On Mon, 2019-06-17 at 20:55 +0800, Yan, Zheng wrote:
> Make write(2) return error prematurely if there is writeback error.
> User can use fsync() or fdatasync() to clear the error.
>
> This change is mainly for reporting errors after blacklist + reconnect.
>
> Signed-off-by: "Yan, Zheng" <zyan@xxxxxxxxxx>
> ---
> fs/ceph/caps.c | 8 ++++++++
> 1 file changed, 8 insertions(+)
>
> diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
> index 57e1447a9d4b..f07767d3864c 100644
> --- a/fs/ceph/caps.c
> +++ b/fs/ceph/caps.c
> @@ -2814,6 +2814,14 @@ int ceph_get_caps(struct file *filp, int need, int want,
> break;
> }
>
> + if (_got & CEPH_CAP_FILE_WR) {
> + ret = filemap_check_wb_err(inode->i_mapping, filp->f_wb_err);
> + if (ret < 0) {
> + ceph_put_cap_refs(ci, _got);
> + return ret;
> + }
> + }
> +
> if ((_got & CEPH_CAP_FILE_RD) && (_got & CEPH_CAP_FILE_CACHE))
> ceph_fscache_revalidate_cookie(ci);
>
This seems wrong. Just because an earlier write failed, we don't
necessarily want to start returning errors on other writes. I'd drop
this patch.
--
Jeff Layton <jlayton@xxxxxxxxxx>
