Merge branch 'inet-implement-lockless-rtm_getnetconf-ops'

Eric Dumazet says:

====================
inet: implement lockless RTM_GETNETCONF ops

This series removes RTNL use for RTM_GETNETCONF operations on AF_INET.

- Annotate data-races to avoid possible KCSAN splats.

- "ip -4 netconf show dev XXX" can be implemented without RTNL [1]

- "ip -4 netconf" dumps can be implemented using RCU instead of RTNL [1]

[1] This only refers to RTM_GETNETCONF operation, "ip" command
    also uses RTM_GETLINK dumps which are using RTNL at this moment.
====================

Link: https://lore.kernel.org/r/20240227092411.2315725-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

include/linux/inetdevice.h

@@ -53,13 +53,15 @@ struct in_device {
 };
 
 #define IPV4_DEVCONF(cnf, attr) ((cnf).data[IPV4_DEVCONF_ ## attr - 1])
+#define IPV4_DEVCONF_RO(cnf, attr) READ_ONCE(IPV4_DEVCONF(cnf, attr))
 #define IPV4_DEVCONF_ALL(net, attr) \
 	IPV4_DEVCONF((*(net)->ipv4.devconf_all), attr)
+#define IPV4_DEVCONF_ALL_RO(net, attr) READ_ONCE(IPV4_DEVCONF_ALL(net, attr))
 
-static inline int ipv4_devconf_get(struct in_device *in_dev, int index)
+static inline int ipv4_devconf_get(const struct in_device *in_dev, int index)
 {
 	index--;
-	return in_dev->cnf.data[index];
+	return READ_ONCE(in_dev->cnf.data[index]);
 }
 
 static inline void ipv4_devconf_set(struct in_device *in_dev, int index,
@@ -67,7 +69,7 @@ static inline void ipv4_devconf_set(struct in_device *in_dev, int index,
 {
 	index--;
 	set_bit(index, in_dev->cnf.state);
-	in_dev->cnf.data[index] = val;
+	WRITE_ONCE(in_dev->cnf.data[index], val);
 }
 
 static inline void ipv4_devconf_setall(struct in_device *in_dev)
@@ -81,18 +83,18 @@ static inline void ipv4_devconf_setall(struct in_device *in_dev)
 	ipv4_devconf_set((in_dev), IPV4_DEVCONF_ ## attr, (val))
 
 #define IN_DEV_ANDCONF(in_dev, attr) \
-	(IPV4_DEVCONF_ALL(dev_net(in_dev->dev), attr) && \
+	(IPV4_DEVCONF_ALL_RO(dev_net(in_dev->dev), attr) && \
 	 IN_DEV_CONF_GET((in_dev), attr))
 
 #define IN_DEV_NET_ORCONF(in_dev, net, attr) \
-	(IPV4_DEVCONF_ALL(net, attr) || \
+	(IPV4_DEVCONF_ALL_RO(net, attr) || \
 	 IN_DEV_CONF_GET((in_dev), attr))
 
 #define IN_DEV_ORCONF(in_dev, attr) \
 	IN_DEV_NET_ORCONF(in_dev, dev_net(in_dev->dev), attr)
 
 #define IN_DEV_MAXCONF(in_dev, attr) \
-	(max(IPV4_DEVCONF_ALL(dev_net(in_dev->dev), attr), \
+	(max(IPV4_DEVCONF_ALL_RO(dev_net(in_dev->dev), attr), \
 	     IN_DEV_CONF_GET((in_dev), attr)))
 
 #define IN_DEV_FORWARD(in_dev)		IN_DEV_CONF_GET((in_dev), FORWARDING)

net/ipv4/devinet.c

@@ -1982,7 +1982,7 @@ static int inet_fill_link_af(struct sk_buff *skb, const struct net_device *dev,
 		return -EMSGSIZE;
 
 	for (i = 0; i < IPV4_DEVCONF_MAX; i++)
-		((u32 *) nla_data(nla))[i] = in_dev->cnf.data[i];
+		((u32 *) nla_data(nla))[i] = READ_ONCE(in_dev->cnf.data[i]);
 
 	return 0;
 }
@@ -2068,9 +2068,9 @@ static int inet_netconf_msgsize_devconf(int type)
 }
 
