1 ========================================================
2 Linux Security Modules: General Security Hooks for Linux
3 ========================================================
5 :Author: Stephen Smalley
6 :Author: Timothy Fraser
11 The APIs described in this book are outdated.
16 In March 2001, the National Security Agency (NSA) gave a presentation
17 about Security-Enhanced Linux (SELinux) at the 2.5 Linux Kernel Summit.
18 SELinux is an implementation of flexible and fine-grained
19 nondiscretionary access controls in the Linux kernel, originally
20 implemented as its own particular kernel patch. Several other security
21 projects (e.g. RSBAC, Medusa) have also developed flexible access
22 control architectures for the Linux kernel, and various projects have
23 developed particular access control models for Linux (e.g. LIDS, DTE,
24 SubDomain). Each project has developed and maintained its own kernel
25 patch to support its security needs.
27 In response to the NSA presentation, Linus Torvalds made a set of
28 remarks that described a security framework he would be willing to
29 consider for inclusion in the mainstream Linux kernel. He described a
30 general framework that would provide a set of security hooks to control
31 operations on kernel objects and a set of opaque security fields in
32 kernel data structures for maintaining security attributes. This
33 framework could then be used by loadable kernel modules to implement any
34 desired model of security. Linus also suggested the possibility of
35 migrating the Linux capabilities code into such a module.
37 The Linux Security Modules (LSM) project was started by WireX to develop
38 such a framework. LSM is a joint development effort by several security
39 projects, including Immunix, SELinux, SGI and Janus, and several
40 individuals, including Greg Kroah-Hartman and James Morris, to develop a
41 Linux kernel patch that implements this framework. The patch is
42 currently tracking the 2.4 series and is targeted for integration into
43 the 2.5 development series. This technical report provides an overview
44 of the framework and the example capabilities security module provided
45 by the LSM kernel patch.
50 The LSM kernel patch provides a general kernel framework to support
51 security modules. In particular, the LSM framework is primarily focused
52 on supporting access control modules, although future development is
53 likely to address other security needs such as auditing. By itself, the
54 framework does not provide any additional security; it merely provides
55 the infrastructure to support security modules. The LSM kernel patch
56 also moves most of the capabilities logic into an optional security
57 module, with the system defaulting to the traditional superuser logic.
58 This capabilities module is discussed further in
59 `LSM Capabilities Module <#cap>`__.
61 The LSM kernel patch adds security fields to kernel data structures and
62 inserts calls to hook functions at critical points in the kernel code to
63 manage the security fields and to perform access control. It also adds
64 functions for registering and unregistering security modules, and adds a
65 general :c:func:`security()` system call to support new system calls
66 for security-aware applications.
68 The LSM security fields are simply ``void*`` pointers. For process and
69 program execution security information, security fields were added to
70 :c:type:`struct task_struct <task_struct>` and
71 :c:type:`struct linux_binprm <linux_binprm>`. For filesystem
72 security information, a security field was added to :c:type:`struct
73 super_block <super_block>`. For pipe, file, and socket security
74 information, security fields were added to :c:type:`struct inode
75 <inode>` and :c:type:`struct file <file>`. For packet and
76 network device security information, security fields were added to
77 :c:type:`struct sk_buff <sk_buff>` and :c:type:`struct
78 net_device <net_device>`. For System V IPC security information,
79 security fields were added to :c:type:`struct kern_ipc_perm
80 <kern_ipc_perm>` and :c:type:`struct msg_msg
81 <msg_msg>`; additionally, the definitions for :c:type:`struct
82 msg_msg <msg_msg>`, struct msg_queue, and struct shmid_kernel
83 were moved to header files (``include/linux/msg.h`` and
84 ``include/linux/shm.h`` as appropriate) to allow the security modules to
85 use these definitions.
87 Each LSM hook is a function pointer in a global table, security_ops.
88 This table is a :c:type:`struct security_operations
89 <security_operations>` structure as defined by
90 ``include/linux/security.h``. Detailed documentation for each hook is
91 included in this header file. At present, this structure consists of a
92 collection of substructures that group related hooks based on the kernel
93 object (e.g. task, inode, file, sk_buff, etc) as well as some top-level
94 hook function pointers for system operations. This structure is likely
95 to be flattened in the future for performance. The placement of the hook
96 calls in the kernel code is described by the "called:" lines in the
97 per-hook documentation in the header file. The hook calls can also be
98 easily found in the kernel code by looking for the string
101 Linus mentioned per-process security hooks in his original remarks as a
102 possible alternative to global security hooks. However, if LSM were to
103 start from the perspective of per-process hooks, then the base framework
104 would have to deal with how to handle operations that involve multiple
105 processes (e.g. kill), since each process might have its own hook for
106 controlling the operation. This would require a general mechanism for
107 composing hooks in the base framework. Additionally, LSM would still
108 need global hooks for operations that have no process context (e.g.
