On Tue, 2010-11-30 at 20:10 -0800, Linus Torvalds wrote: > On Tue, Nov 30, 2010 at 7:47 PM, Trond Myklebust > <Trond.Myklebust@xxxxxxxxxx> wrote: > > Otherwise, the VM may end up removing it while we're reading from it. > > I don't think this is valid. > > Maybe it fixes a bug, but the commit description is misleading at > best. Since you have a reference count to the page, the page is not > going away. Locking may hide some other bug (due to serializing with > other code you care about), but it is _not_ about the "VM may end up > removing it". > > Even from a serialization angle, I think this patch is a bit suspect, > since readdir() will always be called under the inode semaphore, so I > think you'll always be serialized wrt other readdir users. Of course, > you may have invalidation events etc that are outside of readdir, so > ... I'm not worried about other readdir calls invalidating the page. My concern is rather about the VM memory reclaimers ejecting the page from the page cache, and calling nfs_readdir_clear_array while we're referencing the page. This wasn't a problem with the previous readdir code, but it will be with the new incarnation because the actual filenames are stored outside the page itself. As far as I can see, the only way to protect against that is to lock the page, perform the usual tests and then release the page lock when we're done... > Anyway if this patch matters, there's something else going on, and you > need to describe that. No problem. I just wanted to get the patches out so that the people who are reporting regressions can start testing. -- 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