 static int inet_netconf_fill_devconf(struct sk_buff *skb, int ifindex,
-				     struct ipv4_devconf *devconf, u32 portid,
+				     const struct ipv4_devconf *devconf,
-				     u32 seq, int event, unsigned int flags,
+				     u32 portid, u32 seq, int event,
-				     int type)
+				     unsigned int flags, int type)
 {
 	struct nlmsghdr  *nlh;
 	struct netconfmsg *ncm;
@@ -2095,27 +2095,28 @@ static int inet_netconf_fill_devconf(struct sk_buff *skb, int ifindex,
 
 	if ((all || type == NETCONFA_FORWARDING) &&
 	    nla_put_s32(skb, NETCONFA_FORWARDING,
-			IPV4_DEVCONF(*devconf, FORWARDING)) < 0)
+			IPV4_DEVCONF_RO(*devconf, FORWARDING)) < 0)
 		goto nla_put_failure;
 	if ((all || type == NETCONFA_RP_FILTER) &&
 	    nla_put_s32(skb, NETCONFA_RP_FILTER,
-			IPV4_DEVCONF(*devconf, RP_FILTER)) < 0)
+			IPV4_DEVCONF_RO(*devconf, RP_FILTER)) < 0)
 		goto nla_put_failure;
 	if ((all || type == NETCONFA_MC_FORWARDING) &&
 	    nla_put_s32(skb, NETCONFA_MC_FORWARDING,
-			IPV4_DEVCONF(*devconf, MC_FORWARDING)) < 0)
+			IPV4_DEVCONF_RO(*devconf, MC_FORWARDING)) < 0)
 		goto nla_put_failure;
 	if ((all || type == NETCONFA_BC_FORWARDING) &&
 	    nla_put_s32(skb, NETCONFA_BC_FORWARDING,
-			IPV4_DEVCONF(*devconf, BC_FORWARDING)) < 0)
+			IPV4_DEVCONF_RO(*devconf, BC_FORWARDING)) < 0)
 		goto nla_put_failure;
 	if ((all || type == NETCONFA_PROXY_NEIGH) &&
 	    nla_put_s32(skb, NETCONFA_PROXY_NEIGH,
-			IPV4_DEVCONF(*devconf, PROXY_ARP)) < 0)
+			IPV4_DEVCONF_RO(*devconf, PROXY_ARP)) < 0)
 		goto nla_put_failure;
 	if ((all || type == NETCONFA_IGNORE_ROUTES_WITH_LINKDOWN) &&
 	    nla_put_s32(skb, NETCONFA_IGNORE_ROUTES_WITH_LINKDOWN,
-			IPV4_DEVCONF(*devconf, IGNORE_ROUTES_WITH_LINKDOWN)) < 0)
+			IPV4_DEVCONF_RO(*devconf,
+					IGNORE_ROUTES_WITH_LINKDOWN)) < 0)
 		goto nla_put_failure;
 
 out:
@@ -2204,21 +2205,20 @@ static int inet_netconf_get_devconf(struct sk_buff *in_skb,
 				    struct netlink_ext_ack *extack)
 {
 	struct net *net = sock_net(in_skb->sk);
-	struct nlattr *tb[NETCONFA_MAX+1];
+	struct nlattr *tb[NETCONFA_MAX + 1];
+	const struct ipv4_devconf *devconf;
+	struct in_device *in_dev = NULL;
+	struct net_device *dev = NULL;
 	struct sk_buff *skb;
-	struct ipv4_devconf *devconf;
-	struct in_device *in_dev;
-	struct net_device *dev;
 	int ifindex;
 	int err;
 
 	err = inet_netconf_valid_get_req(in_skb, nlh, tb, extack);
 	if (err)
-		goto errout;
+		return err;
 
