Merge branch 'stealer/ipvs/for-davem' of git://git.stealer.net/linux-2.6

include/net/ip_vs.h

@@ -140,8 +140,24 @@ struct ip_vs_seq {
 
 
 /*
- *	IPVS statistics object
+ *	IPVS statistics objects
  */
+struct ip_vs_estimator {
+	struct list_head	list;
+
+	u64			last_inbytes;
+	u64			last_outbytes;
+	u32			last_conns;
+	u32			last_inpkts;
+	u32			last_outpkts;
+
+	u32			cps;
+	u32			inpps;
+	u32			outpps;
+	u32			inbps;
+	u32			outbps;
+};
+
 struct ip_vs_stats
 {
 	__u32                   conns;          /* connections scheduled */
@@ -156,7 +172,15 @@ struct ip_vs_stats
 	__u32			inbps;		/* current in byte rate */
 	__u32			outbps;		/* current out byte rate */
 
+	/*
+	 * Don't add anything before the lock, because we use memcpy() to copy
+	 * the members before the lock to struct ip_vs_stats_user in
+	 * ip_vs_ctl.c.
+	 */
+
 	spinlock_t              lock;           /* spin lock */
+
+	struct ip_vs_estimator	est;		/* estimator */
 };
 
 struct dst_entry;
@@ -440,7 +464,7 @@ struct ip_vs_app
  */
 extern const char *ip_vs_proto_name(unsigned proto);
 extern void ip_vs_init_hash_table(struct list_head *table, int rows);
-#define IP_VS_INIT_HASH_TABLE(t) ip_vs_init_hash_table(t, sizeof(t)/sizeof(t[0]))
+#define IP_VS_INIT_HASH_TABLE(t) ip_vs_init_hash_table((t), ARRAY_SIZE((t)))
 
 #define IP_VS_APP_TYPE_FTP	1
 
@@ -620,7 +644,7 @@ extern int sysctl_ip_vs_expire_quiescent_template;
 extern int sysctl_ip_vs_sync_threshold[2];
 extern int sysctl_ip_vs_nat_icmp_send;
 extern struct ip_vs_stats ip_vs_stats;
-extern struct ctl_path net_vs_ctl_path[];
+extern const struct ctl_path net_vs_ctl_path[];
 
 extern struct ip_vs_service *
 ip_vs_service_get(__u32 fwmark, __u16 protocol, __be32 vaddr, __be16 vport);
@@ -659,7 +683,7 @@ extern void ip_vs_sync_conn(struct ip_vs_conn *cp);
 /*
  *      IPVS rate estimator prototypes (from ip_vs_est.c)
  */
-extern int ip_vs_new_estimator(struct ip_vs_stats *stats);
+extern void ip_vs_new_estimator(struct ip_vs_stats *stats);
 extern void ip_vs_kill_estimator(struct ip_vs_stats *stats);
 extern void ip_vs_zero_estimator(struct ip_vs_stats *stats);
 

net/ipv4/ipvs/ip_vs_app.c

@@ -608,7 +608,7 @@ int ip_vs_skb_replace(struct sk_buff *skb, gfp_t pri,
 }
 
 
-int ip_vs_app_init(void)
+int __init ip_vs_app_init(void)
 {
 	/* we will replace it with proc_net_ipvs_create() soon */
 	proc_net_fops_create(&init_net, "ip_vs_app", 0, &ip_vs_app_fops);

net/ipv4/ipvs/ip_vs_conn.c

@@ -965,7 +965,7 @@ static void ip_vs_conn_flush(void)
 }
 
 
-int ip_vs_conn_init(void)
+int __init ip_vs_conn_init(void)
 {
 	int idx;
 

net/ipv4/ipvs/ip_vs_ctl.c

@@ -683,9 +683,22 @@ static void
 ip_vs_zero_stats(struct ip_vs_stats *stats)
 {
 	spin_lock_bh(&stats->lock);
-	memset(stats, 0, (char *)&stats->lock - (char *)stats);
-	spin_unlock_bh(&stats->lock);
+
+	stats->conns = 0;
+	stats->inpkts = 0;
+	stats->outpkts = 0;
+	stats->inbytes = 0;
+	stats->outbytes = 0;
+
+	stats->cps = 0;
+	stats->inpps = 0;
+	stats->outpps = 0;
+	stats->inbps = 0;
+	stats->outbps = 0;
+
 	ip_vs_zero_estimator(stats);
+
+	spin_unlock_bh(&stats->lock);
 }
 
