Merge branch 'tcp_metrics_saddr'

Christoph Paasch says:

====================
Make tcp-metrics source-address aware

Currently tcp-metrics only stores per-destination addresses. This brings
problems, when a host has multiple interfaces (e.g., a smartphone having
WiFi/3G):

For example, a host contacting a server over WiFi will store the tcp-metrics
per destination IP. If then the host contacts the same server over 3G, the
same tcp-metrics will be used, although the path-characteristics are completly
different (e.g., the ssthresh is probably not the same).

In case of TFO this is not a problem, as the server will provide us a new cookie
once he saw our SYN+DATA with an incorrect cookie.
It may be (in case of carrier-grade NAT), that we keep the same public IP but
have a different private IP. Thus, we better reuse the old cookie even if our
source-IP has changed. However, this scenario is probably very uncommon, as
carriers try to provide the same src-IP to the clients behind their CGN.

Patches 1 + 2 add the source-IP to the tcp metrics.

Patches 3 to 5 modify the netlink-api to support the source-IP. From now on,
when using the command "ip tcp_metrics delete address ADDRESS" all entries
which match this destination IP will be deleted.

Today's iproute2 will complain when doing "ip tcp_metrics flush PREFIX" if
several entries are present for the same destination-IP but with different
source-IPs:

root@client:~/test# ip tcp_metrics
10.2.1.2 age 3.640sec rtt 16250us rttvar 15000us cwnd 10
10.2.1.2 age 4.030sec rtt 18750us rttvar 15000us cwnd 10
root@client:~/test# ip tcp_metrics flush 10.2.1.2/16
Failed to send flush request
: No such process

Follow-up patches will modify iproute2 to handle this correctly and allow
specifying the source-IP in the get/del commands.

v2: Added the patch that allows to selectively get/del of tcp-metrics based
on src-IP and moved the patch that adds the new netlink attribute before
the other patches.
====================

Signed-off-by: David S. Miller <[e-mail]>

include/uapi/linux/tcp_metrics.h

···
 	TCP_METRICS_ATTR_FOPEN_SYN_DROPS,	/* u16, count of drops */
 	TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS,	/* msecs age */
 	TCP_METRICS_ATTR_FOPEN_COOKIE,		/* binary */
+	TCP_METRICS_ATTR_SADDR_IPV4,		/* u32 */
+	TCP_METRICS_ATTR_SADDR_IPV6,		/* binary */
 
 	__TCP_METRICS_ATTR_MAX,
 };

net/ipv4/tcp_metrics.c

···
 
 struct tcp_metrics_block {
 	struct tcp_metrics_block __rcu	*tcpm_next;
-	struct inetpeer_addr		tcpm_addr;
+	struct inetpeer_addr		tcpm_saddr;
+	struct inetpeer_addr		tcpm_daddr;
 	unsigned long			tcpm_stamp;
 	u32				tcpm_ts;
 	u32				tcpm_ts_stamp;
···
 }
 
 static struct tcp_metrics_block *tcpm_new(struct dst_entry *dst,
-					  struct inetpeer_addr *addr,
+					  struct inetpeer_addr *saddr,
+					  struct inetpeer_addr *daddr,
 					  unsigned int hash,
 					  bool reclaim)
 {
···
 		if (!tm)
 			goto out_unlock;
 	}
-	tm->tcpm_addr = *addr;
+	tm->tcpm_saddr = *saddr;
+	tm->tcpm_daddr = *daddr;
 
 	tcpm_suck_dst(tm, dst, true);
 
···
 	return NULL;
 }
 
-static struct tcp_metrics_block *__tcp_get_metrics(const struct inetpeer_addr *addr,
+static struct tcp_metrics_block *__tcp_get_metrics(const struct inetpeer_addr *saddr,
+						   const struct inetpeer_addr *daddr,
 						   struct net *net, unsigned int hash)
 {
 	struct tcp_metrics_block *tm;
···
 
 	for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm;
 	     tm = rcu_dereference(tm->tcpm_next)) {
-		if (addr_same(&tm->tcpm_addr, addr))
+		if (addr_same(&tm->tcpm_saddr, saddr) &&
+		    addr_same(&tm->tcpm_daddr, daddr))
 			break;
 		depth++;
 	}
···
 						       struct dst_entry *dst)
 {
 	struct tcp_metrics_block *tm;
-	struct inetpeer_addr addr;
+	struct inetpeer_addr saddr, daddr;
 	unsigned int hash;
 	struct net *net;
 
