Thomas Gleixner | ec8f24b | 2019-05-19 13:07:45 +0100 | [diff] [blame] | 1 | # SPDX-License-Identifier: GPL-2.0-only |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 2 | # This config refers to the generic KASAN mode. |
Andrey Ryabinin | 0b24bec | 2015-02-13 14:39:17 -0800 | [diff] [blame] | 3 | config HAVE_ARCH_KASAN |
| 4 | bool |
| 5 | |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 6 | config HAVE_ARCH_KASAN_SW_TAGS |
| 7 | bool |
| 8 | |
Andrey Konovalov | 6a63a63 | 2020-12-22 12:01:20 -0800 | [diff] [blame] | 9 | config HAVE_ARCH_KASAN_HW_TAGS |
| 10 | bool |
| 11 | |
| 12 | config HAVE_ARCH_KASAN_VMALLOC |
Daniel Axtens | 3c5c3cf | 2019-11-30 17:54:50 -0800 | [diff] [blame] | 13 | bool |
| 14 | |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 15 | config CC_HAS_KASAN_GENERIC |
| 16 | def_bool $(cc-option, -fsanitize=kernel-address) |
| 17 | |
| 18 | config CC_HAS_KASAN_SW_TAGS |
| 19 | def_bool $(cc-option, -fsanitize=kernel-hwaddress) |
Andrey Ryabinin | 0b24bec | 2015-02-13 14:39:17 -0800 | [diff] [blame] | 20 | |
Andrey Konovalov | 6a63a63 | 2020-12-22 12:01:20 -0800 | [diff] [blame] | 21 | # This option is only required for software KASAN modes. |
| 22 | # Old GCC versions don't have proper support for no_sanitize_address. |
| 23 | # See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89124 for details. |
Marco Elver | 7b861a5 | 2020-06-04 07:58:10 +0200 | [diff] [blame] | 24 | config CC_HAS_WORKING_NOSANITIZE_ADDRESS |
Marco Elver | acf7b0b | 2020-06-23 13:24:48 +0200 | [diff] [blame] | 25 | def_bool !CC_IS_GCC || GCC_VERSION >= 80300 |
Marco Elver | 7b861a5 | 2020-06-04 07:58:10 +0200 | [diff] [blame] | 26 | |
Marco Elver | 7a3767f | 2020-08-06 23:24:28 -0700 | [diff] [blame] | 27 | menuconfig KASAN |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 28 | bool "KASAN: runtime memory debugger" |
Andrey Konovalov | 6a63a63 | 2020-12-22 12:01:20 -0800 | [diff] [blame] | 29 | depends on (((HAVE_ARCH_KASAN && CC_HAS_KASAN_GENERIC) || \ |
| 30 | (HAVE_ARCH_KASAN_SW_TAGS && CC_HAS_KASAN_SW_TAGS)) && \ |
| 31 | CC_HAS_WORKING_NOSANITIZE_ADDRESS) || \ |
| 32 | HAVE_ARCH_KASAN_HW_TAGS |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 33 | depends on (SLUB && SYSFS) || (SLAB && !DEBUG_SLAB) |
Andrey Konovalov | ffcc5ce | 2020-12-22 12:00:42 -0800 | [diff] [blame] | 34 | select STACKDEPOT |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 35 | help |
| 36 | Enables KASAN (KernelAddressSANitizer) - runtime memory debugger, |
| 37 | designed to find out-of-bounds accesses and use-after-free bugs. |
| 38 | See Documentation/dev-tools/kasan.rst for details. |
| 39 | |
Marco Elver | 7a3767f | 2020-08-06 23:24:28 -0700 | [diff] [blame] | 40 | if KASAN |
| 41 | |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 42 | choice |
| 43 | prompt "KASAN mode" |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 44 | default KASAN_GENERIC |
| 45 | help |
Andrey Konovalov | 6a63a63 | 2020-12-22 12:01:20 -0800 | [diff] [blame] | 46 | KASAN has three modes: |
| 47 | 1. generic KASAN (similar to userspace ASan, |
| 48 | x86_64/arm64/xtensa, enabled with CONFIG_KASAN_GENERIC), |
| 49 | 2. software tag-based KASAN (arm64 only, based on software |
| 50 | memory tagging (similar to userspace HWASan), enabled with |
| 51 | CONFIG_KASAN_SW_TAGS), and |
| 52 | 3. hardware tag-based KASAN (arm64 only, based on hardware |
