OpenBSD manual page server

Manual Page Search Parameters
POLL(2) System Calls Manual POLL(2)

poll, ppoll —
synchronous I/O multiplexing

#include <poll.h>

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

int
ppoll(struct pollfd *fds, nfds_t nfds, const struct timespec *timeout, const sigset_t *mask);

poll() provides a mechanism for multiplexing I/O across a set of file descriptors. It is similar in function to select(2). Unlike select(2), however, it is possible to only pass in data corresponding to the file descriptors for which events are wanted. This makes poll() more efficient than select(2) in most cases.

The arguments are as follows:

fds
Points to an array of pollfd structures, which are defined as: 
struct pollfd {
	int fd;
	short events;
	short revents;
};

The fd member is an open file descriptor. If fd is -1, the pollfd structure is considered unused, and revents will be cleared.

The events and revents members are bitmasks of conditions to monitor and conditions found, respectively.

nfds
An unsigned integer specifying the number of pollfd structures in the array.
timeout
Maximum interval to wait for the poll to complete, in milliseconds. If this value is 0, poll() will return immediately. If this value is INFTIM (-1), poll() will block indefinitely until a condition is found.

The calling process sets the events bitmask and poll() sets the revents bitmask. Each call to poll() resets the revents bitmask for accuracy. The condition flags in the bitmasks are defined as:

Data other than high-priority data may be read without blocking.
Normal data may be read without blocking.
Priority data may be read without blocking.
Same as POLLRDNORM. This flag is provided for source code compatibility with older programs and should not be used in new code.
High-priority data may be read without blocking.
Normal data may be written without blocking.
Same as POLLOUT.
Priority data may be written.
An error has occurred on the device or socket. This flag is only valid in the revents bitmask; it is ignored in the events member.
The device or socket has been disconnected. This event and POLLOUT are mutually-exclusive; a descriptor can never be writable if a hangup has occurred. However, this event and POLLIN, POLLRDNORM, POLLRDBAND, or POLLPRI are not mutually-exclusive. This flag is only valid in the revents bitmask; it is ignored in the events member.
The corresponding file descriptor is invalid. This flag is only valid in the revents bitmask; it is ignored in the events member.

The significance and semantics of normal, priority, and high-priority data are device-specific. For example, on OpenBSD, the POLLPRI and POLLRDBAND flags may be used to detect when out-of-band socket data may be read without blocking.

The ppoll() function is similar to poll() except that it specifies the timeout using a timespec structure, and a null pointer is used to specify an indefinite timeout instead of INFTIM. Also, if mask is a non-null pointer, ppoll() atomically sets the calling thread's signal mask to the signal set pointed to by mask for the duration of the function call. In this case, the original signal mask will be restored before ppoll() returns.

Upon error, poll() and ppoll() return -1 and set the global variable errno to indicate the error. If the timeout interval was reached before any events occurred, they return 0. Otherwise, they return the number of pollfd structures for which revents is non-zero.

Care must be taken when converting code from select(2) to poll() as they have slightly different semantics. The first semantic difference is that, unlike select(2), poll() has a way of indicating that one or more file descriptors is invalid by setting a flag in the revents field of corresponding entry of fds, whereas select(2) returns an error (-1) if any of the descriptors with bits set in the fd_set are invalid. The second difference is that on EOF there is no guarantee that POLLIN will be set in revents, the caller must also check for POLLHUP. This differs from select(2) where EOF is considered as a read event.

Consider the following usage of select(2) that implements a read from the standard input with a 60 second time out:

struct timeval timeout;
fd_set readfds;
char buf[BUFSIZ];
int nready;

timeout.tv_sec = 60;
timeout.tv_usec = 0;
FD_ZERO(&readfds);
FD_SET(STDIN_FILENO, &readfds);
nready = select(STDIN_FILENO + 1, &readfds, NULL, NULL, &timeout);
if (nready == -1)
	err(1, "select");
if (nready == 0)
	errx(1, "time out");
if (FD_ISSET(STDIN_FILENO, &readfds)) {
	if (read(STDIN_FILENO, buf, sizeof(buf)) == -1)
		err(1, "read");
}

This can be converted to poll() as follows:

struct pollfd pfd[1];
char buf[BUFSIZ];
int nready;

pfd[0].fd = STDIN_FILENO;
pfd[0].events = POLLIN;
nready = poll(pfd, 1, 60 * 1000);
if (nready == -1)
	err(1, "poll");
if (nready == 0)
	errx(1, "time out");
if ((pfd[0].revents & (POLLERR|POLLNVAL)))
	errx(1, "bad fd %d", pfd[0].fd);
if ((pfd[0].revents & (POLLIN|POLLHUP))) {
	if (read(STDIN_FILENO, buf, sizeof(buf)) == -1)
		err(1, "read");
}

poll() and ppoll() will fail if:
[]
fds points outside the process's allocated address space.
[]
A signal was caught before any polled events occurred and before the timeout elapsed.
[]
nfds was greater than the number of available file descriptors.
[]
The timeout passed was invalid.

clock_gettime(2), getrlimit(2), read(2), select(2), write(2)

The poll() function is compliant with the IEEE Std 1003.1-2008 (“POSIX.1”) specification. The ppoll() function is a Linux extension.

A poll() system call appeared in AT&T System V Release 3 UNIX. The ppoll() function appeared in OpenBSD 5.4.

The POLLWRBAND flag is accepted but ignored by the kernel.

Because OpenBSD does not implement STREAMS, there is no distinction between some of the fields in the events and revents bitmasks. As a result, the POLLIN, POLLNORM, and POLLRDNORM flags are equivalent. Similarly, the POLLPRI and POLLRDBAND flags are also equivalent.

Internally to the kernel, poll() and ppoll() work poorly if multiple processes wait on the same file descriptor.

August 18, 2018 OpenBSD-current