2016-08-08 00:31:03 +03:00
The Kernel Address Sanitizer (KASAN)
====================================
Overview
--------
2020-12-22 23:03:35 +03:00
KernelAddressSANitizer (KASAN) is a dynamic memory safety error detector
designed to find out-of-bound and use-after-free bugs. KASAN has three modes:
2016-08-08 00:31:03 +03:00
2020-12-22 23:02:24 +03:00
1. generic KASAN (similar to userspace ASan),
2. software tag-based KASAN (similar to userspace HWASan),
3. hardware tag-based KASAN (based on hardware memory tagging).
2016-08-08 00:31:03 +03:00
2021-04-30 09:00:15 +03:00
Generic KASAN is mainly used for debugging due to a large memory overhead.
Software tag-based KASAN can be used for dogfood testing as it has a lower
memory overhead that allows using it with real workloads. Hardware tag-based
KASAN comes with low memory and performance overheads and, therefore, can be
used in production. Either as an in-field memory bug detector or as a security
mitigation.
Software KASAN modes (#1 and #2) use compile-time instrumentation to insert
validity checks before every memory access and, therefore, require a compiler
2020-12-22 23:02:24 +03:00
version that supports that.
2016-08-08 00:31:03 +03:00
2021-04-30 09:00:15 +03:00
Generic KASAN is supported in GCC and Clang. With GCC, it requires version
2020-10-14 02:47:51 +03:00
8.3.0 or later. Any supported Clang version is compatible, but detection of
2020-08-07 09:24:31 +03:00
out-of-bounds accesses for global variables is only supported since Clang 11.
2018-12-28 11:31:10 +03:00
2021-04-30 09:00:15 +03:00
Software tag-based KASAN mode is only supported in Clang.
2018-12-28 11:31:10 +03:00
2021-04-30 09:00:15 +03:00
The hardware KASAN mode (#3) relies on hardware to perform the checks but
still requires a compiler version that supports memory tagging instructions.
This mode is supported in GCC 10+ and Clang 11+.
Both software KASAN modes work with SLUB and SLAB memory allocators,
while the hardware tag-based KASAN currently only supports SLUB.
Currently, generic KASAN is supported for the x86_64, arm, arm64, xtensa, s390,
2020-12-22 23:02:24 +03:00
and riscv architectures, and tag-based KASAN modes are supported only for arm64.
2016-08-08 00:31:03 +03:00
Usage
-----
2021-04-30 09:00:18 +03:00
To enable KASAN, configure the kernel with::
2016-08-08 00:31:03 +03:00
2021-04-30 09:00:18 +03:00
CONFIG_KASAN=y
2016-08-08 00:31:03 +03:00
2021-04-30 09:00:18 +03:00
and choose between `` CONFIG_KASAN_GENERIC `` (to enable generic KASAN),
`` CONFIG_KASAN_SW_TAGS `` (to enable software tag-based KASAN), and
`` CONFIG_KASAN_HW_TAGS `` (to enable hardware tag-based KASAN).
2018-12-28 11:31:10 +03:00
2021-04-30 09:00:18 +03:00
For software modes, also choose between `` CONFIG_KASAN_OUTLINE `` and
`` CONFIG_KASAN_INLINE `` . Outline and inline are compiler instrumentation types.
The former produces a smaller binary while the latter is 1.1-2 times faster.
2018-12-28 11:31:10 +03:00
2021-04-30 09:00:18 +03:00
To include alloc and free stack traces of affected slab objects into reports,
enable `` CONFIG_STACKTRACE `` . To include alloc and free stack traces of affected
physical pages, enable `` CONFIG_PAGE_OWNER `` and boot with `` page_owner=on `` .
2019-10-15 00:11:44 +03:00
2016-08-08 00:31:03 +03:00
Error reports
~~~~~~~~~~~~~
2021-04-30 09:00:21 +03:00
A typical KASAN report looks like this::
2016-08-08 00:31:03 +03:00
==================================================================
2018-12-28 11:31:10 +03:00
BUG: KASAN: slab-out-of-bounds in kmalloc_oob_right+0xa8/0xbc [test_kasan]
Write of size 1 at addr ffff8801f44ec37b by task insmod/2760
CPU: 1 PID: 2760 Comm: insmod Not tainted 4.19.0-rc3+ #698
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1 04/01/2014
2016-08-08 00:31:03 +03:00
Call Trace:
2018-12-28 11:31:10 +03:00
dump_stack+0x94/0xd8
print_address_description+0x73/0x280
kasan_report+0x144/0x187
__asan_report_store1_noabort+0x17/0x20
kmalloc_oob_right+0xa8/0xbc [test_kasan]
kmalloc_tests_init+0x16/0x700 [test_kasan]
do_one_initcall+0xa5/0x3ae
do_init_module+0x1b6/0x547
load_module+0x75df/0x8070
__do_sys_init_module+0x1c6/0x200
__x64_sys_init_module+0x6e/0xb0
do_syscall_64+0x9f/0x2c0
entry_SYSCALL_64_after_hwframe+0x44/0xa9
RIP: 0033:0x7f96443109da
RSP: 002b:00007ffcf0b51b08 EFLAGS: 00000202 ORIG_RAX: 00000000000000af
RAX: ffffffffffffffda RBX: 000055dc3ee521a0 RCX: 00007f96443109da
RDX: 00007f96445cff88 RSI: 0000000000057a50 RDI: 00007f9644992000
RBP: 000055dc3ee510b0 R08: 0000000000000003 R09: 0000000000000000
R10: 00007f964430cd0a R11: 0000000000000202 R12: 00007f96445cff88
R13: 000055dc3ee51090 R14: 0000000000000000 R15: 0000000000000000
Allocated by task 2760:
save_stack+0x43/0xd0
kasan_kmalloc+0xa7/0xd0
kmem_cache_alloc_trace+0xe1/0x1b0
kmalloc_oob_right+0x56/0xbc [test_kasan]
kmalloc_tests_init+0x16/0x700 [test_kasan]
do_one_initcall+0xa5/0x3ae
do_init_module+0x1b6/0x547
load_module+0x75df/0x8070
__do_sys_init_module+0x1c6/0x200
__x64_sys_init_module+0x6e/0xb0
do_syscall_64+0x9f/0x2c0
entry_SYSCALL_64_after_hwframe+0x44/0xa9
Freed by task 815:
save_stack+0x43/0xd0
__kasan_slab_free+0x135/0x190
kasan_slab_free+0xe/0x10
kfree+0x93/0x1a0
umh_complete+0x6a/0xa0
call_usermodehelper_exec_async+0x4c3/0x640
ret_from_fork+0x35/0x40
The buggy address belongs to the object at ffff8801f44ec300
which belongs to the cache kmalloc-128 of size 128
The buggy address is located 123 bytes inside of
128-byte region [ffff8801f44ec300, ffff8801f44ec380)
The buggy address belongs to the page:
page:ffffea0007d13b00 count:1 mapcount:0 mapping:ffff8801f7001640 index:0x0
flags: 0x200000000000100(slab)
raw: 0200000000000100 ffffea0007d11dc0 0000001a0000001a ffff8801f7001640
raw: 0000000000000000 0000000080150015 00000001ffffffff 0000000000000000
page dumped because: kasan: bad access detected
2016-08-08 00:31:03 +03:00
Memory state around the buggy address:
2018-12-28 11:31:10 +03:00
ffff8801f44ec200: fc fc fc fc fc fc fc fc fb fb fb fb fb fb fb fb
ffff8801f44ec280: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc
>ffff8801f44ec300: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 03
^
ffff8801f44ec380: fc fc fc fc fc fc fc fc fb fb fb fb fb fb fb fb
ffff8801f44ec400: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc
2016-08-08 00:31:03 +03:00
==================================================================
2021-04-30 09:00:21 +03:00
The report header summarizes what kind of bug happened and what kind of access
caused it. It is followed by a stack trace of the bad access, a stack trace of
where the accessed memory was allocated (in case a slab object was accessed),
and a stack trace of where the object was freed (in case of a use-after-free
bug report). Next comes a description of the accessed slab object and the
information about the accessed memory page.