| 53 | memory tagging, enabled with CONFIG_KASAN_HW_TAGS). |
Marco Elver | ac4766b | 2020-08-06 23:24:31 -0700 | [diff] [blame] | 54 | |
Andrey Konovalov | 6a63a63 | 2020-12-22 12:01:20 -0800 | [diff] [blame] | 55 | All KASAN modes are strictly debugging features. |
| 56 | |
| 57 | For better error reports enable CONFIG_STACKTRACE. |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 58 | |
| 59 | config KASAN_GENERIC |
| 60 | bool "Generic mode" |
| 61 | depends on HAVE_ARCH_KASAN && CC_HAS_KASAN_GENERIC |
Jason A. Donenfeld | dd275ca | 2018-06-27 23:26:20 -0700 | [diff] [blame] | 62 | select SLUB_DEBUG if SLUB |
Andrey Konovalov | 6a63a63 | 2020-12-22 12:01:20 -0800 | [diff] [blame] | 63 | select CONSTRUCTORS |
Andrey Ryabinin | 0b24bec | 2015-02-13 14:39:17 -0800 | [diff] [blame] | 64 | help |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 65 | Enables generic KASAN mode. |
Marco Elver | ac4766b | 2020-08-06 23:24:31 -0700 | [diff] [blame] | 66 | |
| 67 | This mode is supported in both GCC and Clang. With GCC it requires |
Marco Elver | 527f675 | 2020-10-13 16:47:51 -0700 | [diff] [blame] | 68 | version 8.3.0 or later. Any supported Clang version is compatible, |
| 69 | but detection of out-of-bounds accesses for global variables is |
| 70 | supported only since Clang 11. |
Marco Elver | ac4766b | 2020-08-06 23:24:31 -0700 | [diff] [blame] | 71 | |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 72 | This mode consumes about 1/8th of available memory at kernel start |
| 73 | and introduces an overhead of ~x1.5 for the rest of the allocations. |
| 74 | The performance slowdown is ~x3. |
Marco Elver | ac4766b | 2020-08-06 23:24:31 -0700 | [diff] [blame] | 75 | |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 76 | Currently CONFIG_KASAN_GENERIC doesn't work with CONFIG_DEBUG_SLAB |
Alexander Potapenko | 7ed2f9e | 2016-03-25 14:21:59 -0700 | [diff] [blame] | 77 | (the resulting kernel does not boot). |
Andrey Ryabinin | 0b24bec | 2015-02-13 14:39:17 -0800 | [diff] [blame] | 78 | |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 79 | config KASAN_SW_TAGS |
| 80 | bool "Software tag-based mode" |
| 81 | depends on HAVE_ARCH_KASAN_SW_TAGS && CC_HAS_KASAN_SW_TAGS |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 82 | select SLUB_DEBUG if SLUB |
Andrey Konovalov | 6a63a63 | 2020-12-22 12:01:20 -0800 | [diff] [blame] | 83 | select CONSTRUCTORS |
Arnd Bergmann | e7c52b8 | 2018-02-06 15:41:41 -0800 | [diff] [blame] | 84 | help |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 85 | Enables software tag-based KASAN mode. |
Marco Elver | ac4766b | 2020-08-06 23:24:31 -0700 | [diff] [blame] | 86 | |
Andrey Konovalov | 6a63a63 | 2020-12-22 12:01:20 -0800 | [diff] [blame] | 87 | This mode require software memory tagging support in the form of |
| 88 | HWASan-like compiler instrumentation. |
| 89 | |
| 90 | Currently this mode is only implemented for arm64 CPUs and relies on |
| 91 | Top Byte Ignore. This mode requires Clang. |
Marco Elver | ac4766b | 2020-08-06 23:24:31 -0700 | [diff] [blame] | 92 | |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 93 | This mode consumes about 1/16th of available memory at kernel start |
| 94 | and introduces an overhead of ~20% for the rest of the allocations. |
| 95 | This mode may potentially introduce problems relating to pointer |
