On Tue, Nov 10, 2020 at 09:57:39AM -0600, Chris Friesen wrote: > No, there are quite a few of them. I've included them below. I agree, it's > not clear who's holding the lock. Is there a way to find that out? > > Just to be sure, I'm looking for whoever has the BH_Lock bit set on the > buffer_head "b_state" field, right? I don't see any ownership field the way > we have for mutexes. Is there some way to find out who would have locked > the buffer? It's quite possible that the buffer was locked as part of doing I/O, and we are just waiting for the I/O to complete. An example of this is in journal_submit_commit_record(), where we lock the buffer using lock_buffer(), and then call submit_bh() to submit the buffer for I/O. When the I/O is completed, the buffer head will be unlocked, and we can check the buffer_uptodate flag to see if the I/O completed successfully. (See journal_wait_on_commit_record() for an example of this.) So the first thing I'd suggest doing is looking at the console output or dmesg output from the crashdump to see if there are any clues in terms of kernel messages from the device driver before things locked up. This could be as simple as the device falling off the bus, in which case there might be some kernel error messages from the block layer or device driver that would give some insight. Good luck, - Ted
