OS-369 zone hung in shutting down state

3 files changed, 18 insertions, 6 deletions

diff --git a/usr/src/uts/common/os/zone.c b/usr/src/uts/common/os/zone.c
index e818862a65..fff321cf32 100644
--- a/usr/src/uts/common/os/zone.c
+++ b/usr/src/uts/common/os/zone.c
@@ -4848,6 +4848,7 @@ zone_boot(zoneid_t zoneid)
 static int
 zone_empty(zone_t *zone)
 {
+	int cnt = 0;
 	int waitstatus;
 
 	/*
@@ -4858,7 +4859,16 @@ zone_empty(zone_t *zone)
 	ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
 	while ((waitstatus = zone_status_timedwait_sig(zone,
 	    ddi_get_lbolt() + hz, ZONE_IS_EMPTY)) == -1) {
-		killall(zone->zone_id);
+		boolean_t force = B_FALSE;
+
+		/* Every 30 seconds, try harder */
+		if (cnt++ >= 30) {
+			cmn_err(CE_WARN, "attempt to force kill zone %d\n",
+			    zone->zone_id);
+			force = B_TRUE;
+			cnt = 0;
+		}
+		killall(zone->zone_id, force);
 	}
 	/*
 	 * return EINTR if we were signaled
@@ -6743,7 +6753,7 @@ zone_kadmin(int cmd, int fcn, const char *mdep, cred_t *credp)
 	 * zone_ki_call_zoneadmd() will do a more thorough job of this
 	 * later.
 	 */
-	killall(zone->zone_id);
+	killall(zone->zone_id, B_FALSE);
 	/*
 	 * Now, create the thread to contact zoneadmd and do the rest of the
 	 * work.  This thread can't be created in our zone otherwise
diff --git a/usr/src/uts/common/sys/uadmin.h b/usr/src/uts/common/sys/uadmin.h
index c35d0a5cfb..6adeb477bb 100644
--- a/usr/src/uts/common/sys/uadmin.h
+++ b/usr/src/uts/common/sys/uadmin.h
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2011 Joyent, Inc.  All rights reserved.
  */
 
 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
@@ -157,7 +158,7 @@ extern kmutex_t ualock;
 extern void mdboot(int, int, char *, boolean_t);
 extern void mdpreboot(int, int, char *);
 extern int kadmin(int, int, void *, cred_t *);
-extern void killall(zoneid_t);
+extern void killall(zoneid_t, boolean_t);
 #endif
 
 #if defined(__STDC__)
diff --git a/usr/src/uts/common/syscall/uadmin.c b/usr/src/uts/common/syscall/uadmin.c
index 1bdfbbfd0b..dbff1b637c 100644
--- a/usr/src/uts/common/syscall/uadmin.c
+++ b/usr/src/uts/common/syscall/uadmin.c
@@ -22,6 +22,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2011 Joyent, Inc.  All rights reserved.
  */
 
 #include <sys/param.h>
@@ -76,7 +77,7 @@ volatile int fastreboot_dryrun = 0;
  * system with many zones.
  */
 void
-killall(zoneid_t zoneid)
+killall(zoneid_t zoneid, boolean_t force)
 {
 	proc_t *p;
 
@@ -106,7 +107,7 @@ killall(zoneid_t zoneid)
 		    p->p_stat != SIDL &&
 		    p->p_stat != SZOMB) {
 			mutex_enter(&p->p_lock);
-			if (sigismember(&p->p_sig, SIGKILL)) {
+			if (!force && sigismember(&p->p_sig, SIGKILL)) {
 				mutex_exit(&p->p_lock);
 				p = p->p_next;
 			} else {
@@ -243,7 +244,7 @@ kadmin(int cmd, int fcn, void *mdep, cred_t *credp)
 		 */
 		zone_shutdown_global();
 
-		killall(ALL_ZONES);
+		killall(ALL_ZONES, B_FALSE);
 		/*
 		 * If we are calling kadmin() from a kernel context then we
 		 * do not release these resources.