path: root/man/man7/epoll.7

.\"  Copyright (C) 2003  Davide Libenzi
.\"
.\" SPDX-License-Identifier: GPL-2.0-or-later
.\"
.\"  Davide Libenzi <davidel@xmailserver.org>
.\"
.TH epoll 7 2024-05-02 "Linux man-pages (unreleased)"
.SH NAME
epoll \- I/O event notification facility
.SH SYNOPSIS
.nf
.B #include <sys/epoll.h>
.fi
.SH DESCRIPTION
The
.B epoll
API performs a similar task to
.BR poll (2):
monitoring multiple file descriptors to see if I/O is possible on any of them.
The
.B epoll
API can be used either as an edge-triggered or a level-triggered
interface and scales well to large numbers of watched file descriptors.
.P
The central concept of the
.B epoll
API is the
.B epoll
.IR instance ,
an in-kernel data structure which, from a user-space perspective,
can be considered as a container for two lists:
.IP \[bu] 3
The
.I interest
list (sometimes also called the
.B epoll
set): the set of file descriptors that the process has registered
an interest in monitoring.
.IP \[bu]
The
.I ready
list: the set of file descriptors that are "ready" for I/O.
The ready list is a subset of
(or, more precisely, a set of references to)
the file descriptors in the interest list.
The ready list is dynamically populated
by the kernel as a result of I/O activity on those file descriptors.
.P
The following system calls are provided to
create and manage an
.B epoll
instance:
.IP \[bu] 3
.BR epoll_create (2)
creates a new
.B epoll
instance and returns a file descriptor referring to that instance.
(The more recent
.BR epoll_create1 (2)
extends the functionality of
.BR epoll_create (2).)
.IP \[bu]
Interest in particular file descriptors is then registered via
.BR epoll_ctl (2),
which adds items to the interest list of the
.B epoll
instance.
.IP \[bu]
.BR epoll_wait (2)
waits for I/O events,
blocking the calling thread if no events are currently available.
(This system call can be thought of as fetching items from
the ready list of the
.B epoll
instance.)
.\"
.SS Level-triggered and edge-triggered
The
.B epoll
event distribution interface is able to behave both as edge-triggered
(ET) and as level-triggered (LT).
The difference between the two mechanisms
can be described as follows.
Suppose that
this scenario happens:
.IP (1) 5
The file descriptor that represents the read side of a pipe
.RI ( rfd )
is registered on the
.B epoll
instance.
.IP (2)
A pipe writer writes 2\ kB of data on the write side of the pipe.
.IP (3)
A call to
.BR epoll_wait (2)
is done that will return
.I rfd
as a ready file descriptor.
.IP (4)
The pipe reader reads 1\ kB of data from
.IR rfd .
.IP (5)
A call to
.BR epoll_wait (2)
is done.
.P
If the
.I rfd
file descriptor has been added to the
.B epoll
interface using the
.B EPOLLET
(edge-triggered)
flag, the call to
.BR epoll_wait (2)
done in step
.B 5
will probably hang despite the available data still present in the file
input buffer;
meanwhile the remote peer might be expecting a response based on the
data it already sent.
The reason for this is that edge-triggered mode
delivers events only when changes occur on the monitored file descriptor.
So, in step
.B 5
the caller might end up waiting for some data that is already present inside
the input buffer.
In the above example, an event on
.I rfd
will be generated because of the write done in
.B 2
and the event is consumed in
.BR 3 .
Since the read operation done in
.B 4
does not consume the whole buffer data, the call to
.BR epoll_wait (2)
done in step
.B 5
might block indefinitely.
.P
An application that employs the
.B EPOLLET
flag should use nonblocking file descriptors to avoid having a blocking
read or write starve a task that is handling multiple file descriptors.
The suggested way to use
.B epoll
as an edge-triggered
.RB ( EPOLLET )
interface is as follows:
.IP (1) 5
with nonblocking file descriptors; and
.IP (2)
by waiting for an event only after
.BR read (2)
or
.BR write (2)
return
.BR EAGAIN .
.P
By contrast, when used as a level-triggered interface
(the default, when
.B EPOLLET
is not specified),
.B epoll
is simply a faster
.BR poll (2),
and can be used wherever the latter is used since it shares the
same semantics.
