1 @node Low-Level I/O, File System Interface, I/O on Streams, Top
2 @c %MENU% Low-level, less portable I/O
3 @chapter Low-Level Input/Output
5 This chapter describes functions for performing low-level input/output
6 operations on file descriptors. These functions include the primitives
7 for the higher-level I/O functions described in @ref{I/O on Streams}, as
8 well as functions for performing low-level control operations for which
9 there are no equivalents on streams.
11 Stream-level I/O is more flexible and usually more convenient;
12 therefore, programmers generally use the descriptor-level functions only
13 when necessary. These are some of the usual reasons:
17 For reading binary files in large chunks.
20 For reading an entire file into core before parsing it.
23 To perform operations other than data transfer, which can only be done
24 with a descriptor. (You can use @code{fileno} to get the descriptor
25 corresponding to a stream.)
28 To pass descriptors to a child process. (The child can create its own
29 stream to use a descriptor that it inherits, but cannot inherit a stream
34 * Opening and Closing Files:: How to open and close file
36 * I/O Primitives:: Reading and writing data.
37 * File Position Primitive:: Setting a descriptor's file
39 * Descriptors and Streams:: Converting descriptor to stream
41 * Stream/Descriptor Precautions:: Precautions needed if you use both
42 descriptors and streams.
43 * Scatter-Gather:: Fast I/O to discontinuous buffers.
44 * Memory-mapped I/O:: Using files like memory.
45 * Waiting for I/O:: How to check for input or output
46 on multiple file descriptors.
47 * Synchronizing I/O:: Making sure all I/O actions completed.
48 * Asynchronous I/O:: Perform I/O in parallel.
49 * Control Operations:: Various other operations on file
51 * Duplicating Descriptors:: Fcntl commands for duplicating
53 * Descriptor Flags:: Fcntl commands for manipulating
54 flags associated with file
56 * File Status Flags:: Fcntl commands for manipulating
57 flags associated with open files.
58 * File Locks:: Fcntl commands for implementing
60 * Interrupt Input:: Getting an asynchronous signal when
62 * IOCTLs:: Generic I/O Control operations.
66 @node Opening and Closing Files
67 @section Opening and Closing Files
69 @cindex opening a file descriptor
70 @cindex closing a file descriptor
71 This section describes the primitives for opening and closing files
72 using file descriptors. The @code{open} and @code{creat} functions are
73 declared in the header file @file{fcntl.h}, while @code{close} is
74 declared in @file{unistd.h}.
80 @deftypefun int open (const char *@var{filename}, int @var{flags}[, mode_t @var{mode}])
81 The @code{open} function creates and returns a new file descriptor
82 for the file named by @var{filename}. Initially, the file position
83 indicator for the file is at the beginning of the file. The argument
84 @var{mode} is used only when a file is created, but it doesn't hurt
85 to supply the argument in any case.
87 The @var{flags} argument controls how the file is to be opened. This is
88 a bit mask; you create the value by the bitwise OR of the appropriate
89 parameters (using the @samp{|} operator in C).
90 @xref{File Status Flags}, for the parameters available.
92 The normal return value from @code{open} is a non-negative integer file
93 descriptor. In the case of an error, a value of @math{-1} is returned
94 instead. In addition to the usual file name errors (@pxref{File
95 Name Errors}), the following @code{errno} error conditions are defined
100 The file exists but is not readable/writeable as requested by the @var{flags}
101 argument, the file does not exist and the directory is unwriteable so
102 it cannot be created.
105 Both @code{O_CREAT} and @code{O_EXCL} are set, and the named file already
109 The @code{open} operation was interrupted by a signal.
110 @xref{Interrupted Primitives}.
113 The @var{flags} argument specified write access, and the file is a directory.
116 The process has too many files open.
117 The maximum number of file descriptors is controlled by the
118 @code{RLIMIT_NOFILE} resource limit; @pxref{Limits on Resources}.
121 The entire system, or perhaps the file system which contains the
122 directory, cannot support any additional open files at the moment.
123 (This problem cannot happen on the GNU system.)
126 The named file does not exist, and @code{O_CREAT} is not specified.
129 The directory or file system that would contain the new file cannot be
130 extended, because there is no disk space left.
133 @code{O_NONBLOCK} and @code{O_WRONLY} are both set in the @var{flags}
134 argument, the file named by @var{filename} is a FIFO (@pxref{Pipes and
135 FIFOs}), and no process has the file open for reading.
138 The file resides on a read-only file system and any of @w{@code{O_WRONLY}},
139 @code{O_RDWR}, and @code{O_TRUNC} are set in the @var{flags} argument,
140 or @code{O_CREAT} is set and the file does not already exist.
145 If on a 32 bit machine the sources are translated with
146 @code{_FILE_OFFSET_BITS == 64} the function @code{open} returns a file
147 descriptor opened in the large file mode which enables the file handling
148 functions to use files up to @math{2^63} bytes in size and offset from
149 @math{-2^63} to @math{2^63}. This happens transparently for the user
150 since all of the lowlevel file handling functions are equally replaced.
152 This function is a cancellation point in multi-threaded programs. This
153 is a problem if the thread allocates some resources (like memory, file
154 descriptors, semaphores or whatever) at the time @code{open} is
155 called. If the thread gets cancelled these resources stay allocated
156 until the program ends. To avoid this calls to @code{open} should be
157 protected using cancellation handlers.
158 @c ref pthread_cleanup_push / pthread_cleanup_pop
160 The @code{open} function is the underlying primitive for the @code{fopen}
161 and @code{freopen} functions, that create streams.
166 @deftypefun int open64 (const char *@var{filename}, int @var{flags}[, mode_t @var{mode}])
167 This function is similar to @code{open}. It returns a file descriptor
168 which can be used to access the file named by @var{filename}. The only
169 difference is that on 32 bit systems the file is opened in the
170 large file mode. I.e., file length and file offsets can exceed 31 bits.
172 When the sources are translated with @code{_FILE_OFFSET_BITS == 64} this
173 function is actually available under the name @code{open}. I.e., the
174 new, extended API using 64 bit file sizes and offsets transparently
175 replaces the old API.
180 @deftypefn {Obsolete function} int creat (const char *@var{filename}, mode_t @var{mode})
181 This function is obsolete. The call:
184 creat (@var{filename}, @var{mode})
191 open (@var{filename}, O_WRONLY | O_CREAT | O_TRUNC, @var{mode})
194 If on a 32 bit machine the sources are translated with
195 @code{_FILE_OFFSET_BITS == 64} the function @code{creat} returns a file
196 descriptor opened in the large file mode which enables the file handling
197 functions to use files up to @math{2^63} in size and offset from
198 @math{-2^63} to @math{2^63}. This happens transparently for the user
199 since all of the lowlevel file handling functions are equally replaced.
204 @deftypefn {Obsolete function} int creat64 (const char *@var{filename}, mode_t @var{mode})
205 This function is similar to @code{creat}. It returns a file descriptor
206 which can be used to access the file named by @var{filename}. The only
207 the difference is that on 32 bit systems the file is opened in the
208 large file mode. I.e., file length and file offsets can exceed 31 bits.
210 To use this file descriptor one must not use the normal operations but
211 instead the counterparts named @code{*64}, e.g., @code{read64}.
213 When the sources are translated with @code{_FILE_OFFSET_BITS == 64} this
214 function is actually available under the name @code{open}. I.e., the
215 new, extended API using 64 bit file sizes and offsets transparently
216 replaces the old API.
221 @deftypefun int close (int @var{filedes})
222 The function @code{close} closes the file descriptor @var{filedes}.
223 Closing a file has the following consequences:
227 The file descriptor is deallocated.
230 Any record locks owned by the process on the file are unlocked.
233 When all file descriptors associated with a pipe or FIFO have been closed,
234 any unread data is discarded.
237 This function is a cancellation point in multi-threaded programs. This
238 is a problem if the thread allocates some resources (like memory, file
239 descriptors, semaphores or whatever) at the time @code{close} is
240 called. If the thread gets cancelled these resources stay allocated
241 until the program ends. To avoid this, calls to @code{close} should be
242 protected using cancellation handlers.
243 @c ref pthread_cleanup_push / pthread_cleanup_pop
245 The normal return value from @code{close} is @math{0}; a value of @math{-1}
246 is returned in case of failure. The following @code{errno} error
247 conditions are defined for this function:
251 The @var{filedes} argument is not a valid file descriptor.
254 The @code{close} call was interrupted by a signal.
255 @xref{Interrupted Primitives}.
256 Here is an example of how to handle @code{EINTR} properly:
259 TEMP_FAILURE_RETRY (close (desc));
265 When the file is accessed by NFS, these errors from @code{write} can sometimes
266 not be detected until @code{close}. @xref{I/O Primitives}, for details
270 Please note that there is @emph{no} separate @code{close64} function.
271 This is not necessary since this function does not determine nor depend
272 on the mode of the file. The kernel which performs the @code{close}
273 operation knows which mode the descriptor is used for and can handle
277 To close a stream, call @code{fclose} (@pxref{Closing Streams}) instead
278 of trying to close its underlying file descriptor with @code{close}.
279 This flushes any buffered output and updates the stream object to
280 indicate that it is closed.
283 @section Input and Output Primitives
285 This section describes the functions for performing primitive input and
286 output operations on file descriptors: @code{read}, @code{write}, and
287 @code{lseek}. These functions are declared in the header file
293 @deftp {Data Type} ssize_t
294 This data type is used to represent the sizes of blocks that can be
295 read or written in a single operation. It is similar to @code{size_t},
296 but must be a signed type.
299 @cindex reading from a file descriptor
302 @deftypefun ssize_t read (int @var{filedes}, void *@var{buffer}, size_t @var{size})
303 The @code{read} function reads up to @var{size} bytes from the file
304 with descriptor @var{filedes}, storing the results in the @var{buffer}.
305 (This is not necessarily a character string, and no terminating null
308 @cindex end-of-file, on a file descriptor
309 The return value is the number of bytes actually read. This might be
310 less than @var{size}; for example, if there aren't that many bytes left
311 in the file or if there aren't that many bytes immediately available.
312 The exact behavior depends on what kind of file it is. Note that
313 reading less than @var{size} bytes is not an error.
315 A value of zero indicates end-of-file (except if the value of the
316 @var{size} argument is also zero). This is not considered an error.
317 If you keep calling @code{read} while at end-of-file, it will keep
318 returning zero and doing nothing else.
320 If @code{read} returns at least one character, there is no way you can
321 tell whether end-of-file was reached. But if you did reach the end, the
322 next read will return zero.
324 In case of an error, @code{read} returns @math{-1}. The following
325 @code{errno} error conditions are defined for this function:
329 Normally, when no input is immediately available, @code{read} waits for
330 some input. But if the @code{O_NONBLOCK} flag is set for the file
331 (@pxref{File Status Flags}), @code{read} returns immediately without
332 reading any data, and reports this error.
334 @strong{Compatibility Note:} Most versions of BSD Unix use a different
335 error code for this: @code{EWOULDBLOCK}. In the GNU library,
336 @code{EWOULDBLOCK} is an alias for @code{EAGAIN}, so it doesn't matter
339 On some systems, reading a large amount of data from a character special
340 file can also fail with @code{EAGAIN} if the kernel cannot find enough
341 physical memory to lock down the user's pages. This is limited to
342 devices that transfer with direct memory access into the user's memory,
343 which means it does not include terminals, since they always use
344 separate buffers inside the kernel. This problem never happens in the
347 Any condition that could result in @code{EAGAIN} can instead result in a
348 successful @code{read} which returns fewer bytes than requested.
349 Calling @code{read} again immediately would result in @code{EAGAIN}.
352 The @var{filedes} argument is not a valid file descriptor,
353 or is not open for reading.
356 @code{read} was interrupted by a signal while it was waiting for input.
357 @xref{Interrupted Primitives}. A signal will not necessary cause
358 @code{read} to return @code{EINTR}; it may instead result in a
359 successful @code{read} which returns fewer bytes than requested.
362 For many devices, and for disk files, this error code indicates
365 @code{EIO} also occurs when a background process tries to read from the
366 controlling terminal, and the normal action of stopping the process by
367 sending it a @code{SIGTTIN} signal isn't working. This might happen if
368 the signal is being blocked or ignored, or because the process group is
369 orphaned. @xref{Job Control}, for more information about job control,
370 and @ref{Signal Handling}, for information about signals.
373 Please note that there is no function named @code{read64}. This is not
374 necessary since this function does not directly modify or handle the
375 possibly wide file offset. Since the kernel handles this state
376 internally, the @code{read} function can be used for all cases.
378 This function is a cancellation point in multi-threaded programs. This
379 is a problem if the thread allocates some resources (like memory, file
380 descriptors, semaphores or whatever) at the time @code{read} is
381 called. If the thread gets cancelled these resources stay allocated
382 until the program ends. To avoid this, calls to @code{read} should be
383 protected using cancellation handlers.
384 @c ref pthread_cleanup_push / pthread_cleanup_pop
386 The @code{read} function is the underlying primitive for all of the
387 functions that read from streams, such as @code{fgetc}.
392 @deftypefun ssize_t pread (int @var{filedes}, void *@var{buffer}, size_t @var{size}, off_t @var{offset})
393 The @code{pread} function is similar to the @code{read} function. The
394 first three arguments are identical, and the return values and error
395 codes also correspond.
397 The difference is the fourth argument and its handling. The data block
398 is not read from the current position of the file descriptor
399 @code{filedes}. Instead the data is read from the file starting at
400 position @var{offset}. The position of the file descriptor itself is
401 not affected by the operation. The value is the same as before the call.
403 When the source file is compiled with @code{_FILE_OFFSET_BITS == 64} the
404 @code{pread} function is in fact @code{pread64} and the type
405 @code{off_t} has 64 bits, which makes it possible to handle files up to
406 @math{2^63} bytes in length.
408 The return value of @code{pread} describes the number of bytes read.
409 In the error case it returns @math{-1} like @code{read} does and the
410 error codes are also the same, with these additions:
414 The value given for @var{offset} is negative and therefore illegal.
417 The file descriptor @var{filedes} is associate with a pipe or a FIFO and
418 this device does not allow positioning of the file pointer.
421 The function is an extension defined in the Unix Single Specification
427 @deftypefun ssize_t pread64 (int @var{filedes}, void *@var{buffer}, size_t @var{size}, off64_t @var{offset})
428 This function is similar to the @code{pread} function. The difference
429 is that the @var{offset} parameter is of type @code{off64_t} instead of
430 @code{off_t} which makes it possible on 32 bit machines to address
431 files larger than @math{2^31} bytes and up to @math{2^63} bytes. The
432 file descriptor @code{filedes} must be opened using @code{open64} since
433 otherwise the large offsets possible with @code{off64_t} will lead to
434 errors with a descriptor in small file mode.
436 When the source file is compiled with @code{_FILE_OFFSET_BITS == 64} on a
437 32 bit machine this function is actually available under the name
438 @code{pread} and so transparently replaces the 32 bit interface.
441 @cindex writing to a file descriptor
444 @deftypefun ssize_t write (int @var{filedes}, const void *@var{buffer}, size_t @var{size})
445 The @code{write} function writes up to @var{size} bytes from
446 @var{buffer} to the file with descriptor @var{filedes}. The data in
447 @var{buffer} is not necessarily a character string and a null character is
448 output like any other character.
