[PATCH net-next v3 00/11] Multipath TCP: Prerequisites

[PATCH net-next v3 00/11] Multipath TCP: Prerequisites

[Mat Martineau]

v2 -> v3: Ensure sk_type alignment in struct sock (patch 2)

v1 -> v2: sk_pacing_shift left as a regular struct member (patch 2), and
modified SACK space check based on recent -net fix (patch 9).


The MPTCP upstreaming community has been collaborating on an
upstreamable MPTCP implementation that complies with RFC 8684. A minimal
set of features to comply with the specification involves a sizeable set
of code changes, so David requested that we split this work in to
multiple, smaller patch sets to build up MPTCP infrastructure.

The minimal MPTCP feature set we are proposing for review in the v5.6
timeframe begins with these three parts:

Part 1 (this patch set): MPTCP prerequisites. Introduce some MPTCP
definitions, additional ULP and skb extension features, TCP option space
checking, and a few exported symbols.

Part 2: Single subflow implementation and self tests.

Part 3: Switch from MPTCP v0 (RFC 6824) to MPTCP v1 (new RFC 8684,
publication expected in the next few days).

We plan to send those over the next week. Additional patches for
multiple subflow support, path management, active backup, and other
features are in the pipeline for submission after making progress with
the above reviews.

Clone/fetch:
https://github.com/multipath-tcp/mptcp_net-next.git (tag: netdev-v3-part1)

Browse:
https://github.com/multipath-tcp/mptcp_net-next/tree/netdev-v3-part1

Thank you for your review. You can find us at mptcp@lists.01.org and
https://is.gd/mptcp_upstream


Mat Martineau (9):
  net: Make sock protocol value checks more specific
  sock: Make sk_protocol a 16-bit value
  tcp: Define IPPROTO_MPTCP
  tcp: Add MPTCP option number
  tcp, ulp: Add clone operation to tcp_ulp_ops
  mptcp: Add MPTCP to skb extensions
  tcp: Prevent coalesce/collapse when skb has MPTCP extensions
  tcp: Export TCP functions and ops struct
  tcp: Check for filled TCP option space before SACK

Paolo Abeni (2):
  tcp: clean ext on tx recycle
  skb: add helpers to allocate ext independently from sk_buff

 MAINTAINERS                     | 10 ++++++++
 include/linux/skbuff.h          |  6 +++++
 include/net/mptcp.h             | 43 +++++++++++++++++++++++++++++++++
 include/net/sock.h              | 12 ++++-----
 include/net/tcp.h               | 22 +++++++++++++++++
 include/trace/events/sock.h     |  5 ++--
 include/uapi/linux/in.h         |  2 ++
 net/ax25/af_ax25.c              |  2 +-
 net/core/skbuff.c               | 42 ++++++++++++++++++++++++++++++--
 net/decnet/af_decnet.c          |  2 +-
 net/ipv4/inet_connection_sock.c |  2 ++
 net/ipv4/tcp.c                  |  6 ++---
 net/ipv4/tcp_input.c            | 10 ++++++--
 net/ipv4/tcp_ipv4.c             |  2 +-
 net/ipv4/tcp_output.c           | 12 ++++++---
 net/ipv4/tcp_ulp.c              | 12 +++++++++
 net/ipv6/tcp_ipv6.c             |  6 ++---
 tools/include/uapi/linux/in.h   |  2 ++
 18 files changed, 173 insertions(+), 25 deletions(-)
 create mode 100644 include/net/mptcp.h

-- 
2.24.1

[PATCH net-next v3 01/11] net: Make sock protocol value checks more specific

[Mat Martineau]

SK_PROTOCOL_MAX is only used in two places, for DECNet and AX.25. The
limits have more to do with the those protocol definitions than they do
with the data type of sk_protocol, so remove SK_PROTOCOL_MAX and use
U8_MAX directly.

Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
---
 include/net/sock.h     | 1 -
 net/ax25/af_ax25.c     | 2 +-
 net/decnet/af_decnet.c | 2 +-
 3 files changed, 2 insertions(+), 3 deletions(-)

diff --git a/include/net/sock.h b/include/net/sock.h
index 87d54ef57f00..81dc811aad2e 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -458,7 +458,6 @@ struct sock {
 				sk_userlocks : 4,
 				sk_protocol  : 8,
 				sk_type      : 16;
-#define SK_PROTOCOL_MAX U8_MAX
 	u16			sk_gso_max_segs;
 	u8			sk_pacing_shift;
 	unsigned long	        sk_lingertime;
diff --git a/net/ax25/af_ax25.c b/net/ax25/af_ax25.c
index 324306d6fde0..ff57ea89c27e 100644
--- a/net/ax25/af_ax25.c
+++ b/net/ax25/af_ax25.c
@@ -808,7 +808,7 @@ static int ax25_create(struct net *net, struct socket *sock, int protocol,
 	struct sock *sk;
 	ax25_cb *ax25;
 
-	if (protocol < 0 || protocol > SK_PROTOCOL_MAX)
+	if (protocol < 0 || protocol > U8_MAX)
 		return -EINVAL;
 
 	if (!net_eq(net, &init_net))
diff --git a/net/decnet/af_decnet.c b/net/decnet/af_decnet.c
index e19a92a62e14..0a46ea3bddd5 100644
--- a/net/decnet/af_decnet.c
+++ b/net/decnet/af_decnet.c
@@ -670,7 +670,7 @@ static int dn_create(struct net *net, struct socket *sock, int protocol,
 {
 	struct sock *sk;
 
-	if (protocol < 0 || protocol > SK_PROTOCOL_MAX)
+	if (protocol < 0 || protocol > U8_MAX)
 		return -EINVAL;
 
 	if (!net_eq(net, &init_net))
-- 
2.24.1

[PATCH net-next v3 02/11] sock: Make sk_protocol a 16-bit value

[Mat Martineau]

Match the 16-bit width of skbuff->protocol. Fills an 8-bit hole so
sizeof(struct sock) does not change.

v2 -> v3:
 - keep 'sk_type' 2 bytes aligned (Eric)
v1 -> v2:
 - preserve sk_pacing_shift as bit field (Eric)

Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Co-developed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
---
 include/net/sock.h          | 10 +++++-----
 include/trace/events/sock.h |  2 +-
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/include/net/sock.h b/include/net/sock.h
index 81dc811aad2e..b93cadba1a3b 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -451,15 +451,15 @@ struct sock {
 #define SK_FL_TYPE_MASK    0xffff0000
 #endif
 
-	unsigned int		sk_padding : 1,
+	u8			sk_padding : 1,
 				sk_kern_sock : 1,
 				sk_no_check_tx : 1,
 				sk_no_check_rx : 1,
-				sk_userlocks : 4,
-				sk_protocol  : 8,
-				sk_type      : 16;
-	u16			sk_gso_max_segs;
+				sk_userlocks : 4;
 	u8			sk_pacing_shift;
+	u16			sk_type;
+	u16			sk_protocol;
+	u16			sk_gso_max_segs;
 	unsigned long	        sk_lingertime;
 	struct proto		*sk_prot_creator;
 	rwlock_t		sk_callback_lock;
diff --git a/include/trace/events/sock.h b/include/trace/events/sock.h
index 51fe9f6719eb..3ff12b90048d 100644
--- a/include/trace/events/sock.h
+++ b/include/trace/events/sock.h
@@ -147,7 +147,7 @@ TRACE_EVENT(inet_sock_set_state,
 		__field(__u16, sport)
 		__field(__u16, dport)
 		__field(__u16, family)
-		__field(__u8, protocol)
+		__field(__u16, protocol)
 		__array(__u8, saddr, 4)
 		__array(__u8, daddr, 4)
 		__array(__u8, saddr_v6, 16)
-- 
2.24.1

[PATCH net-next v3 03/11] tcp: Define IPPROTO_MPTCP

[Mat Martineau]

To open a MPTCP socket with socket(AF_INET, SOCK_STREAM, IPPROTO_MPTCP),
IPPROTO_MPTCP needs a value that differs from IPPROTO_TCP. The existing
IPPROTO numbers mostly map directly to IANA-specified protocol numbers.
MPTCP does not have a protocol number allocated because MPTCP packets
use the TCP protocol number. Use private number not used OTA.

Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
---
 include/trace/events/sock.h   | 3 ++-
 include/uapi/linux/in.h       | 2 ++
 tools/include/uapi/linux/in.h | 2 ++
 3 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/include/trace/events/sock.h b/include/trace/events/sock.h
index 3ff12b90048d..a966d4b5ab37 100644
--- a/include/trace/events/sock.h
+++ b/include/trace/events/sock.h
@@ -19,7 +19,8 @@
 #define inet_protocol_names		\
 		EM(IPPROTO_TCP)			\
 		EM(IPPROTO_DCCP)		\
-		EMe(IPPROTO_SCTP)
+		EM(IPPROTO_SCTP)		\
+		EMe(IPPROTO_MPTCP)
 
 #define tcp_state_names			\
 		EM(TCP_ESTABLISHED)		\
diff --git a/include/uapi/linux/in.h b/include/uapi/linux/in.h
index e7ad9d350a28..1521073b6348 100644
--- a/include/uapi/linux/in.h
+++ b/include/uapi/linux/in.h
@@ -76,6 +76,8 @@ enum {
 #define IPPROTO_MPLS		IPPROTO_MPLS
   IPPROTO_RAW = 255,		/* Raw IP packets			*/
 #define IPPROTO_RAW		IPPROTO_RAW
+  IPPROTO_MPTCP = 262,		/* Multipath TCP connection		*/
+#define IPPROTO_MPTCP		IPPROTO_MPTCP
   IPPROTO_MAX
 };
 #endif
diff --git a/tools/include/uapi/linux/in.h b/tools/include/uapi/linux/in.h
index e7ad9d350a28..1521073b6348 100644
--- a/tools/include/uapi/linux/in.h
+++ b/tools/include/uapi/linux/in.h
@@ -76,6 +76,8 @@ enum {
 #define IPPROTO_MPLS		IPPROTO_MPLS
   IPPROTO_RAW = 255,		/* Raw IP packets			*/
 #define IPPROTO_RAW		IPPROTO_RAW
+  IPPROTO_MPTCP = 262,		/* Multipath TCP connection		*/
+#define IPPROTO_MPTCP		IPPROTO_MPTCP
   IPPROTO_MAX
 };
 #endif
-- 
2.24.1

[PATCH net-next v3 04/11] tcp: Add MPTCP option number

[Mat Martineau]

TCP option 30 is allocated for MPTCP by the IANA.

Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
---
 include/net/tcp.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index 86b9a8766648..d4b6bf2c5d3c 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -182,6 +182,7 @@ void tcp_time_wait(struct sock *sk, int state, int timeo);
 #define TCPOPT_SACK             5       /* SACK Block */
 #define TCPOPT_TIMESTAMP	8	/* Better RTT estimations/PAWS */
 #define TCPOPT_MD5SIG		19	/* MD5 Signature (RFC2385) */
+#define TCPOPT_MPTCP		30	/* Multipath TCP (RFC6824) */
 #define TCPOPT_FASTOPEN		34	/* Fast open (RFC7413) */
 #define TCPOPT_EXP		254	/* Experimental */
 /* Magic number to be after the option value for sharing TCP
-- 
2.24.1

[PATCH net-next v3 05/11] tcp, ulp: Add clone operation to tcp_ulp_ops

[Mat Martineau]

If ULP is used on a listening socket, icsk_ulp_ops and icsk_ulp_data are
copied when the listener is cloned. Sometimes the clone is immediately
deleted, which will invoke the release op on the clone and likely
corrupt the listening socket's icsk_ulp_data.

The clone operation is invoked immediately after the clone is copied and
gives the ULP type an opportunity to set up the clone socket and its
icsk_ulp_data.

Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
---
 include/net/tcp.h               |  5 +++++
 net/ipv4/inet_connection_sock.c |  2 ++
 net/ipv4/tcp_ulp.c              | 12 ++++++++++++
 3 files changed, 19 insertions(+)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index d4b6bf2c5d3c..c82b2f75d024 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -2145,6 +2145,9 @@ struct tcp_ulp_ops {
 	/* diagnostic */
 	int (*get_info)(const struct sock *sk, struct sk_buff *skb);
 	size_t (*get_info_size)(const struct sock *sk);
+	/* clone ulp */
+	void (*clone)(const struct request_sock *req, struct sock *newsk,
+		      const gfp_t priority);
 
 	char		name[TCP_ULP_NAME_MAX];
 	struct module	*owner;
@@ -2155,6 +2158,8 @@ int tcp_set_ulp(struct sock *sk, const char *name);
 void tcp_get_available_ulp(char *buf, size_t len);
 void tcp_cleanup_ulp(struct sock *sk);
 void tcp_update_ulp(struct sock *sk, struct proto *p);
+void tcp_clone_ulp(const struct request_sock *req,
+		   struct sock *newsk, const gfp_t priority);
 
