nfs: Fixes "Remote I/O error" mount failures

- Fixes issue where NFS mount fail with "Remove I/O error" after the
  target directory has been deleted and re-created after the gNFSd has
  already cached the inode of the first generation of the target
  directory.
- The solution is to follow the guidance of the AFR2 comments and
  refresh the inode by deleting it from cache and looking it up
  again.

BUG: 1258069
Change-Id: I9c7d8bd460ee9e5ea0b5b47d23886b1afcdcd563
Reported-by: Richard Wareing <rwareing@fb.com>
Signed-off-by: Niels de Vos <ndevos@redhat.com>
Reviewed-on: http://review.gluster.org/12045
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Raghavendra Bhat <raghavendra@redhat.com>

2 files changed, 48 insertions, 1 deletions

diff --git a/tests/bugs/bug-1258069.t b/tests/bugs/bug-1258069.t
new file mode 100755
index 00000000000..8df4a8a9e1b
--- /dev/null
+++ b/tests/bugs/bug-1258069.t
@@ -0,0 +1,28 @@
+#!/bin/bash
+
+. $(dirname $0)/../include.rc
+. $(dirname $0)/../volume.rc
+. $(dirname $0)/../nfs.rc
+
+cleanup
+
+TEST glusterd
+TEST pidof glusterd
+TEST $CLI volume create $V0 replica 3 $H0:$B0/${V0}{0,1,2}
+TEST $CLI volume set $V0 cluster.choose-local off
+TEST $CLI volume set $V0 nfs.disable off
+TEST $CLI volume set $V0 diagnostics.client-log-level DEBUG
+TEST $CLI volume start $V0
+
+EXPECT_WITHIN $NFS_EXPORT_TIMEOUT "1" is_nfs_export_available
+TEST mount_nfs $H0:/$V0 $N0 nolock
+TEST mkdir -p $N0/a/b/c
+TEST umount_nfs $N0
+TEST glusterfs --volfile-id=/$V0 --volfile-server=$H0 $M0 --attribute-timeout=0 --entry-timeout=0
+rmdir $M0/a/b/c
+mkdir $M0/a/b/c
+TEST mount_nfs $H0:/$V0/a/b/c $N0 nolock
+TEST umount_nfs $N0
+TEST umount $M0
+
+cleanup
diff --git a/xlators/nfs/server/src/mount3.c b/xlators/nfs/server/src/mount3.c
index 43a77f888cd..58d068c54ba 100644
--- a/xlators/nfs/server/src/mount3.c
+++ b/xlators/nfs/server/src/mount3.c
@@ -896,6 +896,21 @@ err:
         return ret;
 }
 
+int __mnt3_resolve_subdir (mnt3_resolve_t *mres);
+
+/*
+ * Per the AFR2 comments, this function performs the "fresh" lookup
+ * by deleting the inode from cache and calling __mnt3_resolve_subdir
+ * again.
+ */
+int __mnt3_fresh_lookup (mnt3_resolve_t *mres) {
+        inode_unlink (mres->resolveloc.inode,
+                mres->resolveloc.parent, mres->resolveloc.name);
+        strncpy (mres->remainingdir, mres->resolveloc.path,
+                strlen(mres->resolveloc.path));
+        nfs_loc_wipe (&mres->resolveloc);
+        return __mnt3_resolve_subdir (mres);
+}
 
 int32_t
 mnt3_resolve_subdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
@@ -915,7 +930,11 @@ mnt3_resolve_subdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         mres = frame->local;
         mntxl = (xlator_t *)cookie;
-        if (op_ret == -1) {
+        if (op_ret == -1 && op_errno == ESTALE) {
+                /* Nuke inode from cache and try the LOOKUP
+                 * request again. */
+                return __mnt3_fresh_lookup (mres);
+        } else if (op_ret == -1) {
                 gf_log (GF_NFS, GF_LOG_ERROR, "path=%s (%s)",
                         mres->resolveloc.path, strerror (op_errno));
                 mntstat = mnt3svc_errno_to_mnterr (op_errno);