Jan Luca Naumann: > I could reproduce the bug using a 4.15 kernel. Could you please take a > look into this?
Thanx for the interesting bug report. Although I could not reproduce the bug, I'm afraid it is an aufs problem. I will try more and investigate it, but it may take some time. I am interested in why you set '1' to the aufs module parameter "debug". If you had not set, this bug would not appear I guess. Did you see something wrong without setting "debug"? And you tried debugging? If so, I want to know the original problem too. Just for the case you are interested in the details: The line fs/aufs/finfo.c:113 is AuRwDestroy(&finfo->fi_rwsem) which is an aufs internal debugging macro. It checks the lock-status of the semaphore by looking up the LOCKDEP info. If the dying semaphore is still locked, the macro considers it is an unknown bug and produces the messsage (which you got). I've searched the bug where fi_rwsem is still locked, but I could not find. So *currently* I am guessing there is some unmatching between the LOCKDEP info and the real lock-status. As you might know, LOCKDEP is a kernel debugging feature and AuRwDestroy() macro is enabled only when CONFIG_AUFS_DEBUG is enabled. So I guess if you disabled both of them, you wouldn't meet the problem. J. R. Okajima