1 .\" This manpage is Copyright (C) 1992 Drew Eckhardt;
2 .\" and Copyright (C) 1993 Michael Haardt, Ian Jackson.
3 .\" and Copyright (C) 2008 Greg Banks
4 .\" and Copyright (C) 2006, 2008, 2013, 2014 Michael Kerrisk <mtk.manpages@gmail.com>
6 .\" %%%LICENSE_START(VERBATIM)
7 .\" Permission is granted to make and distribute verbatim copies of this
8 .\" manual provided the copyright notice and this permission notice are
9 .\" preserved on all copies.
11 .\" Permission is granted to copy and distribute modified versions of this
12 .\" manual under the conditions for verbatim copying, provided that the
13 .\" entire resulting derived work is distributed under the terms of a
14 .\" permission notice identical to this one.
16 .\" Since the Linux kernel and libraries are constantly changing, this
17 .\" manual page may be incorrect or out-of-date. The author(s) assume no
18 .\" responsibility for errors or omissions, or for damages resulting from
19 .\" the use of the information contained herein. The author(s) may not
20 .\" have taken the same level of care in the production of this manual,
21 .\" which is licensed free of charge, as they might when working
24 .\" Formatted or processed versions of this manual, if unaccompanied by
25 .\" the source, must acknowledge the copyright and authors of this work.
28 .\" Modified 1993-07-21 by Rik Faith <faith@cs.unc.edu>
29 .\" Modified 1994-08-21 by Michael Haardt
30 .\" Modified 1996-04-13 by Andries Brouwer <aeb@cwi.nl>
31 .\" Modified 1996-05-13 by Thomas Koenig
32 .\" Modified 1996-12-20 by Michael Haardt
33 .\" Modified 1999-02-19 by Andries Brouwer <aeb@cwi.nl>
34 .\" Modified 1998-11-28 by Joseph S. Myers <jsm28@hermes.cam.ac.uk>
35 .\" Modified 1999-06-03 by Michael Haardt
36 .\" Modified 2002-05-07 by Michael Kerrisk <mtk.manpages@gmail.com>
37 .\" Modified 2004-06-23 by Michael Kerrisk <mtk.manpages@gmail.com>
38 .\" 2004-12-08, mtk, reordered flags list alphabetically
39 .\" 2004-12-08, Martin Pool <mbp@sourcefrog.net> (& mtk), added O_NOATIME
40 .\" 2007-09-18, mtk, Added description of O_CLOEXEC + other minor edits
41 .\" 2008-01-03, mtk, with input from Trond Myklebust
42 .\" <trond.myklebust@fys.uio.no> and Timo Sirainen <tss@iki.fi>
43 .\" Rewrite description of O_EXCL.
44 .\" 2008-01-11, Greg Banks <gnb@melbourne.sgi.com>: add more detail
46 .\" 2008-02-26, Michael Haardt: Reorganized text for O_CREAT and mode
48 .\" FIXME . Apr 08: The next POSIX revision has O_EXEC, O_SEARCH, and
49 .\" O_TTYINIT. Eventually these may need to be documented. --mtk
51 .TH OPEN 2 2020-02-09 "Linux" "Linux Programmer's Manual"
53 open, openat, creat \- open and possibly create a file
56 .B #include <sys/types.h>
57 .B #include <sys/stat.h>
60 .BI "int open(const char *" pathname ", int " flags );
61 .BI "int open(const char *" pathname ", int " flags ", mode_t " mode );
63 .BI "int creat(const char *" pathname ", mode_t " mode );
65 .BI "int openat(int " dirfd ", const char *" pathname ", int " flags );
66 .BI "int openat(int " dirfd ", const char *" pathname ", int " flags \
69 /* Documented separately, in \fBopenat2\fP(2): */
70 .BI "int openat2(int " dirfd ", const char *" pathname ,
71 .BI " const struct open_how *" how ", size_t " size ");
75 Feature Test Macro Requirements for glibc (see
76 .BR feature_test_macros (7)):
85 _POSIX_C_SOURCE\ >=\ 200809L
95 system call opens the file specified by
97 If the specified file does not exist,
107 is a file descriptor, a small, nonnegative integer that is used
108 in subsequent system calls
109 .RB ( read "(2), " write "(2), " lseek "(2), " fcntl (2),
110 etc.) to refer to the open file.
111 The file descriptor returned by a successful call will be
112 the lowest-numbered file descriptor not currently open for the process.
114 By default, the new file descriptor is set to remain open across an
118 file descriptor flag described in
120 is initially disabled); the
122 flag, described below, can be used to change this default.