450 The return value is the number of bytes actually written. This may be
451 @var{size}, but can always be smaller. Your program should always call
452 @code{write} in a loop, iterating until all the data is written.
454 Once @code{write} returns, the data is enqueued to be written and can be
455 read back right away, but it is not necessarily written out to permanent
456 storage immediately. You can use @code{fsync} when you need to be sure
457 your data has been permanently stored before continuing. (It is more
458 efficient for the system to batch up consecutive writes and do them all
459 at once when convenient. Normally they will always be written to disk
460 within a minute or less.) Modern systems provide another function
461 @code{fdatasync} which guarantees integrity only for the file data and
463 @c !!! xref fsync, fdatasync
464 You can use the @code{O_FSYNC} open mode to make @code{write} always
465 store the data to disk before returning; @pxref{Operating Modes}.
467 In the case of an error, @code{write} returns @math{-1}. The following
468 @code{errno} error conditions are defined for this function:
472 Normally, @code{write} blocks until the write operation is complete.
473 But if the @code{O_NONBLOCK} flag is set for the file (@pxref{Control
474 Operations}), it returns immediately without writing any data and
475 reports this error. An example of a situation that might cause the
476 process to block on output is writing to a terminal device that supports
477 flow control, where output has been suspended by receipt of a STOP
480 @strong{Compatibility Note:} Most versions of BSD Unix use a different
481 error code for this: @code{EWOULDBLOCK}. In the GNU library,
482 @code{EWOULDBLOCK} is an alias for @code{EAGAIN}, so it doesn't matter
485 On some systems, writing a large amount of data from a character special
486 file can also fail with @code{EAGAIN} if the kernel cannot find enough
487 physical memory to lock down the user's pages. This is limited to
488 devices that transfer with direct memory access into the user's memory,
489 which means it does not include terminals, since they always use
490 separate buffers inside the kernel. This problem does not arise in the
494 The @var{filedes} argument is not a valid file descriptor,
495 or is not open for writing.
498 The size of the file would become larger than the implementation can support.
501 The @code{write} operation was interrupted by a signal while it was
502 blocked waiting for completion. A signal will not necessarily cause
503 @code{write} to return @code{EINTR}; it may instead result in a
504 successful @code{write} which writes fewer bytes than requested.
505 @xref{Interrupted Primitives}.
508 For many devices, and for disk files, this error code indicates
512 The device containing the file is full.
515 This error is returned when you try to write to a pipe or FIFO that
516 isn't open for reading by any process. When this happens, a @code{SIGPIPE}
517 signal is also sent to the process; see @ref{Signal Handling}.
520 Unless you have arranged to prevent @code{EINTR} failures, you should
521 check @code{errno} after each failing call to @code{write}, and if the
522 error was @code{EINTR}, you should simply repeat the call.
523 @xref{Interrupted Primitives}. The easy way to do this is with the
524 macro @code{TEMP_FAILURE_RETRY}, as follows:
527 nbytes = TEMP_FAILURE_RETRY (write (desc, buffer, count));
530 Please note that there is no function named @code{write64}. This is not
531 necessary since this function does not directly modify or handle the
532 possibly wide file offset. Since the kernel handles this state
533 internally the @code{write} function can be used for all cases.
535 This function is a cancellation point in multi-threaded programs. This
536 is a problem if the thread allocates some resources (like memory, file
537 descriptors, semaphores or whatever) at the time @code{write} is
538 called. If the thread gets cancelled these resources stay allocated
539 until the program ends. To avoid this, calls to @code{write} should be
540 protected using cancellation handlers.
541 @c ref pthread_cleanup_push / pthread_cleanup_pop
543 The @code{write} function is the underlying primitive for all of the
544 functions that write to streams, such as @code{fputc}.
549 @deftypefun ssize_t pwrite (int @var{filedes}, const void *@var{buffer}, size_t @var{size}, off_t @var{offset})
550 The @code{pwrite} function is similar to the @code{write} function. The
551 first three arguments are identical, and the return values and error codes
554 The difference is the fourth argument and its handling. The data block
555 is not written to the current position of the file descriptor
556 @code{filedes}. Instead the data is written to the file starting at
557 position @var{offset}. The position of the file descriptor itself is
558 not affected by the operation. The value is the same as before the call.
560 When the source file is compiled with @code{_FILE_OFFSET_BITS == 64} the
561 @code{pwrite} function is in fact @code{pwrite64} and the type
562 @code{off_t} has 64 bits, which makes it possible to handle files up to
563 @math{2^63} bytes in length.
565 The return value of @code{pwrite} describes the number of written bytes.
566 In the error case it returns @math{-1} like @code{write} does and the
567 error codes are also the same, with these additions:
571 The value given for @var{offset} is negative and therefore illegal.
574 The file descriptor @var{filedes} is associated with a pipe or a FIFO and
575 this device does not allow positioning of the file pointer.
578 The function is an extension defined in the Unix Single Specification
584 @deftypefun ssize_t pwrite64 (int @var{filedes}, const void *@var{buffer}, size_t @var{size}, off64_t @var{offset})
585 This function is similar to the @code{pwrite} function. The difference
586 is that the @var{offset} parameter is of type @code{off64_t} instead of
587 @code{off_t} which makes it possible on 32 bit machines to address
588 files larger than @math{2^31} bytes and up to @math{2^63} bytes. The
589 file descriptor @code{filedes} must be opened using @code{open64} since
590 otherwise the large offsets possible with @code{off64_t} will lead to
591 errors with a descriptor in small file mode.
593 When the source file is compiled using @code{_FILE_OFFSET_BITS == 64} on a
594 32 bit machine this function is actually available under the name
595 @code{pwrite} and so transparently replaces the 32 bit interface.
599 @node File Position Primitive
600 @section Setting the File Position of a Descriptor
602 Just as you can set the file position of a stream with @code{fseek}, you
603 can set the file position of a descriptor with @code{lseek}. This
604 specifies the position in the file for the next @code{read} or
605 @code{write} operation. @xref{File Positioning}, for more information
606 on the file position and what it means.
608 To read the current file position value from a descriptor, use
609 @code{lseek (@var{desc}, 0, SEEK_CUR)}.
611 @cindex file positioning on a file descriptor
612 @cindex positioning a file descriptor
613 @cindex seeking on a file descriptor
616 @deftypefun off_t lseek (int @var{filedes}, off_t @var{offset}, int @var{whence})
617 The @code{lseek} function is used to change the file position of the
618 file with descriptor @var{filedes}.
620 The @var{whence} argument specifies how the @var{offset} should be
621 interpreted, in the same way as for the @code{fseek} function, and it must
622 be one of the symbolic constants @code{SEEK_SET}, @code{SEEK_CUR}, or
627 Specifies that @var{whence} is a count of characters from the beginning
631 Specifies that @var{whence} is a count of characters from the current
632 file position. This count may be positive or negative.
635 Specifies that @var{whence} is a count of characters from the end of
636 the file. A negative count specifies a position within the current
637 extent of the file; a positive count specifies a position past the
638 current end. If you set the position past the current end, and
639 actually write data, you will extend the file with zeros up to that
643 The return value from @code{lseek} is normally the resulting file
644 position, measured in bytes from the beginning of the file.
645 You can use this feature together with @code{SEEK_CUR} to read the
646 current file position.
648 If you want to append to the file, setting the file position to the
649 current end of file with @code{SEEK_END} is not sufficient. Another
650 process may write more data after you seek but before you write,
651 extending the file so the position you write onto clobbers their data.
652 Instead, use the @code{O_APPEND} operating mode; @pxref{Operating Modes}.
654 You can set the file position past the current end of the file. This
655 does not by itself make the file longer; @code{lseek} never changes the
656 file. But subsequent output at that position will extend the file.
657 Characters between the previous end of file and the new position are
658 filled with zeros. Extending the file in this way can create a
659 ``hole'': the blocks of zeros are not actually allocated on disk, so the
660 file takes up less space than it appears to; it is then called a
663 @cindex holes in files
665 If the file position cannot be changed, or the operation is in some way
666 invalid, @code{lseek} returns a value of @math{-1}. The following
667 @code{errno} error conditions are defined for this function:
671 The @var{filedes} is not a valid file descriptor.
674 The @var{whence} argument value is not valid, or the resulting
675 file offset is not valid. A file offset is invalid.
678 The @var{filedes} corresponds to an object that cannot be positioned,
679 such as a pipe, FIFO or terminal device. (POSIX.1 specifies this error
680 only for pipes and FIFOs, but in the GNU system, you always get
681 @code{ESPIPE} if the object is not seekable.)
684 When the source file is compiled with @code{_FILE_OFFSET_BITS == 64} the
685 @code{lseek} function is in fact @code{lseek64} and the type
686 @code{off_t} has 64 bits which makes it possible to handle files up to
687 @math{2^63} bytes in length.
689 This function is a cancellation point in multi-threaded programs. This
690 is a problem if the thread allocates some resources (like memory, file
691 descriptors, semaphores or whatever) at the time @code{lseek} is
692 called. If the thread gets cancelled these resources stay allocated
693 until the program ends. To avoid this calls to @code{lseek} should be
694 protected using cancellation handlers.
695 @c ref pthread_cleanup_push / pthread_cleanup_pop
697 The @code{lseek} function is the underlying primitive for the
698 @code{fseek}, @code{fseeko}, @code{ftell}, @code{ftello} and
699 @code{rewind} functions, which operate on streams instead of file
705 @deftypefun off64_t lseek64 (int @var{filedes}, off64_t @var{offset}, int @var{whence})
706 This function is similar to the @code{lseek} function. The difference
707 is that the @var{offset} parameter is of type @code{off64_t} instead of
708 @code{off_t} which makes it possible on 32 bit machines to address
709 files larger than @math{2^31} bytes and up to @math{2^63} bytes. The
710 file descriptor @code{filedes} must be opened using @code{open64} since
711 otherwise the large offsets possible with @code{off64_t} will lead to
712 errors with a descriptor in small file mode.
714 When the source file is compiled with @code{_FILE_OFFSET_BITS == 64} on a
715 32 bits machine this function is actually available under the name
716 @code{lseek} and so transparently replaces the 32 bit interface.
719 You can have multiple descriptors for the same file if you open the file
720 more than once, or if you duplicate a descriptor with @code{dup}.
721 Descriptors that come from separate calls to @code{open} have independent
722 file positions; using @code{lseek} on one descriptor has no effect on the
730 d1 = open ("foo", O_RDONLY);
731 d2 = open ("foo", O_RDONLY);
732 lseek (d1, 1024, SEEK_SET);
739 will read the first four characters of the file @file{foo}. (The
740 error-checking code necessary for a real program has been omitted here
743 By contrast, descriptors made by duplication share a common file
744 position with the original descriptor that was duplicated. Anything
745 which alters the file position of one of the duplicates, including
746 reading or writing data, affects all of them alike. Thus, for example,
751 char buf1[4], buf2[4];
752 d1 = open ("foo", O_RDONLY);
755 lseek (d3, 1024, SEEK_SET);
762 will read four characters starting with the 1024'th character of
763 @file{foo}, and then four more characters starting with the 1028'th
768 @deftp {Data Type} off_t
769 This is an arithmetic data type used to represent file sizes.
770 In the GNU system, this is equivalent to @code{fpos_t} or @code{long int}.
772 If the source is compiled with @code{_FILE_OFFSET_BITS == 64} this type
773 is transparently replaced by @code{off64_t}.
778 @deftp {Data Type} off64_t
779 This type is used similar to @code{off_t}. The difference is that even
780 on 32 bit machines, where the @code{off_t} type would have 32 bits,
781 @code{off64_t} has 64 bits and so is able to address files up to
782 @math{2^63} bytes in length.
784 When compiling with @code{_FILE_OFFSET_BITS == 64} this type is
785 available under the name @code{off_t}.
788 These aliases for the @samp{SEEK_@dots{}} constants exist for the sake
789 of compatibility with older BSD systems. They are defined in two
790 different header files: @file{fcntl.h} and @file{sys/file.h}.
794 An alias for @code{SEEK_SET}.
797 An alias for @code{SEEK_CUR}.
800 An alias for @code{SEEK_END}.
803 @node Descriptors and Streams
804 @section Descriptors and Streams
805 @cindex streams, and file descriptors
806 @cindex converting file descriptor to stream
807 @cindex extracting file descriptor from stream
809 Given an open file descriptor, you can create a stream for it with the
810 @code{fdopen} function. You can get the underlying file descriptor for
811 an existing stream with the @code{fileno} function. These functions are
812 declared in the header file @file{stdio.h}.
817 @deftypefun {FILE *} fdopen (int @var{filedes}, const char *@var{opentype})
818 The @code{fdopen} function returns a new stream for the file descriptor
821 The @var{opentype} argument is interpreted in the same way as for the
822 @code{fopen} function (@pxref{Opening Streams}), except that
823 the @samp{b} option is not permitted; this is because GNU makes no
824 distinction between text and binary files. Also, @code{"w"} and
825 @code{"w+"} do not cause truncation of the file; these have an effect only
826 when opening a file, and in this case the file has already been opened.
827 You must make sure that the @var{opentype} argument matches the actual
828 mode of the open file descriptor.
830 The return value is the new stream. If the stream cannot be created
831 (for example, if the modes for the file indicated by the file descriptor
832 do not permit the access specified by the @var{opentype} argument), a
833 null pointer is returned instead.
835 In some other systems, @code{fdopen} may fail to detect that the modes
836 for file descriptor do not permit the access specified by
837 @code{opentype}. The GNU C library always checks for this.
840 For an example showing the use of the @code{fdopen} function,
841 see @ref{Creating a Pipe}.
845 @deftypefun int fileno (FILE *@var{stream})
846 This function returns the file descriptor associated with the stream
847 @var{stream}. If an error is detected (for example, if the @var{stream}
848 is not valid) or if @var{stream} does not do I/O to a file,
849 @code{fileno} returns @math{-1}.
854 @deftypefun int fileno_unlocked (FILE *@var{stream})
855 The @code{fileno_unlocked} function is equivalent to the @code{fileno}
856 function except that it does not implicitly lock the stream if the state
857 is @code{FSETLOCKING_INTERNAL}.
859 This function is a GNU extension.
862 @cindex standard file descriptors
863 @cindex file descriptors, standard
864 There are also symbolic constants defined in @file{unistd.h} for the
865 file descriptors belonging to the standard streams @code{stdin},
866 @code{stdout}, and @code{stderr}; see @ref{Standard Streams}.
874 This macro has value @code{0}, which is the file descriptor for
876 @cindex standard input file descriptor
881 @vindex STDOUT_FILENO
882 This macro has value @code{1}, which is the file descriptor for
884 @cindex standard output file descriptor
889 @vindex STDERR_FILENO
890 This macro has value @code{2}, which is the file descriptor for
891 standard error output.
893 @cindex standard error file descriptor
895 @node Stream/Descriptor Precautions
896 @section Dangers of Mixing Streams and Descriptors
898 @cindex streams and descriptors
899 @cindex descriptors and streams
900 @cindex mixing descriptors and streams
902 You can have multiple file descriptors and streams (let's call both
903 streams and descriptors ``channels'' for short) connected to the same
904 file, but you must take care to avoid confusion between channels. There
905 are two cases to consider: @dfn{linked} channels that share a single
906 file position value, and @dfn{independent} channels that have their own
909 It's best to use just one channel in your program for actual data
910 transfer to any given file, except when all the access is for input.
