDG
84flag, then the descriptor table is shared,
85which will allow sharing of the kqueue between two processes.
86.Pp
87The
88.Fn kevent
89system call
90is used to register events with the queue, and return any pending
91events to the user.
92The
93.Fa changelist
94argument
95is a pointer to an array of
96.Va kevent
97structures, as defined in
98.In sys/event.h .
99All changes contained in the
100.Fa changelist
101are applied before any pending events are read from the queue.
102The
103.Fa nchanges
104argument
105gives the size of
106.Fa changelist .
107The
108.Fa eventlist
109argument
110is a pointer to an array of kevent structures.
111The
112.Fa nevents
113argument
114determines the size of
115.Fa eventlist .
116When
117.Fa nevents
118is zero,
119.Fn kevent
120will return immediately even if there is a
121.Fa timeout
122specified unlike
123.Xr select 2 .
124If
125.Fa timeout
126is a non-NULL pointer, it specifies a maximum interval to wait
127for an event, which will be interpreted as a struct timespec.
128If
129.Fa timeout
130is a NULL pointer,
131.Fn kevent
132waits indefinitely.
133To effect a poll, the
134.Fa timeout
135argument should be non-NULL, pointing to a zero-valued
136.Va timespec
137structure.
138The same array may be used for the
139.Fa changelist
140and
141.Fa eventlist .
142.Pp
143The
144.Fn EV_SET
145macro is provided for ease of initializing a
146kevent structure.
147.Pp
148The
149.Va kevent
150structure is defined as:
151.Bd -literal
152struct kevent {
153	uintptr_t ident;	/* identifier for this event */
154	short	  filter;	/* filter for event */
155	u_short	  flags;	/* action flags for kqueue */
156	u_int	  fflags;	/* filter flag value */
157	intptr_t  data;		/* filter data value */
158	void	  *udata;	/* opaque user data identifier */
159};
160.Ed
161.Pp
162The fields of
163.Fa struct kevent
164are:
165.Bl -tag -width "Fa filter"
166.It Fa ident
167Value used to identify this event.
168The exact interpretation is determined by the attached filter,
169but often is a file descriptor.
170.It Fa filter
171Identifies the kernel filter used to process this event.
172The pre-defined
173system filters are described below.
174.It Fa flags
175Actions to perform on the event.
176.It Fa fflags
177Filter-specific flags.
178.It Fa data
179Filter-specific data value.
180.It Fa udata
181Opaque user-defined value passed through the kernel unchanged.
182.El
183.Pp
184The
185.Va flags
186field can contain the following values:
187.Bl -tag -width EV_DISPATCH
188.It Dv EV_ADD
189Adds the event to the kqueue.
190Re-adding an existing event
191will modify the parameters of the original event, and not result
192in a duplicate entry.
193Adding an event automatically enables it,
194unless overridden by the EV_DISABLE flag.
195.It Dv EV_ENABLE
196Permit
197.Fn kevent
198to return the event if it is triggered.
199.It Dv EV_DISABLE
200Disable the event so
201.Fn kevent
202will not return it.
203The filter itself is not disabled.
204.It Dv EV_DISPATCH
205Disable the event source immediately after delivery of an event.
206See
207.Dv EV_DISABLE
208above.
209.It Dv EV_DELETE
210Removes the event from the kqueue.
211Events which are attached to
212file descriptors are automatically deleted on the last close of
213the descriptor.
214.It Dv EV_RECEIPT
215This flag is useful for making bulk changes to a kqueue without draining
216any pending events.
217When passed as input, it forces
218.Dv EV_ERROR
219to always be returned.
220When a filter is successfully added the
221.Va data
222field will be zero.
223Note that if this flag is encountered and there is no remaining space in
224.Fa eventlist
225to hold the
226.Dv EV_ERROR
227event, then subsequent changes will not get processed.
228.It Dv EV_ONESHOT
229Causes the event to return only the first occurrence of the filter
230being triggered.
231After the user retrieves the event from the kqueue,
232it is deleted.
233.It Dv EV_CLEAR
234After the event is retrieved by the user, its state is reset.
