32 Profile components and syntax #

The default AppArmor profile is attached to a program by its name, so a profile name must match the path to the application it is to confine.

/usr/bin/foo {
...
}

This profile is automatically used whenever an unconfined process executes /usr/bin/foo.

Unattached profiles do not reside in the file system namespace and therefore are not automatically attached to an application. The name of an unattached profile is preceded by the keyword profile. You can freely choose a profile name, except for the following limitations: the name must not begin with a : or . character. If it contains a whitespace, it must be quoted. If the name begins with a /, the profile is considered to be a standard profile, so the following two profiles are identical:

profile /usr/bin/foo {
...
}
/usr/bin/foo {
...
}

Unattached profiles are never used automatically, nor can they be transitioned to through a Px rule. They need to be attached to a program by either using a named profile transition (see Section 32.12.7, “Named profile transitions”) or with the change_profile rule (see Section 32.2.5, “Change rules”).

Unattached profiles are useful for specialized profiles for system utilities that generally should not be confined by a system-wide profile (for example, /bin/bash). They can also be used to set up roles or to confine a user.

Local profiles provide a convenient way to provide specialized confinement for utility programs launched by a confined application. They are specified like standard profiles, except that they are embedded in a parent profile and begin with the profile keyword:

/parent/profile {
   ...
   profile /local/profile {
      ...
   }
}

To transition to a local profile, either use a cx rule (see Section 32.12.2, “Discrete local profile execute mode (cx)”) or a named profile transition (see Section 32.12.7, “Named profile transitions”).

AppArmor "hats" are a local profiles with certain additional restrictions and an implicit rule allowing for change_hat to be used to transition to them. Refer to Chapter 36, Profiling your Web applications using ChangeHat for a detailed description.

AppArmor provides change_hat and change_profile rules that control domain transitioning. change_hat are specified by defining hats in a profile, while change_profile rules refer to another profile and start with the keyword change_profile:

change_profile -> /usr/bin/foobar,

Both change_hat and change_profile provide for an application directed profile transition, without having to launch a separate application. change_profile provides a generic one way transition between any of the loaded profiles. change_hat provides for a returnable parent child transition where an application can switch from the parent profile to the hat profile and if it provides the correct secret key return to the parent profile at a later time.

change_profile is best used in situations where an application goes through a trusted setup phase and then can lower its privilege level. Any resources mapped or opened during the start-up phase may still be accessible after the profile change, but the new profile restricts the opening of new resources, and limits certain resources opened before the switch. Specifically, memory resources are still available while capability and file resources (if they are not memory mapped) can be limited.

change_hat is best used in situations where an application runs a virtual machine or an interpreter that does not provide direct access to the applications resources (for example Apache's mod_php). Since change_hat stores the return secret key in the application's memory the phase of reduced privilege should not have direct access to memory. It is also important that file access is properly separated, since the hat can restrict accesses to a file handle but does not close it. If an application does buffering and provides access to the open files with buffering, the accesses to these files might not be seen by the kernel and hence not restricted by the new profile.

Warning: Safety of domain transitions

The change_hat and change_profile domain transitions are less secure than a domain transition done through an exec because they do not affect a process's memory mappings, nor do they close resources that have already been opened.

Abstractions are includes that are grouped by common application tasks. These tasks include access to authentication mechanisms, access to name service routines, common graphics requirements, and system accounting. Files listed in these abstractions are specific to the named task. Programs that require one of these files also requires other files listed in the abstraction file (depending on the local configuration and the specific requirements of the program). Find abstractions in /etc/apparmor.d/abstractions.

The program-chunks directory (/etc/apparmor.d/program-chunks) contains certain chunks of profiles that are specific to program suites and not generally useful outside of the suite, thus are never suggested for use in profiles by the profile wizards (aa-logprof and aa-genprof). Currently, program chunks are only available for the postfix program suite.

The tunables directory (/etc/apparmor.d/tunables) contains global variable definitions. When used in a profile, these variables expand to a value that can be changed without changing the entire profile. Add all the tunables definitions that should be available to every profile to /etc/apparmor.d/tunables/global.

Profile flags control the behavior of the related profile. You can add profile flags to the profile definition by editing it manually, see the following syntax:

/path/to/profiled/binary flags=(list_of_flags) {
  [...]
}

You can use multiple flags separated by a comma ',' or space ' '. There are three basic types of profile flags: mode, relative, and attach flags.

Mode flag is complain (illegal accesses are allowed and logged). If it is omitted, the profile is in enforce mode (enforces the policy).