 /*
@@ -1589,7 +1602,7 @@ static struct ctl_table vs_vars[] = {
 	{ .ctl_name = 0 }
 };
 
-struct ctl_path net_vs_ctl_path[] = {
+const struct ctl_path net_vs_ctl_path[] = {
 	{ .procname = "net", .ctl_name = CTL_NET, },
 	{ .procname = "ipv4", .ctl_name = NET_IPV4, },
 	{ .procname = "vs", },
@@ -1784,7 +1797,9 @@ static const struct file_operations ip_vs_info_fops = {
 
 #endif
 
-struct ip_vs_stats ip_vs_stats;
+struct ip_vs_stats ip_vs_stats = {
+	.lock = __SPIN_LOCK_UNLOCKED(ip_vs_stats.lock),
+};
 
 #ifdef CONFIG_PROC_FS
 static int ip_vs_stats_show(struct seq_file *seq, void *v)
@@ -2306,7 +2321,7 @@ static struct nf_sockopt_ops ip_vs_sockopts = {
 };
 
 
-int ip_vs_control_init(void)
+int __init ip_vs_control_init(void)
 {
 	int ret;
 	int idx;
@@ -2333,8 +2348,6 @@ int ip_vs_control_init(void)
 		INIT_LIST_HEAD(&ip_vs_rtable[idx]);
 	}
 
-	memset(&ip_vs_stats, 0, sizeof(ip_vs_stats));
-	spin_lock_init(&ip_vs_stats.lock);
 	ip_vs_new_estimator(&ip_vs_stats);
 
 	/* Hook the defense timer */

net/ipv4/ipvs/ip_vs_dh.c

@@ -233,6 +233,7 @@ static struct ip_vs_scheduler ip_vs_dh_scheduler =
 	.name =			"dh",
 	.refcnt =		ATOMIC_INIT(0),
 	.module =		THIS_MODULE,
+	.n_list =		LIST_HEAD_INIT(ip_vs_dh_scheduler.n_list),
 	.init_service =		ip_vs_dh_init_svc,
 	.done_service =		ip_vs_dh_done_svc,
 	.update_service =	ip_vs_dh_update_svc,
@@ -242,7 +243,6 @@ static struct ip_vs_scheduler ip_vs_dh_scheduler =
 
 static int __init ip_vs_dh_init(void)
 {
-	INIT_LIST_HEAD(&ip_vs_dh_scheduler.n_list);
 	return register_ip_vs_scheduler(&ip_vs_dh_scheduler);
 }
 

net/ipv4/ipvs/ip_vs_est.c

@@ -17,6 +17,7 @@
 #include <linux/types.h>
 #include <linux/interrupt.h>
 #include <linux/sysctl.h>
+#include <linux/list.h>
 
 #include <net/ip_vs.h>
 
@@ -44,28 +45,11 @@
  */
 
 
-struct ip_vs_estimator
-{
-	struct ip_vs_estimator	*next;
-	struct ip_vs_stats	*stats;
-
-	u32			last_conns;
-	u32			last_inpkts;
-	u32			last_outpkts;
-	u64			last_inbytes;
-	u64			last_outbytes;
-
-	u32			cps;
-	u32			inpps;
-	u32			outpps;
-	u32			inbps;
-	u32			outbps;
-};
+static void estimation_timer(unsigned long arg);
 
