[PATCH net-next v2 0/4] tcp: receive-side per route dctcp handling

[PATCH net-next v2 0/4] tcp: receive-side per route dctcp handling

[Daniel Borkmann]

Original cover letter:

  Currently, the following case doesn't use DCTCP, even if it should:

    - responder has f.e. cubic as system wide default
    - 'ip route congctl dctcp $src' was set

  Then, DCTCP is NOT used if a DCTCP sender attempts to connect from a
  host in the $src range: ECT(0) is set, but listen_sk is not dctcp, so
  we fail the INET_ECN_is_not_ect sanity check.

  We also have to examine the dst used for the SYN/ACK reply to make
  this case work.

  In order to minimize additional cost, store the 'ecn is must have'
  information is the dst_features field.

  The set targets -next instead of -net since this doesn't seem to be a
  serious bug and to give the change more soak time until it hits linus
  tree.

v1 -> v2:
 - Addressed Dave's feedback, not exposing any bits to user space
 - Added patch 3 to reject incorrect configurations
 - Rest as is, rebased and retested

Thanks!

Daniel Borkmann (3):
  net: fib6: reduce identation in ip6_convert_metrics
  fib, fib6: reject invalid feature bits
  tcp: use dctcp if enabled on the route to the initiator

Florian Westphal (1):
  net: fib: move metrics parsing to a helper

 include/net/dst.h              |  6 ++++
 include/net/tcp.h              |  2 +-
 include/uapi/linux/rtnetlink.h | 11 +++---
 net/core/rtnetlink.c           |  6 ++++
 net/ipv4/fib_semantics.c       | 77 ++++++++++++++++++++++++++----------------
 net/ipv4/tcp_cong.c            |  9 +++--
 net/ipv4/tcp_input.c           |  7 ++--
 net/ipv6/route.c               | 39 ++++++++++++---------
 8 files changed, 101 insertions(+), 56 deletions(-)

-- 
1.9.3

[PATCH net-next v2 1/4] net: fib: move metrics parsing to a helper

[Daniel Borkmann]

From: Florian Westphal <fw@strlen.de>

fib_create_info() is already quite large, so before adding more
code to the metrics section move that to a helper, similar to
ip6_convert_metrics.

Suggested-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Florian Westphal <fw@strlen.de>
---
 net/ipv4/fib_semantics.c | 71 ++++++++++++++++++++++++++++--------------------
 1 file changed, 41 insertions(+), 30 deletions(-)

diff --git a/net/ipv4/fib_semantics.c b/net/ipv4/fib_semantics.c
index 1b2d011..88afbae 100644
--- a/net/ipv4/fib_semantics.c
+++ b/net/ipv4/fib_semantics.c
@@ -876,6 +876,44 @@ static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
 	return true;
 }
 
+static int
+fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
+{
+	struct nlattr *nla;
+	int remaining;
+
+	if (!cfg->fc_mx)
+		return 0;
+
+	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
+		int type = nla_type(nla);
+		u32 val;
+
+		if (!type)
+			continue;
+		if (type > RTAX_MAX)
+			return -EINVAL;
+
+		if (type == RTAX_CC_ALGO) {
+			char tmp[TCP_CA_NAME_MAX];
+
+			nla_strlcpy(tmp, nla, sizeof(tmp));
+			val = tcp_ca_get_key_by_name(tmp);
+			if (val == TCP_CA_UNSPEC)
+				return -EINVAL;
+		} else {
+			val = nla_get_u32(nla);
+		}
+		if (type == RTAX_ADVMSS && val > 65535 - 40)
+			val = 65535 - 40;
+		if (type == RTAX_MTU && val > 65535 - 15)
+			val = 65535 - 15;
+		fi->fib_metrics[type - 1] = val;
+	}
+
+	return 0;
+}
+
 struct fib_info *fib_create_info(struct fib_config *cfg)
 {
 	int err;
@@ -948,36 +986,9 @@ struct fib_info *fib_create_info(struct fib_config *cfg)
 			goto failure;
 	} endfor_nexthops(fi)
 
-	if (cfg->fc_mx) {
-		struct nlattr *nla;
-		int remaining;
-
-		nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
-			int type = nla_type(nla);
-
-			if (type) {
-				u32 val;
-
-				if (type > RTAX_MAX)
-					goto err_inval;
-				if (type == RTAX_CC_ALGO) {
-					char tmp[TCP_CA_NAME_MAX];
-
-					nla_strlcpy(tmp, nla, sizeof(tmp));
-					val = tcp_ca_get_key_by_name(tmp);
-					if (val == TCP_CA_UNSPEC)
-						goto err_inval;
-				} else {
-					val = nla_get_u32(nla);
-				}
-				if (type == RTAX_ADVMSS && val > 65535 - 40)
-					val = 65535 - 40;
-				if (type == RTAX_MTU && val > 65535 - 15)
-					val = 65535 - 15;
-				fi->fib_metrics[type - 1] = val;
-			}
-		}
-	}
+	err = fib_convert_metrics(fi, cfg);
+	if (err)
+		goto failure;
 
 	if (cfg->fc_mp) {
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
-- 
1.9.3