235This is useful for filters which report state transitions
236instead of the current state.
237Note that some filters may automatically
238set this flag internally.
239.It Dv EV_EOF
240Filters may set this flag to indicate filter-specific EOF condition.
241.It Dv EV_ERROR
242See
243.Sx RETURN VALUES
244below.
245.El
246.Pp
247The predefined system filters are listed below.
248Arguments may be passed to and from the filter via the
249.Va fflags
250and
251.Va data
252fields in the kevent structure.
253.Bl -tag -width "Dv EVFILT_PROCDESC"
254.It Dv EVFILT_READ
255Takes a descriptor as the identifier, and returns whenever
256there is data available to read.
257The behavior of the filter is slightly different depending
258on the descriptor type.
259.Bl -tag -width 2n
260.It Sockets
261Sockets which have previously been passed to
262.Fn listen
263return when there is an incoming connection pending.
264.Va data
265contains the size of the listen backlog.
266.Pp
267Other socket descriptors return when there is data to be read,
268subject to the
269.Dv SO_RCVLOWAT
270value of the socket buffer.
271This may be overridden with a per-filter low water mark at the
272time the filter is added by setting the
273.Dv NOTE_LOWAT
274flag in
275.Va fflags ,
276and specifying the new low water mark in
277.Va data .
278On return,
279.Va data
280contains the number of bytes of protocol data available to read.
281.Pp
282If the read direction of the socket has shutdown, then the filter
283also sets
284.Dv EV_EOF
285in
286.Va flags ,
287and returns the socket error (if any) in
288.Va fflags .
289It is possible for EOF to be returned (indicating the connection is gone)
290while there is still data pending in the socket buffer.
291.It Vnodes
292Returns when the file pointer is not at the end of file.
293.Va data
294contains the offset from current position to end of file,
295and may be negative.
296.Pp
297This behavior is different from
298.Xr poll 2 ,
299where read events are triggered for regular files unconditionally.
300This event can be triggered unconditionally by setting the
301.Dv NOTE_FILE_POLL
302flag in
303.Va fflags .
304.It "Fifos, Pipes"
305Returns when the there is data to read;
306.Va data
307contains the number of bytes available.
308.Pp
309When the last writer disconnects, the filter will set
310.Dv EV_EOF
311in
312.Va flags .
313This may be cleared by passing in
314.Dv EV_CLEAR ,
315at which point the
316filter will resume waiting for data to become available before
317returning.
318.It "BPF devices"
319Returns when the BPF buffer is full, the BPF timeout has expired, or
320when the BPF has
321.Dq immediate mode
322enabled and there is any data to read;
323.Va data
324contains the number of bytes available.
325.El
326.It Dv EVFILT_WRITE
327Takes a descriptor as the identifier, and returns whenever
328it is possible to write to the descriptor.
329For sockets, pipes
330and fifos,
331.Va data
332will contain the amount of space remaining in the write buffer.
333The filter will set EV_EOF when the reader disconnects, and for
334the fifo case, this may be cleared by use of
335.Dv EV_CLEAR .
336Note that this filter is not supported for vnodes or BPF devices.
337.Pp
338For sockets, the low water mark and socket error handling is
339identical to the
340.Dv EVFILT_READ
341case.
342.It Dv EVFILT_AIO
343Events for this filter are not registered with
344.Fn kevent
345directly but are registered via the
346.Va aio_sigevent
347member of an asynchronous I/O request when it is scheduled via an
348asynchronous I/O system call such as
349.Fn aio_read .
350The filter returns under the same conditions as
351.Fn aio_error .
352For more details on this filter see
353.Xr sigevent 3 and
354.Xr aio 4 .
355.It Dv EVFILT_VNODE
356Takes a file descriptor as the identifier and the events to watch for in
357.Va fflags ,
358and returns when one or more of the requested events occurs on the descriptor.
359The events to monitor are:
360.Bl -tag -width "Dv NOTE_CLOSE_WRITE"
361.It Dv NOTE_ATTRIB
362The file referenced by the descriptor had its attributes changed.
363.It Dv NOTE_CLOSE
364A file descriptor referencing the monitored file, was closed.
