Blob Blame History Raw
From 2c13d8bd00f1e8ff30e9fc81f183f6450303ac30 Mon Sep 17 00:00:00 2001
From: Alexey Tikhonov <atikhono@redhat.com>
Date: Wed, 11 Dec 2019 18:42:49 +0100
Subject: [PATCH] util/watchdog: fixed watchdog implementation

In case watchdog detected locked process and this process was parent
process it just sent SIGTERM to the whole group of processes, including
itself.
This handling was wrong: generic `server_setup()` installs custom
libtevent handler for SIGTERM signal so this signal is only processed
in the context of tevent mainloop. But if tevent mainloop is stuck
(exactly the case that triggers WD) then event is not processed
and this made watchdog useless.
`watchdog_handler()` and `watchdog_detect_timeshift()` were amended to do
unconditional `_exit()` after optionally sending a signal to the group.

Resolves: https://pagure.io/SSSD/sssd/issue/4089

Reviewed-by: Sumit Bose <sbose@redhat.com>
---
 src/util/util_watchdog.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/src/util/util_watchdog.c b/src/util/util_watchdog.c
index a07275b19..38c248271 100644
--- a/src/util/util_watchdog.c
+++ b/src/util/util_watchdog.c
@@ -54,9 +54,8 @@ static void watchdog_detect_timeshift(void)
         if (write(watchdog_ctx.pipefd[1], "1", 1) != 1) {
             if (getpid() == getpgrp()) {
                 kill(-getpgrp(), SIGTERM);
-            } else {
-                _exit(1);
             }
+            _exit(1);
         }
     }
 }
@@ -75,9 +74,8 @@ static void watchdog_handler(int sig)
     if (__sync_add_and_fetch(&watchdog_ctx.ticks, 1) > WATCHDOG_MAX_TICKS) {
         if (getpid() == getpgrp()) {
             kill(-getpgrp(), SIGTERM);
-        } else {
-            _exit(1);
         }
+        _exit(1);
     }
 }
 
-- 
2.20.1