| Message ID | 20210507025915.1464056-1-pcc@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2] kasan: fix unit tests with CONFIG_UBSAN_LOCAL_BOUNDS enabled | expand |
On Fri, May 7, 2021 at 4:59 AM Peter Collingbourne <pcc@google.com> wrote: > > These tests deliberately access these arrays out of bounds, > which will cause the dynamic local bounds checks inserted by > CONFIG_UBSAN_LOCAL_BOUNDS to fail and panic the kernel. To avoid this > problem, access the arrays via volatile pointers, which will prevent > the compiler from being able to determine the array bounds. Thanks for tracking this down! These crashes have been puzzling me for a while. > These accesses use volatile pointers to char (char *volatile) rather > than the more conventional pointers to volatile char (volatile char *) > because we want to prevent the compiler from making inferences about > the pointer itself (i.e. its array bounds), not the data that it > refers to. > > Signed-off-by: Peter Collingbourne <pcc@google.com> Tested-by: Alexander Potapenko <glider@google.com> (also note you are missing the Acked-by: here that Andrey gave)
On Fri, May 7, 2021 at 4:59 AM Peter Collingbourne <pcc@google.com> wrote: > > These tests deliberately access these arrays out of bounds, > which will cause the dynamic local bounds checks inserted by > CONFIG_UBSAN_LOCAL_BOUNDS to fail and panic the kernel. To avoid this > problem, access the arrays via volatile pointers, which will prevent > the compiler from being able to determine the array bounds. > > These accesses use volatile pointers to char (char *volatile) rather > than the more conventional pointers to volatile char (volatile char *) > because we want to prevent the compiler from making inferences about > the pointer itself (i.e. its array bounds), not the data that it > refers to. > > Signed-off-by: Peter Collingbourne <pcc@google.com> > Cc: stable@vger.kernel.org > Link: https://linux-review.googlesource.com/id/I90b1713fbfa1bf68ff895aef099ea77b98a7c3b9 > --- > lib/test_kasan.c | 29 +++++++++++++++++++++++------ > 1 file changed, 23 insertions(+), 6 deletions(-) > > diff --git a/lib/test_kasan.c b/lib/test_kasan.c > index dc05cfc2d12f..cacbbbdef768 100644 > --- a/lib/test_kasan.c > +++ b/lib/test_kasan.c > @@ -654,8 +654,20 @@ static char global_array[10]; > > static void kasan_global_oob(struct kunit *test) > { > - volatile int i = 3; > - char *p = &global_array[ARRAY_SIZE(global_array) + i]; > + /* > + * Deliberate out-of-bounds access. To prevent CONFIG_UBSAN_LOCAL_BOUNDS > + * from failing here and panicing the kernel, access the array via a > + * volatile pointer, which will prevent the compiler from being able to > + * determine the array bounds. > + * > + * This access uses a volatile pointer to char (char *volatile) rather > + * than the more conventional pointer to volatile char (volatile char *) > + * because we want to prevent the compiler from making inferences about > + * the pointer itself (i.e. its array bounds), not the data that it > + * refers to. > + */ > + char *volatile array = global_array; > + char *p = &array[ARRAY_SIZE(global_array) + 3]; > > /* Only generic mode instruments globals. */ > KASAN_TEST_NEEDS_CONFIG_ON(test, CONFIG_KASAN_GENERIC); > @@ -703,8 +715,9 @@ static void ksize_uaf(struct kunit *test) > static void kasan_stack_oob(struct kunit *test) > { > char stack_array[10]; > - volatile int i = OOB_TAG_OFF; > - char *p = &stack_array[ARRAY_SIZE(stack_array) + i]; > + /* See comment in kasan_global_oob. */ > + char *volatile array = stack_array; > + char *p = &array[ARRAY_SIZE(stack_array) + OOB_TAG_OFF]; > > KASAN_TEST_NEEDS_CONFIG_ON(test, CONFIG_KASAN_STACK); > > @@ -715,7 +728,9 @@ static void kasan_alloca_oob_left(struct kunit *test) > { > volatile int i = 10; > char alloca_array[i]; > - char *p = alloca_array - 1; > + /* See comment in kasan_global_oob. */ > + char *volatile array = alloca_array; > + char *p = array - 1; > > /* Only generic mode instruments dynamic allocas. */ > KASAN_TEST_NEEDS_CONFIG_ON(test, CONFIG_KASAN_GENERIC); > @@ -728,7 +743,9 @@ static void kasan_alloca_oob_right(struct kunit *test) > { > volatile int i = 10; > char alloca_array[i]; > - char *p = alloca_array + i; > + /* See comment in kasan_global_oob. */ > + char *volatile array = alloca_array; > + char *p = array + i; > > /* Only generic mode instruments dynamic allocas. */ > KASAN_TEST_NEEDS_CONFIG_ON(test, CONFIG_KASAN_GENERIC); > -- > 2.31.1.607.g51e8a6a459-goog > Reviewed-by: Andrey Konovalov <andreyknvl@gmail.com> Thanks!