 #define MODULE_ALIAS_TCP_ULP(name)				\
 	__MODULE_INFO(alias, alias_userspace, name);		\
diff --git a/net/ipv4/inet_connection_sock.c b/net/ipv4/inet_connection_sock.c
index e4c6e8b40490..d667f2569f8e 100644
--- a/net/ipv4/inet_connection_sock.c
+++ b/net/ipv4/inet_connection_sock.c
@@ -810,6 +810,8 @@ struct sock *inet_csk_clone_lock(const struct sock *sk,
 		/* Deinitialize accept_queue to trap illegal accesses. */
 		memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 
+		tcp_clone_ulp(req, newsk, priority);
+
 		security_inet_csk_clone(newsk, req);
 	}
 	return newsk;
diff --git a/net/ipv4/tcp_ulp.c b/net/ipv4/tcp_ulp.c
index 12ab5db2b71c..e7a2589d69ee 100644
--- a/net/ipv4/tcp_ulp.c
+++ b/net/ipv4/tcp_ulp.c
@@ -130,6 +130,18 @@ void tcp_cleanup_ulp(struct sock *sk)
 	icsk->icsk_ulp_ops = NULL;
 }
 
+void tcp_clone_ulp(const struct request_sock *req, struct sock *newsk,
+		   const gfp_t priority)
+{
+	struct inet_connection_sock *icsk = inet_csk(newsk);
+
+	if (!icsk->icsk_ulp_ops)
+		return;
+
+	if (icsk->icsk_ulp_ops->clone)
+		icsk->icsk_ulp_ops->clone(req, newsk, priority);
+}
+
 static int __tcp_set_ulp(struct sock *sk, const struct tcp_ulp_ops *ulp_ops)
 {
 	struct inet_connection_sock *icsk = inet_csk(sk);
-- 
2.24.1

[PATCH net-next v3 06/11] mptcp: Add MPTCP to skb extensions

[Mat Martineau]

Add enum value for MPTCP and update config dependencies

Co-developed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
---
 MAINTAINERS            | 10 ++++++++++
 include/linux/skbuff.h |  3 +++
 include/net/mptcp.h    | 27 +++++++++++++++++++++++++++
 net/core/skbuff.c      |  7 +++++++
 4 files changed, 47 insertions(+)
 create mode 100644 include/net/mptcp.h

diff --git a/MAINTAINERS b/MAINTAINERS
index a28c77ee6b0d..132eb3f9fbb7 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -11568,6 +11568,16 @@ F:	net/ipv6/calipso.c
 F:	net/netfilter/xt_CONNSECMARK.c
 F:	net/netfilter/xt_SECMARK.c
 
+NETWORKING [MPTCP]
+M:	Mat Martineau <mathew.j.martineau@linux.intel.com>
+M:	Matthieu Baerts <matthieu.baerts@tessares.net>
+L:	netdev@vger.kernel.org
+L:	mptcp@lists.01.org
+W:	https://github.com/multipath-tcp/mptcp_net-next/wiki
+B:	https://github.com/multipath-tcp/mptcp_net-next/issues
+S:	Maintained
+F:	include/net/mptcp.h
+
 NETWORKING [TCP]
 M:	Eric Dumazet <edumazet@google.com>
 L:	netdev@vger.kernel.org
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index e9133bcf0544..1a261c3ee074 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -4091,6 +4091,9 @@ enum skb_ext_id {
 #endif
 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
 	TC_SKB_EXT,
+#endif
+#if IS_ENABLED(CONFIG_MPTCP)
+	SKB_EXT_MPTCP,
 #endif
 	SKB_EXT_NUM, /* must be last */
 };
diff --git a/include/net/mptcp.h b/include/net/mptcp.h
new file mode 100644
index 000000000000..f9f668ac4339
--- /dev/null
+++ b/include/net/mptcp.h
@@ -0,0 +1,27 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Multipath TCP
+ *
+ * Copyright (c) 2017 - 2019, Intel Corporation.
+ */
+
+#ifndef __NET_MPTCP_H
+#define __NET_MPTCP_H
+
+#include <linux/types.h>
+
+/* MPTCP sk_buff extension data */
+struct mptcp_ext {
+	u64		data_ack;
+	u64		data_seq;
+	u32		subflow_seq;
+	u16		data_len;
+	u8		use_map:1,
+			dsn64:1,
+			data_fin:1,
+			use_ack:1,
+			ack64:1,
+			__unused:2;
+};
+
+#endif /* __NET_MPTCP_H */
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 973a71f4bc89..fa67036dd928 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -68,6 +68,7 @@
 #include <net/ip6_checksum.h>
 #include <net/xfrm.h>
 #include <net/mpls.h>
+#include <net/mptcp.h>
 
