Commit | Line | Data |
---|---|---|
ec8f24b7 | 1 | # SPDX-License-Identifier: GPL-2.0-only |
9f671e58 KC |
2 | menu "Kernel hardening options" |
3 | ||
4 | config 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 | ||
20 | menu "Memory initialization" | |
21 | ||
f0fe00d4 | 22 | config CC_HAS_AUTO_VAR_INIT_PATTERN |
709a972e KC |
23 | def_bool $(cc-option,-ftrivial-auto-var-init=pattern) |
24 | ||
f0fe00d4 | 25 | config 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 | ||
9f671e58 KC |
28 | choice |
29 | prompt "Initialize kernel stack variables at function entry" | |
30 | default GCC_PLUGIN_STRUCTLEAK_BYREF_ALL if COMPILE_TEST && GCC_PLUGINS | |
f0fe00d4 | 31 | default INIT_STACK_ALL_PATTERN if COMPILE_TEST && CC_HAS_AUTO_VAR_INIT_PATTERN |
dcb7c0b9 | 32 | default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_PATTERN |
9f671e58 KC |
33 | default INIT_STACK_NONE |
34 | help | |
35 | This option enables initialization of stack variables at | |
36 | function entry time. This has the possibility to have the | |
37 | greatest coverage (since all functions can have their | |
38 | variables initialized), but the performance impact depends | |
39 | on the function calling complexity of a given workload's | |
40 | syscalls. | |
41 | ||
42 | This chooses the level of coverage over classes of potentially | |
dcb7c0b9 | 43 | uninitialized variables. The selected class of variable will be |
9f671e58 KC |
44 | initialized before use in a function. |
45 | ||
46 | config INIT_STACK_NONE | |
dcb7c0b9 | 47 | bool "no automatic stack variable initialization (weakest)" |
9f671e58 KC |
48 | help |
49 | Disable automatic stack variable initialization. | |
50 | This leaves the kernel vulnerable to the standard | |
51 | classes of uninitialized stack variable exploits | |
52 | and information exposures. | |
53 | ||
54 | config GCC_PLUGIN_STRUCTLEAK_USER | |
55 | bool "zero-init structs marked for userspace (weak)" | |
56 | depends on GCC_PLUGINS | |
57 | select GCC_PLUGIN_STRUCTLEAK | |
58 | help | |
59 | Zero-initialize any structures on the stack containing | |
60 | a __user attribute. This can prevent some classes of | |
61 | uninitialized stack variable exploits and information | |
62 | exposures, like CVE-2013-2141: | |
63 | https://git.kernel.org/linus/b9e146d8eb3b9eca | |
64 | ||
65 | config GCC_PLUGIN_STRUCTLEAK_BYREF | |
66 | bool "zero-init structs passed by reference (strong)" | |
67 | depends on GCC_PLUGINS | |
02c58773 | 68 | depends on !(KASAN && KASAN_STACK) |
9f671e58 KC |
69 | select GCC_PLUGIN_STRUCTLEAK |
70 | help | |
71 | Zero-initialize any structures on the stack that may | |
72 | be passed by reference and had not already been | |
73 | explicitly initialized. This can prevent most classes | |
74 | of uninitialized stack variable exploits and information | |
75 | exposures, like CVE-2017-1000410: | |
76 | https://git.kernel.org/linus/06e7e776ca4d3654 | |
77 | ||
173e6ee2 AB |
78 | As a side-effect, this keeps a lot of variables on the |
79 | stack that can otherwise be optimized out, so combining | |
80 | this with CONFIG_KASAN_STACK can lead to a stack overflow | |
81 | and is disallowed. | |
82 | ||
9f671e58 | 83 | config GCC_PLUGIN_STRUCTLEAK_BYREF_ALL |
dcb7c0b9 | 84 | bool "zero-init everything passed by reference (very strong)" |
9f671e58 | 85 | depends on GCC_PLUGINS |
02c58773 | 86 | depends on !(KASAN && KASAN_STACK) |
9f671e58 KC |
87 | select GCC_PLUGIN_STRUCTLEAK |
88 | help | |
89 | Zero-initialize any stack variables that may be passed | |
90 | by reference and had not already been explicitly | |
91 | initialized. This is intended to eliminate all classes | |
92 | of uninitialized stack variable exploits and information | |
93 | exposures. | |
94 | ||
dcb7c0b9 KC |
95 | As a side-effect, this keeps a lot of variables on the |
96 | stack that can otherwise be optimized out, so combining | |
97 | this with CONFIG_KASAN_STACK can lead to a stack overflow | |
98 | and is disallowed. | |
99 | ||
f0fe00d4 | 100 | config INIT_STACK_ALL_PATTERN |
