Hi folks, These are a couple of shutdown related unmount hang fixes that check-parallel has been tripping over semi-regularly. The first fix is relatively straight forward, but the second is marked as RFC because I've only just got to the bottom of that one and it's quite a bit more complex to solve. The second fix will probably have to be split into 3 smaller patches (tracing needed to find the issue, code rearrangement to make the fix cleaner, and the fix) but before I do that I wanted to get some more eyes on it and feedback whilst I do more testing to confirm that it works as intended and hasn't introduced any regressions. The fundamental problem that the second patch addresses is that stale inode buffers need the last reference to the BLI to complete the stale inodes attached to the buffer when it is released. There's a corner case where this last reference can be the transaction commit context, and that doesn't perform the necessary cleanup. I've kept all the triage notes, analysis and thinking I did to come up with the fix in the commit message for the second patch - hopefully that will help people understand what is actually going on a lot faster than I did.... -Dave.
