259 lines · plain
1Kernel Memory Leak Detector2===========================3 4Kmemleak provides a way of detecting possible kernel memory leaks in a5way similar to a `tracing garbage collector6<https://en.wikipedia.org/wiki/Tracing_garbage_collection>`_,7with the difference that the orphan objects are not freed but only8reported via /sys/kernel/debug/kmemleak. A similar method is used by the9Valgrind tool (``memcheck --leak-check``) to detect the memory leaks in10user-space applications.11 12Usage13-----14 15CONFIG_DEBUG_KMEMLEAK in "Kernel hacking" has to be enabled. A kernel16thread scans the memory every 10 minutes (by default) and prints the17number of new unreferenced objects found. If the ``debugfs`` isn't already18mounted, mount with::19 20 # mount -t debugfs nodev /sys/kernel/debug/21 22To display the details of all the possible scanned memory leaks::23 24 # cat /sys/kernel/debug/kmemleak25 26To trigger an intermediate memory scan::27 28 # echo scan > /sys/kernel/debug/kmemleak29 30To clear the list of all current possible memory leaks::31 32 # echo clear > /sys/kernel/debug/kmemleak33 34New leaks will then come up upon reading ``/sys/kernel/debug/kmemleak``35again.36 37Note that the orphan objects are listed in the order they were allocated38and one object at the beginning of the list may cause other subsequent39objects to be reported as orphan.40 41Memory scanning parameters can be modified at run-time by writing to the42``/sys/kernel/debug/kmemleak`` file. The following parameters are supported:43 44- off45 disable kmemleak (irreversible)46- stack=on47 enable the task stacks scanning (default)48- stack=off49 disable the tasks stacks scanning50- scan=on51 start the automatic memory scanning thread (default)52- scan=off53 stop the automatic memory scanning thread54- scan=<secs>55 set the automatic memory scanning period in seconds56 (default 600, 0 to stop the automatic scanning)57- scan58 trigger a memory scan59- clear60 clear list of current memory leak suspects, done by61 marking all current reported unreferenced objects grey,62 or free all kmemleak objects if kmemleak has been disabled.63- dump=<addr>64 dump information about the object found at <addr>65 66Kmemleak can also be disabled at boot-time by passing ``kmemleak=off`` on67the kernel command line.68 69Memory may be allocated or freed before kmemleak is initialised and70these actions are stored in an early log buffer. The size of this buffer71is configured via the CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE option.72 73If CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF are enabled, the kmemleak is74disabled by default. Passing ``kmemleak=on`` on the kernel command75line enables the function. 76 77If you are getting errors like "Error while writing to stdout" or "write_loop:78Invalid argument", make sure kmemleak is properly enabled.79 80Basic Algorithm81---------------82 83The memory allocations via :c:func:`kmalloc`, :c:func:`vmalloc`,84:c:func:`kmem_cache_alloc` and85friends are traced and the pointers, together with additional86information like size and stack trace, are stored in a rbtree.87The corresponding freeing function calls are tracked and the pointers88removed from the kmemleak data structures.89 90An allocated block of memory is considered orphan if no pointer to its91start address or to any location inside the block can be found by92scanning the memory (including saved registers). This means that there93might be no way for the kernel to pass the address of the allocated94block to a freeing function and therefore the block is considered a95memory leak.96 97The scanning algorithm steps:98 99 1. mark all objects as white (remaining white objects will later be100 considered orphan)101 2. scan the memory starting with the data section and stacks, checking102 the values against the addresses stored in the rbtree. If103 a pointer to a white object is found, the object is added to the104 gray list105 3. scan the gray objects for matching addresses (some white objects106 can become gray and added at the end of the gray list) until the107 gray set is finished108 4. the remaining white objects are considered orphan and reported via109 /sys/kernel/debug/kmemleak110 111Some allocated memory blocks have pointers stored in the kernel's112internal data structures and they cannot be detected as orphans. To113avoid this, kmemleak can also store the number of values pointing to an114address inside the block address range that need to be found so that the115block is not considered a leak. One example is __vmalloc().116 117Testing specific sections with kmemleak118---------------------------------------119 120Upon initial bootup your /sys/kernel/debug/kmemleak output page may be121quite extensive. This can also be the case if you have very buggy code122when doing development. To work around these situations you can use the123'clear' command to clear all reported unreferenced objects from the124/sys/kernel/debug/kmemleak output. By issuing a 'scan' after a 'clear'125you can find new unreferenced objects; this should help with testing126specific sections of code.127 128To test a critical section on demand with a clean kmemleak do::129 130 # echo clear > /sys/kernel/debug/kmemleak131 ... test your kernel or modules ...132 # echo scan > /sys/kernel/debug/kmemleak133 134Then as usual to get your report with::135 136 # cat /sys/kernel/debug/kmemleak137 138Freeing kmemleak internal objects139---------------------------------140 141To allow access to