man2/select.2

   1 .\" This manpage is copyright (C) 1992 Drew Eckhardt,
   2 .\"                 copyright (C) 1995 Michael Shields.
   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  15 .\" manual page may be incorrect or out-of-date.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 .\" Modified 1993-07-24 by Rik Faith <faith@cs.unc.edu>
  27 .\" Modified 1995-05-18 by Jim Van Zandt <jrv@vanzandt.mv.com>
  28 .\" Sun Feb 11 14:07:00 MET 1996  Martin Schulze  <joey@linux.de>
  29 .\"     * layout slightly modified
  30 .\"
  31 .\" Modified Mon Oct 21 23:05:29 EDT 1996 by Eric S. Raymond <esr@thyrsus.com>
  32 .\" Modified Thu Feb 24 01:41:09 CET 2000 by aeb
  33 .\" Modified Thu Feb  9 22:32:09 CET 2001 by bert hubert <ahu@ds9a.nl>, aeb
  34 .\" Modified Mon Nov 11 14:35:00 PST 2002 by Ben Woodard <ben@zork.net>
  35 .\" 2005-03-11, mtk, modified pselect() text (it is now a system
  36 .\"     call in 2.6.16.
  37 .\"
  38 .TH SELECT 2 2016-03-15 "Linux" "Linux Programmer's Manual"
  39 .SH NAME
  40 select, pselect, FD_CLR, FD_ISSET, FD_SET, FD_ZERO \-
  41 synchronous I/O multiplexing
  42 .SH SYNOPSIS
  43 .nf
  44 /* According to POSIX.1-2001, POSIX.1-2008 */
  45 .br
  46 .B #include <sys/select.h>
  47 .sp
  48 /* According to earlier standards */
  49 .br
  50 .B #include <sys/time.h>
  51 .br
  52 .B #include <sys/types.h>
  53 .br
  54 .B #include <unistd.h>
  55 .sp
  56 .BI "int select(int " nfds ", fd_set *" readfds ", fd_set *" writefds ,
  57 .BI "           fd_set *" exceptfds ", struct timeval *" timeout );
  58 .sp
  59 .BI "void FD_CLR(int " fd ", fd_set *" set );
  60 .br
  61 .BI "int  FD_ISSET(int " fd ", fd_set *" set );
  62 .br
  63 .BI "void FD_SET(int " fd ", fd_set *" set );
  64 .br
  65 .BI "void FD_ZERO(fd_set *" set );
  66 .sp
  67 .B #include <sys/select.h>
  68 .sp
  69 .BI "int pselect(int " nfds ", fd_set *" readfds ", fd_set *" writefds ,
  70 .BI "            fd_set *" exceptfds ", const struct timespec *" timeout ,
  71 .BI "            const sigset_t *" sigmask );
  72 .fi
  73 .sp
  74 .in -4n
  75 Feature Test Macro Requirements for glibc (see
  76 .BR feature_test_macros (7)):
  77 .in
  78 .sp
  79 .BR pselect ():
  80 _POSIX_C_SOURCE\ >=\ 200112L
  81 .SH DESCRIPTION
  82 .BR select ()
  83 and
  84 .BR pselect ()
  85 allow a program to monitor multiple file descriptors,
  86 waiting until one or more of the file descriptors become "ready"
  87 for some class of I/O operation (e.g., input possible).
  88 A file descriptor is considered ready if it is possible to
  89 perform a corresponding I/O operation (e.g.,
  90 .BR read (2)
  91 without blocking, or a sufficiently small
  92 .BR write (2)).
  93 .PP
  94 .BR select ()
  95 can monitor only file descriptors numbers that are less than
  96 .BR FD_SETSIZE ;
  97 .BR poll (2)
  98 does not have this limitation.
  99 See BUGS.
 100 .PP
 101 The operation of
 102 .BR select ()
 103 and
 104 .BR pselect ()
 105 is identical, other than these three differences:
 106 .TP
 107 (i)
 108 .BR select ()
 109 uses a timeout that is a
 110 .I struct timeval
 111 (with seconds and microseconds), while
 112 .BR pselect ()
 113 uses a
 114 .I struct timespec
 115 (with seconds and nanoseconds).
 116 .TP
 117 (ii)
 118 .BR select ()
 119 may update the
 120 .I timeout
 121 argument to indicate how much time was left.
