1The Kernel Address Sanitizer (KASAN)
2====================================
3
4Overview
5--------
6
7KernelAddressSANitizer (KASAN) is a dynamic memory safety error detector
8designed to find out-of-bound and use-after-free bugs. KASAN has three modes:
9
101. generic KASAN (similar to userspace ASan),
112. software tag-based KASAN (similar to userspace HWASan),
123. hardware tag-based KASAN (based on hardware memory tagging).
13
14Generic KASAN is mainly used for debugging due to a large memory overhead.
15Software tag-based KASAN can be used for dogfood testing as it has a lower
16memory overhead that allows using it with real workloads. Hardware tag-based
17KASAN comes with low memory and performance overheads and, therefore, can be
18used in production. Either as an in-field memory bug detector or as a security
19mitigation.
20
21Software KASAN modes (#1 and #2) use compile-time instrumentation to insert
22validity checks before every memory access and, therefore, require a compiler
23version that supports that.
24
25Generic KASAN is supported in GCC and Clang. With GCC, it requires version
268.3.0 or later. Any supported Clang version is compatible, but detection of
27out-of-bounds accesses for global variables is only supported since Clang 11.
28
29Software tag-based KASAN mode is only supported in Clang.
30
31The hardware KASAN mode (#3) relies on hardware to perform the checks but
32still requires a compiler version that supports memory tagging instructions.
33This mode is supported in GCC 10+ and Clang 11+.
34
35Both software KASAN modes work with SLUB and SLAB memory allocators,
36while the hardware tag-based KASAN currently only supports SLUB.
37
38Currently, generic KASAN is supported for the x86_64, arm, arm64, xtensa, s390,
39and riscv architectures, and tag-based KASAN modes are supported only for arm64.
40
41Usage
42-----
43
44To enable KASAN, configure the kernel with::
45
46	  CONFIG_KASAN=y
47
48and choose between ``CONFIG_KASAN_GENERIC`` (to enable generic KASAN),
49``CONFIG_KASAN_SW_TAGS`` (to enable software tag-based KASAN), and
50``CONFIG_KASAN_HW_TAGS`` (to enable hardware tag-based KASAN).
51
52For software modes, also choose between ``CONFIG_KASAN_OUTLINE`` and
53``CONFIG_KASAN_INLINE``. Outline and inline are compiler instrumentation types.
54The former produces a smaller binary while the latter is 1.1-2 times faster.
55
56To include alloc and free stack traces of affected slab objects into reports,
57enable ``CONFIG_STACKTRACE``. To include alloc and free stack traces of affected
58physical pages, enable ``CONFIG_PAGE_OWNER`` and boot with ``page_owner=on``.
59
60Error reports
61~~~~~~~~~~~~~
62
63A typical KASAN report looks like this::
64
65    ==================================================================
66    BUG: KASAN: slab-out-of-bounds in kmalloc_oob_right+0xa8/0xbc [test_kasan]
67    Write of size 1 at addr ffff8801f44ec37b by task insmod/2760
68
69    CPU: 1 PID: 2760 Comm: insmod Not tainted 4.19.0-rc3+ #698
70    Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1 04/01/2014
71    Call Trace:
72     dump_stack+0x94/0xd8
73     print_address_description+0x73/0x280
74     kasan_report+0x144/0x187
75     __asan_report_store1_noabort+0x17/0x20
76     kmalloc_oob_right+0xa8/0xbc [test_kasan]
77     kmalloc_tests_init+0x16/0x700 [test_kasan]
78     do_one_initcall+0xa5/0x3ae
79     do_init_module+0x1b6/0x547
80     load_module+0x75df/0x8070
81     __do_sys_init_module+0x1c6/0x200
82     __x64_sys_init_module+0x6e/0xb0
83     do_syscall_64+0x9f/0x2c0
84     entry_SYSCALL_64_after_hwframe+0x44/0xa9
85    RIP: 0033:0x7f96443109da
86    RSP: 002b:00007ffcf0b51b08 EFLAGS: 00000202 ORIG_RAX: 00000000000000af
87    RAX: ffffffffffffffda RBX: 000055dc3ee521a0 RCX: 00007f96443109da
88    RDX: 00007f96445cff88 RSI: 0000000000057a50 RDI: 00007f9644992000
89    RBP: 000055dc3ee510b0 R08: 0000000000000003 R09: 0000000000000000
90    R10: 00007f964430cd0a R11: 0000000000000202 R12: 00007f96445cff88
91    R13: 000055dc3ee51090 R14: 0000000000000000 R15: 0000000000000000
92
93    Allocated by task 2760:
94     save_stack+0x43/0xd0
95     kasan_kmalloc+0xa7/0xd0
96     kmem_cache_alloc_trace+0xe1/0x1b0
97     kmalloc_oob_right+0x56/0xbc [test_kasan]
