On Tue, Jun 20, 2017 at 03:24:49PM -0400, J. Bruce Fields wrote: > On Sun, Jun 18, 2017 at 12:34:38AM -0700, Christoph Hellwig wrote: > > On Sat, Jun 17, 2017 at 01:23:24PM -0400, Chuck Lever wrote: > > > > > > > On Jun 17, 2017, at 11:07 AM, Christoph Hellwig <hch@xxxxxxxxxxxxx> wrote: > > > > > > > > On Fri, Jun 16, 2017 at 11:22:54AM -0400, Chuck Lever wrote: > > > >> Running a multi-threaded 8KB fio test (70/30 mix), three or four out > > > >> of twelve of the jobs fail when using krb5i. The failure is an EIO > > > >> on a read. > > > > > > > > Just curious: what is the backing fs that you tested with? I'd be > > > > curious if you see this on XFS for example. > > > > > > I was able to reproduce this with a tmpfs share and > > > with an XFS share on NVMe LUNs. > > > > Interesting. XFS actually locks out all writes while doing a buffered > > read, so we end up with two different read instance for the hash vs > > the data, which does indeed sound dangerous. > > In the bad case Chuck was seeing, we were going through > splice_direct_to_actor(), is that a buffered read for the purposes of > the above statement? But, even if xfs is holding a lock across that splice operation, nfsd is still left holding references to the pages, and they'll be checksumed and transmitted some time later, so I don't see how xfs has much control over the problem. --b. -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