dcb7c0b9 | 101 | bool "pattern-init everything (strongest)" |
f0fe00d4 | 102 | depends on CC_HAS_AUTO_VAR_INIT_PATTERN |
709a972e | 103 | help |
dcb7c0b9 KC |
104 | Initializes everything on the stack (including padding) |
105 | with a specific debug value. This is intended to eliminate | |
106 | all classes of uninitialized stack variable exploits and | |
107 | information exposures, even variables that were warned about | |
108 | having been left uninitialized. | |
709a972e | 109 | |
f0fe00d4 | 110 | Pattern initialization is known to provoke many existing bugs |
111 | related to uninitialized locals, e.g. pointers receive | |
dcb7c0b9 KC |
112 | non-NULL values, buffer sizes and indices are very big. The |
113 | pattern is situation-specific; Clang on 64-bit uses 0xAA | |
114 | repeating for all types and padding except float and double | |
115 | which use 0xFF repeating (-NaN). Clang on 32-bit uses 0xFF | |
116 | repeating for all types and padding. | |
f0fe00d4 | 117 | |
118 | config INIT_STACK_ALL_ZERO | |
dcb7c0b9 | 119 | bool "zero-init everything (strongest and safest)" |
f0fe00d4 | 120 | depends on CC_HAS_AUTO_VAR_INIT_ZERO |
121 | help | |
dcb7c0b9 KC |
122 | Initializes everything on the stack (including padding) |
123 | with a zero value. This is intended to eliminate all | |
124 | classes of uninitialized stack variable exploits and | |
125 | information exposures, even variables that were warned | |
126 | about having been left uninitialized. | |
127 | ||
128 | Zero initialization provides safe defaults for strings | |
129 | (immediately NUL-terminated), pointers (NULL), indices | |
130 | (index 0), and sizes (0 length), so it is therefore more | |
131 | suitable as a production security mitigation than pattern | |
132 | initialization. | |
f0fe00d4 | 133 | |
9f671e58 KC |
134 | endchoice |
135 | ||
136 | config GCC_PLUGIN_STRUCTLEAK_VERBOSE | |
137 | bool "Report forcefully initialized variables" | |
138 | depends on GCC_PLUGIN_STRUCTLEAK | |
139 | depends on !COMPILE_TEST # too noisy | |
140 | help | |
141 | This option will cause a warning to be printed each time the | |
142 | structleak plugin finds a variable it thinks needs to be | |
143 | initialized. Since not all existing initializers are detected | |
144 | by the plugin, this can produce false positive warnings. | |
145 | ||
b6a6a377 KC |
146 | config GCC_PLUGIN_STACKLEAK |
147 | bool "Poison kernel stack before returning from syscalls" | |
148 | depends on GCC_PLUGINS | |
149 | depends on HAVE_ARCH_STACKLEAK | |
150 | help | |
151 | This option makes the kernel erase the kernel stack before | |
152 | returning from system calls. This has the effect of leaving | |
153 | the stack initialized to the poison value, which both reduces | |
154 | the lifetime of any sensitive stack contents and reduces | |
155 | potential for uninitialized stack variable exploits or information | |
156 | exposures (it does not cover functions reaching the same stack | |
157 | depth as prior functions during the same syscall). This blocks | |
158 | most uninitialized stack variable attacks, with the performance | |
159 | impact being driven by the depth of the stack usage, rather than | |
160 | the function calling complexity. | |
161 | ||
162 | The performance impact on a single CPU system kernel compilation | |
163 | sees a 1% slowdown, other systems and workloads may vary and you | |
164 | are advised to test this feature on your expected workload before | |
165 | deploying it. | |
166 | ||
167 | This plugin was ported from grsecurity/PaX. More information at: | |
168 | * https://grsecurity.net/ | |
169 | * https://pax.grsecurity.net/ | |
170 | ||
171 | config STACKLEAK_TRACK_MIN_SIZE | |
172 | int "Minimum stack frame size of functions tracked by STACKLEAK" | |
173 | default 100 | |
174 | range 0 4096 | |
175 | depends on GCC_PLUGIN_STACKLEAK | |
176 | help | |
177 | The STACKLEAK gcc plugin instruments the kernel code for tracking | |
178 | the lowest border of the kernel stack (and for some other purposes). | |