2016-08-08 00:31:03 +03:00
2021-04-30 09:00:21 +03:00
In the end, the report shows the memory state around the accessed address.
Internally, KASAN tracks memory state separately for each memory granule, which
2020-12-22 23:03:35 +03:00
is either 8 or 16 aligned bytes depending on KASAN mode. Each number in the
memory state section of the report shows the state of one of the memory
granules that surround the accessed address.
2021-04-30 09:00:21 +03:00
For generic KASAN, the size of each memory granule is 8. The state of each
2020-12-22 23:03:35 +03:00
granule is encoded in one shadow byte. Those 8 bytes can be accessible,
2021-04-30 09:00:21 +03:00
partially accessible, freed, or be a part of a redzone. KASAN uses the following
encoding for each shadow byte: 00 means that all 8 bytes of the corresponding
2020-12-22 23:03:35 +03:00
memory region are accessible; number N (1 <= N <= 7) means that the first N
bytes are accessible, and other (8 - N) bytes are not; any negative value
indicates that the entire 8-byte word is inaccessible. KASAN uses different
negative values to distinguish between different kinds of inaccessible memory
like redzones or freed memory (see mm/kasan/kasan.h).
2016-08-08 00:31:03 +03:00
2021-04-30 09:00:21 +03:00
In the report above, the arrow points to the shadow byte `` 03 `` , which means
that the accessed address is partially accessible.
For tag-based KASAN modes, this last report section shows the memory tags around
the accessed address (see the `Implementation details`_ section).
Note that KASAN bug titles (like `` slab-out-of-bounds `` or `` use-after-free `` )
are best-effort: KASAN prints the most probable bug type based on the limited
information it has. The actual type of the bug might be different.
Generic KASAN also reports up to two auxiliary call stack traces. These stack
traces point to places in code that interacted with the object but that are not
directly present in the bad access stack trace. Currently, this includes
call_rcu() and workqueue queuing.
2020-12-22 23:03:35 +03:00
Boot parameters
~~~~~~~~~~~~~~~
2021-04-30 09:00:24 +03:00
KASAN is affected by the generic `` panic_on_warn `` command line parameter.
When it is enabled, KASAN panics the kernel after printing a bug report.
By default, KASAN prints a bug report only for the first invalid memory access.
With `` kasan_multi_shot `` , KASAN prints a report on every invalid access. This
effectively disables `` panic_on_warn `` for KASAN reports.
2021-02-24 23:05:09 +03:00
Hardware tag-based KASAN mode (see the section about various modes below) is
2021-02-26 04:20:38 +03:00
intended for use in production as a security mitigation. Therefore, it supports
2021-04-30 09:00:24 +03:00
boot parameters that allow disabling KASAN or controlling its features.
2020-12-22 23:03:35 +03:00
2021-01-24 08:01:34 +03:00
- `` kasan=off `` or `` =on `` controls whether KASAN is enabled (default: `` on `` ).
2020-12-22 23:03:35 +03:00
2021-03-15 16:20:12 +03:00
- `` kasan.mode=sync `` or `` =async `` controls whether KASAN is configured in
synchronous or asynchronous mode of execution (default: `` sync `` ).
Synchronous mode: a bad access is detected immediately when a tag
check fault occurs.
Asynchronous mode: a bad access detection is delayed. When a tag check
fault occurs, the information is stored in hardware (in the TFSR_EL1
register for arm64). The kernel periodically checks the hardware and
only reports tag faults during these checks.
2021-01-24 08:01:34 +03:00
- `` kasan.stacktrace=off `` or `` =on `` disables or enables alloc and free stack
2021-02-10 00:42:03 +03:00
traces collection (default: `` on `` ).
2020-12-22 23:03:35 +03:00
2021-01-24 08:01:34 +03:00
- `` kasan.fault=report `` or `` =panic `` controls whether to only print a KASAN
2021-04-30 09:00:24 +03:00
report or also panic the kernel (default: `` report `` ). The panic happens even
if `` kasan_multi_shot `` is enabled.
2020-12-22 23:03:35 +03:00
2016-08-08 00:31:03 +03:00
Implementation details
----------------------
2018-12-28 11:31:10 +03:00
Generic KASAN
~~~~~~~~~~~~~
2021-04-30 09:00:27 +03:00
Software KASAN modes use shadow memory to record whether each byte of memory is
safe to access and use compile-time instrumentation to insert shadow memory
checks before each memory access.
2016-08-08 00:31:03 +03:00
2021-04-30 09:00:27 +03:00
Generic KASAN dedicates 1/8th of kernel memory to its shadow memory (16TB
2018-12-28 11:31:10 +03:00
to cover 128TB on x86_64) and uses direct mapping with a scale and offset to
translate a memory address to its corresponding shadow address.
2016-08-08 00:31:03 +03:00
Here is the function which translates an address to its corresponding shadow
address::
static inline void *kasan_mem_to_shadow(const void * addr)
{
2021-04-30 09:00:27 +03:00
return (void *)((unsigned long)addr >> KASAN_SHADOW_SCALE_SHIFT)
2016-08-08 00:31:03 +03:00
+ KASAN_SHADOW_OFFSET;
}
where `` KASAN_SHADOW_SCALE_SHIFT = 3 `` .
2018-12-28 11:31:10 +03:00
Compile-time instrumentation is used to insert memory access checks. Compiler
2021-04-30 09:00:27 +03:00
inserts 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 accesses are valid or not by checking corresponding shadow memory.
2016-08-08 00:31:03 +03:00
2021-04-30 09:00:27 +03:00
With inline instrumentation, instead of making function calls, the compiler
directly inserts the code to check shadow memory. This option significantly
enlarges the kernel, but it gives an x1.1-x2 performance boost over the
outline-instrumented kernel.
2018-12-28 11:31:10 +03:00
2021-04-30 09:00:27 +03:00
Generic KASAN is the only mode that delays the reuse of freed objects via
2020-12-22 23:03:35 +03:00
quarantine (see mm/kasan/quarantine.c for implementation).
