| 1 |
.\" Copyright 1992, Drew Eckhardt <[email protected]> |
| 2 |
.\" Copyright 1993, Ian Jackson <[email protected]> |
| 3 |
.\" Copyright 2007, Michael Kerrisk <[email protected]> |
| 4 |
.\" Copyright, the authors of the Linux man-pages project |
| 5 |
.\" |
| 6 |
.\" SPDX-License-Identifier: Linux-man-pages-copyleft |
| 7 |
.\" |
| 8 |
.TH write 2 2026-02-08 "Linux man-pages 6.18" |
| 9 |
.SH NAME |
| 10 |
write \- write to a file descriptor |
| 11 |
.SH LIBRARY |
| 12 |
Standard C library |
| 13 |
.RI ( libc ,\~ \-lc ) |
| 14 |
.SH SYNOPSIS |
| 15 |
.nf |
| 16 |
.B #include <unistd.h> |
| 17 |
.P |
| 18 |
.BR "ssize_t write(" "size_t count;" |
| 19 |
.BI " int " fd ", const void " buf [ count "], size_t " count ); |
| 20 |
.fi |
| 21 |
.SH DESCRIPTION |
| 22 |
.BR write () |
| 23 |
writes up to |
| 24 |
.I count |
| 25 |
bytes from the buffer starting at |
| 26 |
.I buf |
| 27 |
to the file referred to by the file descriptor |
| 28 |
.IR fd . |
| 29 |
.P |
| 30 |
The number of bytes written may be less than |
| 31 |
.I count |
| 32 |
if, for example, |
| 33 |
there is insufficient space on the underlying physical medium, or the |
| 34 |
.B RLIMIT_FSIZE |
| 35 |
resource limit is encountered (see |
| 36 |
.BR setrlimit (2)), |
| 37 |
or the call was interrupted by a signal |
| 38 |
handler after having written less than |
| 39 |
.I count |
| 40 |
bytes. |
| 41 |
(See also |
| 42 |
.BR pipe (7).) |
| 43 |
.P |
| 44 |
For a seekable file (i.e., one to which |
| 45 |
.BR lseek (2) |
| 46 |
may be applied, for example, a regular file) |
| 47 |
writing takes place at the file offset, |
| 48 |
and the file offset is incremented by |
| 49 |
the number of bytes actually written. |
| 50 |
If the file was |
| 51 |
.BR open (2)ed |
| 52 |
with |
| 53 |
.BR O_APPEND , |
| 54 |
the file offset is first set to the end of the file before writing. |
| 55 |
The adjustment of the file offset and the write operation |
| 56 |
are performed as an atomic step. |
| 57 |
.P |
| 58 |
POSIX requires that a |
| 59 |
.BR read (2) |
| 60 |
that can be proved to occur after a |
| 61 |
.BR write () |
| 62 |
has returned will return the new data. |
| 63 |
Note that not all filesystems are POSIX conforming. |
| 64 |
.P |
| 65 |
According to POSIX.1, if |
| 66 |
.I count |
| 67 |
is greater than |
| 68 |
.BR SSIZE_MAX , |
| 69 |
the result is implementation-defined; |
| 70 |
see NOTES for the upper limit on Linux. |
| 71 |
.SH RETURN VALUE |
| 72 |
On success, the number of bytes written is returned. |
| 73 |
On error, \-1 is returned, and |
| 74 |
.I errno |
| 75 |
is set |
| 76 |
to indicate the error. |
| 77 |
.P |
| 78 |
Note that a successful |
| 79 |
.BR write () |
| 80 |
may transfer fewer than |
| 81 |
.I count |
| 82 |
bytes. |
| 83 |
Such partial writes can occur for various reasons; |
| 84 |
for example, because there was insufficient space on the disk device |
| 85 |
to write all of the requested bytes, or because a blocked |
| 86 |
.BR write () |
| 87 |
to a socket, pipe, or similar was interrupted by a signal handler |
| 88 |
after it had transferred some, but before it had transferred all |
| 89 |
of the requested bytes. |
| 90 |
In the event of a partial write, the caller can make another |
| 91 |
.BR write () |
| 92 |
call to transfer the remaining bytes. |
| 93 |
The subsequent call will either transfer further bytes or |
| 94 |
may result in an error (e.g., if the disk is now full). |
| 95 |
.P |
| 96 |
If |
| 97 |
.I count |
| 98 |
is zero and |
| 99 |
.I fd |
| 100 |
refers to a regular file, then |
| 101 |
.BR write () |
| 102 |
may return a failure status if one of the errors below is detected. |
| 103 |
If no errors are detected, or error detection is not performed, |
| 104 |
0 is returned without causing any other effect. |
| 105 |
If |
| 106 |
.I count |
| 107 |
is zero and |
| 108 |
.I fd |
| 109 |
refers to a file other than a regular file, |
| 110 |
the results are not specified. |
| 111 |
.SH ERRORS |
