Chapter 3: Files and directories

Hard links and symbolic links

A file can have more than one name. Linux offers two ways to do it: a hard link, which is a second name for the same data, and a symbolic link, which is a small file that points to another name. They look alike in daily use but behave very differently when the original is moved or deleted.

• Explain how a file name, its inode and its data relate, and read the link count in ls -l.
• Create hard links and symbolic links with ln and ln -s, and recognise each in a listing.
• Predict what happens to each kind of link when the original name is removed, and know their limits.

Names, inodes and data

On a Linux file system, a file's name and the file itself are separate things. The file is an inode: a numbered record holding its owner, permissions, time stamps and where its data is on disk. A directory is just a list of names, each pointing to an inode number. ls -i shows the inode numbers:

Names, inodes and data: Select a view to compare the two kinds of link.

Hard links: a second name

ln EXISTING NEWNAME adds another name for the same inode. Both names are equally real: neither is the "original" any more. The link count in ls -l (the number after the permissions) goes up:

[student@servera ~]$ echo "Release checklist" > checklist.txt
[student@servera ~]$ ln checklist.txt /tmp/checklist-copy.txt
[student@servera ~]$ ls -li checklist.txt /tmp/checklist-copy.txt
261894 -rw-r--r--. 2 student student 18 Oct  3 09:58 checklist.txt
261894 -rw-r--r--. 2 student student 18 Oct  3 09:58 /tmp/checklist-copy.txt

Same inode number, link count 2. Change the content through one name and the other shows the change, because there is only one file:

[student@servera ~]$ echo "1. Back up the database" >> /tmp/checklist-copy.txt
[student@servera ~]$ cat checklist.txt
Release checklist
1. Back up the database

(>> adds a line to the end of a file; chapter 5 explains it.) Removing a name only removes that name. The data stays until the last name is gone:

[student@servera ~]$ rm checklist.txt
[student@servera ~]$ ls -l /tmp/checklist-copy.txt
-rw-r--r--. 1 student student 42 Oct  3 09:58 /tmp/checklist-copy.txt

Hard links have two limits:

[student@servera ~]$ ln /var/log loghard
ln: /var/log: hard link not allowed for directory
[student@servera ~]$ ln note.txt /dev/shm/note.txt
ln: failed to create hard link '/dev/shm/note.txt' => 'note.txt': Invalid cross-device link

• They work for files only, never directories.
• Both names must be on the same file system, because an inode number only means something within its own file system. df shows which file system a directory is on: here /home is on /dev/sda2 and /dev/shm is a separate in-memory one.

Symbolic links: a pointer to a name

ln -s TARGET LINKNAME creates a symbolic link (or soft link): a tiny file whose content is just a path. Opening the link opens whatever that path leads to. In ls -l it starts with l and shows -> and its target:

[student@servera ~]$ ln -s /tmp/checklist-copy.txt latest.txt
[student@servera ~]$ ls -l latest.txt
lrwxrwxrwx. 1 student student 23 Oct  3 09:58 latest.txt -> /tmp/checklist-copy.txt
[student@servera ~]$ cat latest.txt
Release checklist
1. Back up the database

Symbolic links have neither of the hard link's limits: they can point to directories and across file systems. The price is that they point to a name, not to the data. Remove or rename the target and the link dangles:

[student@servera ~]$ rm /tmp/checklist-copy.txt
[student@servera ~]$ cat latest.txt
cat: latest.txt: No such file or directory

The link still exists, still pointing at a name that no longer exists. Create a new file with that name and the link works again, pointing at the new file.

Links to directories

A symbolic link to a directory behaves like the directory. cd into it and pwd shows the path you came through; pwd -P shows the real one:

[student@servera ~]$ ln -s /var/log logs
[student@servera ~]$ cd logs
[student@servera logs]$ pwd
/home/student/logs
[student@servera logs]$ pwd -P
/var/log

You saw this pattern at the start of the chapter: /bin is a symbolic link to usr/bin. readlink LINK prints where a link points, and readlink -f follows every link to the final real path.

Which one to use

In practice, symbolic links are what you will create most: a short name for a deep directory, a stable name such as current that you repoint at each new release, or a compatibility name like /bin. Hard links appear in backup tools, which use them to store identical files only once.

Check your understanding

• ls -l shows a link count of 3 for a regular file. What does that mean?
• You remove the original name of a file that also has a hard link. What happens to the data?
• Which needs a symbolic link rather than a hard link?
• What does ls -l show for a symbolic link?
• A symbolic link points to a file that has been deleted. What is the link called, and what does cat print?