| 1 |
.\" Copyright 1992, Drew Eckhardt <[email protected]> |
| 2 |
.\" Copyright 2016, Michael Kerrisk <[email protected]> |
| 3 |
.\" Copyright, the authors of the Linux man-pages project |
| 4 |
.\" |
| 5 |
.\" SPDX-License-Identifier: Linux-man-pages-copyleft |
| 6 |
.\" |
| 7 |
.TH close 2 2026-02-08 "Linux man-pages 6.18" |
| 8 |
.SH NAME |
| 9 |
close \- close a file descriptor |
| 10 |
.SH LIBRARY |
| 11 |
Standard C library |
| 12 |
.RI ( libc ,\~ \-lc ) |
| 13 |
.SH SYNOPSIS |
| 14 |
.nf |
| 15 |
.B #include <unistd.h> |
| 16 |
.P |
| 17 |
.BI "int close(int " fd ); |
| 18 |
.fi |
| 19 |
.SH DESCRIPTION |
| 20 |
.BR close () |
| 21 |
closes a file descriptor, so that it no longer refers to any file and |
| 22 |
may be reused. |
| 23 |
Any record locks (see |
| 24 |
.BR fcntl (2)) |
| 25 |
held on the file it was associated with, |
| 26 |
and owned by the process, |
| 27 |
are removed regardless of the file descriptor that was used to obtain the lock. |
| 28 |
This has some unfortunate consequences |
| 29 |
and one should be extra careful when using advisory record locking. |
| 30 |
See |
| 31 |
.BR fcntl (2) |
| 32 |
for discussion of the risks and consequences |
| 33 |
as well as for the (probably preferred) open file description locks. |
| 34 |
.P |
| 35 |
If |
| 36 |
.I fd |
| 37 |
is the last file descriptor referring to the underlying |
| 38 |
open file description (see |
| 39 |
.BR open (2)), |
| 40 |
the resources associated with the open file description are freed; |
| 41 |
if the file descriptor was the last reference to a file which has been |
| 42 |
removed using |
| 43 |
.BR unlink (2), |
| 44 |
the file is deleted. |
| 45 |
.SH RETURN VALUE |
| 46 |
.BR close () |
| 47 |
returns zero on success. |
| 48 |
On error, \-1 is returned, and |
| 49 |
.I errno |
| 50 |
is set to indicate the error. |
| 51 |
.SH ERRORS |
| 52 |
.TP |
| 53 |
.B EBADF |
| 54 |
.I fd |
| 55 |
isn't a valid open file descriptor. |
| 56 |
.TP |
| 57 |
.B EINTR |
| 58 |
.\" Though, it's in doubt whether this error can ever occur; see |
| 59 |
.\" https://lwn.net/Articles/576478/ "Returning EINTR from close()" |
| 60 |
The |
| 61 |
.BR close () |
| 62 |
call was interrupted by a signal; |
| 63 |
see |
| 64 |
.BR signal (7). |
| 65 |
.TP |
| 66 |
.B EIO |
| 67 |
An I/O error occurred. |
| 68 |
.TP |
| 69 |
.B ENOSPC |
| 70 |
.TQ |
| 71 |
.B EDQUOT |
| 72 |
On NFS, these errors are not normally reported against the first write |
| 73 |
which exceeds the available storage space, but instead against a |
| 74 |
subsequent |
| 75 |
.BR write (2), |
| 76 |
.BR fsync (2), |
| 77 |
or |
| 78 |
.BR close (). |
| 79 |
.P |
| 80 |
See CAVEATS for a discussion of why |
| 81 |
.BR close () |
| 82 |
should not be retried after an error. |
| 83 |
.SH STANDARDS |
| 84 |
POSIX.1-2008. |
| 85 |
.SH HISTORY |
| 86 |
4.3BSD, |
| 87 |
SVr4, |
| 88 |
POSIX.1-1988. |
| 89 |
.\" SVr4 documents an additional ENOLINK error condition. |
| 90 |
.SH NOTES |
| 91 |
The close-on-exec file descriptor flag can be used to ensure |
| 92 |
that a file descriptor is automatically closed upon a successful |
| 93 |
.BR execve (2); |
| 94 |
see |
| 95 |
.BR fcntl (2) |
| 96 |
for details. |
| 97 |
.SH CAVEATS |
| 98 |
A successful close does not guarantee that the data has been successfully |
| 99 |
saved to disk, as the kernel uses the buffer cache to defer writes. |
| 100 |