previously found memory leaks after kmemleak has been142disabled by the user or due to an fatal error, internal kmemleak objects143won't be freed when kmemleak is disabled, and those objects may occupy144a large part of physical memory.145 146In this situation, you may reclaim memory with::147 148 # echo clear > /sys/kernel/debug/kmemleak149 150Kmemleak API151------------152 153See the include/linux/kmemleak.h header for the functions prototype.154 155- ``kmemleak_init`` - initialize kmemleak156- ``kmemleak_alloc`` - notify of a memory block allocation157- ``kmemleak_alloc_percpu`` - notify of a percpu memory block allocation158- ``kmemleak_vmalloc`` - notify of a vmalloc() memory allocation159- ``kmemleak_free`` - notify of a memory block freeing160- ``kmemleak_free_part`` - notify of a partial memory block freeing161- ``kmemleak_free_percpu`` - notify of a percpu memory block freeing162- ``kmemleak_update_trace`` - update object allocation stack trace163- ``kmemleak_not_leak`` - mark an object as not a leak164- ``kmemleak_ignore`` - do not scan or report an object as leak165- ``kmemleak_scan_area`` - add scan areas inside a memory block166- ``kmemleak_no_scan`` - do not scan a memory block167- ``kmemleak_erase`` - erase an old value in a pointer variable168- ``kmemleak_alloc_recursive`` - as kmemleak_alloc but checks the recursiveness169- ``kmemleak_free_recursive`` - as kmemleak_free but checks the recursiveness170 171The following functions take a physical address as the object pointer172and only perform the corresponding action if the address has a lowmem173mapping:174 175- ``kmemleak_alloc_phys``176- ``kmemleak_free_part_phys``177- ``kmemleak_ignore_phys``178 179Dealing with false positives/negatives180--------------------------------------181 182The false negatives are real memory leaks (orphan objects) but not183reported by kmemleak because values found during the memory scanning184point to such objects. To reduce the number of false negatives, kmemleak185provides the kmemleak_ignore, kmemleak_scan_area, kmemleak_no_scan and186kmemleak_erase functions (see above). The task stacks also increase the187amount of false negatives and their scanning is not enabled by default.188 189The false positives are objects wrongly reported as being memory leaks190(orphan). For objects known not to be leaks, kmemleak provides the191kmemleak_not_leak function. The kmemleak_ignore could also be used if192the memory block is known not to contain other pointers and it will no193longer be scanned.194 195Some of the reported leaks are only transient, especially on SMP196systems, because of pointers temporarily stored in CPU registers or197stacks. Kmemleak defines MSECS_MIN_AGE (defaulting to 1000) representing198the minimum age of an object to be reported as a memory leak.199 200Limitations and Drawbacks201-------------------------202 203The main drawback is the reduced performance of memory allocation and204freeing. To avoid other penalties, the memory scanning is only performed205when the /sys/kernel/debug/kmemleak file is read. Anyway, this tool is206intended for debugging purposes where the performance might not be the207most important requirement.208 209To keep the algorithm simple, kmemleak scans for values pointing to any210address inside a block's address range. This may lead to an increased211number of false negatives. However, it is likely that a real memory leak212will eventually become visible.213 214Another source of false negatives is the data stored in non-pointer215values. In a future version, kmemleak could only scan the pointer216members in the allocated structures. This feature would solve many of217the false negative cases described above.218 219The tool can report false positives. These are cases where an allocated220block doesn't need to be freed (some cases in the init_call functions),221the pointer is calculated by other methods than the usual container_of222macro or the pointer is stored in a location not scanned by kmemleak.223 224Page allocations and ioremap are not tracked.225 226Testing with kmemleak-test227--------------------------228 229To check if you have all set up to use kmemleak, you can use the kmemleak-test230module, a module that deliberately leaks memory. Set CONFIG_SAMPLE_KMEMLEAK231as module (it can't be used as built-in) and boot the kernel with kmemleak232enabled. Load the module and perform a scan with::233 234 # modprobe kmemleak-test235 # echo scan > /sys/kernel/debug/kmemleak236 237Note that the you may not get results instantly or on the first scanning. When238kmemleak gets results, it'll log ``kmemleak: <count of leaks> new suspected239memory leaks``. Then read the file to see then::240 241 # cat /sys/kernel/debug/kmemleak242 unreferenced object 0xffff89862ca702e8 (size 32):243 comm "modprobe", pid 2088, jiffies 4294680594 (age 375.486s)244 hex dump (first 32 bytes):245 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b kkkkkkkkkkkkkkkk246 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b a5 kkkkkkkkkkkkkkk.247 backtrace:248 [<00000000e0a73ec7>] 0xffffffffc01d2036249 [<000000000c5d2a46>] do_one_initcall+0x41/0x1df250 [<0000000046db7e0a>] do_init_module+0x55/0x200251 [<00000000542b9814>] load_module+0x203c/0x2480252 [<00000000c2850256>] __do_sys_finit_module+0xba/0xe0253 [<000000006564e7ef>] do_syscall_64+0x43/0x110254 [<000000007c873fa6>] entry_SYSCALL_64_after_hwframe+0x44/0xa9255 ...256 257Removing the module with ``rmmod kmemleak_test`` should also trigger some258kmemleak results.259