911 For example, if you open a pipe (something you can only do at the file
912 descriptor level), either do all I/O with the descriptor, or construct a
913 stream from the descriptor with @code{fdopen} and then do all I/O with
917 * Linked Channels:: Dealing with channels sharing a file position.
918 * Independent Channels:: Dealing with separately opened, unlinked channels.
919 * Cleaning Streams:: Cleaning a stream makes it safe to use
923 @node Linked Channels
924 @subsection Linked Channels
925 @cindex linked channels
927 Channels that come from a single opening share the same file position;
928 we call them @dfn{linked} channels. Linked channels result when you
929 make a stream from a descriptor using @code{fdopen}, when you get a
930 descriptor from a stream with @code{fileno}, when you copy a descriptor
931 with @code{dup} or @code{dup2}, and when descriptors are inherited
932 during @code{fork}. For files that don't support random access, such as
933 terminals and pipes, @emph{all} channels are effectively linked. On
934 random-access files, all append-type output streams are effectively
935 linked to each other.
937 @cindex cleaning up a stream
938 If you have been using a stream for I/O, and you want to do I/O using
939 another channel (either a stream or a descriptor) that is linked to it,
940 you must first @dfn{clean up} the stream that you have been using.
941 @xref{Cleaning Streams}.
943 Terminating a process, or executing a new program in the process,
944 destroys all the streams in the process. If descriptors linked to these
945 streams persist in other processes, their file positions become
946 undefined as a result. To prevent this, you must clean up the streams
947 before destroying them.
949 @node Independent Channels
950 @subsection Independent Channels
951 @cindex independent channels
953 When you open channels (streams or descriptors) separately on a seekable
954 file, each channel has its own file position. These are called
955 @dfn{independent channels}.
957 The system handles each channel independently. Most of the time, this
958 is quite predictable and natural (especially for input): each channel
959 can read or write sequentially at its own place in the file. However,
960 if some of the channels are streams, you must take these precautions:
964 You should clean an output stream after use, before doing anything else
965 that might read or write from the same part of the file.
968 You should clean an input stream before reading data that may have been
969 modified using an independent channel. Otherwise, you might read
970 obsolete data that had been in the stream's buffer.
973 If you do output to one channel at the end of the file, this will
974 certainly leave the other independent channels positioned somewhere
975 before the new end. You cannot reliably set their file positions to the
976 new end of file before writing, because the file can always be extended
977 by another process between when you set the file position and when you
978 write the data. Instead, use an append-type descriptor or stream; they
979 always output at the current end of the file. In order to make the
980 end-of-file position accurate, you must clean the output channel you
981 were using, if it is a stream.
983 It's impossible for two channels to have separate file pointers for a
984 file that doesn't support random access. Thus, channels for reading or
985 writing such files are always linked, never independent. Append-type
986 channels are also always linked. For these channels, follow the rules
987 for linked channels; see @ref{Linked Channels}.
989 @node Cleaning Streams
990 @subsection Cleaning Streams
992 On the GNU system, you can clean up any stream with @code{fclean}:
996 @deftypefun int fclean (FILE *@var{stream})
997 Clean up the stream @var{stream} so that its buffer is empty. If
998 @var{stream} is doing output, force it out. If @var{stream} is doing
999 input, give the data in the buffer back to the system, arranging to
1003 On other systems, you can use @code{fflush} to clean a stream in most
1006 You can skip the @code{fclean} or @code{fflush} if you know the stream
1007 is already clean. A stream is clean whenever its buffer is empty. For
1008 example, an unbuffered stream is always clean. An input stream that is
1009 at end-of-file is clean. A line-buffered stream is clean when the last
1010 character output was a newline.
1012 There is one case in which cleaning a stream is impossible on most
1013 systems. This is when the stream is doing input from a file that is not
1014 random-access. Such streams typically read ahead, and when the file is
1015 not random access, there is no way to give back the excess data already
1016 read. When an input stream reads from a random-access file,
1017 @code{fflush} does clean the stream, but leaves the file pointer at an
1018 unpredictable place; you must set the file pointer before doing any
1019 further I/O. On the GNU system, using @code{fclean} avoids both of
1022 Closing an output-only stream also does @code{fflush}, so this is a
1023 valid way of cleaning an output stream. On the GNU system, closing an
1024 input stream does @code{fclean}.
1026 You need not clean a stream before using its descriptor for control
1027 operations such as setting terminal modes; these operations don't affect
1028 the file position and are not affected by it. You can use any
1029 descriptor for these operations, and all channels are affected
1030 simultaneously. However, text already ``output'' to a stream but still
1031 buffered by the stream will be subject to the new terminal modes when
1032 subsequently flushed. To make sure ``past'' output is covered by the
1033 terminal settings that were in effect at the time, flush the output
1034 streams for that terminal before setting the modes. @xref{Terminal
1037 @node Scatter-Gather
1038 @section Fast Scatter-Gather I/O
1039 @cindex scatter-gather
1041 Some applications may need to read or write data to multiple buffers,
1042 which are separated in memory. Although this can be done easily enough
1043 with multiple calls to @code{read} and @code{write}, it is inefficent
1044 because there is overhead associated with each kernel call.
1046 Instead, many platforms provide special high-speed primitives to perform
1047 these @dfn{scatter-gather} operations in a single kernel call. The GNU C
1048 library will provide an emulation on any system that lacks these
1049 primitives, so they are not a portability threat. They are defined in
1052 These functions are controlled with arrays of @code{iovec} structures,
1053 which describe the location and size of each buffer.
1057 @deftp {Data Type} {struct iovec}
1059 The @code{iovec} structure describes a buffer. It contains two fields:
1063 @item void *iov_base
1064 Contains the address of a buffer.
1066 @item size_t iov_len
1067 Contains the length of the buffer.
1074 @deftypefun ssize_t readv (int @var{filedes}, const struct iovec *@var{vector}, int @var{count})
1076 The @code{readv} function reads data from @var{filedes} and scatters it
1077 into the buffers described in @var{vector}, which is taken to be
1078 @var{count} structures long. As each buffer is filled, data is sent to the
1081 Note that @code{readv} is not guaranteed to fill all the buffers.
1082 It may stop at any point, for the same reasons @code{read} would.
1084 The return value is a count of bytes (@emph{not} buffers) read, @math{0}
1085 indicating end-of-file, or @math{-1} indicating an error. The possible
1086 errors are the same as in @code{read}.
1092 @deftypefun ssize_t writev (int @var{filedes}, const struct iovec *@var{vector}, int @var{count})
1094 The @code{writev} function gathers data from the buffers described in
1095 @var{vector}, which is taken to be @var{count} structures long, and writes
1096 them to @code{filedes}. As each buffer is written, it moves on to the
1099 Like @code{readv}, @code{writev} may stop midstream under the same
1100 conditions @code{write} would.
1102 The return value is a count of bytes written, or @math{-1} indicating an
1103 error. The possible errors are the same as in @code{write}.
1107 @c Note - I haven't read this anywhere. I surmised it from my knowledge
1108 @c of computer science. Thus, there could be subtleties I'm missing.
1110 Note that if the buffers are small (under about 1kB), high-level streams
1111 may be easier to use than these functions. However, @code{readv} and
1112 @code{writev} are more efficient when the individual buffers themselves
1113 (as opposed to the total output), are large. In that case, a high-level
1114 stream would not be able to cache the data effectively.
1116 @node Memory-mapped I/O
1117 @section Memory-mapped I/O
1119 On modern operating systems, it is possible to @dfn{mmap} (pronounced
1120 ``em-map'') a file to a region of memory. When this is done, the file can
1121 be accessed just like an array in the program.
1123 This is more efficent than @code{read} or @code{write}, as only the regions
1124 of the file that a program actually accesses are loaded. Accesses to
1125 not-yet-loaded parts of the mmapped region are handled in the same way as
1128 Since mmapped pages can be stored back to their file when physical
1129 memory is low, it is possible to mmap files orders of magnitude larger
1130 than both the physical memory @emph{and} swap space. The only limit is
1131 address space. The theoretical limit is 4GB on a 32-bit machine -
1132 however, the actual limit will be smaller since some areas will be
1133 reserved for other purposes. If the LFS interface is used the file size
1134 on 32-bit systems is not limited to 2GB (offsets are signed which
1135 reduces the addressable area of 4GB by half); the full 64-bit are
1138 Memory mapping only works on entire pages of memory. Thus, addresses
1139 for mapping must be page-aligned, and length values will be rounded up.
1140 To determine the size of a page the machine uses one should use
1142 @vindex _SC_PAGESIZE
1144 size_t page_size = (size_t) sysconf (_SC_PAGESIZE);
1148 These functions are declared in @file{sys/mman.h}.
1152 @deftypefun {void *} mmap (void *@var{address}, size_t @var{length},int @var{protect}, int @var{flags}, int @var{filedes}, off_t @var{offset})
1154 The @code{mmap} function creates a new mapping, connected to bytes
1155 (@var{offset}) to (@var{offset} + @var{length}) in the file open on
1158 @var{address} gives a preferred starting address for the mapping.
1159 @code{NULL} expresses no preference. Any previous mapping at that
1160 address is automatically removed. The address you give may still be
1161 changed, unless you use the @code{MAP_FIXED} flag.
1166 @var{protect} contains flags that control what kind of access is
1167 permitted. They include @code{PROT_READ}, @code{PROT_WRITE}, and
1168 @code{PROT_EXEC}, which permit reading, writing, and execution,
1169 respectively. Inappropriate access will cause a segfault (@pxref{Program
1172 Note that most hardware designs cannot support write permission without
1173 read permission, and many do not distinguish read and execute permission.
1174 Thus, you may receive wider permissions than you ask for, and mappings of
1175 write-only files may be denied even if you do not use @code{PROT_READ}.
1177 @var{flags} contains flags that control the nature of the map.
1178 One of @code{MAP_SHARED} or @code{MAP_PRIVATE} must be specified.
1184 This specifies that writes to the region should never be written back
1185 to the attached file. Instead, a copy is made for the process, and the
1186 region will be swapped normally if memory runs low. No other process will
1189 Since private mappings effectively revert to ordinary memory
1190 when written to, you must have enough virtual memory for a copy of
1191 the entire mmapped region if you use this mode with @code{PROT_WRITE}.
1194 This specifies that writes to the region will be written back to the
1195 file. Changes made will be shared immediately with other processes
1196 mmaping the same file.
1198 Note that actual writing may take place at any time. You need to use
1199 @code{msync}, described below, if it is important that other processes
1200 using conventional I/O get a consistent view of the file.
1203 This forces the system to use the exact mapping address specified in
1204 @var{address} and fail if it can't.
1206 @c One of these is official - the other is obviously an obsolete synonym
1210 This flag tells the system to create an anonymous mapping, not connected
1211 to a file. @var{filedes} and @var{off} are ignored, and the region is
1212 initialized with zeros.
1214 Anonymous maps are used as the basic primitive to extend the heap on some
1215 systems. They are also useful to share data between multiple tasks
1216 without creating a file.
1218 On some systems using private anonymous mmaps is more efficient than using
1219 @code{malloc} for large blocks. This is not an issue with the GNU C library,
1220 as the included @code{malloc} automatically uses @code{mmap} where appropriate.
1222 @c Linux has some other MAP_ options, which I have not discussed here.
1223 @c MAP_DENYWRITE, MAP_EXECUTABLE and MAP_GROWSDOWN don't seem applicable to
1224 @c user programs (and I don't understand the last two). MAP_LOCKED does
1225 @c not appear to be implemented.
1229 @code{mmap} returns the address of the new mapping, or @math{-1} for an
1232 Possible errors include:
1238 Either @var{address} was unusable, or inconsistent @var{flags} were
1243 @var{filedes} was not open for the type of access specified in @var{protect}.
1247 Either there is not enough memory for the operation, or the process is
1248 out of address space.
1252 This file is of a type that doesn't support mapping.
1256 The file is on a filesystem that doesn't support mapping.
1258 @c On Linux, EAGAIN will appear if the file has a conflicting mandatory lock.
1259 @c However mandatory locks are not discussed in this manual.
1261 @c Similarly, ETXTBSY will occur if the MAP_DENYWRITE flag (not documented
1262 @c here) is used and the file is already open for writing.
1270 @deftypefun {void *} mmap64 (void *@var{address}, size_t @var{length},int @var{protect}, int @var{flags}, int @var{filedes}, off64_t @var{offset})
1271 The @code{mmap64} function is equivalent to the @code{mmap} function but
1272 the @var{offset} parameter is of type @code{off64_t}. On 32-bit systems
1273 this allows the file associated with the @var{filedes} descriptor to be
1274 larger than 2GB. @var{filedes} must be a descriptor returned from a
1275 call to @code{open64} or @code{fopen64} and @code{freopen64} where the
1276 descriptor is retrieved with @code{fileno}.
1278 When the sources are translated with @code{_FILE_OFFSET_BITS == 64} this
1279 function is actually available under the name @code{mmap}. I.e., the
1280 new, extended API using 64 bit file sizes and offsets transparently
1281 replaces the old API.
1286 @deftypefun int munmap (void *@var{addr}, size_t @var{length})
1288 @code{munmap} removes any memory maps from (@var{addr}) to (@var{addr} +
1289 @var{length}). @var{length} should be the length of the mapping.
1291 It is safe to unmap multiple mappings in one command, or include unmapped
1292 space in the range. It is also possible to unmap only part of an existing
1293 mapping. However, only entire pages can be removed. If @var{length} is not
1294 an even number of pages, it will be rounded up.
1296 It returns @math{0} for success and @math{-1} for an error.
1298 One error is possible:
1303 The memory range given was outside the user mmap range or wasn't page
1312 @deftypefun int msync (void *@var{address}, size_t @var{length}, int @var{flags})
1314 When using shared mappings, the kernel can write the file at any time
1315 before the mapping is removed. To be certain data has actually been
1316 written to the file and will be accessible to non-memory-mapped I/O, it
1317 is necessary to use this function.
1319 It operates on the region @var{address} to (@var{address} + @var{length}).
1320 It may be used on part of a mapping or multiple mappings, however the
1321 region given should not contain any unmapped space.
1323 @var{flags} can contain some options:
1329 This flag makes sure the data is actually written @emph{to disk}.
1330 Normally @code{msync} only makes sure that accesses to a file with
1331 conventional I/O reflect the recent changes.
1335 This tells @code{msync} to begin the synchronization, but not to wait for
1338 @c Linux also has MS_INVALIDATE, which I don't understand.
1342 @code{msync} returns @math{0} for success and @math{-1} for
1343 error. Errors include:
1348 An invalid region was given, or the @var{flags} were invalid.
1351 There is no existing mapping in at least part of the given region.
1359 @deftypefun {void *} mremap (void *@var{address}, size_t @var{length}, size_t @var{new_length}, int @var{flag})
1361 This function can be used to change the size of an existing memory
1362 area. @var{address} and @var{length} must cover a region entirely mapped
1363 in the same @code{mmap} statement. A new mapping with the same
1364 characteristics will be returned with the length @var{new_length}.
1366 One option is possible, @code{MREMAP_MAYMOVE}. If it is given in
1367 @var{flags}, the system may remove the existing mapping and create a new
1368 one of the desired length in another location.