| 96 | casting and comparison, as it embeds tags into the top byte of each |
| 97 | pointer. |
Marco Elver | ac4766b | 2020-08-06 23:24:31 -0700 | [diff] [blame] | 98 | |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 99 | Currently CONFIG_KASAN_SW_TAGS doesn't work with CONFIG_DEBUG_SLAB |
| 100 | (the resulting kernel does not boot). |
Arnd Bergmann | e7c52b8 | 2018-02-06 15:41:41 -0800 | [diff] [blame] | 101 | |
Andrey Konovalov | 6a63a63 | 2020-12-22 12:01:20 -0800 | [diff] [blame] | 102 | config KASAN_HW_TAGS |
| 103 | bool "Hardware tag-based mode" |
| 104 | depends on HAVE_ARCH_KASAN_HW_TAGS |
| 105 | depends on SLUB |
| 106 | help |
| 107 | Enables hardware tag-based KASAN mode. |
| 108 | |
| 109 | This mode requires hardware memory tagging support, and can be used |
| 110 | by any architecture that provides it. |
| 111 | |
| 112 | Currently this mode is only implemented for arm64 CPUs starting from |
| 113 | ARMv8.5 and relies on Memory Tagging Extension and Top Byte Ignore. |
| 114 | |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 115 | endchoice |
| 116 | |
Andrey Ryabinin | 0b24bec | 2015-02-13 14:39:17 -0800 | [diff] [blame] | 117 | choice |
| 118 | prompt "Instrumentation type" |
Andrey Konovalov | 6a63a63 | 2020-12-22 12:01:20 -0800 | [diff] [blame] | 119 | depends on KASAN_GENERIC || KASAN_SW_TAGS |
Andrey Ryabinin | 0b24bec | 2015-02-13 14:39:17 -0800 | [diff] [blame] | 120 | default KASAN_OUTLINE |
| 121 | |
| 122 | config KASAN_OUTLINE |
| 123 | bool "Outline instrumentation" |
| 124 | help |
| 125 | Before every memory access compiler insert function call |
| 126 | __asan_load*/__asan_store*. These functions performs check |
| 127 | of shadow memory. This is slower than inline instrumentation, |
| 128 | however it doesn't bloat size of kernel's .text section so |
| 129 | much as inline does. |
| 130 | |
| 131 | config KASAN_INLINE |
| 132 | bool "Inline instrumentation" |
| 133 | help |
| 134 | Compiler directly inserts code checking shadow memory before |
| 135 | memory accesses. This is faster than outline (in some workloads |
| 136 | it gives about x2 boost over outline instrumentation), but |
| 137 | make kernel's .text size much bigger. |
| 138 | |
| 139 | endchoice |
| 140 | |
Arnd Bergmann | 6baec88 | 2019-02-28 16:21:58 -0800 | [diff] [blame] | 141 | config KASAN_STACK_ENABLE |
| 142 | bool "Enable stack instrumentation (unsafe)" if CC_IS_CLANG && !COMPILE_TEST |
Andrey Konovalov | 6a63a63 | 2020-12-22 12:01:20 -0800 | [diff] [blame] | 143 | depends on KASAN_GENERIC || KASAN_SW_TAGS |
Arnd Bergmann | 6baec88 | 2019-02-28 16:21:58 -0800 | [diff] [blame] | 144 | help |
| 145 | The LLVM stack address sanitizer has a know problem that |
| 146 | causes excessive stack usage in a lot of functions, see |
| 147 | https://bugs.llvm.org/show_bug.cgi?id=38809 |
| 148 | Disabling asan-stack makes it safe to run kernels build |
| 149 | with clang-8 with KASAN enabled, though it loses some of |
| 150 | the functionality. |
Arnd Bergmann | ebb6d35 | 2019-08-02 21:48:54 -0700 | [diff] [blame] | 151 | This feature is always disabled when compile-testing with clang |
| 152 | to avoid cluttering the output in stack overflow warnings, |
| 153 | but clang users can still enable it for builds without |
| 154 | CONFIG_COMPILE_TEST. On gcc it is assumed to always be safe |
| 155 | to use and enabled by default. |
