RENAME(2) Linux Programmer's Manual RENAME(2)
NAME
rename, renameat, renameat2 - change the name or location of a file
SYNOPSIS
#include <stdio.h>
int rename(const char *oldpath, const char *newpath);
#include <fcntl.h> /* Definition of AT_* constants */
#include <stdio.h>
int renameat(int olddirfd, const char *oldpath,
int newdirfd, const char *newpath);
int renameat2(int olddirfd, const char *oldpath,
int newdirfd, const char *newpath, unsigned int flags);
Note: There is no glibc wrapper for renameat2(); see NOTES.
Feature Test Macro Requirements for glibc (see feature_test_macros(7)):
renameat():
Since glibc 2.10:
_POSIX_C_SOURCE >= 200809L
Before glibc 2.10:
_ATFILE_SOURCE
DESCRIPTION
rename() renames a file, moving it between directories if required.
Any other hard links to the file (as created using link(2)) are unaf-
fected. Open file descriptors for oldpath are also unaffected.
Various restrictions determine whether or not the rename operation suc-
ceeds: see ERRORS below.
If newpath already exists, it will be atomically replaced, so that
there is no point at which another process attempting to access newpath
will find it missing. However, there will probably be a window in
which both oldpath and newpath refer to the file being renamed.
If oldpath and newpath are existing hard links referring to the same
file, then rename() does nothing, and returns a success status.
If newpath exists but the operation fails for some reason, rename()
guarantees to leave an instance of newpath in place.
oldpath can specify a directory. In this case, newpath must either not
exist, or it must specify an empty directory.
If oldpath refers to a symbolic link, the link is renamed; if newpath
refers to a symbolic link, the link will be overwritten.
renameat()
The renameat() system call operates in exactly the same way as
rename(), except for the differences described here.
If the pathname given in oldpath is relative, then it is interpreted
relative to the directory referred to by the file descriptor olddirfd
(rather than relative to the current working directory of the calling
process, as is done by rename() for a relative pathname).
If oldpath is relative and olddirfd is the special value AT_FDCWD, then
oldpath is interpreted relative to the current working directory of the
calling process (like rename()).
If oldpath is absolute, then olddirfd is ignored.
The interpretation of newpath is as for oldpath, except that a relative
pathname is interpreted relative to the directory referred to by the
file descriptor newdirfd.
See openat(2) for an explanation of the need for renameat().
renameat2()
renameat2() has an additional flags argument. A renameat2() call with
a zero flags argument is equivalent to renameat().
The flags argument is a bit mask consisting of zero or more of the fol-
lowing flags:
RENAME_EXCHANGE
Atomically exchange oldpath and newpath. Both pathnames must
exist but may be of different types (e.g., one could be a non-
empty directory and the other a symbolic link).
RENAME_NOREPLACE
Don't overwrite newpath of the rename. Return an error if new-
path already exists.
RENAME_NOREPLACE can't be employed together with
RENAME_EXCHANGE.
RENAME_WHITEOUT (since Linux 3.18)
This operation makes sense only for overlay/union filesystem
implementations.
Specifying RENAME_WHITEOUT creates a "whiteout" object at the
source of the rename at the same time as performing the rename.
The whole operation is atomic, so that if the rename succeeds
then the whiteout will also have been created.
A "whiteout" is an object that has special meaning in
union/overlay filesystem constructs. In these constructs, mul-
tiple layers exist and only the top one is ever modified. A
whiteout on an upper layer will effectively hide a matching file
in the lower layer, making it appear as if the file didn't
exist.
When a file that exists on the lower layer is renamed, the file
is first copied up (if not already on the upper layer) and then
renamed on the upper, read-write layer. At the same time, the
source file needs to be "whiteouted" (so that the version of the
source file in the lower layer is rendered invisible). The
whole operation needs to be done atomically.
When not part of a union/overlay, the whiteout appears as a
character device with a {0,0} device number.
RENAME_WHITEOUT requires the same privileges as creating a
device node (i.e., the CAP_MKNOD capability).
RENAME_WHITEOUT can't be employed together with RENAME_EXCHANGE.
RENAME_WHITEOUT requires support from the underlying filesystem.
Among the filesystems that provide that support are shmem (since
Linux 3.18), ext4 (since Linux 3.18), and XFS (since Linux 4.1).
RETURN VALUE
On success, zero is returned. On error, -1 is returned, and errno is
set appropriately.
ERRORS
EACCES Write permission is denied for the directory containing oldpath
or newpath, or, search permission is denied for one of the
directories in the path prefix of oldpath or newpath, or oldpath
is a directory and does not allow write permission (needed to
update the .. entry). (See also path_resolution(7).)