123 The file offset is set to the beginning of the file (see
129 .IR "open file description" ,
130 an entry in the system-wide table of open files.
131 The open file description records the file offset and the file status flags
133 A file descriptor is a reference to an open file description;
134 this reference is unaffected if
136 is subsequently removed or modified to refer to a different file.
137 For further details on open file descriptions, see NOTES.
141 must include one of the following
143 .BR O_RDONLY ", " O_WRONLY ", or " O_RDWR .
144 These request opening the file read-only, write-only, or read/write,
147 In addition, zero or more file creation flags and file status flags
153 .I file creation flags
166 are all of the remaining flags listed below.
167 .\" SUSv4 divides the flags into:
171 .\" * Other (O_CLOEXEC, O_DIRECTORY, O_NOFOLLOW)
172 .\" though it's not clear what the difference between "other" and
173 .\" "File creation" flags is. I raised an Aardvark to see if this
174 .\" can be clarified in SUSv4; 10 Oct 2008.
175 .\" http://thread.gmane.org/gmane.comp.standards.posix.austin.general/64/focus=67
176 .\" TC1 (balloted in 2013), resolved this, so that those three constants
177 .\" are also categorized" as file status flags.
179 The distinction between these two groups of flags is that
180 the file creation flags affect the semantics of the open operation itself,
181 while the file status flags affect the semantics of subsequent I/O operations.
182 The file status flags can be retrieved and (in some cases)
187 The full list of file creation flags and file status flags is as follows:
190 The file is opened in append mode.
193 the file offset is positioned at the end of the file,
196 The modification of the file offset and the write operation
197 are performed as a single atomic step.
200 may lead to corrupted files on NFS filesystems if more than one process
201 appends data to a file at once.
202 .\" For more background, see
203 .\" http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=453946
204 .\" http://nfs.sourceforge.net/
205 This is because NFS does not support
206 appending to a file, so the client kernel has to simulate it, which
207 can't be done without a race condition.
210 Enable signal-driven I/O:
213 by default, but this can be changed via
215 when input or output becomes possible on this file descriptor.
216 This feature is available only for terminals, pseudoterminals,
217 sockets, and (since Linux 2.6) pipes and FIFOs.
221 See also BUGS, below.
223 .BR O_CLOEXEC " (since Linux 2.6.23)"
224 .\" NOTE! several other man pages refer to this text
225 Enable the close-on-exec flag for the new file descriptor.
226 .\" FIXME . for later review when Issue 8 is one day released...
227 .\" POSIX proposes to fix many APIs that provide hidden FDs
228 .\" http://austingroupbugs.net/tag_view_page.php?tag_id=8
229 .\" http://austingroupbugs.net/view.php?id=368
230 Specifying this flag permits a program to avoid additional
233 operations to set the
237 Note that the use of this flag is essential in some multithreaded programs,
238 because using a separate
243 flag does not suffice to avoid race conditions
244 where one thread opens a file descriptor and
245 attempts to set its close-on-exec flag using
247 at the same time as another thread does a
251 Depending on the order of execution,
252 the race may lead to the file descriptor returned by
254 being unintentionally leaked to the program executed by the child process
257 (This kind of race is in principle possible for any system call
258 that creates a file descriptor whose close-on-exec flag should be set,
259 and various other Linux system calls provide an equivalent of the
261 flag to deal with this problem.)
262 .\" This flag fixes only one form of the race condition;
263 .\" The race can also occur with, for example, file descriptors
264 .\" returned by accept(), pipe(), etc.
269 does not exist, create it as a regular file.
271 The owner (user ID) of the new file is set to the effective user ID
274 The group ownership (group ID) of the new file is set either to
275 the effective group ID of the process (System V semantics)
276 or to the group ID of the parent directory (BSD semantics).
277 On Linux, the behavior depends on whether the
278 set-group-ID mode bit is set on the parent directory:
279 if that bit is set, then BSD semantics apply;
280 otherwise, System V semantics apply.
281 For some filesystems, the behavior also depends on the
285 mount options described in
287 .\" As at 2.6.25, bsdgroups is supported by ext2, ext3, ext4, and
288 .\" XFS (since 2.6.14).
293 argument specifies the file mode bits be applied when a new file is created.
294 This argument must be supplied when
307 The effective mode is modified by the process's
309 in the usual way: in the absence of a default ACL, the mode of the
311 .IR "(mode\ &\ ~umask)" .
