NFSv4: Ensure that we recover from the OPEN + OPEN_CONFIRM BAD_STATEID race

 If the server is in the unconfirmed OPEN state for a given open owner
 and receives a second OPEN for the same open owner, it will cancel the
 state of the first request and set up an OPEN_CONFIRM for the second.

 This can cause a race that is discussed in rfc3530 on page 181.

 The following patch allows the client to recover by retrying the
 original open request.

 Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

1 files changed, 10 insertions, 0 deletions

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index a8be9af610a..ff378126ccd 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -785,6 +785,16 @@ static struct nfs4_state *nfs4_do_open(struct inode *dir, struct dentry *dentry,
 			exception.retry = 1;
 			continue;
 		}
+		/*
+		 * BAD_STATEID on OPEN means that the server cancelled our
+		 * state before it received the OPEN_CONFIRM.
+		 * Recover by retrying the request as per the discussion
+		 * on Page 181 of RFC3530.
+		 */
+		if (status == -NFS4ERR_BAD_STATEID) {
+			exception.retry = 1;
+			continue;
+		}
 		res = ERR_PTR(nfs4_handle_exception(NFS_SERVER(dir),
 					status, &exception));
 	} while (exception.retry);