From 2263fb23bdbbf313535edf54440fe20627b57d7f Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Pavel=20B=C5=99ezina?= Date: Mon, 22 Aug 2016 13:15:04 +0200 Subject: [PATCH 08/39] watchdog: cope with time shift MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit When a time is changed into the past during sssd runtime (e.g. on boot during time correction), it is possible that we never hit watchdog tevent timer since it is based on system time. This patch adds a past-time shift detection mechanism. If a time shift is detected we restart watchdog. Resolves: https://fedorahosted.org/sssd/ticket/3154 Reviewed-by: Lukáš Slebodník Reviewed-by: Fabiano Fidêncio (cherry picked from commit b8ceaeb80cffb00c26390913ea959b77f7e848b9) --- src/util/util_watchdog.c | 41 +++++++++++++++++++++++++++++++++++++++++ 1 file changed, 41 insertions(+) diff --git a/src/util/util_watchdog.c b/src/util/util_watchdog.c index 5032fddba1b94b3fc7e560162c392dfa57d699cf..1c27d73f13b3042ecb549a2184e1368e8339d199 100644 --- a/src/util/util_watchdog.c +++ b/src/util/util_watchdog.c @@ -29,8 +29,39 @@ struct watchdog_ctx { struct timeval interval; struct tevent_timer *te; volatile int ticks; + + /* To detect time shift. */ + struct tevent_context *ev; + int input_interval; + time_t timestamp; } watchdog_ctx; +static bool watchdog_detect_timeshift(void) +{ + time_t prev_time; + time_t cur_time; + errno_t ret; + + prev_time = watchdog_ctx.timestamp; + cur_time = watchdog_ctx.timestamp = time(NULL); + if (cur_time < prev_time) { + /* Time shift detected. We need to restart watchdog. */ + DEBUG(SSSDBG_IMPORTANT_INFO, "Time shift detected, " + "restarting watchdog!\n"); + teardown_watchdog(); + ret = setup_watchdog(watchdog_ctx.ev, watchdog_ctx.input_interval); + if (ret != EOK) { + DEBUG(SSSDBG_FATAL_FAILURE, "Unable to restart watchdog " + "[%d]: %s\n", ret, sss_strerror(ret)); + orderly_shutdown(1); + } + + return true; + } + + return false; +} + /* the watchdog is purposefully *not* handled by the tevent * signal handler as it is meant to check if the daemon is * still processing the event queue itself. A stuck process @@ -38,6 +69,12 @@ struct watchdog_ctx { * signals either */ static void watchdog_handler(int sig) { + /* Do not count ticks if time shift was detected + * since watchdog was restarted. */ + if (watchdog_detect_timeshift()) { + return; + } + /* if 3 ticks passed by kills itself */ if (__sync_add_and_fetch(&watchdog_ctx.ticks, 1) > 3) { @@ -101,6 +138,10 @@ int setup_watchdog(struct tevent_context *ev, int interval) watchdog_ctx.interval.tv_sec = interval; watchdog_ctx.interval.tv_usec = 0; + watchdog_ctx.ev = ev; + watchdog_ctx.input_interval = interval; + watchdog_ctx.timestamp = time(NULL); + /* Start the timer */ /* we give 1 second head start to the watchdog event */ its.it_value.tv_sec = interval + 1; -- 2.9.3