[PATCH net-next v2 2/4] net: fib6: reduce identation in ip6_convert_metrics

[Daniel Borkmann]

Reduce the identation a bit, there's no need to artificically have
it increased.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
---
 net/ipv6/route.c | 32 ++++++++++++++++----------------
 1 file changed, 16 insertions(+), 16 deletions(-)

diff --git a/net/ipv6/route.c b/net/ipv6/route.c
index 308dd5f..0261b72 100644
--- a/net/ipv6/route.c
+++ b/net/ipv6/route.c
@@ -1711,26 +1711,26 @@ static int ip6_convert_metrics(struct mx6_config *mxc,
 
 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 		int type = nla_type(nla);
+		u32 val;
 
-		if (type) {
-			u32 val;
-
-			if (unlikely(type > RTAX_MAX))
-				goto err;
-			if (type == RTAX_CC_ALGO) {
-				char tmp[TCP_CA_NAME_MAX];
+		if (!type)
+			continue;
+		if (unlikely(type > RTAX_MAX))
+			goto err;
 
-				nla_strlcpy(tmp, nla, sizeof(tmp));
-				val = tcp_ca_get_key_by_name(tmp);
-				if (val == TCP_CA_UNSPEC)
-					goto err;
-			} else {
-				val = nla_get_u32(nla);
-			}
+		if (type == RTAX_CC_ALGO) {
+			char tmp[TCP_CA_NAME_MAX];
 
-			mp[type - 1] = val;
-			__set_bit(type - 1, mxc->mx_valid);
+			nla_strlcpy(tmp, nla, sizeof(tmp));
+			val = tcp_ca_get_key_by_name(tmp);
+			if (val == TCP_CA_UNSPEC)
+				goto err;
+		} else {
+			val = nla_get_u32(nla);
 		}
+
+		mp[type - 1] = val;
+		__set_bit(type - 1, mxc->mx_valid);
 	}
 
 	mxc->mx = mp;
-- 
1.9.3

[PATCH net-next v2 3/4] fib, fib6: reject invalid feature bits

[Daniel Borkmann]

Feature bits that are invalid should not be accepted by the kernel,
only the lower 4 bits may be configured, but not the remaining ones.
Even from these 4, 2 of them are unused.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
---
 include/uapi/linux/rtnetlink.h | 11 +++++++----
 net/ipv4/fib_semantics.c       |  2 ++
 net/ipv6/route.c               |  2 ++
 3 files changed, 11 insertions(+), 4 deletions(-)

diff --git a/include/uapi/linux/rtnetlink.h b/include/uapi/linux/rtnetlink.h
index 0d3d3cc..7020247 100644
--- a/include/uapi/linux/rtnetlink.h
+++ b/include/uapi/linux/rtnetlink.h
@@ -418,10 +418,13 @@ enum {
 
 #define RTAX_MAX (__RTAX_MAX - 1)
 
-#define RTAX_FEATURE_ECN	0x00000001
-#define RTAX_FEATURE_SACK	0x00000002
-#define RTAX_FEATURE_TIMESTAMP	0x00000004
-#define RTAX_FEATURE_ALLFRAG	0x00000008
+#define RTAX_FEATURE_ECN	(1 << 0)
+#define RTAX_FEATURE_SACK	(1 << 1)
+#define RTAX_FEATURE_TIMESTAMP	(1 << 2)
+#define RTAX_FEATURE_ALLFRAG	(1 << 3)
+
+#define RTAX_FEATURE_MASK	(RTAX_FEATURE_ECN | RTAX_FEATURE_SACK | \
+				 RTAX_FEATURE_TIMESTAMP | RTAX_FEATURE_ALLFRAG)
 
 struct rta_session {
 	__u8	proto;
diff --git a/net/ipv4/fib_semantics.c b/net/ipv4/fib_semantics.c
index 88afbae..115a08e 100644
--- a/net/ipv4/fib_semantics.c
+++ b/net/ipv4/fib_semantics.c
@@ -908,6 +908,8 @@ fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
 			val = 65535 - 40;
 		if (type == RTAX_MTU && val > 65535 - 15)
 			val = 65535 - 15;
+		if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
+			return -EINVAL;
 		fi->fib_metrics[type - 1] = val;
 	}
 
diff --git a/net/ipv6/route.c b/net/ipv6/route.c
index 0261b72..8771530 100644
--- a/net/ipv6/route.c
+++ b/net/ipv6/route.c
@@ -1728,6 +1728,8 @@ static int ip6_convert_metrics(struct mx6_config *mxc,
 		} else {
 			val = nla_get_u32(nla);
 		}
+		if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
+			goto err;
 
 		mp[type - 1] = val;
 		__set_bit(type - 1, mxc->mx_valid);
-- 
1.9.3

[PATCH net-next v2 4/4] tcp: use dctcp if enabled on the route to the initiator

[Daniel Borkmann]