-
-static struct ip_vs_estimator *est_list = NULL;
-static DEFINE_RWLOCK(est_lock);
-static struct timer_list est_timer;
+static LIST_HEAD(est_list);
+static DEFINE_SPINLOCK(est_lock);
+static DEFINE_TIMER(est_timer, estimation_timer, 0, 0);
 
 static void estimation_timer(unsigned long arg)
 {
@@ -76,9 +60,9 @@ static void estimation_timer(unsigned long arg)
 	u64 n_inbytes, n_outbytes;
 	u32 rate;
 
-	read_lock(&est_lock);
-	for (e = est_list; e; e = e->next) {
-		s = e->stats;
+	spin_lock(&est_lock);
+	list_for_each_entry(e, &est_list, list) {
+		s = container_of(e, struct ip_vs_stats, est);
 
 		spin_lock(&s->lock);
 		n_conns = s->conns;
@@ -114,19 +98,16 @@ static void estimation_timer(unsigned long arg)
 		s->outbps = (e->outbps+0xF)>>5;
 		spin_unlock(&s->lock);
 	}
-	read_unlock(&est_lock);
+	spin_unlock(&est_lock);
 	mod_timer(&est_timer, jiffies + 2*HZ);
 }
 
-int ip_vs_new_estimator(struct ip_vs_stats *stats)
+void ip_vs_new_estimator(struct ip_vs_stats *stats)
 {
-	struct ip_vs_estimator *est;
+	struct ip_vs_estimator *est = &stats->est;
 
-	est = kzalloc(sizeof(*est), GFP_KERNEL);
-	if (est == NULL)
-		return -ENOMEM;
+	INIT_LIST_HEAD(&est->list);
 
-	est->stats = stats;
 	est->last_conns = stats->conns;
 	est->cps = stats->cps<<10;
 
@@ -142,59 +123,40 @@ int ip_vs_new_estimator(struct ip_vs_stats *stats)
 	est->last_outbytes = stats->outbytes;
 	est->outbps = stats->outbps<<5;
 
-	write_lock_bh(&est_lock);
-	est->next = est_list;
-	if (est->next == NULL) {
-		setup_timer(&est_timer, estimation_timer, 0);
-		est_timer.expires = jiffies + 2*HZ;
-		add_timer(&est_timer);
-	}
-	est_list = est;
-	write_unlock_bh(&est_lock);
-	return 0;
+	spin_lock_bh(&est_lock);
+	if (list_empty(&est_list))
+		mod_timer(&est_timer, jiffies + 2 * HZ);
+	list_add(&est->list, &est_list);
+	spin_unlock_bh(&est_lock);
 }
 
 void ip_vs_kill_estimator(struct ip_vs_stats *stats)
 {
-	struct ip_vs_estimator *est, **pest;
-	int killed = 0;
-
-	write_lock_bh(&est_lock);
-	pest = &est_list;
-	while ((est=*pest) != NULL) {
-		if (est->stats != stats) {
-			pest = &est->next;
-			continue;
+	struct ip_vs_estimator *est = &stats->est;
+
+	spin_lock_bh(&est_lock);
+	list_del(&est->list);
+	while (list_empty(&est_list) && try_to_del_timer_sync(&est_timer) < 0) {
+		spin_unlock_bh(&est_lock);
+		cpu_relax();
+		spin_lock_bh(&est_lock);
 	}
-		*pest = est->next;
-		kfree(est);
-		killed++;
-	}
-	if (killed && est_list == NULL)
-		del_timer_sync(&est_timer);
-	write_unlock_bh(&est_lock);
+	spin_unlock_bh(&est_lock);
 }
 
