nfs41: wait for LAYOUTRETURN before retrying LAYOUTGET

Also take care to stop waiting if someone clears retry bit.

Signed-off-by: Peng Tao <tao.peng@primarydata.com>

3 files changed, 45 insertions, 3 deletions

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 7e1a97a54f9..44c600aac90 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -7796,7 +7796,9 @@ static void nfs4_layoutreturn_release(void *calldata)
 	spin_lock(&lo->plh_inode->i_lock);
 	if (lrp->res.lrs_present)
 		pnfs_set_layout_stateid(lo, &lrp->res.stateid, true);
-	clear_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
+	clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
+	smp_mb__after_atomic();
+	wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
 	clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, &lo->plh_flags);
 	rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
 	lo->plh_block_lgets--;
diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 893f6b5afe6..c4c9fe606ae 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -1398,6 +1398,26 @@ static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
 	return ret;
 }
 
+/* stop waiting if someone clears NFS_LAYOUT_RETRY_LAYOUTGET bit. */
+static int pnfs_layoutget_retry_bit_wait(struct wait_bit_key *key)
+{
+	if (!test_bit(NFS_LAYOUT_RETRY_LAYOUTGET, key->flags))
+		return 1;
+	return nfs_wait_bit_killable(key);
+}
+
+static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
+{
+	/*
+	 * send layoutcommit as it can hold up layoutreturn due to lseg
+	 * reference
+	 */
+	pnfs_layoutcommit_inode(lo->plh_inode, false);
+	return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
+				   pnfs_layoutget_retry_bit_wait,
+				   TASK_UNINTERRUPTIBLE);
+}
+
 /*
  * Layout segment is retreived from the server if not cached.
  * The appropriate layout segment is referenced and returned to the caller.
@@ -1444,7 +1464,8 @@ lookup_again:
 	}
 
 	/* if LAYOUTGET already failed once we don't try again */
-	if (pnfs_layout_io_test_failed(lo, iomode))
+	if (pnfs_layout_io_test_failed(lo, iomode) &&
+	    !pnfs_should_retry_layoutget(lo))
 		goto out_unlock;
 
 	first = list_empty(&lo->plh_segs);
@@ -1469,6 +1490,22 @@ lookup_again:
 			goto out_unlock;
 	}
 
+	/*
+	 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
+	 * for LAYOUTRETURN even if first is true.
+	 */
+	if (!lseg && pnfs_should_retry_layoutget(lo) &&
+	    test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
+		spin_unlock(&ino->i_lock);
+		dprintk("%s wait for layoutreturn\n", __func__);
+		if (pnfs_prepare_to_retry_layoutget(lo)) {
+			pnfs_put_layout_hdr(lo);
+			dprintk("%s retrying\n", __func__);
+			goto lookup_again;
+		}
+		goto out_put_layout_hdr;
+	}
+
 	if (pnfs_layoutgets_blocked(lo, &arg, 0))
 		goto out_unlock;
 	atomic_inc(&lo->plh_outstanding);
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index fed6ae067ac..49a46670840 100644
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -359,8 +359,11 @@ static inline void pnfs_set_retry_layoutget(struct pnfs_layout_hdr *lo)
 
 static inline void pnfs_clear_retry_layoutget(struct pnfs_layout_hdr *lo)
 {
-	if (test_and_clear_bit(NFS_LAYOUT_RETRY_LAYOUTGET, &lo->plh_flags))
+	if (test_and_clear_bit(NFS_LAYOUT_RETRY_LAYOUTGET, &lo->plh_flags)) {
 		atomic_dec(&lo->plh_refcount);
+		/* wake up waiters for LAYOUTRETURN as that is not needed */
+		wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
+	}
 }
 
 static inline bool pnfs_should_retry_layoutget(struct pnfs_layout_hdr *lo)