 #include <linux/uaccess.h>
 #include <trace/events/skb.h>
@@ -4109,6 +4110,9 @@ static const u8 skb_ext_type_len[] = {
 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
 	[TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
 #endif
+#if IS_ENABLED(CONFIG_MPTCP)
+	[SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
+#endif
 };
 
 static __always_inline unsigned int skb_ext_total_length(void)
@@ -4122,6 +4126,9 @@ static __always_inline unsigned int skb_ext_total_length(void)
 #endif
 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
 		skb_ext_type_len[TC_SKB_EXT] +
+#endif
+#if IS_ENABLED(CONFIG_MPTCP)
+		skb_ext_type_len[SKB_EXT_MPTCP] +
 #endif
 		0;
 }
-- 
2.24.1

[PATCH net-next v3 07/11] tcp: Prevent coalesce/collapse when skb has MPTCP extensions

[Mat Martineau]

The MPTCP extension data needs to be preserved as it passes through the
TCP stack. Make sure that these skbs are not appended to others during
coalesce or collapse, so the data remains associated with the payload of
the given skb.

Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
---
 include/net/mptcp.h   | 16 ++++++++++++++++
 include/net/tcp.h     |  8 ++++++++
 net/ipv4/tcp_input.c  | 10 ++++++++--
 net/ipv4/tcp_output.c |  2 +-
 4 files changed, 33 insertions(+), 3 deletions(-)

diff --git a/include/net/mptcp.h b/include/net/mptcp.h
index f9f668ac4339..43ddfdf9e4a3 100644
--- a/include/net/mptcp.h
+++ b/include/net/mptcp.h
@@ -8,6 +8,7 @@
 #ifndef __NET_MPTCP_H
 #define __NET_MPTCP_H
 
+#include <linux/skbuff.h>
 #include <linux/types.h>
 
 /* MPTCP sk_buff extension data */
@@ -24,4 +25,19 @@ struct mptcp_ext {
 			__unused:2;
 };
 
+#ifdef CONFIG_MPTCP
+
+static inline bool mptcp_skb_ext_exist(const struct sk_buff *skb)
+{
+	return skb_ext_exist(skb, SKB_EXT_MPTCP);
+}
+
+#else
+
+static inline bool mptcp_skb_ext_exist(const struct sk_buff *skb)
+{
+	return false;
+}
+
+#endif /* CONFIG_MPTCP */
 #endif /* __NET_MPTCP_H */
diff --git a/include/net/tcp.h b/include/net/tcp.h
index c82b2f75d024..c483c73b8d41 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -39,6 +39,7 @@
 #include <net/tcp_states.h>
 #include <net/inet_ecn.h>
 #include <net/dst.h>
+#include <net/mptcp.h>
 
 #include <linux/seq_file.h>
 #include <linux/memcontrol.h>
@@ -978,6 +979,13 @@ static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
 	return likely(!TCP_SKB_CB(skb)->eor);
 }
 
+static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
+					const struct sk_buff *from)
+{
+	return likely(tcp_skb_can_collapse_to(to) &&
+		      !mptcp_skb_ext_exist(from));
+}
+
 /* Events passed to congestion control interface */
 enum tcp_ca_event {
 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 88b987ca9ebb..55b460a2ece2 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -1422,7 +1422,7 @@ static struct sk_buff *tcp_shift_skb_data(struct sock *sk, struct sk_buff *skb,
 	if ((TCP_SKB_CB(prev)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED)
 		goto fallback;
 
-	if (!tcp_skb_can_collapse_to(prev))
+	if (!tcp_skb_can_collapse(prev, skb))
 		goto fallback;
 
 	in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) &&
@@ -4420,6 +4420,9 @@ static bool tcp_try_coalesce(struct sock *sk,
 	if (TCP_SKB_CB(from)->seq != TCP_SKB_CB(to)->end_seq)
 		return false;
 
+	if (mptcp_skb_ext_exist(from))
+		return false;
+
 #ifdef CONFIG_TLS_DEVICE
 	if (from->decrypted != to->decrypted)
 		return false;
@@ -4928,10 +4931,12 @@ tcp_collapse(struct sock *sk, struct sk_buff_head *list, struct rb_root *root,
 