-	err = -EINVAL;
 	if (!tb[NETCONFA_IFINDEX])
-		goto errout;
+		return -EINVAL;
 
 	ifindex = nla_get_s32(tb[NETCONFA_IFINDEX]);
 	switch (ifindex) {
@@ -2229,10 +2229,10 @@ static int inet_netconf_get_devconf(struct sk_buff *in_skb,
 		devconf = net->ipv4.devconf_dflt;
 		break;
 	default:
-		dev = __dev_get_by_index(net, ifindex);
+		err = -ENODEV;
-		if (!dev)
+		dev = dev_get_by_index(net, ifindex);
-			goto errout;
+		if (dev)
-		in_dev = __in_dev_get_rtnl(dev);
+			in_dev = in_dev_get(dev);
 		if (!in_dev)
 			goto errout;
 		devconf = &in_dev->cnf;
@@ -2256,6 +2256,9 @@ static int inet_netconf_get_devconf(struct sk_buff *in_skb,
 	}
 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
 errout:
+	if (in_dev)
+		in_dev_put(in_dev);
+	dev_put(dev);
 	return err;
 }
 
@@ -2264,11 +2267,13 @@ static int inet_netconf_dump_devconf(struct sk_buff *skb,
 {
 	const struct nlmsghdr *nlh = cb->nlh;
 	struct net *net = sock_net(skb->sk);
-	int h, s_h;
+	struct {
-	int idx, s_idx;
+		unsigned long ifindex;
+		unsigned int all_default;
+	} *ctx = (void *)cb->ctx;
+	const struct in_device *in_dev;
 	struct net_device *dev;
-	struct in_device *in_dev;
+	int err = 0;
-	struct hlist_head *head;
 
 	if (cb->strict_check) {
 		struct netlink_ext_ack *extack = cb->extack;
@@ -2285,64 +2290,47 @@ static int inet_netconf_dump_devconf(struct sk_buff *skb,
 		}
 	}
 
-	s_h = cb->args[0];
+	rcu_read_lock();
-	s_idx = idx = cb->args[1];
+	for_each_netdev_dump(net, dev, ctx->ifindex) {
-
+		in_dev = __in_dev_get_rcu(dev);
-	for (h = s_h; h < NETDEV_HASHENTRIES; h++, s_idx = 0) {
+		if (!in_dev)
-		idx = 0;
+			continue;
-		head = &net->dev_index_head[h];
+		err = inet_netconf_fill_devconf(skb, dev->ifindex,
-		rcu_read_lock();
+						&in_dev->cnf,
-		cb->seq = inet_base_seq(net);
+						NETLINK_CB(cb->skb).portid,
-		hlist_for_each_entry_rcu(dev, head, index_hlist) {
+						nlh->nlmsg_seq,
-			if (idx < s_idx)
+						RTM_NEWNETCONF, NLM_F_MULTI,
-				goto cont;
+						NETCONFA_ALL);
-			in_dev = __in_dev_get_rcu(dev);
+		if (err < 0)
-			if (!in_dev)
-				goto cont;
-
-			if (inet_netconf_fill_devconf(skb, dev->ifindex,
-						      &in_dev->cnf,
-						      NETLINK_CB(cb->skb).portid,
-						      nlh->nlmsg_seq,
-						      RTM_NEWNETCONF,
-						      NLM_F_MULTI,
-						      NETCONFA_ALL) < 0) {
-				rcu_read_unlock();
-				goto done;
-			}
-			nl_dump_check_consistent(cb, nlmsg_hdr(skb));
-cont:
-			idx++;
-		}
-		rcu_read_unlock();
-	}
-	if (h == NETDEV_HASHENTRIES) {
-		if (inet_netconf_fill_devconf(skb, NETCONFA_IFINDEX_ALL,
-					      net->ipv4.devconf_all,
-					      NETLINK_CB(cb->skb).portid,
-					      nlh->nlmsg_seq,
-					      RTM_NEWNETCONF, NLM_F_MULTI,
-					      NETCONFA_ALL) < 0)
 			goto done;
-		else
-			h++;
 	}
-	if (h == NETDEV_HASHENTRIES + 1) {
+	if (ctx->all_default == 0) {
-		if (inet_netconf_fill_devconf(skb, NETCONFA_IFINDEX_DEFAULT,
+		err = inet_netconf_fill_devconf(skb, NETCONFA_IFINDEX_ALL,
-					      net->ipv4.devconf_dflt,
+						net->ipv4.devconf_all,
-					      NETLINK_CB(cb->skb).portid,
+						NETLINK_CB(cb->skb).portid,
-					      nlh->nlmsg_seq,
+						nlh->nlmsg_seq,
-					      RTM_NEWNETCONF, NLM_F_MULTI,
+						RTM_NEWNETCONF, NLM_F_MULTI,
-					      NETCONFA_ALL) < 0)
+						NETCONFA_ALL);
+		if (err < 0)
 			goto done;
-		else
+		ctx->all_default++;
-			h++;
+	}
+	if (ctx->all_default == 1) {
+		err = inet_netconf_fill_devconf(skb, NETCONFA_IFINDEX_DEFAULT,
+						net->ipv4.devconf_dflt,
+						NETLINK_CB(cb->skb).portid,
+						nlh->nlmsg_seq,
+						RTM_NEWNETCONF, NLM_F_MULTI,
+						NETCONFA_ALL);
+		if (err < 0)
+			goto done;
+		ctx->all_default++;
 	}
 done:
-	cb->args[0] = h;
+	if (err < 0 && likely(skb->len))
-	cb->args[1] = idx;
+		err = skb->len;
-
+	rcu_read_unlock();
-	return skb->len;
+	return err;
 }
 