diff --git a/lib/test_kasan.c b/lib/test_kasan.c index dc05cfc2d12f..cacbbbdef768 100644 --- a/lib/test_kasan.c +++ b/lib/test_kasan.c @@ -654,8 +654,20 @@ static char global_array[10]; static void kasan_global_oob(struct kunit *test) { - volatile int i = 3; - char *p = &global_array[ARRAY_SIZE(global_array) + i]; + /* + * Deliberate out-of-bounds access. To prevent CONFIG_UBSAN_LOCAL_BOUNDS + * from failing here and panicing the kernel, access the array via a + * volatile pointer, which will prevent the compiler from being able to + * determine the array bounds. + * + * This access uses a volatile pointer to char (char *volatile) rather + * than the more conventional pointer to volatile char (volatile char *) + * because we want to prevent the compiler from making inferences about + * the pointer itself (i.e. its array bounds), not the data that it + * refers to. + */ + char *volatile array = global_array; + char *p = &array[ARRAY_SIZE(global_array) + 3]; /* Only generic mode instruments globals. */ KASAN_TEST_NEEDS_CONFIG_ON(test, CONFIG_KASAN_GENERIC); @@ -703,8 +715,9 @@ static void ksize_uaf(struct kunit *test) static void kasan_stack_oob(struct kunit *test) { char stack_array[10]; - volatile int i = OOB_TAG_OFF; - char *p = &stack_array[ARRAY_SIZE(stack_array) + i]; + /* See comment in kasan_global_oob. */ + char *volatile array = stack_array; + char *p = &array[ARRAY_SIZE(stack_array) + OOB_TAG_OFF]; KASAN_TEST_NEEDS_CONFIG_ON(test, CONFIG_KASAN_STACK); @@ -715,7 +728,9 @@ static void kasan_alloca_oob_left(struct kunit *test) { volatile int i = 10; char alloca_array[i]; - char *p = alloca_array - 1; + /* See comment in kasan_global_oob. */ + char *volatile array = alloca_array; + char *p = array - 1; /* Only generic mode instruments dynamic allocas. */ KASAN_TEST_NEEDS_CONFIG_ON(test, CONFIG_KASAN_GENERIC); @@ -728,7 +743,9 @@ static void kasan_alloca_oob_right(struct kunit *test) { volatile int i = 10; char alloca_array[i]; - char *p = alloca_array + i; + /* See comment in kasan_global_oob. */ + char *volatile array = alloca_array; + char *p = array + i; /* Only generic mode instruments dynamic allocas. */ KASAN_TEST_NEEDS_CONFIG_ON(test, CONFIG_KASAN_GENERIC);
These tests deliberately access these arrays out of bounds, which will cause the dynamic local bounds checks inserted by CONFIG_UBSAN_LOCAL_BOUNDS to fail and panic the kernel. To avoid this problem, access the arrays via volatile pointers, which will prevent the compiler from being able to determine the array bounds. These accesses use volatile pointers to char (char *volatile) rather than the more conventional pointers to volatile char (volatile char *) because we want to prevent the compiler from making inferences about the pointer itself (i.e. its array bounds), not the data that it refers to. Signed-off-by: Peter Collingbourne <pcc@google.com> Cc: stable@vger.kernel.org Link: https://linux-review.googlesource.com/id/I90b1713fbfa1bf68ff895aef099ea77b98a7c3b9 --- lib/test_kasan.c | 29 +++++++++++++++++++++++------ 1 file changed, 23 insertions(+), 6 deletions(-)