Typically, filesystems do not flush buffers when a file is closed. |
| 101 |
If you need to be sure that |
| 102 |
the data is physically stored on the underlying disk, use |
| 103 |
.BR fsync (2). |
| 104 |
(It will depend on the disk hardware at this point.) |
| 105 |
.\" |
| 106 |
.SS Multithreaded processes and close() |
| 107 |
It is probably unwise to close file descriptors while |
| 108 |
they may be in use by system calls in |
| 109 |
other threads in the same process. |
| 110 |
Since a file descriptor may be reused, |
| 111 |
there are some obscure race conditions |
| 112 |
that may cause unintended side effects. |
| 113 |
.\" Date: Tue, 4 Sep 2007 13:57:35 +0200 |
| 114 |
.\" From: Fredrik Noring <[email protected]> |
| 115 |
.\" One such race involves signals and ERESTARTSYS. If a file descriptor |
| 116 |
.\" in use by a system call is closed and then reused by, e.g., an |
| 117 |
.\" independent open() in some unrelated thread, before the original system |
| 118 |
.\" call has restarted after ERESTARTSYS, the original system call will |
| 119 |
.\" later restart with the reused file descriptor. This is most likely a |
| 120 |
.\" serious programming error. |
| 121 |
.P |
| 122 |
Furthermore, consider the following scenario where two threads are |
| 123 |
performing operations on the same file descriptor: |
| 124 |
.IP (1) 5 |
| 125 |
One thread is blocked in an I/O system call on the file descriptor. |
| 126 |
For example, it is trying to |
| 127 |
.BR write (2) |
| 128 |
to a pipe that is already full, or trying to |
| 129 |
.BR read (2) |
| 130 |
from a stream socket which currently has no available data. |
| 131 |
.IP (2) |
| 132 |
Another thread closes the file descriptor. |
| 133 |
.P |
| 134 |
The behavior in this situation varies across systems. |
| 135 |
On some systems, when the file descriptor is closed, |
| 136 |
the blocking system call returns immediately with an error. |
| 137 |
.P |
| 138 |
On Linux (and possibly some other systems), the behavior is different: |
| 139 |
the blocking I/O system call holds a reference to the underlying |
| 140 |
open file description, and this reference keeps the description open |
| 141 |
until the I/O system call completes. |
| 142 |
.\" 'struct file' in kernel-speak |
| 143 |
(See |
| 144 |
.BR open (2) |
| 145 |
for a discussion of open file descriptions.) |
| 146 |
Thus, the blocking system call in the first thread may successfully |
| 147 |
complete after the |
| 148 |
.BR close () |
| 149 |
in the second thread. |
| 150 |
.\" |
| 151 |
.SS Dealing with error returns from close() |
| 152 |
A careful programmer will check the return value of |
| 153 |
.BR close (), |
| 154 |
since it is quite possible that errors on a previous |
| 155 |
.BR write (2) |
| 156 |
operation are reported only on the final |
| 157 |
.BR close () |
| 158 |
that releases the open file description. |
| 159 |
Failing to check the return value when closing a file may lead to |
| 160 |
.I silent |
| 161 |
loss of data. |
| 162 |
This can especially be observed with NFS and with disk quota. |
| 163 |
.P |
| 164 |
Note, however, that a failure return should be used only for |
| 165 |
diagnostic purposes (i.e., a warning to the application that there |
| 166 |
may still be I/O pending or there may have been failed I/O) |
| 167 |
or remedial purposes |
| 168 |
(e.g., writing the file once more or creating a backup). |
| 169 |
.P |
