On 5 Feb 2019, at 5:23, Benjamin Coddington wrote: > On 4 Feb 2019, at 16:40, Trond Myklebust wrote: >> What about the case when the page is in the swap cache? Is that also >> safe? > > I think the case you'd like me to consider is when a file on NFS has its > mapping backed by pages that are in swap. In that case, it makes no sense > to check the count of pages in the swap file's mapping to determine whether > to set NFS_INO_INVALID_DATA. We actually want to optimize away the > revalidation if the NFS file lacks mapped pages. All this makes me nervous about breaking swap, and is probably not appropriate for a fix here so I am going to drop this approach. Ben
