hacktricks/macos-hardening/macos-security-and-privilege-escalation/macos-security-protections/macos-macf-mandatory-access-control-framework.md
2024-12-12 11:39:29 +01:00

17 KiB

macOS MACF

{% hint style="success" %} Learn & practice AWS Hacking:HackTricks Training AWS Red Team Expert (ARTE)
Learn & practice GCP Hacking: HackTricks Training GCP Red Team Expert (GRTE)

Support HackTricks
{% endhint %}

Basic Information

MACF stands for Mandatory Access Control Framework, which is a security system built into the operating system to help protect your computer. It works by setting strict rules about who or what can access certain parts of the system, such as files, applications, and system resources. By enforcing these rules automatically, MACF ensures that only authorized users and processes can perform specific actions, reducing the risk of unauthorized access or malicious activities.

Note that MACF doesn't really make any decisions as it just intercepts actions, it leaves the decisions to the policy modules (kernel extensions) it calls like AppleMobileFileIntegrity.kext, Quarantine.kext, Sandbox.kext, TMSafetyNet.kext and mcxalr.kext.

Flow

  1. Process performs a syscall/mach trap
  2. The relevant function is called inside the kernel
  3. Function calls MACF
  4. MACF checks policy modules that requested to hook that function in their policy
  5. MACF calls the relevant policies
  6. Policies indicates if they allow or deny the action

{% hint style="danger" %} Apple is the only one that can use the MAC Framework KPI. {% endhint %}

Labels

MACF use labels that then the policies checking if they should grant some access or not will use. The code of the labels struct declaration can be found here, which is then used inside the struct ucred in here in the cr_label part. The label contains flags and s number of slots that can be used by MACF policies to allocate pointers. For example Sanbox will point to the container profile

MACF Policies

A MACF Policy defined rule and conditions to be applied in certain kernel operations.

A kernel extension could configure a mac_policy_conf struct and then register it calling mac_policy_register. From here:

 #define mpc_t	struct mac_policy_conf *

/**
  @brief Mac policy configuration

  This structure specifies the configuration information for a
  MAC policy module.  A policy module developer must supply
  a short unique policy name, a more descriptive full name, a list of label
  namespaces and count, a pointer to the registered enty point operations,
  any load time flags, and optionally, a pointer to a label slot identifier.

  The Framework will update the runtime flags (mpc_runtime_flags) to
  indicate that the module has been registered.

  If the label slot identifier (mpc_field_off) is NULL, the Framework
  will not provide label storage for the policy.  Otherwise, the
  Framework will store the label location (slot) in this field.

  The mpc_list field is used by the Framework and should not be
  modified by policies.
*/
/* XXX - reorder these for better aligment on 64bit platforms */
struct mac_policy_conf {
	const char		*mpc_name;		/** policy name */
	const char		*mpc_fullname;		/** full name */
	const char		**mpc_labelnames;	/** managed label namespaces */
	unsigned int		 mpc_labelname_count;	/** number of managed label namespaces */
	struct mac_policy_ops	*mpc_ops;		/** operation vector */
	int			 mpc_loadtime_flags;	/** load time flags */
	int			*mpc_field_off;		/** label slot */
	int			 mpc_runtime_flags;	/** run time flags */
	mpc_t			 mpc_list;		/** List reference */
	void			*mpc_data;		/** module data */
};

It's easy to identify the kernel extensions configuring these policies by checking calls to mac_policy_register. Moreover, checking the disassemble of the extension it's also possible to find the used mac_policy_conf struct.

Note that MACF policies can be registered and unregistered also dynamically.

One of the main fields of the mac_policy_conf is the mpc_ops. This fied specifies which opreations the policy is interested in. Note that there are hundres of them, so it's possible to zero all of them and then select just the ones the policy is interested on. From here:

struct mac_policy_ops {
	mpo_audit_check_postselect_t		*mpo_audit_check_postselect;
	mpo_audit_check_preselect_t		*mpo_audit_check_preselect;
	mpo_bpfdesc_label_associate_t		*mpo_bpfdesc_label_associate;
	mpo_bpfdesc_label_destroy_t		*mpo_bpfdesc_label_destroy;
	mpo_bpfdesc_label_init_t		*mpo_bpfdesc_label_init;
	mpo_bpfdesc_check_receive_t		*mpo_bpfdesc_check_receive;
	mpo_cred_check_label_update_execve_t	*mpo_cred_check_label_update_execve;
	mpo_cred_check_label_update_t		*mpo_cred_check_label_update;
[...]

