decnet: RCU conversion and get rid of dev_base_lock

While tracking dev_base_lock users, I found decnet used it in
dnet_select_source(), but for a wrong purpose:

Writers only hold RTNL, not dev_base_lock, so readers must use RCU if
they cannot use RTNL.

Adds an rcu_head in struct dn_ifaddr and handle proper RCU management.

Adds __rcu annotation in dn_route as well.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Steven Whitehouse <swhiteho@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 39 insertions, 29 deletions

diff --git a/net/decnet/dn_route.c b/net/decnet/dn_route.c
index df0f3e54ff8..94a9eb1d313 100644
--- a/net/decnet/dn_route.c
+++ b/net/decnet/dn_route.c
@@ -93,7 +93,7 @@
 
 struct dn_rt_hash_bucket
 {
-	struct dn_route *chain;
+	struct dn_route __rcu *chain;
 	spinlock_t lock;
 };
 
@@ -157,15 +157,17 @@ static inline void dnrt_drop(struct dn_route *rt)
 static void dn_dst_check_expire(unsigned long dummy)
 {
 	int i;
-	struct dn_route *rt, **rtp;
+	struct dn_route *rt;
+	struct dn_route __rcu **rtp;
 	unsigned long now = jiffies;
 	unsigned long expire = 120 * HZ;
 
-	for(i = 0; i <= dn_rt_hash_mask; i++) {
+	for (i = 0; i <= dn_rt_hash_mask; i++) {
 		rtp = &dn_rt_hash_table[i].chain;
 
 		spin_lock(&dn_rt_hash_table[i].lock);
-		while((rt=*rtp) != NULL) {
+		while ((rt = rcu_dereference_protected(*rtp,
+						lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
 			if (atomic_read(&rt->dst.__refcnt) ||
 					(now - rt->dst.lastuse) < expire) {
 				rtp = &rt->dst.dn_next;
@@ -186,17 +188,19 @@ static void dn_dst_check_expire(unsigned long dummy)
 
 static int dn_dst_gc(struct dst_ops *ops)
 {
-	struct dn_route *rt, **rtp;
+	struct dn_route *rt;
+	struct dn_route __rcu **rtp;
 	int i;
 	unsigned long now = jiffies;
 	unsigned long expire = 10 * HZ;
 
-	for(i = 0; i <= dn_rt_hash_mask; i++) {
+	for (i = 0; i <= dn_rt_hash_mask; i++) {
 
 		spin_lock_bh(&dn_rt_hash_table[i].lock);
 		rtp = &dn_rt_hash_table[i].chain;
 
-		while((rt=*rtp) != NULL) {
+		while ((rt = rcu_dereference_protected(*rtp,
+						lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
 			if (atomic_read(&rt->dst.__refcnt) ||
 					(now - rt->dst.lastuse) < expire) {
 				rtp = &rt->dst.dn_next;
@@ -227,7 +231,7 @@ static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
 {
 	u32 min_mtu = 230;
 	struct dn_dev *dn = dst->neighbour ?
-			    (struct dn_dev *)dst->neighbour->dev->dn_ptr : NULL;
+			    rcu_dereference_raw(dst->neighbour->dev->dn_ptr) : NULL;
 
 	if (dn && dn->use_long == 0)
 		min_mtu -= 6;
@@ -277,13 +281,15 @@ static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
 
 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
 {
-	struct dn_route *rth, **rthp;
+	struct dn_route *rth;
+	struct dn_route __rcu **rthp;
 	unsigned long now = jiffies;
 
 	rthp = &dn_rt_hash_table[hash].chain;
 
 	spin_lock_bh(&dn_rt_hash_table[hash].lock);
-	while((rth = *rthp) != NULL) {
+	while ((rth = rcu_dereference_protected(*rthp,
+						lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
 		if (compare_keys(&rth->fl, &rt->fl)) {
 			/* Put it first */
 			*rthp = rth->dst.dn_next;
@@ -315,15 +321,15 @@ static void dn_run_flush(unsigned long dummy)
 	int i;
 	struct dn_route *rt, *next;
 
-	for(i = 0; i < dn_rt_hash_mask; i++) {
+	for (i = 0; i < dn_rt_hash_mask; i++) {
 		spin_lock_bh(&dn_rt_hash_table[i].lock);
 