-	addr.family = req->rsk_ops->family;
-	switch (addr.family) {
+	saddr.family = req->rsk_ops->family;
+	daddr.family = req->rsk_ops->family;
+	switch (daddr.family) {
 	case AF_INET:
-		addr.addr.a4 = inet_rsk(req)->ir_rmt_addr;
-		hash = (__force unsigned int) addr.addr.a4;
+		saddr.addr.a4 = inet_rsk(req)->ir_loc_addr;
+		daddr.addr.a4 = inet_rsk(req)->ir_rmt_addr;
+		hash = (__force unsigned int) daddr.addr.a4;
 		break;
 #if IS_ENABLED(CONFIG_IPV6)
 	case AF_INET6:
-		*(struct in6_addr *)addr.addr.a6 = inet_rsk(req)->ir_v6_rmt_addr;
+		*(struct in6_addr *)saddr.addr.a6 = inet_rsk(req)->ir_v6_loc_addr;
+		*(struct in6_addr *)daddr.addr.a6 = inet_rsk(req)->ir_v6_rmt_addr;
 		hash = ipv6_addr_hash(&inet_rsk(req)->ir_v6_rmt_addr);
 		break;
 #endif
···
 
 	for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm;
 	     tm = rcu_dereference(tm->tcpm_next)) {
-		if (addr_same(&tm->tcpm_addr, &addr))
+		if (addr_same(&tm->tcpm_saddr, &saddr) &&
+		    addr_same(&tm->tcpm_daddr, &daddr))
 			break;
 	}
 	tcpm_check_stamp(tm, dst);
···
 static struct tcp_metrics_block *__tcp_get_metrics_tw(struct inet_timewait_sock *tw)
 {
 	struct tcp_metrics_block *tm;
-	struct inetpeer_addr addr;
+	struct inetpeer_addr saddr, daddr;
 	unsigned int hash;
 	struct net *net;
 
-	addr.family = tw->tw_family;
-	switch (addr.family) {
+	saddr.family = tw->tw_family;
+	daddr.family = tw->tw_family;
+	switch (daddr.family) {
 	case AF_INET:
-		addr.addr.a4 = tw->tw_daddr;
-		hash = (__force unsigned int) addr.addr.a4;
+		saddr.addr.a4 = tw->tw_rcv_saddr;
+		daddr.addr.a4 = tw->tw_daddr;
+		hash = (__force unsigned int) daddr.addr.a4;
 		break;
 #if IS_ENABLED(CONFIG_IPV6)
 	case AF_INET6:
-		*(struct in6_addr *)addr.addr.a6 = tw->tw_v6_daddr;
+		*(struct in6_addr *)saddr.addr.a6 = tw->tw_v6_rcv_saddr;
+		*(struct in6_addr *)daddr.addr.a6 = tw->tw_v6_daddr;
 		hash = ipv6_addr_hash(&tw->tw_v6_daddr);
 		break;
 #endif
···
 
 	for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm;
 	     tm = rcu_dereference(tm->tcpm_next)) {
-		if (addr_same(&tm->tcpm_addr, &addr))
+		if (addr_same(&tm->tcpm_saddr, &saddr) &&
+		    addr_same(&tm->tcpm_daddr, &daddr))
 			break;
 	}
 	return tm;
···
 						 bool create)
 {
 	struct tcp_metrics_block *tm;
-	struct inetpeer_addr addr;
+	struct inetpeer_addr saddr, daddr;
 	unsigned int hash;
 	struct net *net;
 	bool reclaim;
 