109 network input operations). Consequently, LSM provides global security
110 hooks, but a security module is free to implement per-process hooks
111 (where that makes sense) by storing a security_ops table in each
112 process' security field and then invoking these per-process hooks from
113 the global hooks. The problem of composition is thus deferred to the
116 The global security_ops table is initialized to a set of hook functions
117 provided by a dummy security module that provides traditional superuser
118 logic. A :c:func:`register_security()` function (in
119 ``security/security.c``) is provided to allow a security module to set
120 security_ops to refer to its own hook functions, and an
121 :c:func:`unregister_security()` function is provided to revert
122 security_ops to the dummy module hooks. This mechanism is used to set
123 the primary security module, which is responsible for making the final
124 decision for each hook.
126 LSM also provides a simple mechanism for stacking additional security
127 modules with the primary security module. It defines
128 :c:func:`register_security()` and
129 :c:func:`unregister_security()` hooks in the :c:type:`struct
130 security_operations <security_operations>` structure and
131 provides :c:func:`mod_reg_security()` and
132 :c:func:`mod_unreg_security()` functions that invoke these hooks
133 after performing some sanity checking. A security module can call these
134 functions in order to stack with other modules. However, the actual
135 details of how this stacking is handled are deferred to the module,
136 which can implement these hooks in any way it wishes (including always
137 returning an error if it does not wish to support stacking). In this
138 manner, LSM again defers the problem of composition to the module.
140 Although the LSM hooks are organized into substructures based on kernel
141 object, all of the hooks can be viewed as falling into two major
142 categories: hooks that are used to manage the security fields and hooks
143 that are used to perform access control. Examples of the first category
144 of hooks include the :c:func:`alloc_security()` and
145 :c:func:`free_security()` hooks defined for each kernel data
146 structure that has a security field. These hooks are used to allocate
147 and free security structures for kernel objects. The first category of
148 hooks also includes hooks that set information in the security field
149 after allocation, such as the :c:func:`post_lookup()` hook in
150 :c:type:`struct inode_security_ops <inode_security_ops>`.
151 This hook is used to set security information for inodes after
152 successful lookup operations. An example of the second category of hooks
153 is the :c:func:`permission()` hook in :c:type:`struct
154 inode_security_ops <inode_security_ops>`. This hook checks
155 permission when accessing an inode.
157 LSM Capabilities Module
158 =======================
160 The LSM kernel patch moves most of the existing POSIX.1e capabilities
161 logic into an optional security module stored in the file
162 ``security/capability.c``. This change allows users who do not want to
163 use capabilities to omit this code entirely from their kernel, instead
164 using the dummy module for traditional superuser logic or any other
165 module that they desire. This change also allows the developers of the
166 capabilities logic to maintain and enhance their code more freely,
167 without needing to integrate patches back into the base kernel.
169 In addition to moving the capabilities logic, the LSM kernel patch could
170 move the capability-related fields from the kernel data structures into
171 the new security fields managed by the security modules. However, at
172 present, the LSM kernel patch leaves the capability fields in the kernel
173 data structures. In his original remarks, Linus suggested that this
174 might be preferable so that other security modules can be easily stacked
175 with the capabilities module without needing to chain multiple security
176 structures on the security field. It also avoids imposing extra overhead
177 on the capabilities module to manage the security fields. However, the
178 LSM framework could certainly support such a move if it is determined to
179 be desirable, with only a few additional changes described below.
181 At present, the capabilities logic for computing process capabilities on
182 :c:func:`execve()` and :c:func:`set\*uid()`, checking
183 capabilities for a particular process, saving and checking capabilities
184 for netlink messages, and handling the :c:func:`capget()` and
185 :c:func:`capset()` system calls have been moved into the
186 capabilities module. There are still a few locations in the base kernel
187 where capability-related fields are directly examined or modified, but
188 the current version of the LSM patch does allow a security module to
189 completely replace the assignment and testing of capabilities. These few
190 locations would need to be changed if the capability-related fields were
191 moved into the security field. The following is a list of known
192 locations that still perform such direct examination or modification of
193 capability-related fields:
195 - ``fs/open.c``::c:func:`sys_access()`
197 - ``fs/lockd/host.c``::c:func:`nlm_bind_host()`
199 - ``fs/nfsd/auth.c``::c:func:`nfsd_setuser()`
201 - ``fs/proc/array.c``::c:func:`task_cap()`