2018-12-28 11:31:10 +03:00
Software tag-based KASAN
~~~~~~~~~~~~~~~~~~~~~~~~
2020-12-22 23:02:24 +03:00
Software tag-based KASAN requires software memory tagging support in the form
of HWASan-like compiler instrumentation (see HWASan documentation for details).
Software tag-based KASAN is currently only implemented for arm64 architecture.
Software tag-based KASAN uses the Top Byte Ignore (TBI) feature of arm64 CPUs
to store a pointer tag in the top byte of kernel pointers. Like generic KASAN
it uses shadow memory to store memory tags associated with each 16-byte memory
2018-12-28 11:31:10 +03:00
cell (therefore it dedicates 1/16th of the kernel memory for shadow memory).
2020-12-22 23:02:24 +03:00
On each memory allocation software tag-based KASAN generates a random tag, tags
the allocated memory with this tag, and embeds this tag into the returned
pointer.
2018-12-28 11:31:10 +03:00
Software tag-based KASAN uses compile-time instrumentation to insert checks
before each memory access. These checks make sure that tag of the memory that
is being accessed is equal to tag of the pointer that is used to access this
2020-12-22 23:02:24 +03:00
memory. In case of a tag mismatch software tag-based KASAN prints a bug report.
2018-12-28 11:31:10 +03:00
Software tag-based KASAN also has two instrumentation modes (outline, that
emits callbacks to check memory accesses; and inline, that performs the shadow
memory checks inline). With outline instrumentation mode, a bug report is
simply printed from the function that performs the access check. With inline
instrumentation a brk instruction is emitted by the compiler, and a dedicated
brk handler is used to print bug reports.
2020-12-22 23:02:24 +03:00
Software tag-based KASAN uses 0xFF as a match-all pointer tag (accesses through
pointers with 0xFF pointer tag aren't checked). The value 0xFE is currently
reserved to tag freed memory regions.
Software tag-based KASAN currently only supports tagging of
kmem_cache_alloc/kmalloc and page_alloc memory.
Hardware tag-based KASAN
~~~~~~~~~~~~~~~~~~~~~~~~
Hardware tag-based KASAN is similar to the software mode in concept, but uses
hardware memory tagging support instead of compiler instrumentation and
shadow memory.
Hardware tag-based KASAN is currently only implemented for arm64 architecture
and based on both arm64 Memory Tagging Extension (MTE) introduced in ARMv8.5
Instruction Set Architecture, and Top Byte Ignore (TBI).
Special arm64 instructions are used to assign memory tags for each allocation.
Same tags are assigned to pointers to those allocations. On every memory
access, hardware makes sure that tag of the memory that is being accessed is
equal to tag of the pointer that is used to access this memory. In case of a
tag mismatch a fault is generated and a report is printed.
Hardware tag-based KASAN uses 0xFF as a match-all pointer tag (accesses through
pointers with 0xFF pointer tag aren't checked). The value 0xFE is currently
reserved to tag freed memory regions.
Hardware tag-based KASAN currently only supports tagging of
kmem_cache_alloc/kmalloc and page_alloc memory.
kasan: support backing vmalloc space with real shadow memory
Patch series "kasan: support backing vmalloc space with real shadow
memory", v11.
Currently, vmalloc space is backed by the early shadow page. This means
that kasan is incompatible with VMAP_STACK.
This series provides a mechanism to back vmalloc space with real,
dynamically allocated memory. I have only wired up x86, because that's
the only currently supported arch I can work with easily, but it's very
easy to wire up other architectures, and it appears that there is some
work-in-progress code to do this on arm64 and s390.
This has been discussed before in the context of VMAP_STACK:
- https://bugzilla.kernel.org/show_bug.cgi?id=202009
- https://lkml.org/lkml/2018/7/22/198
- https://lkml.org/lkml/2019/7/19/822
In terms of implementation details:
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=1)
This is unfortunate but given that this is a debug feature only, not the
end of the world. The benchmarks are also a stress-test for the vmalloc
subsystem: they're not indicative of an overall 2x slowdown!
This patch (of 4):
Hook into vmalloc and vmap, and dynamically allocate real shadow memory
to back the mappings.
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
To avoid the difficulties around swapping mappings around, this code
expects that the part of the shadow region that covers the vmalloc space
will not be covered by the early shadow page, but will be left unmapped.
This will require changes in arch-specific code.
This allows KASAN with VMAP_STACK, and may be helpful for architectures
that do not have a separate module space (e.g. powerpc64, which I am
currently working on). It also allows relaxing the module alignment
back to PAGE_SIZE.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=3D1)
This is unfortunate but given that this is a debug feature only, not the
end of the world.
The full benchmark results are:
Performance
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 662004 11404956 17.23 19144610 28.92 1.68
full_fit_alloc_test 710950 12029752 16.92 13184651 18.55 1.10
long_busy_list_alloc_test 9431875 43990172 4.66 82970178 8.80 1.89
random_size_alloc_test 5033626 23061762 4.58 47158834 9.37 2.04
fix_align_alloc_test 1252514 15276910 12.20 31266116 24.96 2.05
random_size_align_alloc_te 1648501 14578321 8.84 25560052 15.51 1.75
align_shift_alloc_test 147 830 5.65 5692 38.72 6.86
pcpu_alloc_test 80732 125520 1.55 140864 1.74 1.12
Total Cycles 119240774314 763211341128 6.40 1390338696894 11.66 1.82
Sequential, 2 cpus
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 1423150 14276550 10.03 27733022 19.49 1.94
full_fit_alloc_test 1754219 14722640 8.39 15030786 8.57 1.02
long_busy_list_alloc_test 11451858 52154973 4.55 107016027 9.34 2.05
random_size_alloc_test 5989020 26735276 4.46 68885923 11.50 2.58
fix_align_alloc_test 2050976 20166900 9.83 50491675 24.62 2.50
random_size_align_alloc_te 2858229 17971700 6.29 38730225 13.55 2.16
align_shift_alloc_test 405 6428 15.87 26253 64.82 4.08
pcpu_alloc_test 127183 151464 1.19 216263 1.70 1.43
Total Cycles 54181269392 308723699764 5.70 650772566394 12.01 2.11
fix_size_alloc_test 1420404 14289308 10.06 27790035 19.56 1.94
full_fit_alloc_test 1736145 14806234 8.53 15274301 8.80 1.03
long_busy_list_alloc_test 11404638 52270785 4.58 107550254 9.43 2.06
random_size_alloc_test 6017006 26650625 4.43 68696127 11.42 2.58
fix_align_alloc_test 2045504 20280985 9.91 50414862 24.65 2.49
random_size_align_alloc_te 2845338 17931018 6.30 38510276 13.53 2.15
align_shift_alloc_test 472 3760 7.97 9656 20.46 2.57
pcpu_alloc_test 118643 132732 1.12 146504 1.23 1.10
Total Cycles 54040011688 309102805492 5.72 651325675652 12.05 2.11
[dja@axtens.net: fixups]
Link: http://lkml.kernel.org/r/20191120052719.7201-1-dja@axtens.net
Link: https://bugzilla.kernel.org/show_bug.cgi?id=3D202009
Link: http://lkml.kernel.org/r/20191031093909.9228-2-dja@axtens.net
Signed-off-by: Mark Rutland <mark.rutland@arm.com> [shadow rework]
Signed-off-by: Daniel Axtens <dja@axtens.net>
Co-developed-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Vasily Gorbik <gor@linux.ibm.com>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-12-01 04:54:50 +03:00
2021-02-24 23:05:09 +03:00
If the hardware doesn't support MTE (pre ARMv8.5), hardware tag-based KASAN
won't be enabled. In this case all boot parameters are ignored.