-	addr.family = sk->sk_family;
-	switch (addr.family) {
+	saddr.family = sk->sk_family;
+	daddr.family = sk->sk_family;
+	switch (daddr.family) {
 	case AF_INET:
-		addr.addr.a4 = inet_sk(sk)->inet_daddr;
-		hash = (__force unsigned int) addr.addr.a4;
+		saddr.addr.a4 = inet_sk(sk)->inet_saddr;
+		daddr.addr.a4 = inet_sk(sk)->inet_daddr;
+		hash = (__force unsigned int) daddr.addr.a4;
 		break;
 #if IS_ENABLED(CONFIG_IPV6)
 	case AF_INET6:
-		*(struct in6_addr *)addr.addr.a6 = sk->sk_v6_daddr;
+		*(struct in6_addr *)saddr.addr.a6 = sk->sk_v6_rcv_saddr;
+		*(struct in6_addr *)daddr.addr.a6 = sk->sk_v6_daddr;
 		hash = ipv6_addr_hash(&sk->sk_v6_daddr);
 		break;
 #endif
···
 	net = dev_net(dst->dev);
 	hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log);
 
-	tm = __tcp_get_metrics(&addr, net, hash);
+	tm = __tcp_get_metrics(&saddr, &daddr, net, hash);
 	reclaim = false;
 	if (tm == TCP_METRICS_RECLAIM_PTR) {
 		reclaim = true;
 		tm = NULL;
 	}
 	if (!tm && create)
-		tm = tcpm_new(dst, &addr, hash, reclaim);
+		tm = tcpm_new(dst, &saddr, &daddr, hash, reclaim);
 	else
 		tcpm_check_stamp(tm, dst);
 
···
 	struct nlattr *nest;
 	int i;
 
-	switch (tm->tcpm_addr.family) {
+	switch (tm->tcpm_daddr.family) {
 	case AF_INET:
 		if (nla_put_be32(msg, TCP_METRICS_ATTR_ADDR_IPV4,
-				tm->tcpm_addr.addr.a4) < 0)
+				tm->tcpm_daddr.addr.a4) < 0)
+			goto nla_put_failure;
+		if (nla_put_be32(msg, TCP_METRICS_ATTR_SADDR_IPV4,
+				tm->tcpm_saddr.addr.a4) < 0)
 			goto nla_put_failure;
 		break;
 	case AF_INET6:
 		if (nla_put(msg, TCP_METRICS_ATTR_ADDR_IPV6, 16,
-			    tm->tcpm_addr.addr.a6) < 0)
+			    tm->tcpm_daddr.addr.a6) < 0)
+			goto nla_put_failure;
+		if (nla_put(msg, TCP_METRICS_ATTR_SADDR_IPV6, 16,
+			    tm->tcpm_saddr.addr.a6) < 0)
 			goto nla_put_failure;
 		break;
 	default:
···
 	return skb->len;
 }
 
-static int parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr,
-			 unsigned int *hash, int optional)
+static int __parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr,
+			   unsigned int *hash, int optional, int v4, int v6)
 {
 	struct nlattr *a;
 
-	a = info->attrs[TCP_METRICS_ATTR_ADDR_IPV4];
+	a = info->attrs[v4];
 	if (a) {
 		addr->family = AF_INET;
 		addr->addr.a4 = nla_get_be32(a);
-		*hash = (__force unsigned int) addr->addr.a4;
+		if (hash)
+			*hash = (__force unsigned int) addr->addr.a4;
 		return 0;
 	}
-	a = info->attrs[TCP_METRICS_ATTR_ADDR_IPV6];
+	a = info->attrs[v6];
 	if (a) {
 		if (nla_len(a) != sizeof(struct in6_addr))
 			return -EINVAL;
 		addr->family = AF_INET6;
 		memcpy(addr->addr.a6, nla_data(a), sizeof(addr->addr.a6));
-		*hash = ipv6_addr_hash((struct in6_addr *) addr->addr.a6);
+		if (hash)
+			*hash = ipv6_addr_hash((struct in6_addr *) addr->addr.a6);
 		return 0;
 	}
 	return optional ? 1 : -EAFNOSUPPORT;
 }
 