Currently, the following case doesn't use DCTCP, even if it should:
A responder has f.e. Cubic as system wide default, but for a specific
route to the initiating host, DCTCP is being set in RTAX_CC_ALGO. The
initiating host then uses DCTCP as congestion control, but since the
initiator sets ECT(0), tcp_ecn_create_request() doesn't set ecn_ok,
and we have to fall back to Reno after 3WHS completes.

We were thinking on how to solve this in a minimal, non-intrusive
way without bloating tcp_ecn_create_request() needlessly: lets cache
the CA ecn option flag in RTAX_FEATURES. In other words, when ECT(0)
is set on the SYN packet, set ecn_ok=1 iff route RTAX_FEATURES
contains the unexposed (internal-only) DST_FEATURE_ECN_CA. This allows
to only do a single metric feature lookup inside tcp_ecn_create_request().

Joint work with Florian Westphal.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Florian Westphal <fw@strlen.de>
---
 include/net/dst.h        | 6 ++++++
 include/net/tcp.h        | 2 +-
 net/core/rtnetlink.c     | 6 ++++++
 net/ipv4/fib_semantics.c | 6 +++++-
 net/ipv4/tcp_cong.c      | 9 ++++++---
 net/ipv4/tcp_input.c     | 7 +++++--
 net/ipv6/route.c         | 9 +++++++--
 7 files changed, 36 insertions(+), 9 deletions(-)

diff --git a/include/net/dst.h b/include/net/dst.h
index 4c48016..9261d92 100644
--- a/include/net/dst.h
+++ b/include/net/dst.h
@@ -207,6 +207,12 @@ static inline void dst_metric_set(struct dst_entry *dst, int metric, u32 val)
 		p[metric-1] = val;
 }
 
+/* Kernel-internal feature bits that are unallocated in user space. */
+#define DST_FEATURE_ECN_CA	(1 << 31)
+
+#define DST_FEATURE_MASK	(DST_FEATURE_ECN_CA)
+#define DST_FEATURE_ECN_MASK	(DST_FEATURE_ECN_CA | RTAX_FEATURE_ECN)
+
 static inline u32
 dst_feature(const struct dst_entry *dst, u32 feature)
 {
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 4a7b039..0cab28c 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -888,7 +888,7 @@ void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
 extern struct tcp_congestion_ops tcp_reno;
 
 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
-u32 tcp_ca_get_key_by_name(const char *name);
+u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
 #ifdef CONFIG_INET
 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
 #else
diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
index 788ceed..a466821 100644
--- a/net/core/rtnetlink.c
+++ b/net/core/rtnetlink.c
@@ -678,6 +678,12 @@ int rtnetlink_put_metrics(struct sk_buff *skb, u32 *metrics)
 					continue;
 				if (nla_put_string(skb, i + 1, name))
 					goto nla_put_failure;
+			} else if (i == RTAX_FEATURES - 1) {
+				u32 user_features = metrics[i] & RTAX_FEATURE_MASK;
+
+				BUILD_BUG_ON(RTAX_FEATURE_MASK & DST_FEATURE_MASK);
+				if (nla_put_u32(skb, i + 1, user_features))
+					goto nla_put_failure;
 			} else {
 				if (nla_put_u32(skb, i + 1, metrics[i]))
 					goto nla_put_failure;
diff --git a/net/ipv4/fib_semantics.c b/net/ipv4/fib_semantics.c
index 115a08e..992a959 100644
--- a/net/ipv4/fib_semantics.c
+++ b/net/ipv4/fib_semantics.c
@@ -879,6 +879,7 @@ static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
 static int
 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
 {
+	bool ecn_ca = false;
 	struct nlattr *nla;
 	int remaining;
 
@@ -898,7 +899,7 @@ fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
 			char tmp[TCP_CA_NAME_MAX];
 
 			nla_strlcpy(tmp, nla, sizeof(tmp));
-			val = tcp_ca_get_key_by_name(tmp);
+			val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
 			if (val == TCP_CA_UNSPEC)
 				return -EINVAL;
 		} else {
@@ -913,6 +914,9 @@ fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
 		fi->fib_metrics[type - 1] = val;
 	}
 