Almost all the hooks will be called back by MACF when one of those operations are intercepted. However, mpo_policy_* hooks are an exception because mpo_hook_policy_init() is a callback called upon registration (so after mac_policy_register()) and mpo_hook_policy_initbsd() is called during late registration once the BSD subsystem has initialised properly.

Moreover, the mpo_policy_syscall hook can be registered by any kext to expose a private ioctl style call interface. Then, a user client will be able to call mac_syscall (#381) specifying as parameters the policy name with an integer code and optional arguments.
For example, the Sandbox.kext uses this a lot.

Checking the kext's __DATA.__const* is possible to identify the mac_policy_ops structure used when registering the policy. It's possible to find it because its pointer is at an offset inside mpo_policy_conf and also because the amount of NULL pointers that will be in that area.

Moreover, it's also possible to get the list of kexts that have configured a policy by dumping from memory the struct _mac_policy_list which is updated with every policy that is registered.

MACF Initialization

MACF is initialised very soon. It's set up in XNU's bootstrap_thread: after ipc_bootstrap a call to mac_policy_init() which initializes the mac_policy_list and moments later mac_policy_initmach() is called. Among other things, this function will get all the Apple kexts with the AppleSecurityExtension key in their Info.plist like ALF.kext, AppleMobileFileIntegrity.kext, Quarantine.kext, Sandbox.kext and TMSafetyNet.kext and loads them.

MACF Callouts

It's common to find callouts to MACF defined in code like: #if CONFIG_MAC conditional blocks. Moreover, inside these blocks it's possible to find calls to mac_proc_check* which calls MACF to check for permissions to perform certain actions. Moreover, the format of the MACF callouts is: mac_<object>_<opType>_opName.

The object is one of the following: bpfdesc, cred, file, proc, vnode, mount, devfs, ifnet, inpcb, mbuf, ipq, pipe, sysv[msg/msq/shm/sem], posix[shm/sem], socket, kext.
The opType is usually check which will be used to allow or deny the action. However, it's also possible to find notify, which will allow the kext to react to the given action.

You can find an example in https://github.com/apple-oss-distributions/xnu/blob/94d3b452840153a99b38a3a9659680b2a006908e/bsd/kern/kern_mman.c#L621:

int
mmap(proc_t p, struct mmap_args *uap, user_addr_t *retval)
{
[...]
#if CONFIG_MACF
			error = mac_file_check_mmap(vfs_context_ucred(ctx),
			    fp->fp_glob, prot, flags, file_pos + pageoff,
			    &maxprot);
			if (error) {
				(void)vnode_put(vp);
				goto bad;
			}
#endif /* MAC */
[...]

Then, it's possible to find the code of mac_file_check_mmap in https://github.com/apple-oss-distributions/xnu/blob/94d3b452840153a99b38a3a9659680b2a006908e/security/mac_file.c#L174

mac_file_check_mmap(struct ucred *cred, struct fileglob *fg, int prot,
    int flags, uint64_t offset, int *maxprot)
{
	int error;
	int maxp;

	maxp = *maxprot;
	MAC_CHECK(file_check_mmap, cred, fg, NULL, prot, flags, offset, &maxp);
	if ((maxp | *maxprot) != *maxprot) {
		panic("file_check_mmap increased max protections");
	}
	*maxprot = maxp;
	return error;
}

Which is calling the MAC_CHECK macro, whose code can be found in https://github.com/apple-oss-distributions/xnu/blob/94d3b452840153a99b38a3a9659680b2a006908e/security/mac_internal.h#L261

/*
 * MAC_CHECK performs the designated check by walking the policy
 * module list and checking with each as to how it feels about the
 * request.  Note that it returns its value via 'error' in the scope
 * of the caller.
 */
#define MAC_CHECK(check, args...) do {                              \
    error = 0;                                                      \
    MAC_POLICY_ITERATE({                                            \
	    if (mpc->mpc_ops->mpo_ ## check != NULL) {              \
	            DTRACE_MACF3(mac__call__ ## check, void *, mpc, int, error, int, MAC_ITERATE_CHECK); \
	            int __step_err = mpc->mpc_ops->mpo_ ## check (args); \
	            DTRACE_MACF2(mac__rslt__ ## check, void *, mpc, int, __step_err); \
	            error = mac_error_select(__step_err, error);         \
	    }                                                           \
    });                                                             \
} while (0)

Which will go over all the registered mac policies calling their functions and storing the output inside the error variable, which will only be overridable by mac_error_select by success codes so if any check fails the complete check will fail and the action won't be allowed.

