man2/poll.2

   1 .\" Copyright (C) 1997 Andries Brouwer (aeb@cwi.nl)
   2 .\" and Copyright (C) 2006, Michael Kerrisk <mtk.manpages@gmail.com>
   3 .\"
   4 .\" %%%LICENSE_START(VERBATIM)
   5 .\" Permission is granted to make and distribute verbatim copies of this
   6 .\" manual provided the copyright notice and this permission notice are
   7 .\" preserved on all copies.
   8 .\"
   9 .\" Permission is granted to copy and distribute modified versions of this
  10 .\" manual under the conditions for verbatim copying, provided that the
  11 .\" entire resulting derived work is distributed under the terms of a
  12 .\" permission notice identical to this one.
  13 .\"
  14 .\" Since the Linux kernel and libraries are constantly changing, this
  15 .\" manual page may be incorrect or out-of-date.  The author(s) assume no
  16 .\" responsibility for errors or omissions, or for damages resulting from
  17 .\" the use of the information contained herein.  The author(s) may not
  18 .\" have taken the same level of care in the production of this manual,
  19 .\" which is licensed free of charge, as they might when working
  20 .\" professionally.
  21 .\"
  22 .\" Formatted or processed versions of this manual, if unaccompanied by
  23 .\" the source, must acknowledge the copyright and authors of this work.
  24 .\" %%%LICENSE_END
  25 .\"
  26 .\" Additions from Richard Gooch <rgooch@atnf.CSIRO.AU> and aeb, 971207
  27 .\" 2006-03-13, mtk, Added ppoll() + various other rewordings
  28 .\" 2006-07-01, mtk, Added POLLRDHUP + various other wording and
  29 .\"     formatting changes.
  30 .\"
  31 .TH POLL 2 2019-08-02 "Linux" "Linux Programmer's Manual"
  32 .SH NAME
  33 poll, ppoll \- wait for some event on a file descriptor
  34 .SH SYNOPSIS
  35 .nf
  36 .B #include <poll.h>
  37 .PP
  38 .BI "int poll(struct pollfd *" fds ", nfds_t " nfds ", int " timeout );
  39
  40 .BR "#define _GNU_SOURCE" "         /* See feature_test_macros(7) */"
  41 .B #include <signal.h>
  42 .B #include <poll.h>
  43 .PP
  44 .BI "int ppoll(struct pollfd *" fds ", nfds_t " nfds ", "
  45 .BI "        const struct timespec *" tmo_p ", const sigset_t *" sigmask );
  46 .fi
  47 .SH DESCRIPTION
  48 .BR poll ()
  49 performs a similar task to
  50 .BR select (2):
  51 it waits for one of a set of file descriptors to become ready
  52 to perform I/O.
  53 .PP
  54 The set of file descriptors to be monitored is specified in the
  55 .I fds
  56 argument, which is an array of structures of the following form:
  57 .PP
  58 .in +4n
  59 .EX
  60 struct pollfd {
  61     int   fd;         /* file descriptor */
  62     short events;     /* requested events */
  63     short revents;    /* returned events */
  64 };
  65 .EE
  66 .in
  67 .PP
  68 The caller should specify the number of items in the
  69 .I fds
  70 array in
  71 .IR nfds .
  72 .PP
  73 The field
  74 .I fd
  75 contains a file descriptor for an open file.
  76 If this field is negative, then the corresponding
  77 .I events
  78 field is ignored and the
  79 .I revents
  80 field returns zero.
  81 (This provides an easy way of ignoring a
  82 file descriptor for a single
  83 .BR poll ()
  84 call: simply negate the
  85 .I fd
  86 field.
  87 Note, however, that this technique can't be used to ignore file descriptor 0.)
  88 .PP
  89 The field
  90 .I events
  91 is an input parameter, a bit mask specifying the events the application
  92 is interested in for the file descriptor
  93 .IR fd .
