On 16 Jan 2020, at 10:38, Trond Myklebust wrote: > On Thu, 2020-01-16 at 10:13 -0500, Benjamin Coddington wrote: >> On 16 Jan 2020, at 9:35, Trond Myklebust wrote: > > nfs_file_open() is completely the wrong place to perform a lookup. Its > purpose in the VFS is to allow the filesystem to set up state *after* > we've already looked up the dentry, revalidated it and therefore > decided which file to open. > The NFSv4.0 behaviour of performing a new lookup is actually the > aberration here, and is due to the fact that it does not have an open- > by-filehandle operation, so we have no alternative. Ok, that makes good sense to me. > As I said, if you want stronger semantics, there are lookupcache mount > options that allow you to tune things. I therefore see no valid reason > to change the existing behaviour, which also matches that of older NFS > versions (i.e. v3 and v2). Thanks for the discussion. Your point above about v4.0 CLAIM_NULL actually doing another lookup after we already did a lookup throws things in sharp relief. Ben