312 Note that this mode applies only to future accesses of the
313 newly created file; the
315 call that creates a read-only file may well return a read/write
318 The following symbolic constants are provided for
322 00700 user (file owner) has read, write, and execute permission
325 00400 user has read permission
328 00200 user has write permission
331 00100 user has execute permission
334 00070 group has read, write, and execute permission
337 00040 group has read permission
340 00020 group has write permission
343 00010 group has execute permission
346 00007 others have read, write, and execute permission
349 00004 others have read permission
352 00002 others have write permission
355 00001 others have execute permission
358 According to POSIX, the effect when other bits are set in
361 On Linux, the following bits are also honored in
366 0004000 set-user-ID bit
369 0002000 set-group-ID bit (see
373 0001000 sticky bit (see
377 .BR O_DIRECT " (since Linux 2.4.10)"
378 Try to minimize cache effects of the I/O to and from this file.
379 In general this will degrade performance, but it is useful in
380 special situations, such as when applications do their own caching.
381 File I/O is done directly to/from user-space buffers.
384 flag on its own makes an effort to transfer data synchronously,
385 but does not give the guarantees of the
387 flag that data and necessary metadata are transferred.
388 To guarantee synchronous I/O,
390 must be used in addition to
392 See NOTES below for further discussion.
394 A semantically similar (but deprecated) interface for block devices
399 If \fIpathname\fP is not a directory, cause the open to fail.
400 .\" But see the following and its replies:
401 .\" http://marc.theaimsgroup.com/?t=112748702800001&r=1&w=2
402 .\" [PATCH] open: O_DIRECTORY and O_CREAT together should fail
403 .\" O_DIRECTORY | O_CREAT causes O_DIRECTORY to be ignored.
404 This flag was added in kernel version 2.1.126, to
405 avoid denial-of-service problems if
411 Write operations on the file will complete according to the requirements of
414 integrity completion.
419 return, the output data
420 has been transferred to the underlying hardware,
421 along with any file metadata that would be required to retrieve that data
422 (i.e., as though each
424 was followed by a call to
426 .IR "See NOTES below" .
429 Ensure that this call creates the file:
430 if this flag is specified in conjunction with
439 When these two flags are specified, symbolic links are not followed:
440 .\" POSIX.1-2001 explicitly requires this behavior.
443 is a symbolic link, then
445 fails regardless of where the symbolic link points.
447 In general, the behavior of
449 is undefined if it is used without
451 There is one exception: on Linux 2.6 and later,
457 refers to a block device.
458 If the block device is in use by the system (e.g., mounted),
465 is supported only when using NFSv3 or later on kernel 2.6 or later.
466 In NFS environments where
468 support is not provided, programs that rely on it
469 for performing locking tasks will contain a race condition.
470 Portable programs that want to perform atomic file locking using a lockfile,
471 and need to avoid reliance on NFS support for
473 can create a unique file on
474 the same filesystem (e.g., incorporating hostname and PID), and use
476 to make a link to the lockfile.
479 returns 0, the lock is successful.
482 on the unique file to check if its link count has increased to 2,
483 in which case the lock is also successful.
487 Allow files whose sizes cannot be represented in an
489 (but can be represented in an
493 .B _LARGEFILE64_SOURCE
494 macro must be defined
498 in order to obtain this definition.
501 feature test macro to 64 (rather than using
504 method of accessing large files on 32-bit systems (see
505 .BR feature_test_macros (7)).
507 .BR O_NOATIME " (since Linux 2.6.8)"
508 Do not update the file last access time
514 This flag can be employed only if one of the following conditions is true:
517 The effective UID of the process
518 .\" Strictly speaking: the filesystem UID
519 matches the owner UID of the file.
521 The calling process has the
523 capability in its user namespace and
524 the owner UID of the file has a mapping in the namespace.
527 This flag is intended for use by indexing or backup programs,
528 where its use can significantly reduce the amount of disk activity.
529 This flag may not be effective on all filesystems.
530 One example is NFS, where the server maintains the access time.
531 .\" The O_NOATIME flag also affects the treatment of st_atime
532 .\" by mmap() and readdir(2), MTK, Dec 04.
537 refers to a terminal device\(emsee
539 will not become the process's controlling terminal even if the
540 process does not have one.
543 If the trailing component (i.e., basename) of
545 is a symbolic link, then the open fails, with the error
547 Symbolic links in earlier components of the pathname will still be
551 error that can occur in this case is indistinguishable from the case where
552 an open fails because there are too many symbolic links found
553 while resolving components in the prefix part of the pathname.)