 void ip_vs_zero_estimator(struct ip_vs_stats *stats)
 {
-	struct ip_vs_estimator *e;
-
-	write_lock_bh(&est_lock);
-	for (e = est_list; e; e = e->next) {
-		if (e->stats != stats)
-			continue;
-
-		/* set counters zero */
-		e->last_conns = 0;
-		e->last_inpkts = 0;
-		e->last_outpkts = 0;
-		e->last_inbytes = 0;
-		e->last_outbytes = 0;
-		e->cps = 0;
-		e->inpps = 0;
-		e->outpps = 0;
-		e->inbps = 0;
-		e->outbps = 0;
-	}
-	write_unlock_bh(&est_lock);
+	struct ip_vs_estimator *est = &stats->est;
+
+	/* set counters zero, caller must hold the stats->lock lock */
+	est->last_inbytes = 0;
+	est->last_outbytes = 0;
+	est->last_conns = 0;
+	est->last_inpkts = 0;
+	est->last_outpkts = 0;
+	est->cps = 0;
+	est->inpps = 0;
+	est->outpps = 0;
+	est->inbps = 0;
+	est->outbps = 0;
 }

net/ipv4/ipvs/ip_vs_lblc.c

@@ -539,6 +539,7 @@ static struct ip_vs_scheduler ip_vs_lblc_scheduler =
 	.name =			"lblc",
 	.refcnt =		ATOMIC_INIT(0),
 	.module =		THIS_MODULE,
+	.n_list =		LIST_HEAD_INIT(ip_vs_lblc_scheduler.n_list),
 	.init_service =		ip_vs_lblc_init_svc,
 	.done_service =		ip_vs_lblc_done_svc,
 	.update_service =	ip_vs_lblc_update_svc,
@@ -550,7 +551,6 @@ static int __init ip_vs_lblc_init(void)
 {
 	int ret;
 
-	INIT_LIST_HEAD(&ip_vs_lblc_scheduler.n_list);
 	sysctl_header = register_sysctl_paths(net_vs_ctl_path, vs_vars_table);
 	ret = register_ip_vs_scheduler(&ip_vs_lblc_scheduler);
 	if (ret)

net/ipv4/ipvs/ip_vs_lblcr.c

@@ -728,6 +728,7 @@ static struct ip_vs_scheduler ip_vs_lblcr_scheduler =
 	.name =			"lblcr",
 	.refcnt =		ATOMIC_INIT(0),
 	.module =		THIS_MODULE,
+	.n_list =		LIST_HEAD_INIT(ip_vs_lblcr_scheduler.n_list),
 	.init_service =		ip_vs_lblcr_init_svc,
 	.done_service =		ip_vs_lblcr_done_svc,
 	.update_service =	ip_vs_lblcr_update_svc,
@@ -739,7 +740,6 @@ static int __init ip_vs_lblcr_init(void)
 {
 	int ret;
 
-	INIT_LIST_HEAD(&ip_vs_lblcr_scheduler.n_list);
 	sysctl_header = register_sysctl_paths(net_vs_ctl_path, vs_vars_table);
 	ret = register_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
 	if (ret)

net/ipv4/ipvs/ip_vs_lc.c

@@ -98,6 +98,7 @@ static struct ip_vs_scheduler ip_vs_lc_scheduler = {
 	.name =			"lc",
 	.refcnt =		ATOMIC_INIT(0),
 	.module =		THIS_MODULE,
+	.n_list =		LIST_HEAD_INIT(ip_vs_lc_scheduler.n_list),
 	.init_service =		ip_vs_lc_init_svc,
 	.done_service =		ip_vs_lc_done_svc,
 	.update_service =	ip_vs_lc_update_svc,
@@ -107,7 +108,6 @@ static struct ip_vs_scheduler ip_vs_lc_scheduler = {
 
 static int __init ip_vs_lc_init(void)
 {
-	INIT_LIST_HEAD(&ip_vs_lc_scheduler.n_list);
 	return register_ip_vs_scheduler(&ip_vs_lc_scheduler) ;
 }
 

