1.\" $NetBSD: kqueue.2,v 1.46 2017/07/03 21:32:50 wiz Exp $ 2.\" 3.\" Copyright (c) 2000 Jonathan Lemon 4.\" All rights reserved. 5.\" 6.\" Copyright (c) 2001, 2002, 2003 The NetBSD Foundation, Inc. 7.\" All rights reserved. 8.\" 9.\" Portions of this documentation is derived from text contributed by 10.\" Luke Mewburn. 11.\" 12.\" Redistribution and use in source and binary forms, with or without 13.\" modification, are permitted provided that the following conditions 14.\" are met: 15.\" 1. Redistributions of source code must retain the above copyright 16.\" notice, this list of conditions and the following disclaimer. 17.\" 2. Redistributions in binary form must reproduce the above copyright 18.\" notice, this list of conditions and the following disclaimer in the 19.\" documentation and/or other materials provided with the distribution. 20.\" 21.\" THIS SOFTWARE IS PROVIDED ``AS IS'' AND 22.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24.\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 25.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31.\" SUCH DAMAGE. 32.\" 33.\" $FreeBSD: src/lib/libc/sys/kqueue.2,v 1.22 2001/06/27 19:55:57 dd Exp $ 34.\" 35.Dd June 6, 2017 36.Dt KQUEUE 2 37.Os 38.Sh NAME 39.Nm kqueue , 40.Nm kqueue1 , 41.Nm kevent , 42.Nm EV_SET 43.Nd kernel event notification mechanism 44.Sh LIBRARY 45.Lb libc 46.Sh SYNOPSIS 47.In sys/event.h 48.In sys/time.h 49.Ft int 50.Fn kqueue "void" 51.Ft int 52.Fn kqueue1 "int flags" 53.Ft int 54.Fn kevent "int kq" "const struct kevent *changelist" "size_t nchanges" "struct kevent *eventlist" "size_t nevents" "const struct timespec *timeout" 55.Fn EV_SET "&kev" ident filter flags fflags data udata 56.Sh DESCRIPTION 57.Fn kqueue 58provides a generic method of notifying the user when an event 59happens or a condition holds, based on the results of small 60pieces of kernel code termed filters. 61A kevent is identified by the (ident, filter) pair; there may only 62be one unique kevent per kqueue. 63.Pp 64The filter is executed upon the initial registration of a kevent 65in order to detect whether a preexisting condition is present, and is also 66executed whenever an event is passed to the filter for evaluation. 67If the filter determines that the condition should be reported, 68then the kevent is placed on the kqueue for the user to retrieve. 69.Pp 70The filter is also run when the user attempts to retrieve the kevent 71from the kqueue. 72If the filter indicates that the condition that triggered 73the event no longer holds, the kevent is removed from the kqueue and 74is not returned. 75.Pp 76Multiple events which trigger the filter do not result in multiple 77kevents being placed on the kqueue; instead, the filter will aggregate 78the events into a single struct kevent. 79Calling 80.Xr close 2 81on a file descriptor will remove any kevents that reference the descriptor. 82.Pp 83.Fn kqueue 84creates a new kernel event queue and returns a descriptor. 85.Pp 86The 87.Fn kqueue1 88function also allows to set the following 89.Fa flags 90on the returned file descriptor: 91.Bl -column O_NONBLOCK -offset indent 92.It Dv O_CLOEXEC 93Set the close on exec property. 94.It Dv O_NONBLOCK 95Set non-blocking I/O. 96.It Dv O_NOSIGPIPE 97Return 98.Er EPIPE 99instead of raising 100.Dv SIGPIPE . 101.El 102.Pp 103The queue is not inherited by a child created with 104.Xr fork 2 . 105.\" However, if 106.\" .Xr rfork 2 107.\" is called without the 108.\" .Dv RFFDG 109.\" flag, then the descriptor table is shared, 110.\" which will allow sharing of the kqueue between two processes. 111.Pp 112.Fn kevent 113is used to register events with the queue, and return any pending 114events to the user. 115.Fa changelist 116is a pointer to an array of 117.Va kevent 118structures, as defined in 119.In sys/event.h . 120All changes contained in the 121.Fa changelist 122are applied before any pending events are read from the queue. 123.Fa nchanges 124gives the size of 125.Fa changelist . 