555 This flag is a FreeBSD extension, which was added to Linux in version 2.1.126,
556 and has subsequently been standardized in POSIX.1-2008.
561 .\" The headers from glibc 2.0.100 and later include a
562 .\" definition of this flag; \fIkernels before 2.1.126 will ignore it if
565 .BR O_NONBLOCK " or " O_NDELAY
566 When possible, the file is opened in nonblocking mode.
569 nor any subsequent I/O operations on the file descriptor which is
570 returned will cause the calling process to wait.
572 Note that the setting of this flag has no effect on the operation of
577 since those interfaces merely inform the caller about whether
578 a file descriptor is "ready",
579 meaning that an I/O operation performed on
580 the file descriptor with the
586 Note that this flag has no effect for regular files and block devices;
587 that is, I/O operations will (briefly) block when device activity
588 is required, regardless of whether
593 semantics might eventually be implemented,
594 applications should not depend upon blocking behavior
595 when specifying this flag for regular files and block devices.
597 For the handling of FIFOs (named pipes), see also
599 For a discussion of the effect of
601 in conjunction with mandatory file locks and with file leases, see
604 .BR O_PATH " (since Linux 2.6.39)"
605 .\" commit 1abf0c718f15a56a0a435588d1b104c7a37dc9bd
606 .\" commit 326be7b484843988afe57566b627fb7a70beac56
607 .\" commit 65cfc6722361570bfe255698d9cd4dccaf47570d
609 .\" http://thread.gmane.org/gmane.linux.man/2790/focus=3496
610 .\" Subject: Re: [PATCH] open(2): document O_PATH
611 .\" Newsgroups: gmane.linux.man, gmane.linux.kernel
613 Obtain a file descriptor that can be used for two purposes:
614 to indicate a location in the filesystem tree and
615 to perform operations that act purely at the file descriptor level.
616 The file itself is not opened, and other file operations (e.g.,
627 The following operations
629 be performed on the resulting file descriptor:
635 if the file descriptor refers to a directory
637 .\" commit 332a2e1244bd08b9e3ecd378028513396a004a24
642 .\" fstat(): commit 55815f70147dcfa3ead5738fd56d3574e2e3c1c2
645 .\" fstatfs(): commit 9d05746e7b16d8565dddbe3200faa1e669d23bbf
647 Duplicating the file descriptor
653 Getting and setting file descriptor flags
659 Retrieving open file status flags using the
662 operation: the returned flags will include the bit
665 Passing the file descriptor as the
669 and the other "*at()" system calls.
675 .BR AT_SYMLINK_FOLLOW )
676 even if the file is not a directory.
678 Passing the file descriptor to another process via a UNIX domain socket
696 Opening a file or directory with the
698 flag requires no permissions on the object itself
699 (but does require execute permission on the directories in the path prefix).
700 Depending on the subsequent operation,
701 a check for suitable file permissions may be performed (e.g.,
703 requires execute permission on the directory referred to
704 by its file descriptor argument).
706 obtaining a reference to a filesystem object by opening it with the
708 flag requires that the caller have read permission on the object,
709 even when the subsequent operation (e.g.,
712 does not require read permission on the object.
716 is a symbolic link and the
718 flag is also specified,
719 then the call returns a file descriptor referring to the symbolic link.
720 This file descriptor can be used as the
728 with an empty pathname to have the calls operate on the symbolic link.
732 refers to an automount point that has not yet been triggered, so no
733 other filesystem is mounted on it, then the call returns a file
734 descriptor referring to the automount directory without triggering a mount.
736 can then be used to determine if it is, in fact, an untriggered
738 .RB ( ".f_type == AUTOFS_SUPER_MAGIC" ).
742 for regular files is to provide the equivalent of POSIX.1's
745 This permits us to open a file for which we have execute
746 permission but not read permission, and then execute that file,
747 with steps something like the following:
752 fd = open("some_prog", O_PATH);
753 snprintf(buf, PATH_MAX, "/proc/self/fd/%d", fd);
754 execl(buf, "some_prog", (char *) NULL);
760 file descriptor can also be passed as the argument of
764 Write operations on the file will complete according to the requirements of
768 (by contrast with the
778 returns, the output data and associated file metadata
779 have been transferred to the underlying hardware
780 (i.e., as though each
782 was followed by a call to
784 .IR "See NOTES below" .
786 .BR O_TMPFILE " (since Linux 3.11)"
787 .\" commit 60545d0d4610b02e55f65d141c95b18ccf855b6e
788 .\" commit f4e0c30c191f87851c4a53454abb55ee276f4a7e
789 .\" commit bb458c644a59dbba3a1fe59b27106c5e68e1c4bd
790 Create an unnamed temporary regular file.
793 argument specifies a directory;
794 an unnamed inode will be created in that directory's filesystem.
