On 5/14/16, 21:06, "Jeff Layton" <jlayton@xxxxxxxxxxxxxxx> wrote:
>Currently, the code will clear the fail bit if we get back a fatal
>error. I don't think that's correct -- we only want to clear that
>bit if the layoutget succeeds.
>
>Fixes: 0bcbf039f6 (nfs: handle request add failure properly)
>Signed-off-by: Jeff Layton <jeff.layton@xxxxxxxxxxxxxxx>
>---
> fs/nfs/pnfs.c | 8 +++++---
> 1 file changed, 5 insertions(+), 3 deletions(-)
>
>diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
>index e6cad5ee5d29..5f6ed295acb5 100644
>--- a/fs/nfs/pnfs.c
>+++ b/fs/nfs/pnfs.c
>@@ -876,11 +876,13 @@ send_layoutget(struct pnfs_layout_hdr *lo,
> lseg = nfs4_proc_layoutget(lgp, gfp_flags);
> } while (lseg == ERR_PTR(-EAGAIN));
>
>- if (IS_ERR(lseg) && !nfs_error_is_fatal(PTR_ERR(lseg)))
>- lseg = NULL;
>- else
>+ if (IS_ERR(lseg)) {
>+ if (!nfs_error_is_fatal(PTR_ERR(lseg)))
>+ lseg = NULL;
>+ } else {
> pnfs_layout_clear_fail_bit(lo,
> pnfs_iomode_to_fail_bit(range->iomode));
>+ }
No… The intention was indeed that we clear the fail bit in all cases except fatal errors. Otherwise, the client won’t attempt another layoutget.
��.n��������+%������w��{.n�����{��w��� jg��� �����ݢj����G���� ���j:+v���w�m������w�������h�����٥