Tip

A more flexible way of setting the whole profile into complain mode is to create a symbolic link from the profile file inside the /etc/apparmor.d/force-complain/ directory.

ln -s /etc/apparmor.d/bin.ping /etc/apparmor.d/force-complain/bin.ping

Relative flags are chroot_relative (states that the profile is relative to the chroot instead of namespace) or namespace_relative (the default, with the path being relative to outside the chroot). They are mutually exclusive.

Attach flags consist of two pairs of mutually exclusive flags: attach_disconnected or no_attach_disconnected (determine if path names resolved to be outside of the namespace are attached to the root, which means they have the '/' character at the beginning), and chroot_attach or chroot_no_attach (control path name generation when in a chroot environment while a file is accessed that is external to the chroot but within the namespace).

AppArmor allows to use variables holding paths in profiles. Use global variables to make your profiles portable and local variables to create shortcuts for paths.

A typical example of when global variables come in handy are network scenarios in which user home directories are mounted in different locations. Instead of rewriting paths to home directories in all affected profiles, you only need to change the value of a variable. Global variables are defined under /etc/apparmor.d/tunables and need to be made available via an include statement. Find the variable definitions for this use case (@{HOME} and @{HOMEDIRS}) in the /etc/apparmor.d/tunables/home file.

Local variables are defined at the head of a profile. This is useful to provide the base of for a chrooted path, for example:

@{CHROOT_BASE}=/tmp/foo
/sbin/rsyslogd {
...
# chrooted applications
@{CHROOT_BASE}/var/lib/*/dev/log w,
@{CHROOT_BASE}/var/log/** w,
...
}

In the following example, while @{HOMEDIRS} lists where all the user home directories are stored, @{HOME} is a space-separated list of home directories. Later on, @{HOMEDIRS} is expanded by two new specific places where user home directories are stored.

@{HOMEDIRS}=/home/
@{HOME}=@{HOMEDIRS}/*/ /root/
[...]
@{HOMEDIRS}+=/srv/nfs/home/ /mnt/home/

Note

With the current AppArmor tools, variables can only be used when manually editing and maintaining a profile.

Profile names can contain globbing expressions allowing the profile to match against multiple binaries.

The following example is valid for systems where the foo binary resides either in /usr/bin or /bin.

/{usr/,}bin/foo { ... }

In the following example, when matching against the executable /bin/foo, the /bin/foo profile is an exact match so it is chosen. For the executable /bin/fat, the profile /bin/foo does not match, and because the /bin/f* profile is more specific (less general) than /bin/**, the /bin/f* profile is chosen.

/bin/foo { ... }

/bin/f*  { ... }

/bin/**  { ... }

For more information on profile name globbing examples, see the man page of AppArmor, man 5 apparmor.d,, section Globbing.

Namespaces are used to provide different profiles sets. Say one for the system, another for a chroot environment or container. Namespaces are hierarchical—a namespace can see its children but a child cannot see its parent. Namespace names start with a colon : followed by an alphanumeric string, a trailing colon : and an optional double slash //, such as

:childNameSpace://

Profiles loaded to a child namespace are prefixed with their namespace name (viewed from a parent's perspective):

:childNameSpace://apache

Namespaces can be entered via the change_profile API, or named profile transitions:

/path/to/executable px -> :childNameSpace://apache

Profiles can have a name, and an attachment specification. This allows for profiles with a logical name that can be more meaningful to users/administrators than a profile name that contains pattern matching (see Section 32.6.3, “Pattern matching”). For example, the default profile

/** { ... }

can be named

profile default /** { ... }

Also, a profile with pattern matching can be named. For example:

/usr/lib64/firefox*/firefox-*bin { ... }

can be named

profile firefox /usr/lib64/firefox*/firefox-*bin { ... }

Alias rules provide an alternative way to manipulate profile path mappings to site specific layouts. They are an alternative form of path rewriting to using variables, and are done post variable resolution. The alias rule says to treat rules that have the same source prefix as if the rules are at target prefix.

alias /home/ -> /usr/home/

All the rules that have a prefix match to /home/ provide access to /usr/home/. For example:

/home/username/** r,

allows as well access to

/usr/home/username/** r,

Aliases provide a quick way of remapping rules without the need to rewrite them. They keep the source path still accessible—in our example, the alias rule keeps the paths under /home/ still accessible.

With the alias rule, you can point to multiple targets at the same time.

alias /home/ -> /usr/home/
alias /home/ -> /mnt/home/

Note

With the current AppArmor tools, alias rules can only be used when manually editing and maintaining a profile.