126.Fa eventlist 127is a pointer to an array of kevent structures. 128.Fa nevents 129determines the size of 130.Fa eventlist . 131If 132.Fa timeout 133is a 134.No non- Ns Dv NULL 135pointer, it specifies a maximum interval to wait 136for an event, which will be interpreted as a 137.Li struct timespec . 138If 139.Fa timeout 140is a 141.Dv NULL 142pointer, 143.Fn kevent 144waits indefinitely. 145To effect a poll, the 146.Fa timeout 147argument should be 148.No non- Ns Dv NULL , 149pointing to a zero-valued 150.Xr timespec 3 151structure. 152The same array may be used for the 153.Fa changelist 154and 155.Fa eventlist . 156.Pp 157.Fn EV_SET 158is a macro which is provided for ease of initializing a kevent structure. 159This macro does not evaluate its parameters multiple times. 160.Pp 161The 162.Va kevent 163structure is defined as: 164.Bd -literal 165struct kevent { 166 uintptr_t ident; /* identifier for this event */ 167 uint32_t filter; /* filter for event */ 168 uint32_t flags; /* action flags for kqueue */ 169 uint32_t fflags; /* filter flag value */ 170 int64_t data; /* filter data value */ 171 intptr_t udata; /* opaque user data identifier */ 172}; 173.Ed 174.Pp 175The fields of 176.Fa struct kevent 177are: 178.Bl -tag -width XXXfilter -offset indent 179.It ident 180Value used to identify this event. 181The exact interpretation is determined by the attached filter, 182but often is a file descriptor. 183.It filter 184Identifies the kernel filter used to process this event. 185There are pre-defined system filters (which are described below), and 186other filters may be added by kernel subsystems as necessary. 187.It flags 188Actions to perform on the event. 189.It fflags 190Filter-specific flags. 191.It data 192Filter-specific data value. 193.It udata 194Opaque user-defined value passed through the kernel unchanged. 195.El 196.Pp 197The 198.Va flags 199field can contain the following values: 200.Bl -tag -width XXXEV_ONESHOT -offset indent 201.It EV_ADD 202Adds the event to the kqueue. 203Re-adding an existing event will modify the parameters of the original 204event, and not result in a duplicate entry. 205Adding an event automatically enables it, 206unless overridden by the EV_DISABLE flag. 207.It EV_ENABLE 208Permit 209.Fn kevent 210to return the event if it is triggered. 211.It EV_DISABLE 212Disable the event so 213.Fn kevent 214will not return it. 215The filter itself is not disabled. 216.It EV_DISPATCH 217Disable the event source immediately after delivery of an event. 218See 219.Dv EV_DISABLE 220above. 221.It EV_DELETE 222Removes the event from the kqueue. 223Events which are attached to file descriptors are automatically deleted 224on the last close of the descriptor. 225.It EV_RECEIPT 226This flag is useful for making bulk changes to a kqueue without draining 227any pending events. 228When passed as input, it forces 229.Dv EV_ERROR 230to always be returned. 231When a filter is successfully added the 232.Va data 233field will be zero. 234.It EV_ONESHOT 235Causes the event to return only the first occurrence of the filter 236being triggered. 237After the user retrieves the event from the kqueue, it is deleted. 238.It EV_CLEAR 239After the event is retrieved by the user, its state is reset. 240This is useful for filters which report state transitions 241instead of the current state. 242Note that some filters may automatically set this flag internally. 243.It EV_EOF 244Filters may set this flag to indicate filter-specific EOF condition. 245.It EV_ERROR 246See 247.Sx RETURN VALUES 248below. 249.El 250.Ss Filters 251Filters are identified by a number. 252There are two types of filters; pre-defined filters which 253are described below, and third-party filters that may be added with 254.Xr kfilter_register 9 255by kernel sub-systems, third-party device drivers, or loadable 256kernel modules. 