1370 The address of the resulting mapping is returned, or @math{-1}. Possible
1371 error codes include:
1376 There is no existing mapping in at least part of the original region, or
1377 the region covers two or more distinct mappings.
1380 The address given is misaligned or inappropriate.
1383 The region has pages locked, and if extended it would exceed the
1384 process's resource limit for locked pages. @xref{Limits on Resources}.
1387 The region is private writeable, and insufficent virtual memory is
1388 available to extend it. Also, this error will occur if
1389 @code{MREMAP_MAYMOVE} is not given and the extension would collide with
1390 another mapped region.
1395 This function is only available on a few systems. Except for performing
1396 optional optimizations one should not rely on this function.
1398 Not all file descriptors may be mapped. Sockets, pipes, and most devices
1399 only allow sequential access and do not fit into the mapping abstraction.
1400 In addition, some regular files may not be mmapable, and older kernels may
1401 not support mapping at all. Thus, programs using @code{mmap} should
1402 have a fallback method to use should it fail. @xref{Mmap,,,standards,GNU
1405 @c XXX madvice documentation missing
1407 @node Waiting for I/O
1408 @section Waiting for Input or Output
1409 @cindex waiting for input or output
1410 @cindex multiplexing input
1411 @cindex input from multiple files
1413 Sometimes a program needs to accept input on multiple input channels
1414 whenever input arrives. For example, some workstations may have devices
1415 such as a digitizing tablet, function button box, or dial box that are
1416 connected via normal asynchronous serial interfaces; good user interface
1417 style requires responding immediately to input on any device. Another
1418 example is a program that acts as a server to several other processes
1419 via pipes or sockets.
1421 You cannot normally use @code{read} for this purpose, because this
1422 blocks the program until input is available on one particular file
1423 descriptor; input on other channels won't wake it up. You could set
1424 nonblocking mode and poll each file descriptor in turn, but this is very
1427 A better solution is to use the @code{select} function. This blocks the
1428 program until input or output is ready on a specified set of file
1429 descriptors, or until a timer expires, whichever comes first. This
1430 facility is declared in the header file @file{sys/types.h}.
1433 In the case of a server socket (@pxref{Listening}), we say that
1434 ``input'' is available when there are pending connections that could be
1435 accepted (@pxref{Accepting Connections}). @code{accept} for server
1436 sockets blocks and interacts with @code{select} just as @code{read} does
1439 @cindex file descriptor sets, for @code{select}
1440 The file descriptor sets for the @code{select} function are specified
1441 as @code{fd_set} objects. Here is the description of the data type
1442 and some macros for manipulating these objects.
1444 @comment sys/types.h
1446 @deftp {Data Type} fd_set
1447 The @code{fd_set} data type represents file descriptor sets for the
1448 @code{select} function. It is actually a bit array.
1451 @comment sys/types.h
1453 @deftypevr Macro int FD_SETSIZE
1454 The value of this macro is the maximum number of file descriptors that a
1455 @code{fd_set} object can hold information about. On systems with a
1456 fixed maximum number, @code{FD_SETSIZE} is at least that number. On
1457 some systems, including GNU, there is no absolute limit on the number of
1458 descriptors open, but this macro still has a constant value which
1459 controls the number of bits in an @code{fd_set}; if you get a file
1460 descriptor with a value as high as @code{FD_SETSIZE}, you cannot put
1461 that descriptor into an @code{fd_set}.
1464 @comment sys/types.h
1466 @deftypefn Macro void FD_ZERO (fd_set *@var{set})
1467 This macro initializes the file descriptor set @var{set} to be the
1471 @comment sys/types.h
1473 @deftypefn Macro void FD_SET (int @var{filedes}, fd_set *@var{set})
1474 This macro adds @var{filedes} to the file descriptor set @var{set}.
1477 @comment sys/types.h
1479 @deftypefn Macro void FD_CLR (int @var{filedes}, fd_set *@var{set})
1480 This macro removes @var{filedes} from the file descriptor set @var{set}.
1483 @comment sys/types.h
1485 @deftypefn Macro int FD_ISSET (int @var{filedes}, fd_set *@var{set})
1486 This macro returns a nonzero value (true) if @var{filedes} is a member
1487 of the file descriptor set @var{set}, and zero (false) otherwise.
1490 Next, here is the description of the @code{select} function itself.
1492 @comment sys/types.h
1494 @deftypefun int select (int @var{nfds}, fd_set *@var{read-fds}, fd_set *@var{write-fds}, fd_set *@var{except-fds}, struct timeval *@var{timeout})
1495 The @code{select} function blocks the calling process until there is
1496 activity on any of the specified sets of file descriptors, or until the
1497 timeout period has expired.
1499 The file descriptors specified by the @var{read-fds} argument are
1500 checked to see if they are ready for reading; the @var{write-fds} file
1501 descriptors are checked to see if they are ready for writing; and the
1502 @var{except-fds} file descriptors are checked for exceptional
1503 conditions. You can pass a null pointer for any of these arguments if
1504 you are not interested in checking for that kind of condition.
1506 A file descriptor is considered ready for reading if it is not at end of
1507 file. A server socket is considered ready for reading if there is a
1508 pending connection which can be accepted with @code{accept};
1509 @pxref{Accepting Connections}. A client socket is ready for writing when
1510 its connection is fully established; @pxref{Connecting}.
1512 ``Exceptional conditions'' does not mean errors---errors are reported
1513 immediately when an erroneous system call is executed, and do not
1514 constitute a state of the descriptor. Rather, they include conditions
1515 such as the presence of an urgent message on a socket. (@xref{Sockets},
1516 for information on urgent messages.)
1518 The @code{select} function checks only the first @var{nfds} file
1519 descriptors. The usual thing is to pass @code{FD_SETSIZE} as the value
1522 The @var{timeout} specifies the maximum time to wait. If you pass a
1523 null pointer for this argument, it means to block indefinitely until one
1524 of the file descriptors is ready. Otherwise, you should provide the
1525 time in @code{struct timeval} format; see @ref{High-Resolution
1526 Calendar}. Specify zero as the time (a @code{struct timeval} containing
1527 all zeros) if you want to find out which descriptors are ready without
1528 waiting if none are ready.
1530 The normal return value from @code{select} is the total number of ready file
1531 descriptors in all of the sets. Each of the argument sets is overwritten
1532 with information about the descriptors that are ready for the corresponding
1533 operation. Thus, to see if a particular descriptor @var{desc} has input,
1534 use @code{FD_ISSET (@var{desc}, @var{read-fds})} after @code{select} returns.
1536 If @code{select} returns because the timeout period expires, it returns
1539 Any signal will cause @code{select} to return immediately. So if your
1540 program uses signals, you can't rely on @code{select} to keep waiting
1541 for the full time specified. If you want to be sure of waiting for a
1542 particular amount of time, you must check for @code{EINTR} and repeat
1543 the @code{select} with a newly calculated timeout based on the current
1544 time. See the example below. See also @ref{Interrupted Primitives}.
1546 If an error occurs, @code{select} returns @code{-1} and does not modify
1547 the argument file descriptor sets. The following @code{errno} error
1548 conditions are defined for this function:
1552 One of the file descriptor sets specified an invalid file descriptor.
1555 The operation was interrupted by a signal. @xref{Interrupted Primitives}.
1558 The @var{timeout} argument is invalid; one of the components is negative
1563 @strong{Portability Note:} The @code{select} function is a BSD Unix
1566 Here is an example showing how you can use @code{select} to establish a
1567 timeout period for reading from a file descriptor. The @code{input_timeout}
1568 function blocks the calling process until input is available on the
1569 file descriptor, or until the timeout period expires.
1572 @include select.c.texi
1575 There is another example showing the use of @code{select} to multiplex
1576 input from multiple sockets in @ref{Server Example}.
1579 @node Synchronizing I/O
1580 @section Synchronizing I/O operations
1582 @cindex synchronizing
1583 In most modern operating systems the normal I/O operations are not
1584 executed synchronously. I.e., even if a @code{write} system call
1585 returns this does not mean the data is actually written to the media,
1588 In situations where synchronization points are necessary,you can use
1589 special functions which ensure that all operations finish before
1594 @deftypefun int sync (void)
1595 A call to this function will not return as long as there is data which
1596 has not been written to the device. All dirty buffers in the kernel will
1597 be written and so an overall consistent system can be achieved (if no
1598 other process in parallel writes data).
1600 A prototype for @code{sync} can be found in @file{unistd.h}.
1602 The return value is zero to indicate no error.
1605 Programs more often want to ensure that data written to a given file is
1606 committed, rather than all data in the system. For this, @code{sync} is overkill.
1611 @deftypefun int fsync (int @var{fildes})
1612 The @code{fsync} can be used to make sure all data associated with the
1613 open file @var{fildes} is written to the device associated with the
1614 descriptor. The function call does not return unless all actions have
1617 A prototype for @code{fsync} can be found in @file{unistd.h}.
1619 This function is a cancellation point in multi-threaded programs. This
1620 is a problem if the thread allocates some resources (like memory, file
1621 descriptors, semaphores or whatever) at the time @code{fsync} is
1622 called. If the thread gets cancelled these resources stay allocated
1623 until the program ends. To avoid this, calls to @code{fsync} should be
1624 protected using cancellation handlers.
1625 @c ref pthread_cleanup_push / pthread_cleanup_pop
1627 The return value of the function is zero if no error occurred. Otherwise
1628 it is @math{-1} and the global variable @var{errno} is set to the
1632 The descriptor @var{fildes} is not valid.
1635 No synchronization is possible since the system does not implement this.
1639 Sometimes it is not even necessary to write all data associated with a
1640 file descriptor. E.g., in database files which do not change in size it
1641 is enough to write all the file content data to the device.
1642 Meta-information like the modification time etc. are not that important
1643 and leaving such information uncommitted does not prevent a successful
1644 recovering of the file in case of a problem.
1648 @deftypefun int fdatasync (int @var{fildes})
1649 When a call to the @code{fdatasync} function returns, it is ensured
1650 that all of the file data is written to the device. For all pending I/O
1651 operations, the parts guaranteeing data integrity finished.
1653 Not all systems implement the @code{fdatasync} operation. On systems
1654 missing this functionality @code{fdatasync} is emulated by a call to
1655 @code{fsync} since the performed actions are a superset of those
1656 required by @code{fdatasyn}.
1658 The prototype for @code{fdatasync} is in @file{unistd.h}.
1660 The return value of the function is zero if no error occurred. Otherwise
1661 it is @math{-1} and the global variable @var{errno} is set to the
1665 The descriptor @var{fildes} is not valid.
1668 No synchronization is possible since the system does not implement this.
1673 @node Asynchronous I/O
1674 @section Perform I/O Operations in Parallel
1676 The POSIX.1b standard defines a new set of I/O operations which can
1677 significantly reduce the time an application spends waiting at I/O. The
1678 new functions allow a program to initiate one or more I/O operations and
1679 then immediately resume normal work while the I/O operations are
1680 executed in parallel. This functionality is available if the
1681 @file{unistd.h} file defines the symbol @code{_POSIX_ASYNCHRONOUS_IO}.
1683 These functions are part of the library with realtime functions named
1684 @file{librt}. They are not actually part of the @file{libc} binary.
1685 The implementation of these functions can be done using support in the
1686 kernel (if available) or using an implementation based on threads at
1687 userlevel. In the latter case it might be necessary to link applications
1688 with the thread library @file{libpthread} in addition to @file{librt}.
1690 All AIO operations operate on files which were opened previously. There
1691 might be arbitrarily many operations running for one file. The
1692 asynchronous I/O operations are controlled using a data structure named
1693 @code{struct aiocb} (@dfn{AIO control block}). It is defined in
1694 @file{aio.h} as follows.
1698 @deftp {Data Type} {struct aiocb}
1699 The POSIX.1b standard mandates that the @code{struct aiocb} structure
1700 contains at least the members described in the following table. There
1701 might be more elements which are used by the implementation, but
1702 depending on these elements is not portable and is highly deprecated.
1705 @item int aio_fildes
1706 This element specifies the file descriptor which is used for the
1707 operation. It must be a legal descriptor since otherwise the operation
1710 The device on which the file is opened must allow the seek operation.
1711 I.e., it is not possible to use any of the AIO operations on devices
1712 like terminals where an @code{lseek} call would lead to an error.
1714 @item off_t aio_offset
1715 This element specifies at which offset in the file the operation (input
1716 or output) is performed. Since the operations are carried out in arbitrary
1717 order and more than one operation for one file descriptor can be
1718 started, one cannot expect a current read/write position of the file
1721 @item volatile void *aio_buf
1722 This is a pointer to the buffer with the data to be written or the place
1723 where the read data is stored.
1725 @item size_t aio_nbytes
1726 This element specifies the length of the buffer pointed to by @code{aio_buf}.
1728 @item int aio_reqprio
1729 If the platform has defined @code{_POSIX_PRIORITIZED_IO} and
1730 @code{_POSIX_PRIORITY_SCHEDULING} the AIO requests are
1731 processed based on the current scheduling priority. The
1732 @code{aio_reqprio} element can then be used to lower the priority of the
1735 @item struct sigevent aio_sigevent
1736 This element specifies how the calling process is notified once the
1737 operation terminates. If the @code{sigev_notify} element is
1738 @code{SIGEV_NONE} no notification is send. If it is @code{SIGEV_SIGNAL}
1739 the signal determined by @code{sigev_signo} is send. Otherwise
1740 @code{sigev_notify} must be @code{SIGEV_THREAD}. In this case a thread
1741 is created which starts executing the function pointed to by
1742 @code{sigev_notify_function}.
1744 @item int aio_lio_opcode
1745 This element is only used by the @code{lio_listio} and
1746 @code{lio_listio64} functions. Since these functions allow an
1747 arbitrary number of operations to start at once, and each operation can be
1748 input or output (or nothing), the information must be stored in the
1749 control block. The possible values are:
1753 Start a read operation. Read from the file at position
1754 @code{aio_offset} and store the next @code{aio_nbytes} bytes in the
1755 buffer pointed to by @code{aio_buf}.
1758 Start a write operation. Write @code{aio_nbytes} bytes starting at
1759 @code{aio_buf} into the file starting at position @code{aio_offset}.
1762 Do nothing for this control block. This value is useful sometimes when
1763 an array of @code{struct aiocb} values contains holes, i.e., some of the
1764 values must not be handled although the whole array is presented to the
1765 @code{lio_listio} function.
1769 When the sources are compiled using @code{_FILE_OFFSET_BITS == 64} on a
1770 32 bit machine this type is in fact @code{struct aiocb64} since the LFS
1771 interface transparently replaces the @code{struct aiocb} definition.
1774 For use with the AIO functions defined in the LFS there is a similar type
1775 defined which replaces the types of the appropriate members with larger
1776 types but otherwise is equivalent to @code{struct aiocb}. Particularly,
1777 all member names are the same.
1781 @deftp {Data Type} {struct aiocb64}
1783 @item int aio_fildes
1784 This element specifies the file descriptor which is used for the
1785 operation. It must be a legal descriptor since otherwise the operation
1786 fails for obvious reasons.
1788 The device on which the file is opened must allow the seek operation.
1789 I.e., it is not possible to use any of the AIO operations on devices
1790 like terminals where an @code{lseek} call would lead to an error.