{% hint style="success" %} However, remember that not all MACF callouts are used only to deny actions. For example, mac_priv_grant calls the macro MAC_GRANT, which will grant the requested privilege if any policy answers with a 0:

/*
 * MAC_GRANT performs the designated check by walking the policy
 * module list and checking with each as to how it feels about the
 * request.  Unlike MAC_CHECK, it grants if any policies return '0',
 * and otherwise returns EPERM.  Note that it returns its value via
 * 'error' in the scope of the caller.
 */
#define MAC_GRANT(check, args...) do {                              \
    error = EPERM;                                                  \
    MAC_POLICY_ITERATE({                                            \
	if (mpc->mpc_ops->mpo_ ## check != NULL) {                  \
	        DTRACE_MACF3(mac__call__ ## check, void *, mpc, int, error, int, MAC_ITERATE_GRANT); \
	        int __step_res = mpc->mpc_ops->mpo_ ## check (args); \
	        if (__step_res == 0) {                              \
	                error = 0;                                  \
	        }                                                   \
	        DTRACE_MACF2(mac__rslt__ ## check, void *, mpc, int, __step_res); \
	    }                                                           \
    });                                                             \
} while (0)

{% endhint %}

priv_check & priv_grant

These callas are meant to check and provide (tens of) privileges defined in bsd/sys/priv.h.
Some kernel code would call priv_check_cred() from bsd/kern/kern_priv.c with the KAuth credentials of the process and one of the privileges code which will call mac_priv_check to see if any policy denies giving the privilege and then it calls mac_priv_grant to see if any policy grants the privilege.

proc_check_syscall_unix

This hook allows to intercept all system calls. In bsd/dev/[i386|arm]/systemcalls.c it's possible to see the declared function unix_syscall, which contains this code:

#if CONFIG_MACF
	if (__improbable(proc_syscall_filter_mask(proc) != NULL && !bitstr_test(proc_syscall_filter_mask(proc), syscode))) {
		error = mac_proc_check_syscall_unix(proc, syscode);
		if (error) {
			goto skip_syscall;
		}
	}
#endif /* CONFIG_MACF */

Which will check in the calling process bitmask if the current syscall should call mac_proc_check_syscall_unix. This is because syscalls are called so frequently that it's interesting to avoid calling mac_proc_check_syscall_unix every time.

Note that the function proc_set_syscall_filter_mask(), which set the bitmask syscalls in a process is called by Sandbox to set masks on sandboxed processes.

Exposed MACF syscalls

It's possible to interact with MACF through some syscalls defined in security/mac.h:

/*
 * Extended non-POSIX.1e interfaces that offer additional services
 * available from the userland and kernel MAC frameworks.
 */
#ifdef __APPLE_API_PRIVATE
__BEGIN_DECLS
int      __mac_execve(char *fname, char **argv, char **envv, mac_t _label);
int      __mac_get_fd(int _fd, mac_t _label);
int      __mac_get_file(const char *_path, mac_t _label);
int      __mac_get_link(const char *_path, mac_t _label);
int      __mac_get_pid(pid_t _pid, mac_t _label);
int      __mac_get_proc(mac_t _label);
int      __mac_set_fd(int _fildes, const mac_t _label);
int      __mac_set_file(const char *_path, mac_t _label);
int      __mac_set_link(const char *_path, mac_t _label);
int      __mac_mount(const char *type, const char *path, int flags, void *data,
    struct mac *label);
int      __mac_get_mount(const char *path, struct mac *label);
int      __mac_set_proc(const mac_t _label);
int      __mac_syscall(const char *_policyname, int _call, void *_arg);
__END_DECLS
#endif /*__APPLE_API_PRIVATE*/

References

{% hint style="success" %} Learn & practice AWS Hacking:HackTricks Training AWS Red Team Expert (ARTE)
Learn & practice GCP Hacking: HackTricks Training GCP Red Team Expert (GRTE)

Support HackTricks
{% endhint %}