[PATCH v11 00/19] Kernel address sanitizer - runtime memory debugger.

From: Andrey Ryabinin
Date: Tue Feb 03 2015 - 12:43:32 EST


KASan is a runtime memory debugger designed to find use-after-free
and out-of-bounds bugs.

Currently KASAN supported only for x86_64 architecture and requires kernel
to be build with SLUB allocator.
KASAN uses compile-time instrumentation for checking every memory access, therefore you
will need a fresh GCC >= v4.9.2

Patches also available in git:

git://github.com/aryabinin/linux --branch=kasan/kasan_v11

Changes since v10:
- Address comments from Andrew.
Note: I didn't fix log level inconsistency between pr_err()/dump_stack()
yet. This doesn't seems like super important right now, I don't want to bloat
this patchset even more. I think it would be better to do this in separate series,
since if we wan't to fix this, we will need to fix slub code too (object_err()
which is used by KASan).


Historical background of address sanitizer from Dmitry Vyukov <dvyukov@xxxxxxxxxx>:
"We've developed the set of tools, AddressSanitizer (Asan),
ThreadSanitizer and MemorySanitizer, for user space. We actively use
them for testing inside of Google (continuous testing, fuzzing,
running prod services). To date the tools have found more than 10'000
scary bugs in Chromium, Google internal codebase and various
open-source projects (Firefox, OpenSSL, gcc, clang, ffmpeg, MySQL and
lots of others):
https://code.google.com/p/address-sanitizer/wiki/FoundBugs
https://code.google.com/p/thread-sanitizer/wiki/FoundBugs
https://code.google.com/p/memory-sanitizer/wiki/FoundBugs
The tools are part of both gcc and clang compilers.

We have not yet done massive testing under the Kernel AddressSanitizer
(it's kind of chicken and egg problem, you need it to be upstream to
start applying it extensively). To date it has found about 50 bugs.
Bugs that we've found in upstream kernel are listed here:
https://code.google.com/p/address-sanitizer/wiki/AddressSanitizerForKernel#Trophies
We've also found ~20 bugs in out internal version of the kernel. Also
people from Samsung and Oracle have found some. It's somewhat expected
that when we boot the kernel and run a trivial workload, we do not
find hundreds of bugs -- most of the harmful bugs in kernel codebase
were already fixed the hard way (the kernel is quite stable, right).
Based on our experience with user-space version of the tool, most of
the bugs will be discovered by continuously testing new code (new bugs
discovered the easy way), running fuzzers (that can discover existing
bugs that are not hit frequently enough) and running end-to-end tests
of production systems.

As others noted, the main feature of AddressSanitizer is its
performance due to inline compiler instrumentation and simple linear
shadow memory. User-space Asan has ~2x slowdown on computational
programs and ~2x memory consumption increase. Taking into account that
kernel usually consumes only small fraction of CPU and memory when
running real user-space programs, I would expect that kernel Asan will
have ~10-30% slowdown and similar memory consumption increase (when we
finish all tuning).

I agree that Asan can well replace kmemcheck. We have plans to start
working on Kernel MemorySanitizer that finds uses of uninitialized
memory. Asan+Msan will provide feature-parity with kmemcheck. As
others noted, Asan will unlikely replace debug slab and pagealloc that
can be enabled at runtime. Asan uses compiler instrumentation, so even
if it is disabled, it still incurs visible overheads.

Asan technology is easily portable to other architectures. Compiler
instrumentation is fully portable. Runtime has some arch-dependent
parts like shadow mapping and atomic operation interception. They are
relatively easy to port.

Thanks"


Comparison with other debugging features:
=======================================

KMEMCHECK:
- KASan can do almost everything that kmemcheck can. KASan uses compile-time
instrumentation, which makes it significantly faster than kmemcheck.
The only advantage of kmemcheck over KASan is detection of uninitialized
memory reads.