98     kmalloc_tests_init+0x16/0x700 [test_kasan]
99     do_one_initcall+0xa5/0x3ae
100     do_init_module+0x1b6/0x547
101     load_module+0x75df/0x8070
102     __do_sys_init_module+0x1c6/0x200
103     __x64_sys_init_module+0x6e/0xb0
104     do_syscall_64+0x9f/0x2c0
105     entry_SYSCALL_64_after_hwframe+0x44/0xa9
106
107    Freed by task 815:
108     save_stack+0x43/0xd0
109     __kasan_slab_free+0x135/0x190
110     kasan_slab_free+0xe/0x10
111     kfree+0x93/0x1a0
112     umh_complete+0x6a/0xa0
113     call_usermodehelper_exec_async+0x4c3/0x640
114     ret_from_fork+0x35/0x40
115
116    The buggy address belongs to the object at ffff8801f44ec300
117     which belongs to the cache kmalloc-128 of size 128
118    The buggy address is located 123 bytes inside of
119     128-byte region [ffff8801f44ec300, ffff8801f44ec380)
120    The buggy address belongs to the page:
121    page:ffffea0007d13b00 count:1 mapcount:0 mapping:ffff8801f7001640 index:0x0
122    flags: 0x200000000000100(slab)
123    raw: 0200000000000100 ffffea0007d11dc0 0000001a0000001a ffff8801f7001640
124    raw: 0000000000000000 0000000080150015 00000001ffffffff 0000000000000000
125    page dumped because: kasan: bad access detected
126
127    Memory state around the buggy address:
128     ffff8801f44ec200: fc fc fc fc fc fc fc fc fb fb fb fb fb fb fb fb
129     ffff8801f44ec280: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc
130    >ffff8801f44ec300: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 03
131                                                                    ^
132     ffff8801f44ec380: fc fc fc fc fc fc fc fc fb fb fb fb fb fb fb fb
133     ffff8801f44ec400: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc
134    ==================================================================
135
136The report header summarizes what kind of bug happened and what kind of access
137caused it. It is followed by a stack trace of the bad access, a stack trace of
138where the accessed memory was allocated (in case a slab object was accessed),
139and a stack trace of where the object was freed (in case of a use-after-free
140bug report). Next comes a description of the accessed slab object and the
141information about the accessed memory page.
142
143In the end, the report shows the memory state around the accessed address.
144Internally, KASAN tracks memory state separately for each memory granule, which
145is either 8 or 16 aligned bytes depending on KASAN mode. Each number in the
146memory state section of the report shows the state of one of the memory
147granules that surround the accessed address.
148
149For generic KASAN, the size of each memory granule is 8. The state of each
150granule is encoded in one shadow byte. Those 8 bytes can be accessible,
151partially accessible, freed, or be a part of a redzone. KASAN uses the following
152encoding for each shadow byte: 00 means that all 8 bytes of the corresponding
153memory region are accessible; number N (1 <= N <= 7) means that the first N
154bytes are accessible, and other (8 - N) bytes are not; any negative value
155indicates that the entire 8-byte word is inaccessible. KASAN uses different
156negative values to distinguish between different kinds of inaccessible memory
157like redzones or freed memory (see mm/kasan/kasan.h).
158
159In the report above, the arrow points to the shadow byte ``03``, which means
160that the accessed address is partially accessible.
161
162For tag-based KASAN modes, this last report section shows the memory tags around
163the accessed address (see the `Implementation details`_ section).
164
165Note that KASAN bug titles (like ``slab-out-of-bounds`` or ``use-after-free``)
166are best-effort: KASAN prints the most probable bug type based on the limited
167information it has. The actual type of the bug might be different.
168
169Generic KASAN also reports up to two auxiliary call stack traces. These stack
170traces point to places in code that interacted with the object but that are not
171directly present in the bad access stack trace. Currently, this includes
172call_rcu() and workqueue queuing.
173
174Boot parameters
175~~~~~~~~~~~~~~~
176
177KASAN is affected by the generic ``panic_on_warn`` command line parameter.