1792 @item off64_t aio_offset
1793 This element specifies at which offset in the file the operation (input
1794 or output) is performed. Since the operation are carried in arbitrary
1795 order and more than one operation for one file descriptor can be
1796 started, one cannot expect a current read/write position of the file
1799 @item volatile void *aio_buf
1800 This is a pointer to the buffer with the data to be written or the place
1801 where the ead data is stored.
1803 @item size_t aio_nbytes
1804 This element specifies the length of the buffer pointed to by @code{aio_buf}.
1806 @item int aio_reqprio
1807 If for the platform @code{_POSIX_PRIORITIZED_IO} and
1808 @code{_POSIX_PRIORITY_SCHEDULING} are defined the AIO requests are
1809 processed based on the current scheduling priority. The
1810 @code{aio_reqprio} element can then be used to lower the priority of the
1813 @item struct sigevent aio_sigevent
1814 This element specifies how the calling process is notified once the
1815 operation terminates. If the @code{sigev_notify} element is
1816 @code{SIGEV_NONE} no notification is sent. If it is @code{SIGEV_SIGNAL}
1817 the signal determined by @code{sigev_signo} is sent. Otherwise
1818 @code{sigev_notify} must be @code{SIGEV_THREAD} in which case a thread
1819 which starts executing the function pointed to by
1820 @code{sigev_notify_function}.
1822 @item int aio_lio_opcode
1823 This element is only used by the @code{lio_listio} and
1824 @code{[lio_listio64} functions. Since these functions allow an
1825 arbitrary number of operations to start at once, and since each operation can be
1826 input or output (or nothing), the information must be stored in the
1827 control block. See the description of @code{struct aiocb} for a description
1828 of the possible values.
1831 When the sources are compiled using @code{_FILE_OFFSET_BITS == 64} on a
1832 32 bit machine this type is available under the name @code{struct
1833 aiocb64} since the LFS replaces transparently the old interface.
1837 * Asynchronous Reads/Writes:: Asynchronous Read and Write Operations.
1838 * Status of AIO Operations:: Getting the Status of AIO Operations.
1839 * Synchronizing AIO Operations:: Getting into a consistent state.
1840 * Cancel AIO Operations:: Cancellation of AIO Operations.
1841 * Configuration of AIO:: How to optimize the AIO implementation.
1844 @node Asynchronous Reads/Writes
1845 @subsection Asynchronous Read and Write Operations
1849 @deftypefun int aio_read (struct aiocb *@var{aiocbp})
1850 This function initiates an asynchronous read operation. It
1851 immediately returns after the operation was enqueued or when an
1852 error was encountered.
1854 The first @code{aiocbp->aio_nbytes} bytes of the file for which
1855 @code{aiocbp->aio_fildes} is a descriptor are written to the buffer
1856 starting at @code{aiocbp->aio_buf}. Reading starts at the absolute
1857 position @code{aiocbp->aio_offset} in the file.
1859 If prioritized I/O is supported by the platform the
1860 @code{aiocbp->aio_reqprio} value is used to adjust the priority before
1861 the request is actually enqueued.
1863 The calling process is notified about the termination of the read
1864 request according to the @code{aiocbp->aio_sigevent} value.
1866 When @code{aio_read} returns, the return value is zero if no error
1867 occurred that can be found before the process is enqueued. If such an
1868 early error is found, the function returns @math{-1} and sets
1869 @code{errno} to one of the following values:
1873 The request was not enqueued due to (temporarily) exceeded resource
1876 The @code{aio_read} function is not implemented.
1878 The @code{aiocbp->aio_fildes} descriptor is not valid. This condition
1879 need not be recognized before enqueueing the request and so this error
1880 might also be signaled asynchronously.
1882 The @code{aiocbp->aio_offset} or @code{aiocbp->aio_reqpiro} value is
1883 invalid. This condition need not be recognized before enqueueing the
1884 request and so this error might also be signaled asynchronously.
1887 If @code{aio_read} returns zero, the current status of the request
1888 can be queried using @code{aio_error} and @code{aio_return} functions.
1889 As long as the value returned by @code{aio_error} is @code{EINPROGRESS}
1890 the operation has not yet completed. If @code{aio_error} returns zero,
1891 the operation successfully terminated, otherwise the value is to be
1892 interpreted as an error code. If the function terminated, the result of
1893 the operation can be obtained using a call to @code{aio_return}. The
1894 returned value is the same as an equivalent call to @code{read} would
1895 have returned. Possible error codes returned by @code{aio_error} are:
1899 The @code{aiocbp->aio_fildes} descriptor is not valid.
1901 The operation was cancelled before the operation was finished
1902 (@pxref{Cancel AIO Operations})
1904 The @code{aiocbp->aio_offset} value is invalid.
1907 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
1908 function is in fact @code{aio_read64} since the LFS interface transparently
1909 replaces the normal implementation.
1914 @deftypefun int aio_read64 (struct aiocb *@var{aiocbp})
1915 This function is similar to the @code{aio_read} function. The only
1916 difference is that on @w{32 bit} machines the file descriptor should
1917 be opened in the large file mode. Internally @code{aio_read64} uses
1918 functionality equivalent to @code{lseek64} (@pxref{File Position
1919 Primitive}) to position the file descriptor correctly for the reading,
1920 as opposed to @code{lseek} functionality used in @code{aio_read}.
1922 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
1923 function is available under the name @code{aio_read} and so transparently
1924 replaces the interface for small files on 32 bit machines.
1927 To write data asynchronously to a file there exists an equivalent pair
1928 of functions with a very similar interface.
1932 @deftypefun int aio_write (struct aiocb *@var{aiocbp})
1933 This function initiates an asynchronous write operation. The function
1934 call immediately returns after the operation was enqueued or if before
1935 this happens an error was encountered.
1937 The first @code{aiocbp->aio_nbytes} bytes from the buffer starting at
1938 @code{aiocbp->aio_buf} are written to the file for which
1939 @code{aiocbp->aio_fildes} is an descriptor, starting at the absolute
1940 position @code{aiocbp->aio_offset} in the file.
1942 If prioritized I/O is supported by the platform the
1943 @code{aiocbp->aio_reqprio} value is used to adjust the priority before
1944 the request is actually enqueued.
1946 The calling process is notified about the termination of the read
1947 request according to the @code{aiocbp->aio_sigevent} value.
1949 When @code{aio_write} returns the return value is zero if no error
1950 occurred that can be found before the process is enqueued. If such an
1951 early error is found the function returns @math{-1} and sets
1952 @code{errno} to one of the following values.
1956 The request was not enqueued due to (temporarily) exceeded resource
1959 The @code{aio_write} function is not implemented.
1961 The @code{aiocbp->aio_fildes} descriptor is not valid. This condition
1962 needs not be recognized before enqueueing the request and so this error
1963 might also be signaled asynchronously.
1965 The @code{aiocbp->aio_offset} or @code{aiocbp->aio_reqpiro} value is
1966 invalid. This condition needs not be recognized before enqueueing the
1967 request and so this error might also be signaled asynchronously.
1970 In the case @code{aio_write} returns zero the current status of the
1971 request can be queried using @code{aio_error} and @code{aio_return}
1972 functions. As long as the value returned by @code{aio_error} is
1973 @code{EINPROGRESS} the operation has not yet completed. If
1974 @code{aio_error} returns zero the operation successfully terminated,
1975 otherwise the value is to be interpreted as an error code. If the
1976 function terminated the result of the operation can be get using a call
1977 to @code{aio_return}. The returned value is the same as an equivalent
1978 call to @code{read} would have returned. Possible error code returned
1979 by @code{aio_error} are:
1983 The @code{aiocbp->aio_fildes} descriptor is not valid.
1985 The operation was cancelled before the operation was finished
1986 (@pxref{Cancel AIO Operations})
1988 The @code{aiocbp->aio_offset} value is invalid.
1991 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
1992 function is in fact @code{aio_write64} since the LFS interface transparently
1993 replaces the normal implementation.
1998 @deftypefun int aio_write64 (struct aiocb *@var{aiocbp})
1999 This function is similar to the @code{aio_write} function. The only
2000 difference is that on @w{32 bit} machines the file descriptor should
2001 be opened in the large file mode. Internally @code{aio_write64} uses
2002 functionality equivalent to @code{lseek64} (@pxref{File Position
2003 Primitive}) to position the file descriptor correctly for the writing,
2004 as opposed to @code{lseek} functionality used in @code{aio_write}.
2006 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
2007 function is available under the name @code{aio_write} and so transparently
2008 replaces the interface for small files on 32 bit machines.
2011 Beside these functions with the more or less traditional interface
2012 POSIX.1b also defines a function with can initiate more than one
2013 operation at once and which can handled freely mixed read and write
2014 operation. It is therefore similar to a combination of @code{readv} and
2019 @deftypefun int lio_listio (int @var{mode}, struct aiocb *const @var{list}[], int @var{nent}, struct sigevent *@var{sig})
2020 The @code{lio_listio} function can be used to enqueue an arbitrary
2021 number of read and write requests at one time. The requests can all be
2022 meant for the same file, all for different files or every solution in
2025 @code{lio_listio} gets the @var{nent} requests from the array pointed to
2026 by @var{list}. What operation has to be performed is determined by the
2027 @code{aio_lio_opcode} member in each element of @var{list}. If this
2028 field is @code{LIO_READ} an read operation is queued, similar to a call
2029 of @code{aio_read} for this element of the array (except that the way
2030 the termination is signalled is different, as we will see below). If
2031 the @code{aio_lio_opcode} member is @code{LIO_WRITE} an write operation
2032 is enqueued. Otherwise the @code{aio_lio_opcode} must be @code{LIO_NOP}
2033 in which case this element of @var{list} is simply ignored. This
2034 ``operation'' is useful in situations where one has a fixed array of
2035 @code{struct aiocb} elements from which only a few need to be handled at
2036 a time. Another situation is where the @code{lio_listio} call was
2037 cancelled before all requests are processed (@pxref{Cancel AIO
2038 Operations}) and the remaining requests have to be reissued.
2040 The other members of each element of the array pointed to by
2041 @code{list} must have values suitable for the operation as described in
2042 the documentation for @code{aio_read} and @code{aio_write} above.
2044 The @var{mode} argument determines how @code{lio_listio} behaves after
2045 having enqueued all the requests. If @var{mode} is @code{LIO_WAIT} it
2046 waits until all requests terminated. Otherwise @var{mode} must be
2047 @code{LIO_NOWAIT} and in this case the function returns immediately after
2048 having enqueued all the requests. In this case the caller gets a
2049 notification of the termination of all requests according to the
2050 @var{sig} parameter. If @var{sig} is @code{NULL} no notification is
2051 send. Otherwise a signal is sent or a thread is started, just as
2052 described in the description for @code{aio_read} or @code{aio_write}.
2054 If @var{mode} is @code{LIO_WAIT} the return value of @code{lio_listio}
2055 is @math{0} when all requests completed successfully. Otherwise the
2056 function return @math{-1} and @code{errno} is set accordingly. To find
2057 out which request or requests failed one has to use the @code{aio_error}
2058 function on all the elements of the array @var{list}.
2060 In case @var{mode} is @code{LIO_NOWAIT} the function return @math{0} if
2061 all requests were enqueued correctly. The current state of the requests
2062 can be found using @code{aio_error} and @code{aio_return} as described
2063 above. In case @code{lio_listio} returns @math{-1} in this mode the
2064 global variable @code{errno} is set accordingly. If a request did not
2065 yet terminate a call to @code{aio_error} returns @code{EINPROGRESS}. If
2066 the value is different the request is finished and the error value (or
2067 @math{0}) is returned and the result of the operation can be retrieved
2068 using @code{aio_return}.
2070 Possible values for @code{errno} are:
2074 The resources necessary to queue all the requests are not available in
2075 the moment. The error status for each element of @var{list} must be
2076 checked which request failed.
2078 Another reason could be that the system wide limit of AIO requests is
2079 exceeded. This cannot be the case for the implementation on GNU systems
2080 since no arbitrary limits exist.
2082 The @var{mode} parameter is invalid or @var{nent} is larger than
2083 @code{AIO_LISTIO_MAX}.
2085 One or more of the request's I/O operations failed. The error status of
2086 each request should be checked for which one failed.
2088 The @code{lio_listio} function is not supported.
2091 If the @var{mode} parameter is @code{LIO_NOWAIT} and the caller cancels
2092 an request the error status for this request returned by
2093 @code{aio_error} is @code{ECANCELED}.
2095 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
2096 function is in fact @code{lio_listio64} since the LFS interface
2097 transparently replaces the normal implementation.
2102 @deftypefun int lio_listio64 (int @var{mode}, struct aiocb *const @var{list}, int @var{nent}, struct sigevent *@var{sig})
2103 This function is similar to the @code{aio_listio} function. The only
2104 difference is that only @w{32 bit} machines the file descriptor should
2105 be opened in the large file mode. Internally @code{lio_listio64} uses
2106 functionality equivalent to @code{lseek64} (@pxref{File Position
2107 Primitive}) to position the file descriptor correctly for the reading or
2108 writing, as opposed to @code{lseek} functionality used in
2111 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
2112 function is available under the name @code{lio_listio} and so
2113 transparently replaces the interface for small files on 32 bit
2117 @node Status of AIO Operations
2118 @subsection Getting the Status of AIO Operations
2120 As already described in the documentation of the functions in the last
2121 section, it must be possible to get information about the status of an I/O
2122 request. When the operation is performed truly asynchronously (as with
2123 @code{aio_read} and @code{aio_write} and with @code{aio_listio} when the
2124 mode is @code{LIO_NOWAIT}) one sometimes needs to know whether a
2125 specific request already terminated and if yes, what the result was.
2126 The following two functions allow you to get this kind of information.
2130 @deftypefun int aio_error (const struct aiocb *@var{aiocbp})
2131 This function determines the error state of the request described by the
2132 @code{struct aiocb} variable pointed to by @var{aiocbp}. If the
2133 request has not yet terminated the value returned is always
2134 @code{EINPROGRESS}. Once the request has terminated the value
2135 @code{aio_error} returns is either @math{0} if the request completed
2136 successfully or it returns the value which would be stored in the
2137 @code{errno} variable if the request would have been done using
2138 @code{read}, @code{write}, or @code{fsync}.
2140 The function can return @code{ENOSYS} if it is not implemented. It
2141 could also return @code{EINVAL} if the @var{aiocbp} parameter does not
2142 refer to an asynchronous operation whose return status is not yet known.
2144 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
2145 function is in fact @code{aio_error64} since the LFS interface
2146 transparently replaces the normal implementation.
2151 @deftypefun int aio_error64 (const struct aiocb64 *@var{aiocbp})
2152 This function is similar to @code{aio_error} with the only difference
2153 that the argument is a reference to a variable of type @code{struct
2156 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
2157 function is available under the name @code{aio_error} and so
2158 transparently replaces the interface for small files on 32 bit
2164 @deftypefun ssize_t aio_return (const struct aiocb *@var{aiocbp})
2165 This function can be used to retrieve the return status of the operation
2166 carried out by the request described in the variable pointed to by
2167 @var{aiocbp}. As long as the error status of this request as returned
2168 by @code{aio_error} is @code{EINPROGRESS} the return of this function is
2171 Once the request is finished this function can be used exactly once to
2172 retrieve the return value. Following calls might lead to undefined
2173 behaviour. The return value itself is the value which would have been
2174 returned by the @code{read}, @code{write}, or @code{fsync} call.