795 Anything written to the resulting file will be lost when
796 the last file descriptor is closed, unless the file is given a name.
799 must be specified with one of
807 is not specified, then
809 can be used to link the temporary file into the filesystem, making it
810 permanent, using code like the following:
815 fd = open("/path/to/dir", O_TMPFILE | O_RDWR,
818 /* File I/O on 'fd'... */
820 linkat(fd, NULL, AT_FDCWD, "/path/for/file", AT_EMPTY_PATH);
822 /* If the caller doesn't have the CAP_DAC_READ_SEARCH
823 capability (needed to use AT_EMPTY_PATH with linkat(2)),
824 and there is a proc(5) filesystem mounted, then the
825 linkat(2) call above can be replaced with:
827 snprintf(path, PATH_MAX, "/proc/self/fd/%d", fd);
828 linkat(AT_FDCWD, path, AT_FDCWD, "/path/for/file",
838 argument determines the file permission mode, as with
845 prevents a temporary file from being linked into the filesystem
847 (Note that the meaning of
849 in this case is different from the meaning of
853 There are two main use cases for
854 .\" Inspired by http://lwn.net/Articles/559147/
860 functionality: race-free creation of temporary files that
861 (1) are automatically deleted when closed;
862 (2) can never be reached via any pathname;
863 (3) are not subject to symlink attacks; and
864 (4) do not require the caller to devise unique names.
866 Creating a file that is initially invisible, which is then populated
867 with data and adjusted to have appropriate filesystem attributes
872 before being atomically linked into the filesystem
873 in a fully formed state (using
879 requires support by the underlying filesystem;
880 only a subset of Linux filesystems provide that support.
881 In the initial implementation, support was provided in
882 the ext2, ext3, ext4, UDF, Minix, and shmem filesystems.
883 .\" To check for support, grep for "tmpfile" in kernel sources
884 Support for other filesystems has subsequently been added as follows:
886 .\" commit 99b6436bc29e4f10e4388c27a3e4810191cc4788
887 .\" commit ab29743117f9f4c22ac44c13c1647fb24fb2bafe
889 .\" commit ef3b9af50bfa6a1f02cd7b3f5124b712b1ba3e3c
891 .\" commit 50732df02eefb39ab414ef655979c2c9b64ad21c
892 and ubifs (Linux 4.9)
895 If the file already exists and is a regular file and the access mode allows
900 it will be truncated to length 0.
901 If the file is a FIFO or terminal device file, the
904 Otherwise, the effect of
910 is equivalent to calling
915 .BR O_CREAT|O_WRONLY|O_TRUNC .
919 system call operates in exactly the same way as
921 except for the differences described here.
923 If the pathname given in
925 is relative, then it is interpreted relative to the directory
926 referred to by the file descriptor
928 (rather than relative to the current working directory of
929 the calling process, as is done by
931 for a relative pathname).
941 is interpreted relative to the current working
942 directory of the calling process (like
954 system call is an extension of
956 and provides a superset of the features of
958 It is separate documented, in
965 return the new file descriptor, or \-1 if an error occurred
968 is set appropriately).
974 can fail with the following errors:
977 The requested access to the file is not allowed, or search permission
978 is denied for one of the directories in the path prefix of
980 or the file did not exist yet and write access to the parent directory
983 .BR path_resolution (7).)
988 is specified, the file does not exist, and the user's quota of disk
989 blocks or inodes on the filesystem has been exhausted.
994 .BR O_CREAT " and " O_EXCL
999 points outside your accessible address space.
1006 While blocked waiting to complete an open of a slow device
1009 the call was interrupted by a signal handler; see
1013 The filesystem does not support the
1018 for more information.
1022 .\" In particular, __O_TMPFILE instead of O_TMPFILE
1039 and the final component ("basename") of the new file's
1042 (e.g., it contains characters not permitted by the underlying filesystem).
1046 The final component ("basename") of
1049 (e.g., it contains characters not permitted by the underlying filesystem).
1053 refers to a directory and the access requested involved writing
1062 refers to an existing directory,
1070 but this kernel version does not provide the
1075 Too many symbolic links were encountered in resolving
1080 was a symbolic link, and
1088 The per-process limit on the number of open file descriptors has been reached
1089 (see the description of
1099 The system-wide limit on the total number of open files has been reached.
1103 refers to a device special file and no corresponding device exists.
