1Kernel Memory Leak Detector 2=========================== 3 4Introduction 5------------ 6 7Kmemleak provides a way of detecting possible kernel memory leaks in a 8way similar to a tracing garbage collector 9(http://en.wikipedia.org/wiki/Garbage_collection_%28computer_science%29#Tracing_garbage_collectors), 10with the difference that the orphan objects are not freed but only 11reported via /sys/kernel/debug/kmemleak. A similar method is used by the 12Valgrind tool (memcheck --leak-check) to detect the memory leaks in 13user-space applications. 14 15Please check DEBUG_KMEMLEAK dependencies in lib/Kconfig.debug for supported 16architectures. 17 18Usage 19----- 20 21CONFIG_DEBUG_KMEMLEAK in "Kernel hacking" has to be enabled. A kernel 22thread scans the memory every 10 minutes (by default) and prints the 23number of new unreferenced objects found. To display the details of all 24the possible memory leaks: 25 26 # mount -t debugfs nodev /sys/kernel/debug/ 27 # cat /sys/kernel/debug/kmemleak 28 29To trigger an intermediate memory scan: 30 31 # echo scan > /sys/kernel/debug/kmemleak 32 33To clear the list of all current possible memory leaks: 34 35 # echo clear > /sys/kernel/debug/kmemleak 36 37New leaks will then come up upon reading /sys/kernel/debug/kmemleak 38again. 39 40Note that the orphan objects are listed in the order they were allocated 41and one object at the beginning of the list may cause other subsequent 42objects to be reported as orphan. 43 44Memory scanning parameters can be modified at run-time by writing to the 45/sys/kernel/debug/kmemleak file. The following parameters are supported: 46 47 off - disable kmemleak (irreversible) 48 stack=on - enable the task stacks scanning (default) 49 stack=off - disable the tasks stacks scanning 50 scan=on - start the automatic memory scanning thread (default) 51 scan=off - stop the automatic memory scanning thread 52 scan=<secs> - set the automatic memory scanning period in seconds 53 (default 600, 0 to stop the automatic scanning) 54 scan - trigger a memory scan 55 clear - clear list of current memory leak suspects, done by 56 marking all current reported unreferenced objects grey 57 dump=<addr> - dump information about the object found at <addr> 58 59Kmemleak can also be disabled at boot-time by passing "kmemleak=off" on 60the kernel command line. 61 62Memory may be allocated or freed before kmemleak is initialised and 63these actions are stored in an early log buffer. The size of this buffer 64is configured via the CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE option. 65 66Basic Algorithm 67--------------- 68 69The memory allocations via kmalloc, vmalloc, kmem_cache_alloc and 70friends are traced and the pointers, together with additional 71information like size and stack trace, are stored in a prio search tree. 72The corresponding freeing function calls are tracked and the pointers 73removed from the kmemleak data structures. 74 75An allocated block of memory is considered orphan if no pointer to its 76start address or to any location inside the block can be found by 77scanning the memory (including saved registers). This means that there 78might be no way for the kernel to pass the address of the allocated 79block to a freeing function and therefore the block is considered a 80memory leak. 81 82The scanning algorithm steps: 83 84 1. mark all objects as white (remaining white objects will later be 85 considered orphan) 86 2. scan the memory starting with the data section and stacks, checking 87 the values against the addresses stored in the prio search tree. If 88 a pointer to a white object is found, the object is added to the 89 gray list 90 3. scan the gray objects for matching addresses (some white objects 91 can become gray and added at the end of the gray list) until the 92 gray set is finished 93 4. the remaining white objects are considered orphan and reported via 94 /sys/kernel/debug/kmemleak 95 96Some allocated memory blocks have pointers stored in the kernel's 97internal data structures and they cannot be detected as orphans. To 98avoid this, kmemleak can also store the number of values pointing to an 99address inside the block address range that need to be found so that the 100block is not considered a leak. One example is __vmalloc(). 101 102Testing specific sections with kmemleak 103--------------------------------------- 104 105Upon initial bootup your /sys/kernel/debug/kmemleak output page may be 106quite extensive. This can also be the case if you have very buggy code 107when doing development. To work around these situations you can use the 108'clear' command to clear all reported unreferenced objects from the 109/sys/kernel/debug/kmemleak output. By issuing a 'scan' after a 'clear' 110you can find new unreferenced objects; this should help with testing 111specific sections of code. 112 113To test a critical section on demand with a clean kmemleak do: 114 115 # echo clear > /sys/kernel/debug/kmemleak 116 ... test your kernel or modules ... 117 # echo scan > /sys/kernel/debug/kmemleak 118 119Then as usual to get your report with: 120 121 # cat /sys/kernel/debug/kmemleak 122 123Kmemleak API 124------------ 125 126See the include/linux/kmemleak.h header for the functions prototype. 127 128kmemleak_init - initialize kmemleak 129kmemleak_alloc - notify of a memory block allocation 130kmemleak_alloc_percpu - notify of a percpu memory block allocation 131kmemleak_free - notify of a memory block freeing 132kmemleak_free_part - notify of a partial memory block freeing 133kmemleak_free_percpu - notify of a percpu memory block freeing 134kmemleak_not_leak - mark an object as not a leak 135kmemleak_ignore - do not scan or report an object as leak 136kmemleak_scan_area - add scan areas inside a memory block 137kmemleak_no_scan - do not scan a memory block 138kmemleak_erase - erase an old value in a pointer variable 139kmemleak_alloc_recursive - as kmemleak_alloc but checks the recursiveness 140kmemleak_free_recursive - as kmemleak_free but checks the recursiveness 141 142Dealing with false positives/negatives 143-------------------------------------- 144 145The false negatives are real memory leaks (orphan objects) but not 146reported by kmemleak because values found during the memory scanning 147point to such objects. To reduce the number of false negatives, kmemleak 148provides the kmemleak_ignore, kmemleak_scan_area, kmemleak_no_scan and 149kmemleak_erase functions (see above). The task stacks also increase the 150amount of false negatives and their scanning is not enabled by default. 151 152The false positives are objects wrongly reported as being memory leaks 153(orphan). For objects known not to be leaks, kmemleak provides the 154kmemleak_not_leak function. The kmemleak_ignore could also be used if 155the memory block is known not to contain other pointers and it will no 156longer be scanned. 157 158Some of the reported leaks are only transient, especially on SMP 159systems, because of pointers temporarily stored in CPU registers or 160stacks. Kmemleak defines MSECS_MIN_AGE (defaulting to 1000) representing 161the minimum age of an object to be reported as a memory leak. 162 163Limitations and Drawbacks 164------------------------- 165 166The main drawback is the reduced performance of memory allocation and 167freeing. To avoid other penalties, the memory scanning is only performed 168when the /sys/kernel/debug/kmemleak file is read. Anyway, this tool is 169intended for debugging purposes where the performance might not be the 170most important requirement. 171 172To keep the algorithm simple, kmemleak scans for values pointing to any 173address inside a block's address range. This may lead to an increased 174number of false negatives. However, it is likely that a real memory leak 175will eventually become visible. 176 177Another source of false negatives is the data stored in non-pointer 178values. In a future version, kmemleak could only scan the pointer 179members in the allocated structures. This feature would solve many of 180the false negative cases described above. 181 182The tool can report false positives. These are cases where an allocated 183block doesn't need to be freed (some cases in the init_call functions), 184the pointer is calculated by other methods than the usual container_of 185macro or the pointer is stored in a location not scanned by kmemleak. 186 187Page allocations and ioremap are not tracked. 188