179 | It inserts the stackleak_track_stack() call for the functions with | |
180 | a stack frame size greater than or equal to this parameter. | |
181 | If unsure, leave the default value 100. | |
182 | ||
183 | config STACKLEAK_METRICS | |
184 | bool "Show STACKLEAK metrics in the /proc file system" | |
185 | depends on GCC_PLUGIN_STACKLEAK | |
186 | depends on PROC_FS | |
187 | help | |
188 | If this is set, STACKLEAK metrics for every task are available in | |
189 | the /proc file system. In particular, /proc/<pid>/stack_depth | |
190 | shows the maximum kernel stack consumption for the current and | |
191 | previous syscalls. Although this information is not precise, it | |
192 | can be useful for estimating the STACKLEAK performance impact for | |
193 | your workloads. | |
194 | ||
195 | config STACKLEAK_RUNTIME_DISABLE | |
196 | bool "Allow runtime disabling of kernel stack erasing" | |
197 | depends on GCC_PLUGIN_STACKLEAK | |
198 | help | |
199 | This option provides 'stack_erasing' sysctl, which can be used in | |
200 | runtime to control kernel stack erasing for kernels built with | |
201 | CONFIG_GCC_PLUGIN_STACKLEAK. | |
202 | ||
6471384a AP |
203 | config INIT_ON_ALLOC_DEFAULT_ON |
204 | bool "Enable heap memory zeroing on allocation by default" | |
205 | help | |
206 | This has the effect of setting "init_on_alloc=1" on the kernel | |
207 | command line. This can be disabled with "init_on_alloc=0". | |
208 | When "init_on_alloc" is enabled, all page allocator and slab | |
209 | allocator memory will be zeroed when allocated, eliminating | |
210 | many kinds of "uninitialized heap memory" flaws, especially | |
211 | heap content exposures. The performance impact varies by | |
212 | workload, but most cases see <1% impact. Some synthetic | |
213 | workloads have measured as high as 7%. | |
214 | ||
215 | config INIT_ON_FREE_DEFAULT_ON | |
216 | bool "Enable heap memory zeroing on free by default" | |
217 | help | |
218 | This has the effect of setting "init_on_free=1" on the kernel | |
219 | command line. This can be disabled with "init_on_free=0". | |
220 | Similar to "init_on_alloc", when "init_on_free" is enabled, | |
221 | all page allocator and slab allocator memory will be zeroed | |
222 | when freed, eliminating many kinds of "uninitialized heap memory" | |
223 | flaws, especially heap content exposures. The primary difference | |
224 | with "init_on_free" is that data lifetime in memory is reduced, | |
225 | as anything freed is wiped immediately, making live forensics or | |
226 | cold boot memory attacks unable to recover freed memory contents. | |
227 | The performance impact varies by workload, but is more expensive | |
228 | than "init_on_alloc" due to the negative cache effects of | |
229 | touching "cold" memory areas. Most cases see 3-5% impact. Some | |
230 | synthetic workloads have measured as high as 8%. | |
231 | ||
a82adfd5 KC |
232 | config CC_HAS_ZERO_CALL_USED_REGS |
233 | def_bool $(cc-option,-fzero-call-used-regs=used-gpr) | |
234 | ||
235 | config ZERO_CALL_USED_REGS | |
236 | bool "Enable register zeroing on function exit" | |
237 | depends on CC_HAS_ZERO_CALL_USED_REGS | |
238 | help | |
239 | At the end of functions, always zero any caller-used register | |
240 | contents. This helps ensure that temporary values are not | |
241 | leaked beyond the function boundary. This means that register | |
242 | contents are less likely to be available for side channels | |
243 | and information exposures. Additionally, this helps reduce the | |
244 | number of useful ROP gadgets by about 20% (and removes compiler | |
245 | generated "write-what-where" gadgets) in the resulting kernel | |
246 | image. This has a less than 1% performance impact on most | |
247 | workloads. Image size growth depends on architecture, and should | |
248 | be evaluated for suitability. For example, x86_64 grows by less | |
249 | than 1%, and arm64 grows by about 5%. | |
250 | ||
9f671e58 KC |
251 | endmenu |
252 | ||
253 | endmenu |