 122 .BR pselect ()
 123 does not change this argument.
 124 .TP
 125 (iii)
 126 .BR select ()
 127 has no
 128 .I sigmask
 129 argument, and behaves as
 130 .BR pselect ()
 131 called with NULL
 132 .IR sigmask .
 133 .PP
 134 Three independent sets of file descriptors are watched.
 135 Those listed in
 136 .I readfds
 137 will be watched to see if characters become
 138 available for reading (more precisely, to see if a read will not
 139 block; in particular, a file descriptor is also ready on end-of-file),
 140 those in
 141 .I writefds
 142 will be watched to see if space is available for write (though a large
 143 write may still block), and those in
 144 .I exceptfds
 145 will be watched for exceptions.
 146 On exit, the sets are modified in place
 147 to indicate which file descriptors actually changed status.
 148 Each of the three file descriptor sets may be specified as NULL
 149 if no file descriptors are to be watched for the corresponding class
 150 of events.
 151 .PP
 152 Four macros are provided to manipulate the sets.
 153 .BR FD_ZERO ()
 154 clears a set.
 155 .BR FD_SET ()
 156 and
 157 .BR FD_CLR ()
 158 respectively add and remove a given file descriptor from a set.
 159 .BR FD_ISSET ()
 160 tests to see if a file descriptor is part of the set;
 161 this is useful after
 162 .BR select ()
 163 returns.
 164 .PP
 165 .I nfds
 166 is the highest-numbered file descriptor in any of the three sets, plus 1.
 167 .PP
 168 The
 169 .I timeout
 170 argument specifies the interval that
 171 .BR select ()
 172 should block waiting for a file descriptor to become ready.
 173 The call will block until either:
 174 .IP * 3
 175 a file descriptor becomes ready;
 176 .IP *
 177 the call is interrupted by a signal handler; or
 178 .IP *
 179 the timeout expires.
 180 .PP
 181 Note that the
 182 .I timeout
 183 interval will be rounded up to the system clock granularity,
 184 and kernel scheduling delays mean that the blocking interval
 185 may overrun by a small amount.
 186 If both fields of the
 187 .I timeval
 188 structure are zero, then
 189 .BR select ()
 190 returns immediately.
 191 (This is useful for polling.)
 192 If
 193 .I timeout
 194 is NULL (no timeout),
 195 .BR select ()
 196 can block indefinitely.
 197 .PP
 198 .I sigmask
 199 is a pointer to a signal mask (see
 200 .BR sigprocmask (2));
 201 if it is not NULL, then
 202 .BR pselect ()
 203 first replaces the current signal mask by the one pointed to by
 204 .IR sigmask ,
 205 then does the "select" function, and then restores the original
 206 signal mask.
 207 .PP
 208 Other than the difference in the precision of the
 209 .I timeout
 210 argument, the following
 211 .BR pselect ()
 212 call:
 213 .nf
 214
 215     ready = pselect(nfds, &readfds, &writefds, &exceptfds,
 216                     timeout, &sigmask);
 217
 218 .fi
 219 is equivalent to
 220 .I atomically
 221 executing the following calls:
 222 .nf
 223
 224     sigset_t origmask;
 225
 226     pthread_sigmask(SIG_SETMASK, &sigmask, &origmask);
 227     ready = select(nfds, &readfds, &writefds, &exceptfds, timeout);
 228     pthread_sigmask(SIG_SETMASK, &origmask, NULL);
 229 .fi
 230 .PP
 231 The reason that
 232 .BR pselect ()
 233 is needed is that if one wants to wait for either a signal
 234 or for a file descriptor to become ready, then
 235 an atomic test is needed to prevent race conditions.
 236 (Suppose the signal handler sets a global flag and
 237 returns.
 238 Then a test of this global flag followed by a call of
 239 .BR select ()
 240 could hang indefinitely if the signal arrived just after the test
 241 but just before the call.
 242 By contrast,
 243 .BR pselect ()
 244 allows one to first block signals, handle the signals that have come in,
 245 then call
 246 .BR pselect ()
 247 with the desired
 248 .IR sigmask ,
 249 avoiding the race.)