Note, that enabling CONFIG_KASAN_HW_TAGS always results in in-kernel TBI being
enabled. Even when kasan.mode=off is provided, or when the hardware doesn't
support MTE (but supports TBI).
2021-02-26 04:20:38 +03:00
Hardware tag-based KASAN only reports the first found bug. After that MTE tag
checking gets disabled.
2021-04-30 09:00:12 +03:00
Shadow memory
-------------
kasan: support backing vmalloc space with real shadow memory
Patch series "kasan: support backing vmalloc space with real shadow
memory", v11.
Currently, vmalloc space is backed by the early shadow page. This means
that kasan is incompatible with VMAP_STACK.
This series provides a mechanism to back vmalloc space with real,
dynamically allocated memory. I have only wired up x86, because that's
the only currently supported arch I can work with easily, but it's very
easy to wire up other architectures, and it appears that there is some
work-in-progress code to do this on arm64 and s390.
This has been discussed before in the context of VMAP_STACK:
- https://bugzilla.kernel.org/show_bug.cgi?id=202009
- https://lkml.org/lkml/2018/7/22/198
- https://lkml.org/lkml/2019/7/19/822
In terms of implementation details:
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=1)
This is unfortunate but given that this is a debug feature only, not the
end of the world. The benchmarks are also a stress-test for the vmalloc
subsystem: they're not indicative of an overall 2x slowdown!
This patch (of 4):
Hook into vmalloc and vmap, and dynamically allocate real shadow memory
to back the mappings.
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
To avoid the difficulties around swapping mappings around, this code
expects that the part of the shadow region that covers the vmalloc space
will not be covered by the early shadow page, but will be left unmapped.
This will require changes in arch-specific code.
This allows KASAN with VMAP_STACK, and may be helpful for architectures
that do not have a separate module space (e.g. powerpc64, which I am
currently working on). It also allows relaxing the module alignment
back to PAGE_SIZE.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=3D1)
This is unfortunate but given that this is a debug feature only, not the
end of the world.
The full benchmark results are:
Performance
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 662004 11404956 17.23 19144610 28.92 1.68
full_fit_alloc_test 710950 12029752 16.92 13184651 18.55 1.10
long_busy_list_alloc_test 9431875 43990172 4.66 82970178 8.80 1.89
random_size_alloc_test 5033626 23061762 4.58 47158834 9.37 2.04
fix_align_alloc_test 1252514 15276910 12.20 31266116 24.96 2.05
random_size_align_alloc_te 1648501 14578321 8.84 25560052 15.51 1.75
align_shift_alloc_test 147 830 5.65 5692 38.72 6.86
pcpu_alloc_test 80732 125520 1.55 140864 1.74 1.12
Total Cycles 119240774314 763211341128 6.40 1390338696894 11.66 1.82
Sequential, 2 cpus
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 1423150 14276550 10.03 27733022 19.49 1.94
full_fit_alloc_test 1754219 14722640 8.39 15030786 8.57 1.02
long_busy_list_alloc_test 11451858 52154973 4.55 107016027 9.34 2.05
random_size_alloc_test 5989020 26735276 4.46 68885923 11.50 2.58
fix_align_alloc_test 2050976 20166900 9.83 50491675 24.62 2.50
random_size_align_alloc_te 2858229 17971700 6.29 38730225 13.55 2.16
align_shift_alloc_test 405 6428 15.87 26253 64.82 4.08
pcpu_alloc_test 127183 151464 1.19 216263 1.70 1.43
Total Cycles 54181269392 308723699764 5.70 650772566394 12.01 2.11
fix_size_alloc_test 1420404 14289308 10.06 27790035 19.56 1.94
full_fit_alloc_test 1736145 14806234 8.53 15274301 8.80 1.03
long_busy_list_alloc_test 11404638 52270785 4.58 107550254 9.43 2.06
random_size_alloc_test 6017006 26650625 4.43 68696127 11.42 2.58
fix_align_alloc_test 2045504 20280985 9.91 50414862 24.65 2.49
random_size_align_alloc_te 2845338 17931018 6.30 38510276 13.53 2.15
align_shift_alloc_test 472 3760 7.97 9656 20.46 2.57
pcpu_alloc_test 118643 132732 1.12 146504 1.23 1.10
Total Cycles 54040011688 309102805492 5.72 651325675652 12.05 2.11
[dja@axtens.net: fixups]
Link: http://lkml.kernel.org/r/20191120052719.7201-1-dja@axtens.net
Link: https://bugzilla.kernel.org/show_bug.cgi?id=3D202009
Link: http://lkml.kernel.org/r/20191031093909.9228-2-dja@axtens.net
Signed-off-by: Mark Rutland <mark.rutland@arm.com> [shadow rework]
Signed-off-by: Daniel Axtens <dja@axtens.net>
Co-developed-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Vasily Gorbik <gor@linux.ibm.com>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-12-01 04:54:50 +03:00
The kernel maps memory in a number of different parts of the address
space. This poses something of a problem for KASAN, which requires
that all addresses accessed by instrumented code have a valid shadow
region.
The range of kernel virtual addresses is large: there is not enough
real memory to support a real shadow region for every address that
could be accessed by the kernel.
2021-04-30 09:00:12 +03:00
Default behaviour
~~~~~~~~~~~~~~~~~
kasan: support backing vmalloc space with real shadow memory
Patch series "kasan: support backing vmalloc space with real shadow
memory", v11.
Currently, vmalloc space is backed by the early shadow page. This means
that kasan is incompatible with VMAP_STACK.
This series provides a mechanism to back vmalloc space with real,
dynamically allocated memory. I have only wired up x86, because that's
the only currently supported arch I can work with easily, but it's very
easy to wire up other architectures, and it appears that there is some
work-in-progress code to do this on arm64 and s390.
This has been discussed before in the context of VMAP_STACK:
- https://bugzilla.kernel.org/show_bug.cgi?id=202009
- https://lkml.org/lkml/2018/7/22/198
- https://lkml.org/lkml/2019/7/19/822
In terms of implementation details:
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=1)
This is unfortunate but given that this is a debug feature only, not the
end of the world. The benchmarks are also a stress-test for the vmalloc
subsystem: they're not indicative of an overall 2x slowdown!
This patch (of 4):
Hook into vmalloc and vmap, and dynamically allocate real shadow memory
to back the mappings.
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
To avoid the difficulties around swapping mappings around, this code
expects that the part of the shadow region that covers the vmalloc space
will not be covered by the early shadow page, but will be left unmapped.
