1Kernel Memory Leak Detector
2===========================
3
4Kmemleak provides a way of detecting possible kernel memory leaks in a
5way similar to a `tracing garbage collector
6<https://en.wikipedia.org/wiki/Tracing_garbage_collection>`_,
7with the difference that the orphan objects are not freed but only
8reported via /sys/kernel/debug/kmemleak. A similar method is used by the
9Valgrind tool (``memcheck --leak-check``) to detect the memory leaks in
10user-space applications.
11Kmemleak is supported on x86, arm, powerpc, sparc, sh, microblaze, ppc, mips, s390 and tile.
12
13Usage
14-----
15
16CONFIG_DEBUG_KMEMLEAK in "Kernel hacking" has to be enabled. A kernel
17thread scans the memory every 10 minutes (by default) and prints the
18number of new unreferenced objects found. If the ``debugfs`` isn't already
19mounted, mount with::
20
21  # mount -t debugfs nodev /sys/kernel/debug/
22
23To display the details of all the possible scanned memory leaks::
24
25  # cat /sys/kernel/debug/kmemleak
26
27To trigger an intermediate memory scan::
28
29  # echo scan > /sys/kernel/debug/kmemleak
30
31To clear the list of all current possible memory leaks::
32
33  # echo clear > /sys/kernel/debug/kmemleak
34
35New leaks will then come up upon reading ``/sys/kernel/debug/kmemleak``
36again.
37
38Note that the orphan objects are listed in the order they were allocated
39and one object at the beginning of the list may cause other subsequent
40objects to be reported as orphan.
41
42Memory scanning parameters can be modified at run-time by writing to the
43``/sys/kernel/debug/kmemleak`` file. The following parameters are supported:
44
45- off
46    disable kmemleak (irreversible)
47- stack=on
48    enable the task stacks scanning (default)
49- stack=off
50    disable the tasks stacks scanning
51- scan=on
52    start the automatic memory scanning thread (default)
53- scan=off
54    stop the automatic memory scanning thread
55- scan=<secs>
56    set the automatic memory scanning period in seconds
57    (default 600, 0 to stop the automatic scanning)
58- scan
59    trigger a memory scan
60- clear
61    clear list of current memory leak suspects, done by
62    marking all current reported unreferenced objects grey,
63    or free all kmemleak objects if kmemleak has been disabled.
64- dump=<addr>
65    dump information about the object found at <addr>
66
67Kmemleak can also be disabled at boot-time by passing ``kmemleak=off`` on
68the kernel command line.
69
70Memory may be allocated or freed before kmemleak is initialised and
71these actions are stored in an early log buffer. The size of this buffer
72is configured via the CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE option.
73
74If CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF are enabled, the kmemleak is
75disabled by default. Passing ``kmemleak=on`` on the kernel command
76line enables the function.
77
78If you are getting errors like "Error while writing to stdout" or "write_loop:
79Invalid argument", make sure kmemleak is properly enabled.
80
81Basic Algorithm
82---------------
83
84The memory allocations via :c:func:`kmalloc`, :c:func:`vmalloc`,
85:c:func:`kmem_cache_alloc` and
86friends are traced and the pointers, together with additional
87information like size and stack trace, are stored in a rbtree.
88The corresponding freeing function calls are tracked and the pointers
89removed from the kmemleak data structures.
90
91An allocated block of memory is considered orphan if no pointer to its
92start address or to any location inside the block can be found by
93scanning the memory (including saved registers). This means that there
94might be no way for the kernel to pass the address of the allocated
95block to a freeing function and therefore the block is considered a
96memory leak.
97
98The scanning algorithm steps:
99
100  1. mark all objects as white (remaining white objects will later be
101     considered orphan)
102  2. scan the memory starting with the data section and stacks, checking
103     the values against the addresses stored in the rbtree. If
104     a pointer to a white object is found, the object is added to the
105     gray list
106  3. scan the gray objects for matching addresses (some white objects
107     can become gray and added at the end of the gray list) until the
108     gray set is finished
109  4. the remaining white objects are considered orphan and reported via
110     /sys/kernel/debug/kmemleak
111
112Some allocated memory blocks have pointers stored in the kernel's
113internal data structures and they cannot be detected as orphans. To
114avoid this, kmemleak can also store the number of values pointing to an
115address inside the block address range that need to be found so that the
116block is not considered a leak. One example is __vmalloc().
117
118Testing specific sections with kmemleak
119---------------------------------------
120
121Upon initial bootup your /sys/kernel/debug/kmemleak output page may be
122quite extensive. This can also be the case if you have very buggy code
123when doing development. To work around these situations you can use the
124'clear' command to clear all reported unreferenced objects from the
125/sys/kernel/debug/kmemleak output. By issuing a 'scan' after a 'clear'
126you can find new unreferenced objects; this should help with testing
127specific sections of code.
128
129To test a critical section on demand with a clean kmemleak do::
130
131  # echo clear > /sys/kernel/debug/kmemleak
132  ... test your kernel or modules ...
133  # echo scan > /sys/kernel/debug/kmemleak
134
135Then as usual to get your report with::
136
137  # cat /sys/kernel/debug/kmemleak
138
139Freeing kmemleak internal objects
140---------------------------------
141
142To allow access to previously found memory leaks after kmemleak has been
143disabled by the user or due to an fatal error, internal kmemleak objects
144won't be freed when kmemleak is disabled, and those objects may occupy
145a large part of physical memory.
