Possible init bug
james at digitalmatter.com.au
Tue Jan 25 02:25:52 UTC 2011
I am using busybox 1.17.1 in an ARM embedded system, and I have found a bug in _something_, though it may not be busybox itself. When I have a respawn line in my inittab, pointing to a non-existent or quickly exiting binary, I find that memory is slowly leaked - roughly 12 kB every 2 or 3 seconds. The memory doesn't go to any user processes, and when the kernel is compiled with kmemleak, I see an ever increasing number of kmemleak_object's in /proc/slabinfo (but no leaks reported by kmemleak). Eventually, the oom killer wipes the whole board out. I had a brief look at the init code in busybox, but couldn't think of any obvious cause. In fact, I'm inclined to think that it is a kernel bug. However, normal spawning of processes (from the shell for instance) is not leaking memory on my board.
So, I was hoping someone with a similar system could try adding an erroneous / quickly exiting process to their inittab, to see if the problem is reproducible. Any ideas as to what the extra kobjects might be, or how to find out, would also be appreciated.
More information about the busybox