Some brief performance testing showed that kasan could be x500-x600 times
faster than kmemcheck:

$ netperf -l 30
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to localhost (127.0.0.1) port 0 AF_INET
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec

no debug: 87380 16384 16384 30.00 41624.72

kasan inline: 87380 16384 16384 30.00 12870.54

kasan outline: 87380 16384 16384 30.00 10586.39

kmemcheck: 87380 16384 16384 30.03 20.23

- Also kmemcheck couldn't work on several CPUs. It always sets number of CPUs to 1.
KASan doesn't have such limitation.

DEBUG_PAGEALLOC:
- KASan is slower than DEBUG_PAGEALLOC, but KASan works on sub-page
granularity level, so it able to find more bugs.

SLUB_DEBUG (poisoning, redzones):
- SLUB_DEBUG has lower overhead than KASan.

- SLUB_DEBUG in most cases are not able to detect bad reads,
KASan able to detect both reads and writes.

- In some cases (e.g. redzone overwritten) SLUB_DEBUG detect
bugs only on allocation/freeing of object. KASan catch
bugs right before it will happen, so we always know exact
place of first bad read/write.

Basic idea:
===========

The main idea of KASAN is to use shadow memory to record whether each byte of memory
is safe to access or not, and use compiler's instrumentation to check the shadow memory
on each memory access.

Address sanitizer uses 1/8 of the memory addressable in kernel for shadow memory
(on x86_64 16TB of virtual address space reserved for shadow to cover all 128TB)
and uses direct mapping with a scale and offset to translate a memory
address to its corresponding shadow address.

Here is function to translate address to corresponding shadow address:

unsigned long kasan_mem_to_shadow(unsigned long addr)
{
return (addr >> KASAN_SHADOW_SCALE_SHIFT) + KASAN_SHADOW_OFFSET;
}
where KASAN_SHADOW_SCALE_SHIFT = 3.

So for every 8 bytes there is one corresponding byte of shadow memory.
The following encoding used for each shadow byte: 0 means that all 8 bytes of the
corresponding memory region are valid for access; k (1 <= k <= 7) means that
the first k bytes are valid for access, and other (8 - k) bytes are not;
Any negative value indicates that the entire 8-bytes are inaccessible.
Different negative values used to distinguish between different kinds of
inaccessible memory (redzones, freed memory) (see mm/kasan/kasan.h).

To be able to detect accesses to bad memory we need a special compiler.
Such compiler inserts a specific function calls (__asan_load*(addr), __asan_store*(addr))
before each memory access of size 1, 2, 4, 8 or 16.

These functions check whether memory region is valid to access or not by checking
corresponding shadow memory. If access is not valid an error printed.


Changelog for previous versions:
===============================

Changes since v9:
- Makefile changes per discussion with Michal Marek
- Fixed false positive reports that could happen on module_freeing.

Changes since v8:
- Fixed unpoisoned redzones for not-allocated-yet object
in newly allocated slab page. (from Dmitry C.)

- Some minor non-function cleanups in kasan internals.

- Added ack from Catalin

- Added stack instrumentation. With this we could detect
out of bounds accesses in stack variables. (patch 12)

- Added globals instrumentation - catching out of bounds in
global varibles. (patches 13-17)

- Shadow moved out from vmalloc into hole between vmemmap
and %esp fixup stacks. For globals instrumentation
we will need shadow backing modules addresses.
So we need some sort of a shadow memory allocator
(something like vmmemap_populate() function, except
that it should be available after boot).

__vmalloc_node_range() suits that purpose, except that
it can't be used for allocating for shadow in vmalloc
area because shadow in vmalloc is already 'allocated'
to protect us from other vmalloc users. So we need
16TB of unused addresses. And we have big enough hole
between vmemmap and %esp fixup stacks. So I moved shadow
there.


Changes since v7:
- Fix build with CONFIG_KASAN_INLINE=y from Sasha.

- Don't poison redzone on freeing, since it is poisend already from Dmitry Chernenkov.