178When it is enabled, KASAN panics the kernel after printing a bug report.
179
180By default, KASAN prints a bug report only for the first invalid memory access.
181With ``kasan_multi_shot``, KASAN prints a report on every invalid access. This
182effectively disables ``panic_on_warn`` for KASAN reports.
183
184Hardware tag-based KASAN mode (see the section about various modes below) is
185intended for use in production as a security mitigation. Therefore, it supports
186boot parameters that allow disabling KASAN or controlling its features.
187
188- ``kasan=off`` or ``=on`` controls whether KASAN is enabled (default: ``on``).
189
190- ``kasan.mode=sync`` or ``=async`` controls whether KASAN is configured in
191  synchronous or asynchronous mode of execution (default: ``sync``).
192  Synchronous mode: a bad access is detected immediately when a tag
193  check fault occurs.
194  Asynchronous mode: a bad access detection is delayed. When a tag check
195  fault occurs, the information is stored in hardware (in the TFSR_EL1
196  register for arm64). The kernel periodically checks the hardware and
197  only reports tag faults during these checks.
198
199- ``kasan.stacktrace=off`` or ``=on`` disables or enables alloc and free stack
200  traces collection (default: ``on``).
201
202- ``kasan.fault=report`` or ``=panic`` controls whether to only print a KASAN
203  report or also panic the kernel (default: ``report``). The panic happens even
204  if ``kasan_multi_shot`` is enabled.
205
206Implementation details
207----------------------
208
209Generic KASAN
210~~~~~~~~~~~~~
211
212Software KASAN modes use shadow memory to record whether each byte of memory is
213safe to access and use compile-time instrumentation to insert shadow memory
214checks before each memory access.
215
216Generic KASAN dedicates 1/8th of kernel memory to its shadow memory (16TB
217to cover 128TB on x86_64) and uses direct mapping with a scale and offset to
218translate a memory address to its corresponding shadow address.
219
220Here is the function which translates an address to its corresponding shadow
221address::
222
223    static inline void *kasan_mem_to_shadow(const void *addr)
224    {
225	return (void *)((unsigned long)addr >> KASAN_SHADOW_SCALE_SHIFT)
226		+ KASAN_SHADOW_OFFSET;
227    }
228
229where ``KASAN_SHADOW_SCALE_SHIFT = 3``.
230
231Compile-time instrumentation is used to insert memory access checks. Compiler
232inserts function calls (``__asan_load*(addr)``, ``__asan_store*(addr)``) before
233each memory access of size 1, 2, 4, 8, or 16. These functions check whether
234memory accesses are valid or not by checking corresponding shadow memory.
235
236With inline instrumentation, instead of making function calls, the compiler
237directly inserts the code to check shadow memory. This option significantly
238enlarges the kernel, but it gives an x1.1-x2 performance boost over the
239outline-instrumented kernel.
240
241Generic KASAN is the only mode that delays the reuse of freed objects via
242quarantine (see mm/kasan/quarantine.c for implementation).
243
244Software tag-based KASAN
245~~~~~~~~~~~~~~~~~~~~~~~~
246
247Software tag-based KASAN uses a software memory tagging approach to checking
248access validity. It is currently only implemented for the arm64 architecture.
249
250Software tag-based KASAN uses the Top Byte Ignore (TBI) feature of arm64 CPUs
251to store a pointer tag in the top byte of kernel pointers. It uses shadow memory
252to store memory tags associated with each 16-byte memory cell (therefore, it
253dedicates 1/16th of the kernel memory for shadow memory).
254
255On each memory allocation, software tag-based KASAN generates a random tag, tags
256the allocated memory with this tag, and embeds the same tag into the returned
257pointer.
258
259Software tag-based KASAN uses compile-time instrumentation to insert checks
260before each memory access. These checks make sure that the tag of the memory
261that is being accessed is equal to the tag of the pointer that is used to access
262this memory. In case of a tag mismatch, software tag-based KASAN prints a bug
263report.
264
265Software tag-based KASAN also has two instrumentation modes (outline, which
266emits callbacks to check memory accesses; and inline, which performs the shadow
267memory checks inline). With outline instrumentation mode, a bug report is
268printed from the function that performs the access check. With inline
269instrumentation, a ``brk`` instruction is emitted by the compiler, and a
270dedicated ``brk`` handler is used to print bug reports.
