File permissions¶

Access to files and directories is controlled by setting permissions, on a UNIX system the base permissions are set using POSIX and can be extended with Access Control Lists (ACLs).

POSIX¶

Most UNIX file systems use the Portable Operating System Interface (POSIX) to control who can access files and directories. POSIX uses a combination of three groups of users, known as access classes:

user each user identified by a user-id
group a group of users identified by a group-id, on NCI groups are usually associated to a project, and occasionally to a specific role
other the definition of this class depends on each file/directory context, as it represents anyone else who is not the user or a member of the group.

Each file or directory is owned by only one user and one group, only the owner or a system administrator can change its permissions.

There are three kinds of permissions:

read - the permission to read a file. When set for a directory, only names of files in the directory can be listed, other details like file type, size, ownership, permissions etc. are not visible.
write - the permission to modify or remove a file. When set for a directory, it grants the ability to modify entries in the directory, including creating, deleting, and renaming files. Note that, without execute permission on a directory, the write permission does not take any effect.
execute - the permission to execute a file. When set for a directory, it is interpreted as the search permission: it grants the ability to access file contents and meta-information if its name is known, but does not list files inside the directory, unless read is set.

Hidden permissions¶

Most of this section is taken from the NCI user guide.

In addition to the 9 bits required to store the user, group, and other permissions of a file, there are three kinds of “additional permissions” that are rarely used but are extremely important to understand. The additional permissions are known as the setuid bit, setgid bit, and sticky bit. Each special bit has a different function depending on where it is used.

setuid

The setuid bit on an executable file means that the file will run as the userid of the file’s owner as opposed to the userid of the user executing the file. It is considered as a security hazard and will have no effect when setting the execute bit to s in the user owner triad on Gadi.

setgid

The setgid bit is similar to the setuid bit but set the permissions for its group owner. It enables a command to be run as the project that owns the file rather than the user’s default project. On Gadi, again, for security concerns, it is ignored when set on standard files, but it can be used on directories to force new files and directories created inside them to inherit their group ownership.

chmod g+s / g-s 

sticky bit

When a directory has the sticky bit set, its files can be deleted or renamed only by the file owner, directory owner and the root user. The command below shows how the sticky bit can be set.

chmod +t / -t

This is useful to keep writing permissions for anyone in the group but avoid someone accidentally deleting a file they don’t own.

umask

The default permission for newly created files is set in the umask value. On Gadi, umask 022 is the default configuration in the system .bashrc file. Users can overwrite it in their own file ~/.bashrc.

The command umask sets the inverse mask of the default file access permissions. The OS restricts the permissions of newly created files to no more than the defined permissions by the umask value.

Note

What happens when a file is created on Gadi

Every time a new file or directory is created, the first set of permissions is automatically created based on the user who created them and their account settings.

all project’s folders have rwxrws--- permissions, hence a user has to be part of the group to read/write in them. The s in the group execute position shows that the setgid is active, so any new file will inherit the directory group rather than the user group
the owner of the file is the user who first creates the file
if the file is created in $HOME then the file group will be the user’s $PROJECT, this is usually set in $HOME/.rashrc
the permissions of the file are determined by the umask file
if a file is created by running a program with active setuid and setgid, then they will determine the permissions
if the file is in a directory with the sticky bit activated, then only the file owner, directory owner or root will be able to remove the file, regardless of other permissions.

Changing permissions¶

A file/directory owner is the only user that can set and change a file’s permissions and/or group (aside from a system administrator). Unix commands are:

chmod to change permissions;
chgrp to change group chgrp;
chown to change owner.

Warning

chown can only be used by an NCI system administrator. A request can be made to the NCI helpdesk and both the current and new owner should be cc-ed, occasionally NCI might also request the Lead CI of the project to authorise the transaction.

In most cases just using the POSIX permission system is sufficient to handle who can view or write a specific file. The limitations of this system are obvious when there is the need to set different levels of permissions for different groups of users. In that case, the distinction between other and group might not be sufficient.

