[12/19] NFS/flexfiles: Speed up read failover when DSes are down
diff mbox series

Message ID 20190301192455.104943-13-trond.myklebust@hammerspace.com
State New
Headers show
Series
  • Updates for pNFS/flexfiles
Related show

Commit Message

Trond Myklebust March 1, 2019, 7:24 p.m. UTC
If we notice that a DS may be down, we should attempt to read from the
other mirrors first before we go back to retry the dead DS.

Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
---
 fs/nfs/flexfilelayout/flexfilelayout.c | 74 +++++++++++++++++++++-----
 fs/nfs/pnfs.h                          |  1 +
 fs/nfs/pnfs_dev.c                      | 10 ++++
 3 files changed, 73 insertions(+), 12 deletions(-)

Patch
diff mbox series

diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c
index c4a9f7f2f0e4..ba5ed07eac92 100644
--- a/fs/nfs/flexfilelayout/flexfilelayout.c
+++ b/fs/nfs/flexfilelayout/flexfilelayout.c
@@ -788,10 +788,28 @@  ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
 	}
 }
 
+static void
+ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, int idx)
+{
+	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
+
+	if (devid)
+		nfs4_mark_deviceid_unavailable(devid);
+}
+
+static void
+ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, int idx)
+{
+	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
+
+	if (devid)
+		nfs4_mark_deviceid_available(devid);
+}
+
 static struct nfs4_pnfs_ds *
-ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
-				  int start_idx,
-				  int *best_idx)
+ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
+			     int start_idx, int *best_idx,
+			     bool check_device)
 {
 	struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 	struct nfs4_ff_layout_mirror *mirror;
@@ -799,25 +817,53 @@  ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
 	bool fail_return = false;
 	int idx;
 
-	/* mirrors are sorted by efficiency */
+	/* mirrors are initially sorted by efficiency */
 	for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
 		if (idx+1 == fls->mirror_array_cnt)
-			fail_return = true;
+			fail_return = !check_device;
 
 		mirror = FF_LAYOUT_COMP(lseg, idx);
-		if (ff_layout_test_devid_unavailable(&mirror->mirror_ds->id_node))
+		ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
+		if (!ds)
 			continue;
 
-		ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
-		if (ds) {
-			*best_idx = idx;
-			return ds;
-		}
+		if (check_device &&
+		    nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
+			continue;
+
+		*best_idx = idx;
+		return ds;
 	}
 
 	return NULL;
 }
 
+static struct nfs4_pnfs_ds *
+ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
+				 int start_idx, int *best_idx)
+{
+	return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
+}
+
+static struct nfs4_pnfs_ds *
+ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
+				   int start_idx, int *best_idx)
+{
+	return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
+}
+
+static struct nfs4_pnfs_ds *
+ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
+				  int start_idx, int *best_idx)
+{
+	struct nfs4_pnfs_ds *ds;
+
+	ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
+	if (ds)
+		return ds;
+	return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
+}
+
 static void
 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
 		      struct nfs_page *req,
@@ -1167,8 +1213,10 @@  static int ff_layout_async_handle_error(struct rpc_task *task,
 {
 	int vers = clp->cl_nfs_mod->rpc_vers->number;
 
-	if (task->tk_status >= 0)
+	if (task->tk_status >= 0) {
+		ff_layout_mark_ds_reachable(lseg, idx);
 		return 0;
+	}
 
 	/* Handle the case of an invalid layout segment */
 	if (!pnfs_is_valid_lseg(lseg))
@@ -1231,6 +1279,8 @@  static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
 	err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
 				       mirror, offset, length, status, opnum,
 				       GFP_NOIO);
+	if (status == NFS4ERR_NXIO)
+		ff_layout_mark_ds_unreachable(lseg, idx);
 	pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
 	dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
 }
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index 56659ccce1d8..c0420b979d88 100644
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -350,6 +350,7 @@  void nfs4_delete_deviceid(const struct pnfs_layoutdriver_type *, const struct nf
 void nfs4_init_deviceid_node(struct nfs4_deviceid_node *, struct nfs_server *,
 			     const struct nfs4_deviceid *);
 bool nfs4_put_deviceid_node(struct nfs4_deviceid_node *);
+void nfs4_mark_deviceid_available(struct nfs4_deviceid_node *node);
 void nfs4_mark_deviceid_unavailable(struct nfs4_deviceid_node *node);
 bool nfs4_test_deviceid_unavailable(struct nfs4_deviceid_node *node);
 void nfs4_deviceid_purge_client(const struct nfs_client *);
diff --git a/fs/nfs/pnfs_dev.c b/fs/nfs/pnfs_dev.c
index ec6b607ff1d7..537b80d693f1 100644
--- a/fs/nfs/pnfs_dev.c
+++ b/fs/nfs/pnfs_dev.c
@@ -283,6 +283,16 @@  nfs4_put_deviceid_node(struct nfs4_deviceid_node *d)
 }
 EXPORT_SYMBOL_GPL(nfs4_put_deviceid_node);
 
+void
+nfs4_mark_deviceid_available(struct nfs4_deviceid_node *node)
+{
+	if (test_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags)) {
+		clear_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags);
+		smp_mb__after_atomic();
+	}
+}
+EXPORT_SYMBOL_GPL(nfs4_mark_deviceid_available);
+
 void
 nfs4_mark_deviceid_unavailable(struct nfs4_deviceid_node *node)
 {