257.Pp 258As a third-party filter is referenced by a well-known name instead 259of a statically assigned number, two 260.Xr ioctl 2 Ns s 261are supported on the file descriptor returned by 262.Fn kqueue 263to map a filter name to a filter number, and vice-versa (passing 264arguments in a structure described below): 265.Bl -tag -width KFILTER_BYFILTER -offset indent 266.It KFILTER_BYFILTER 267Map 268.Va filter 269to 270.Va name , 271which is of size 272.Va len . 273.It KFILTER_BYNAME 274Map 275.Va name 276to 277.Va filter . 278.Va len 279is ignored. 280.El 281.Pp 282The following structure is used to pass arguments in and out of the 283.Xr ioctl 2 : 284.Bd -literal -offset indent 285struct kfilter_mapping { 286 char *name; /* name to lookup or return */ 287 size_t len; /* length of name */ 288 uint32_t filter; /* filter to lookup or return */ 289}; 290.Ed 291.Pp 292Arguments may be passed to and from the filter via the 293.Va fflags 294and 295.Va data 296fields in the kevent structure. 297.Pp 298The predefined system filters are: 299.Bl -tag -width EVFILT_SIGNAL 300.It EVFILT_READ 301Takes a descriptor as the identifier, and returns whenever 302there is data available to read. 303The behavior of the filter is slightly different depending 304on the descriptor type. 305.Bl -tag -width 2n 306.It Sockets 307Sockets which have previously been passed to 308.Xr listen 2 309return when there is an incoming connection pending. 310.Va data 311contains the size of the listen backlog (i.e., the number of 312connections ready to be accepted with 313.Xr accept 2 . ) 314.Pp 315Other socket descriptors return when there is data to be read, 316subject to the 317.Dv SO_RCVLOWAT 318value of the socket buffer. 319This may be overridden with a per-filter low water mark at the 320time the filter is added by setting the 321NOTE_LOWAT 322flag in 323.Va fflags , 324and specifying the new low water mark in 325.Va data . 326On return, 327.Va data 328contains the number of bytes in the socket buffer. 329.Pp 330If the read direction of the socket has shutdown, then the filter 331also sets EV_EOF in 332.Va flags , 333and returns the socket error (if any) in 334.Va fflags . 335It is possible for EOF to be returned (indicating the connection is gone) 336while there is still data pending in the socket buffer. 337.It Vnodes 338Returns when the file pointer is not at the end of file. 339.Va data 340contains the offset from current position to end of file, 341and may be negative. 342.It "Fifos, Pipes" 343Returns when there is data to read; 344.Va data 345contains the number of bytes available. 346.Pp 347When the last writer disconnects, the filter will set EV_EOF in 348.Va flags . 349This may be cleared by passing in EV_CLEAR, at which point the 350filter will resume waiting for data to become available before 351returning. 352.El 353.It EVFILT_WRITE 354Takes a descriptor as the identifier, and returns whenever 355it is possible to write to the descriptor. 356For sockets, pipes, fifos, and ttys, 357.Va data 358will contain the amount of space remaining in the write buffer. 359The filter will set EV_EOF when the reader disconnects, and for 360the fifo case, this may be cleared by use of EV_CLEAR. 361Note that this filter is not supported for vnodes. 362.Pp 363For sockets, the low water mark and socket error handling is 364identical to the EVFILT_READ case. 365.It EVFILT_AIO 366This is not implemented in 367.Nx . 368.ig 369The sigevent portion of the AIO request is filled in, with 370.Va sigev_notify_kqueue 371containing the descriptor of the kqueue that the event should 372be attached to, 373.Va sigev_value 374containing the udata value, and 375.Va sigev_notify 376set to SIGEV_EVENT. 377When the aio_* function is called, the event will be registered 378with the specified kqueue, and the 379.Va ident 380argument set to the 381.Fa struct aiocb 382returned by the aio_* function. 383The filter returns under the same conditions as aio_error. 384.Pp 385Alternatively, a kevent structure may be initialized, with 386.Va ident 387containing the descriptor of the kqueue, and the 388address of the kevent structure placed in the 389.Va aio_lio_opcode 390field of the AIO request. 