- Fix altinstruction_entry for memcpy.

- Move kasan_slab_free() call after debug_obj_free to prevent some false-positives
with CONFIG_DEBUG_OBJECTS=y

- Drop -pg flag for kasan internals to avoid recursion with function tracer
enabled.

- Added ack from Christoph.


Changes since v6:
- New patch 'x86_64: kasan: add interceptors for memset/memmove/memcpy functions'
Recently instrumentation of builtin functions calls (memset/memmove/memcpy)
was removed in GCC 5.0. So to check the memory accessed by such functions,
we now need interceptors for them.

- Added kasan's die notifier which prints a hint message before General protection fault,
explaining that GPF could be caused by NULL-ptr dereference or user memory access.

- Minor refactoring in 3/n patch. Rename kasan_map_shadow() to kasan_init() and call it
from setup_arch() instead of zone_sizes_init().

- Slightly tweak kasan's report layout.

- Update changelog for 1/n patch.

Changes since v5:
- Added __printf(3, 4) to slab_err to catch format mismatches (Joe Perches)

- Changed in Documentation/kasan.txt per Jonathan.

- Patch for inline instrumentation support merged to the first patch.
GCC 5.0 finally has support for this.
- Patch 'kasan: Add support for upcoming GCC 5.0 asan ABI changes' also merged into the first.
Those GCC ABI changes are in GCC's master branch now.

- Added information about instrumentation types to documentation.

- Added -fno-conserve-stack to CFLAGS for mm/kasan/kasan.c file, because -fconserve-stack is bogus
and it causing unecessary split in __asan_load1/__asan_store1. Because of this split
kasan_report() is actually not inlined (even though it __always_inline) and _RET_IP_ gives
unexpected value. GCC bugzilla entry: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63533

Changes since v4:
- rebased on top of mmotm-2014-10-23-16-26

- merge patch 'efi: libstub: disable KASAN for efistub in' into the first patch.
No reason to keep it separate.

- Added support for upcoming asan ABI changes in GCC 5.0 (second patch).
GCC patch has not been published/upstreamed yet, but to will be soon. I'm adding this in advance
in order to avoid breaking kasan with future GCC update.
Details about gcc ABI changes in this thread: https://gcc.gnu.org/ml/gcc-patches/2014-10/msg02510.html

- Updated GCC verison requirements in doc (GCC kasan patches were backported into 4.9 branch)

- Dropped last patch with inline instrumentation support. At first let's wait for merging GCC patches.

Changes since v3:

- rebased on last mm
- Added comment about rcu slabs.
- Removed useless kasan_free_slab_pages().
- Removed __asan_init_v*() stub. GCC doesn't generate this call anymore:
https://gcc.gnu.org/ml/gcc-patches/2014-10/msg00269.html
- Replaced CALL_KASAN_REPORT define with inline function

Changes since v2:

- Shadow moved to vmalloc area.
- Added posion page. This page mapped to shadow correspondig to
shadow region itself:
[kasan_mem_to_shadow(KASAN_SHADOW_START) - kasan_mem_to_shadow(KASAN_SHADOW_END)]
It used to catch memory access to shadow outside mm/kasan/.

- Fixed boot with CONFIG_DEBUG_VIRTUAL=y
- Fixed boot with KASan and stack protector enabled
(patch "x86_64: load_percpu_segment: read irq_stack_union.gs_base before load_segment")

- Fixed build with CONFIG_EFI_STUB=y
- Some slub specific stuf moved from mm/slab.h to include/linux/slub_def.h
- Fixed Kconfig dependency. CONFIG_KASAN depends on CONFIG_SLUB_DEBUG.
- Optimizations of __asan_load/__asan_store.
- Spelling fixes from Randy.
- Misc minor cleanups in different places.