.P
Since even with edge-triggered
.BR epoll ,
multiple events can be generated upon receipt of multiple chunks of data,
the caller has the option to specify the
.B EPOLLONESHOT
flag, to tell
.B epoll
to disable the associated file descriptor after the receipt of an event with
.BR epoll_wait (2).
When the
.B EPOLLONESHOT
flag is specified,
it is the caller's responsibility to rearm the file descriptor using
.BR epoll_ctl (2)
with
.BR EPOLL_CTL_MOD .
.P
If multiple threads
(or processes, if child processes have inherited the
.B epoll
file descriptor across
.BR fork (2))
are blocked in
.BR epoll_wait (2)
waiting on the same epoll file descriptor and a file descriptor
in the interest list that is marked for edge-triggered
.RB ( EPOLLET )
notification becomes ready,
just one of the threads (or processes) is awoken from
.BR epoll_wait (2).
This provides a useful optimization for avoiding "thundering herd" wake-ups
in some scenarios.
.\"
.SS Interaction with autosleep
If the system is in
.B autosleep
mode via
.I /sys/power/autosleep
and an event happens which wakes the device from sleep, the device
driver will keep the device awake only until that event is queued.
To keep the device awake until the event has been processed,
it is necessary to use the
.BR epoll_ctl (2)
.B EPOLLWAKEUP
flag.
.P
When the
.B EPOLLWAKEUP
flag is set in the
.B events
field for a
.IR "struct epoll_event" ,
the system will be kept awake from the moment the event is queued,
through the
.BR epoll_wait (2)
call which returns the event until the subsequent
.BR epoll_wait (2)
call.
If the event should keep the system awake beyond that time,
then a separate
.I wake_lock
should be taken before the second
.BR epoll_wait (2)
call.
.SS /proc interfaces
The following interfaces can be used to limit the amount of
kernel memory consumed by epoll:
.\" Following was added in Linux 2.6.28, but them removed in Linux 2.6.29
.\" .TP
.\" .IR /proc/sys/fs/epoll/max_user_instances " (since Linux 2.6.28)"
.\" This specifies an upper limit on the number of epoll instances
.\" that can be created per real user ID.
.TP
.IR /proc/sys/fs/epoll/max_user_watches " (since Linux 2.6.28)"
This specifies a limit on the total number of
file descriptors that a user can register across
all epoll instances on the system.
The limit is per real user ID.
Each registered file descriptor costs roughly 90 bytes on a 32-bit kernel,
and roughly 160 bytes on a 64-bit kernel.
Currently,
.\" Linux 2.6.29 (in Linux 2.6.28, the default was 1/32 of lowmem)
the default value for
.I max_user_watches
is 1/25 (4%) of the available low memory,
divided by the registration cost in bytes.
.SS Example for suggested usage
While the usage of
.B epoll
when employed as a level-triggered interface does have the same
semantics as
.BR poll (2),
the edge-triggered usage requires more clarification to avoid stalls
in the application event loop.
In this example, listener is a
nonblocking socket on which
.BR listen (2)
has been called.
The function
.I do_use_fd()
uses the new ready file descriptor until
.B EAGAIN
is returned by either
.BR read (2)
or
.BR write (2).
An event-driven state machine application should, after having received
.BR EAGAIN ,
record its current state so that at the next call to
.I do_use_fd()
it will continue to
.BR read (2)
or
.BR write (2)
from where it stopped before.
.P
.in +4n
.EX
#define MAX_EVENTS 10
struct epoll_event ev, events[MAX_EVENTS];
int listen_sock, conn_sock, nfds, epollfd;
\&
/* Code to set up listening socket, \[aq]listen_sock\[aq],
   (socket(), bind(), listen()) omitted. */
\&
epollfd = epoll_create1(0);
if (epollfd == \-1) {
    perror("epoll_create1");
    exit(EXIT_FAILURE);
}
\&
ev.events = EPOLLIN;
ev.data.fd = listen_sock;
if (epoll_ctl(epollfd, EPOLL_CTL_ADD, listen_sock, &ev) == \-1) {
    perror("epoll_ctl: listen_sock");
    exit(EXIT_FAILURE);
}
\&
for (;;) {
    nfds = epoll_wait(epollfd, events, MAX_EVENTS, \-1);
    if (nfds == \-1) {
        perror("epoll_wait");
        exit(EXIT_FAILURE);
    }
\&
    for (n = 0; n < nfds; ++n) {
        if (events[n].data.fd == listen_sock) {
            conn_sock = accept(listen_sock,
                               (struct sockaddr *) &addr, &addrlen);
            if (conn_sock == \-1) {
                perror("accept");
                exit(EXIT_FAILURE);
            }
            setnonblocking(conn_sock);
            ev.events = EPOLLIN | EPOLLET;
            ev.data.fd = conn_sock;
            if (epoll_ctl(epollfd, EPOLL_CTL_ADD, conn_sock,
                        &ev) == \-1) {
                perror("epoll_ctl: conn_sock");
                exit(EXIT_FAILURE);
            }
        } else {
            do_use_fd(events[n].data.fd);
        }
    }
}
.EE
.in
.P
When used as an edge-triggered interface, for performance reasons, it is
possible to add the file descriptor inside the
.B epoll
interface
.RB ( EPOLL_CTL_ADD )
once by specifying
.RB ( EPOLLIN | EPOLLOUT ).
