blob: f63822195871bed47e6bd7c29076587cec361510 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001<?xml version="1.0" encoding="UTF-8"?>
2<!DOCTYPE article PUBLIC "-//OASIS//DTD DocBook XML V4.1.2//EN"
3 "http://www.oasis-open.org/docbook/xml/4.1.2/docbookx.dtd" []>
4
5<article class="whitepaper" id="LinuxSecurityModule" lang="en">
6 <articleinfo>
7 <title>Linux Security Modules: General Security Hooks for Linux</title>
8 <authorgroup>
9 <author>
10 <firstname>Stephen</firstname>
11 <surname>Smalley</surname>
12 <affiliation>
13 <orgname>NAI Labs</orgname>
14 <address><email>ssmalley@nai.com</email></address>
15 </affiliation>
16 </author>
17 <author>
18 <firstname>Timothy</firstname>
19 <surname>Fraser</surname>
20 <affiliation>
21 <orgname>NAI Labs</orgname>
22 <address><email>tfraser@nai.com</email></address>
23 </affiliation>
24 </author>
25 <author>
26 <firstname>Chris</firstname>
27 <surname>Vance</surname>
28 <affiliation>
29 <orgname>NAI Labs</orgname>
30 <address><email>cvance@nai.com</email></address>
31 </affiliation>
32 </author>
33 </authorgroup>
34 </articleinfo>
35
36<sect1><title>Introduction</title>
37
38<para>
39In March 2001, the National Security Agency (NSA) gave a presentation
40about Security-Enhanced Linux (SELinux) at the 2.5 Linux Kernel
41Summit. SELinux is an implementation of flexible and fine-grained
42nondiscretionary access controls in the Linux kernel, originally
43implemented as its own particular kernel patch. Several other
44security projects (e.g. RSBAC, Medusa) have also developed flexible
45access control architectures for the Linux kernel, and various
46projects have developed particular access control models for Linux
47(e.g. LIDS, DTE, SubDomain). Each project has developed and
48maintained its own kernel patch to support its security needs.
49</para>
50
51<para>
52In response to the NSA presentation, Linus Torvalds made a set of
53remarks that described a security framework he would be willing to
54consider for inclusion in the mainstream Linux kernel. He described a
55general framework that would provide a set of security hooks to
56control operations on kernel objects and a set of opaque security
57fields in kernel data structures for maintaining security attributes.
58This framework could then be used by loadable kernel modules to
59implement any desired model of security. Linus also suggested the
60possibility of migrating the Linux capabilities code into such a
61module.
62</para>
63
64<para>
65The Linux Security Modules (LSM) project was started by WireX to
66develop such a framework. LSM is a joint development effort by
67several security projects, including Immunix, SELinux, SGI and Janus,
68and several individuals, including Greg Kroah-Hartman and James
69Morris, to develop a Linux kernel patch that implements this
70framework. The patch is currently tracking the 2.4 series and is
71targeted for integration into the 2.5 development series. This
72technical report provides an overview of the framework and the example
73capabilities security module provided by the LSM kernel patch.
74</para>
75
76</sect1>
77
78<sect1 id="framework"><title>LSM Framework</title>
79
80<para>
81The LSM kernel patch provides a general kernel framework to support
82security modules. In particular, the LSM framework is primarily
83focused on supporting access control modules, although future
84development is likely to address other security needs such as
85auditing. By itself, the framework does not provide any additional
86security; it merely provides the infrastructure to support security
87modules. The LSM kernel patch also moves most of the capabilities
88logic into an optional security module, with the system defaulting
89to the traditional superuser logic. This capabilities module
90is discussed further in <xref linkend="cap"/>.
91</para>
92
93<para>
94The LSM kernel patch adds security fields to kernel data structures
95and inserts calls to hook functions at critical points in the kernel
96code to manage the security fields and to perform access control. It
97also adds functions for registering and unregistering security
98modules, and adds a general <function>security</function> system call
99to support new system calls for security-aware applications.
100</para>
101
102<para>
103The LSM security fields are simply <type>void*</type> pointers. For
104process and program execution security information, security fields
105were added to <structname>struct task_struct</structname> and
106<structname>struct linux_binprm</structname>. For filesystem security
107information, a security field was added to
108<structname>struct super_block</structname>. For pipe, file, and socket
109security information, security fields were added to
110<structname>struct inode</structname> and
111<structname>struct file</structname>. For packet and network device security
112information, security fields were added to
113<structname>struct sk_buff</structname> and
114<structname>struct net_device</structname>. For System V IPC security
115information, security fields were added to
116<structname>struct kern_ipc_perm</structname> and
117<structname>struct msg_msg</structname>; additionally, the definitions
118for <structname>struct msg_msg</structname>, <structname>struct
119msg_queue</structname>, and <structname>struct
120shmid_kernel</structname> were moved to header files
121(<filename>include/linux/msg.h</filename> and
122<filename>include/linux/shm.h</filename> as appropriate) to allow
123the security modules to use these definitions.
124</para>
125
126<para>
127Each LSM hook is a function pointer in a global table,
128security_ops. This table is a
129<structname>security_operations</structname> structure as defined by
130<filename>include/linux/security.h</filename>. Detailed documentation
131for each hook is included in this header file. At present, this
132structure consists of a collection of substructures that group related
133hooks based on the kernel object (e.g. task, inode, file, sk_buff,
134etc) as well as some top-level hook function pointers for system
135operations. This structure is likely to be flattened in the future
136for performance. The placement of the hook calls in the kernel code
137is described by the "called:" lines in the per-hook documentation in
138the header file. The hook calls can also be easily found in the
139kernel code by looking for the string "security_ops->".