2176 The function can return @code{ENOSYS} if it is not implemented. It
2177 could also return @code{EINVAL} if the @var{aiocbp} parameter does not
2178 refer to an asynchronous operation whose return status is not yet known.
2180 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
2181 function is in fact @code{aio_return64} since the LFS interface
2182 transparently replaces the normal implementation.
2187 @deftypefun int aio_return64 (const struct aiocb64 *@var{aiocbp})
2188 This function is similar to @code{aio_return} with the only difference
2189 that the argument is a reference to a variable of type @code{struct
2192 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
2193 function is available under the name @code{aio_return} and so
2194 transparently replaces the interface for small files on 32 bit
2198 @node Synchronizing AIO Operations
2199 @subsection Getting into a Consistent State
2201 When dealing with asynchronous operations it is sometimes necessary to
2202 get into a consistent state. This would mean for AIO that one wants to
2203 know whether a certain request or a group of request were processed.
2204 This could be done by waiting for the notification sent by the system
2205 after the operation terminated, but this sometimes would mean wasting
2206 resources (mainly computation time). Instead POSIX.1b defines two
2207 functions which will help with most kinds of consistency.
2209 The @code{aio_fsync} and @code{aio_fsync64} functions are only available
2210 if in @file{unistd.h} the symbol @code{_POSIX_SYNCHRONIZED_IO} is
2213 @cindex synchronizing
2216 @deftypefun int aio_fsync (int @var{op}, struct aiocb *@var{aiocbp})
2217 Calling this function forces all I/O operations operating queued at the
2218 time of the function call operating on the file descriptor
2219 @code{aiocbp->aio_fildes} into the synchronized I/O completion state
2220 (@pxref{Synchronizing I/O}). The @code{aio_fsync} function returns
2221 immediately but the notification through the method described in
2222 @code{aiocbp->aio_sigevent} will happen only after all requests for this
2223 file descriptor have terminated and the file is synchronized. This also
2224 means that requests for this very same file descriptor which are queued
2225 after the synchronization request are not affected.
2227 If @var{op} is @code{O_DSYNC} the synchronization happens as with a call
2228 to @code{fdatasync}. Otherwise @var{op} should be @code{O_SYNC} and
2229 the synchronization happens as with @code{fsync}.
2231 As long as the synchronization has not happened a call to
2232 @code{aio_error} with the reference to the object pointed to by
2233 @var{aiocbp} returns @code{EINPROGRESS}. Once the synchronization is
2234 done @code{aio_error} return @math{0} if the synchronization was not
2235 successful. Otherwise the value returned is the value to which the
2236 @code{fsync} or @code{fdatasync} function would have set the
2237 @code{errno} variable. In this case nothing can be assumed about the
2238 consistency for the data written to this file descriptor.
2240 The return value of this function is @math{0} if the request was
2241 successfully filed. Otherwise the return value is @math{-1} and
2242 @code{errno} is set to one of the following values:
2246 The request could not be enqueued due to temporary lack of resources.
2248 The file descriptor @code{aiocbp->aio_fildes} is not valid or not open
2251 The implementation does not support I/O synchronization or the @var{op}
2252 parameter is other than @code{O_DSYNC} and @code{O_SYNC}.
2254 This function is not implemented.
2257 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
2258 function is in fact @code{aio_return64} since the LFS interface
2259 transparently replaces the normal implementation.
2264 @deftypefun int aio_fsync64 (int @var{op}, struct aiocb64 *@var{aiocbp})
2265 This function is similar to @code{aio_fsync} with the only difference
2266 that the argument is a reference to a variable of type @code{struct
2269 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
2270 function is available under the name @code{aio_fsync} and so
2271 transparently replaces the interface for small files on 32 bit
2275 Another method of synchronization is to wait until one or more requests of a
2276 specific set terminated. This could be achieved by the @code{aio_*}
2277 functions to notify the initiating process about the termination but in
2278 some situations this is not the ideal solution. In a program which
2279 constantly updates clients somehow connected to the server it is not
2280 always the best solution to go round robin since some connections might
2281 be slow. On the other hand letting the @code{aio_*} function notify the
2282 caller might also be not the best solution since whenever the process
2283 works on preparing data for on client it makes no sense to be
2284 interrupted by a notification since the new client will not be handled
2285 before the current client is served. For situations like this
2286 @code{aio_suspend} should be used.
2290 @deftypefun int aio_suspend (const struct aiocb *const @var{list}[], int @var{nent}, const struct timespec *@var{timeout})
2291 When calling this function the calling thread is suspended until at
2292 least one of the requests pointed to by the @var{nent} elements of the
2293 array @var{list} has completed. If any of the requests already has
2294 completed at the time @code{aio_suspend} is called the function returns
2295 immediately. Whether a request has terminated or not is done by
2296 comparing the error status of the request with @code{EINPROGRESS}. If
2297 an element of @var{list} is @code{NULL} the entry is simply ignored.
2299 If no request has finished the calling process is suspended. If
2300 @var{timeout} is @code{NULL} the process is not waked until a request
2301 finished. If @var{timeout} is not @code{NULL} the process remains
2302 suspended at as long as specified in @var{timeout}. In this case
2303 @code{aio_suspend} returns with an error.
2305 The return value of the function is @math{0} if one or more requests
2306 from the @var{list} have terminated. Otherwise the function returns
2307 @math{-1} and @code{errno} is set to one of the following values:
2311 None of the requests from the @var{list} completed in the time specified
2314 A signal interrupted the @code{aio_suspend} function. This signal might
2315 also be sent by the AIO implementation while signalling the termination
2316 of one of the requests.
2318 The @code{aio_suspend} function is not implemented.
2321 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
2322 function is in fact @code{aio_suspend64} since the LFS interface
2323 transparently replaces the normal implementation.
2328 @deftypefun int aio_suspend64 (const struct aiocb64 *const @var{list}[], int @var{nent}, const struct timespec *@var{timeout})
2329 This function is similar to @code{aio_suspend} with the only difference
2330 that the argument is a reference to a variable of type @code{struct
2333 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
2334 function is available under the name @code{aio_suspend} and so
2335 transparently replaces the interface for small files on 32 bit
2339 @node Cancel AIO Operations
2340 @subsection Cancellation of AIO Operations
2342 When one or more requests are asynchronously processed it might be
2343 useful in some situations to cancel a selected operation, e.g., if it
2344 becomes obvious that the written data is not anymore accurate and would
2345 have to be overwritten soon. As an example assume an application, which
2346 writes data in files in a situation where new incoming data would have
2347 to be written in a file which will be updated by an enqueued request.
2348 The POSIX AIO implementation provides such a function but this function
2349 is not capable to force the cancellation of the request. It is up to the
2350 implementation to decide whether it is possible to cancel the operation
2351 or not. Therefore using this function is merely a hint.
2355 @deftypefun int aio_cancel (int @var{fildes}, struct aiocb *@var{aiocbp})
2356 The @code{aio_cancel} function can be used to cancel one or more
2357 outstanding requests. If the @var{aiocbp} parameter is @code{NULL} the
2358 function tries to cancel all outstanding requests which would process
2359 the file descriptor @var{fildes} (i.e.,, whose @code{aio_fildes} member
2360 is @var{fildes}). If @var{aiocbp} is not @code{NULL} the very specific
2361 request pointed to by @var{aiocbp} is tried to be cancelled.
2363 For requests which were successfully cancelled the normal notification
2364 about the termination of the request should take place. I.e., depending
2365 on the @code{struct sigevent} object which controls this, nothing
2366 happens, a signal is sent or a thread is started. If the request cannot
2367 be cancelled it terminates the usual way after performing te operation.
2369 After a request is successfully cancelled a call to @code{aio_error} with
2370 a reference to this request as the parameter will return
2371 @code{ECANCELED} and a call to @code{aio_return} will return @math{-1}.
2372 If the request wasn't cancelled and is still running the error status is
2373 still @code{EINPROGRESS}.
2375 The return value of the function is @code{AIO_CANCELED} if there were
2376 requests which haven't terminated and which successfully were cancelled.
2377 If there is one or more request left which couldn't be cancelled the
2378 return value is @code{AIO_NOTCANCELED}. In this case @code{aio_error}
2379 must be used to find out which of the perhaps multiple requests (in
2380 @var{aiocbp} is @code{NULL}) wasn't successfully cancelled. If all
2381 requests already terminated at the time @code{aio_cancel} is called the
2382 return value is @code{AIO_ALLDONE}.
2384 If an error occurred during the execution of @code{aio_cancel} the
2385 function returns @math{-1} and sets @code{errno} to one of the following
2390 The file descriptor @var{fildes} is not valid.
2392 @code{aio_cancel} is not implemented.
2395 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
2396 function is in fact @code{aio_cancel64} since the LFS interface
2397 transparently replaces the normal implementation.
2402 @deftypefun int aio_cancel64 (int @var{fildes}, struct aiocb *@var{aiocbp})
2403 This function is similar to @code{aio_cancel} with the only difference
2404 that the argument is a reference to a variable of type @code{struct
2407 When the sources are compiled with @code{_FILE_OFFSET_BITS == 64} this
2408 function is available under the name @code{aio_cancel} and so
2409 transparently replaces the interface for small files on 32 bit
2413 @node Configuration of AIO
2414 @subsection How to optimize the AIO implementation
2416 The POSIX standard does not specify how the AIO functions are
2417 implemented. They could be system calls but it is also possible to
2418 emulate them at userlevel.
2420 At least the available implementation at the point of this writing is a
2421 userlevel implementation which uses threads for handling the enqueued
2422 requests. This implementation requires to make some decisions about
2423 limitations but hard limitations are something which better should be
2424 avoided the GNU C library implementation provides a mean to tune the AIO
2425 implementation individually for each use.
2429 @deftp {Data Type} {struct aioinit}
2430 This data type is used to pass the configuration or tunable parameters
2431 to the implementation. The program has to initialize the members of
2432 this struct and pass it to the implementation using the @code{aio_init}
2436 @item int aio_threads
2437 This member specifies the maximal number of threads which must be used
2440 This number provides an estimate on the maximal number of simultaneously
2444 @item int aio_usedba
2448 @item int aio_numusers
2450 @item int aio_reserved[2]
2457 @deftypefun void aio_init (const struct aioinit *@var{init})
2458 This function must be called before any other AIO function. Calling it
2459 is completely voluntarily since it only is meant to help the AIO
2460 implementation to perform better.
2462 Before calling the @code{aio_init} function the members of a variable of
2463 type @code{struct aioinit} must be initialized. Then a reference to
2464 this variable is passed as the parameter to @code{aio_init} which itself
2465 may or may not pay attention to the hints.
2467 The function has no return value and no error cases are defined. It is
2468 a extension which follows a proposal from the SGI implementation in
2469 @w{Irix 6}. It is not covered by POSIX.1b or Unix98.
2472 @node Control Operations
2473 @section Control Operations on Files
2475 @cindex control operations on files
2476 @cindex @code{fcntl} function
2477 This section describes how you can perform various other operations on
2478 file descriptors, such as inquiring about or setting flags describing
2479 the status of the file descriptor, manipulating record locks, and the
2480 like. All of these operations are performed by the function @code{fcntl}.
2482 The second argument to the @code{fcntl} function is a command that
2483 specifies which operation to perform. The function and macros that name
2484 various flags that are used with it are declared in the header file
2485 @file{fcntl.h}. Many of these flags are also used by the @code{open}
2486 function; see @ref{Opening and Closing Files}.
2491 @deftypefun int fcntl (int @var{filedes}, int @var{command}, @dots{})
2492 The @code{fcntl} function performs the operation specified by
2493 @var{command} on the file descriptor @var{filedes}. Some commands
2494 require additional arguments to be supplied. These additional arguments
2495 and the return value and error conditions are given in the detailed
2496 descriptions of the individual commands.
2498 Briefly, here is a list of what the various commands are.
2502 Duplicate the file descriptor (return another file descriptor pointing
2503 to the same open file). @xref{Duplicating Descriptors}.
2506 Get flags associated with the file descriptor. @xref{Descriptor Flags}.
2509 Set flags associated with the file descriptor. @xref{Descriptor Flags}.
2512 Get flags associated with the open file. @xref{File Status Flags}.
2515 Set flags associated with the open file. @xref{File Status Flags}.
2518 Get a file lock. @xref{File Locks}.
2521 Set or clear a file lock. @xref{File Locks}.
2524 Like @code{F_SETLK}, but wait for completion. @xref{File Locks}.
2527 Get process or process group ID to receive @code{SIGIO} signals.
2528 @xref{Interrupt Input}.
2531 Set process or process group ID to receive @code{SIGIO} signals.
2532 @xref{Interrupt Input}.
2535 This function is a cancellation point in multi-threaded programs. This
2536 is a problem if the thread allocates some resources (like memory, file
2537 descriptors, semaphores or whatever) at the time @code{fcntl} is
2538 called. If the thread gets cancelled these resources stay allocated
2539 until the program ends. To avoid this calls to @code{fcntl} should be
2540 protected using cancellation handlers.
2541 @c ref pthread_cleanup_push / pthread_cleanup_pop
2545 @node Duplicating Descriptors
2546 @section Duplicating Descriptors
2548 @cindex duplicating file descriptors
2549 @cindex redirecting input and output
2551 You can @dfn{duplicate} a file descriptor, or allocate another file
2552 descriptor that refers to the same open file as the original. Duplicate
2553 descriptors share one file position and one set of file status flags
2554 (@pxref{File Status Flags}), but each has its own set of file descriptor
2555 flags (@pxref{Descriptor Flags}).
2557 The major use of duplicating a file descriptor is to implement
2558 @dfn{redirection} of input or output: that is, to change the
2559 file or pipe that a particular file descriptor corresponds to.
2561 You can perform this operation using the @code{fcntl} function with the
2562 @code{F_DUPFD} command, but there are also convenient functions
2563 @code{dup} and @code{dup2} for duplicating descriptors.
2567 The @code{fcntl} function and flags are declared in @file{fcntl.h},
2568 while prototypes for @code{dup} and @code{dup2} are in the header file
2573 @deftypefun int dup (int @var{old})
2574 This function copies descriptor @var{old} to the first available
2575 descriptor number (the first number not currently open). It is
2576 equivalent to @code{fcntl (@var{old}, F_DUPFD, 0)}.
2581 @deftypefun int dup2 (int @var{old}, int @var{new})
2582 This function copies the descriptor @var{old} to descriptor number
2585 If @var{old} is an invalid descriptor, then @code{dup2} does nothing; it
2586 does not close @var{new}. Otherwise, the new duplicate of @var{old}
2587 replaces any previous meaning of descriptor @var{new}, as if @var{new}
2590 If @var{old} and @var{new} are different numbers, and @var{old} is a
2591 valid descriptor number, then @code{dup2} is equivalent to:
2595 fcntl (@var{old}, F_DUPFD, @var{new})
2598 However, @code{dup2} does this atomically; there is no instant in the
2599 middle of calling @code{dup2} at which @var{new} is closed and not yet a
2600 duplicate of @var{old}.
2605 @deftypevr Macro int F_DUPFD
2606 This macro is used as the @var{command} argument to @code{fcntl}, to
2607 copy the file descriptor given as the first argument.