| 170 |
Retrying the |
| 171 |
.BR close () |
| 172 |
after a failure return is the wrong thing to do, |
| 173 |
.\" The file descriptor is released early in close(); |
| 174 |
.\" close() ==> __close_fd(): |
| 175 |
.\" __put_unused_fd() ==> __clear_open_fd() |
| 176 |
.\" return filp_close(file, files); |
| 177 |
.\" |
| 178 |
.\" The errors are returned by filp_close() after the FD has been |
| 179 |
.\" cleared for re-use. |
| 180 |
since this may cause a reused file descriptor |
| 181 |
from another thread to be closed. |
| 182 |
This can occur because the Linux kernel |
| 183 |
.I always |
| 184 |
releases the file descriptor early in the close |
| 185 |
operation, freeing it for reuse; |
| 186 |
the steps that may return an error, |
| 187 |
.\" filp_close() |
| 188 |
such as flushing data to the filesystem or device, |
| 189 |
occur only later in the close operation. |
| 190 |
.P |
| 191 |
Many other implementations similarly always close the file descriptor |
| 192 |
.\" FreeBSD documents this explicitly. From the look of the source code |
| 193 |
.\" SVR4, ancient SunOS, later Solaris, and AIX all do this. |
| 194 |
(except in the case of |
| 195 |
.BR EBADF , |
| 196 |
meaning that the file descriptor was invalid) |
| 197 |
even if they subsequently report an error on return from |
| 198 |
.BR close (). |
| 199 |
POSIX.1-2008 was silent on this point. |
| 200 |
.P |
| 201 |
A careful programmer who wants to know about I/O errors may precede |
| 202 |
.BR close () |
| 203 |
with a call to |
| 204 |
.BR fsync (2). |
| 205 |
.P |
| 206 |
The |
| 207 |
.B EINTR |
| 208 |
error is a somewhat special case. |
| 209 |
Regarding the |
| 210 |
.B EINTR |
| 211 |
error, POSIX.1-2008 said: |
| 212 |
.P |
| 213 |
.RS |
| 214 |
If |
| 215 |
.BR close () |
| 216 |
is interrupted by a signal that is to be caught, it shall return \-1 with |
| 217 |
.I errno |
| 218 |
set to |
| 219 |
.B EINTR |
| 220 |
and the state of |
| 221 |
.I fd |
| 222 |
is unspecified. |
| 223 |
.RE |
| 224 |
.P |
| 225 |
This permits the behavior that occurs on Linux and |
| 226 |
many other implementations, where, |
| 227 |
as with other errors that may be reported by |
| 228 |
.BR close (), |
| 229 |
the file descriptor is guaranteed to be closed. |
| 230 |
However, it also permits another possibility: |
| 231 |
that the implementation returns an |
| 232 |
.B EINTR |
| 233 |
error and keeps the file descriptor open. |
| 234 |
(According to its documentation, HP-UX's |
| 235 |
.BR close () |
| 236 |
does this.) |
| 237 |
The caller must then once more use |
| 238 |
.BR close () |
| 239 |
to close the file descriptor, to avoid file descriptor leaks. |
| 240 |
This divergence in implementation behaviors provides |
| 241 |
a difficult hurdle for portable applications, since on many implementations, |
| 242 |
.BR close () |
| 243 |
must not be called again after an |
| 244 |
.B EINTR |
| 245 |
error, and on at least one, |
| 246 |
.BR close () |
| 247 |
must be called again. |
| 248 |
.P |
| 249 |
POSIX.1-2024 standardized the behavior of HP-UX, |
| 250 |
making Linux and many other implementations non-conforming. |
| 251 |
There are no plans to change the behavior on Linux. |
| 252 |
.SH SEE ALSO |
| 253 |
.BR close_range (2), |
| 254 |
.BR fcntl (2), |
| 255 |
.BR fsync (2), |
| 256 |
.BR open (2), |
| 257 |
.BR shutdown (2), |
| 258 |
.BR unlink (2), |
| 259 |
.BR fclose (3) |