On Wed, 2010-12-01 at 14:38 -0800, Andrew Morton wrote: > On Wed, 1 Dec 2010 14:24:42 -0800 > Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote: > > > On Wed, Dec 1, 2010 at 2:13 PM, Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote: > > > On Wed, 01 Dec 2010 16:51:12 -0500 > > > Trond Myklebust <Trond.Myklebust@xxxxxxxxxx> wrote: > > > > > >> On Wed, 2010-12-01 at 13:38 -0800, Andrew Morton wrote: > > >> > Probably on most call paths we'll be OK - if a process is in the middle > > >> > of a file truncate, holdin a file* ref which holds an inode ref then > > >> > nobody will be unmounting that fs and hence nobody will be unloading > > >> > that module. > > >> > > > >> > However on the random_code->alloc_page->vmscan->releasepage path, none > > >> > of that applies. > > >> > > >> Just out of interest, what ensures that the mapping is still around for > > >> the 'spin_unlock_irq(&mapping->tree_lock);' in __remove_mapping()? > > > > > > Nothing, afacit. > > > > No, we're good. > > > > Module unload has to go through a "stop_machine()" cycle, and that in > > turn requires an idle period for everything. And just a preemption > > reschedule isn't enough for that. > > > > So what is sufficient is that > > > > - we had the page locked and on the mapping > > > > This implies that we had an inode reference to the module, and the > > page lock means that the inode reference cannot go away (because it > > will involve invalidate-pages etc) > > > > - we're not sleeping after __remove_mapping, so unload can't happen afterwards. > > > > A _lot_ of the module races depend on that latter thing. We have > > almost no cases that are strictly about actual reference counts etc. > > > > OK, the stop_machine() plugs a lot of potential race-vs-module-unload > things. But Trond is referring to races against vmscan inode reclaim, > unmount, etc. > Yes, that was my question. However, Linus' explanation appears to answer one of Hugh's objections: we can apparently use a preempt-disable() in order to ensure that the module containing the mapping->a_ops is not unloaded until after the ->freepage() call is complete. That would imply that ->freepage() cannot sleep, but I don't think that is too nasty a restriction. -- Trond Myklebust Linux NFS client maintainer NetApp Trond.Myklebust@xxxxxxxxxx www.netapp.com -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html