OS-375 i_dls_mgmt_upcall()/dlmgmt_zfop() deadlock in dlmgmtd

OS-376 vnic_dev_create()/dlmgmt_table_lock()/vnic_dev_delete() deadlock
OS-378 dlmgmt_zonehalt()/dlmgmt_upcall_getattr() deadlock

4 files changed, 54 insertions, 19 deletions

diff --git a/usr/src/cmd/dlmgmtd/dlmgmt_door.c b/usr/src/cmd/dlmgmtd/dlmgmt_door.c
index e3e7fccd84..acb44870a8 100644
--- a/usr/src/cmd/dlmgmtd/dlmgmt_door.c
+++ b/usr/src/cmd/dlmgmtd/dlmgmt_door.c
@@ -59,7 +59,9 @@
 #include <libsysevent.h>
 #include <libdlmgmt.h>
 #include <librcm.h>
-#include <thread.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
 #include "dlmgmt_impl.h"
 
 typedef void dlmgmt_door_handler_t(void *, void *, size_t *, zoneid_t,
@@ -1318,14 +1320,6 @@ done:
 	retvalp->lr_err = err;
 }
 
-static void
-do_zonehalt(void *zid)
-{
-	dlmgmt_table_lock(B_TRUE);
-	dlmgmt_db_fini((zoneid_t)zid);
-	dlmgmt_table_unlock();
-}
-
 /* ARGSUSED */
 static void
 dlmgmt_zonehalt(void *argp, void *retp, size_t *sz, zoneid_t zoneid,
@@ -1350,17 +1344,22 @@ dlmgmt_zonehalt(void *argp, void *retp, size_t *sz, zoneid_t zoneid,
 			 * dls_devnet_destroy calls mac_perim_enter_by_mh
 			 * which could lead to deadlock if another process is
 			 * holding the mac perimeter then made an upcall to
-			 * dlmgmtd.  To avoid this, we run the dlmgmt_db_fini
-			 * in a thread so that we can continue to service
-			 * upcalls in parallel with cleaning up zones.
+			 * dlmgmtd.  To try to avoid this, we serialize zone
+			 * activity on the /etc/dladm/zone.lck file.
 			 */
-			err = thr_create(NULL, 0,
-			    (void *(*)(void *))do_zonehalt,
-			    (void *)(zonehalt->ld_zoneid), THR_DAEMON, NULL);
-			if (err != 0)
-				dlmgmt_log(LOG_ERR, "unable to create thread "
-				    "to clean up zone %d: %s",
-				    zonehalt->ld_zoneid, strerror(err));
+			int fd;
+
+			while ((fd = open(ZONE_LOCK,
+			    O_WRONLY | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR))
+			    < 0)
+				sleep(1);
+        		close(fd);
+
+			dlmgmt_table_lock(B_TRUE);
+			dlmgmt_db_fini(zonehalt->ld_zoneid);
+			dlmgmt_table_unlock();
+
+			(void) unlink(ZONE_LOCK);
 		}
 	}
 	retvalp->lr_err = err;
diff --git a/usr/src/cmd/dlmgmtd/dlmgmt_impl.h b/usr/src/cmd/dlmgmtd/dlmgmt_impl.h
index cdfd0d8a4d..780a18ce5a 100644
--- a/usr/src/cmd/dlmgmtd/dlmgmt_impl.h
+++ b/usr/src/cmd/dlmgmtd/dlmgmt_impl.h
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2011, Joyent Inc. All rights reserved.
  */
 
 /*
@@ -84,6 +85,8 @@ typedef struct dlmgmt_dlconf_s {
 	avl_node_t		ld_node;
 } dlmgmt_dlconf_t;
 
+#define ZONE_LOCK	"/etc/dladm/zone.lck"
+
 extern boolean_t	debug;
 extern const char	*progname;
 extern char		cachefile[];
diff --git a/usr/src/cmd/dlmgmtd/dlmgmt_main.c b/usr/src/cmd/dlmgmtd/dlmgmt_main.c
index 75e620a36b..7a26440a3f 100644
--- a/usr/src/cmd/dlmgmtd/dlmgmt_main.c
+++ b/usr/src/cmd/dlmgmtd/dlmgmt_main.c
@@ -287,6 +287,8 @@ dlmgmt_init(void)
 		return (err);
 	}
 
+	(void) unlink(ZONE_LOCK);
+
 	/*
 	 * First derive the name of the cache file from the FMRI name. This
 	 * cache name is used to keep active datalink configuration.
diff --git a/usr/src/lib/brand/joyent/zone/statechange.ksh b/usr/src/lib/brand/joyent/zone/statechange.ksh
index 5b9fe14aac..6511227691 100644
--- a/usr/src/lib/brand/joyent/zone/statechange.ksh
+++ b/usr/src/lib/brand/joyent/zone/statechange.ksh
@@ -55,6 +55,29 @@ ZONEPATH=$3
 state=$4
 cmd=$5
 
+LOCKFILE=/etc/dladm/zone.lck
+
+#
+# Create a lock file which we use to serialize datalink operations across zones.
+#
+lock_file()
+{
+	while true; do
+		if (set -o noclobber; echo "$$" >$LOCKFILE) 2>/dev/null; then
+			trap 'rm -f $LOCKFILE; exit $?' INT TERM EXIT
+			break;
+		else
+			sleep 1
+		fi
+	done
+}
+
+unlock_file()
+{
+	rm -f $LOCKFILE
+	trap - INT TERM EXIT
+}
+
 #
 # Set up the vnic(s) for the zone.
 #
@@ -99,6 +122,8 @@ setup_net()
 			exit 1
 		fi
 
+		lock_file
+
 		#
 		# Create the vnic.
 		#
@@ -252,6 +277,8 @@ setup_net()
 			done
 			IFS=$OLDIFS
 		fi
+
+		unlock_file
 	done
 }
 
@@ -287,12 +314,16 @@ cleanup_net()
 	# Cleanup any flows that were setup.
 	for nic in $_ZONECFG_net_resources
 	do
+		lock_file
+
 		flowadm remove-flow -t -z $ZONENAME -l $nic
 		if (( $? != 0 )); then
 			echo "error removing flows for $nic"
 			logger -p daemon.err "zone $ZONENAME " \
 			    "error removing flows for $nic"
 		fi
+
+		unlock_file
 	done
 }