POLL(2) System Calls Manual POLL(2)

poll, pollts, ppoll — synchronous I/O multiplexing

Standard C Library (libc, -lc)

#include <poll.h>

int
poll(struct pollfd *fds, nfds_t nfds, int timeout);

#include <poll.h>
#include <signal.h>
#include <time.h>

int
pollts(struct pollfd * restrict fds, nfds_t nfds, const struct timespec * restrict ts, const sigset_t * restrict sigmask);

#include <poll.h>
#include <signal.h>
#include <time.h>

int
ppoll(struct pollfd * restrict fds, nfds_t nfds, const struct timespec * restrict ts, const sigset_t * restrict sigmask);

(), pollts() and ppoll() examine a set of file descriptors to see if some of them are ready for I/O. For each object inspected, the caller provides a list of conditions (called ``events'') to check for, and the kernel returns a list of conditions that are true. The intent, as with select(2), is to check for whether I/O is possible before performing any, so as to permit a top-level event loop to process input from many sources (and output to many destinations) without blocking on any of them and thus becoming stuck.

The fds argument is a pointer to an array of pollfd structures, one per file to inspect, as defined in <poll.h> (shown below). The nfds argument gives the size of the fds array.

If timeout is neither zero nor INFTIM (-1), it specifies a maximum interval to wait for any file descriptor to become ready, in milliseconds. If timeout is INFTIM (-1), then () blocks indefinitely. If timeout is zero, then poll() will return without blocking.

Similarly, if ts is not a null pointer, it references a timespec structure which specifies a maximum interval to wait for any file descriptor to become ready. If ts is a null pointer, () and () block indefinitely. If ts is not a null pointer, referencing a zero-valued timespec structure, then pollts() and ppoll() will return without blocking.

If sigmask is not a null pointer, then the () and () functions replace the signal mask of the caller by the set of signals pointed to by sigmask while the call is in progress, and restore the caller's original signal mask before returning.

The pollfd structure:

struct pollfd {
    int    fd;       /* file descriptor */
    short  events;   /* events to look for */
    short  revents;  /* events returned */
};

The fields of struct pollfd are as follows:

fd
File descriptor to poll. (Input)
events
Conditions to poll for. (Input)
revents
Conditions that are true. (Output)

There are four conditions that can be placed in events and reported in revents:

POLLRDNORM
Normal data may be read without blocking.
POLLRDBAND
Urgent/out-of-band data may be read without blocking.
POLLWRNORM
Normal data may be written without blocking.
POLLWRBAND
Urgent/out-of-band data may be written without blocking.

There are three more conditions that are always checked for regardless of events and thus may always be reported in revents:

POLLERR
An exceptional condition has occurred on the object.
POLLHUP
The object has been disconnected.
POLLNVAL
The file descriptor is not open.

The following additional flags are defined:

POLLIN
Synonym for POLLRDNORM.
POLLOUT
Synonym for POLLWRNORM.
POLLPRI
Synonym for POLLRDBAND.

If the value passed in fd is negative, the entry is ignored and revents is set to 0. (POLLNVAL is set.)

No file descriptor will ever produce POLLHUP at the same time as POLLWRNORM.

Sockets produce POLLIN rather than POLLHUP when the remote end is closed.

poll() returns the number of descriptors that are ready for I/O, or returns -1 and sets errno if an error occurred. If the time limit expires, poll() returns 0. If poll() returns with an error, including one due to an interrupted call, the fds array will be unmodified.

This implementation differs from the historical one in that no individual file descriptor may cause poll() to return with an error. In cases where this would have happened in the historical implementation (e.g. trying to poll a revoke(2)d descriptor), this implementation instead copies the events bitmask to the revents bitmask. Attempting to perform I/O on this descriptor will then return an error. This behavior is believed to be more useful.

The ppoll() function is a wrapper for pollts() to provide compatibility with the Linux implementation.

An error return from poll() indicates:

[]
fds points outside the process's allocated address space.
[]
A signal was delivered before the time limit expired and before any of the selected events occurred.
[]
The specified time limit is negative or the number of pollfd structures specified is larger than the current file descriptor resource limit.

accept(2), connect(2), read(2), recv(2), select(2), send(2), write(2)

The poll() function appeared in AT&T System V Release 3 UNIX, and was added to NetBSD in NetBSD 1.3. The pollts() function first appeared in NetBSD 3.0. The ppoll() function first appeared in NetBSD 10.0.

As of this writing, in the underlying implementation, POLLIN and POLLPRI are distinct flags from POLLRDNORM and POLLRDBAND (respectively) and the behavior is not always exactly identical.

The detailed behavior of specific flags is not very portable from one OS to another.

February 8, 2021 NetBSD 11.0