365The closed file descriptor did not have write access.
366.It Dv NOTE_CLOSE_WRITE
367A file descriptor referencing the monitored file, was closed.
368The closed file descriptor had write access.
369.Pp
370This note, as well as
371.Dv NOTE_CLOSE ,
372are not activated when files are closed forcibly by
373.Xr unmount 2 or
374.Xr revoke 2 .
375Instead,
376.Dv NOTE_REVOKE
377is sent for such events.
378.It Dv NOTE_DELETE
379The
380.Fn unlink
381system call was called on the file referenced by the descriptor.
382.It Dv NOTE_EXTEND
383For regular file, the file referenced by the descriptor was extended.
384.Pp
385For directory, reports that a directory entry was added or removed,
386as the result of rename operation.
387The
388.Dv NOTE_EXTEND
389event is not reported when a name is changed inside the directory.
390.It Dv NOTE_LINK
391The link count on the file changed.
392In particular, the
393.Dv NOTE_LINK
394event is reported if a subdirectory was created or deleted inside
395the directory referenced by the descriptor.
396.It Dv NOTE_OPEN
397The file referenced by the descriptor was opened.
398.It Dv NOTE_READ
399A read occurred on the file referenced by the descriptor.
400.It Dv NOTE_RENAME
401The file referenced by the descriptor was renamed.
402.It Dv NOTE_REVOKE
403Access to the file was revoked via
404.Xr revoke 2
405or the underlying file system was unmounted.
406.It Dv NOTE_WRITE
407A write occurred on the file referenced by the descriptor.
408.El
409.Pp
410On return,
411.Va fflags
412contains the events which triggered the filter.
413.It Dv EVFILT_PROC
414Takes the process ID to monitor as the identifier and the events to watch for
415in
416.Va fflags ,
417and returns when the process performs one or more of the requested events.
418If a process can normally see another process, it can attach an event to it.
419The events to monitor are:
420.Bl -tag -width "Dv NOTE_TRACKERR"
421.It Dv NOTE_EXIT
422The process has exited.
423The exit status will be stored in
424.Va data .
425.It Dv NOTE_FORK
426The process has called
427.Fn fork .
428.It Dv NOTE_EXEC
429The process has executed a new process via
430.Xr execve 2
431or a similar call.
432.It Dv NOTE_TRACK
433Follow a process across
434.Fn fork
435calls.
436The parent process registers a new kevent to monitor the child process
437using the same
438.Va fflags
439as the original event.
440The child process will signal an event with
441.Dv NOTE_CHILD
442set in
443.Va fflags
444and the parent PID in
445.Va data .
446.Pp
447If the parent process fails to register a new kevent
448.Pq usually due to resource limitations ,
449it will signal an event with
450.Dv NOTE_TRACKERR
451set in
452.Va fflags ,
453and the child process will not signal a
454.Dv NOTE_CHILD
455event.
456.El
457.Pp
458On return,
459.Va fflags
460contains the events which triggered the filter.
461.It Dv EVFILT_PROCDESC
462Takes the process descriptor created by
463.Xr pdfork 2
464to monitor as the identifier and the events to watch for in
465.Va fflags ,
466and returns when the associated process performs one or more of the
467requested events.
468The events to monitor are:
469.Bl -tag -width "Dv NOTE_EXIT"
470.It Dv NOTE_EXIT
471The process has exited.
472The exit status will be stored in
473.Va data .
474.El
475.Pp
476On return,
477.Va fflags
478contains the events which triggered the filter.
479.It Dv EVFILT_SIGNAL
480Takes the signal number to monitor as the identifier and returns
481when the given signal is delivered to the process.
482This coexists with the
483.Fn signal
484and
485.Fn sigaction
486facilities, and has a lower precedence.
487The filter will record
488all attempts to deliver a signal to a process, even if the signal has
489been marked as
490.Dv SIG_IGN ,
491except for the
492.Dv SIGCHLD
493signal, which, if ignored, will not be recorded by the filter.
494Event notification happens after normal
495signal delivery processing.
496.Va data
497returns the number of times the signal has occurred since the last call to
498.Fn kevent .