  94 This field may be specified as zero,
  95 in which case the only events that can be returned in
  96 .I revents
  97 are
  98 .BR POLLHUP ,
  99 .BR POLLERR ,
 100 and
 101 .B POLLNVAL
 102 (see below).
 103 .PP
 104 The field
 105 .I revents
 106 is an output parameter, filled by the kernel with the events that
 107 actually occurred.
 108 The bits returned in
 109 .I revents
 110 can include any of those specified in
 111 .IR events ,
 112 or one of the values
 113 .BR POLLERR ,
 114 .BR POLLHUP ,
 115 or
 116 .BR POLLNVAL .
 117 (These three bits are meaningless in the
 118 .I events
 119 field, and will be set in the
 120 .I revents
 121 field whenever the corresponding condition is true.)
 122 .PP
 123 If none of the events requested (and no error) has occurred for any
 124 of the file descriptors, then
 125 .BR poll ()
 126 blocks until one of the events occurs.
 127 .PP
 128 The
 129 .I timeout
 130 argument specifies the number of milliseconds that
 131 .BR poll ()
 132 should block waiting for a file descriptor to become ready.
 133 The call will block until either:
 134 .IP * 3
 135 a file descriptor becomes ready;
 136 .IP *
 137 the call is interrupted by a signal handler; or
 138 .IP *
 139 the timeout expires.
 140 .PP
 141 Note that the
 142 .I timeout
 143 interval will be rounded up to the system clock granularity,
 144 and kernel scheduling delays mean that the blocking interval
 145 may overrun by a small amount.
 146 Specifying a negative value in
 147 .I timeout
 148 means an infinite timeout.
 149 Specifying a
 150 .I timeout
 151 of zero causes
 152 .BR poll ()
 153 to return immediately, even if no file descriptors are ready.
 154 .PP
 155 The bits that may be set/returned in
 156 .I events
 157 and
 158 .I revents
 159 are defined in \fI<poll.h>\fP:
 160 .TP
 161 .B POLLIN
 162 There is data to read.
 163 .TP
 164 .B POLLPRI
 165 There is some exceptional condition on the file descriptor.
 166 Possibilities include:
 167 .RS
 168 .IP * 3
 169 There is out-of-band data on a TCP socket (see
 170 .BR tcp (7)).
 171 .IP *
 172 A pseudoterminal master in packet mode has seen a state change on the slave
 173 (see
 174 .BR ioctl_tty (2)).
 175 .IP *
 176 A
 177 .I cgroup.events
 178 file has been modified (see
 179 .BR cgroups (7)).
 180 .RE
 181 .TP
 182 .B POLLOUT
 183 Writing is now possible, though a write larger that the available space
 184 in a socket or pipe will still block (unless
 185 .B O_NONBLOCK
 186 is set).
 187 .TP
 188 .BR POLLRDHUP " (since Linux 2.6.17)"
 189 Stream socket peer closed connection,
 190 or shut down writing half of connection.
 191 The
 192 .B _GNU_SOURCE
 193 feature test macro must be defined
 194 (before including
 195 .I any
 196 header files)
 197 in order to obtain this definition.
 198 .TP
 199 .B POLLERR
 200 Error condition (only returned in
 201 .IR revents ;
 202 ignored in
 203 .IR events ).
 204 This bit is also set for a file descriptor referring
 205 to the write end of a pipe when the read end has been closed.
 206 .TP
 207 .B POLLHUP
 208 Hang up (only returned in
 209 .IR revents ;
 210 ignored in
 211 .IR events ).
 212 Note that when reading from a channel such as a pipe or a stream socket,
 213 this event merely indicates that the peer closed its end of the channel.
 214 Subsequent reads from the channel will return 0 (end of file)
 215 only after all outstanding data in the channel has been consumed.
 216 .TP
 217 .B POLLNVAL
 218 Invalid request:
 219 .I fd
 220 not open (only returned in
 221 .IR revents ;
 222 ignored in
 223 .IR events ).