An example of this is when managing a data collection at NCI. NCI policy dictates only users in the group can view a project’s files. It is not allowed to have an other reading permission at the main project directory level. This means that it is not possible to have only a subset of the group users (i.e., the data managers) having writing permission. This is where Access Control Lists (ACLs) enter into play, as with ACLs is possible to set permissions for a different group and/or user. In the data management context this usually happens by setting a writers group which can be joined by all the managers and then using ACLs to give this group writing permission.

Access Control Lists¶

Access Control Lists (ACLs) are used to impose a finer level of permissions control on files and directories.

ACLs cheat sheet¶

Warning

Only the file/directory owner can change ACLs, as for any other permissions. If a user runs this command on a directory containing some files they do not own, they will get a warning like “Operation not permitted”. These warnings can be safely ignored, if there are files that they own, the ACLs will be successfully updated for these.

Troubleshooting¶

ACLs are very useful but tend to get broken easily.

Let’s see first how to interpret the getfacl output

getfacl Download/
# file: Download/
# owner: pxp581
# group: ua8
# flags: -s-
user::rwx
user:ua8_nfs:rwx
group::r-x
mask::rwx
other::--x
default:user::rwx
default:user:pxp581:rwx
default:group::r-x
default:mask::rwx
default:other::r-x

The first three lines refer to the normal POSIX permissions, in this case we can see that pxp581 is the owner, ua8 the group and that this directory has setgid (flag: s) set. So, any new file in it will belong to ua8 too.

The next 5 lines show the POSIX permissions (user, group and other), a permission for user ua8_nfs to rwx set by ACLs and the mask also rwx. What the mask does is to set the upper boundary of what permissions can be. It affects only the permissions of the group and/or any extended user and group added by ACLs. The mask has no affect on other or the main user permissions. However, as the directory has setgid, set the directory other permissions act as a mask for the file other permissions. So, if a file with write permissions was copied in here, the new file will still have only r-x permissions, as the file can’t get more permissions on ‘other’ than the directory other r-x permissions.

Order of operations

When a process requests access to a file system object, two steps are performed.

The ACL entry that most closely matches the requesting process is selected. The ACL entries are looked at in the following order: owner, named users, (owning or named) groups, others. Only the highest ranking entry determines access: i.e. if the user is the owner this will have precedence on the group permissions.
The system checks if the matching entry contains sufficient permissions to execute action.

The effective permissions are set to the permissions defined in the mode, minus the permissions set in the current umask. However, the umask has no affect if a default ACL exists.

The mask entry is automatically created when needed but not provided. Its permissions are set to the union of the permissions of all entries that are in the group class, so initially the mask entry does not mask any permissions.

Here is an example of ‘broken’ ACLs:

# file: test.txt 
# owner: pxp581
# group: hh5
user::rw-
group::r-x			#effective:r--
group:w40:rwx			#effective:rw-
group:hh5_w:rwx			#effective:rw-
mask::rw-
other::r--

As some process has set the mask to rw-, the group will have effectively r-- instead of r-x permissions. And similarly, the groups w40 and hh5_w ACL entry contains permissions that are disabled by the mask entry, so getfacl adds a comment that shows the effective set of permissions granted by that entry.
If we grant again writing permissions to the group using chmod the mask will change, and the ACLs will be again effective. The ACLs can be disabled but not altered by chmod.

Any command that sets the mode parameter can have unexpected effects on ACLs. Aside from setacls and chmod, these are usually commands used to create, move, and copy files.

Warning

Please note that if the ACLs were ignored because mv or cp -a were used, simply copying the file again with cp only won’t preserve the ACLs, the file needs to be deleted first and then copied again.

References

Climate Data Guidelines

File permissions

Contents

File permissions¶

POSIX¶

Hidden permissions¶

Changing permissions¶

Access Control Lists¶

ACLs cheat sheet¶

Troubleshooting¶