| 112 |
.TP |
| 113 |
.B EAGAIN |
| 114 |
The file descriptor |
| 115 |
.I fd |
| 116 |
refers to a file other than a socket and has been marked nonblocking |
| 117 |
.RB ( O_NONBLOCK ), |
| 118 |
and the write would block. |
| 119 |
See |
| 120 |
.BR open (2) |
| 121 |
for further details on the |
| 122 |
.B O_NONBLOCK |
| 123 |
flag. |
| 124 |
.TP |
| 125 |
.BR EAGAIN " or " EWOULDBLOCK |
| 126 |
.\" Actually EAGAIN on Linux |
| 127 |
The file descriptor |
| 128 |
.I fd |
| 129 |
refers to a socket and has been marked nonblocking |
| 130 |
.RB ( O_NONBLOCK ), |
| 131 |
and the write would block. |
| 132 |
POSIX.1-2001 allows either error to be returned for this case, |
| 133 |
and does not require these constants to have the same value, |
| 134 |
so a portable application should check for both possibilities. |
| 135 |
.TP |
| 136 |
.B EBADF |
| 137 |
.I fd |
| 138 |
is not a valid file descriptor or is not open for writing. |
| 139 |
.TP |
| 140 |
.B EDESTADDRREQ |
| 141 |
.I fd |
| 142 |
refers to a datagram socket for which a peer address has not been set using |
| 143 |
.BR connect (2). |
| 144 |
.TP |
| 145 |
.B EDQUOT |
| 146 |
The user's quota of disk blocks on the filesystem containing the file |
| 147 |
referred to by |
| 148 |
.I fd |
| 149 |
has been exhausted. |
| 150 |
.TP |
| 151 |
.B EFAULT |
| 152 |
.I buf |
| 153 |
is outside your accessible address space. |
| 154 |
.TP |
| 155 |
.B EFBIG |
| 156 |
An attempt was made to write a file that exceeds the implementation-defined |
| 157 |
maximum file size or the process's file size limit, |
| 158 |
or to write at a position past the maximum allowed offset. |
| 159 |
.TP |
| 160 |
.B EINTR |
| 161 |
The call was interrupted by a signal before any data was written; |
| 162 |
see |
| 163 |
.BR signal (7). |
| 164 |
.TP |
| 165 |
.B EINVAL |
| 166 |
.I fd |
| 167 |
is attached to an object which is unsuitable for writing; |
| 168 |
or the file was opened with the |
| 169 |
.B O_DIRECT |
| 170 |
flag, and either the address specified in |
| 171 |
.IR buf , |
| 172 |
the value specified in |
| 173 |
.IR count , |
| 174 |
or the file offset is not suitably aligned. |
| 175 |
.TP |
| 176 |
.B EIO |
| 177 |
A low-level I/O error occurred while modifying the inode. |
| 178 |
This error may relate to the write-back of data written by an earlier |
| 179 |
.BR write (), |
| 180 |
which may have been issued to a different file descriptor on |
| 181 |
the same file. |
| 182 |
Since Linux 4.13, errors from write-back come |
| 183 |
with a promise that they |
| 184 |
.I may |
| 185 |
be reported by subsequent. |
| 186 |
.BR write () |
| 187 |
requests, and |
| 188 |
.I will |
| 189 |
be reported by a subsequent |
| 190 |
.BR fsync (2) |
| 191 |
(whether or not they were also reported by |
| 192 |
.BR write ()). |
| 193 |
.\" commit 088737f44bbf6378745f5b57b035e57ee3dc4750 |
| 194 |
An alternate cause of |
| 195 |
.B EIO |
| 196 |
on networked filesystems is when an advisory lock had been taken out |
| 197 |
on the file descriptor and this lock has been lost. |
| 198 |
See the |
| 199 |
.I "Lost locks" |
| 200 |
section of |
| 201 |
.BR fcntl (2) |
| 202 |
for further details. |
| 203 |
.TP |
| 204 |
.B ENOSPC |
| 205 |
The device containing the file referred to by |
| 206 |
.I fd |
| 207 |
has no room for the data. |
| 208 |
.TP |
| 209 |
.B EPERM |
| 210 |
The operation was prevented by a file seal; |
| 211 |
see |
| 212 |
.BR fcntl (2). |
| 213 |
.TP |
| 214 |
.B EPIPE |
| 215 |
.I fd |
| 216 |
is connected to a pipe or socket whose reading end is closed. |
| 217 |
When this happens the writing process will also receive a |
| 218 |
.B SIGPIPE |
| 219 |
signal. |
| 220 |
(Thus, the write return value is seen only if the program |
| 221 |
catches, blocks or ignores this signal.) |
| 222 |
.P |
| 223 |
Other errors may occur, depending on the object connected to |
| 224 |
.IR fd . |
| 225 |
.SH STANDARDS |
| 226 |
POSIX.1-2024. |
| 227 |
.SH HISTORY |