 224 .PP
 225 When compiling with
 226 .B _XOPEN_SOURCE
 227 defined, one also has the following,
 228 which convey no further information beyond the bits listed above:
 229 .TP
 230 .B POLLRDNORM
 231 Equivalent to
 232 .BR POLLIN .
 233 .TP
 234 .B POLLRDBAND
 235 Priority band data can be read (generally unused on Linux).
 236 .\" POLLRDBAND is used in the DECnet protocol.
 237 .TP
 238 .B POLLWRNORM
 239 Equivalent to
 240 .BR POLLOUT .
 241 .TP
 242 .B POLLWRBAND
 243 Priority data may be written.
 244 .PP
 245 Linux also knows about, but does not use
 246 .BR POLLMSG .
 247 .SS ppoll()
 248 The relationship between
 249 .BR poll ()
 250 and
 251 .BR ppoll ()
 252 is analogous to the relationship between
 253 .BR select (2)
 254 and
 255 .BR pselect (2):
 256 like
 257 .BR pselect (2),
 258 .BR ppoll ()
 259 allows an application to safely wait until either a file descriptor
 260 becomes ready or until a signal is caught.
 261 .PP
 262 Other than the difference in the precision of the
 263 .I timeout
 264 argument, the following
 265 .BR ppoll ()
 266 call:
 267 .PP
 268 .in +4n
 269 .EX
 270 ready = ppoll(&fds, nfds, tmo_p, &sigmask);
 271 .EE
 272 .in
 273 .PP
 274 is nearly equivalent to
 275 .I atomically
 276 executing the following calls:
 277 .PP
 278 .in +4n
 279 .EX
 280 sigset_t origmask;
 281 int timeout;
 282
 283 timeout = (tmo_p == NULL) ? \-1 :
 284           (tmo_p\->tv_sec * 1000 + tmo_p\->tv_nsec / 1000000);
 285 pthread_sigmask(SIG_SETMASK, &sigmask, &origmask);
 286 ready = poll(&fds, nfds, timeout);
 287 pthread_sigmask(SIG_SETMASK, &origmask, NULL);
 288 .EE
 289 .in
 290 .PP
 291 The above code segment is described as
 292 .I nearly
 293 equivalent because whereas a negative
 294 .I timeout
 295 value for
 296 .BR poll ()
 297 is interpreted as an infinite timeout, a negative value expressed in
 298 .IR *tmo_p
 299 results in an error from
 300 .BR ppoll ().
 301 .PP
 302 See the description of
 303 .BR pselect (2)
 304 for an explanation of why
 305 .BR ppoll ()
 306 is necessary.
 307 .PP
 308 If the
 309 .I sigmask
 310 argument is specified as NULL, then
 311 no signal mask manipulation is performed
 312 (and thus
 313 .BR ppoll ()
 314 differs from
 315 .BR poll ()
 316 only in the precision of the
 317 .I timeout
 318 argument).
 319 .PP
 320 The
 321 .I tmo_p
 322 argument specifies an upper limit on the amount of time that
 323 .BR ppoll ()
 324 will block.
 325 This argument is a pointer to a structure of the following form:
 326 .PP
 327 .in +4n
 328 .EX
 329 struct timespec {
 330     long    tv_sec;         /* seconds */
 331     long    tv_nsec;        /* nanoseconds */
 332 };
 333 .EE
 334 .in
 335 .PP
 336 If
 337 .I tmo_p
 338 is specified as NULL, then
 339 .BR ppoll ()
 340 can block indefinitely.
 341 .SH RETURN VALUE
 342 On success, a positive number is returned; this is
 343 the number of structures which have nonzero
 344 .I revents
 345 fields (in other words, those descriptors with events or errors reported).
 346 A value of 0 indicates that the call timed out and no file
 347 descriptors were ready.
 348 On error, \-1 is returned, and
 349 .I errno
 350 is set appropriately.
