On Fri, Nov 15, 2019 at 10:10:37PM +0000, Al Viro wrote: > On Fri, Nov 15, 2019 at 04:26:09PM -0500, Steven Rostedt wrote: > > On Fri, 15 Nov 2019 21:18:20 +0000 > > Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote: > > > > > OK... debugfs and tracefs definitely convert to that; so do, AFAICS, > > > spufs and selinuxfs, and I wouldn't be surprised if it could be > > > used in a few more places... securityfs, almost certainly qibfs, > > > gadgetfs looks like it could make use of that. Maybe subrpc > > > as well, but I'll need to look in details. configfs won't, > > > unfortunately... > > > > Thanks Al for looking into this. > > > > I'll try to test it in tracefs, and see if anything breaks. But > > probably wont get to it till next week. > > I'll probably throw that into #next.dcache - if nothing else, > that cuts down on the size of patch converting d_subdirs/d_child > from list to hlist... > > Need to get some sleep first, though - only 5 hours today, so > I want to take another look at that thing tomorrow morning - > I don't trust my ability to spot obvious bugs right now... ;-/ > > Oh, well - that at least might finally push the old "kernel-side > rm -rf done right" pile of half-baked patches into more useful > state, probably superseding most of them. Thanks for doing this. Sorry for the delay in getting back to this, was on a long-haul flight... Anyway, this looks sane to me. debugfs "should" not be having a file added while a directory is being removed at the same time, but I really can't guarantee that someone is trying to do something crazy like that. So "heavy" locking is fine with me, this never has to be a "fast" operation, it's much more important to get it "correct". thanks, greg k-h