This will require changes in arch-specific code.
This allows KASAN with VMAP_STACK, and may be helpful for architectures
that do not have a separate module space (e.g. powerpc64, which I am
currently working on). It also allows relaxing the module alignment
back to PAGE_SIZE.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=3D1)
This is unfortunate but given that this is a debug feature only, not the
end of the world.
The full benchmark results are:
Performance
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 662004 11404956 17.23 19144610 28.92 1.68
full_fit_alloc_test 710950 12029752 16.92 13184651 18.55 1.10
long_busy_list_alloc_test 9431875 43990172 4.66 82970178 8.80 1.89
random_size_alloc_test 5033626 23061762 4.58 47158834 9.37 2.04
fix_align_alloc_test 1252514 15276910 12.20 31266116 24.96 2.05
random_size_align_alloc_te 1648501 14578321 8.84 25560052 15.51 1.75
align_shift_alloc_test 147 830 5.65 5692 38.72 6.86
pcpu_alloc_test 80732 125520 1.55 140864 1.74 1.12
Total Cycles 119240774314 763211341128 6.40 1390338696894 11.66 1.82
Sequential, 2 cpus
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 1423150 14276550 10.03 27733022 19.49 1.94
full_fit_alloc_test 1754219 14722640 8.39 15030786 8.57 1.02
long_busy_list_alloc_test 11451858 52154973 4.55 107016027 9.34 2.05
random_size_alloc_test 5989020 26735276 4.46 68885923 11.50 2.58
fix_align_alloc_test 2050976 20166900 9.83 50491675 24.62 2.50
random_size_align_alloc_te 2858229 17971700 6.29 38730225 13.55 2.16
align_shift_alloc_test 405 6428 15.87 26253 64.82 4.08
pcpu_alloc_test 127183 151464 1.19 216263 1.70 1.43
Total Cycles 54181269392 308723699764 5.70 650772566394 12.01 2.11
fix_size_alloc_test 1420404 14289308 10.06 27790035 19.56 1.94
full_fit_alloc_test 1736145 14806234 8.53 15274301 8.80 1.03
long_busy_list_alloc_test 11404638 52270785 4.58 107550254 9.43 2.06
random_size_alloc_test 6017006 26650625 4.43 68696127 11.42 2.58
fix_align_alloc_test 2045504 20280985 9.91 50414862 24.65 2.49
random_size_align_alloc_te 2845338 17931018 6.30 38510276 13.53 2.15
align_shift_alloc_test 472 3760 7.97 9656 20.46 2.57
pcpu_alloc_test 118643 132732 1.12 146504 1.23 1.10
Total Cycles 54040011688 309102805492 5.72 651325675652 12.05 2.11
[dja@axtens.net: fixups]
Link: http://lkml.kernel.org/r/20191120052719.7201-1-dja@axtens.net
Link: https://bugzilla.kernel.org/show_bug.cgi?id=3D202009
Link: http://lkml.kernel.org/r/20191031093909.9228-2-dja@axtens.net
Signed-off-by: Mark Rutland <mark.rutland@arm.com> [shadow rework]
Signed-off-by: Daniel Axtens <dja@axtens.net>
Co-developed-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Vasily Gorbik <gor@linux.ibm.com>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-12-01 04:54:50 +03:00
By default, architectures only map real memory over the shadow region
for the linear mapping (and potentially other small areas). For all
other areas - such as vmalloc and vmemmap space - a single read-only
page is mapped over the shadow area. This read-only shadow page
declares all memory accesses as permitted.
This presents a problem for modules: they do not live in the linear
mapping, but in a dedicated module space. By hooking in to the module
allocator, KASAN can temporarily map real shadow memory to cover
them. This allows detection of invalid accesses to module globals, for
example.
This also creates an incompatibility with `` VMAP_STACK `` : if the stack
lives in vmalloc space, it will be shadowed by the read-only page, and
the kernel will fault when trying to set up the shadow data for stack
variables.
CONFIG_KASAN_VMALLOC
~~~~~~~~~~~~~~~~~~~~
With `` CONFIG_KASAN_VMALLOC `` , KASAN can cover vmalloc space at the
cost of greater memory usage. Currently this is only supported on x86.
This works by hooking into vmalloc and vmap, and dynamically
allocating real shadow memory to back the mappings.
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
2020-12-22 23:00:24 +03:00
`` KASAN_GRANULE_SIZE * PAGE_SIZE `` .
kasan: support backing vmalloc space with real shadow memory
Patch series "kasan: support backing vmalloc space with real shadow
memory", v11.
Currently, vmalloc space is backed by the early shadow page. This means
that kasan is incompatible with VMAP_STACK.
This series provides a mechanism to back vmalloc space with real,
dynamically allocated memory. I have only wired up x86, because that's
the only currently supported arch I can work with easily, but it's very
easy to wire up other architectures, and it appears that there is some
work-in-progress code to do this on arm64 and s390.
This has been discussed before in the context of VMAP_STACK:
- https://bugzilla.kernel.org/show_bug.cgi?id=202009
- https://lkml.org/lkml/2018/7/22/198
- https://lkml.org/lkml/2019/7/19/822
In terms of implementation details:
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=1)
This is unfortunate but given that this is a debug feature only, not the
end of the world. The benchmarks are also a stress-test for the vmalloc
subsystem: they're not indicative of an overall 2x slowdown!
This patch (of 4):
Hook into vmalloc and vmap, and dynamically allocate real shadow memory
to back the mappings.
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
To avoid the difficulties around swapping mappings around, this code
expects that the part of the shadow region that covers the vmalloc space
will not be covered by the early shadow page, but will be left unmapped.
This will require changes in arch-specific code.
This allows KASAN with VMAP_STACK, and may be helpful for architectures
that do not have a separate module space (e.g. powerpc64, which I am
currently working on). It also allows relaxing the module alignment
back to PAGE_SIZE.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=3D1)
This is unfortunate but given that this is a debug feature only, not the
end of the world.