net/ipv4/ipvs/ip_vs_nq.c

@@ -136,6 +136,7 @@ static struct ip_vs_scheduler ip_vs_nq_scheduler =
 	.name =			"nq",
 	.refcnt =		ATOMIC_INIT(0),
 	.module =		THIS_MODULE,
+	.n_list =		LIST_HEAD_INIT(ip_vs_nq_scheduler.n_list),
 	.init_service =		ip_vs_nq_init_svc,
 	.done_service =		ip_vs_nq_done_svc,
 	.update_service =	ip_vs_nq_update_svc,
@@ -145,7 +146,6 @@ static struct ip_vs_scheduler ip_vs_nq_scheduler =
 
 static int __init ip_vs_nq_init(void)
 {
-	INIT_LIST_HEAD(&ip_vs_nq_scheduler.n_list);
 	return register_ip_vs_scheduler(&ip_vs_nq_scheduler);
 }
 

net/ipv4/ipvs/ip_vs_proto.c

@@ -43,7 +43,7 @@ static struct ip_vs_protocol *ip_vs_proto_table[IP_VS_PROTO_TAB_SIZE];
 /*
  *	register an ipvs protocol
  */
-static int __used register_ip_vs_protocol(struct ip_vs_protocol *pp)
+static int __used __init register_ip_vs_protocol(struct ip_vs_protocol *pp)
 {
 	unsigned hash = IP_VS_PROTO_HASH(pp->protocol);
 
@@ -190,7 +190,7 @@ ip_vs_tcpudp_debug_packet(struct ip_vs_protocol *pp,
 }
 
 
-int ip_vs_protocol_init(void)
+int __init ip_vs_protocol_init(void)
 {
 	char protocols[64];
 #define REGISTER_PROTOCOL(p)			\

net/ipv4/ipvs/ip_vs_rr.c

@@ -94,6 +94,7 @@ static struct ip_vs_scheduler ip_vs_rr_scheduler = {
 	.name =			"rr",			/* name */
 	.refcnt =		ATOMIC_INIT(0),
 	.module =		THIS_MODULE,
+	.n_list =		LIST_HEAD_INIT(ip_vs_rr_scheduler.n_list),
 	.init_service =		ip_vs_rr_init_svc,
 	.done_service =		ip_vs_rr_done_svc,
 	.update_service =	ip_vs_rr_update_svc,
@@ -102,7 +103,6 @@ static struct ip_vs_scheduler ip_vs_rr_scheduler = {
 
 static int __init ip_vs_rr_init(void)
 {
-	INIT_LIST_HEAD(&ip_vs_rr_scheduler.n_list);
 	return register_ip_vs_scheduler(&ip_vs_rr_scheduler);
 }
 

net/ipv4/ipvs/ip_vs_sched.c

@@ -184,7 +184,7 @@ int register_ip_vs_scheduler(struct ip_vs_scheduler *scheduler)
 
 	write_lock_bh(&__ip_vs_sched_lock);
 
-	if (scheduler->n_list.next != &scheduler->n_list) {
+	if (!list_empty(&scheduler->n_list)) {
 		write_unlock_bh(&__ip_vs_sched_lock);
 		ip_vs_use_count_dec();
 		IP_VS_ERR("register_ip_vs_scheduler(): [%s] scheduler "
@@ -229,7 +229,7 @@ int unregister_ip_vs_scheduler(struct ip_vs_scheduler *scheduler)
 	}
 
 	write_lock_bh(&__ip_vs_sched_lock);
-	if (scheduler->n_list.next == &scheduler->n_list) {
+	if (list_empty(&scheduler->n_list)) {
 		write_unlock_bh(&__ip_vs_sched_lock);
 		IP_VS_ERR("unregister_ip_vs_scheduler(): [%s] scheduler "
 			  "is not in the list. failed\n", scheduler->name);

net/ipv4/ipvs/ip_vs_sed.c

@@ -138,6 +138,7 @@ static struct ip_vs_scheduler ip_vs_sed_scheduler =
 	.name =			"sed",
 	.refcnt =		ATOMIC_INIT(0),
 	.module =		THIS_MODULE,
+	.n_list =		LIST_HEAD_INIT(ip_vs_sed_scheduler.n_list),
 	.init_service =		ip_vs_sed_init_svc,
 	.done_service =		ip_vs_sed_done_svc,
 	.update_service =	ip_vs_sed_update_svc,
@@ -147,7 +148,6 @@ static struct ip_vs_scheduler ip_vs_sed_scheduler =
 