- Added inline instrumentation in last patch. This will require two not
yet-in-trunk-patches for GCC:
https://gcc.gnu.org/ml/gcc-patches/2014-09/msg00452.html
https://gcc.gnu.org/ml/gcc-patches/2014-09/msg00605.html

Changes since v1:

- The main change is in shadow memory laoyut.
Now for shadow memory we reserve 1/8 of all virtual addresses available for kernel.
16TB on x86_64 to cover all 128TB of kernel's address space.
At early stage we map whole shadow region with zero page.
Latter, after physical memory mapped to direct mapping address range
we unmap zero pages from corresponding shadow and allocate and map a real
memory.

- Since per-arch work is much bigger now, support for arm/x86_32 platforms was dropped.

- CFLAGS was change from -fsanitize=address with different --params to -fsanitize=kernel-address

- If compiler doesn't support -fsanitize=kernel-address warning printed and build continues without -fsanitize

- Removed kasan_memset/kasan_memcpy/kasan_memmove hooks. It turned out that this hooks are not needed. Compiler
already instrument memset/memcpy/memmove (inserts __asan_load/__asan_store call before mem*() calls).

- branch profiling disabled for mm/kasan/kasan.c to avoid recursion (__asan_load -> ftrace_likely_update -> __asan_load -> ...)

- kasan hooks for buddy allocator moved to right places

Andrey Ryabinin (19):
compiler: introduce __alias(symbol) shortcut
Add kernel address sanitizer infrastructure.
kasan: disable memory hotplug
x86_64: add KASan support
mm: page_alloc: add kasan hooks on alloc and free paths
mm: slub: introduce virt_to_obj function.
mm: slub: share object_err function
mm: slub: introduce metadata_access_enable()/metadata_access_disable()
mm: slub: add kernel address sanitizer support for slub allocator
fs: dcache: manually unpoison dname after allocation to shut up
kasan's reports
kmemleak: disable kasan instrumentation for kmemleak
lib: add kasan test module
x86_64: kasan: add interceptors for memset/memmove/memcpy functions
kasan: enable stack instrumentation
mm: vmalloc: add flag preventing guard hole allocation
mm: vmalloc: pass additional vm_flags to __vmalloc_node_range()
kernel: add support for .init_array.* constructors
module: fix types of device tables aliases
kasan: enable instrumentation of global variables