Tip

Insert global alias definitions in the file /etc/apparmor.d/tunables/alias.

Allows the program to have read access to the resource. Read access is required for shell scripts and other interpreted content and determines if an executing process can core dump.

Allows the program to have write access to the resource. Files must have this permission if they are to be unlinked (removed).

Allows a program to write to the end of a file. In contrast to the w mode, the append mode does not include the ability to overwrite data, to rename, or to remove a file. The append permission is typically used with applications who need to be able to write to log files, but which should not be able to manipulate any existing data in the log files. As the append permission is a subset of the permissions associated with the write mode, the w and a permission flags cannot be used together and are mutually exclusive.

The application can take file locks. Former versions of AppArmor allowed files to be locked if an application had access to them. By using a separate file locking mode, AppArmor makes sure locking is restricted only to those files which need file locking and tightens security as locking can be used in several denial-of-service attack scenarios.

The link mode mediates access to hard links. When a link is created, the target file must have the same access permissions as the link created (but the destination does not need link access).

The link mode grants permission to link to arbitrary files, provided the link has a subset of the permissions granted by the target (subset permission test).

/srv/www/htdocs/index.html rl,

By specifying origin and destination, the link pair rule provides greater control over how hard links are created. Link pair rules by default do not enforce the link subset permission test that the standard rules link permission requires.

link /srv/www/htdocs/index.html -> /var/www/index.html

To force the rule to require the test, the subset keyword is used. The following rules are equivalent:

/var/www/index.html l,
link subset /var/www/index.html -> /**,

Note

Currently link pair rules are not supported by YaST and the command-line tools. Manually edit your profiles to use them. Updating such profiles using the tools is safe, because the link pair entries are not touched.

The allow prefix is optional, and it is idiomatically implied if not specified and the deny (see Section 32.7.9, “Deny rules”) keyword is not used.

allow file /example r,
allow /example r,
allow network,

You can also use the optional file keyword. If you omit it and there are no other rule types that start with a keyword, such as network or mount, it is automatically implied.

file /example/rule r,

is equivalent to

/example/rule r,

The following rule grants access to all files:

file,

which is equal to

/** rwmlk,

File rules can use leading or trailing permissions. The permissions should not be specified as a trailing permission, but rather used at the start of the rule. This is important in that it makes file rules behave like any other rule types.

/path rw,            # old style
rw /path,            # leading permission
file rw /path,       # with explicit 'file' keyword
allow file rw /path, # optional 'allow' keyword added

The file rules can be extended so that they can be conditional upon the user being the owner of the file (the fsuid needs to match the file's uid). For this purpose the owner keyword is put in front of the rule. Owner conditional rules accumulate like regular file rules do.

owner /home/*/** rw

When using file ownership conditions with link rules the ownership test is done against the target file so the user must own the file to be able to link to it.

Note: Precedence of regular file rules

Owner conditional rules are considered a subset of regular file rules. If a regular file rule overlaps with an owner conditional file rule, the rules are merged. Consider the following example.

/foo r,
owner /foo rw,  # or w,

The rules are merged—it results in r for everybody, and w for the owner only.

Tip

To address everybody but the owner of the file, use the keyword other.

owner /foo rw,
other /foo r,

Deny rules can be used to annotate or quiet known rejects. The profile generating tools do not ask about a known reject treated with a deny rule. Such a reject does not show up in the audit logs when denied, keeping the log files lean. If this is not desired, put the keyword audit in front of the deny entry.

It is also possible to use deny rules in combination with allow rules. This allows you to specify a broad allow rule, and then subtract a few known files that should not be allowed. Deny rules can also be combined with owner rules, to deny files owned by the user. The following example allows read/write access to everything in a users directory except write access to the .ssh/ files:

deny /home/*/.ssh/** w,
owner /home/*/** rw,

The extensive use of deny rules is generally not encouraged, because it makes it much harder to understand what a profile does. However a judicious use of deny rules can simplify profiles. Therefore the tools only generate profiles denying specific files and do not use globbing in deny rules. Manually edit your profiles to add deny rules using globbing. Updating such profiles using the tools is safe, because the deny entries are not touched.

This mode requires that a discrete security profile is defined for a resource executed at an AppArmor domain transition. If there is no profile defined, the access is denied.

Incompatible with Ux, ux, px, and ix.

As Px, but instead of searching the global profile set, Cx only searches the local profiles of the current profile. This profile transition provides a way for an application to have alternate profiles for helper applications.