-		if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
+		if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
 			goto nothing_to_declare;
 
-		for(; rt; rt=next) {
-			next = rt->dst.dn_next;
-			rt->dst.dn_next = NULL;
+		for(; rt; rt = next) {
+			next = rcu_dereference_raw(rt->dst.dn_next);
+			RCU_INIT_POINTER(rt->dst.dn_next, NULL);
 			dst_free((struct dst_entry *)rt);
 		}
 
@@ -458,15 +464,16 @@ static int dn_return_long(struct sk_buff *skb)
  */
 static int dn_route_rx_packet(struct sk_buff *skb)
 {
-	struct dn_skb_cb *cb = DN_SKB_CB(skb);
+	struct dn_skb_cb *cb;
 	int err;
 
 	if ((err = dn_route_input(skb)) == 0)
 		return dst_input(skb);
 
+	cb = DN_SKB_CB(skb);
 	if (decnet_debug_level & 4) {
 		char *devname = skb->dev ? skb->dev->name : "???";
-		struct dn_skb_cb *cb = DN_SKB_CB(skb);
+
 		printk(KERN_DEBUG
 			"DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
 			(int)cb->rt_flags, devname, skb->len,
@@ -573,7 +580,7 @@ int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type
 	struct dn_skb_cb *cb;
 	unsigned char flags = 0;
 	__u16 len = le16_to_cpu(*(__le16 *)skb->data);
-	struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
+	struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
 	unsigned char padlen = 0;
 
 	if (!net_eq(dev_net(dev), &init_net))
@@ -728,7 +735,7 @@ static int dn_forward(struct sk_buff *skb)
 {
 	struct dn_skb_cb *cb = DN_SKB_CB(skb);
 	struct dst_entry *dst = skb_dst(skb);
-	struct dn_dev *dn_db = dst->dev->dn_ptr;
+	struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
 	struct dn_route *rt;
 	struct neighbour *neigh = dst->neighbour;
 	int header_len;
@@ -835,13 +842,16 @@ static inline int dn_match_addr(__le16 addr1, __le16 addr2)
 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
 {
 	__le16 saddr = 0;
-	struct dn_dev *dn_db = dev->dn_ptr;
+	struct dn_dev *dn_db;
 	struct dn_ifaddr *ifa;
 	int best_match = 0;
 	int ret;
 
-	read_lock(&dev_base_lock);
-	for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
+	rcu_read_lock();
+	dn_db = rcu_dereference(dev->dn_ptr);
+	for (ifa = rcu_dereference(dn_db->ifa_list);
+	     ifa != NULL;
+	     ifa = rcu_dereference(ifa->ifa_next)) {
 		if (ifa->ifa_scope > scope)
 			continue;
 		if (!daddr) {
@@ -854,7 +864,7 @@ static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int
 		if (best_match == 0)
 			saddr = ifa->ifa_local;
 	}
-	read_unlock(&dev_base_lock);
+	rcu_read_unlock();
 
 	return saddr;
 }
@@ -1020,7 +1030,7 @@ source_ok:
 		err = -ENODEV;
 		if (dev_out == NULL)
 			goto out;
-		dn_db = dev_out->dn_ptr;
+		dn_db = rcu_dereference_raw(dev_out->dn_ptr);
 		/* Possible improvement - check all devices for local addr */
 		if (dn_dev_islocal(dev_out, fl.fld_dst)) {
 			dev_put(dev_out);
@@ -1233,7 +1243,7 @@ static int dn_route_input_slow(struct sk_buff *skb)
 
 	dev_hold(in_dev);
 
-	if ((dn_db = in_dev->dn_ptr) == NULL)
+	if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
 		goto out;
 
 	/* Zero source addresses are not allowed */
@@ -1677,15 +1687,15 @@ static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_rou
 {
 	struct dn_rt_cache_iter_state *s = seq->private;
 
-	rt = rt->dst.dn_next;
-	while(!rt) {
+	rt = rcu_dereference_bh(rt->dst.dn_next);
+	while (!rt) {
 		rcu_read_unlock_bh();
 		if (--s->bucket < 0)
 			break;
 		rcu_read_lock_bh();
-		rt = dn_rt_hash_table[s->bucket].chain;
+		rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
 	}
-	return rcu_dereference_bh(rt);
+	return rt;
 }
 
 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)