271
272Software tag-based KASAN uses 0xFF as a match-all pointer tag (accesses through
273pointers with the 0xFF pointer tag are not checked). The value 0xFE is currently
274reserved to tag freed memory regions.
275
276Software tag-based KASAN currently only supports tagging of slab and page_alloc
277memory.
278
279Hardware tag-based KASAN
280~~~~~~~~~~~~~~~~~~~~~~~~
281
282Hardware tag-based KASAN is similar to the software mode in concept but uses
283hardware memory tagging support instead of compiler instrumentation and
284shadow memory.
285
286Hardware tag-based KASAN is currently only implemented for arm64 architecture
287and based on both arm64 Memory Tagging Extension (MTE) introduced in ARMv8.5
288Instruction Set Architecture and Top Byte Ignore (TBI).
289
290Special arm64 instructions are used to assign memory tags for each allocation.
291Same tags are assigned to pointers to those allocations. On every memory
292access, hardware makes sure that the tag of the memory that is being accessed is
293equal to the tag of the pointer that is used to access this memory. In case of a
294tag mismatch, a fault is generated, and a report is printed.
295
296Hardware tag-based KASAN uses 0xFF as a match-all pointer tag (accesses through
297pointers with the 0xFF pointer tag are not checked). The value 0xFE is currently
298reserved to tag freed memory regions.
299
300Hardware tag-based KASAN currently only supports tagging of slab and page_alloc
301memory.
302
303If the hardware does not support MTE (pre ARMv8.5), hardware tag-based KASAN
304will not be enabled. In this case, all KASAN boot parameters are ignored.
305
306Note that enabling CONFIG_KASAN_HW_TAGS always results in in-kernel TBI being
307enabled. Even when ``kasan.mode=off`` is provided or when the hardware does not
308support MTE (but supports TBI).
309
310Hardware tag-based KASAN only reports the first found bug. After that, MTE tag
311checking gets disabled.
312
313Shadow memory
314-------------
315
316The kernel maps memory in several different parts of the address space.
317The range of kernel virtual addresses is large: there is not enough real
318memory to support a real shadow region for every address that could be
319accessed by the kernel. Therefore, KASAN only maps real shadow for certain
320parts of the address space.
321
322Default behaviour
323~~~~~~~~~~~~~~~~~
324
325By default, architectures only map real memory over the shadow region
326for the linear mapping (and potentially other small areas). For all
327other areas - such as vmalloc and vmemmap space - a single read-only
328page is mapped over the shadow area. This read-only shadow page
329declares all memory accesses as permitted.
330
331This presents a problem for modules: they do not live in the linear
332mapping but in a dedicated module space. By hooking into the module
333allocator, KASAN temporarily maps real shadow memory to cover them.
334This allows detection of invalid accesses to module globals, for example.
335
336This also creates an incompatibility with ``VMAP_STACK``: if the stack
337lives in vmalloc space, it will be shadowed by the read-only page, and
338the kernel will fault when trying to set up the shadow data for stack
339variables.
340
341CONFIG_KASAN_VMALLOC
342~~~~~~~~~~~~~~~~~~~~
343
344With ``CONFIG_KASAN_VMALLOC``, KASAN can cover vmalloc space at the
345cost of greater memory usage. Currently, this is supported on x86,
346riscv, s390, and powerpc.
347
348This works by hooking into vmalloc and vmap and dynamically
349allocating real shadow memory to back the mappings.
350
351Most mappings in vmalloc space are small, requiring less than a full
352page of shadow space. Allocating a full shadow page per mapping would
353therefore be wasteful. Furthermore, to ensure that different mappings
354use different shadow pages, mappings would have to be aligned to
355``KASAN_GRANULE_SIZE * PAGE_SIZE``.
356
357Instead, KASAN shares backing space across multiple mappings. It allocates
358a backing page when a mapping in vmalloc space uses a particular page
359of the shadow region. This page can be shared by other vmalloc
360mappings later on.
361
362KASAN hooks into the vmap infrastructure to lazily clean up unused shadow
363memory.
364
365To avoid the difficulties around swapping mappings around, KASAN expects
366that the part of the shadow region that covers the vmalloc space will
367not be covered by the early shadow page but will be left unmapped.
368This will require changes in arch-specific code.
369
370This allows ``VMAP_STACK`` support on x86 and can simplify support of
371architectures that do not have a fixed module region.
372
373For developers
374--------------
375
376Ignoring accesses
377~~~~~~~~~~~~~~~~~
378
379Software KASAN modes use compiler instrumentation to insert validity checks.