The full benchmark results are:
Performance
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 662004 11404956 17.23 19144610 28.92 1.68
full_fit_alloc_test 710950 12029752 16.92 13184651 18.55 1.10
long_busy_list_alloc_test 9431875 43990172 4.66 82970178 8.80 1.89
random_size_alloc_test 5033626 23061762 4.58 47158834 9.37 2.04
fix_align_alloc_test 1252514 15276910 12.20 31266116 24.96 2.05
random_size_align_alloc_te 1648501 14578321 8.84 25560052 15.51 1.75
align_shift_alloc_test 147 830 5.65 5692 38.72 6.86
pcpu_alloc_test 80732 125520 1.55 140864 1.74 1.12
Total Cycles 119240774314 763211341128 6.40 1390338696894 11.66 1.82
Sequential, 2 cpus
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 1423150 14276550 10.03 27733022 19.49 1.94
full_fit_alloc_test 1754219 14722640 8.39 15030786 8.57 1.02
long_busy_list_alloc_test 11451858 52154973 4.55 107016027 9.34 2.05
random_size_alloc_test 5989020 26735276 4.46 68885923 11.50 2.58
fix_align_alloc_test 2050976 20166900 9.83 50491675 24.62 2.50
random_size_align_alloc_te 2858229 17971700 6.29 38730225 13.55 2.16
align_shift_alloc_test 405 6428 15.87 26253 64.82 4.08
pcpu_alloc_test 127183 151464 1.19 216263 1.70 1.43
Total Cycles 54181269392 308723699764 5.70 650772566394 12.01 2.11
fix_size_alloc_test 1420404 14289308 10.06 27790035 19.56 1.94
full_fit_alloc_test 1736145 14806234 8.53 15274301 8.80 1.03
long_busy_list_alloc_test 11404638 52270785 4.58 107550254 9.43 2.06
random_size_alloc_test 6017006 26650625 4.43 68696127 11.42 2.58
fix_align_alloc_test 2045504 20280985 9.91 50414862 24.65 2.49
random_size_align_alloc_te 2845338 17931018 6.30 38510276 13.53 2.15
align_shift_alloc_test 472 3760 7.97 9656 20.46 2.57
pcpu_alloc_test 118643 132732 1.12 146504 1.23 1.10
Total Cycles 54040011688 309102805492 5.72 651325675652 12.05 2.11
[dja@axtens.net: fixups]
Link: http://lkml.kernel.org/r/20191120052719.7201-1-dja@axtens.net
Link: https://bugzilla.kernel.org/show_bug.cgi?id=3D202009
Link: http://lkml.kernel.org/r/20191031093909.9228-2-dja@axtens.net
Signed-off-by: Mark Rutland <mark.rutland@arm.com> [shadow rework]
Signed-off-by: Daniel Axtens <dja@axtens.net>
Co-developed-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Vasily Gorbik <gor@linux.ibm.com>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-12-01 04:54:50 +03:00
2020-12-22 23:03:35 +03:00
Instead, KASAN shares backing space across multiple mappings. It allocates
kasan: support backing vmalloc space with real shadow memory
Patch series "kasan: support backing vmalloc space with real shadow
memory", v11.
Currently, vmalloc space is backed by the early shadow page. This means
that kasan is incompatible with VMAP_STACK.
This series provides a mechanism to back vmalloc space with real,
dynamically allocated memory. I have only wired up x86, because that's
the only currently supported arch I can work with easily, but it's very
easy to wire up other architectures, and it appears that there is some
work-in-progress code to do this on arm64 and s390.
This has been discussed before in the context of VMAP_STACK:
- https://bugzilla.kernel.org/show_bug.cgi?id=202009
- https://lkml.org/lkml/2018/7/22/198
- https://lkml.org/lkml/2019/7/19/822
In terms of implementation details:
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=1)
This is unfortunate but given that this is a debug feature only, not the
end of the world. The benchmarks are also a stress-test for the vmalloc
subsystem: they're not indicative of an overall 2x slowdown!
This patch (of 4):
Hook into vmalloc and vmap, and dynamically allocate real shadow memory
to back the mappings.
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
To avoid the difficulties around swapping mappings around, this code
expects that the part of the shadow region that covers the vmalloc space
will not be covered by the early shadow page, but will be left unmapped.
This will require changes in arch-specific code.
This allows KASAN with VMAP_STACK, and may be helpful for architectures
that do not have a separate module space (e.g. powerpc64, which I am
currently working on). It also allows relaxing the module alignment
back to PAGE_SIZE.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=3D1)
This is unfortunate but given that this is a debug feature only, not the
end of the world.
The full benchmark results are:
Performance
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 662004 11404956 17.23 19144610 28.92 1.68
full_fit_alloc_test 710950 12029752 16.92 13184651 18.55 1.10
long_busy_list_alloc_test 9431875 43990172 4.66 82970178 8.80 1.89
random_size_alloc_test 5033626 23061762 4.58 47158834 9.37 2.04
fix_align_alloc_test 1252514 15276910 12.20 31266116 24.96 2.05
random_size_align_alloc_te 1648501 14578321 8.84 25560052 15.51 1.75
align_shift_alloc_test 147 830 5.65 5692 38.72 6.86
pcpu_alloc_test 80732 125520 1.55 140864 1.74 1.12
Total Cycles 119240774314 763211341128 6.40 1390338696894 11.66 1.82
Sequential, 2 cpus
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 1423150 14276550 10.03 27733022 19.49 1.94
full_fit_alloc_test 1754219 14722640 8.39 15030786 8.57 1.02
long_busy_list_alloc_test 11451858 52154973 4.55 107016027 9.34 2.05
random_size_alloc_test 5989020 26735276 4.46 68885923 11.50 2.58
fix_align_alloc_test 2050976 20166900 9.83 50491675 24.62 2.50
random_size_align_alloc_te 2858229 17971700 6.29 38730225 13.55 2.16
align_shift_alloc_test 405 6428 15.87 26253 64.82 4.08
pcpu_alloc_test 127183 151464 1.19 216263 1.70 1.43
Total Cycles 54181269392 308723699764 5.70 650772566394 12.01 2.11
fix_size_alloc_test 1420404 14289308 10.06 27790035 19.56 1.94
full_fit_alloc_test 1736145 14806234 8.53 15274301 8.80 1.03
long_busy_list_alloc_test 11404638 52270785 4.58 107550254 9.43 2.06
random_size_alloc_test 6017006 26650625 4.43 68696127 11.42 2.58
fix_align_alloc_test 2045504 20280985 9.91 50414862 24.65 2.49
random_size_align_alloc_te 2845338 17931018 6.30 38510276 13.53 2.15
align_shift_alloc_test 472 3760 7.97 9656 20.46 2.57
pcpu_alloc_test 118643 132732 1.12 146504 1.23 1.10
Total Cycles 54040011688 309102805492 5.72 651325675652 12.05 2.11
[dja@axtens.net: fixups]
Link: http://lkml.kernel.org/r/20191120052719.7201-1-dja@axtens.net
Link: https://bugzilla.kernel.org/show_bug.cgi?id=3D202009
Link: http://lkml.kernel.org/r/20191031093909.9228-2-dja@axtens.net
Signed-off-by: Mark Rutland <mark.rutland@arm.com> [shadow rework]
Signed-off-by: Daniel Axtens <dja@axtens.net>
Co-developed-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Vasily Gorbik <gor@linux.ibm.com>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-12-01 04:54:50 +03:00
a backing page when a mapping in vmalloc space uses a particular page
of the shadow region. This page can be shared by other vmalloc
mappings later on.
2020-12-22 23:03:35 +03:00
KASAN hooks into the vmap infrastructure to lazily clean up unused shadow
kasan: support backing vmalloc space with real shadow memory
Patch series "kasan: support backing vmalloc space with real shadow
memory", v11.
Currently, vmalloc space is backed by the early shadow page. This means
that kasan is incompatible with VMAP_STACK.
This series provides a mechanism to back vmalloc space with real,
dynamically allocated memory. I have only wired up x86, because that's
the only currently supported arch I can work with easily, but it's very
easy to wire up other architectures, and it appears that there is some
work-in-progress code to do this on arm64 and s390.
