]> git.ipfire.org Git - people/ms/linux.git/blob - security/Kconfig.hardening
Merge tag 'usb-6.0-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
[people/ms/linux.git] / security / Kconfig.hardening
1 # SPDX-License-Identifier: GPL-2.0-only
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
22 config CC_HAS_AUTO_VAR_INIT_PATTERN
23 def_bool $(cc-option,-ftrivial-auto-var-init=pattern)
24
25 config CC_HAS_AUTO_VAR_INIT_ZERO
26 # GCC ignores the -enable flag, so we can test for the feature with
27 # a single invocation using the flag, but drop it as appropriate in
28 # the Makefile, depending on the presence of Clang.
29 def_bool $(cc-option,-ftrivial-auto-var-init=zero -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang)
30
31 choice
32 prompt "Initialize kernel stack variables at function entry"
33 default GCC_PLUGIN_STRUCTLEAK_BYREF_ALL if COMPILE_TEST && GCC_PLUGINS
34 default INIT_STACK_ALL_PATTERN if COMPILE_TEST && CC_HAS_AUTO_VAR_INIT_PATTERN
35 default INIT_STACK_ALL_ZERO if CC_HAS_AUTO_VAR_INIT_ZERO
36 default INIT_STACK_NONE
37 help
38 This option enables initialization of stack variables at
39 function entry time. This has the possibility to have the
40 greatest coverage (since all functions can have their
41 variables initialized), but the performance impact depends
42 on the function calling complexity of a given workload's
43 syscalls.
44
45 This chooses the level of coverage over classes of potentially
46 uninitialized variables. The selected class of variable will be
47 initialized before use in a function.
48
49 config INIT_STACK_NONE
50 bool "no automatic stack variable initialization (weakest)"
51 help
52 Disable automatic stack variable initialization.
53 This leaves the kernel vulnerable to the standard
54 classes of uninitialized stack variable exploits
55 and information exposures.
56
57 config GCC_PLUGIN_STRUCTLEAK_USER
58 bool "zero-init structs marked for userspace (weak)"
59 # Plugin can be removed once the kernel only supports GCC 12+
60 depends on GCC_PLUGINS && !CC_HAS_AUTO_VAR_INIT_ZERO
61 select GCC_PLUGIN_STRUCTLEAK
62 help
63 Zero-initialize any structures on the stack containing
64 a __user attribute. This can prevent some classes of
65 uninitialized stack variable exploits and information
66 exposures, like CVE-2013-2141:
67 https://git.kernel.org/linus/b9e146d8eb3b9eca
68
69 config GCC_PLUGIN_STRUCTLEAK_BYREF
70 bool "zero-init structs passed by reference (strong)"
71 # Plugin can be removed once the kernel only supports GCC 12+
72 depends on GCC_PLUGINS && !CC_HAS_AUTO_VAR_INIT_ZERO
73 depends on !(KASAN && KASAN_STACK)
74 select GCC_PLUGIN_STRUCTLEAK
75 help
76 Zero-initialize any structures on the stack that may
77 be passed by reference and had not already been
78 explicitly initialized. This can prevent most classes
79 of uninitialized stack variable exploits and information
80 exposures, like CVE-2017-1000410:
81 https://git.kernel.org/linus/06e7e776ca4d3654
82
83 As a side-effect, this keeps a lot of variables on the
84 stack that can otherwise be optimized out, so combining
85 this with CONFIG_KASAN_STACK can lead to a stack overflow
86 and is disallowed.
87
88 config GCC_PLUGIN_STRUCTLEAK_BYREF_ALL
89 bool "zero-init everything passed by reference (very strong)"
90 # Plugin can be removed once the kernel only supports GCC 12+
91 depends on GCC_PLUGINS && !CC_HAS_AUTO_VAR_INIT_ZERO
92 depends on !(KASAN && KASAN_STACK)
93 select GCC_PLUGIN_STRUCTLEAK
94 help
95 Zero-initialize any stack variables that may be passed
96 by reference and had not already been explicitly
97 initialized. This is intended to eliminate all classes
98 of uninitialized stack variable exploits and information
99 exposures.
100
101 As a side-effect, this keeps a lot of variables on the
102 stack that can otherwise be optimized out, so combining
103 this with CONFIG_KASAN_STACK can lead to a stack overflow
104 and is disallowed.
105
106 config INIT_STACK_ALL_PATTERN
107 bool "pattern-init everything (strongest)"
108 depends on CC_HAS_AUTO_VAR_INIT_PATTERN
109 help
110 Initializes everything on the stack (including padding)
111 with a specific debug value. This is intended to eliminate
112 all classes of uninitialized stack variable exploits and
113 information exposures, even variables that were warned about
114 having been left uninitialized.