391However, this approach will not work on 392architectures with 64-bit pointers, and should be considered deprecated. 393.. 394.It EVFILT_VNODE 395Takes a file descriptor as the identifier and the events to watch for in 396.Va fflags , 397and returns when one or more of the requested events occurs on the descriptor. 398The events to monitor are: 399.Bl -tag -width XXNOTE_RENAME 400.It NOTE_DELETE 401.Xr unlink 2 402was called on the file referenced by the descriptor. 403.It NOTE_WRITE 404A write occurred on the file referenced by the descriptor. 405.It NOTE_EXTEND 406The file referenced by the descriptor was extended. 407.It NOTE_ATTRIB 408The file referenced by the descriptor had its attributes changed. 409.It NOTE_LINK 410The link count on the file changed. 411.It NOTE_RENAME 412The file referenced by the descriptor was renamed. 413.It NOTE_REVOKE 414Access to the file was revoked via 415.Xr revoke 2 416or the underlying file system was unmounted. 417.El 418.Pp 419On return, 420.Va fflags 421contains the events which triggered the filter. 422.It EVFILT_PROC 423Takes the process ID to monitor as the identifier and the events to watch for 424in 425.Va fflags , 426and returns when the process performs one or more of the requested events. 427If a process can normally see another process, it can attach an event to it. 428The events to monitor are: 429.Bl -tag -width XXNOTE_TRACKERR 430.It NOTE_EXIT 431The process has exited. 432The exit code of the process is stored in 433.Va data . 434.It NOTE_FORK 435The process has called 436.Xr fork 2 . 437.It NOTE_EXEC 438The process has executed a new process via 439.Xr execve 2 440or similar call. 441.It NOTE_TRACK 442Follow a process across 443.Xr fork 2 444calls. 445The parent process will return with NOTE_TRACK set in the 446.Va fflags 447field, while the child process will return with NOTE_CHILD set in 448.Va fflags 449and the parent PID in 450.Va data . 451.It NOTE_TRACKERR 452This flag is returned if the system was unable to attach an event to 453the child process, usually due to resource limitations. 454.El 455.Pp 456On return, 457.Va fflags 458contains the events which triggered the filter. 459.It EVFILT_SIGNAL 460Takes the signal number to monitor as the identifier and returns 461when the given signal is delivered to the current process. 462This coexists with the 463.Xr signal 3 464and 465.Xr sigaction 2 466facilities, and has a lower precedence. 467The filter will record 468all attempts to deliver a signal to a process, even if the signal has 469been marked as SIG_IGN. 470Event notification happens after normal signal delivery processing. 471.Va data 472returns the number of times the signal has occurred since the last call to 473.Fn kevent . 474This filter automatically sets the EV_CLEAR flag internally. 475.It EVFILT_TIMER 476Establishes an arbitrary timer identified by 477.Va ident . 478When adding a timer, 479.Va data 480specifies the timeout period in milliseconds. 481The timer will be periodic unless EV_ONESHOT is specified. 482On return, 483.Va data 484contains the number of times the timeout has expired since the last call to 485.Fn kevent . 486This filter automatically sets the EV_CLEAR flag internally. 487.El 488.Sh RETURN VALUES 489.Fn kqueue 490creates a new kernel event queue and returns a file descriptor. 491If there was an error creating the kernel event queue, a value of \-1 is 492returned and 493.Dv errno 494is set. 495.Pp 496.Fn kevent 497returns the number of events placed in the 498.Fa eventlist , 499up to the value given by 500.Fa nevents . 501If an error occurs while processing an element of the 502.Fa changelist 503and there is enough room in the 504.Fa eventlist , 505then the event will be placed in the 506.Fa eventlist 507with 508.Dv EV_ERROR 509set in 510.Va flags 511and the system error in 512.Va data . 513Otherwise, 514.Dv \-1 515will be returned, and 516.Dv errno 517will be set to indicate the error condition. 