| KASAN(9) | MidnightBSD Kernel Developer's Manual | KASAN(9) |
kasan — kernel
address sanitizer
To compile KASAN into the kernel, place the following line in your kernel configuration file:
options KASAN
void
kasan_mark(const
void *addr, size_t
size, size_t
redzsize, uint8_t
code);
kasan is a subsystem which leverages
compiler instrumentation to detect invalid memory accesses in the kernel.
Currently it is implemented only on the amd64 platform.
When kasan is compiled
into the kernel, the compiler is configured to emit function calls upon
every memory access. The functions are implemented by
kasan and permit run-time detection of several types
of bugs including use-after-frees, double frees and frees of invalid
pointers, and out-of-bounds accesses. These protections apply to memory
allocated by uma(9),
malloc(9) and related
functions, and
kmem_malloc()
and related functions, as well as global variables and kernel stacks.
kasan is conservative and will not detect all
instances of these types of bugs. Memory accesses through the kernel map are
sanitized, but accesses via the direct map are not. When
kasan is configured, the kernel aims to minimize its
use of the direct map.
kasan is implemented using compiler
instrumentation and a kernel runtime. When a kernel is built with the KASAN
option enabled, the compiler inserts function calls before most memory
accesses in the generated code. The runtime implements the corresponding
functions, which decide whether a given access is valid. If not, the runtime
prints a warning or panics the kernel, depending on the value of the
debug.kasan.panic_on_violation
sysctl/tunable.
The kasan
runtime in a KASAN-configured kernel can be disabled by setting the loader
tunable
debug.kasan.disable=1.
The kasan runtime works by maintaining a
shadow map for the kernel map. There exists a linear mapping between
addresses in the kernel map and addresses in the shadow map. The shadow map
is used to store information about the current state of allocations from the
kernel map. For example, when a buffer is returned by
malloc(9), the
corresponding region of the shadow map is marked to indicate that the buffer
is valid. When it is freed, the shadow map is updated to mark the buffer as
invalid. Accesses to the buffer are intercepted by the
kasan runtime and validated using the contents of
the shadow map.
Upon booting, all kernel memory is marked as valid. Kernel
allocators must mark cached but free buffers as invalid, and must mark them
valid before freeing the kernel virtual address range. This slightly reduces
the effectiveness of kasan but simplifies its
maintenance and integration into the kernel.
Updates to the shadow map are performed by calling
kasan_mark(). Parameter addr
is the address of the buffer whose shadow is to be updated,
size is the usable size of the buffer, and
redzsize is the full size of the buffer allocated from
lower layers of the system. redzsize must be greater
than or equal to size. In some cases kernel allocators
will return a buffer larger than that requested by the consumer; the unused
space at the end is referred to as a red zone and is always marked as
invalid. code allows the caller to specify an
identifier used when marking a buffer as invalid. The identifier is included
in any reports generated by kasan and helps identify
the source of the invalid access. For instance, when an item is freed to a
uma(9) zone, the item is
marked with KASAN_UMA_FREED. See
<sys/asan.h> for the
available identifiers. If the entire buffer is to be marked valid, i.e.,
size and redzsize are equal,
code should be 0.
kasan first appeared in
FreeBSD 14.0.
Accesses to kernel memory outside of the kernel map are ignored by
the kasan runtime. When
kasan is configured, the kernel memory allocators
are configured to use the kernel map, but some uses of the direct map
remain. For example, on amd64, accesses to page table pages are not
tracked.
Some kernel memory allocators explicitly permit accesses after an
object has been freed. These cannot be sanitized by
kasan. For example, memory from all
uma(9) zones initialized
with the UMA_ZONE_NOFREE flag are not sanitized.
| October 13, 2023 | midnightbsd-4.0 |