This allows you to avoid
continuously switching between
.B EPOLLIN
and
.B EPOLLOUT
calling
.BR epoll_ctl (2)
with
.BR EPOLL_CTL_MOD .
.SS Questions and answers
.IP \[bu] 3
What is the key used to distinguish the file descriptors registered in an
interest list?
.IP
The key is the combination of the file descriptor number and
the open file description
(also known as an "open file handle",
the kernel's internal representation of an open file).
.IP \[bu]
What happens if you register the same file descriptor on an
.B epoll
instance twice?
.IP
You will probably get
.BR EEXIST .
However, it is possible to add a duplicate
.RB ( dup (2),
.BR dup2 (2),
.BR fcntl (2)
.BR F_DUPFD )
file descriptor to the same
.B epoll
instance.
.\" But a file descriptor duplicated by fork(2) can't be added to the
.\" set, because the [file *, fd] pair is already in the epoll set.
.\" That is a somewhat ugly inconsistency.  On the one hand, a child process
.\" cannot add the duplicate file descriptor to the epoll set.  (In every
.\" other case that I can think of, file descriptors duplicated by fork have
.\" similar semantics to file descriptors duplicated by dup() and friends.)  On
.\" the other hand, the very fact that the child has a duplicate of the
.\" file descriptor means that even if the parent closes its file descriptor,
.\" then epoll_wait() in the parent will continue to receive notifications for
.\" that file descriptor because of the duplicated file descriptor in the child.
.\"
.\" See http://thread.gmane.org/gmane.linux.kernel/596462/
.\" "epoll design problems with common fork/exec patterns"
.\"
.\" mtk, Feb 2008
This can be a useful technique for filtering events,
if the duplicate file descriptors are registered with different
.I events
masks.
.IP \[bu]
Can two
.B epoll
instances wait for the same file descriptor?
If so, are events reported to both
.B epoll
file descriptors?
.IP
Yes, and events would be reported to both.
However, careful programming may be needed to do this correctly.
.IP \[bu]
Is the
.B epoll
file descriptor itself poll/epoll/selectable?
.IP
Yes.
If an
.B epoll
file descriptor has events waiting, then it will
indicate as being readable.
.IP \[bu]
What happens if one attempts to put an
.B epoll
file descriptor into its own file descriptor set?
.IP
The
.BR epoll_ctl (2)
call fails
.RB ( EINVAL ).
However, you can add an
.B epoll
file descriptor inside another
.B epoll
file descriptor set.
.IP \[bu]
Can I send an
.B epoll
file descriptor over a UNIX domain socket to another process?
.IP
Yes, but it does not make sense to do this, since the receiving process
would not have copies of the file descriptors in the interest list.
.IP \[bu]
Will closing a file descriptor cause it to be removed from all
.B epoll
interest lists?
.IP
Yes, but be aware of the following point.
A file descriptor is a reference to an open file description (see
.BR open (2)).
Whenever a file descriptor is duplicated via
.BR dup (2),
.BR dup2 (2),
.BR fcntl (2)
.BR F_DUPFD ,
or
.BR fork (2),
a new file descriptor referring to the same open file description is
created.
An open file description continues to exist until all
file descriptors referring to it have been closed.
.IP
A file descriptor is removed from an
interest list only after all the file descriptors referring to the underlying
open file description have been closed.