Note: Limitations of the discrete local profile execute mode (cx)

Currently, Cx transitions are limited to top level profiles and cannot be used in hats and children profiles. This restriction will be removed in the future.

Incompatible with Ux, ux, Px, px, cx, and ix.

Allows the program to execute the resource without any AppArmor profile applied to the executed resource. This mode is useful when a confined program needs to be able to perform a privileged operation, such as rebooting the machine. By placing the privileged section in another executable and granting unconfined execution rights, it is possible to bypass the mandatory constraints imposed on all confined processes. Allowing a root process to go unconfined means it can change AppArmor policy itself. For more information about what is constrained, see the apparmor(7) man page.

This mode is incompatible with ux, px, Px, and ix.

Use the lowercase versions of exec modes—px, cx, ux—only in special cases. They do not scrub the environment of variables such as LD_PRELOAD. As a result, the calling domain may have an undue amount of influence over the called resource. Use these modes only if the child absolutely must be run unconfined and LD_PRELOAD must be used. Any profile using such modes provides negligible security. Use at your own risk.

ix prevents the normal AppArmor domain transition on execve(2) when the profiled program executes the named program. Instead, the executed resource inherits the current profile.

This mode is useful when a confined program needs to call another confined program without gaining the permissions of the target's profile or losing the permissions of the current profile. There is no version to scrub the environment because ix executions do not change privileges.

Incompatible with cx, ux, and px. Implies m.

This mode allows a file to be mapped into memory using mmap(2)'s PROT_EXEC flag. This flag marks the pages executable. It is used on certain architectures to provide non executable data pages, which can complicate exploit attempts. AppArmor uses this mode to limit which files a well-behaved program (or all programs on architectures that enforce non executable memory access controls) may use as libraries, to limit the effect of invalid -L flags given to ld(1) and LD_PRELOAD, LD_LIBRARY_PATH, given to ld.so(8).

By default, the px and cx (and their clean exec variants, too) transition to a profile whose name matches the executable name. With named profile transitions, you can specify a profile to be transitioned to. This is useful if multiple binaries need to share a single profile, or if they need to use a different profile than their name would specify. Named profile transitions can be used with cx, Cx, px and Px. Currently there is a limit of twelve named profile transitions per profile.

Named profile transitions use -> to indicate the name of the profile that needs to be transitioned to:

/usr/bin/foo
{
  /bin/** px -> shared_profile,
  ...
  /usr/*bash cx -> local_profile,
  ...
  profile local_profile
  {
    ...
  }
}

Note: Difference between normal and named transitions

When used with globbing, normal transitions provide a “one to many” relationship—/bin/** px transitions to /bin/ping, /bin/cat, etc, depending on the program being run.

Named transitions provide a “many to one” relationship—all programs that match the rule regardless of their name transition to the specified profile.

Named profile transitions show up in the log as having the mode Nx. The name of the profile to be changed to is listed in the name2 field.

The px and cx transitions specify a hard dependency—if the specified profile does not exist, the exec fails. With the inheritance fallback, the execution succeeds but inherits the current profile. To specify inheritance fallback, ix is combined with cx, Cx, px and Px into the modes cix, Cix, pix and Pix.

/path Cix -> profile_name,

or

Cix /path -> profile_name,

where -> profile_name is optional.

The same applies if you add the unconfined ux mode, where the resulting modes are cux, CUx, pux and PUx. These modes allow falling back to “unconfined” when the specified profile is not found.

/path PUx -> profile_name,

or

PUx /path -> profile_name,

where -> profile_name is optional.

The fallback modes can be used with named profile transitions, too.

When choosing one of the Px, Cx or Ux execution modes, take into account that the following environment variables are removed from the environment before the child process inherits it. As a consequence, applications or processes relying on any of these variables do not work anymore if the profile applied to them carries Px, Cx or Ux flags:

You can use the safe and unsafe keywords for rules instead of using the case modifier of execution modes. For example

/example_rule Px,

is the same as any of the following

safe /example_rule px,
safe /example_rule Px,
safe px /example_rule,
safe Px /example_rule,

and the rule

/example_rule px,

is the same as any of

unsafe /example_rule px,
unsafe /example_rule Px,
unsafe px /example_rule,
unsafe Px /example_rule,

The safe/unsafe keywords are mutually exclusive and can be used in a file rule after the owner keyword, so the order of rule keywords is

[audit] [deny] [owner] [safe|unsafe] file_rule