146
147In this situation, you may reclaim memory with::
148
149  # echo clear > /sys/kernel/debug/kmemleak
150
151Kmemleak API
152------------
153
154See the include/linux/kmemleak.h header for the functions prototype.
155
156- ``kmemleak_init``		 - initialize kmemleak
157- ``kmemleak_alloc``		 - notify of a memory block allocation
158- ``kmemleak_alloc_percpu``	 - notify of a percpu memory block allocation
159- ``kmemleak_vmalloc``		 - notify of a vmalloc() memory allocation
160- ``kmemleak_free``		 - notify of a memory block freeing
161- ``kmemleak_free_part``	 - notify of a partial memory block freeing
162- ``kmemleak_free_percpu``	 - notify of a percpu memory block freeing
163- ``kmemleak_update_trace``	 - update object allocation stack trace
164- ``kmemleak_not_leak``	 - mark an object as not a leak
165- ``kmemleak_ignore``		 - do not scan or report an object as leak
166- ``kmemleak_scan_area``	 - add scan areas inside a memory block
167- ``kmemleak_no_scan``	 - do not scan a memory block
168- ``kmemleak_erase``		 - erase an old value in a pointer variable
169- ``kmemleak_alloc_recursive`` - as kmemleak_alloc but checks the recursiveness
170- ``kmemleak_free_recursive``	 - as kmemleak_free but checks the recursiveness
171
172The following functions take a physical address as the object pointer
173and only perform the corresponding action if the address has a lowmem
174mapping:
175
176- ``kmemleak_alloc_phys``
177- ``kmemleak_free_part_phys``
178- ``kmemleak_not_leak_phys``
179- ``kmemleak_ignore_phys``
180
181Dealing with false positives/negatives
182--------------------------------------
183
184The false negatives are real memory leaks (orphan objects) but not
185reported by kmemleak because values found during the memory scanning
186point to such objects. To reduce the number of false negatives, kmemleak
187provides the kmemleak_ignore, kmemleak_scan_area, kmemleak_no_scan and
188kmemleak_erase functions (see above). The task stacks also increase the
189amount of false negatives and their scanning is not enabled by default.
190
191The false positives are objects wrongly reported as being memory leaks
192(orphan). For objects known not to be leaks, kmemleak provides the
193kmemleak_not_leak function. The kmemleak_ignore could also be used if
194the memory block is known not to contain other pointers and it will no
195longer be scanned.
196
197Some of the reported leaks are only transient, especially on SMP
198systems, because of pointers temporarily stored in CPU registers or
199stacks. Kmemleak defines MSECS_MIN_AGE (defaulting to 1000) representing
200the minimum age of an object to be reported as a memory leak.
201
202Limitations and Drawbacks
203-------------------------
204
205The main drawback is the reduced performance of memory allocation and
206freeing. To avoid other penalties, the memory scanning is only performed
207when the /sys/kernel/debug/kmemleak file is read. Anyway, this tool is
208intended for debugging purposes where the performance might not be the
209most important requirement.
210
211To keep the algorithm simple, kmemleak scans for values pointing to any
212address inside a block's address range. This may lead to an increased
213number of false negatives. However, it is likely that a real memory leak
214will eventually become visible.
215
216Another source of false negatives is the data stored in non-pointer
217values. In a future version, kmemleak could only scan the pointer
218members in the allocated structures. This feature would solve many of
219the false negative cases described above.
220
221The tool can report false positives. These are cases where an allocated
222block doesn't need to be freed (some cases in the init_call functions),
223the pointer is calculated by other methods than the usual container_of
224macro or the pointer is stored in a location not scanned by kmemleak.
225
226Page allocations and ioremap are not tracked.
227
228Testing with kmemleak-test
229--------------------------
230
231To check if you have all set up to use kmemleak, you can use the kmemleak-test
232module, a module that deliberately leaks memory. Set CONFIG_DEBUG_KMEMLEAK_TEST
233as module (it can't be used as bult-in) and boot the kernel with kmemleak
234enabled. Load the module and perform a scan with::
235
236        # modprobe kmemleak-test
237        # echo scan > /sys/kernel/debug/kmemleak
238
239Note that the you may not get results instantly or on the first scanning. When
240kmemleak gets results, it'll log ``kmemleak: <count of leaks> new suspected
241memory leaks``. Then read the file to see then::
242
243        # cat /sys/kernel/debug/kmemleak
244        unreferenced object 0xffff89862ca702e8 (size 32):
245          comm "modprobe", pid 2088, jiffies 4294680594 (age 375.486s)
246          hex dump (first 32 bytes):
247            6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b  kkkkkkkkkkkkkkkk
248            6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b a5  kkkkkkkkkkkkkkk.
249          backtrace:
250            [<00000000e0a73ec7>] 0xffffffffc01d2036
251            [<000000000c5d2a46>] do_one_initcall+0x41/0x1df
252            [<0000000046db7e0a>] do_init_module+0x55/0x200
253            [<00000000542b9814>] load_module+0x203c/0x2480
254            [<00000000c2850256>] __do_sys_finit_module+0xba/0xe0
255            [<000000006564e7ef>] do_syscall_64+0x43/0x110
256            [<000000007c873fa6>] entry_SYSCALL_64_after_hwframe+0x44/0xa9
257        ...
258
259Removing the module with ``rmmod kmemleak_test`` should also trigger some
260kmemleak results.
261