This has been discussed before in the context of VMAP_STACK:
- https://bugzilla.kernel.org/show_bug.cgi?id=202009
- https://lkml.org/lkml/2018/7/22/198
- https://lkml.org/lkml/2019/7/19/822
In terms of implementation details:
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=1)
This is unfortunate but given that this is a debug feature only, not the
end of the world. The benchmarks are also a stress-test for the vmalloc
subsystem: they're not indicative of an overall 2x slowdown!
This patch (of 4):
Hook into vmalloc and vmap, and dynamically allocate real shadow memory
to back the mappings.
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
To avoid the difficulties around swapping mappings around, this code
expects that the part of the shadow region that covers the vmalloc space
will not be covered by the early shadow page, but will be left unmapped.
This will require changes in arch-specific code.
This allows KASAN with VMAP_STACK, and may be helpful for architectures
that do not have a separate module space (e.g. powerpc64, which I am
currently working on). It also allows relaxing the module alignment
back to PAGE_SIZE.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=3D1)
This is unfortunate but given that this is a debug feature only, not the
end of the world.
The full benchmark results are:
Performance
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 662004 11404956 17.23 19144610 28.92 1.68
full_fit_alloc_test 710950 12029752 16.92 13184651 18.55 1.10
long_busy_list_alloc_test 9431875 43990172 4.66 82970178 8.80 1.89
random_size_alloc_test 5033626 23061762 4.58 47158834 9.37 2.04
fix_align_alloc_test 1252514 15276910 12.20 31266116 24.96 2.05
random_size_align_alloc_te 1648501 14578321 8.84 25560052 15.51 1.75
align_shift_alloc_test 147 830 5.65 5692 38.72 6.86
pcpu_alloc_test 80732 125520 1.55 140864 1.74 1.12
Total Cycles 119240774314 763211341128 6.40 1390338696894 11.66 1.82
Sequential, 2 cpus
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 1423150 14276550 10.03 27733022 19.49 1.94
full_fit_alloc_test 1754219 14722640 8.39 15030786 8.57 1.02
long_busy_list_alloc_test 11451858 52154973 4.55 107016027 9.34 2.05
random_size_alloc_test 5989020 26735276 4.46 68885923 11.50 2.58
fix_align_alloc_test 2050976 20166900 9.83 50491675 24.62 2.50
random_size_align_alloc_te 2858229 17971700 6.29 38730225 13.55 2.16
align_shift_alloc_test 405 6428 15.87 26253 64.82 4.08
pcpu_alloc_test 127183 151464 1.19 216263 1.70 1.43
Total Cycles 54181269392 308723699764 5.70 650772566394 12.01 2.11
fix_size_alloc_test 1420404 14289308 10.06 27790035 19.56 1.94
full_fit_alloc_test 1736145 14806234 8.53 15274301 8.80 1.03
long_busy_list_alloc_test 11404638 52270785 4.58 107550254 9.43 2.06
random_size_alloc_test 6017006 26650625 4.43 68696127 11.42 2.58
fix_align_alloc_test 2045504 20280985 9.91 50414862 24.65 2.49
random_size_align_alloc_te 2845338 17931018 6.30 38510276 13.53 2.15
align_shift_alloc_test 472 3760 7.97 9656 20.46 2.57
pcpu_alloc_test 118643 132732 1.12 146504 1.23 1.10
Total Cycles 54040011688 309102805492 5.72 651325675652 12.05 2.11
[dja@axtens.net: fixups]
Link: http://lkml.kernel.org/r/20191120052719.7201-1-dja@axtens.net
Link: https://bugzilla.kernel.org/show_bug.cgi?id=3D202009
Link: http://lkml.kernel.org/r/20191031093909.9228-2-dja@axtens.net
Signed-off-by: Mark Rutland <mark.rutland@arm.com> [shadow rework]
Signed-off-by: Daniel Axtens <dja@axtens.net>
Co-developed-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Vasily Gorbik <gor@linux.ibm.com>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-12-01 04:54:50 +03:00
memory.
2020-12-22 23:03:35 +03:00
To avoid the difficulties around swapping mappings around, KASAN expects
kasan: support backing vmalloc space with real shadow memory
Patch series "kasan: support backing vmalloc space with real shadow
memory", v11.
Currently, vmalloc space is backed by the early shadow page. This means
that kasan is incompatible with VMAP_STACK.
This series provides a mechanism to back vmalloc space with real,
dynamically allocated memory. I have only wired up x86, because that's
the only currently supported arch I can work with easily, but it's very
easy to wire up other architectures, and it appears that there is some
work-in-progress code to do this on arm64 and s390.
This has been discussed before in the context of VMAP_STACK:
- https://bugzilla.kernel.org/show_bug.cgi?id=202009
- https://lkml.org/lkml/2018/7/22/198
- https://lkml.org/lkml/2019/7/19/822
In terms of implementation details:
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=1)
This is unfortunate but given that this is a debug feature only, not the
end of the world. The benchmarks are also a stress-test for the vmalloc
subsystem: they're not indicative of an overall 2x slowdown!
This patch (of 4):
Hook into vmalloc and vmap, and dynamically allocate real shadow memory
to back the mappings.
Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
Instead, share backing space across multiple mappings. Allocate a
backing page when a mapping in vmalloc space uses a particular page of
the shadow region. This page can be shared by other vmalloc mappings
later on.
We hook in to the vmap infrastructure to lazily clean up unused shadow
memory.
To avoid the difficulties around swapping mappings around, this code
expects that the part of the shadow region that covers the vmalloc space
will not be covered by the early shadow page, but will be left unmapped.
This will require changes in arch-specific code.
This allows KASAN with VMAP_STACK, and may be helpful for architectures
that do not have a separate module space (e.g. powerpc64, which I am
currently working on). It also allows relaxing the module alignment
back to PAGE_SIZE.
Testing with test_vmalloc.sh on an x86 VM with 2 vCPUs shows that:
- Turning on KASAN, inline instrumentation, without vmalloc, introuduces
a 4.1x-4.2x slowdown in vmalloc operations.
- Turning this on introduces the following slowdowns over KASAN:
* ~1.76x slower single-threaded (test_vmalloc.sh performance)
* ~2.18x slower when both cpus are performing operations
simultaneously (test_vmalloc.sh sequential_test_order=3D1)
This is unfortunate but given that this is a debug feature only, not the
end of the world.