 351 .SH ERRORS
 352 .TP
 353 .B EFAULT
 354 The array given as argument was not contained in the calling program's
 355 address space.
 356 .TP
 357 .B EINTR
 358 A signal occurred before any requested event; see
 359 .BR signal (7).
 360 .TP
 361 .B EINVAL
 362 The
 363 .I nfds
 364 value exceeds the
 365 .B RLIMIT_NOFILE
 366 value.
 367 .TP
 368 .B EINVAL
 369 .RB ( ppoll ())
 370 The timeout value expressed in
 371 .IR *ip
 372 is invalid (negative).
 373 .TP
 374 .B ENOMEM
 375 There was no space to allocate file descriptor tables.
 376 .SH VERSIONS
 377 The
 378 .BR poll ()
 379 system call was introduced in Linux 2.1.23.
 380 On older kernels that lack this system call,
 381 .\" library call was introduced in libc 5.4.28
 382 the glibc (and the old Linux libc)
 383 .BR poll ()
 384 wrapper function provides emulation using
 385 .BR select (2).
 386 .PP
 387 The
 388 .BR ppoll ()
 389 system call was added to Linux in kernel 2.6.16.
 390 The
 391 .BR ppoll ()
 392 library call was added in glibc 2.4.
 393 .SH CONFORMING TO
 394 .BR poll ()
 395 conforms to POSIX.1-2001 and POSIX.1-2008.
 396 .BR ppoll ()
 397 is Linux-specific.
 398 .\" NetBSD 3.0 has a pollts() which is like Linux ppoll().
 399 .SH NOTES
 400 The operation of
 401 .BR poll ()
 402 and
 403 .BR ppoll ()
 404 is not affected by the
 405 .BR O_NONBLOCK
 406 flag.
 407 .PP
 408 On some other UNIX systems,
 409 .\" Darwin, according to a report by Jeremy Sequoia, relayed by Josh Triplett
 410 .BR poll ()
 411 can fail with the error
 412 .B EAGAIN
 413 if the system fails to allocate kernel-internal resources, rather than
 414 .B ENOMEM
 415 as Linux does.
 416 POSIX permits this behavior.
 417 Portable programs may wish to check for
 418 .B EAGAIN
 419 and loop, just as with
 420 .BR EINTR .
 421 .PP
 422 Some implementations define the nonstandard constant
 423 .B INFTIM
 424 with the value \-1 for use as a
 425 .IR timeout
 426 for
 427 .BR poll ().
 428 This constant is not provided in glibc.
 429 .PP
 430 For a discussion of what may happen if a file descriptor being monitored by
 431 .BR poll ()
 432 is closed in another thread, see
 433 .BR select (2).
 434 .SS C library/kernel differences
 435 The Linux
 436 .BR ppoll ()
 437 system call modifies its
 438 .I tmo_p
 439 argument.
 440 However, the glibc wrapper function hides this behavior
 441 by using a local variable for the timeout argument that
 442 is passed to the system call.
 443 Thus, the glibc
 444 .BR ppoll ()
 445 function does not modify its
 446 .I tmo_p
 447 argument.
 448 .PP
 449 The raw
 450 .BR ppoll ()
 451 system call has a fifth argument,
 452 .IR "size_t sigsetsize" ,
 453 which specifies the size in bytes of the
 454 .IR sigmask
 455 argument.
 456 The glibc
 457 .BR ppoll ()
 458 wrapper function specifies this argument as a fixed value
 459 (equal to
 460 .IR sizeof(kernel_sigset_t) ).
 461 See
 462 .BR sigprocmask (2)
 463 for a discussion on the differences between the kernel and the libc
 464 notion of the sigset.
 465 .SH BUGS
 466 See the discussion of spurious readiness notifications under the
 467 BUGS section of
 468 .BR select (2).
 469 .SH SEE ALSO
 470 .BR restart_syscall (2),
 471 .BR select (2),
 472 .BR select_tut (2),
 473 .BR epoll (7),
 474 .BR time (7)