NFS: Don't reset pg_moreio in __nfs_pageio_add_request
authorTrond Myklebust <trond.myklebust@primarydata.com>
Sun, 13 Jul 2014 19:13:19 +0000 (15:13 -0400)
committerTrond Myklebust <trond.myklebust@primarydata.com>
Sun, 13 Jul 2014 19:18:44 +0000 (15:18 -0400)
Once we've started sending unstable NFS writes, we do not want to
clear pg_moreio, or we may end up sending the very last request as
a stable write if the commit lists are still empty.

Do, however, reset pg_moreio in the case where we end up having to
recoalesce the write if an attempt to use pNFS failed.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
fs/nfs/pagelist.c

index 0aefc8102b6b40ddc61713d2a524377317105124..17fab89f635898ca2e82fae1a8e84bfee05b1a07 100644 (file)
@@ -935,7 +935,6 @@ static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
                        nfs_pageio_doio(desc);
                        if (desc->pg_error < 0)
                                return 0;
-                       desc->pg_moreio = 0;
                        if (desc->pg_recoalesce)
                                return 0;
                        /* retry add_request for this subreq */
@@ -982,6 +981,7 @@ static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
                desc->pg_count = 0;
                desc->pg_base = 0;
                desc->pg_recoalesce = 0;
+               desc->pg_moreio = 0;
 
                while (!list_empty(&head)) {
                        struct nfs_page *req;