1104 (This is a Linux kernel bug; in this situation
1110 is not set and the named file does not exist.
1113 A directory component in
1115 does not exist or is a dangling symbolic link.
1119 refers to a nonexistent directory,
1127 but this kernel version does not provide the
1132 The named file is a FIFO,
1133 but memory for the FIFO buffer can't be allocated because
1134 the per-user hard limit on memory allocation for pipes has been reached
1135 and the caller is not privileged; see
1139 Insufficient kernel memory was available.
1143 was to be created but the device containing
1145 has no room for the new file.
1148 A component used as a directory in
1150 is not, in fact, a directory, or \fBO_DIRECTORY\fP was specified and
1152 was not a directory.
1155 .BR O_NONBLOCK " | " O_WRONLY
1156 is set, the named file is a FIFO, and
1157 no process has the FIFO open for reading.
1160 The file is a device special file and no corresponding device exists.
1163 The file is a UNIX domain socket.
1166 The filesystem containing
1173 refers to a regular file that is too large to be opened.
1174 The usual scenario here is that an application compiled
1175 on a 32-bit platform without
1176 .I -D_FILE_OFFSET_BITS=64
1177 tried to open a file whose size exceeds
1183 This is the error specified by POSIX.1;
1184 in kernels before 2.6.24, Linux gave the error
1187 .\" See http://bugzilla.kernel.org/show_bug.cgi?id=7253
1188 .\" "Open of a large file on 32-bit fails with EFBIG, should be EOVERFLOW"
1189 .\" Reported 2006-10-03
1194 flag was specified, but the effective user ID of the caller
1195 .\" Strictly speaking, it's the filesystem UID... (MTK)
1196 did not match the owner of the file and the caller was not privileged.
1199 The operation was prevented by a file seal; see
1204 refers to a file on a read-only filesystem and write access was
1209 refers to an executable image which is currently being executed and
1210 write access was requested.
1214 refers to a file that is currently in use as a swap file, and the
1220 refers to a file that is currently being read by the kernel (e.g. for
1221 module/firmware loading), and write access was requested.
1226 flag was specified, and an incompatible lease was held on the file
1230 The following additional errors can occur for
1235 is not a valid file descriptor.
1239 is a relative pathname and
1241 is a file descriptor referring to a file other than a directory.
1244 was added to Linux in kernel 2.6.16;
1245 library support was added to glibc in version 2.4.
1249 SVr4, 4.3BSD, POSIX.1-2001, POSIX.1-2008.
1263 flags are Linux-specific.
1266 to obtain their definitions.
1273 flags are not specified in POSIX.1-2001,
1274 but are specified in POSIX.1-2008.
1275 Since glibc 2.12, one can obtain their definitions by defining either
1277 with a value greater than or equal to 200809L or
1279 with a value greater than or equal to 700.
1280 In glibc 2.11 and earlier, one obtains the definitions by defining
1284 .BR feature_test_macros (7),
1285 feature test macros such as
1286 .BR _POSIX_C_SOURCE ,
1290 must be defined before including
1296 flag is sometimes used in cases where one wants to open
1297 but does not necessarily have the intention to read or write.
1299 this may be used to open a device in order to get a file descriptor
1303 The (undefined) effect of
1304 .B O_RDONLY | O_TRUNC
1305 varies among implementations.
1306 On many systems the file is actually truncated.
1307 .\" Linux 2.0, 2.5: truncate
1308 .\" Solaris 5.7, 5.8: truncate
1309 .\" Irix 6.5: truncate
1310 .\" Tru64 5.1B: truncate
1311 .\" HP-UX 11.22: truncate
1312 .\" FreeBSD 4.7: truncate
1316 can open device special files, but
1318 cannot create them; use
1322 If the file is newly created, its
1327 (respectively, time of last access, time of last status change, and
1328 time of last modification; see
1331 to the current time, and so are the
1337 Otherwise, if the file is modified because of the
1343 fields are set to the current time.
1347 directory show the open file descriptors of the process with the PID
1350 .I /proc/[pid]/fdinfo
1351 directory show even more information about these file descriptors.
1354 for further details of both of these directories.
1356 The Linux header file
1362 synonym is defined instead.
1365 .SS Open file descriptions
1366 The term open file description is the one used by POSIX to refer to the
1367 entries in the system-wide table of open files.