2609 The form of the call in this case is:
2612 fcntl (@var{old}, F_DUPFD, @var{next-filedes})
2615 The @var{next-filedes} argument is of type @code{int} and specifies that
2616 the file descriptor returned should be the next available one greater
2617 than or equal to this value.
2619 The return value from @code{fcntl} with this command is normally the value
2620 of the new file descriptor. A return value of @math{-1} indicates an
2621 error. The following @code{errno} error conditions are defined for
2626 The @var{old} argument is invalid.
2629 The @var{next-filedes} argument is invalid.
2632 There are no more file descriptors available---your program is already
2633 using the maximum. In BSD and GNU, the maximum is controlled by a
2634 resource limit that can be changed; @pxref{Limits on Resources}, for
2635 more information about the @code{RLIMIT_NOFILE} limit.
2638 @code{ENFILE} is not a possible error code for @code{dup2} because
2639 @code{dup2} does not create a new opening of a file; duplicate
2640 descriptors do not count toward the limit which @code{ENFILE}
2641 indicates. @code{EMFILE} is possible because it refers to the limit on
2642 distinct descriptor numbers in use in one process.
2645 Here is an example showing how to use @code{dup2} to do redirection.
2646 Typically, redirection of the standard streams (like @code{stdin}) is
2647 done by a shell or shell-like program before calling one of the
2648 @code{exec} functions (@pxref{Executing a File}) to execute a new
2649 program in a child process. When the new program is executed, it
2650 creates and initializes the standard streams to point to the
2651 corresponding file descriptors, before its @code{main} function is
2654 So, to redirect standard input to a file, the shell could do something
2665 file = TEMP_FAILURE_RETRY (open (filename, O_RDONLY));
2666 dup2 (file, STDIN_FILENO);
2667 TEMP_FAILURE_RETRY (close (file));
2668 execv (program, NULL);
2672 There is also a more detailed example showing how to implement redirection
2673 in the context of a pipeline of processes in @ref{Launching Jobs}.
2676 @node Descriptor Flags
2677 @section File Descriptor Flags
2678 @cindex file descriptor flags
2680 @dfn{File descriptor flags} are miscellaneous attributes of a file
2681 descriptor. These flags are associated with particular file
2682 descriptors, so that if you have created duplicate file descriptors
2683 from a single opening of a file, each descriptor has its own set of flags.
2685 Currently there is just one file descriptor flag: @code{FD_CLOEXEC},
2686 which causes the descriptor to be closed if you use any of the
2687 @code{exec@dots{}} functions (@pxref{Executing a File}).
2689 The symbols in this section are defined in the header file
2695 @deftypevr Macro int F_GETFD
2696 This macro is used as the @var{command} argument to @code{fcntl}, to
2697 specify that it should return the file descriptor flags associated
2698 with the @var{filedes} argument.
2700 The normal return value from @code{fcntl} with this command is a
2701 nonnegative number which can be interpreted as the bitwise OR of the
2702 individual flags (except that currently there is only one flag to use).
2704 In case of an error, @code{fcntl} returns @math{-1}. The following
2705 @code{errno} error conditions are defined for this command:
2709 The @var{filedes} argument is invalid.
2716 @deftypevr Macro int F_SETFD
2717 This macro is used as the @var{command} argument to @code{fcntl}, to
2718 specify that it should set the file descriptor flags associated with the
2719 @var{filedes} argument. This requires a third @code{int} argument to
2720 specify the new flags, so the form of the call is:
2723 fcntl (@var{filedes}, F_SETFD, @var{new-flags})
2726 The normal return value from @code{fcntl} with this command is an
2727 unspecified value other than @math{-1}, which indicates an error.
2728 The flags and error conditions are the same as for the @code{F_GETFD}
2732 The following macro is defined for use as a file descriptor flag with
2733 the @code{fcntl} function. The value is an integer constant usable
2734 as a bit mask value.
2738 @deftypevr Macro int FD_CLOEXEC
2739 @cindex close-on-exec (file descriptor flag)
2740 This flag specifies that the file descriptor should be closed when
2741 an @code{exec} function is invoked; see @ref{Executing a File}. When
2742 a file descriptor is allocated (as with @code{open} or @code{dup}),
2743 this bit is initially cleared on the new file descriptor, meaning that
2744 descriptor will survive into the new program after @code{exec}.
2747 If you want to modify the file descriptor flags, you should get the
2748 current flags with @code{F_GETFD} and modify the value. Don't assume
2749 that the flags listed here are the only ones that are implemented; your
2750 program may be run years from now and more flags may exist then. For
2751 example, here is a function to set or clear the flag @code{FD_CLOEXEC}
2752 without altering any other flags:
2755 /* @r{Set the @code{FD_CLOEXEC} flag of @var{desc} if @var{value} is nonzero,}
2756 @r{or clear the flag if @var{value} is 0.}
2757 @r{Return 0 on success, or -1 on error with @code{errno} set.} */
2760 set_cloexec_flag (int desc, int value)
2762 int oldflags = fcntl (desc, F_GETFD, 0);
2763 /* @r{If reading the flags failed, return error indication now.}
2766 /* @r{Set just the flag we want to set.} */
2768 oldflags |= FD_CLOEXEC;
2770 oldflags &= ~FD_CLOEXEC;
2771 /* @r{Store modified flag word in the descriptor.} */
2772 return fcntl (desc, F_SETFD, oldflags);
2776 @node File Status Flags
2777 @section File Status Flags
2778 @cindex file status flags
2780 @dfn{File status flags} are used to specify attributes of the opening of a
2781 file. Unlike the file descriptor flags discussed in @ref{Descriptor
2782 Flags}, the file status flags are shared by duplicated file descriptors
2783 resulting from a single opening of the file. The file status flags are
2784 specified with the @var{flags} argument to @code{open};
2785 @pxref{Opening and Closing Files}.
2787 File status flags fall into three categories, which are described in the
2792 @ref{Access Modes}, specify what type of access is allowed to the
2793 file: reading, writing, or both. They are set by @code{open} and are
2794 returned by @code{fcntl}, but cannot be changed.
2797 @ref{Open-time Flags}, control details of what @code{open} will do.
2798 These flags are not preserved after the @code{open} call.
2801 @ref{Operating Modes}, affect how operations such as @code{read} and
2802 @code{write} are done. They are set by @code{open}, and can be fetched or
2803 changed with @code{fcntl}.
2806 The symbols in this section are defined in the header file
2811 * Access Modes:: Whether the descriptor can read or write.
2812 * Open-time Flags:: Details of @code{open}.
2813 * Operating Modes:: Special modes to control I/O operations.
2814 * Getting File Status Flags:: Fetching and changing these flags.
2818 @subsection File Access Modes
2820 The file access modes allow a file descriptor to be used for reading,
2821 writing, or both. (In the GNU system, they can also allow none of these,
2822 and allow execution of the file as a program.) The access modes are chosen
2823 when the file is opened, and never change.
2827 @deftypevr Macro int O_RDONLY
2828 Open the file for read access.
2833 @deftypevr Macro int O_WRONLY
2834 Open the file for write access.
2839 @deftypevr Macro int O_RDWR
2840 Open the file for both reading and writing.
2843 In the GNU system (and not in other systems), @code{O_RDONLY} and
2844 @code{O_WRONLY} are independent bits that can be bitwise-ORed together,
2845 and it is valid for either bit to be set or clear. This means that
2846 @code{O_RDWR} is the same as @code{O_RDONLY|O_WRONLY}. A file access
2847 mode of zero is permissible; it allows no operations that do input or
2848 output to the file, but does allow other operations such as
2849 @code{fchmod}. On the GNU system, since ``read-only'' or ``write-only''
2850 is a misnomer, @file{fcntl.h} defines additional names for the file
2851 access modes. These names are preferred when writing GNU-specific code.
2852 But most programs will want to be portable to other POSIX.1 systems and
2853 should use the POSIX.1 names above instead.
2857 @deftypevr Macro int O_READ
2858 Open the file for reading. Same as @code{O_RDWR}; only defined on GNU.
2863 @deftypevr Macro int O_WRITE
2864 Open the file for reading. Same as @code{O_WRONLY}; only defined on GNU.
2869 @deftypevr Macro int O_EXEC
2870 Open the file for executing. Only defined on GNU.
2873 To determine the file access mode with @code{fcntl}, you must extract
2874 the access mode bits from the retrieved file status flags. In the GNU
2875 system, you can just test the @code{O_READ} and @code{O_WRITE} bits in
2876 the flags word. But in other POSIX.1 systems, reading and writing
2877 access modes are not stored as distinct bit flags. The portable way to
2878 extract the file access mode bits is with @code{O_ACCMODE}.
2882 @deftypevr Macro int O_ACCMODE
2883 This macro stands for a mask that can be bitwise-ANDed with the file
2884 status flag value to produce a value representing the file access mode.
2885 The mode will be @code{O_RDONLY}, @code{O_WRONLY}, or @code{O_RDWR}.
2886 (In the GNU system it could also be zero, and it never includes the
2890 @node Open-time Flags
2891 @subsection Open-time Flags
2893 The open-time flags specify options affecting how @code{open} will behave.
2894 These options are not preserved once the file is open. The exception to
2895 this is @code{O_NONBLOCK}, which is also an I/O operating mode and so it
2896 @emph{is} saved. @xref{Opening and Closing Files}, for how to call
2899 There are two sorts of options specified by open-time flags.
2903 @dfn{File name translation flags} affect how @code{open} looks up the
2904 file name to locate the file, and whether the file can be created.
2905 @cindex file name translation flags
2906 @cindex flags, file name translation
2909 @dfn{Open-time action flags} specify extra operations that @code{open} will
2910 perform on the file once it is open.
2911 @cindex open-time action flags
2912 @cindex flags, open-time action
2915 Here are the file name translation flags.
2919 @deftypevr Macro int O_CREAT
2920 If set, the file will be created if it doesn't already exist.
2921 @c !!! mode arg, umask
2922 @cindex create on open (file status flag)
2927 @deftypevr Macro int O_EXCL
2928 If both @code{O_CREAT} and @code{O_EXCL} are set, then @code{open} fails
2929 if the specified file already exists. This is guaranteed to never
2930 clobber an existing file.
2935 @deftypevr Macro int O_NONBLOCK
2936 @cindex non-blocking open
2937 This prevents @code{open} from blocking for a ``long time'' to open the
2938 file. This is only meaningful for some kinds of files, usually devices
2939 such as serial ports; when it is not meaningful, it is harmless and
2940 ignored. Often opening a port to a modem blocks until the modem reports
2941 carrier detection; if @code{O_NONBLOCK} is specified, @code{open} will
2942 return immediately without a carrier.
2944 Note that the @code{O_NONBLOCK} flag is overloaded as both an I/O operating
2945 mode and a file name translation flag. This means that specifying
2946 @code{O_NONBLOCK} in @code{open} also sets nonblocking I/O mode;
2947 @pxref{Operating Modes}. To open the file without blocking but do normal
2948 I/O that blocks, you must call @code{open} with @code{O_NONBLOCK} set and
2949 then call @code{fcntl} to turn the bit off.
2954 @deftypevr Macro int O_NOCTTY
2955 If the named file is a terminal device, don't make it the controlling
2956 terminal for the process. @xref{Job Control}, for information about
2957 what it means to be the controlling terminal.
2959 In the GNU system and 4.4 BSD, opening a file never makes it the
2960 controlling terminal and @code{O_NOCTTY} is zero. However, other
2961 systems may use a nonzero value for @code{O_NOCTTY} and set the
2962 controlling terminal when you open a file that is a terminal device; so
2963 to be portable, use @code{O_NOCTTY} when it is important to avoid this.
2964 @cindex controlling terminal, setting
2967 The following three file name translation flags exist only in the GNU system.
2971 @deftypevr Macro int O_IGNORE_CTTY
2972 Do not recognize the named file as the controlling terminal, even if it
2973 refers to the process's existing controlling terminal device. Operations
2974 on the new file descriptor will never induce job control signals.
2980 @deftypevr Macro int O_NOLINK
2981 If the named file is a symbolic link, open the link itself instead of
2982 the file it refers to. (@code{fstat} on the new file descriptor will
2983 return the information returned by @code{lstat} on the link's name.)
2984 @cindex symbolic link, opening
2989 @deftypevr Macro int O_NOTRANS
2990 If the named file is specially translated, do not invoke the translator.
2991 Open the bare file the translator itself sees.
2995 The open-time action flags tell @code{open} to do additional operations
2996 which are not really related to opening the file. The reason to do them
2997 as part of @code{open} instead of in separate calls is that @code{open}
2998 can do them @i{atomically}.
3002 @deftypevr Macro int O_TRUNC
3003 Truncate the file to zero length. This option is only useful for
3004 regular files, not special files such as directories or FIFOs. POSIX.1
3005 requires that you open the file for writing to use @code{O_TRUNC}. In
3006 BSD and GNU you must have permission to write the file to truncate it,
3007 but you need not open for write access.
3009 This is the only open-time action flag specified by POSIX.1. There is
3010 no good reason for truncation to be done by @code{open}, instead of by
3011 calling @code{ftruncate} afterwards. The @code{O_TRUNC} flag existed in
3012 Unix before @code{ftruncate} was invented, and is retained for backward
3016 The remaining operating modes are BSD extensions. They exist only
3017 on some systems. On other systems, these macros are not defined.
3021 @deftypevr Macro int O_SHLOCK
3022 Acquire a shared lock on the file, as with @code{flock}.
3025 If @code{O_CREAT} is specified, the locking is done atomically when
3026 creating the file. You are guaranteed that no other process will get
3027 the lock on the new file first.
3032 @deftypevr Macro int O_EXLOCK
3033 Acquire an exclusive lock on the file, as with @code{flock}.
3034 @xref{File Locks}. This is atomic like @code{O_SHLOCK}.
3037 @node Operating Modes
3038 @subsection I/O Operating Modes
3040 The operating modes affect how input and output operations using a file
3041 descriptor work. These flags are set by @code{open} and can be fetched
3042 and changed with @code{fcntl}.
3046 @deftypevr Macro int O_APPEND
3047 The bit that enables append mode for the file. If set, then all
3048 @code{write} operations write the data at the end of the file, extending
3049 it, regardless of the current file position. This is the only reliable
3050 way to append to a file. In append mode, you are guaranteed that the
3051 data you write will always go to the current end of the file, regardless
3052 of other processes writing to the file. Conversely, if you simply set
3053 the file position to the end of file and write, then another process can
3054 extend the file after you set the file position but before you write,
3055 resulting in your data appearing someplace before the real end of file.
3060 @deftypevr Macro int O_NONBLOCK
3061 The bit that enables nonblocking mode for the file. If this bit is set,
3062 @code{read} requests on the file can return immediately with a failure
3063 status if there is no input immediately available, instead of blocking.
3064 Likewise, @code{write} requests can also return immediately with a
3065 failure status if the output can't be written immediately.
3067 Note that the @code{O_NONBLOCK} flag is overloaded as both an I/O
3068 operating mode and a file name translation flag; @pxref{Open-time Flags}.
3073 @deftypevr Macro int O_NDELAY
3074 This is an obsolete name for @code{O_NONBLOCK}, provided for
3075 compatibility with BSD. It is not defined by the POSIX.1 standard.
3078 The remaining operating modes are BSD and GNU extensions. They exist only
3079 on some systems. On other systems, these macros are not defined.
