Merge branch 'ipmr-remove-rwlocks'

Eric Dumazet says:

====================
ipmr: get rid of rwlocks

We need to get rid of rwlocks in networking stacks,
if read_lock() is (ab)used from softirq context.

As discussed recently [1], rwlock are unfair by design in this case,
and writers can starve and trigger soft lockups.

This series convert ipmr code (both IPv4 and IPv6 families)
to RCU and spinlocks.

[1] https://lkml.org/lkml/2022/6/17/272

v2: fixed two typos, and resent because patch 19/19
    did not make it to patchwork.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

include/linux/mroute_base.h

@@ -26,7 +26,7 @@
  * @remote: Remote address for tunnels
  */
 struct vif_device {
-	struct net_device *dev;
+	struct net_device __rcu *dev;
 	netdevice_tracker dev_tracker;
 	unsigned long bytes_in, bytes_out;
 	unsigned long pkt_in, pkt_out;
@@ -52,6 +52,7 @@ static inline int mr_call_vif_notifier(struct notifier_block *nb,
 				       unsigned short family,
 				       enum fib_event_type event_type,
 				       struct vif_device *vif,
+				       struct net_device *vif_dev,
 				       unsigned short vif_index, u32 tb_id,
 				       struct netlink_ext_ack *extack)
 {
@@ -60,7 +61,7 @@ static inline int mr_call_vif_notifier(struct notifier_block *nb,
 			.family = family,
 			.extack = extack,
 		},
-		.dev = vif->dev,
+		.dev = vif_dev,
 		.vif_index = vif_index,
 		.vif_flags = vif->flags,
 		.tb_id = tb_id,
@@ -73,6 +74,7 @@ static inline int mr_call_vif_notifiers(struct net *net,
 					unsigned short family,
 					enum fib_event_type event_type,
 					struct vif_device *vif,
+					struct net_device *vif_dev,
 					unsigned short vif_index, u32 tb_id,
 					unsigned int *ipmr_seq)
 {
@@ -80,7 +82,7 @@ static inline int mr_call_vif_notifiers(struct net *net,
 		.info = {
 			.family = family,
 		},
-		.dev = vif->dev,
+		.dev = vif_dev,
 		.vif_index = vif_index,
 		.vif_flags = vif->flags,
 		.tb_id = tb_id,
@@ -98,7 +100,8 @@ static inline int mr_call_vif_notifiers(struct net *net,
 #define MAXVIFS	32
 #endif
 
-#define VIF_EXISTS(_mrt, _idx) (!!((_mrt)->vif_table[_idx].dev))
+/* Note: This helper is deprecated. */
+#define VIF_EXISTS(_mrt, _idx) (!!rcu_access_pointer((_mrt)->vif_table[_idx].dev))
 
 /* mfc_flags:
  * MFC_STATIC - the entry was added statically (not by a routing daemon)
@@ -305,7 +308,7 @@ int mr_dump(struct net *net, struct notifier_block *nb, unsigned short family,
 			      struct netlink_ext_ack *extack),
 	    struct mr_table *(*mr_iter)(struct net *net,
 					struct mr_table *mrt),
-	    rwlock_t *mrt_lock, struct netlink_ext_ack *extack);
+	    struct netlink_ext_ack *extack);
 #else
 static inline void vif_device_init(struct vif_device *v,
 				   struct net_device *dev,
@@ -360,7 +363,7 @@ static inline int mr_dump(struct net *net, struct notifier_block *nb,
 					    struct netlink_ext_ack *extack),
 			  struct mr_table *(*mr_iter)(struct net *net,
 						      struct mr_table *mrt),
-			  rwlock_t *mrt_lock, struct netlink_ext_ack *extack)
+			  struct netlink_ext_ack *extack)
 {
 	return -EINVAL;
 }

net/ipv4/ipmr_base.c

@@ -13,7 +13,7 @@ void vif_device_init(struct vif_device *v,
 		     unsigned short flags,
 		     unsigned short get_iflink_mask)
 {
-	v->dev = NULL;
+	RCU_INIT_POINTER(v->dev, NULL);
 	v->bytes_in = 0;
 	v->bytes_out = 0;
 	v->pkt_in = 0;
@@ -208,6 +208,7 @@ EXPORT_SYMBOL(mr_mfc_seq_next);
 int mr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 		   struct mr_mfc *c, struct rtmsg *rtm)
 {
+	struct net_device *vif_dev;
 	struct rta_mfc_stats mfcs;
 	struct nlattr *mp_attr;
 	struct rtnexthop *nhp;
@@ -220,10 +221,13 @@ int mr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 		return -ENOENT;
 	}
 