140
141</para>
142
143<para>
144Linus mentioned per-process security hooks in his original remarks as a
145possible alternative to global security hooks. However, if LSM were
146to start from the perspective of per-process hooks, then the base
147framework would have to deal with how to handle operations that
148involve multiple processes (e.g. kill), since each process might have
149its own hook for controlling the operation. This would require a
150general mechanism for composing hooks in the base framework.
151Additionally, LSM would still need global hooks for operations that
152have no process context (e.g. network input operations).
153Consequently, LSM provides global security hooks, but a security
154module is free to implement per-process hooks (where that makes sense)
155by storing a security_ops table in each process' security field and
156then invoking these per-process hooks from the global hooks.
157The problem of composition is thus deferred to the module.
158</para>
159
160<para>
161The global security_ops table is initialized to a set of hook
162functions provided by a dummy security module that provides
163traditional superuser logic. A <function>register_security</function>
164function (in <filename>security/security.c</filename>) is provided to
165allow a security module to set security_ops to refer to its own hook
166functions, and an <function>unregister_security</function> function is
167provided to revert security_ops to the dummy module hooks. This
168mechanism is used to set the primary security module, which is
169responsible for making the final decision for each hook.
170</para>
171
172<para>
173LSM also provides a simple mechanism for stacking additional security
174modules with the primary security module. It defines
175<function>register_security</function> and
176<function>unregister_security</function> hooks in the
177<structname>security_operations</structname> structure and provides
178<function>mod_reg_security</function> and
179<function>mod_unreg_security</function> functions that invoke these
180hooks after performing some sanity checking. A security module can
181call these functions in order to stack with other modules. However,
182the actual details of how this stacking is handled are deferred to the
183module, which can implement these hooks in any way it wishes
184(including always returning an error if it does not wish to support
185stacking). In this manner, LSM again defers the problem of
186composition to the module.
187</para>
188
189<para>
190Although the LSM hooks are organized into substructures based on
191kernel object, all of the hooks can be viewed as falling into two
192major categories: hooks that are used to manage the security fields
193and hooks that are used to perform access control. Examples of the
194first category of hooks include the
195<function>alloc_security</function> and
196<function>free_security</function> hooks defined for each kernel data
197structure that has a security field. These hooks are used to allocate
198and free security structures for kernel objects. The first category
199of hooks also includes hooks that set information in the security
200field after allocation, such as the <function>post_lookup</function>
201hook in <structname>struct inode_security_ops</structname>. This hook
202is used to set security information for inodes after successful lookup
203operations. An example of the second category of hooks is the
204<function>permission</function> hook in
205<structname>struct inode_security_ops</structname>. This hook checks
206permission when accessing an inode.
207</para>
208
209</sect1>
210
211<sect1 id="cap"><title>LSM Capabilities Module</title>
212
213<para>
214The LSM kernel patch moves most of the existing POSIX.1e capabilities
215logic into an optional security module stored in the file
216<filename>security/capability.c</filename>. This change allows
217users who do not want to use capabilities to omit this code entirely
218from their kernel, instead using the dummy module for traditional
219superuser logic or any other module that they desire. This change
220also allows the developers of the capabilities logic to maintain and
221enhance their code more freely, without needing to integrate patches
222back into the base kernel.
223</para>
224
225<para>
226In addition to moving the capabilities logic, the LSM kernel patch
227could move the capability-related fields from the kernel data
228structures into the new security fields managed by the security
229modules. However, at present, the LSM kernel patch leaves the
230capability fields in the kernel data structures. In his original
231remarks, Linus suggested that this might be preferable so that other
232security modules can be easily stacked with the capabilities module
233without needing to chain multiple security structures on the security field.
234It also avoids imposing extra overhead on the capabilities module
235to manage the security fields. However, the LSM framework could
236certainly support such a move if it is determined to be desirable,
237with only a few additional changes described below.
238</para>
239
240<para>
241At present, the capabilities logic for computing process capabilities
242on <function>execve</function> and <function>set*uid</function>,
243checking capabilities for a particular process, saving and checking
244capabilities for netlink messages, and handling the
245<function>capget</function> and <function>capset</function> system
246calls have been moved into the capabilities module. There are still a
247few locations in the base kernel where capability-related fields are
248directly examined or modified, but the current version of the LSM
249patch does allow a security module to completely replace the
250assignment and testing of capabilities. These few locations would
251need to be changed if the capability-related fields were moved into
252the security field. The following is a list of known locations that
253still perform such direct examination or modification of
254capability-related fields:
255<itemizedlist>
256<listitem><para><filename>fs/open.c</filename>:<function>sys_access</function></para></listitem>
257<listitem><para><filename>fs/lockd/host.c</filename>:<function>nlm_bind_host</function></para></listitem>
258<listitem><para><filename>fs/nfsd/auth.c</filename>:<function>nfsd_setuser</function></para></listitem>
259<listitem><para><filename>fs/proc/array.c</filename>:<function>task_cap</function></para></listitem>
260</itemizedlist>
261</para>
262
263</sect1>
264
265</article>