3083 @deftypevr Macro int O_ASYNC
3084 The bit that enables asynchronous input mode. If set, then @code{SIGIO}
3085 signals will be generated when input is available. @xref{Interrupt Input}.
3087 Asynchronous input mode is a BSD feature.
3092 @deftypevr Macro int O_FSYNC
3093 The bit that enables synchronous writing for the file. If set, each
3094 @code{write} call will make sure the data is reliably stored on disk before
3095 returning. @c !!! xref fsync
3097 Synchronous writing is a BSD feature.
3102 @deftypevr Macro int O_SYNC
3103 This is another name for @code{O_FSYNC}. They have the same value.
3108 @deftypevr Macro int O_NOATIME
3109 If this bit is set, @code{read} will not update the access time of the
3110 file. @xref{File Times}. This is used by programs that do backups, so
3111 that backing a file up does not count as reading it.
3112 Only the owner of the file or the superuser may use this bit.
3114 This is a GNU extension.
3117 @node Getting File Status Flags
3118 @subsection Getting and Setting File Status Flags
3120 The @code{fcntl} function can fetch or change file status flags.
3124 @deftypevr Macro int F_GETFL
3125 This macro is used as the @var{command} argument to @code{fcntl}, to
3126 read the file status flags for the open file with descriptor
3129 The normal return value from @code{fcntl} with this command is a
3130 nonnegative number which can be interpreted as the bitwise OR of the
3131 individual flags. Since the file access modes are not single-bit values,
3132 you can mask off other bits in the returned flags with @code{O_ACCMODE}
3135 In case of an error, @code{fcntl} returns @math{-1}. The following
3136 @code{errno} error conditions are defined for this command:
3140 The @var{filedes} argument is invalid.
3146 @deftypevr Macro int F_SETFL
3147 This macro is used as the @var{command} argument to @code{fcntl}, to set
3148 the file status flags for the open file corresponding to the
3149 @var{filedes} argument. This command requires a third @code{int}
3150 argument to specify the new flags, so the call looks like this:
3153 fcntl (@var{filedes}, F_SETFL, @var{new-flags})
3156 You can't change the access mode for the file in this way; that is,
3157 whether the file descriptor was opened for reading or writing.
3159 The normal return value from @code{fcntl} with this command is an
3160 unspecified value other than @math{-1}, which indicates an error. The
3161 error conditions are the same as for the @code{F_GETFL} command.
3164 If you want to modify the file status flags, you should get the current
3165 flags with @code{F_GETFL} and modify the value. Don't assume that the
3166 flags listed here are the only ones that are implemented; your program
3167 may be run years from now and more flags may exist then. For example,
3168 here is a function to set or clear the flag @code{O_NONBLOCK} without
3169 altering any other flags:
3173 /* @r{Set the @code{O_NONBLOCK} flag of @var{desc} if @var{value} is nonzero,}
3174 @r{or clear the flag if @var{value} is 0.}
3175 @r{Return 0 on success, or -1 on error with @code{errno} set.} */
3178 set_nonblock_flag (int desc, int value)
3180 int oldflags = fcntl (desc, F_GETFL, 0);
3181 /* @r{If reading the flags failed, return error indication now.} */
3184 /* @r{Set just the flag we want to set.} */
3186 oldflags |= O_NONBLOCK;
3188 oldflags &= ~O_NONBLOCK;
3189 /* @r{Store modified flag word in the descriptor.} */
3190 return fcntl (desc, F_SETFL, oldflags);
3199 @cindex record locking
3200 The remaining @code{fcntl} commands are used to support @dfn{record
3201 locking}, which permits multiple cooperating programs to prevent each
3202 other from simultaneously accessing parts of a file in error-prone
3205 @cindex exclusive lock
3207 An @dfn{exclusive} or @dfn{write} lock gives a process exclusive access
3208 for writing to the specified part of the file. While a write lock is in
3209 place, no other process can lock that part of the file.
3213 A @dfn{shared} or @dfn{read} lock prohibits any other process from
3214 requesting a write lock on the specified part of the file. However,
3215 other processes can request read locks.
3217 The @code{read} and @code{write} functions do not actually check to see
3218 whether there are any locks in place. If you want to implement a
3219 locking protocol for a file shared by multiple processes, your application
3220 must do explicit @code{fcntl} calls to request and clear locks at the
3223 Locks are associated with processes. A process can only have one kind
3224 of lock set for each byte of a given file. When any file descriptor for
3225 that file is closed by the process, all of the locks that process holds
3226 on that file are released, even if the locks were made using other
3227 descriptors that remain open. Likewise, locks are released when a
3228 process exits, and are not inherited by child processes created using
3229 @code{fork} (@pxref{Creating a Process}).
3231 When making a lock, use a @code{struct flock} to specify what kind of
3232 lock and where. This data type and the associated macros for the
3233 @code{fcntl} function are declared in the header file @file{fcntl.h}.
3238 @deftp {Data Type} {struct flock}
3239 This structure is used with the @code{fcntl} function to describe a file
3240 lock. It has these members:
3243 @item short int l_type
3244 Specifies the type of the lock; one of @code{F_RDLCK}, @code{F_WRLCK}, or
3247 @item short int l_whence
3248 This corresponds to the @var{whence} argument to @code{fseek} or
3249 @code{lseek}, and specifies what the offset is relative to. Its value
3250 can be one of @code{SEEK_SET}, @code{SEEK_CUR}, or @code{SEEK_END}.
3253 This specifies the offset of the start of the region to which the lock
3254 applies, and is given in bytes relative to the point specified by
3255 @code{l_whence} member.
3258 This specifies the length of the region to be locked. A value of
3259 @code{0} is treated specially; it means the region extends to the end of
3263 This field is the process ID (@pxref{Process Creation Concepts}) of the
3264 process holding the lock. It is filled in by calling @code{fcntl} with
3265 the @code{F_GETLK} command, but is ignored when making a lock.
3271 @deftypevr Macro int F_GETLK
3272 This macro is used as the @var{command} argument to @code{fcntl}, to
3273 specify that it should get information about a lock. This command
3274 requires a third argument of type @w{@code{struct flock *}} to be passed
3275 to @code{fcntl}, so that the form of the call is:
3278 fcntl (@var{filedes}, F_GETLK, @var{lockp})
3281 If there is a lock already in place that would block the lock described
3282 by the @var{lockp} argument, information about that lock overwrites
3283 @code{*@var{lockp}}. Existing locks are not reported if they are
3284 compatible with making a new lock as specified. Thus, you should
3285 specify a lock type of @code{F_WRLCK} if you want to find out about both
3286 read and write locks, or @code{F_RDLCK} if you want to find out about
3289 There might be more than one lock affecting the region specified by the
3290 @var{lockp} argument, but @code{fcntl} only returns information about
3291 one of them. The @code{l_whence} member of the @var{lockp} structure is
3292 set to @code{SEEK_SET} and the @code{l_start} and @code{l_len} fields
3293 set to identify the locked region.
3295 If no lock applies, the only change to the @var{lockp} structure is to
3296 update the @code{l_type} to a value of @code{F_UNLCK}.
3298 The normal return value from @code{fcntl} with this command is an
3299 unspecified value other than @math{-1}, which is reserved to indicate an
3300 error. The following @code{errno} error conditions are defined for
3305 The @var{filedes} argument is invalid.
3308 Either the @var{lockp} argument doesn't specify valid lock information,
3309 or the file associated with @var{filedes} doesn't support locks.
3315 @deftypevr Macro int F_SETLK
3316 This macro is used as the @var{command} argument to @code{fcntl}, to
3317 specify that it should set or clear a lock. This command requires a
3318 third argument of type @w{@code{struct flock *}} to be passed to
3319 @code{fcntl}, so that the form of the call is:
3322 fcntl (@var{filedes}, F_SETLK, @var{lockp})
3325 If the process already has a lock on any part of the region, the old lock
3326 on that part is replaced with the new lock. You can remove a lock
3327 by specifying a lock type of @code{F_UNLCK}.
3329 If the lock cannot be set, @code{fcntl} returns immediately with a value
3330 of @math{-1}. This function does not block waiting for other processes
3331 to release locks. If @code{fcntl} succeeds, it return a value other
3334 The following @code{errno} error conditions are defined for this
3340 The lock cannot be set because it is blocked by an existing lock on the
3341 file. Some systems use @code{EAGAIN} in this case, and other systems
3342 use @code{EACCES}; your program should treat them alike, after
3343 @code{F_SETLK}. (The GNU system always uses @code{EAGAIN}.)
3346 Either: the @var{filedes} argument is invalid; you requested a read lock
3347 but the @var{filedes} is not open for read access; or, you requested a
3348 write lock but the @var{filedes} is not open for write access.
3351 Either the @var{lockp} argument doesn't specify valid lock information,
3352 or the file associated with @var{filedes} doesn't support locks.
3355 The system has run out of file lock resources; there are already too
3356 many file locks in place.
3358 Well-designed file systems never report this error, because they have no
3359 limitation on the number of locks. However, you must still take account
3360 of the possibility of this error, as it could result from network access
3361 to a file system on another machine.
3367 @deftypevr Macro int F_SETLKW
3368 This macro is used as the @var{command} argument to @code{fcntl}, to
3369 specify that it should set or clear a lock. It is just like the
3370 @code{F_SETLK} command, but causes the process to block (or wait)
3371 until the request can be specified.
3373 This command requires a third argument of type @code{struct flock *}, as
3374 for the @code{F_SETLK} command.
3376 The @code{fcntl} return values and errors are the same as for the
3377 @code{F_SETLK} command, but these additional @code{errno} error conditions
3378 are defined for this command:
3382 The function was interrupted by a signal while it was waiting.
3383 @xref{Interrupted Primitives}.
3386 The specified region is being locked by another process. But that
3387 process is waiting to lock a region which the current process has
3388 locked, so waiting for the lock would result in deadlock. The system
3389 does not guarantee that it will detect all such conditions, but it lets
3390 you know if it notices one.
3395 The following macros are defined for use as values for the @code{l_type}
3396 member of the @code{flock} structure. The values are integer constants.
3403 This macro is used to specify a read (or shared) lock.
3409 This macro is used to specify a write (or exclusive) lock.
3415 This macro is used to specify that the region is unlocked.
3418 As an example of a situation where file locking is useful, consider a
3419 program that can be run simultaneously by several different users, that
3420 logs status information to a common file. One example of such a program
3421 might be a game that uses a file to keep track of high scores. Another
3422 example might be a program that records usage or accounting information
3423 for billing purposes.
3425 Having multiple copies of the program simultaneously writing to the
3426 file could cause the contents of the file to become mixed up. But
3427 you can prevent this kind of problem by setting a write lock on the
3428 file before actually writing to the file.
3430 If the program also needs to read the file and wants to make sure that
3431 the contents of the file are in a consistent state, then it can also use
3432 a read lock. While the read lock is set, no other process can lock
3433 that part of the file for writing.
3435 @c ??? This section could use an example program.
3437 Remember that file locks are only a @emph{voluntary} protocol for
3438 controlling access to a file. There is still potential for access to
3439 the file by programs that don't use the lock protocol.
3441 @node Interrupt Input
3442 @section Interrupt-Driven Input
3444 @cindex interrupt-driven input
3445 If you set the @code{O_ASYNC} status flag on a file descriptor
3446 (@pxref{File Status Flags}), a @code{SIGIO} signal is sent whenever
3447 input or output becomes possible on that file descriptor. The process
3448 or process group to receive the signal can be selected by using the
3449 @code{F_SETOWN} command to the @code{fcntl} function. If the file
3450 descriptor is a socket, this also selects the recipient of @code{SIGURG}
3451 signals that are delivered when out-of-band data arrives on that socket;
3452 see @ref{Out-of-Band Data}. (@code{SIGURG} is sent in any situation
3453 where @code{select} would report the socket as having an ``exceptional
3454 condition''. @xref{Waiting for I/O}.)
3456 If the file descriptor corresponds to a terminal device, then @code{SIGIO}
3457 signals are sent to the foreground process group of the terminal.
3461 The symbols in this section are defined in the header file
3466 @deftypevr Macro int F_GETOWN
3467 This macro is used as the @var{command} argument to @code{fcntl}, to
3468 specify that it should get information about the process or process
3469 group to which @code{SIGIO} signals are sent. (For a terminal, this is
3470 actually the foreground process group ID, which you can get using
3471 @code{tcgetpgrp}; see @ref{Terminal Access Functions}.)
3473 The return value is interpreted as a process ID; if negative, its
3474 absolute value is the process group ID.
3476 The following @code{errno} error condition is defined for this command:
3480 The @var{filedes} argument is invalid.
3486 @deftypevr Macro int F_SETOWN
3487 This macro is used as the @var{command} argument to @code{fcntl}, to
3488 specify that it should set the process or process group to which
3489 @code{SIGIO} signals are sent. This command requires a third argument
3490 of type @code{pid_t} to be passed to @code{fcntl}, so that the form of
3494 fcntl (@var{filedes}, F_SETOWN, @var{pid})
3497 The @var{pid} argument should be a process ID. You can also pass a
3498 negative number whose absolute value is a process group ID.
3500 The return value from @code{fcntl} with this command is @math{-1}
3501 in case of error and some other value if successful. The following
3502 @code{errno} error conditions are defined for this command:
3506 The @var{filedes} argument is invalid.
3509 There is no process or process group corresponding to @var{pid}.
3513 @c ??? This section could use an example program.
3516 @section Generic I/O Control operations
3517 @cindex generic i/o control operations
3520 The GNU system can handle most input/output operations on many different
3521 devices and objects in terms of a few file primitives - @code{read},
3522 @code{write} and @code{lseek}. However, most devices also have a few
3523 peculiar operations which do not fit into this model. Such as:
3528 Changing the character font used on a terminal.
3531 Telling a magnetic tape system to rewind or fast forward. (Since they
3532 cannot move in byte increments, @code{lseek} is inapplicable).
3535 Ejecting a disk from a drive.
3538 Playing an audio track from a CD-ROM drive.
3541 Maintaining routing tables for a network.
3545 Although some such objects such as sockets and terminals
3546 @footnote{Actually, the terminal-specific functions are implemented with
3547 IOCTLs on many platforms.} have special functions of their own, it would
3548 not be practical to create functions for all these cases.
3550 Instead these minor operations, known as @dfn{IOCTL}s, are assigned code
3551 numbers and multiplexed through the @code{ioctl} function, defined in
3552 @code{sys/ioctl.h}. The code numbers themselves are defined in many
3555 @deftypefun int ioctl (int @var{filedes}, int @var{command}, @dots{})
3557 The @code{ioctl} function performs the generic I/O operation
3558 @var{command} on @var{filedes}.
3560 A third argument is usually present, either a single number or a pointer
3561 to a structure. The meaning of this argument, the returned value, and
3562 any error codes depends upon the command used. Often @math{-1} is
3563 returned for a failure.
3567 On some systems, IOCTLs used by different devices share the same numbers.
3568 Thus, although use of an inappropriate IOCTL @emph{usually} only produces
3569 an error, you should not attempt to use device-specific IOCTLs on an
3572 Most IOCTLs are OS-specific and/or only used in special system utilities,
3573 and are thus beyond the scope of this document. For an example of the use
3574 of an IOCTL, see @ref{Out-of-Band Data}.