 static int __init ip_vs_sed_init(void)
 {
-	INIT_LIST_HEAD(&ip_vs_sed_scheduler.n_list);
 	return register_ip_vs_scheduler(&ip_vs_sed_scheduler);
 }
 

net/ipv4/ipvs/ip_vs_sh.c

@@ -230,6 +230,7 @@ static struct ip_vs_scheduler ip_vs_sh_scheduler =
 	.name =			"sh",
 	.refcnt =		ATOMIC_INIT(0),
 	.module =		THIS_MODULE,
+	.n_list	 =		LIST_HEAD_INIT(ip_vs_sh_scheduler.n_list),
 	.init_service =		ip_vs_sh_init_svc,
 	.done_service =		ip_vs_sh_done_svc,
 	.update_service =	ip_vs_sh_update_svc,
@@ -239,7 +240,6 @@ static struct ip_vs_scheduler ip_vs_sh_scheduler =
 
 static int __init ip_vs_sh_init(void)
 {
-	INIT_LIST_HEAD(&ip_vs_sh_scheduler.n_list);
 	return register_ip_vs_scheduler(&ip_vs_sh_scheduler);
 }
 

net/ipv4/ipvs/ip_vs_sync.c

@@ -904,9 +904,9 @@ int stop_sync_thread(int state)
 		 * progress of stopping the master sync daemon.
 		 */
 
-		spin_lock(&ip_vs_sync_lock);
+		spin_lock_bh(&ip_vs_sync_lock);
 		ip_vs_sync_state &= ~IP_VS_STATE_MASTER;
-		spin_unlock(&ip_vs_sync_lock);
+		spin_unlock_bh(&ip_vs_sync_lock);
 		kthread_stop(sync_master_thread);
 		sync_master_thread = NULL;
 	} else if (state == IP_VS_STATE_BACKUP) {

net/ipv4/ipvs/ip_vs_wlc.c

@@ -126,6 +126,7 @@ static struct ip_vs_scheduler ip_vs_wlc_scheduler =
 	.name =			"wlc",
 	.refcnt =		ATOMIC_INIT(0),
 	.module =		THIS_MODULE,
+	.n_list =		LIST_HEAD_INIT(ip_vs_wlc_scheduler.n_list),
 	.init_service =		ip_vs_wlc_init_svc,
 	.done_service =		ip_vs_wlc_done_svc,
 	.update_service =	ip_vs_wlc_update_svc,
@@ -135,7 +136,6 @@ static struct ip_vs_scheduler ip_vs_wlc_scheduler =
 
 static int __init ip_vs_wlc_init(void)
 {
-	INIT_LIST_HEAD(&ip_vs_wlc_scheduler.n_list);
 	return register_ip_vs_scheduler(&ip_vs_wlc_scheduler);
 }
 

net/ipv4/ipvs/ip_vs_wrr.c

@@ -212,6 +212,7 @@ static struct ip_vs_scheduler ip_vs_wrr_scheduler = {
 	.name =			"wrr",
 	.refcnt =		ATOMIC_INIT(0),
 	.module =		THIS_MODULE,
+	.n_list =		LIST_HEAD_INIT(ip_vs_wrr_scheduler.n_list),
 	.init_service =		ip_vs_wrr_init_svc,
 	.done_service =		ip_vs_wrr_done_svc,
 	.update_service =	ip_vs_wrr_update_svc,
@@ -220,7 +221,6 @@ static struct ip_vs_scheduler ip_vs_wrr_scheduler = {
 
 static int __init ip_vs_wrr_init(void)
 {
-	INIT_LIST_HEAD(&ip_vs_wrr_scheduler.n_list);
 	return register_ip_vs_scheduler(&ip_vs_wrr_scheduler) ;
 }