+	if (ecn_ca)
+		fi->fib_metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
+
 	return 0;
 }
 
diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c
index a2ed23c..93c4dc3 100644
--- a/net/ipv4/tcp_cong.c
+++ b/net/ipv4/tcp_cong.c
@@ -114,16 +114,19 @@ void tcp_unregister_congestion_control(struct tcp_congestion_ops *ca)
 }
 EXPORT_SYMBOL_GPL(tcp_unregister_congestion_control);
 
-u32 tcp_ca_get_key_by_name(const char *name)
+u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca)
 {
 	const struct tcp_congestion_ops *ca;
-	u32 key;
+	u32 key = TCP_CA_UNSPEC;
 
 	might_sleep();
 
 	rcu_read_lock();
 	ca = __tcp_ca_find_autoload(name);
-	key = ca ? ca->key : TCP_CA_UNSPEC;
+	if (ca) {
+		key = ca->key;
+		*ecn_ca = ca->flags & TCP_CONG_NEEDS_ECN;
+	}
 	rcu_read_unlock();
 
 	return key;
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index dc08e23..a8f515b 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -6003,14 +6003,17 @@ static void tcp_ecn_create_request(struct request_sock *req,
 	const struct net *net = sock_net(listen_sk);
 	bool th_ecn = th->ece && th->cwr;
 	bool ect, ecn_ok;
+	u32 ecn_ok_dst;
 
 	if (!th_ecn)
 		return;
 
 	ect = !INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield);
-	ecn_ok = net->ipv4.sysctl_tcp_ecn || dst_feature(dst, RTAX_FEATURE_ECN);
+	ecn_ok_dst = dst_feature(dst, DST_FEATURE_ECN_MASK);
+	ecn_ok = net->ipv4.sysctl_tcp_ecn || ecn_ok_dst;
 
-	if ((!ect && ecn_ok) || tcp_ca_needs_ecn(listen_sk))
+	if ((!ect && ecn_ok) || tcp_ca_needs_ecn(listen_sk) ||
+	    (ecn_ok_dst & DST_FEATURE_ECN_CA))
 		inet_rsk(req)->ecn_ok = 1;
 }
 
diff --git a/net/ipv6/route.c b/net/ipv6/route.c
index 8771530..f45cac6 100644
--- a/net/ipv6/route.c
+++ b/net/ipv6/route.c
@@ -1698,6 +1698,7 @@ out:
 static int ip6_convert_metrics(struct mx6_config *mxc,
 			       const struct fib6_config *cfg)
 {
+	bool ecn_ca = false;
 	struct nlattr *nla;
 	int remaining;
 	u32 *mp;
@@ -1722,7 +1723,7 @@ static int ip6_convert_metrics(struct mx6_config *mxc,
 			char tmp[TCP_CA_NAME_MAX];
 
 			nla_strlcpy(tmp, nla, sizeof(tmp));
-			val = tcp_ca_get_key_by_name(tmp);
+			val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
 			if (val == TCP_CA_UNSPEC)
 				goto err;
 		} else {
@@ -1735,8 +1736,12 @@ static int ip6_convert_metrics(struct mx6_config *mxc,
 		__set_bit(type - 1, mxc->mx_valid);
 	}
 
-	mxc->mx = mp;
+	if (ecn_ca) {
+		__set_bit(RTAX_FEATURES - 1, mxc->mx_valid);
+		mp[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
+	}
 
+	mxc->mx = mp;
 	return 0;
  err:
 	kfree(mp);
-- 
1.9.3

Re: [PATCH net-next v2 0/4] tcp: receive-side per route dctcp handling

[David Miller]

From: Daniel Borkmann <daniel@iogearbox.net>
Date: Mon, 31 Aug 2015 15:58:43 +0200

> Original cover letter:
> 
>   Currently, the following case doesn't use DCTCP, even if it should:
> 
>     - responder has f.e. cubic as system wide default
>     - 'ip route congctl dctcp $src' was set
> 
>   Then, DCTCP is NOT used if a DCTCP sender attempts to connect from a
>   host in the $src range: ECT(0) is set, but listen_sk is not dctcp, so
>   we fail the INET_ECN_is_not_ect sanity check.
> 
>   We also have to examine the dst used for the SYN/ACK reply to make
>   this case work.
> 
>   In order to minimize additional cost, store the 'ecn is must have'
>   information is the dst_features field.
> 
>   The set targets -next instead of -net since this doesn't seem to be a
>   serious bug and to give the change more soak time until it hits linus
>   tree.
> 
> v1 -> v2:
>  - Addressed Dave's feedback, not exposing any bits to user space
>  - Added patch 3 to reject incorrect configurations
>  - Rest as is, rebased and retested

Series applied, thanks Daniel.