 250 .SS The timeout
 251 The time structures involved are defined in
 252 .I <sys/time.h>
 253 and look like
 254
 255 .in +4n
 256 .nf
 257 struct timeval {
 258     long    tv_sec;         /* seconds */
 259     long    tv_usec;        /* microseconds */
 260 };
 261 .fi
 262 .in
 263
 264 and
 265
 266 .in +4n
 267 .nf
 268 struct timespec {
 269     long    tv_sec;         /* seconds */
 270     long    tv_nsec;        /* nanoseconds */
 271 };
 272 .fi
 273 .in
 274
 275 (However, see below on the POSIX.1 versions.)
 276 .PP
 277 Some code calls
 278 .BR select ()
 279 with all three sets empty,
 280 .I nfds
 281 zero, and a non-NULL
 282 .I timeout
 283 as a fairly portable way to sleep with subsecond precision.
 284 .PP
 285 On Linux,
 286 .BR select ()
 287 modifies
 288 .I timeout
 289 to reflect the amount of time not slept; most other implementations
 290 do not do this.
 291 (POSIX.1 permits either behavior.)
 292 This causes problems both when Linux code which reads
 293 .I timeout
 294 is ported to other operating systems, and when code is ported to Linux
 295 that reuses a \fIstruct timeval\fP for multiple
 296 .BR select ()s
 297 in a loop without reinitializing it.
 298 Consider
 299 .I timeout
 300 to be undefined after
 301 .BR select ()
 302 returns.
 303 .\" .PP - it is rumored that:
 304 .\" On BSD, when a timeout occurs, the file descriptor bits are not changed.
 305 .\" - it is certainly true that:
 306 .\" Linux follows SUSv2 and sets the bit masks to zero upon a timeout.
 307 .SH RETURN VALUE
 308 On success,
 309 .BR select ()
 310 and
 311 .BR pselect ()
 312 return the number of file descriptors contained in the three returned
 313 descriptor sets (that is, the total number of bits that are set in
 314 .IR readfds ,
 315 .IR writefds ,
 316 .IR exceptfds )
 317 which may be zero if the timeout expires before anything interesting happens.
 318 On error, \-1 is returned, and
 319 .I errno
 320 is set to indicate the error;
 321 the file descriptor sets are unmodified,
 322 and
 323 .I timeout
 324 becomes undefined.
 325 .SH ERRORS
 326 .TP
 327 .B EBADF
 328 An invalid file descriptor was given in one of the sets.
 329 (Perhaps a file descriptor that was already closed,
 330 or one on which an error has occurred.)
 331 .TP
 332 .B EINTR
 333 A signal was caught; see
 334 .BR signal (7).
 335 .TP
 336 .B EINVAL
 337 .I nfds
 338 is negative or exceeds the
 339 .BR RLIMIT_NOFILE
 340 resource limit (see
 341 .BR getrlimit (2)).
 342 .TP
 343 .B EINVAL
 344 The value contained within
 345 .I timeout
 346 is invalid.
 347 .TP
 348 .B ENOMEM
 349 Unable to allocate memory for internal tables.
 350 .SH VERSIONS
 351 .BR pselect ()
 352 was added to Linux in kernel 2.6.16.
 353 Prior to this,
 354 .BR pselect ()
 355 was emulated in glibc (but see BUGS).
 356 .SH CONFORMING TO
 357 .BR select ()
 358 conforms to POSIX.1-2001, POSIX.1-2008, and
 359 4.4BSD
 360 .RB ( select ()
 361 first appeared in 4.2BSD).
 362 Generally portable to/from
 363 non-BSD systems supporting clones of the BSD socket layer (including
 364 System\ V variants).
 365 However, note that the System\ V variant typically
 366 sets the timeout variable before exit, but the BSD variant does not.
 367 .PP
 368 .BR pselect ()
 369 is defined in POSIX.1g, and in
 370 POSIX.1-2001 and POSIX.1-2008.
 371 .SH NOTES
 372 An
 373 .I fd_set
 374 is a fixed size buffer.
 375 Executing
 376 .BR FD_CLR ()
 377 or
 378 .BR FD_SET ()
 379 with a value of
 380 .I fd
 381 that is negative or is equal to or larger than
 382 .B FD_SETSIZE
 383 will result
 384 in undefined behavior.
 385 Moreover, POSIX requires
 386 .I fd
 387 to be a valid file descriptor.