This means that even after a file descriptor that is part of an
interest list has been closed,
events may be reported for that file descriptor if other file
descriptors referring to the same underlying file description remain open.
To prevent this happening,
the file descriptor must be explicitly removed from the interest list (using
.BR epoll_ctl (2)
.BR EPOLL_CTL_DEL )
before it is duplicated.
Alternatively,
the application must ensure that all file descriptors are closed
(which may be difficult if file descriptors were duplicated
behind the scenes by library functions that used
.BR dup (2)
or
.BR fork (2)).
.IP \[bu]
If more than one event occurs between
.BR epoll_wait (2)
calls, are they combined or reported separately?
.IP
They will be combined.
.IP \[bu]
Does an operation on a file descriptor affect the
already collected but not yet reported events?
.IP
You can do two operations on an existing file descriptor.
Remove would be meaningless for
this case.
Modify will reread available I/O.
.IP \[bu]
Do I need to continuously read/write a file descriptor
until
.B EAGAIN
when using the
.B EPOLLET
flag (edge-triggered behavior)?
.IP
Receiving an event from
.BR epoll_wait (2)
should suggest to you that such
file descriptor is ready for the requested I/O operation.
You must consider it ready until the next (nonblocking)
read/write yields
.BR EAGAIN .
When and how you will use the file descriptor is entirely up to you.
.IP
For packet/token-oriented files (e.g., datagram socket,
terminal in canonical mode),
the only way to detect the end of the read/write I/O space
is to continue to read/write until
.BR EAGAIN .
.IP
For stream-oriented files (e.g., pipe, FIFO, stream socket), the
condition that the read/write I/O space is exhausted can also be detected by
checking the amount of data read from / written to the target file
descriptor.
For example, if you call
.BR read (2)
by asking to read a certain amount of data and
.BR read (2)
returns a lower number of bytes, you
can be sure of having exhausted the read I/O space for the file
descriptor.
The same is true when writing using
.BR write (2).
(Avoid this latter technique if you cannot guarantee that
the monitored file descriptor always refers to a stream-oriented file.)
.SS Possible pitfalls and ways to avoid them
.IP \[bu] 3
.B Starvation (edge-triggered)
.IP
If there is a large amount of I/O space,
it is possible that by trying to drain
it the other files will not get processed causing starvation.
(This problem is not specific to
.BR epoll .)
.IP
The solution is to maintain a ready list
and mark the file descriptor as ready
in its associated data structure, thereby allowing the application to
remember which files need to be processed but still round robin amongst
all the ready files.
This also supports ignoring subsequent events you
receive for file descriptors that are already ready.
.IP \[bu]
.B If using an event cache...
.IP
If you use an event cache or store all the file descriptors returned from
.BR epoll_wait (2),
then make sure to provide a way to mark
its closure dynamically (i.e., caused by
a previous event's processing).
Suppose you receive 100 events from
.BR epoll_wait (2),
and in event #47 a condition causes event #13 to be closed.
If you remove the structure and
.BR close (2)
the file descriptor for event #13, then your
event cache might still say there are events waiting for that
file descriptor causing confusion.
.IP
One solution for this is to call, during the processing of event 47,
.BR epoll_ctl ( EPOLL_CTL_DEL )
to delete file descriptor 13 and
.BR close (2),
then mark its associated
data structure as removed and link it to a cleanup list.
If you find another
event for file descriptor 13 in your batch processing,
you will discover the file descriptor had been
previously removed and there will be no confusion.
.SH VERSIONS
Some other systems provide similar mechanisms;
for example,
FreeBSD has
.IR kqueue ,
and Solaris has
.IR /dev/poll .
.SH STANDARDS
Linux.
.SH HISTORY
Linux 2.5.44.
.\" Its interface should be finalized in Linux 2.5.66.
glibc 2.3.2.
.SH NOTES
The set of file descriptors that is being monitored via
an epoll file descriptor can be viewed via the entry for
the epoll file descriptor in the process's
.IR /proc/ pid /fdinfo
directory.
See
.BR proc (5)
for further details.
.P
The
.BR kcmp (2)
.B KCMP_EPOLL_TFD
operation can be used to test whether a file descriptor
is present in an epoll instance.
.SH SEE ALSO
.BR epoll_create (2),
.BR epoll_create1 (2),
.BR epoll_ctl (2),
.BR epoll_wait (2),
.BR poll (2),
.BR select (2)