1368 In other contexts, this object is
1369 variously also called an "open file object",
1370 a "file handle", an "open file table entry",
1371 or\(emin kernel-developer parlance\(ema
1374 When a file descriptor is duplicated (using
1377 the duplicate refers to the same open file description
1378 as the original file descriptor,
1379 and the two file descriptors consequently share
1380 the file offset and file status flags.
1381 Such sharing can also occur between processes:
1382 a child process created via
1384 inherits duplicates of its parent's file descriptors,
1385 and those duplicates refer to the same open file descriptions.
1389 of a file creates a new open file description;
1390 thus, there may be multiple open file descriptions
1391 corresponding to a file inode.
1393 On Linux, one can use the
1396 operation to test whether two file descriptors
1397 (in the same process or in two different processes)
1398 refer to the same open file description.
1401 .SS Synchronized I/O
1402 The POSIX.1-2008 "synchronized I/O" option
1403 specifies different variants of synchronized I/O,
1411 for controlling the behavior.
1412 Regardless of whether an implementation supports this option,
1413 it must at least support the use of
1423 Somewhat incorrectly, glibc defines
1425 to have the same value as
1428 is defined in the Linux header file
1430 on HP PA-RISC, but it is not used.)
1433 provides synchronized I/O
1435 integrity completion,
1436 meaning write operations will flush data and all associated metadata
1437 to the underlying hardware.
1439 provides synchronized I/O
1441 integrity completion,
1442 meaning write operations will flush data
1443 to the underlying hardware,
1444 but will only flush metadata updates that are required
1445 to allow a subsequent read operation to complete successfully.
1446 Data integrity completion can reduce the number of disk operations
1447 that are required for applications that don't need the guarantees
1448 of file integrity completion.
1450 To understand the difference between the two types of completion,
1451 consider two pieces of file metadata:
1452 the file last modification timestamp
1454 and the file length.
1455 All write operations will update the last file modification timestamp,
1456 but only writes that add data to the end of the
1457 file will change the file length.
1458 The last modification timestamp is not needed to ensure that
1459 a read completes successfully, but the file length is.
1462 would only guarantee to flush updates to the file length metadata
1465 would also always flush the last modification timestamp metadata).
1467 Before Linux 2.6.33, Linux implemented only the
1471 However, when that flag was specified,
1472 most filesystems actually provided the equivalent of synchronized I/O
1474 integrity completion (i.e.,
1476 was actually implemented as the equivalent of
1479 Since Linux 2.6.33, proper
1481 support is provided.
1482 However, to ensure backward binary compatibility,
1484 was defined with the same value as the historical
1488 was defined as a new (two-bit) flag value that includes the
1491 This ensures that applications compiled against
1492 new headers get at least
1494 semantics on pre-2.6.33 kernels.
1496 .SS C library/kernel differences
1498 the glibc wrapper function for
1502 system call, rather than the kernel's
1505 For certain architectures, this is also true in glibc versions before 2.26.
1508 There are many infelicities in the protocol underlying NFS, affecting
1510 .BR O_SYNC " and " O_NDELAY .
1512 On NFS filesystems with UID mapping enabled,
1515 return a file descriptor but, for example,
1519 This is because the client performs
1522 permissions, but UID mapping is performed by the server upon
1523 read and write requests.
1527 Opening the read or write end of a FIFO blocks until the other
1528 end is also opened (by another process or thread).
1531 for further details.
1534 .SS File access mode
1535 Unlike the other values that can be specified in
1540 .BR O_RDONLY ", " O_WRONLY ", and " O_RDWR
1541 do not specify individual bits.
1542 Rather, they define the low order two bits of
1544 and are defined respectively as 0, 1, and 2.
1545 In other words, the combination
1546 .B "O_RDONLY | O_WRONLY"
1547 is a logical error, and certainly does not have the same meaning as
1550 Linux reserves the special, nonstandard access mode 3 (binary 11) in
1553 check for read and write permission on the file and return a file descriptor
1554 that can't be used for reading or writing.
1555 This nonstandard access mode is used by some Linux drivers to return a
1556 file descriptor that is to be used only for device-specific
1559 .\" See for example util-linux's disk-utils/setfdprm.c
1560 .\" For some background on access mode 3, see
1561 .\" http://thread.gmane.org/gmane.linux.kernel/653123
1562 .\" "[RFC] correct flags to f_mode conversion in __dentry_open"
1563 .\" LKML, 12 Mar 2008
1566 .SS Rationale for openat() and other "directory file descriptor" APIs
1568 and the other system calls and library functions that take
1569 a directory file descriptor argument
1573 .BR fanotify_mark (2),
1581 .BR name_to_handle_at (2),
1591 address two problems with the older interfaces that preceded them.