518If the time limit expires, then 519.Fn kevent 520returns 0. 521.Sh EXAMPLES 522The following example program monitors a file (provided to it as the first 523argument) and prints information about some common events it receives 524notifications for: 525.Bd -literal -offset indent 526#include <sys/types.h> 527#include <sys/event.h> 528#include <sys/time.h> 529#include <stdio.h> 530#include <unistd.h> 531#include <stdlib.h> 532#include <fcntl.h> 533#include <err.h> 534 535int 536main(int argc, char *argv[]) 537{ 538 int fd, kq, nev; 539 struct kevent ev; 540 static const struct timespec tout = { 1, 0 }; 541 542 if ((fd = open(argv[1], O_RDONLY)) == -1) 543 err(1, "Cannot open `%s'", argv[1]); 544 545 if ((kq = kqueue()) == -1) 546 err(1, "Cannot create kqueue"); 547 548 EV_SET(&ev, fd, EVFILT_VNODE, EV_ADD | EV_ENABLE | EV_CLEAR, 549 NOTE_DELETE|NOTE_WRITE|NOTE_EXTEND|NOTE_ATTRIB|NOTE_LINK| 550 NOTE_RENAME|NOTE_REVOKE, 0, 0); 551 if (kevent(kq, &ev, 1, NULL, 0, &tout) == -1) 552 err(1, "kevent"); 553 for (;;) { 554 nev = kevent(kq, NULL, 0, &ev, 1, &tout); 555 if (nev == -1) 556 err(1, "kevent"); 557 if (nev == 0) 558 continue; 559 if (ev.fflags & NOTE_DELETE) { 560 printf("deleted "); 561 ev.fflags &= ~NOTE_DELETE; 562 } 563 if (ev.fflags & NOTE_WRITE) { 564 printf("written "); 565 ev.fflags &= ~NOTE_WRITE; 566 } 567 if (ev.fflags & NOTE_EXTEND) { 568 printf("extended "); 569 ev.fflags &= ~NOTE_EXTEND; 570 } 571 if (ev.fflags & NOTE_ATTRIB) { 572 printf("chmod/chown/utimes "); 573 ev.fflags &= ~NOTE_ATTRIB; 574 } 575 if (ev.fflags & NOTE_LINK) { 576 printf("hardlinked "); 577 ev.fflags &= ~NOTE_LINK; 578 } 579 if (ev.fflags & NOTE_RENAME) { 580 printf("renamed "); 581 ev.fflags &= ~NOTE_RENAME; 582 } 583 if (ev.fflags & NOTE_REVOKE) { 584 printf("revoked "); 585 ev.fflags &= ~NOTE_REVOKE; 586 } 587 printf("\\n"); 588 if (ev.fflags) 589 warnx("unknown event 0x%x\\n", ev.fflags); 590 } 591} 592.Ed 593.Sh ERRORS 594The 595.Fn kqueue 596function fails if: 597.Bl -tag -width Er 598.It Bq Er EMFILE 599The per-process descriptor table is full. 600.It Bq Er ENFILE 601The system file table is full. 602.It Bq Er ENOMEM 603The kernel failed to allocate enough memory for the kernel queue. 604.El 605.Pp 606The 607.Fn kevent 608function fails if: 609.Bl -tag -width Er 610.It Bq Er EACCES 611The process does not have permission to register a filter. 612.It Bq Er EBADF 613The specified descriptor is invalid. 614.It Bq Er EFAULT 615There was an error reading or writing the 616.Va kevent 617structure. 618.It Bq Er EINTR 619A signal was delivered before the timeout expired and before any 620events were placed on the kqueue for return. 621.It Bq Er EINVAL 622The specified time limit or filter is invalid. 623.It Bq Er ENOENT 624The event could not be found to be modified or deleted. 625.It Bq Er ENOMEM 626No memory was available to register the event. 627.It Bq Er EOPNOTSUPP 628This type of file descriptor is not supported for 629.Fn kevent 630operations. 631.It Bq Er ESRCH 632The specified process to attach to does not exist. 633.El 634.Sh SEE ALSO 635.\" .Xr aio_error 2 , 636.\" .Xr aio_read 2 , 637.\" .Xr aio_return 2 , 638.Xr fork 2 , 639.Xr ioctl 2 , 640.Xr listen 2 , 641.Xr poll 2 , 642.Xr read 2 , 643.Xr select 2 , 644.Xr sigaction 2 , 645.Xr unlink 2 , 646.Xr write 2 , 647.Xr signal 3 , 648.Xr timespec 3 , 649.Xr kfilter_register 9 , 650.Xr knote 9 651.Rs 652.%A Jonathan Lemon 653.%T "Kqueue: A Generic and Scalable Event Notification Facility" 654.%I USENIX Association 655.%B Proceedings of the FREENIX Track: 2001 USENIX Annual Technical Conference 656.%D June 25-30, 2001 657.%U http://www.usenix.org/event/usenix01/freenix01/full_papers/lemon/lemon.pdf 658.Re 659.Sh HISTORY 660The 661.Fn kqueue 662and 663.Fn kevent 664functions first appeared in 665.Fx 4.1 , 666and then in 667.Nx 2.0 . 668The 669.Fn kqueue1 670function first appeared in 671.Nx 6.0 . 672.Pp 673The 674.Fn EV_SET 675macro was protected from evaluating multiple times the first argument in 676.Nx 8.0 . 677