 388
 389 On some other UNIX systems,
 390 .\" Darwin, according to a report by Jeremy Sequoia, relayed by Josh Triplett
 391 .BR select ()
 392 can fail with the error
 393 .B EAGAIN
 394 if the system fails to allocate kernel-internal resources, rather than
 395 .B ENOMEM
 396 as Linux does.
 397 POSIX specifies this error for
 398 .BR poll (2),
 399 but not for
 400 .BR select ().
 401 Portable programs may wish to check for
 402 .B EAGAIN
 403 and loop, just as with
 404 .BR EINTR .
 405
 406 Concerning the types involved, the classical situation is that
 407 the two fields of a
 408 .I timeval
 409 structure are typed as
 410 .I long
 411 (as shown above), and the structure is defined in
 412 .IR <sys/time.h> .
 413 The POSIX.1 situation is
 414
 415 .in +4n
 416 .nf
 417 struct timeval {
 418     time_t         tv_sec;     /* seconds */
 419     suseconds_t    tv_usec;    /* microseconds */
 420 };
 421 .fi
 422 .in
 423
 424 where the structure is defined in
 425 .I <sys/select.h>
 426 and the data types
 427 .I time_t
 428 and
 429 .I suseconds_t
 430 are defined in
 431 .IR <sys/types.h> .
 432 .LP
 433 Concerning prototypes, the classical situation is that one should
 434 include
 435 .I <time.h>
 436 for
 437 .BR select ().
 438 The POSIX.1 situation is that one should include
 439 .I <sys/select.h>
 440 for
 441 .BR select ()
 442 and
 443 .BR pselect ().
 444
 445 Under glibc 2.0,
 446 .I <sys/select.h>
 447 gives the wrong prototype for
 448 .BR pselect ().
 449 Under glibc 2.1 to 2.2.1, it gives
 450 .BR pselect ()
 451 when
 452 .B _GNU_SOURCE
 453 is defined.
 454 Since glibc 2.2.2, the requirements are as shown in the SYNOPSIS.
 455 .SS Multithreaded applications
 456 If a file descriptor being monitored by
 457 .BR select ()
 458 is closed in another thread, the result is unspecified.
 459 On some UNIX systems,
 460 .BR select ()
 461 unblocks and returns, with an indication that the file descriptor is ready
 462 (a subsequent I/O operation will likely fail with an error,
 463 unless another the file descriptor reopened between the time
 464 .BR select ()
 465 returned and the I/O operations was performed).
 466 On Linux (and some other systems),
 467 closing the file descriptor in another thread has no effect on
 468 .BR select ().
 469 In summary, any application that relies on a particular behavior
 470 in this scenario must be considered buggy.
 471 .\"
 472 .SS C library/kernel differences
 473 The Linux kernel allows file descriptor sets of arbitrary size,
 474 determining the length of the sets to be checked from the value of
 475 .IR nfds .
 476 However, in the glibc implementation, the
 477 .IR fd_set
 478 type is fixed in size.
 479 See also BUGS.
 480
 481 The
 482 .BR pselect ()
 483 interface described in this page is implemented by glibc.
 484 The underlying Linux system call is named
 485 .BR pselect6 ().
 486 This system call has somewhat different behavior from the glibc
 487 wrapper function.
 488
 489 The Linux
 490 .BR pselect6 ()
 491 system call modifies its
 492 .I timeout
 493 argument.
 494 However, the glibc wrapper function hides this behavior
 495 by using a local variable for the timeout argument that
 496 is passed to the system call.
 497 Thus, the glibc
 498 .BR pselect ()
 499 function does not modify its
 500 .I timeout
 501 argument;
 502 this is the behavior required by POSIX.1-2001.
 503
 504 The final argument of the
 505 .BR pselect6 ()
 506 system call is not a
 507 .I "sigset_t\ *"
 508 pointer, but is instead a structure of the form:
 509 .in +4
 510 .nf
 511
 512 struct {
 513     const sigset_t *ss;     /* Pointer to signal set */
 514     size_t          ss_len; /* Size (in bytes) of object pointed
 515                                to by 'ss' */
 516 };
 517
 518 .fi
 519 .in
 520 This allows the system call to obtain both
 521 a pointer to the signal set and its size,
 522 while allowing for the fact that most architectures
 523 support a maximum of 6 arguments to a system call.
