On Mon, 2004-05-03 at 10:17, Lever, Charles wrote:

does anyone have plans to address this problem?

invalidate_inode_pages2() should already address the problem of clearing
out locked pages, and indeed that is what we currently use for the NFS
client in 2.6.x.

As for the case of returning EIO on read failure, perhaps that could be
made conditional on PageError() actually being set? Currently that
appears to be the logic used in the case of writes (see
wait_on_page_writeback_range()), but for some reason, not for reads.

Cheers,
  Trond