115
116 Pattern initialization is known to provoke many existing bugs
117 related to uninitialized locals, e.g. pointers receive
118 non-NULL values, buffer sizes and indices are very big. The
119 pattern is situation-specific; Clang on 64-bit uses 0xAA
120 repeating for all types and padding except float and double
121 which use 0xFF repeating (-NaN). Clang on 32-bit uses 0xFF
122 repeating for all types and padding.
123
124 config INIT_STACK_ALL_ZERO
125 bool "zero-init everything (strongest and safest)"
126 depends on CC_HAS_AUTO_VAR_INIT_ZERO
127 help
128 Initializes everything on the stack (including padding)
129 with a zero value. This is intended to eliminate all
130 classes of uninitialized stack variable exploits and
131 information exposures, even variables that were warned
132 about having been left uninitialized.
133
134 Zero initialization provides safe defaults for strings
135 (immediately NUL-terminated), pointers (NULL), indices
136 (index 0), and sizes (0 length), so it is therefore more
137 suitable as a production security mitigation than pattern
138 initialization.
139
140 endchoice
141
142 config GCC_PLUGIN_STRUCTLEAK_VERBOSE
143 bool "Report forcefully initialized variables"
144 depends on GCC_PLUGIN_STRUCTLEAK
145 depends on !COMPILE_TEST # too noisy
146 help
147 This option will cause a warning to be printed each time the
148 structleak plugin finds a variable it thinks needs to be
149 initialized. Since not all existing initializers are detected
150 by the plugin, this can produce false positive warnings.
151
152 config GCC_PLUGIN_STACKLEAK
153 bool "Poison kernel stack before returning from syscalls"
154 depends on GCC_PLUGINS
155 depends on HAVE_ARCH_STACKLEAK
156 help
157 This option makes the kernel erase the kernel stack before
158 returning from system calls. This has the effect of leaving
159 the stack initialized to the poison value, which both reduces
160 the lifetime of any sensitive stack contents and reduces
161 potential for uninitialized stack variable exploits or information
162 exposures (it does not cover functions reaching the same stack
163 depth as prior functions during the same syscall). This blocks
164 most uninitialized stack variable attacks, with the performance
165 impact being driven by the depth of the stack usage, rather than
166 the function calling complexity.
167
168 The performance impact on a single CPU system kernel compilation
169 sees a 1% slowdown, other systems and workloads may vary and you
170 are advised to test this feature on your expected workload before
171 deploying it.
172
173 This plugin was ported from grsecurity/PaX. More information at:
174 * https://grsecurity.net/
175 * https://pax.grsecurity.net/
176
177 config GCC_PLUGIN_STACKLEAK_VERBOSE
178 bool "Report stack depth analysis instrumentation" if EXPERT
179 depends on GCC_PLUGIN_STACKLEAK
180 depends on !COMPILE_TEST # too noisy
181 help
182 This option will cause a warning to be printed each time the
183 stackleak plugin finds a function it thinks needs to be
184 instrumented. This is useful for comparing coverage between
185 builds.
186
187 config STACKLEAK_TRACK_MIN_SIZE
188 int "Minimum stack frame size of functions tracked by STACKLEAK"
189 default 100
190 range 0 4096
191 depends on GCC_PLUGIN_STACKLEAK
192 help
193 The STACKLEAK gcc plugin instruments the kernel code for tracking
194 the lowest border of the kernel stack (and for some other purposes).
195 It inserts the stackleak_track_stack() call for the functions with
196 a stack frame size greater than or equal to this parameter.
197 If unsure, leave the default value 100.
198
199 config STACKLEAK_METRICS
200 bool "Show STACKLEAK metrics in the /proc file system"
201 depends on GCC_PLUGIN_STACKLEAK
202 depends on PROC_FS
203 help
204 If this is set, STACKLEAK metrics for every task are available in
205 the /proc file system. In particular, /proc/<pid>/stack_depth
206 shows the maximum kernel stack consumption for the current and
207 previous syscalls. Although this information is not precise, it
208 can be useful for estimating the STACKLEAK performance impact for
209 your workloads.
210
211 config STACKLEAK_RUNTIME_DISABLE
212 bool "Allow runtime disabling of kernel stack erasing"
213 depends on GCC_PLUGIN_STACKLEAK
214 help
215 This option provides 'stack_erasing' sysctl, which can be used in
216 runtime to control kernel stack erasing for kernels built with
217 CONFIG_GCC_PLUGIN_STACKLEAK.
218
219 config INIT_ON_ALLOC_DEFAULT_ON
220 bool "Enable heap memory zeroing on allocation by default"
221 help
222 This has the effect of setting "init_on_alloc=1" on the kernel
223 command line. This can be disabled with "init_on_alloc=0".
224 When "init_on_alloc" is enabled, all page allocator and slab
225 allocator memory will be zeroed when allocated, eliminating
226 many kinds of "uninitialized heap memory" flaws, especially
227 heap content exposures. The performance impact varies by
228 workload, but most cases see <1% impact. Some synthetic
229 workloads have measured as high as 7%.
