shm_sub BUGFIX clear left over event and orig_cid

orig_cid should always be cleared when the originator stops waiting for
the subscribers. If not, subsequent events cannot be generated on the
subscription SHM, until the previous originator process is restarted.

Normally, the wait logic takes care of this. However, in case a
subscription evpipe could not be notified we bail on waiting and return
with error. However, in this case orig_cid and event are left behind
never to be cleared.

src/shm_sub.c

@@ -1353,6 +1353,11 @@ sr_shmsub_change_notify_update(struct sr_mod_info_s *mod_info, const char *orig_
     goto cleanup;
 
 cleanup_wrunlock:
+    /* clear the event unless a subscriber reported an error, in which case further clean up will happen */
+    if (sub_shm->event != SR_SUB_EV_ERROR) {
+        ATOMIC_STORE_RELAXED(sub_shm->event, SR_SUB_EV_NONE);
+        sub_shm->orig_cid = 0;
+    }
     /* SUB WRITE UNLOCK */
     sr_rwunlock(&sub_shm->lock, 0, SR_LOCK_WRITE, cid, __func__);
 
@@ -1637,6 +1642,12 @@ sr_shmsub_change_notify_change(struct sr_mod_info_s *mod_info, const char *orig_
 cleanup:
     for (i = 0; i < notify_count; ++i) {
         if (notify_subs[i].lock) {
+            /* clear the event unless a subscriber reported an error and an abort will be sent */
+            if (notify_subs[i].sub_shm->event != SR_SUB_EV_ERROR) {
+                ATOMIC_STORE_RELAXED(notify_subs[i].sub_shm->event, SR_SUB_EV_NONE);
+                notify_subs[i].sub_shm->orig_cid = 0;
+            }
+
             /* SUB UNLOCK */
             sr_rwunlock(&notify_subs[i].sub_shm->lock, 0, notify_subs[i].lock, cid, __func__);
             notify_subs[i].lock = SR_LOCK_NONE;
@@ -1798,6 +1809,10 @@ sr_shmsub_change_notify_change_done(struct sr_mod_info_s *mod_info, const char *
 cleanup:
     for (i = 0; i < notify_count; ++i) {
         if (notify_subs[i].lock) {
+            /* always clear the event */
+            ATOMIC_STORE_RELAXED(notify_subs[i].sub_shm->event, SR_SUB_EV_NONE);
+            notify_subs[i].sub_shm->orig_cid = 0;
+
             /* SUB UNLOCK */
             sr_rwunlock(&notify_subs[i].sub_shm->lock, 0, notify_subs[i].lock, cid, __func__);
             notify_subs[i].lock = SR_LOCK_NONE;
@@ -2021,6 +2036,10 @@ sr_shmsub_change_notify_change_abort(struct sr_mod_info_s *mod_info, const char
 cleanup:
     for (i = 0; i < notify_count; ++i) {
         if (notify_subs[i].lock) {
+            /* always clear the event */
+            ATOMIC_STORE_RELAXED(notify_subs[i].sub_shm->event, SR_SUB_EV_NONE);
+            notify_subs[i].sub_shm->orig_cid = 0;
+
             /* SUB UNLOCK */
             sr_rwunlock(&notify_subs[i].sub_shm->lock, 0, notify_subs[i].lock, cid, __func__);
             notify_subs[i].lock = SR_LOCK_NONE;
@@ -2188,6 +2207,10 @@ sr_shmsub_oper_get_notify(struct sr_mod_info_mod_s *mod, const char *xpath, cons
 cleanup:
     for (i = 0; i < notify_count; ++i) {
         if (notify_subs[i].lock) {
+            /* clear any event left behind due to an error (eg. could not notify evpipe) */
+            ATOMIC_STORE_RELAXED(notify_subs[i].sub_shm->event, SR_SUB_EV_NONE);
+            notify_subs[i].sub_shm->orig_cid = 0;
+
             /* SUB UNLOCK */
             sr_rwunlock(&notify_subs[i].sub_shm->lock, 0, notify_subs[i].lock, cid, __func__);
             notify_subs[i].lock = SR_LOCK_NONE;
@@ -2701,6 +2724,12 @@ sr_shmsub_rpc_notify(sr_conn_ctx_t *conn, sr_rwlock_t *sub_lock, off_t *subs, ui
     } while (subscriber_count);
 
 cleanup_wrunlock:
+    /* clear the event unless a subscriber reported an error and an abort will be sent */
+    if (sub_shm->event != SR_SUB_EV_ERROR) {
+        sub_shm->event = SR_SUB_EV_NONE;
+        sub_shm->orig_cid = 0;
+    }
+
     /* SUB WRITE UNLOCK */
     sr_rwunlock(&sub_shm->lock, 0, SR_LOCK_WRITE, conn->cid, __func__);
 
@@ -2830,6 +2859,10 @@ sr_shmsub_rpc_notify_abort(sr_conn_ctx_t *conn, sr_rwlock_t *sub_lock, off_t *su
     return err_info;
 
 cleanup_wrunlock:
+    /* always clear the event */
+    ATOMIC_STORE_RELAXED(sub_shm->event, SR_SUB_EV_NONE);
+    sub_shm->orig_cid = 0;
+
     /* SUB WRITE UNLOCK */
     sr_rwunlock(&sub_shm->lock, 0, SR_LOCK_WRITE, conn->cid, __func__);
 
@@ -2967,6 +3000,12 @@ sr_shmsub_notif_notify(sr_conn_ctx_t *conn, const struct lyd_node *notif, struct
     goto cleanup;
 
 cleanup_ext_sub_unlock:
+    /* always clear the event if an error occurred */
+    if (err_info) {
+        ATOMIC_STORE_RELAXED(sub_shm->event, SR_SUB_EV_NONE);
+        sub_shm->orig_cid = 0;
+    }
+
     /* SUB WRITE UNLOCK */
     sr_rwunlock(&sub_shm->lock, 0, SR_LOCK_WRITE, conn->cid, __func__);