afr: more quorum checks in lookup and new entry marking

Problem: See github issue for details.

Fix:
-In lookup if the entry exists in 2 out of 3 bricks, don't fail the
lookup with ENOENT just because there is an entrylk on the parent.
Consider quorum before deciding.

-If entry FOP does not succeed on quorum no. of bricks, do not perform
new entry mark.

Fixes: #1303
Change-Id: I56df8c89ad53b29fa450c7930a7b7ccec9f4a6c5
Signed-off-by: Ravishankar N <ravishankar@redhat.com>

3 files changed, 25 insertions, 11 deletions

diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index e6a76791e96..8dbdb572abd 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -45,7 +45,7 @@ afr_quorum_errno(afr_private_t *priv)
     return ENOTCONN;
 }
 
-static void
+void
 afr_fill_success_replies(afr_local_t *local, afr_private_t *priv,
                          unsigned char *replies)
 {
@@ -2877,6 +2877,7 @@ afr_lookup_done(call_frame_t *frame, xlator_t *this)
         0,
     };
     gf_boolean_t locked_entry = _gf_false;
+    gf_boolean_t in_flight_create = _gf_false;
     gf_boolean_t can_interpret = _gf_true;
     inode_t *parent = NULL;
     ia_type_t ia_type = IA_INVAL;
@@ -2920,17 +2921,12 @@ afr_lookup_done(call_frame_t *frame, xlator_t *this)
         if (!replies[i].valid)
             continue;
 
-        if (locked_entry && replies[i].op_ret == -1 &&
-            replies[i].op_errno == ENOENT) {
-            /* Second, check entry is still
-               "underway" in creation */
-            local->op_ret = -1;
-            local->op_errno = ENOENT;
-            goto error;
-        }
-
-        if (replies[i].op_ret == -1)
+        if (replies[i].op_ret == -1) {
+            if (locked_entry && replies[i].op_errno == ENOENT) {
+                in_flight_create = _gf_true;
+            }
             continue;
+        }
 
         if (read_subvol == -1 || !readable[read_subvol]) {
             read_subvol = i;
@@ -2940,6 +2936,12 @@ afr_lookup_done(call_frame_t *frame, xlator_t *this)
         }
     }
 
+    if (in_flight_create && !afr_has_quorum(success_replies, this, NULL)) {
+        local->op_ret = -1;
+        local->op_errno = ENOENT;
+        goto error;
+    }
+
     if (read_subvol == -1)
         goto error;
     /* We now have a read_subvol, which is readable[] (if there
diff --git a/xlators/cluster/afr/src/afr-dir-write.c b/xlators/cluster/afr/src/afr-dir-write.c
index f366df7ec76..b7cceb79158 100644
--- a/xlators/cluster/afr/src/afr-dir-write.c
+++ b/xlators/cluster/afr/src/afr-dir-write.c
@@ -345,6 +345,7 @@ afr_mark_entry_pending_changelog(call_frame_t *frame, xlator_t *this)
     afr_private_t *priv = NULL;
     int pre_op_count = 0;
     int failed_count = 0;
+    unsigned char *success_replies = NULL;
 
     local = frame->local;
     priv = this->private;
@@ -360,9 +361,16 @@ afr_mark_entry_pending_changelog(call_frame_t *frame, xlator_t *this)
     failed_count = AFR_COUNT(local->transaction.failed_subvols,
                              priv->child_count);
 
+    /* FOP succeeded on all bricks. */
     if (pre_op_count == priv->child_count && !failed_count)
         return;
 
+    /* FOP did not suceed on quorum no. of bricks. */
+    success_replies = alloca0(priv->child_count);
+    afr_fill_success_replies(local, priv, success_replies);
+    if (!afr_has_quorum(success_replies, this, NULL))
+        return;
+
     if (priv->thin_arbiter_count) {
         /*Mark new entry using ta file*/
         local->is_new_entry = _gf_true;
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 322bb691e44..1fff5640940 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -1405,4 +1405,8 @@ afr_is_lock_mode_mandatory(dict_t *xdata);
 
 void
 afr_dom_lock_release(call_frame_t *frame);
+
+void
+afr_fill_success_replies(afr_local_t *local, afr_private_t *priv,
+                         unsigned char *replies);
 #endif /* __AFR_H__ */