Merge branch 'sverker/erts/fix-aborted-pending-connection-race/OTP-15296' into maint-21

* sverker/erts/fix-aborted-pending-connection-race/OTP-15296:
  erts: Fix "Prevent inconsistent node lists" fix

3 files changed, 57 insertions, 51 deletions

diff --git a/erts/emulator/beam/dist.c b/erts/emulator/beam/dist.c
index 16c4d689a5..1e822d5c7b 100644
--- a/erts/emulator/beam/dist.c
+++ b/erts/emulator/beam/dist.c
@@ -3563,20 +3563,18 @@ BIF_RETTYPE erts_internal_new_connection_1(BIF_ALIST_1)
     BIF_RET(TUPLE2(hp, make_small(conn_id), dhandle));
 }
 
-static Sint abort_connection(DistEntry* dep, Uint32 conn_id)
+Sint erts_abort_connection_rwunlock(DistEntry* dep)
 {
-    erts_de_rwlock(dep);
+    Sint reds = 0;
+    ERTS_LC_ASSERT(erts_lc_is_de_rwlocked(dep));
 
-    if (dep->connection_id != conn_id)
-        ;
-    else if (dep->state == ERTS_DE_STATE_CONNECTED) {
+    if (dep->state == ERTS_DE_STATE_CONNECTED) {
         kill_connection(dep);
     }
     else if (dep->state == ERTS_DE_STATE_PENDING) {
         ErtsAtomCache *cache;
         ErtsDistOutputBuf *obuf;
         ErtsProcList *resume_procs;
-        Sint reds = 0;
         ErtsMonLnkDist *mld;
 
 	ASSERT(is_nil(dep->cid));
@@ -3621,17 +3619,18 @@ static Sint abort_connection(DistEntry* dep, Uint32 conn_id)
         erts_de_rwlock(dep);
         ASSERT(dep->state == ERTS_DE_STATE_EXITING);
         dep->state = ERTS_DE_STATE_IDLE;
-        erts_de_rwunlock(dep);
-        return reds;
     }
     erts_de_rwunlock(dep);
-    return 0;
+    return reds;
 }
 
-Sint
-erts_abort_connection(DistEntry *dep, Uint32 conn_id)
+static Sint abort_connection(DistEntry *dep, Uint32 conn_id)
 {
-    return abort_connection(dep, conn_id);
+    erts_de_rwlock(dep);
+    if (dep->connection_id == conn_id)
+        return erts_abort_connection_rwunlock(dep);
+    erts_de_rwunlock(dep);
+    return 0;
 }
 
 BIF_RETTYPE erts_internal_abort_connection_2(BIF_ALIST_2)
diff --git a/erts/emulator/beam/dist.h b/erts/emulator/beam/dist.h
index 30b4b35c20..75cb865390 100644
--- a/erts/emulator/beam/dist.h
+++ b/erts/emulator/beam/dist.h
@@ -399,7 +399,7 @@ extern void erts_kill_dist_connection(DistEntry *dep, Uint32);
 
 extern Uint erts_dist_cache_size(void);
 
-extern Sint erts_abort_connection(DistEntry *dep, Uint32 conn_id);
+extern Sint erts_abort_connection_rwunlock(DistEntry *dep);
 
 
 #endif
diff --git a/erts/emulator/beam/erl_node_tables.c b/erts/emulator/beam/erl_node_tables.c
index 9b34af1480..f4a36d124a 100644
--- a/erts/emulator/beam/erl_node_tables.c
+++ b/erts/emulator/beam/erl_node_tables.c
@@ -60,6 +60,10 @@ static int references_atoms_need_init = 1;
 static ErtsMonotonicTime orig_node_tab_delete_delay;
 static ErtsMonotonicTime node_tab_delete_delay;
 
+
+static void report_gc_active_dist_entry(Eterm sysname, enum dist_entry_state);
+
+
 /* -- The distribution table ---------------------------------------------- */
 
 #define ErtsBin2DistEntry(B) \
@@ -412,44 +416,6 @@ static void schedule_delete_dist_entry(DistEntry* dep)
 static void
 start_timer_delete_dist_entry(void *vdep)
 {
-    DistEntry *dep = vdep;
-    Eterm sysname;
-    enum dist_entry_state state;
-    Uint32 connection_id;
-
-    erts_de_rlock(dep);
-    state = dep->state;
-    connection_id = dep->connection_id;
-    sysname = dep->sysname;
-    erts_de_runlock(dep);
-
-    if (state != ERTS_DE_STATE_IDLE) {
-        char *state_str;
-        erts_dsprintf_buf_t *dsbuf = erts_create_logger_dsbuf();
-        switch (state) {
-        case ERTS_DE_STATE_CONNECTED:
-            state_str = "connected";
-            break;
-        case ERTS_DE_STATE_PENDING:
-            state_str = "pending connect";
-            break;
-        case ERTS_DE_STATE_EXITING:
-            state_str = "exiting";
-            break;
-        case ERTS_DE_STATE_IDLE:
-            state_str = "idle";
-            break;
-        default:
-            state_str = "unknown";
-            break;
-        }
-        erts_dsprintf(dsbuf, "Garbage collecting distribution "
-                      "entry for node %T in state: %s",
-                      sysname, state_str);
-        erts_send_error_to_logger_nogl(dsbuf);
-        erts_abort_connection(dep, connection_id);
-    }
-    
     if (node_tab_delete_delay == 0) {
         prepare_try_delete_dist_entry(vdep);
     }
@@ -489,6 +455,19 @@ static void try_delete_dist_entry(DistEntry* dep)
 {
     erts_aint_t refc;
 
+    erts_de_rwlock(dep);
+    if (dep->state != ERTS_DE_STATE_IDLE && de_refc_read(dep,0) == 0) {
+        Eterm sysname = dep->sysname;
+        enum dist_entry_state state  = dep->state;
+
+        if (dep->state != ERTS_DE_STATE_PENDING)
+            ERTS_INTERNAL_ERROR("Garbage collecting connected distribution entry");
+        erts_abort_connection_rwunlock(dep);
+        report_gc_active_dist_entry(sysname, state);
+    }
+    else
+        erts_de_rwunlock(dep);
+
     erts_rwmtx_rwlock(&erts_dist_table_rwmtx);
     /*
      * Another thread might have looked up this dist entry after
@@ -515,6 +494,34 @@ static void try_delete_dist_entry(DistEntry* dep)
     }
 }
 
+static void report_gc_active_dist_entry(Eterm sysname,
+                                        enum dist_entry_state state)
+{
+    char *state_str;
+    erts_dsprintf_buf_t *dsbuf = erts_create_logger_dsbuf();
+    switch (state) {
+    case ERTS_DE_STATE_CONNECTED:
+        state_str = "connected";
+        break;
+    case ERTS_DE_STATE_PENDING:
+        state_str = "pending connect";
+        break;
+    case ERTS_DE_STATE_EXITING:
+        state_str = "exiting";
+        break;
+    case ERTS_DE_STATE_IDLE:
+        state_str = "idle";
+        break;
+    default:
+        state_str = "unknown";
+        break;
+    }
+    erts_dsprintf(dsbuf, "Garbage collecting distribution "
+                  "entry for node %T in state: %s",
+                  sysname, state_str);
+    erts_send_error_to_logger_nogl(dsbuf);
+}
+
 int erts_dist_entry_destructor(Binary *bin)
 {
     DistEntry *dep = ErtsBin2DistEntry(bin);