On Thu, Nov 07, 2019 at 02:52:34PM -0600, Eric Sandeen wrote: > When a filesystem is unmounted, we currently call fsnotify_sb_delete() > before evict_inodes(), which means that fsnotify_unmount_inodes() > must iterate over all inodes on the superblock, even though it will > only act on inodes with a refcount. This is inefficient and can lead > to livelocks as it iterates over many unrefcounted inodes. > > However, since fsnotify_sb_delete() and evict_inodes() are working > on orthogonal sets of inodes (fsnotify_sb_delete() only cares about > nonzero refcount, and evict_inodes() only cares about zero refcount), > we can swap the order of the calls. The fsnotify call will then have > a much smaller list to walk (any refcounted inodes). > > This should speed things up overall, and avoid livelocks in > fsnotify_unmount_inodes(). Umm... The critical part you've omitted here is that at this stage any final iput() done by fsnotify_sb_delete() (or anybody else, really) will forcibly evict the sucker there and then. So it's not as if any inodes were *added* to the evictable set by fsnotify_sb_delete() to be picked by evict_inodes() - any candidate is immediately disposed of. The crucial point is that SB_ACTIVE is already cleared by that stage - without that the patch would've been badly broken. That aside, both patches look sane. Could you update the commit message and resend the second one?
