ec: Fix mutex related coverity scan issues

This patch solves 3 issues detected by coverity scan:

    CID1241484 Data race condition
    CID1241486 Data race condition
    CID1256173 Thread deadlock
    CID1257622 Thread deadlock

With this patch, inode lock is never acquired inside a region locked
with fop->lock.

This is a backport of http://review.gluster.org/9230/ and
http://review.gluster.org/9263/

Change-Id: I35c4633efd1b68b9f72b42661fa7c728b1f52c6a
BUG: 1170954
Signed-off-by: Xavier Hernandez <xhernandez@datalab.es>
Reviewed-on: http://review.gluster.org/9244
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Dan Lambright <dlambrig@redhat.com>
Reviewed-by: Raghavendra Bhat <raghavendra@redhat.com>

2 files changed, 19 insertions, 11 deletions

diff --git a/xlators/cluster/ec/src/ec-combine.c b/xlators/cluster/ec/src/ec-combine.c
index a98380adfb8..e605f7ba567 100644
--- a/xlators/cluster/ec/src/ec-combine.c
+++ b/xlators/cluster/ec/src/ec-combine.c
@@ -851,12 +851,9 @@ void ec_combine(ec_cbk_data_t * cbk, ec_combine_f combine)
     ec_trace("ANSWER", fop, "combine=%s[%d]",
              ec_bin(str, sizeof(str), cbk->mask, 0), cbk->count);
 
-    if ((cbk->count == fop->expected) && (fop->answer == NULL))
-    {
+    if ((cbk->count == fop->expected) && (fop->answer == NULL)) {
         fop->answer = cbk;
 
-        ec_update_bad(fop, cbk->mask);
-
         resume = 1;
     }
 
@@ -865,12 +862,11 @@ void ec_combine(ec_cbk_data_t * cbk, ec_combine_f combine)
 
     UNLOCK(&fop->lock);
 
-    if (needed > 0)
-    {
+    if (needed > 0) {
         ec_dispatch_next(fop, cbk->idx);
-    }
-    else if (resume)
-    {
+    } else if (resume) {
+        ec_update_bad(fop, cbk->mask);
+
         ec_resume(fop, 0);
     }
 }
diff --git a/xlators/cluster/ec/src/ec-common.c b/xlators/cluster/ec/src/ec-common.c
index 88cc103e513..dbf2238ea8e 100644
--- a/xlators/cluster/ec/src/ec-common.c
+++ b/xlators/cluster/ec/src/ec-common.c
@@ -322,7 +322,7 @@ void ec_resume_parent(ec_fop_data_t * fop, int32_t error)
 void ec_complete(ec_fop_data_t * fop)
 {
     ec_cbk_data_t * cbk = NULL;
-    int32_t resume = 0;
+    int32_t resume = 0, update = 0;
 
     LOCK(&fop->lock);
 
@@ -336,7 +336,7 @@ void ec_complete(ec_fop_data_t * fop)
                     ((cbk->op_ret >= 0) || (cbk->op_errno != ENOTCONN))) {
                     fop->answer = cbk;
 
-                    ec_update_bad(fop, cbk->mask);
+                    update = 1;
                 }
             }
 
@@ -350,6 +350,14 @@ void ec_complete(ec_fop_data_t * fop)
 
     UNLOCK(&fop->lock);
 
+    /* ec_update_bad() locks inode->lock. This may cause deadlocks with
+       fop->lock when used in another order. Since ec_update_bad() will not
+       be called more than once for each fop, it can be called from outside
+       the fop->lock locked region. */
+    if (update) {
+        ec_update_bad(fop, cbk->mask);
+    }
+
     if (resume)
     {
         ec_resume(fop, 0);
@@ -862,9 +870,13 @@ void ec_lock(ec_fop_data_t * fop)
 
             list_add_tail(&fop->locks[fop->locked].wait_list, &lock->waiting);
 
+            LOCK(&fop->lock);
+
             fop->jobs++;
             fop->refs++;
 
+            UNLOCK(&fop->lock);
+
             UNLOCK(&lock->loc.inode->lock);
 
             break;