kmemleak.rst 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259
  1. Kernel Memory Leak Detector
  2. ===========================
  3. Kmemleak provides a way of detecting possible kernel memory leaks in a
  4. way similar to a `tracing garbage collector
  5. <https://en.wikipedia.org/wiki/Tracing_garbage_collection>`_,
  6. with the difference that the orphan objects are not freed but only
  7. reported via /sys/kernel/debug/kmemleak. A similar method is used by the
  8. Valgrind tool (``memcheck --leak-check``) to detect the memory leaks in
  9. user-space applications.
  10. Usage
  11. -----
  12. CONFIG_DEBUG_KMEMLEAK in "Kernel hacking" has to be enabled. A kernel
  13. thread scans the memory every 10 minutes (by default) and prints the
  14. number of new unreferenced objects found. If the ``debugfs`` isn't already
  15. mounted, mount with::
  16. # mount -t debugfs nodev /sys/kernel/debug/
  17. To display the details of all the possible scanned memory leaks::
  18. # cat /sys/kernel/debug/kmemleak
  19. To trigger an intermediate memory scan::
  20. # echo scan > /sys/kernel/debug/kmemleak
  21. To clear the list of all current possible memory leaks::
  22. # echo clear > /sys/kernel/debug/kmemleak
  23. New leaks will then come up upon reading ``/sys/kernel/debug/kmemleak``
  24. again.
  25. Note that the orphan objects are listed in the order they were allocated
  26. and one object at the beginning of the list may cause other subsequent
  27. objects to be reported as orphan.
  28. Memory scanning parameters can be modified at run-time by writing to the
  29. ``/sys/kernel/debug/kmemleak`` file. The following parameters are supported:
  30. - off
  31. disable kmemleak (irreversible)
  32. - stack=on
  33. enable the task stacks scanning (default)
  34. - stack=off
  35. disable the tasks stacks scanning
  36. - scan=on
  37. start the automatic memory scanning thread (default)
  38. - scan=off
  39. stop the automatic memory scanning thread
  40. - scan=<secs>
  41. set the automatic memory scanning period in seconds
  42. (default 600, 0 to stop the automatic scanning)
  43. - scan
  44. trigger a memory scan
  45. - clear
  46. clear list of current memory leak suspects, done by
  47. marking all current reported unreferenced objects grey,
  48. or free all kmemleak objects if kmemleak has been disabled.
  49. - dump=<addr>
  50. dump information about the object found at <addr>
  51. Kmemleak can also be disabled at boot-time by passing ``kmemleak=off`` on
  52. the kernel command line.
  53. Memory may be allocated or freed before kmemleak is initialised and
  54. these actions are stored in an early log buffer. The size of this buffer
  55. is configured via the CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE option.
  56. If CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF are enabled, the kmemleak is
  57. disabled by default. Passing ``kmemleak=on`` on the kernel command
  58. line enables the function.
  59. If you are getting errors like "Error while writing to stdout" or "write_loop:
  60. Invalid argument", make sure kmemleak is properly enabled.
  61. Basic Algorithm
  62. ---------------
  63. The memory allocations via :c:func:`kmalloc`, :c:func:`vmalloc`,
  64. :c:func:`kmem_cache_alloc` and
  65. friends are traced and the pointers, together with additional
  66. information like size and stack trace, are stored in a rbtree.
  67. The corresponding freeing function calls are tracked and the pointers
  68. removed from the kmemleak data structures.
  69. An allocated block of memory is considered orphan if no pointer to its
  70. start address or to any location inside the block can be found by
  71. scanning the memory (including saved registers). This means that there
  72. might be no way for the kernel to pass the address of the allocated
  73. block to a freeing function and therefore the block is considered a
  74. memory leak.
  75. The scanning algorithm steps:
  76. 1. mark all objects as white (remaining white objects will later be
  77. considered orphan)
  78. 2. scan the memory starting with the data section and stacks, checking
  79. the values against the addresses stored in the rbtree. If
  80. a pointer to a white object is found, the object is added to the
  81. gray list
  82. 3. scan the gray objects for matching addresses (some white objects
  83. can become gray and added at the end of the gray list) until the
  84. gray set is finished
  85. 4. the remaining white objects are considered orphan and reported via
  86. /sys/kernel/debug/kmemleak
  87. Some allocated memory blocks have pointers stored in the kernel's
  88. internal data structures and they cannot be detected as orphans. To
  89. avoid this, kmemleak can also store the number of values pointing to an
  90. address inside the block address range that need to be found so that the
  91. block is not considered a leak. One example is __vmalloc().
  92. Testing specific sections with kmemleak
  93. ---------------------------------------
  94. Upon initial bootup your /sys/kernel/debug/kmemleak output page may be
  95. quite extensive. This can also be the case if you have very buggy code
  96. when doing development. To work around these situations you can use the
  97. 'clear' command to clear all reported unreferenced objects from the
  98. /sys/kernel/debug/kmemleak output. By issuing a 'scan' after a 'clear'
  99. you can find new unreferenced objects; this should help with testing
  100. specific sections of code.
  101. To test a critical section on demand with a clean kmemleak do::
  102. # echo clear > /sys/kernel/debug/kmemleak
  103. ... test your kernel or modules ...
  104. # echo scan > /sys/kernel/debug/kmemleak
  105. Then as usual to get your report with::
  106. # cat /sys/kernel/debug/kmemleak
  107. Freeing kmemleak internal objects
  108. ---------------------------------
  109. To allow access to previously found memory leaks after kmemleak has been
  110. disabled by the user or due to an fatal error, internal kmemleak objects
  111. won't be freed when kmemleak is disabled, and those objects may occupy
  112. a large part of physical memory.