380Such instrumentation might be incompatible with some parts of the kernel, and
381therefore needs to be disabled.
382
383Other parts of the kernel might access metadata for allocated objects.
384Normally, KASAN detects and reports such accesses, but in some cases (e.g.,
385in memory allocators), these accesses are valid.
386
387For software KASAN modes, to disable instrumentation for a specific file or
388directory, add a ``KASAN_SANITIZE`` annotation to the respective kernel
389Makefile:
390
391- For a single file (e.g., main.o)::
392
393    KASAN_SANITIZE_main.o := n
394
395- For all files in one directory::
396
397    KASAN_SANITIZE := n
398
399For software KASAN modes, to disable instrumentation on a per-function basis,
400use the KASAN-specific ``__no_sanitize_address`` function attribute or the
401generic ``noinstr`` one.
402
403Note that disabling compiler instrumentation (either on a per-file or a
404per-function basis) makes KASAN ignore the accesses that happen directly in
405that code for software KASAN modes. It does not help when the accesses happen
406indirectly (through calls to instrumented functions) or with the hardware
407tag-based mode that does not use compiler instrumentation.
408
409For software KASAN modes, to disable KASAN reports in a part of the kernel code
410for the current task, annotate this part of the code with a
411``kasan_disable_current()``/``kasan_enable_current()`` section. This also
412disables the reports for indirect accesses that happen through function calls.
413
414For tag-based KASAN modes (include the hardware one), to disable access
415checking, use ``kasan_reset_tag()`` or ``page_kasan_tag_reset()``. Note that
416temporarily disabling access checking via ``page_kasan_tag_reset()`` requires
417saving and restoring the per-page KASAN tag via
418``page_kasan_tag``/``page_kasan_tag_set``.
419
420Tests
421~~~~~
422
423There are KASAN tests that allow verifying that KASAN works and can detect
424certain types of memory corruptions. The tests consist of two parts:
425
4261. Tests that are integrated with the KUnit Test Framework. Enabled with
427``CONFIG_KASAN_KUNIT_TEST``. These tests can be run and partially verified
428automatically in a few different ways; see the instructions below.
429
4302. Tests that are currently incompatible with KUnit. Enabled with
431``CONFIG_KASAN_MODULE_TEST`` and can only be run as a module. These tests can
432only be verified manually by loading the kernel module and inspecting the
433kernel log for KASAN reports.
434
435Each KUnit-compatible KASAN test prints one of multiple KASAN reports if an
436error is detected. Then the test prints its number and status.
437
438When a test passes::
439
440        ok 28 - kmalloc_double_kzfree
441
442When a test fails due to a failed ``kmalloc``::
443
444        # kmalloc_large_oob_right: ASSERTION FAILED at lib/test_kasan.c:163
445        Expected ptr is not null, but is
446        not ok 4 - kmalloc_large_oob_right
447
448When a test fails due to a missing KASAN report::
449
450        # kmalloc_double_kzfree: EXPECTATION FAILED at lib/test_kasan.c:974
451        KASAN failure expected in "kfree_sensitive(ptr)", but none occurred
452        not ok 44 - kmalloc_double_kzfree
453
454
455At the end the cumulative status of all KASAN tests is printed. On success::
456
457        ok 1 - kasan
458
459Or, if one of the tests failed::
460
461        not ok 1 - kasan
462
463There are a few ways to run KUnit-compatible KASAN tests.
464
4651. Loadable module
466
467   With ``CONFIG_KUNIT`` enabled, KASAN-KUnit tests can be built as a loadable
468   module and run by loading ``test_kasan.ko`` with ``insmod`` or ``modprobe``.
469
4702. Built-In
471
472   With ``CONFIG_KUNIT`` built-in, KASAN-KUnit tests can be built-in as well.
473   In this case, the tests will run at boot as a late-init call.
474
4753. Using kunit_tool
476
477   With ``CONFIG_KUNIT`` and ``CONFIG_KASAN_KUNIT_TEST`` built-in, it is also
478   possible to use ``kunit_tool`` to see the results of KUnit tests in a more
479   readable way. This will not print the KASAN reports of the tests that passed.
480   See `KUnit documentation <https://www.kernel.org/doc/html/latest/dev-tools/kunit/index.html>`_
481   for more up-to-date information on ``kunit_tool``.
482
483.. _KUnit: https://www.kernel.org/doc/html/latest/dev-tools/kunit/index.html
484