# HG changeset patch
# Parent 2b91a85590f2846e352b51c6276fbb10203411f0

diff -r 2b91a85590f2 sys/netinet/cc.h
--- a/sys/netinet/cc.h	Mon Nov 15 11:36:36 2010 +1100
+++ b/sys/netinet/cc.h	Tue Nov 16 01:19:20 2010 +1100
@@ -58,6 +58,10 @@
 extern const int tcprexmtthresh;
 extern struct cc_algo newreno_cc_algo;
 
+/* Per-netstack bits. */
+VNET_DECLARE(struct cc_algo *, default_cc_ptr);
+#define	V_default_cc_ptr VNET(default_cc_ptr)
+
 /* Define the new net.inet.tcp.cc sysctl tree. */
 SYSCTL_DECL(_net_inet_tcp_cc);
 
@@ -146,7 +150,7 @@
 #define	CC_DATA(tp)	((tp)->ccv->cc_data)
 
 /* Macro to obtain the system default CC algo's struct ptr. */
-#define	CC_DEFAULT()	STAILQ_FIRST(&cc_list)
+#define	CC_DEFAULT()	V_default_cc_ptr
 
 extern struct rwlock cc_list_lock;
 #define	CC_LIST_LOCK_INIT()	rw_init(&cc_list_lock, "cc_list")
@@ -155,6 +159,6 @@
 #define	CC_LIST_RUNLOCK()	rw_runlock(&cc_list_lock)
 #define	CC_LIST_WLOCK()		rw_wlock(&cc_list_lock)
 #define	CC_LIST_WUNLOCK()	rw_wunlock(&cc_list_lock)
-#define	CC_LIST_WLOCK_ASSERT()	rw_assert(&cc_list_lock, RA_WLOCKED)
+#define	CC_LIST_LOCK_ASSERT()	rw_assert(&cc_list_lock, RA_LOCKED)
 
 #endif /* _NETINET_CC_H_ */
diff -r 2b91a85590f2 sys/netinet/cc/cc.c
--- a/sys/netinet/cc/cc.c	Mon Nov 15 11:36:36 2010 +1100
+++ b/sys/netinet/cc/cc.c	Tue Nov 16 01:19:20 2010 +1100
@@ -81,24 +81,7 @@
 /* Protects the cc_list TAILQ. */
 struct rwlock cc_list_lock;
 
-/*
- * Set the default CC algorithm to new_default. The default is identified
- * by being the first element in the cc_list TAILQ.
- */
-static void
-cc_set_default(struct cc_algo *new_default)
-{
-	CC_LIST_WLOCK_ASSERT();
-
-	/*
-	 * Make the requested system default CC algorithm the first element in
-	 * the list if it isn't already.
-	 */
-	if (new_default != CC_DEFAULT()) {
-		STAILQ_REMOVE(&cc_list, new_default, cc_algo, entries);
-		STAILQ_INSERT_HEAD(&cc_list, new_default, entries);
-	}
-}
+VNET_DEFINE(struct cc_algo *, default_cc_ptr) = &newreno_cc_algo;
 
 /*
  * Sysctl handler to show and change the default CC algorithm.
@@ -106,14 +89,13 @@
 static int
 cc_default_algo(SYSCTL_HANDLER_ARGS)
 {
+	char default_cc[TCP_CA_NAME_MAX];
 	struct cc_algo *funcs;
 	int err, found;
 
 	err = found = 0;
 
 	if (req->newptr == NULL) {
-		char default_cc[TCP_CA_NAME_MAX];
-
 		/* Just print the current default. */
 		CC_LIST_RLOCK();
 		strlcpy(default_cc, CC_DEFAULT()->name, TCP_CA_NAME_MAX);
@@ -121,15 +103,15 @@
 		err = sysctl_handle_string(oidp, default_cc, 1, req);
 	} else {
 		/* Find algo with specified name and set it to default. */
-		CC_LIST_WLOCK();
+		CC_LIST_RLOCK();
 		STAILQ_FOREACH(funcs, &cc_list, entries) {
 			if (strncmp((char *)req->newptr, funcs->name,
 			    TCP_CA_NAME_MAX) == 0) {
 				found = 1;
-				cc_set_default(funcs);
+				CC_DEFAULT() = funcs;
 			}
 		}
-		CC_LIST_WUNLOCK();
+		CC_LIST_RUNLOCK();
 
 		if (!found)
 			err = ESRCH;
@@ -173,6 +155,28 @@
 	return (err);
 }
 
+static void
+cc_checkreset_default(struct cc_algo *remove_cc)
+{
+	VNET_ITERATOR_DECL(vnet_iter);
+
+	CC_LIST_LOCK_ASSERT();
+
+	/*
+	 * Check the default CC algo for each vnet and reset to NewReno if the
+	 * vnet is using the algo that is about to go away.
+	 */
+	VNET_LIST_RLOCK_NOSLEEP();
+	VNET_FOREACH(vnet_iter) {
+		CURVNET_SET(vnet_iter);
+		if (strncmp(CC_DEFAULT()->name, remove_cc->name,
+		    TCP_CA_NAME_MAX) == 0)
+			CC_DEFAULT() = &newreno_cc_algo;
+		CURVNET_RESTORE();
+	}
+	VNET_LIST_RUNLOCK_NOSLEEP();
+}
+
 /*
  * Initialise CC subsystem on system boot.
  */
@@ -202,14 +206,7 @@
 	CC_LIST_WLOCK();
 	STAILQ_FOREACH_SAFE(funcs, &cc_list, entries, tmpfuncs) {
 		if (funcs == remove_cc) {
-			/*
-			 * If we're removing the current system default,
-			 * reset the default to newreno.
-			 */
-			if (strncmp(CC_DEFAULT()->name, remove_cc->name,
-			    TCP_CA_NAME_MAX) == 0)
-				cc_set_default(&newreno_cc_algo);
-
+			cc_checkreset_default(remove_cc);
 			STAILQ_REMOVE(&cc_list, funcs, cc_algo, entries);
 			err = 0;
 			break;
@@ -303,7 +300,7 @@
 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, cc, CTLFLAG_RW, NULL,
     "congestion control related settings");
 
-SYSCTL_PROC(_net_inet_tcp_cc, OID_AUTO, algorithm, CTLTYPE_STRING|CTLFLAG_RW,
+SYSCTL_VNET_PROC(_net_inet_tcp_cc, OID_AUTO, algorithm, CTLTYPE_STRING|CTLFLAG_RW,
     NULL, 0, cc_default_algo, "A", "default congestion control algorithm");
 
 SYSCTL_PROC(_net_inet_tcp_cc, OID_AUTO, available, CTLTYPE_STRING|CTLFLAG_RD,