cluster/distribute: If cached_subvol is down, return ENOTCONN in lookup

When we follow a linkfile, and the lookup returns a ENOTCONN error, return
the error, as the cached subvol is down, and lookup_everywhere wont succeed,
but actually ends up clearing the linkfile, and clearing the namespace.

Change-Id: I772bf71531bc646e8fb62d3e8549a5fe0f3896da
BUG: 893378
Signed-off-by: shishir gowda <sgowda@redhat.com>
Reviewed-on: http://review.gluster.org/4383
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>

2 files changed, 81 insertions, 1 deletions

diff --git a/tests/bugs/bug-893378.t b/tests/bugs/bug-893378.t
new file mode 100755
index 000000000..470589f53
--- /dev/null
+++ b/tests/bugs/bug-893378.t
@@ -0,0 +1,71 @@
+#!/bin/bash
+
+. $(dirname $0)/../include.rc
+cleanup;
+BRICK_COUNT=3
+
+function file_has_linkfile()
+{
+        i=0
+        j=0
+        while [ $i -lt $BRICK_COUNT ]
+        do
+                stat=`stat $B0/${V0}$i/$1 2>/dev/null`
+                if [ $? -eq 0 ]
+                then
+                        let j++
+                        let "BRICK${j}=$i"
+
+                fi
+                let i++
+        done
+        return $j
+}
+
+function get_cached_brick()
+{
+        i=1
+        while [ $i -lt 3 ]
+        do
+                test=`getfattr -n trusted.glusterfs.dht.linkto -e text $B0/${V0}$BRICK$i 2>&1`
+                if [ $? -eq 1 ]
+                then
+                        cached=$BRICK"$i"
+                        i=$(( $i+3 ))
+                fi
+                let i++
+        done
+
+        return $cached
+}
+
+TEST glusterd
+TEST pidof glusterd
+
+TEST $CLI volume create $V0 $H0:$B0/${V0}0 $H0:$B0/${V0}1 $H0:$B0/${V0}2
+TEST $CLI volume start $V0
+
+## Mount FUSE
+TEST glusterfs --attribute-timeout=0 --entry-timeout=0 -s $H0 --volfile-id $V0 $M0;
+
+## create a linkfile on subvolume 0
+TEST touch $M0/1
+TEST mv $M0/1 $M0/2
+
+file_has_linkfile 2
+has_link=$?
+if [ $has_link -eq 2 ]
+then
+        get_cached_brick
+        CACHED=$?
+        # Kill a brick process
+        kill -9 `cat /var/lib/glusterd/vols/$V0/run/$H0-d-backends-${V0}$CACHED.pid`;
+fi
+
+## trigger a lookup
+ls -l $M0/2 2>/dev/null
+
+## fail dd if file exists.
+
+dd if=/dev/zero of=$M0/2 bs=1 count=1 conv=excl 2>/dev/null
+EXPECT "1" echo $?
diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index ded86bcc3..540c2ed87 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -1095,7 +1095,16 @@ dht_lookup_linkfile_cbk (call_frame_t *frame, void *cookie,
                 gf_log (this->name, GF_LOG_INFO,
                         "lookup of %s on %s (following linkfile) failed (%s)",
                         local->loc.path, subvol->name, strerror (op_errno));
-                goto err;
+
+                /* If cached subvol returned ENOTCONN, do not do
+                lookup_everywhere. We need to make sure linkfile does not get
+                removed, which can take away the namespace, and subvol is
+                anyways down. */
+
+                if (op_errno != ENOTCONN)
+                        goto err;
+                else
+                        goto unwind;
         }
 
         if (check_is_dir (inode, stbuf, xattr)) {