  113. In this situation, you may reclaim memory with::
  114. # echo clear > /sys/kernel/debug/kmemleak
  115. Kmemleak API
  116. ------------
  117. See the include/linux/kmemleak.h header for the functions prototype.
  118. - ``kmemleak_init`` - initialize kmemleak
  119. - ``kmemleak_alloc`` - notify of a memory block allocation
  120. - ``kmemleak_alloc_percpu`` - notify of a percpu memory block allocation
  121. - ``kmemleak_vmalloc`` - notify of a vmalloc() memory allocation
  122. - ``kmemleak_free`` - notify of a memory block freeing
  123. - ``kmemleak_free_part`` - notify of a partial memory block freeing
  124. - ``kmemleak_free_percpu`` - notify of a percpu memory block freeing
  125. - ``kmemleak_update_trace`` - update object allocation stack trace
  126. - ``kmemleak_not_leak`` - mark an object as not a leak
  127. - ``kmemleak_ignore`` - do not scan or report an object as leak
  128. - ``kmemleak_scan_area`` - add scan areas inside a memory block
  129. - ``kmemleak_no_scan`` - do not scan a memory block
  130. - ``kmemleak_erase`` - erase an old value in a pointer variable
  131. - ``kmemleak_alloc_recursive`` - as kmemleak_alloc but checks the recursiveness
  132. - ``kmemleak_free_recursive`` - as kmemleak_free but checks the recursiveness
  133. The following functions take a physical address as the object pointer
  134. and only perform the corresponding action if the address has a lowmem
  135. mapping:
  136. - ``kmemleak_alloc_phys``
  137. - ``kmemleak_free_part_phys``
  138. - ``kmemleak_not_leak_phys``
  139. - ``kmemleak_ignore_phys``
  140. Dealing with false positives/negatives
  141. --------------------------------------
  142. The false negatives are real memory leaks (orphan objects) but not
  143. reported by kmemleak because values found during the memory scanning
  144. point to such objects. To reduce the number of false negatives, kmemleak
  145. provides the kmemleak_ignore, kmemleak_scan_area, kmemleak_no_scan and
  146. kmemleak_erase functions (see above). The task stacks also increase the
  147. amount of false negatives and their scanning is not enabled by default.
  148. The false positives are objects wrongly reported as being memory leaks
  149. (orphan). For objects known not to be leaks, kmemleak provides the
  150. kmemleak_not_leak function. The kmemleak_ignore could also be used if
  151. the memory block is known not to contain other pointers and it will no
  152. longer be scanned.
  153. Some of the reported leaks are only transient, especially on SMP
  154. systems, because of pointers temporarily stored in CPU registers or
  155. stacks. Kmemleak defines MSECS_MIN_AGE (defaulting to 1000) representing
  156. the minimum age of an object to be reported as a memory leak.
  157. Limitations and Drawbacks
  158. -------------------------
  159. The main drawback is the reduced performance of memory allocation and
  160. freeing. To avoid other penalties, the memory scanning is only performed
  161. when the /sys/kernel/debug/kmemleak file is read. Anyway, this tool is
  162. intended for debugging purposes where the performance might not be the
  163. most important requirement.
  164. To keep the algorithm simple, kmemleak scans for values pointing to any
  165. address inside a block's address range. This may lead to an increased
  166. number of false negatives. However, it is likely that a real memory leak
  167. will eventually become visible.
  168. Another source of false negatives is the data stored in non-pointer
  169. values. In a future version, kmemleak could only scan the pointer
  170. members in the allocated structures. This feature would solve many of
  171. the false negative cases described above.
  172. The tool can report false positives. These are cases where an allocated
  173. block doesn't need to be freed (some cases in the init_call functions),
  174. the pointer is calculated by other methods than the usual container_of
  175. macro or the pointer is stored in a location not scanned by kmemleak.
  176. Page allocations and ioremap are not tracked.
  177. Testing with kmemleak-test
  178. --------------------------
  179. To check if you have all set up to use kmemleak, you can use the kmemleak-test
  180. module, a module that deliberately leaks memory. Set CONFIG_DEBUG_KMEMLEAK_TEST
  181. as module (it can't be used as built-in) and boot the kernel with kmemleak
  182. enabled. Load the module and perform a scan with::
  183. # modprobe kmemleak-test
  184. # echo scan > /sys/kernel/debug/kmemleak
  185. Note that the you may not get results instantly or on the first scanning. When
  186. kmemleak gets results, it'll log ``kmemleak: <count of leaks> new suspected
  187. memory leaks``. Then read the file to see then::
  188. # cat /sys/kernel/debug/kmemleak
  189. unreferenced object 0xffff89862ca702e8 (size 32):
  190. comm "modprobe", pid 2088, jiffies 4294680594 (age 375.486s)
  191. hex dump (first 32 bytes):
  192. 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b kkkkkkkkkkkkkkkk
  193. 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b a5 kkkkkkkkkkkkkkk.
  194. backtrace:
  195. [<00000000e0a73ec7>] 0xffffffffc01d2036
  196. [<000000000c5d2a46>] do_one_initcall+0x41/0x1df
  197. [<0000000046db7e0a>] do_init_module+0x55/0x200
  198. [<00000000542b9814>] load_module+0x203c/0x2480
  199. [<00000000c2850256>] __do_sys_finit_module+0xba/0xe0
  200. [<000000006564e7ef>] do_syscall_64+0x43/0x110
  201. [<000000007c873fa6>] entry_SYSCALL_64_after_hwframe+0x44/0xa9
  202. ...
  203. Removing the module with ``rmmod kmemleak_test`` should also trigger some
  204. kmemleak results.