EBUSY The rename fails because oldpath or newpath is a directory that
is in use by some process (perhaps as current working directory,
or as root directory, or because it was open for reading) or is
in use by the system (for example as mount point), while the
system considers this an error. (Note that there is no require-
ment to return EBUSY in such cases--there is nothing wrong with
doing the rename anyway--but it is allowed to return EBUSY if
the system cannot otherwise handle such situations.)
EDQUOT The user's quota of disk blocks on the filesystem has been
exhausted.
EFAULT oldpath or newpath points outside your accessible address space.
EINVAL The new pathname contained a path prefix of the old, or, more
generally, an attempt was made to make a directory a subdirec-
tory of itself.
EISDIR newpath is an existing directory, but oldpath is not a direc-
tory.
ELOOP Too many symbolic links were encountered in resolving oldpath or
newpath.
EMLINK oldpath already has the maximum number of links to it, or it was
a directory and the directory containing newpath has the maximum
number of links.
ENAMETOOLONG
oldpath or newpath was too long.
ENOENT The link named by oldpath does not exist; or, a directory compo-
nent in newpath does not exist; or, oldpath or newpath is an
empty string.
ENOMEM Insufficient kernel memory was available.
ENOSPC The device containing the file has no room for the new directory
entry.
ENOTDIR
A component used as a directory in oldpath or newpath is not, in
fact, a directory. Or, oldpath is a directory, and newpath
exists but is not a directory.
ENOTEMPTY or EEXIST
newpath is a nonempty directory, that is, contains entries other
than "." and "..".
EPERM or EACCES
The directory containing oldpath has the sticky bit (S_ISVTX)
set and the process's effective user ID is neither the user ID
of the file to be deleted nor that of the directory containing
it, and the process is not privileged (Linux: does not have the
CAP_FOWNER capability); or newpath is an existing file and the
directory containing it has the sticky bit set and the process's
effective user ID is neither the user ID of the file to be
replaced nor that of the directory containing it, and the
process is not privileged (Linux: does not have the CAP_FOWNER
capability); or the filesystem containing pathname does not sup-
port renaming of the type requested.
EROFS The file is on a read-only filesystem.
EXDEV oldpath and newpath are not on the same mounted filesystem.
(Linux permits a filesystem to be mounted at multiple points,
but rename() does not work across different mount points, even
if the same filesystem is mounted on both.)
The following additional errors can occur for renameat() and
renameat2():
EBADF olddirfd or newdirfd is not a valid file descriptor.
ENOTDIR
oldpath is relative and olddirfd is a file descriptor referring
to a file other than a directory; or similar for newpath and
newdirfd
The following additional errors can occur for renameat2():
EEXIST flags contains RENAME_NOREPLACE and newpath already exists.
EINVAL An invalid flag was specified in flags.
EINVAL Both RENAME_NOREPLACE and RENAME_EXCHANGE were specified in
flags.
EINVAL Both RENAME_WHITEOUT and RENAME_EXCHANGE were specified in
flags.
EINVAL The filesystem does not support one of the flags in flags.
ENOENT flags contains RENAME_EXCHANGE and newpath does not exist.
EPERM RENAME_WHITEOUT was specified in flags, but the caller does not
have the CAP_MKNOD capability.
VERSIONS
renameat() was added to Linux in kernel 2.6.16; library support was
added to glibc in version 2.4.
renameat2() was added to Linux in kernel 3.15.
CONFORMING TO
rename(): 4.3BSD, C89, C99, POSIX.1-2001, POSIX.1-2008.
renameat(): POSIX.1-2008.
renameat2() is Linux-specific.
NOTES
Glibc does not provide a wrapper for the renameat2() system call; call
it using syscall(2).
Glibc notes
On older kernels where renameat() is unavailable, the glibc wrapper
function falls back to the use of rename(). When oldpath and newpath
are relative pathnames, glibc constructs pathnames based on the sym-
bolic links in /proc/self/fd that correspond to the olddirfd and
newdirfd arguments.
BUGS
On NFS filesystems, you can not assume that if the operation failed,
the file was not renamed. If the server does the rename operation and
then crashes, the retransmitted RPC which will be processed when the
server is up again causes a failure. The application is expected to
deal with this. See link(2) for a similar problem.
SEE ALSO
mv(1), chmod(2), link(2), symlink(2), unlink(2), path_resolution(7),
symlink(7)
COLOPHON
This page is part of release 4.15 of the Linux man-pages project. A
description of the project, information about reporting bugs, and the
latest version of this page, can be found at
https://www.kernel.org/doc/man-pages/.
Linux 2017-09-15 RENAME(2)