Re: [NFS] [PATCH] mmap corruption

[Trond Myklebust <trond.myklebust@fys.uio.no>]

>>>>> " " == Steve Dickson <SteveD@RedHat.com> writes:

     > OK, I understand your point.  And Yes, ndirty and ncommit
     > always seem to be zero when nfs_wb_all() returns. Only when
     > npages != 0 is when I get the corruption.

     > I didn't realize that npages != 0 meant there are only pending
     > reads *not* pending writes... Thanks for that clarification....

My mistake. npages counts only writes...

However, I still stand by my statement that nfs_wb_all() is supposed
to ensure that *all* pending writes have been cleared.
The only explanation for npages != 0 is if

  a) an error occurred with nfs_wb_all() (we should perhaps test the
     return value of nfs_wb_all() there). Under normal circumstances,
     an error should only occur if you're using soft mounts, though.

  b) somebody redirtied the page *after* nfs_wb_all() was done.

Cheers,
  Trond