+static int parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr,
+			 unsigned int *hash, int optional)
+{
+	return __parse_nl_addr(info, addr, hash, optional,
+			       TCP_METRICS_ATTR_ADDR_IPV4,
+			       TCP_METRICS_ATTR_ADDR_IPV6);
+}
+
+static int parse_nl_saddr(struct genl_info *info, struct inetpeer_addr *addr)
+{
+	return __parse_nl_addr(info, addr, NULL, 0,
+			       TCP_METRICS_ATTR_SADDR_IPV4,
+			       TCP_METRICS_ATTR_SADDR_IPV6);
+}
+
 static int tcp_metrics_nl_cmd_get(struct sk_buff *skb, struct genl_info *info)
 {
 	struct tcp_metrics_block *tm;
-	struct inetpeer_addr addr;
+	struct inetpeer_addr saddr, daddr;
 	unsigned int hash;
 	struct sk_buff *msg;
 	struct net *net = genl_info_net(info);
 	void *reply;
 	int ret;
+	bool src = true;
 
-	ret = parse_nl_addr(info, &addr, &hash, 0);
+	ret = parse_nl_addr(info, &daddr, &hash, 0);
 	if (ret < 0)
 		return ret;
+
+	ret = parse_nl_saddr(info, &saddr);
+	if (ret < 0)
+		src = false;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 	if (!msg)
···
 	rcu_read_lock();
 	for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm;
 	     tm = rcu_dereference(tm->tcpm_next)) {
-		if (addr_same(&tm->tcpm_addr, &addr)) {
+		if (addr_same(&tm->tcpm_daddr, &daddr) &&
+		    (!src || addr_same(&tm->tcpm_saddr, &saddr))) {
 			ret = tcp_metrics_fill_info(msg, tm);
 			break;
 		}
···
 static int tcp_metrics_nl_cmd_del(struct sk_buff *skb, struct genl_info *info)
 {
 	struct tcpm_hash_bucket *hb;
-	struct tcp_metrics_block *tm;
+	struct tcp_metrics_block *tm, *tmlist = NULL;
 	struct tcp_metrics_block __rcu **pp;
-	struct inetpeer_addr addr;
+	struct inetpeer_addr saddr, daddr;
 	unsigned int hash;
 	struct net *net = genl_info_net(info);
 	int ret;
+	bool src = true;
 
-	ret = parse_nl_addr(info, &addr, &hash, 1);
+	ret = parse_nl_addr(info, &daddr, &hash, 1);
 	if (ret < 0)
 		return ret;
 	if (ret > 0)
 		return tcp_metrics_flush_all(net);
+	ret = parse_nl_saddr(info, &saddr);
+	if (ret < 0)
+		src = false;
 
 	hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log);
 	hb = net->ipv4.tcp_metrics_hash + hash;
 	pp = &hb->chain;
 	spin_lock_bh(&tcp_metrics_lock);
-	for (tm = deref_locked_genl(*pp); tm;
-	     pp = &tm->tcpm_next, tm = deref_locked_genl(*pp)) {
-		if (addr_same(&tm->tcpm_addr, &addr)) {
+	for (tm = deref_locked_genl(*pp); tm; tm = deref_locked_genl(*pp)) {
+		if (addr_same(&tm->tcpm_daddr, &daddr) &&
+		    (!src || addr_same(&tm->tcpm_saddr, &saddr))) {
 			*pp = tm->tcpm_next;
-			break;
+			tm->tcpm_next = tmlist;
+			tmlist = tm;
+		} else {
+			pp = &tm->tcpm_next;
 		}
 	}
 	spin_unlock_bh(&tcp_metrics_lock);
-	if (!tm)
+	if (!tmlist)
 		return -ESRCH;
-	kfree_rcu(tm, rcu_head);
+	for (tm = tmlist; tm; tm = tmlist) {
+		tmlist = tm->tcpm_next;
+		kfree_rcu(tm, rcu_head);
+	}
 	return 0;
 }