 #ifdef CONFIG_SYSCTL
@@ -2825,5 +2813,6 @@ void __init devinet_init(void)
 	rtnl_register(PF_INET, RTM_DELADDR, inet_rtm_deladdr, NULL, 0);
 	rtnl_register(PF_INET, RTM_GETADDR, NULL, inet_dump_ifaddr, 0);
 	rtnl_register(PF_INET, RTM_GETNETCONF, inet_netconf_get_devconf,
-		      inet_netconf_dump_devconf, 0);
+		      inet_netconf_dump_devconf,
+		      RTNL_FLAG_DOIT_UNLOCKED | RTNL_FLAG_DUMP_UNLOCKED);
 }

net/ipv4/igmp.c

@@ -120,12 +120,12 @@
  */
 
 #define IGMP_V1_SEEN(in_dev) \
-	(IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 1 || \
+	(IPV4_DEVCONF_ALL_RO(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 1 || \
 	 IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 1 || \
 	 ((in_dev)->mr_v1_seen && \
 	  time_before(jiffies, (in_dev)->mr_v1_seen)))
 #define IGMP_V2_SEEN(in_dev) \
-	(IPV4_DEVCONF_ALL(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 2 || \
+	(IPV4_DEVCONF_ALL_RO(dev_net(in_dev->dev), FORCE_IGMP_VERSION) == 2 || \
 	 IN_DEV_CONF_GET((in_dev), FORCE_IGMP_VERSION) == 2 || \
 	 ((in_dev)->mr_v2_seen && \
 	  time_before(jiffies, (in_dev)->mr_v2_seen)))

net/ipv4/proc.c

@@ -395,7 +395,7 @@ static int snmp_seq_show_ipstats(struct seq_file *seq, void *v)
 		seq_printf(seq, " %s", snmp4_ipstats_list[i].name);
 
 	seq_printf(seq, "\nIp: %d %d",
-		   IPV4_DEVCONF_ALL(net, FORWARDING) ? 1 : 2,
+		   IPV4_DEVCONF_ALL_RO(net, FORWARDING) ? 1 : 2,
 		   READ_ONCE(net->ipv4.sysctl_ip_default_ttl));
 
 	BUILD_BUG_ON(offsetof(struct ipstats_mib, mibs) != 0);

net/ipv4/route.c

@@ -2313,7 +2313,7 @@ static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
 		if (IN_DEV_BFORWARD(in_dev))
 			goto make_route;
 		/* not do cache if bc_forwarding is enabled */
-		if (IPV4_DEVCONF_ALL(net, BC_FORWARDING))
+		if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING))
 			do_cache = false;
 		goto brd_input;
 	}
@@ -2993,7 +2993,7 @@ static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
 #ifdef CONFIG_IP_MROUTE
 			if (ipv4_is_multicast(dst) &&
 			    !ipv4_is_local_multicast(dst) &&
-			    IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
+			    IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) {
 				int err = ipmr_get_route(net, skb,
 							 fl4->saddr, fl4->daddr,
 							 r, portid);