1592 Here, the explanation is in terms of the
1594 call, but the rationale is analogous for the other interfaces.
1598 allows an application to avoid race conditions that could
1601 to open files in directories other than the current working directory.
1602 These race conditions result from the fact that some component
1603 of the directory prefix given to
1605 could be changed in parallel with the call to
1607 Suppose, for example, that we wish to create the file
1608 .I dir1/dir2/xxx.dep
1612 The problem is that between the existence check and the file-creation step,
1616 (which might be symbolic links)
1617 could be modified to point to a different location.
1618 Such races can be avoided by
1619 opening a file descriptor for the target directory,
1620 and then specifying that file descriptor as the
1628 file descriptor also has other benefits:
1630 the file descriptor is a stable reference to the directory,
1631 even if the directory is renamed; and
1633 the open file descriptor prevents the underlying filesystem from
1635 just as when a process has a current working directory on a filesystem.
1639 allows the implementation of a per-thread "current working
1640 directory", via file descriptor(s) maintained by the application.
1641 (This functionality can also be obtained by tricks based
1643 .IR /proc/self/fd/ dirfd,
1644 but less efficiently.)
1651 flag may impose alignment restrictions on the length and address
1652 of user-space buffers and the file offset of I/Os.
1654 restrictions vary by filesystem and kernel version and might be
1656 However there is currently no filesystem\-independent
1657 interface for an application to discover these restrictions for a given
1659 Some filesystems provide their own interfaces
1660 for doing so, for example the
1665 Under Linux 2.4, transfer sizes, and the alignment of the user buffer
1666 and the file offset must all be multiples of the logical block size
1668 Since Linux 2.6.0, alignment to the logical block size of the
1669 underlying storage (typically 512 bytes) suffices.
1670 The logical block size can be determined using the
1673 operation or from the shell using the command:
1680 I/Os should never be run concurrently with the
1683 if the memory buffer is a private mapping
1684 (i.e., any mapping created with the
1688 this includes memory allocated on the heap and statically allocated buffers).
1689 Any such I/Os, whether submitted via an asynchronous I/O interface or from
1690 another thread in the process,
1691 should be completed before
1694 Failure to do so can result in data corruption and undefined behavior in
1695 parent and child processes.
1696 This restriction does not apply when the memory buffer for the
1698 I/Os was created using
1705 Nor does this restriction apply when the memory buffer has been advised as
1709 ensuring that it will not be available
1715 flag was introduced in SGI IRIX, where it has alignment
1716 restrictions similar to those of Linux 2.4.
1719 call to query appropriate alignments, and sizes.
1720 FreeBSD 4.x introduced
1721 a flag of the same name, but without alignment restrictions.
1724 support was added under Linux in kernel version 2.4.10.
1725 Older Linux kernels simply ignore this flag.
1726 Some filesystems may not implement the flag, in which case
1728 fails with the error
1732 Applications should avoid mixing
1734 and normal I/O to the same file,
1735 and especially to overlapping byte regions in the same file.
1736 Even when the filesystem correctly handles the coherency issues in
1737 this situation, overall I/O throughput is likely to be slower than
1738 using either mode alone.
1739 Likewise, applications should avoid mixing
1741 of files with direct I/O to the same files.
1745 with NFS will differ from local filesystems.
1747 kernels configured in certain ways, may not support this combination.
1748 The NFS protocol does not support passing the flag to the server, so
1750 I/O will bypass the page cache only on the client; the server may
1751 still cache the I/O.
1752 The client asks the server to make the I/O
1753 synchronous to preserve the synchronous semantics of
1755 Some servers will perform poorly under these circumstances, especially
1756 if the I/O size is small.
1757 Some servers may also be configured to
1758 lie to clients about the I/O having reached stable storage; this
1759 will avoid the performance penalty at some risk to data integrity
1760 in the event of server power failure.
1761 The Linux NFS client places no alignment restrictions on
1767 is a potentially powerful tool that should be used with caution.
1768 It is recommended that applications treat use of
1770 as a performance option which is disabled by default.
1772 Currently, it is not possible to enable signal-driven
1779 to enable this flag.
1780 .\" FIXME . Check bugzilla report on open(O_ASYNC)
1781 .\" See http://bugzilla.kernel.org/show_bug.cgi?id=5993
1783 One must check for two different error codes,
1787 when trying to determine whether the kernel supports
1797 and the file specified by
1801 will create a regular file (i.e.,
1816 .BR open_by_handle_at (2),
1827 .BR path_resolution (7),