-	if (VIF_EXISTS(mrt, c->mfc_parent) &&
-	    nla_put_u32(skb, RTA_IIF,
-			mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
+	rcu_read_lock();
+	vif_dev = rcu_dereference(mrt->vif_table[c->mfc_parent].dev);
+	if (vif_dev && nla_put_u32(skb, RTA_IIF, vif_dev->ifindex) < 0) {
+		rcu_read_unlock();
 		return -EMSGSIZE;
+	}
+	rcu_read_unlock();
 
 	if (c->mfc_flags & MFC_OFFLOAD)
 		rtm->rtm_flags |= RTNH_F_OFFLOAD;
@@ -232,23 +236,27 @@ int mr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 	if (!mp_attr)
 		return -EMSGSIZE;
 
+	rcu_read_lock();
 	for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
-		if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
-			struct vif_device *vif;
+		struct vif_device *vif = &mrt->vif_table[ct];
+
+		vif_dev = rcu_dereference(vif->dev);
+		if (vif_dev && c->mfc_un.res.ttls[ct] < 255) {
 
 			nhp = nla_reserve_nohdr(skb, sizeof(*nhp));
 			if (!nhp) {
+				rcu_read_unlock();
 				nla_nest_cancel(skb, mp_attr);
 				return -EMSGSIZE;
 			}
 
 			nhp->rtnh_flags = 0;
 			nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
-			vif = &mrt->vif_table[ct];
-			nhp->rtnh_ifindex = vif->dev->ifindex;
+			nhp->rtnh_ifindex = vif_dev->ifindex;
 			nhp->rtnh_len = sizeof(*nhp);
 		}
 	}
+	rcu_read_unlock();
 
 	nla_nest_end(skb, mp_attr);
 
@@ -275,13 +283,14 @@ static bool mr_mfc_uses_dev(const struct mr_table *mrt,
 	int ct;
 
 	for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
-		if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
-			const struct vif_device *vif;
-
-			vif = &mrt->vif_table[ct];
-			if (vif->dev == dev)
-				return true;
-		}
+		const struct net_device *vif_dev;
+		const struct vif_device *vif;
+
+		vif = &mrt->vif_table[ct];
+		vif_dev = rcu_access_pointer(vif->dev);
+		if (vif_dev && c->mfc_un.res.ttls[ct] < 255 &&
+		    vif_dev == dev)
+			return true;
 	}
 	return false;
 }
@@ -390,7 +399,6 @@ int mr_dump(struct net *net, struct notifier_block *nb, unsigned short family,
 			      struct netlink_ext_ack *extack),
 	    struct mr_table *(*mr_iter)(struct net *net,
 					struct mr_table *mrt),
-	    rwlock_t *mrt_lock,
 	    struct netlink_ext_ack *extack)
 {
 	struct mr_table *mrt;
@@ -402,22 +410,25 @@ int mr_dump(struct net *net, struct notifier_block *nb, unsigned short family,
 
 	for (mrt = mr_iter(net, NULL); mrt; mrt = mr_iter(net, mrt)) {
 		struct vif_device *v = &mrt->vif_table[0];
+		struct net_device *vif_dev;
 		struct mr_mfc *mfc;
 		int vifi;
 
 		/* Notifiy on table VIF entries */
-		read_lock(mrt_lock);
+		rcu_read_lock();
 		for (vifi = 0; vifi < mrt->maxvif; vifi++, v++) {
-			if (!v->dev)
+			vif_dev = rcu_dereference(v->dev);
+			if (!vif_dev)
 				continue;
 
 			err = mr_call_vif_notifier(nb, family,
-						   FIB_EVENT_VIF_ADD,
-						   v, vifi, mrt->id, extack);
+						   FIB_EVENT_VIF_ADD, v,
+						   vif_dev, vifi,
+						   mrt->id, extack);
 			if (err)
 				break;
 		}
-		read_unlock(mrt_lock);
+		rcu_read_unlock();
 
 		if (err)
 			return err;