Copy a directory

Written by masteryeti on December 30, 2020.

Using cp (POSIX)

The following command is used to copy a directory from one place to another:


cp -aTu /path/to/directory-1/ /path/to/directory-2/

cp is short for "copy".
-a is short for "archive", which means that:
- no symbolic links (shortcuts) from the source directory are followed;
- all attributes are preserved (mode, ownership, timestamps, context, links, xattr, etc.);
- all files and directories within the source directory will be copied recursively.
-T is an alias for "--no-target-directory", which means that instead of putting the source directory into the target directory, it puts the contents of the source directory into the target directory (similar to a move).
-u is short for "update", which means that only files are copied which either do not yet exist in the target directory or have a more recent time (last modification date). This is optional, and is lazy, which sometimes means saving a lot of time.
/path/to/directory-1/ is the source directory from which to copy.
/path/to/directory-2/ is the target directory.

Using ssh and tar

In order to copy a local directory onto another machine, using ssh:


tar cf - /path/to/directory-1/ | ssh -T -p 22 username@example.com tar xfC - /path/to/directory-2/

Or similarly, to copy a remote directory to a local machine:


ssh -T -p 22 username@example.com tar cf - /path/to/directory-1/ | tar xfC - /path/to/directory-2/

Using rsync

If either the source or target directory is on another machine, you may transfer it over an ssh tunnel, using rsync.


rsync -aAX -P -c -e "ssh -p 22" --inplace --no-whole-file /path/to/directory-1/ username@example.com:/path/to/directory-2/

-aAX makes an exact copy of the directory (recursively). This includes reproducing character/block devices, special files (sockets, named pipes, etc.), symlinks (so no follow), permissions, times (last modified, last access, etc.), owner/group, ACL entries, extended attributes, and all that recursively. Except hardlinks, for this, specify -H.
-P is short for --partial --progress, to show progress and allow partial files so that rsync may be interrupted halfway and then later continue.
-c enables checksum checking, instead of transferring files based on their last modified time.
--inplace --no-whole-file allows rsync to seamlessly continue transfer after interrupted (i.e. CTRL+C or connection lost). Use this only for potentially unstable network connections, and only on files that are not actively in use.

In order to synchronize, instead of copying, add --delete. This makes rsync delete files in the target directory, that do not exist in the source directory.

For local file transfers, not over a network connection, use:


rsync -aAX -P -c /path/to/directory-1/ /path/to/directory-2/

Using sshfs

This allows any copy utility to be used for any remote location accessible by ssh, including cp. It works by mounting a remote directory into a local mount point. Under the hood it uses FUSE for the filesystem integration, and SFTP for the actual transfer.

The following copies the local contents of /path/to/directory-1/ into the remote directory /path/to/directory-2/.


mkdir /tmp/remote
sshfs -p 22 username@example.com:/path/to/directory-2/ /tmp/remote/
cp -a /path/to/directory-1/ /mnt/remote/
umount /tmp/remote && rmdir /tmp/remote

Whereas the following copies the directory itself, like so: /path/to/directory-2/directory-1/.


mkdir /tmp/remote
sshfs -p 22 username@example.com:/path/to/ /tmp/remote/
cp -a /path/to/directory-1/ /mnt/remote/directory-2/
umount /tmp/remote && rmdir /tmp/remote