Documentation/kasan.txt | 170 +++++++++++
Documentation/x86/x86_64/mm.txt | 2 +
Makefile | 3 +-
arch/arm/kernel/module.c | 2 +-
arch/arm64/kernel/module.c | 4 +-
arch/mips/kernel/module.c | 2 +-
arch/parisc/kernel/module.c | 2 +-
arch/s390/kernel/module.c | 2 +-
arch/sparc/kernel/module.c | 2 +-
arch/unicore32/kernel/module.c | 2 +-
arch/x86/Kconfig | 1 +
arch/x86/boot/Makefile | 2 +
arch/x86/boot/compressed/Makefile | 2 +
arch/x86/boot/compressed/eboot.c | 3 +-
arch/x86/boot/compressed/misc.h | 1 +
arch/x86/include/asm/kasan.h | 31 ++
arch/x86/include/asm/page_64_types.h | 12 +-
arch/x86/include/asm/string_64.h | 18 +-
arch/x86/kernel/Makefile | 4 +
arch/x86/kernel/dumpstack.c | 5 +-
arch/x86/kernel/head64.c | 9 +-
arch/x86/kernel/head_64.S | 30 ++
arch/x86/kernel/module.c | 14 +-
arch/x86/kernel/setup.c | 3 +
arch/x86/kernel/x8664_ksyms_64.c | 10 +-
arch/x86/lib/memcpy_64.S | 6 +-
arch/x86/lib/memmove_64.S | 4 +
arch/x86/lib/memset_64.S | 10 +-
arch/x86/mm/Makefile | 3 +
arch/x86/mm/kasan_init_64.c | 206 +++++++++++++
arch/x86/realmode/Makefile | 2 +-
arch/x86/realmode/rm/Makefile | 1 +
arch/x86/vdso/Makefile | 1 +
drivers/firmware/efi/libstub/Makefile | 1 +
drivers/firmware/efi/libstub/efistub.h | 4 +
fs/dcache.c | 5 +
include/asm-generic/vmlinux.lds.h | 1 +
include/linux/compiler-gcc.h | 1 +
include/linux/compiler-gcc4.h | 4 +
include/linux/compiler-gcc5.h | 2 +
include/linux/init_task.h | 8 +
include/linux/kasan.h | 89 ++++++
include/linux/module.h | 2 +-
include/linux/sched.h | 3 +
include/linux/slab.h | 11 +-
include/linux/slub_def.h | 19 ++
include/linux/vmalloc.h | 13 +-
kernel/module.c | 2 +
lib/Kconfig.debug | 2 +
lib/Kconfig.kasan | 54 ++++
lib/Makefile | 1 +
lib/test_kasan.c | 277 ++++++++++++++++++
mm/Makefile | 4 +
mm/compaction.c | 2 +
mm/kasan/Makefile | 8 +
mm/kasan/kasan.c | 516 +++++++++++++++++++++++++++++++++
mm/kasan/kasan.h | 75 +++++
mm/kasan/report.c | 269 +++++++++++++++++
mm/kmemleak.c | 6 +
mm/page_alloc.c | 3 +
mm/slab_common.c | 5 +-
mm/slub.c | 58 +++-
mm/vmalloc.c | 16 +-
scripts/Makefile.kasan | 26 ++
scripts/Makefile.lib | 10 +
scripts/module-common.lds | 3 +
66 files changed, 2022 insertions(+), 47 deletions(-)
create mode 100644 Documentation/kasan.txt
create mode 100644 arch/x86/include/asm/kasan.h
create mode 100644 arch/x86/mm/kasan_init_64.c
create mode 100644 include/linux/kasan.h
create mode 100644 lib/Kconfig.kasan
create mode 100644 lib/test_kasan.c
create mode 100644 mm/kasan/Makefile
create mode 100644 mm/kasan/kasan.c
create mode 100644 mm/kasan/kasan.h
create mode 100644 mm/kasan/report.c
create mode 100644 scripts/Makefile.kasan

--
--
Cc: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
Cc: Konstantin Serebryany <kcc@xxxxxxxxxx>
Cc: Dmitry Chernenkov <dmitryc@xxxxxxxxxx>
Cc: Andrey Konovalov <adech.fo@xxxxxxxxx>
Cc: Yuri Gribov <tetra2005@xxxxxxxxx>
Cc: Konstantin Khlebnikov <koct9i@xxxxxxxxx>
Cc: Sasha Levin <sasha.levin@xxxxxxxxxx>
Cc: Michal Marek <mmarek@xxxxxxx>
Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Cc: Ingo Molnar <mingo@xxxxxxxxxx>
Cc: Christoph Lameter <cl@xxxxxxxxx>
Cc: Pekka Enberg <penberg@xxxxxxxxxx>
Cc: David Rientjes <rientjes@xxxxxxxxxx>
Cc: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx>
Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
Cc: Dave Hansen <dave.hansen@xxxxxxxxx>
Cc: Andi Kleen <andi@xxxxxxxxxxxxxx>
Cc: Vegard Nossum <vegard.nossum@xxxxxxxxx>
Cc: H. Peter Anvin <hpa@xxxxxxxxx>
Cc: <x86@xxxxxxxxxx>
Cc: <linux-mm@xxxxxxxxx>
Cc: Randy Dunlap <rdunlap@xxxxxxxxxxxxx>
Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Cc: Alexander Viro <viro@xxxxxxxxxxxxxxxxxx>
Cc: Dave Jones <davej@xxxxxxxxxx>
Cc: Jonathan Corbet <corbet@xxxxxxx>
Cc: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
--
2.2.2

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/