| 228 |
SVr4, 4.3BSD, POSIX.1-2001. |
| 229 |
.\" SVr4 documents additional error |
| 230 |
.\" conditions EDEADLK, ENOLCK, ENOLNK, ENOSR, ENXIO, or ERANGE. |
| 231 |
.P |
| 232 |
Under SVr4 a write may be interrupted and return |
| 233 |
.B EINTR |
| 234 |
at any point, |
| 235 |
not just before any data is written. |
| 236 |
.SH NOTES |
| 237 |
A successful return from |
| 238 |
.BR write () |
| 239 |
does not make any guarantee that data has been committed to disk. |
| 240 |
On some filesystems, including NFS, it does not even guarantee |
| 241 |
that space has successfully been reserved for the data. |
| 242 |
In this case, |
| 243 |
some errors might be delayed until a future |
| 244 |
.BR write (), |
| 245 |
.BR fsync (2), |
| 246 |
or even |
| 247 |
.BR close (2). |
| 248 |
The only way to be sure is to call |
| 249 |
.BR fsync (2) |
| 250 |
after you are done writing all your data. |
| 251 |
.P |
| 252 |
If a |
| 253 |
.BR write () |
| 254 |
is interrupted by a signal handler before any bytes are written, |
| 255 |
then the call fails with the error |
| 256 |
.BR EINTR ; |
| 257 |
if it is interrupted after at least one byte has been written, |
| 258 |
the call succeeds, and returns the number of bytes written. |
| 259 |
.P |
| 260 |
On Linux, |
| 261 |
.BR write () |
| 262 |
(and similar system calls) will transfer at most |
| 263 |
0x7ffff000 (2,147,479,552) bytes, |
| 264 |
returning the number of bytes actually transferred. |
| 265 |
.\" commit e28cc71572da38a5a12c1cfe4d7032017adccf69 |
| 266 |
(This is true on both 32-bit and 64-bit systems.) |
| 267 |
.P |
| 268 |
An error return value while performing |
| 269 |
.BR write () |
| 270 |
using direct I/O does not mean the |
| 271 |
entire write has failed. |
| 272 |
Partial data may be written |
| 273 |
and the data at the file offset on which the |
| 274 |
.BR write () |
| 275 |
was attempted should be considered inconsistent. |
| 276 |
.SH BUGS |
| 277 |
According to POSIX.1-2008/SUSv4 Section XSI 2.9.7 |
| 278 |
("Thread Interactions with Regular File Operations"): |
| 279 |
.P |
| 280 |
.RS 4 |
| 281 |
All of the following functions shall be atomic with respect to |
| 282 |
each other in the effects specified in POSIX.1-2008 when they |
| 283 |
operate on regular files or symbolic links: ... |
| 284 |
.RE |
| 285 |
.P |
| 286 |
Among the APIs subsequently listed are |
| 287 |
.BR write () |
| 288 |
and |
| 289 |
.BR writev (2). |
| 290 |
And among the effects that should be atomic across threads (and processes) |
| 291 |
are updates of the file offset. |
| 292 |
However, before Linux 3.14, |
| 293 |
this was not the case: if two processes that share |
| 294 |
an open file description (see |
| 295 |
.BR open (2)) |
| 296 |
perform a |
| 297 |
.BR write () |
| 298 |
(or |
| 299 |
.BR writev (2)) |
| 300 |
at the same time, then the I/O operations were not atomic |
| 301 |
with respect to updating the file offset, |
| 302 |
with the result that the blocks of data output by the two processes |
| 303 |
might (incorrectly) overlap. |
| 304 |
This problem was fixed in Linux 3.14. |
| 305 |
.\" http://thread.gmane.org/gmane.linux.kernel/1649458 |
| 306 |
.\" From: Michael Kerrisk (man-pages <mtk.manpages <at> gmail.com> |
| 307 |
.\" Subject: Update of file offset on write() etc. is non-atomic with I/O |
| 308 |
.\" Date: 2014-02-17 15:41:37 GMT |
| 309 |
.\" Newsgroups: gmane.linux.kernel, gmane.linux.file-systems |
| 310 |
.\" commit 9c225f2655e36a470c4f58dbbc99244c5fc7f2d4 |
| 311 |
.\" Author: Linus Torvalds <[email protected]> |
| 312 |
.\" Date: Mon Mar 3 09:36:58 2014 -0800 |
| 313 |
.\" |
| 314 |
.\" vfs: atomic f_pos accesses as per POSIX |
| 315 |
.SH SEE ALSO |
| 316 |
.BR close (2), |
| 317 |
.BR fcntl (2), |
| 318 |
.BR fsync (2), |
| 319 |
.BR ioctl (2), |
| 320 |
.BR lseek (2), |
| 321 |
.BR open (2), |
| 322 |
.BR pwrite (2), |
| 323 |
.BR read (2), |
| 324 |
.BR select (2), |
| 325 |
.BR writev (2), |
| 326 |
.BR fwrite (3) |