499This filter automatically sets the
500.Dv EV_CLEAR
501flag internally.
502.It Dv EVFILT_TIMER
503Establishes an arbitrary timer identified by
504.Va ident .
505When adding a timer,
506.Va data
507specifies the timeout period.
508The timer will be periodic unless
509.Dv EV_ONESHOT
510is specified.
511On return,
512.Va data
513contains the number of times the timeout has expired since the last call to
514.Fn kevent .
515This filter automatically sets the EV_CLEAR flag internally.
516.Bl -tag -width "Dv NOTE_USECONDS"
517.It Dv NOTE_SECONDS
518.Va data
519is in seconds.
520.It Dv NOTE_MSECONDS
521.Va data
522is in milliseconds.
523.It Dv NOTE_USECONDS
524.Va data
525is in microseconds.
526.It Dv NOTE_NSECONDS
527.Va data
528is in nanoseconds.
529.El
530.Pp
531If
532.Va fflags
533is not set, the default is milliseconds.
534On return,
535.Va fflags
536contains the events which triggered the filter.
537.Pp
538If an existing timer is re-added, the existing timer will be
539effectively canceled (throwing away any undelivered record of previous
540timer expiration) and re-started using the new parameters contained in
541.Va data
542and
543.Va fflags .
544.Pp
545There is a system wide limit on the number of timers
546which is controlled by the
547.Va kern.kq_calloutmax
548sysctl.
549.It Dv EVFILT_USER
550Establishes a user event identified by
551.Va ident
552which is not associated with any kernel mechanism but is triggered by
553user level code.
554The lower 24 bits of the
555.Va fflags
556may be used for user defined flags and manipulated using the following:
557.Bl -tag -width "Dv NOTE_FFLAGSMASK"
558.It Dv NOTE_FFNOP
559Ignore the input
560.Va fflags .
561.It Dv NOTE_FFAND
562Bitwise AND
563.Va fflags .
564.It Dv NOTE_FFOR
565Bitwise OR
566.Va fflags .
567.It Dv NOTE_FFCOPY
568Copy
569.Va fflags .
570.It Dv NOTE_FFCTRLMASK
571Control mask for
572.Va fflags .
573.It Dv NOTE_FFLAGSMASK
574User defined flag mask for
575.Va fflags .
576.El
577.Pp
578A user event is triggered for output with the following:
579.Bl -tag -width "Dv NOTE_FFLAGSMASK"
580.It Dv NOTE_TRIGGER
581Cause the event to be triggered.
582.El
583.Pp
584On return,
585.Va fflags
586contains the users defined flags in the lower 24 bits.
587.El
588.Sh CANCELLATION BEHAVIOUR
589If
590.Fa nevents
591is non-zero, i.e., the function is potentially blocking, the call
592is a cancellation point.
593Otherwise, i.e., if
594.Fa nevents
595is zero, the call is not cancellable.
596Cancellation can only occur before any changes are made to the kqueue,
597or when the call was blocked and no changes to the queue were requested.
598.Sh RETURN VALUES
599The
600.Fn kqueue
601system call
602creates a new kernel event queue and returns a file descriptor.
603If there was an error creating the kernel event queue, a value of -1 is
604returned and errno set.
605.Pp
606The
607.Fn kevent
608system call
609returns the number of events placed in the
610.Fa eventlist ,
611up to the value given by
612.Fa nevents .
613If an error occurs while processing an element of the
614.Fa changelist
615and there is enough room in the
616.Fa eventlist ,
617then the event will be placed in the
618.Fa eventlist
619with
620.Dv EV_ERROR
621set in
622.Va flags
623and the system error in
624.Va data .
625Otherwise,
626.Dv -1
627will be returned, and
628.Dv errno
629will be set to indicate the error condition.
630If the time limit expires, then
631.Fn kevent
632returns 0.
