cluster/afr: Don't start heal when lookup succeeds on < 2 children

        Backport of http://review.gluster.org/8698

Problem:
When self-heal code doesn't see at least 2 successes on looking up children,
then self-heal can't be done. What is happening now is if all the lookups fail
then the pending changelog is all zeros in xattrs so all the children are
becoming sources and leading to crashes when the code paths further assume that
some data structures are populated properly

Fix:
Don't proceed with self-heals when < 2 children succeed lookups.

BUG: 1145726
Change-Id: I65465843f0e554c8ccdd8fa930ab42ac123ec023
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/8824
Reviewed-by: Krutika Dhananjay <kdhananj@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>

1 files changed, 17 insertions, 0 deletions

diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c
index 0158948d728..b104e6b7869 100644
--- a/xlators/cluster/afr/src/afr-self-heal-common.c
+++ b/xlators/cluster/afr/src/afr-self-heal-common.c
@@ -324,6 +324,12 @@ afr_selfheal_find_direction (xlator_t *this, struct afr_reply *replies,
 	accused = alloca0 (priv->child_count);
 	matrix = ALLOC_MATRIX(priv->child_count, int);
 
+        if (afr_success_count (replies,
+                               priv->child_count) < AFR_SH_MIN_PARTICIPANTS) {
+                /* Treat this just like locks not being acquired */
+                return -ENOTCONN;
+        }
+
 	/* First construct the pending matrix for further analysis */
 	afr_selfheal_extract_xattr (this, replies, type, dirty, matrix);
 
@@ -502,6 +508,17 @@ afr_selfheal_unlocked_discover (call_frame_t *frame, inode_t *inode,
 						  priv->child_up);
 }
 
+unsigned int
+afr_success_count (struct afr_reply *replies, unsigned int count)
+{
+        int     i = 0;
+        unsigned int success = 0;
+
+        for (i = 0; i < count; i++)
+                if (replies[i].valid && replies[i].op_ret == 0)
+                        success++;
+        return success;
+}
 
 int
 afr_selfheal_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,