blob: 269967c4fc1b67e81b403e9925497620974c68f8 [file] [log] [blame]
Thomas Gleixnerec8f24b2019-05-19 13:07:45 +01001# SPDX-License-Identifier: GPL-2.0-only
Kees Cook9f671e52019-04-10 08:23:44 -07002menu "Kernel hardening options"
3
4config GCC_PLUGIN_STRUCTLEAK
5 bool
6 help
7 While the kernel is built with warnings enabled for any missed
8 stack variable initializations, this warning is silenced for
9 anything passed by reference to another function, under the
10 occasionally misguided assumption that the function will do
11 the initialization. As this regularly leads to exploitable
12 flaws, this plugin is available to identify and zero-initialize
13 such variables, depending on the chosen level of coverage.
14
15 This plugin was originally ported from grsecurity/PaX. More
16 information at:
17 * https://grsecurity.net/
18 * https://pax.grsecurity.net/
19
20menu "Memory initialization"
21
glider@google.comf0fe00d2020-06-16 10:34:35 +020022config CC_HAS_AUTO_VAR_INIT_PATTERN
Kees Cook709a9722019-04-10 08:48:31 -070023 def_bool $(cc-option,-ftrivial-auto-var-init=pattern)
24
glider@google.comf0fe00d2020-06-16 10:34:35 +020025config CC_HAS_AUTO_VAR_INIT_ZERO
26 def_bool $(cc-option,-ftrivial-auto-var-init=zero -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang)
27
Kees Cook9f671e52019-04-10 08:23:44 -070028choice
29 prompt "Initialize kernel stack variables at function entry"
30 default GCC_PLUGIN_STRUCTLEAK_BYREF_ALL if COMPILE_TEST && GCC_PLUGINS
glider@google.comf0fe00d2020-06-16 10:34:35 +020031 default INIT_STACK_ALL_PATTERN if COMPILE_TEST && CC_HAS_AUTO_VAR_INIT_PATTERN
Kees Cook9f671e52019-04-10 08:23:44 -070032 default INIT_STACK_NONE
33 help
34 This option enables initialization of stack variables at
35 function entry time. This has the possibility to have the
36 greatest coverage (since all functions can have their
37 variables initialized), but the performance impact depends
38 on the function calling complexity of a given workload's
39 syscalls.
40
41 This chooses the level of coverage over classes of potentially
42 uninitialized variables. The selected class will be
43 initialized before use in a function.
44
45 config INIT_STACK_NONE
46 bool "no automatic initialization (weakest)"
47 help
48 Disable automatic stack variable initialization.
49 This leaves the kernel vulnerable to the standard
50 classes of uninitialized stack variable exploits
51 and information exposures.
52
53 config GCC_PLUGIN_STRUCTLEAK_USER
54 bool "zero-init structs marked for userspace (weak)"
55 depends on GCC_PLUGINS
56 select GCC_PLUGIN_STRUCTLEAK
57 help
58 Zero-initialize any structures on the stack containing
59 a __user attribute. This can prevent some classes of
60 uninitialized stack variable exploits and information
61 exposures, like CVE-2013-2141:
62 https://git.kernel.org/linus/b9e146d8eb3b9eca
63
64 config GCC_PLUGIN_STRUCTLEAK_BYREF
65 bool "zero-init structs passed by reference (strong)"
66 depends on GCC_PLUGINS
Arnd Bergmann173e6ee2019-07-22 13:41:20 +020067 depends on !(KASAN && KASAN_STACK=1)
Kees Cook9f671e52019-04-10 08:23:44 -070068 select GCC_PLUGIN_STRUCTLEAK
69 help
70 Zero-initialize any structures on the stack that may
71 be passed by reference and had not already been
72 explicitly initialized. This can prevent most classes
73 of uninitialized stack variable exploits and information
74 exposures, like CVE-2017-1000410:
75 https://git.kernel.org/linus/06e7e776ca4d3654
76
Arnd Bergmann173e6ee2019-07-22 13:41:20 +020077 As a side-effect, this keeps a lot of variables on the
78 stack that can otherwise be optimized out, so combining
79 this with CONFIG_KASAN_STACK can lead to a stack overflow
80 and is disallowed.
81
Kees Cook9f671e52019-04-10 08:23:44 -070082 config GCC_PLUGIN_STRUCTLEAK_BYREF_ALL
83 bool "zero-init anything passed by reference (very strong)"
84 depends on GCC_PLUGINS
Arnd Bergmann173e6ee2019-07-22 13:41:20 +020085 depends on !(KASAN && KASAN_STACK=1)
Kees Cook9f671e52019-04-10 08:23:44 -070086 select GCC_PLUGIN_STRUCTLEAK
87 help
88 Zero-initialize any stack variables that may be passed
89 by reference and had not already been explicitly
90 initialized. This is intended to eliminate all classes
91 of uninitialized stack variable exploits and information
92 exposures.
93
glider@google.comf0fe00d2020-06-16 10:34:35 +020094 config INIT_STACK_ALL_PATTERN
Kees Cook709a9722019-04-10 08:48:31 -070095 bool "0xAA-init everything on the stack (strongest)"
glider@google.comf0fe00d2020-06-16 10:34:35 +020096 depends on CC_HAS_AUTO_VAR_INIT_PATTERN
Kees Cook709a9722019-04-10 08:48:31 -070097 help
98 Initializes everything on the stack with a 0xAA
99 pattern. This is intended to eliminate all classes
100 of uninitialized stack variable exploits and information
101 exposures, even variables that were warned to have been
102 left uninitialized.
