The Filesystem

This contains links to information used in the lectures.

Logical Structure

The UNIX filesystem, with symbolic links, is technically a general graph, since it can have loops. In the absence of such errors, it is an acyclic graph. Nonetheless it is still commonly referred to as a tree structure rooted at the root directory /. An important concept is the pathname, or route from / through the directory structure to a file. Terminology:

full pathname - Every file and directory in the file system can be identified by a complete list of the names of the directories that are on the route from the root directory to that file or directory.
Relative pathnames - You can define a file or directory by its location in relation to your current directory. The pathname, which must not start with slash (/) is given as / (slash) separated list of the directories on the route to the file (or directory) from your current directory.
Filename - the pathname with the directory part.

Limitations:

directory names must be less than 256 characters
pathnames names must be less than 1023 characters

Filesystems

Filesystems are subtrees of the directory hierarchy rooted at a particular directory called the filesystem root.

They are attached to the file tree with mount command, which maps an existing directory in the tree, called the mount point to the filesystem root.
Note that the previous contents of the mount point and all its subdirectories become inaccessible.
Mount points are usually empty directories.

Filesystems are unmounted using the umount command. Normally you cannot unmount a busy filesystem, that is there must be no open files and no processes executing in a directory there. BSD and OSF/1 allow forcing unmounting of busy filesystem using umount -f - this is a bad idea.

Finding the processes that are preventing unmounting is a pain. There is a program lsof - list of open files - by Victor Abell of Purdue which lists them. Click here to obtain by ftp.

Organization of the Filesystem

The Unix filesystem was never well organized and has become worse as time has elapsed, with the proliferation of symbolic links. In principle, one would like files which change to be in separate locations to those which are static, but this is not the case. There are some almost standard directories.

File Types

Regular - just a sequence of bytes
Directories: names of files and number of the associated i-node ; mkdir, rmdir, rm -r
Hard Links - just an additional name for a file, actually two directory entries which point to same inode. Cannot have links across disk partitions. Unix keeps count of number of links to a file. Created with ln.
Symbolic Links - actually a file with an absolute or relative path to the file in it. Unix can distinguish them by bit in inode. Create with ln -s. Be aware of problem with using .. after traversing symbolic link.
Special (i/o devices usually in /dev)
- Character - raw unbuffered in kernel
- Block - kernel performs buffering, written in blocks
  - Device numbers (major, minor) - major gives device driver, minor individual device
  - Some device use minor device to determine density (tapes), or outgoing v incoming modems (terminal)
- UNIX domain sockets- allow communication between 2 unrelated processes on same host. They cannot be read or written by processes not involved in connection. Used in printing, X, etc. Create with socket, remove with rm or unlink.
- Named pipes (ATT) - allow communication between 2 unrelated processes on same host. Create with mknod.
Except directories all can be removed with rm.
Results of ls -l on different file types
```
dr-xr-xr-x  12 bin      bin         4096 Jan 25 14:37 /dev/
-rwxr-xr-x   1 farrell  faculty    16384 Sep 28  1994 access*
-rw-r-xr-x   1 farrell  faculty      542 Sep 26  1994 access.c*
prw-r-xr-x   1 farrell  faculty        0 Sep 26  1994 anamedpipe*
-rw-r-xr-x   2 farrell  faculty       50 Sep 26  1994 bar*
lrwxr-xr-x   1 farrell  faculty        2 Sep 22 01:04 here@ -> ..
lrwxr-xr-x   1 farrell  faculty        3 Sep 22 01:04 new@ -> bar
crw-rw-rw-   2 root     other     17 0x000020 Apr  2  1993 /dev/ttyr0
crw-rw-rw-   2 root     other     17 0x000021 Apr  2  1993 /dev/ttyr1
crw-r-----   1 root     sys       47 0x201000 Dec  1  1993 /dev/rdsk/c201d0s0
crw-r-----   1 root     sys       47 0x201100 Apr  2  1993 /dev/rdsk/c201d1s0
brw-r-----   1 root     sys        7 0x201000 Dec  1  1993 /dev/dsk/c201d0s0
brw-r-----   1 root     sys        7 0x201100 Apr  2  1993 /dev/dsk/c201d1s0
```
Permissions
- Normal permissions - bits with octal values 0 through 0777
- Meaning on Files and Directories
- Setuid and Setgid bits - bits with octal value 4000 and 2000
  Set to permit executables to execute with effective UID or effective GID of the file rather than of the person executing it. This gives accesses not otherwise available. In SunOS can setgid on directory to control group of files created there.
- Sticky Bit - bit with octal value 1000
  Obsolete - used on small memory systems (PDP 11) to keep programs resident. In some versions of Unix it has been reused - if set on directories it prevents deletion or renaming unless you are owner of the directory or file, or superuser. This is to improve privacy of directories such as /tmp which are writable by all.
Changing permission
Changing access permissions can be done symbolically or numerically.
Default permissions
Default permissions can be set by changing the file-creation mode mask which is a number consisting of three octal digits which is subtracted from the mode. This is set with umask.
Inodes
The kernel maintains file information in a structure called the inode. These are organized in a table, which is created when the filesystem is created. The inode contains 40 or more fields, the most interesting of which are:
- link count - number of hard links
- owner
- group
- mode
- size
- last access time
- last modification time
- last change time
- type (regular, directory, special file)
There may be multiple directory entries (hardlinks) pointing to a file, but there is only one inode. The number can be determined with ls -l.
Changing owner and group
This is accomplished with chown and chgrp. On HP-UX in order to change the owner or group, you must own the file and have the CHOWN privilege (see setprivgrp(1M)). If either command is invoked on a regular file by other than the superuser, the set-user-ID and set-group-ID bits of the file mode (04000 and 02000 respectively) are cleared. Note that a given user's or group's ability to use this command can be restricted by setprivgrp.

Paul A. Farrell
Monday, January 30, 1996

The Filesystem

Logical Structure

Filesystems

Organization of the Filesystem

File Types

Permissions

Changing permission

Default permissions

Inodes

Changing owner and group