633.Sh EXAMPLES
634.Bd -literal -compact
635#include <sys/types.h>
636#include <sys/event.h>
637#include <sys/time.h>
638#include <err.h>
639#include <fcntl.h>
640#include <stdio.h>
641#include <stdlib.h>
642#include <string.h>
643#include <unistd.h>
644
645int
646main(int argc, char **argv)
647{
648    struct kevent event;    /* Event we want to monitor */
649    struct kevent tevent;   /* Event triggered */
650    int kq, fd, ret;
651
652    if (argc != 2)
653	err(EXIT_FAILURE, "Usage: %s path\en", argv[0]);
654    fd = open(argv[1], O_RDONLY);
655    if (fd == -1)
656	err(EXIT_FAILURE, "Failed to open '%s'", argv[1]);
657
658    /* Create kqueue. */
659    kq = kqueue();
660    if (kq == -1)
661	err(EXIT_FAILURE, "kqueue() failed");
662
663    /* Initialize kevent structure. */
664    EV_SET(&event, fd, EVFILT_VNODE, EV_ADD | EV_CLEAR, NOTE_WRITE,
665	0, NULL);
666    /* Attach event to the kqueue. */
667    ret = kevent(kq, &event, 1, NULL, 0, NULL);
668    if (ret == -1)
669	err(EXIT_FAILURE, "kevent register");
670    if (event.flags & EV_ERROR)
671	errx(EXIT_FAILURE, "Event error: %s", strerror(event.data));
672
673    for (;;) {
674	/* Sleep until something happens. */
675	ret = kevent(kq, NULL, 0, &tevent, 1, NULL);
676	if (ret == -1) {
677	    err(EXIT_FAILURE, "kevent wait");
678	} else if (ret > 0) {
679	    printf("Something was written in '%s'\en", argv[1]);
680	}
681    }
682}
683.Ed
684.Sh ERRORS
685The
686.Fn kqueue
687system call fails if:
688.Bl -tag -width Er
689.It Bq Er ENOMEM
690The kernel failed to allocate enough memory for the kernel queue.
691.It Bq Er ENOMEM
692The
693.Dv RLIMIT_KQUEUES
694rlimit
695(see
696.Xr getrlimit 2 )
697for the current user would be exceeded.
698.It Bq Er EMFILE
699The per-process descriptor table is full.
700.It Bq Er ENFILE
701The system file table is full.
702.El
703.Pp
704The
705.Fn kevent
706system call fails if:
707.Bl -tag -width Er
708.It Bq Er EACCES
709The process does not have permission to register a filter.
710.It Bq Er EFAULT
711There was an error reading or writing the
712.Va kevent
713structure.
714.It Bq Er EBADF
715The specified descriptor is invalid.
716.It Bq Er EINTR
717A signal was delivered before the timeout expired and before any
718events were placed on the kqueue for return.
719.It Bq Er EINTR
720A cancellation request was delivered to the thread, but not yet handled.
721.It Bq Er EINVAL
722The specified time limit or filter is invalid.
723.It Bq Er ENOENT
724The event could not be found to be modified or deleted.
725.It Bq Er ENOMEM
726No memory was available to register the event
727or, in the special case of a timer, the maximum number of
728timers has been exceeded.
729This maximum is configurable via the
730.Va kern.kq_calloutmax
731sysctl.
732.It Bq Er ESRCH
733The specified process to attach to does not exist.
734.El
735.Pp
736When
737.Fn kevent
738call fails with
739.Er EINTR
740error, all changes in the
741.Fa changelist
742have been applied.
743.Sh SEE ALSO
744.Xr aio_error 2 ,
745.Xr aio_read 2 ,
746.Xr aio_return 2 ,
747.Xr poll 2 ,
748.Xr read 2 ,
749.Xr select 2 ,
750.Xr sigaction 2 ,
751.Xr write 2 ,
752.Xr pthread_setcancelstate 3 ,
753.Xr signal 3
754.Sh HISTORY
755The
756.Fn kqueue
757and
758.Fn kevent
759system calls first appeared in
760.Fx 4.1 .
761.Sh AUTHORS
762The
763.Fn kqueue
764system and this manual page were written by
765.An Jonathan Lemon Aq Mt jlemon@FreeBSD.org .
766.Sh BUGS
767The
768.Fa timeout
769value is limited to 24 hours; longer timeouts will be silently
770reinterpreted as 24 hours.
771