Arnd Bergmann | 6baec88 | 2019-02-28 16:21:58 -0800 | [diff] [blame] | 156 | |
| 157 | config KASAN_STACK |
| 158 | int |
| 159 | default 1 if KASAN_STACK_ENABLE || CC_IS_GCC |
| 160 | default 0 |
| 161 | |
Walter Wu | ae8f06b | 2019-09-23 15:34:13 -0700 | [diff] [blame] | 162 | config KASAN_SW_TAGS_IDENTIFY |
| 163 | bool "Enable memory corruption identification" |
| 164 | depends on KASAN_SW_TAGS |
| 165 | help |
| 166 | This option enables best-effort identification of bug type |
| 167 | (use-after-free or out-of-bounds) at the cost of increased |
| 168 | memory consumption. |
| 169 | |
Daniel Axtens | 3c5c3cf | 2019-11-30 17:54:50 -0800 | [diff] [blame] | 170 | config KASAN_VMALLOC |
| 171 | bool "Back mappings in vmalloc space with real shadow memory" |
Andrey Konovalov | 71f6af6 | 2020-12-22 12:00:10 -0800 | [diff] [blame] | 172 | depends on KASAN_GENERIC && HAVE_ARCH_KASAN_VMALLOC |
Daniel Axtens | 3c5c3cf | 2019-11-30 17:54:50 -0800 | [diff] [blame] | 173 | help |
| 174 | By default, the shadow region for vmalloc space is the read-only |
| 175 | zero page. This means that KASAN cannot detect errors involving |
| 176 | vmalloc space. |
| 177 | |
| 178 | Enabling this option will hook in to vmap/vmalloc and back those |
| 179 | mappings with real shadow memory allocated on demand. This allows |
| 180 | for KASAN to detect more sorts of errors (and to support vmapped |
| 181 | stacks), but at the cost of higher memory usage. |
| 182 | |
Patricia Alfonso | 73228c7 | 2020-10-13 16:55:06 -0700 | [diff] [blame] | 183 | config KASAN_KUNIT_TEST |
| 184 | tristate "KUnit-compatible tests of KASAN bug detection capabilities" if !KUNIT_ALL_TESTS |
| 185 | depends on KASAN && KUNIT |
| 186 | default KUNIT_ALL_TESTS |
Andrey Ryabinin | 3f15801 | 2015-02-13 14:39:53 -0800 | [diff] [blame] | 187 | help |
Patricia Alfonso | 73228c7 | 2020-10-13 16:55:06 -0700 | [diff] [blame] | 188 | This is a KUnit test suite doing various nasty things like |
| 189 | out of bounds and use after free accesses. It is useful for testing |
Andrey Konovalov | 2bd926b | 2018-12-28 00:29:53 -0800 | [diff] [blame] | 190 | kernel debugging features like KASAN. |
Marco Elver | 7a3767f | 2020-08-06 23:24:28 -0700 | [diff] [blame] | 191 | |
Patricia Alfonso | 73228c7 | 2020-10-13 16:55:06 -0700 | [diff] [blame] | 192 | For more information on KUnit and unit tests in general, please refer |
Andrey Konovalov | f05842c | 2021-02-24 12:05:26 -0800 | [diff] [blame] | 193 | to the KUnit documentation in Documentation/dev-tools/kunit. |
Patricia Alfonso | 73228c7 | 2020-10-13 16:55:06 -0700 | [diff] [blame] | 194 | |
Andrey Konovalov | 5d92bdf | 2021-02-24 12:05:29 -0800 | [diff] [blame] | 195 | config KASAN_MODULE_TEST |
Patricia Alfonso | 73228c7 | 2020-10-13 16:55:06 -0700 | [diff] [blame] | 196 | tristate "KUnit-incompatible tests of KASAN bug detection capabilities" |
Andrey Konovalov | f05842c | 2021-02-24 12:05:26 -0800 | [diff] [blame] | 197 | depends on m && KASAN && !KASAN_HW_TAGS |
Patricia Alfonso | 73228c7 | 2020-10-13 16:55:06 -0700 | [diff] [blame] | 198 | help |
| 199 | This is a part of the KASAN test suite that is incompatible with |
| 200 | KUnit. Currently includes tests that do bad copy_from/to_user |
| 201 | accesses. |
| 202 | |
Marco Elver | 7a3767f | 2020-08-06 23:24:28 -0700 | [diff] [blame] | 203 | endif # KASAN |