On 03/01/17(Tue) 14:08, Ted Unangst wrote: > Martin Pieuchot wrote: > > It seems that most of the problems exposed by the introduction of the > > NET_LOCK() are related to the non-recursive nature of the rwlock. Some > > known issues involve pflow(4), cloning interfaces an NFS. > > > > Diff below makes use of a recursive-rwlock instead. I just finished a > > build on NFS with it, so I'd like to know if it works for your use case. > > I don't want to interfere with progress, so if you think this is best, carry > on. > > But I'll note that using recursive locks leads to sloppy locking discipline, > which is exactly how we find ourselves in this mess today. Need the lock? Not > sure? Grab it anyway. > > The rrwlock was added for the particular use case of vfs locking, which is > already a recursive cluster fuck, but wasn't supposed to be used for new code. > Of course, the network stack also qualifies as legacy code, so can't object > much to using it there either. > > I imagine you've already thought of this. :)
I did :) To be honest I'm not sure if we should take this direction. For the moment I'm trying to learn as much as possible about the problems we're facing. If it appears to be just sloppy recursive code then we can decide what to do with it and when. I am also looking for more feedbacks and/or inputs so I appreciate your email on the matter.