 524 .SH BUGS
 525 POSIX allows an implementation to define an upper limit,
 526 advertised via the constant
 527 .BR FD_SETSIZE ,
 528 on the range of file descriptors that can be specified
 529 in a file descriptor set.
 530 The Linux kernel imposes no fixed limit, but the glibc implementation makes
 531 .IR fd_set
 532 a fixed-size type, with
 533 .BR FD_SETSIZE
 534 defined as 1024, and the
 535 .BR FD_* ()
 536 macros operating according to that limit.
 537 To monitor file descriptors greater than 1023, use
 538 .BR poll (2)
 539 instead.
 540
 541 Glibc 2.0 provided a version of
 542 .BR pselect ()
 543 that did not take a
 544 .I sigmask
 545 argument.
 546
 547 Starting with version 2.1, glibc provided an emulation of
 548 .BR pselect ()
 549 that was implemented using
 550 .BR sigprocmask (2)
 551 and
 552 .BR select ().
 553 This implementation remained vulnerable to the very race condition that
 554 .BR pselect ()
 555 was designed to prevent.
 556 Modern versions of glibc use the (race-free)
 557 .BR pselect ()
 558 system call on kernels where it is provided.
 559
 560 On systems that lack
 561 .BR pselect (),
 562 reliable (and more portable) signal trapping can be achieved
 563 using the self-pipe trick.
 564 In this technique,
 565 a signal handler writes a byte to a pipe whose other end
 566 is monitored by
 567 .BR select ()
 568 in the main program.
 569 (To avoid possibly blocking when writing to a pipe that may be full
 570 or reading from a pipe that may be empty,
 571 nonblocking I/O is used when reading from and writing to the pipe.)
 572
 573 Under Linux,
 574 .BR select ()
 575 may report a socket file descriptor as "ready for reading", while
 576 nevertheless a subsequent read blocks.
 577 This could for example
 578 happen when data has arrived but upon examination has wrong
 579 checksum and is discarded.
 580 There may be other circumstances
 581 in which a file descriptor is spuriously reported as ready.
 582 .\" Stevens discusses a case where accept can block after select
 583 .\" returns successfully because of an intervening RST from the client.
 584 Thus it may be safer to use
 585 .B O_NONBLOCK
 586 on sockets that should not block.
 587 .\" Maybe the kernel should have returned EIO in such a situation?
 588
 589 On Linux,
 590 .BR select ()
 591 also modifies
 592 .I timeout
 593 if the call is interrupted by a signal handler (i.e., the
 594 .B EINTR
 595 error return).
 596 This is not permitted by POSIX.1.
 597 The Linux
 598 .BR pselect ()
 599 system call has the same behavior,
 600 but the glibc wrapper hides this behavior by internally copying the
 601 .I timeout
 602 to a local variable and passing that variable to the system call.
 603 .SH EXAMPLE
 604 .nf
 605 #include <stdio.h>
 606 #include <stdlib.h>
 607 #include <sys/time.h>
 608 #include <sys/types.h>
 609 #include <unistd.h>
 610
 611 int
 612 main(void)
 613 {
 614     fd_set rfds;
 615     struct timeval tv;
 616     int retval;
 617
 618     /* Watch stdin (fd 0) to see when it has input. */
 619
 620     FD_ZERO(&rfds);
 621     FD_SET(0, &rfds);
 622
 623     /* Wait up to five seconds. */
 624
 625     tv.tv_sec = 5;
 626     tv.tv_usec = 0;
 627
 628     retval = select(1, &rfds, NULL, NULL, &tv);
 629     /* Don't rely on the value of tv now! */
 630
 631     if (retval == \-1)
 632         perror("select()");
 633     else if (retval)
 634         printf("Data is available now.\\n");
 635         /* FD_ISSET(0, &rfds) will be true. */
 636     else
 637         printf("No data within five seconds.\\n");
 638
 639     exit(EXIT_SUCCESS);
 640 }
 641 .fi
 642 .SH SEE ALSO
 643 .BR accept (2),
 644 .BR connect (2),
 645 .BR poll (2),
 646 .BR read (2),
 647 .BR recv (2),
 648 .BR restart_syscall (2),
 649 .BR send (2),
 650 .BR sigprocmask (2),
 651 .BR write (2),
 652 .BR epoll (7),
 653 .BR time (7)
 654
 655 For a tutorial with discussion and examples, see
 656 .BR select_tut (2).