The full benchmark results are:
Performance
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 662004 11404956 17.23 19144610 28.92 1.68
full_fit_alloc_test 710950 12029752 16.92 13184651 18.55 1.10
long_busy_list_alloc_test 9431875 43990172 4.66 82970178 8.80 1.89
random_size_alloc_test 5033626 23061762 4.58 47158834 9.37 2.04
fix_align_alloc_test 1252514 15276910 12.20 31266116 24.96 2.05
random_size_align_alloc_te 1648501 14578321 8.84 25560052 15.51 1.75
align_shift_alloc_test 147 830 5.65 5692 38.72 6.86
pcpu_alloc_test 80732 125520 1.55 140864 1.74 1.12
Total Cycles 119240774314 763211341128 6.40 1390338696894 11.66 1.82
Sequential, 2 cpus
No KASAN KASAN original x baseline KASAN vmalloc x baseline x KASAN
fix_size_alloc_test 1423150 14276550 10.03 27733022 19.49 1.94
full_fit_alloc_test 1754219 14722640 8.39 15030786 8.57 1.02
long_busy_list_alloc_test 11451858 52154973 4.55 107016027 9.34 2.05
random_size_alloc_test 5989020 26735276 4.46 68885923 11.50 2.58
fix_align_alloc_test 2050976 20166900 9.83 50491675 24.62 2.50
random_size_align_alloc_te 2858229 17971700 6.29 38730225 13.55 2.16
align_shift_alloc_test 405 6428 15.87 26253 64.82 4.08
pcpu_alloc_test 127183 151464 1.19 216263 1.70 1.43
Total Cycles 54181269392 308723699764 5.70 650772566394 12.01 2.11
fix_size_alloc_test 1420404 14289308 10.06 27790035 19.56 1.94
full_fit_alloc_test 1736145 14806234 8.53 15274301 8.80 1.03
long_busy_list_alloc_test 11404638 52270785 4.58 107550254 9.43 2.06
random_size_alloc_test 6017006 26650625 4.43 68696127 11.42 2.58
fix_align_alloc_test 2045504 20280985 9.91 50414862 24.65 2.49
random_size_align_alloc_te 2845338 17931018 6.30 38510276 13.53 2.15
align_shift_alloc_test 472 3760 7.97 9656 20.46 2.57
pcpu_alloc_test 118643 132732 1.12 146504 1.23 1.10
Total Cycles 54040011688 309102805492 5.72 651325675652 12.05 2.11
[dja@axtens.net: fixups]
Link: http://lkml.kernel.org/r/20191120052719.7201-1-dja@axtens.net
Link: https://bugzilla.kernel.org/show_bug.cgi?id=3D202009
Link: http://lkml.kernel.org/r/20191031093909.9228-2-dja@axtens.net
Signed-off-by: Mark Rutland <mark.rutland@arm.com> [shadow rework]
Signed-off-by: Daniel Axtens <dja@axtens.net>
Co-developed-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Vasily Gorbik <gor@linux.ibm.com>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-12-01 04:54:50 +03:00
that the part of the shadow region that covers the vmalloc space will
not be covered by the early shadow page, but will be left
unmapped. This will require changes in arch-specific code.
This allows `` VMAP_STACK `` support on x86, and can simplify support of
architectures that do not have a fixed module region.
2020-10-14 02:55:09 +03:00
2021-04-30 09:00:12 +03:00
For developers
--------------
Ignoring accesses
~~~~~~~~~~~~~~~~~
Software KASAN modes use compiler instrumentation to insert validity checks.
Such instrumentation might be incompatible with some part of the kernel, and
therefore needs to be disabled. To disable instrumentation for specific files
or directories, add a line similar to the following to the respective kernel
Makefile:
- For a single file (e.g. main.o)::
KASAN_SANITIZE_main.o := n
- For all files in one directory::
KASAN_SANITIZE := n
Tests
~~~~~
2020-10-14 02:55:09 +03:00
2021-02-24 23:05:29 +03:00
KASAN tests consist of two parts:
2020-12-22 23:03:35 +03:00
1. Tests that are integrated with the KUnit Test Framework. Enabled with
`` CONFIG_KASAN_KUNIT_TEST `` . These tests can be run and partially verified
automatically in a few different ways, see the instructions below.
2020-10-14 02:55:09 +03:00
2020-12-22 23:03:35 +03:00
2. Tests that are currently incompatible with KUnit. Enabled with
2021-02-24 23:05:29 +03:00
`` CONFIG_KASAN_MODULE_TEST `` and can only be run as a module. These tests can
2020-12-22 23:03:35 +03:00
only be verified manually, by loading the kernel module and inspecting the
kernel log for KASAN reports.
2020-10-14 02:55:09 +03:00
2020-12-22 23:03:35 +03:00
Each KUnit-compatible KASAN test prints a KASAN report if an error is detected.
Then the test prints its number and status.
When a test passes::
2020-10-14 02:55:09 +03:00
ok 28 - kmalloc_double_kzfree
2020-10-27 12:51:09 +03:00
2020-12-22 23:03:35 +03:00
When a test fails due to a failed `` kmalloc `` ::
2020-10-14 02:55:09 +03:00
# kmalloc_large_oob_right: ASSERTION FAILED at lib/test_kasan.c:163
Expected ptr is not null, but is
not ok 4 - kmalloc_large_oob_right
2020-10-27 12:51:09 +03:00
2020-12-22 23:03:35 +03:00
When a test fails due to a missing KASAN report::
2020-10-14 02:55:09 +03:00
# kmalloc_double_kzfree: EXPECTATION FAILED at lib/test_kasan.c:629
Expected kasan_data->report_expected == kasan_data->report_found, but
kasan_data->report_expected == 1
kasan_data->report_found == 0
not ok 28 - kmalloc_double_kzfree
2020-12-22 23:03:35 +03:00
At the end the cumulative status of all KASAN tests is printed. On success::
2020-10-14 02:55:09 +03:00
ok 1 - kasan
2020-12-22 23:03:35 +03:00
Or, if one of the tests failed::
2020-10-14 02:55:09 +03:00
not ok 1 - kasan
2020-12-22 23:03:35 +03:00
There are a few ways to run KUnit-compatible KASAN tests.
1. Loadable module
2020-10-14 02:55:09 +03:00
With `` CONFIG_KUNIT `` enabled, `` CONFIG_KASAN_KUNIT_TEST `` can be built as
2020-12-22 23:03:35 +03:00
a loadable module and run on any architecture that supports KASAN by loading
the module with insmod or modprobe. The module is called `` test_kasan `` .
2020-10-14 02:55:09 +03:00
2020-12-22 23:03:35 +03:00
2. Built-In
2020-10-14 02:55:09 +03:00
With `` CONFIG_KUNIT `` built-in, `` CONFIG_KASAN_KUNIT_TEST `` can be built-in
2020-12-22 23:03:35 +03:00
on any architecure that supports KASAN. These and any other KUnit tests enabled
will run and print the results at boot as a late-init call.
2020-10-14 02:55:09 +03:00
2020-12-22 23:03:35 +03:00
3. Using kunit_tool
2020-10-14 02:55:09 +03:00
2020-12-22 23:03:35 +03:00
With `` CONFIG_KUNIT `` and `` CONFIG_KASAN_KUNIT_TEST `` built-in, it's also
possible use `` kunit_tool `` to see the results of these and other KUnit tests
in a more readable way. This will not print the KASAN reports of the tests that
passed. Use `KUnit documentation <https://www.kernel.org/doc/html/latest/dev-tools/kunit/index.html> `_
for more up-to-date information on `` kunit_tool `` .
2020-10-14 02:55:09 +03:00
.. _KUnit: https://www.kernel.org/doc/html/latest/dev-tools/kunit/index.html