FS: Use stable writes when not doing a bulk flush

If we're only doing a single write, and there are no other unstable
writes being queued up, we might want to just flip to using a stable
write RPC call.

Reviewed-by: NeilBrown <neilb@suse.de>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

4 files changed, 19 insertions, 3 deletions

diff --git a/fs/nfs/pagelist.c b/fs/nfs/pagelist.c
index 23e794410669..fd85618149a1 100644
--- a/fs/nfs/pagelist.c
+++ b/fs/nfs/pagelist.c
@@ -223,6 +223,7 @@ void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
         desc->pg_count = 0;
         desc->pg_bsize = bsize;
         desc->pg_base = 0;
+        desc->pg_moreio = 0;
         desc->pg_inode = inode;
         desc->pg_doio = doio;
         desc->pg_ioflags = io_flags;
@@ -335,9 +336,11 @@ int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
                            struct nfs_page *req)
 {
         while (!nfs_pageio_do_add_request(desc, req)) {
+                desc->pg_moreio = 1;
                 nfs_pageio_doio(desc);
                 if (desc->pg_error < 0)
                         return 0;
+                desc->pg_moreio = 0;
         }
         return 1;
 }
diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index 47a3ad63e0d5..4d686ee53244 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -179,8 +179,8 @@ static int wb_priority(struct writeback_control *wbc)
         if (wbc->for_reclaim)
                 return FLUSH_HIGHPRI | FLUSH_STABLE;
         if (wbc->for_kupdate || wbc->for_background)
-                return FLUSH_LOWPRI;
-        return 0;
+                return FLUSH_LOWPRI | FLUSH_COND_STABLE;
+        return FLUSH_COND_STABLE;
 }
 
 /*
@@ -863,7 +863,7 @@ static int nfs_write_rpcsetup(struct nfs_page *req,
         data->args.context = get_nfs_open_context(req->wb_context);
         data->args.lock_context = req->wb_lock_context;
         data->args.stable  = NFS_UNSTABLE;
-        if (how & FLUSH_STABLE) {
+        if (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
                 data->args.stable = NFS_DATA_SYNC;
                 if (!nfs_need_commit(NFS_I(inode)))
                         data->args.stable = NFS_FILE_SYNC;
@@ -912,6 +912,12 @@ static int nfs_flush_multi(struct nfs_pageio_descriptor *desc)
 
         nfs_list_remove_request(req);
 
+        if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
+            (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit ||
+             desc->pg_count > wsize))
+                desc->pg_ioflags &= ~FLUSH_COND_STABLE;
+
+
         nbytes = desc->pg_count;
         do {
                 size_t len = min(nbytes, wsize);
@@ -1002,6 +1008,10 @@ static int nfs_flush_one(struct nfs_pageio_descriptor *desc)
         if ((!lseg) && list_is_singular(&data->pages))
                 lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, IOMODE_RW);
 
+        if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
+            (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit))
+                desc->pg_ioflags &= ~FLUSH_COND_STABLE;
+
         /* Set up the argument struct */
         ret = nfs_write_rpcsetup(req, data, &nfs_write_full_ops, desc->pg_count, 0, lseg, desc->pg_ioflags);
 out:
diff --git a/include/linux/nfs_fs.h b/include/linux/nfs_fs.h
index f88522b10a38..cb2add401f25 100644
--- a/include/linux/nfs_fs.h
+++ b/include/linux/nfs_fs.h
@@ -33,6 +33,8 @@
 #define FLUSH_STABLE            4       /* commit to stable storage */
 #define FLUSH_LOWPRI            8       /* low priority background flush */
 #define FLUSH_HIGHPRI           16      /* high priority memory reclaim flush */
+#define FLUSH_COND_STABLE       32      /* conditional stable write - only stable
+                                         * if everything fits in one RPC */
 
 #ifdef __KERNEL__
 
diff --git a/include/linux/nfs_page.h b/include/linux/nfs_page.h
index 90907ada6d52..92d54c81f51e 100644
--- a/include/linux/nfs_page.h
+++ b/include/linux/nfs_page.h
@@ -57,6 +57,7 @@ struct nfs_pageio_descriptor {
         size_t                  pg_count;
         size_t                  pg_bsize;
         unsigned int            pg_base;
+        char                    pg_moreio;
 
         struct inode            *pg_inode;
         int                     (*pg_doio)(struct nfs_pageio_descriptor *);