103
glider@google.comf0fe00d2020-06-16 10:34:35 +0200104 Pattern initialization is known to provoke many existing bugs
105 related to uninitialized locals, e.g. pointers receive
106 non-NULL values, buffer sizes and indices are very big.
107
108 config INIT_STACK_ALL_ZERO
109 bool "zero-init everything on the stack (strongest and safest)"
110 depends on CC_HAS_AUTO_VAR_INIT_ZERO
111 help
112 Initializes everything on the stack with a zero
113 value. This is intended to eliminate all classes
114 of uninitialized stack variable exploits and information
115 exposures, even variables that were warned to have been
116 left uninitialized.
117
118 Zero initialization provides safe defaults for strings,
119 pointers, indices and sizes, and is therefore
120 more suitable as a security mitigation measure.
121
Kees Cook9f671e52019-04-10 08:23:44 -0700122endchoice
123
124config GCC_PLUGIN_STRUCTLEAK_VERBOSE
125 bool "Report forcefully initialized variables"
126 depends on GCC_PLUGIN_STRUCTLEAK
127 depends on !COMPILE_TEST # too noisy
128 help
129 This option will cause a warning to be printed each time the
130 structleak plugin finds a variable it thinks needs to be
131 initialized. Since not all existing initializers are detected
132 by the plugin, this can produce false positive warnings.
133
Kees Cookb6a6a372019-04-10 09:04:40 -0700134config GCC_PLUGIN_STACKLEAK
135 bool "Poison kernel stack before returning from syscalls"
136 depends on GCC_PLUGINS
137 depends on HAVE_ARCH_STACKLEAK
138 help
139 This option makes the kernel erase the kernel stack before
140 returning from system calls. This has the effect of leaving
141 the stack initialized to the poison value, which both reduces
142 the lifetime of any sensitive stack contents and reduces
143 potential for uninitialized stack variable exploits or information
144 exposures (it does not cover functions reaching the same stack
145 depth as prior functions during the same syscall). This blocks
146 most uninitialized stack variable attacks, with the performance
147 impact being driven by the depth of the stack usage, rather than
148 the function calling complexity.
149
150 The performance impact on a single CPU system kernel compilation
151 sees a 1% slowdown, other systems and workloads may vary and you
152 are advised to test this feature on your expected workload before
153 deploying it.
154
155 This plugin was ported from grsecurity/PaX. More information at:
156 * https://grsecurity.net/
157 * https://pax.grsecurity.net/
158
159config STACKLEAK_TRACK_MIN_SIZE
160 int "Minimum stack frame size of functions tracked by STACKLEAK"
161 default 100
162 range 0 4096
163 depends on GCC_PLUGIN_STACKLEAK
164 help
165 The STACKLEAK gcc plugin instruments the kernel code for tracking
166 the lowest border of the kernel stack (and for some other purposes).
167 It inserts the stackleak_track_stack() call for the functions with
168 a stack frame size greater than or equal to this parameter.
169 If unsure, leave the default value 100.
170
171config STACKLEAK_METRICS
172 bool "Show STACKLEAK metrics in the /proc file system"
173 depends on GCC_PLUGIN_STACKLEAK
174 depends on PROC_FS
175 help
176 If this is set, STACKLEAK metrics for every task are available in
177 the /proc file system. In particular, /proc/<pid>/stack_depth
178 shows the maximum kernel stack consumption for the current and
179 previous syscalls. Although this information is not precise, it
180 can be useful for estimating the STACKLEAK performance impact for
181 your workloads.
182
183config STACKLEAK_RUNTIME_DISABLE
184 bool "Allow runtime disabling of kernel stack erasing"
185 depends on GCC_PLUGIN_STACKLEAK
186 help
187 This option provides 'stack_erasing' sysctl, which can be used in
188 runtime to control kernel stack erasing for kernels built with
189 CONFIG_GCC_PLUGIN_STACKLEAK.
190
Alexander Potapenko64713842019-07-11 20:59:19 -0700191config INIT_ON_ALLOC_DEFAULT_ON
192 bool "Enable heap memory zeroing on allocation by default"
193 help
194 This has the effect of setting "init_on_alloc=1" on the kernel
195 command line. This can be disabled with "init_on_alloc=0".
196 When "init_on_alloc" is enabled, all page allocator and slab
197 allocator memory will be zeroed when allocated, eliminating
198 many kinds of "uninitialized heap memory" flaws, especially
199 heap content exposures. The performance impact varies by
200 workload, but most cases see <1% impact. Some synthetic
201 workloads have measured as high as 7%.
202
203config INIT_ON_FREE_DEFAULT_ON
204 bool "Enable heap memory zeroing on free by default"
205 help
206 This has the effect of setting "init_on_free=1" on the kernel
207 command line. This can be disabled with "init_on_free=0".
208 Similar to "init_on_alloc", when "init_on_free" is enabled,
209 all page allocator and slab allocator memory will be zeroed
210 when freed, eliminating many kinds of "uninitialized heap memory"
211 flaws, especially heap content exposures. The primary difference
212 with "init_on_free" is that data lifetime in memory is reduced,
213 as anything freed is wiped immediately, making live forensics or
214 cold boot memory attacks unable to recover freed memory contents.
215 The performance impact varies by workload, but is more expensive
216 than "init_on_alloc" due to the negative cache effects of
217 touching "cold" memory areas. Most cases see 3-5% impact. Some
218 synthetic workloads have measured as high as 8%.
219
Kees Cook9f671e52019-04-10 08:23:44 -0700220endmenu
221
222endmenu