230
231 config INIT_ON_FREE_DEFAULT_ON
232 bool "Enable heap memory zeroing on free by default"
233 help
234 This has the effect of setting "init_on_free=1" on the kernel
235 command line. This can be disabled with "init_on_free=0".
236 Similar to "init_on_alloc", when "init_on_free" is enabled,
237 all page allocator and slab allocator memory will be zeroed
238 when freed, eliminating many kinds of "uninitialized heap memory"
239 flaws, especially heap content exposures. The primary difference
240 with "init_on_free" is that data lifetime in memory is reduced,
241 as anything freed is wiped immediately, making live forensics or
242 cold boot memory attacks unable to recover freed memory contents.
243 The performance impact varies by workload, but is more expensive
244 than "init_on_alloc" due to the negative cache effects of
245 touching "cold" memory areas. Most cases see 3-5% impact. Some
246 synthetic workloads have measured as high as 8%.
247
248 config CC_HAS_ZERO_CALL_USED_REGS
249 def_bool $(cc-option,-fzero-call-used-regs=used-gpr)
250
251 config ZERO_CALL_USED_REGS
252 bool "Enable register zeroing on function exit"
253 depends on CC_HAS_ZERO_CALL_USED_REGS
254 help
255 At the end of functions, always zero any caller-used register
256 contents. This helps ensure that temporary values are not
257 leaked beyond the function boundary. This means that register
258 contents are less likely to be available for side channels
259 and information exposures. Additionally, this helps reduce the
260 number of useful ROP gadgets by about 20% (and removes compiler
261 generated "write-what-where" gadgets) in the resulting kernel
262 image. This has a less than 1% performance impact on most
263 workloads. Image size growth depends on architecture, and should
264 be evaluated for suitability. For example, x86_64 grows by less
265 than 1%, and arm64 grows by about 5%.
266
267 endmenu
268
269 config CC_HAS_RANDSTRUCT
270 def_bool $(cc-option,-frandomize-layout-seed-file=/dev/null)
271
272 choice
273 prompt "Randomize layout of sensitive kernel structures"
274 default RANDSTRUCT_FULL if COMPILE_TEST && (GCC_PLUGINS || CC_HAS_RANDSTRUCT)
275 default RANDSTRUCT_NONE
276 help
277 If you enable this, the layouts of structures that are entirely
278 function pointers (and have not been manually annotated with
279 __no_randomize_layout), or structures that have been explicitly
280 marked with __randomize_layout, will be randomized at compile-time.
281 This can introduce the requirement of an additional information
282 exposure vulnerability for exploits targeting these structure
283 types.
284
285 Enabling this feature will introduce some performance impact,
286 slightly increase memory usage, and prevent the use of forensic
287 tools like Volatility against the system (unless the kernel
288 source tree isn't cleaned after kernel installation).
289
290 The seed used for compilation is in scripts/basic/randomize.seed.
291 It remains after a "make clean" to allow for external modules to
292 be compiled with the existing seed and will be removed by a
293 "make mrproper" or "make distclean". This file should not be made
294 public, or the structure layout can be determined.
295
296 config RANDSTRUCT_NONE
297 bool "Disable structure layout randomization"
298 help
299 Build normally: no structure layout randomization.
300
301 config RANDSTRUCT_FULL
302 bool "Fully randomize structure layout"
303 depends on CC_HAS_RANDSTRUCT || GCC_PLUGINS
304 select MODVERSIONS if MODULES
305 help
306 Fully randomize the member layout of sensitive
307 structures as much as possible, which may have both a
308 memory size and performance impact.
309
310 One difference between the Clang and GCC plugin
311 implementations is the handling of bitfields. The GCC
312 plugin treats them as fully separate variables,
313 introducing sometimes significant padding. Clang tries
314 to keep adjacent bitfields together, but with their bit
315 ordering randomized.
316
317 config RANDSTRUCT_PERFORMANCE
318 bool "Limit randomization of structure layout to cache-lines"
319 depends on GCC_PLUGINS
320 select MODVERSIONS if MODULES
321 help
322 Randomization of sensitive kernel structures will make a
323 best effort at restricting randomization to cacheline-sized
324 groups of members. It will further not randomize bitfields
325 in structures. This reduces the performance hit of RANDSTRUCT
326 at the cost of weakened randomization.
327 endchoice
328
329 config RANDSTRUCT
330 def_bool !RANDSTRUCT_NONE
331
332 config GCC_PLUGIN_RANDSTRUCT
333 def_bool GCC_PLUGINS && RANDSTRUCT
334 help
335 Use GCC plugin to randomize structure layout.
336
337 This plugin was ported from grsecurity/PaX. More
338 information at:
339 * https://grsecurity.net/
340 * https://pax.grsecurity.net/
341
342 endmenu