On Wed, 2019-04-03 at 20:35 +0100, Al Viro wrote: > On Wed, Apr 03, 2019 at 11:55:01AM +1100, NeilBrown wrote: > > > It seems very likely that this was just a subtlety that I missed. > > I doesn't help that "ino" isn't actually and inode and isn't freed like > > an inode, but that is no excuse. > > > > When we add the rcu_head linkage to 'struct autofs_info', we might as > > well remove the 'struct inode' from there - it doesn't seem to have been > > used for years. > > Umm... I can do that, but then we get greater potential for conflicts > and the whole thing might be better off in autofs tree. Ian, up to you - > I can throw both into a never-rebased branch, so you could merge it > into your tree; the first one is -stable fodder, though... Ha, as unlikely as that usually is, given the low churn in the autofs code, conflicts might be a problem. I'm likely to change the autofs info flags field to unsigned as part of implementing a suggestion from Neil (quite some time ago now) to fix the AT_NO_AUTOMOUNT handling. So I'll hold onto the patch to remove the inode field, include it as part of that change, and forward it as low priority via Andrew. I'll send your rcu free patch back to you (with an added comment in the description) as a higher priority change. Ian