 		/* The first skb to collapse is:
 		 * - not SYN/FIN and
+		 * - does not include a MPTCP skb extension
 		 * - bloated or contains data before "start" or
 		 *   overlaps to the next one.
 		 */
 		if (!(TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) &&
+		    !mptcp_skb_ext_exist(skb) &&
 		    (tcp_win_from_space(sk, skb->truesize) > skb->len ||
 		     before(TCP_SKB_CB(skb)->seq, start))) {
 			end_of_skbs = false;
@@ -4947,7 +4952,7 @@ tcp_collapse(struct sock *sk, struct sk_buff_head *list, struct rb_root *root,
 		/* Decided to skip this, advance start seq. */
 		start = TCP_SKB_CB(skb)->end_seq;
 	}
-	if (end_of_skbs ||
+	if (end_of_skbs || mptcp_skb_ext_exist(skb) ||
 	    (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)))
 		return;
 
@@ -4990,6 +4995,7 @@ tcp_collapse(struct sock *sk, struct sk_buff_head *list, struct rb_root *root,
 				skb = tcp_collapse_one(sk, skb, list, root);
 				if (!skb ||
 				    skb == tail ||
+				    mptcp_skb_ext_exist(skb) ||
 				    (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)))
 					goto end;
 #ifdef CONFIG_TLS_DEVICE
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index b184f03d7437..9e04d45bc0e4 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -2854,7 +2854,7 @@ static void tcp_retrans_try_collapse(struct sock *sk, struct sk_buff *to,
 		if (!tcp_can_collapse(sk, skb))
 			break;
 
-		if (!tcp_skb_can_collapse_to(to))
+		if (!tcp_skb_can_collapse(to, skb))
 			break;
 
 		space -= skb->len;
-- 
2.24.1

Re: [PATCH net-next v3 07/11] tcp: Prevent coalesce/collapse when skb has MPTCP extensions

[Eric Dumazet]

On 12/17/19 12:38 PM, Mat Martineau wrote:
> The MPTCP extension data needs to be preserved as it passes through the
> TCP stack. Make sure that these skbs are not appended to others during
> coalesce or collapse, so the data remains associated with the payload of
> the given skb.
>


This seems a very pessimistic change to me.

Are you planing later to refine this limitation ?

Surely if a sender sends TSO packet, we allow all the segments
being aggregated at receive side either by GRO or TCP coalescing.

Re: [PATCH net-next v3 07/11] tcp: Prevent coalesce/collapse when skb has MPTCP extensions

[David Miller]

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Wed, 18 Dec 2019 11:50:24 -0800

> On 12/17/19 12:38 PM, Mat Martineau wrote:
>> The MPTCP extension data needs to be preserved as it passes through the
>> TCP stack. Make sure that these skbs are not appended to others during
>> coalesce or collapse, so the data remains associated with the payload of
>> the given skb.
> 
> This seems a very pessimistic change to me.
> 
> Are you planing later to refine this limitation ?
> 
> Surely if a sender sends TSO packet, we allow all the segments
> being aggregated at receive side either by GRO or TCP coalescing.

This turns off absolutely crucial functional elements of our TCP
stack, and will avoid all of the machinery that avoids wastage in TCP
packets sitting in the various queues.  skb->truesize management, etc.

I will not apply these patches with such a non-trivial regression in
place for MPTCP streams, sorry.

Re: [PATCH net-next v3 07/11] tcp: Prevent coalesce/collapse when skb has MPTCP extensions

[Mat Martineau]

On Wed, 18 Dec 2019, David Miller wrote:

> From: Eric Dumazet <eric.dumazet@gmail.com>
> Date: Wed, 18 Dec 2019 11:50:24 -0800
>
>> On 12/17/19 12:38 PM, Mat Martineau wrote:
>>> The MPTCP extension data needs to be preserved as it passes through the
>>> TCP stack. Make sure that these skbs are not appended to others during
>>> coalesce or collapse, so the data remains associated with the payload of
>>> the given skb.
>>
>> This seems a very pessimistic change to me.
>>
>> Are you planing later to refine this limitation ?
>>
>> Surely if a sender sends TSO packet, we allow all the segments
>> being aggregated at receive side either by GRO or TCP coalescing.
>
> This turns off absolutely crucial functional elements of our TCP
> stack, and will avoid all of the machinery that avoids wastage in TCP
> packets sitting in the various queues.  skb->truesize management, etc.
>
> I will not apply these patches with such a non-trivial regression in
> place for MPTCP streams, sorry.
>

Ok, understood. Not every packet has this MPTCP extension data so 
coalescing was not always turned off, but given the importance of avoiding 
this memory waste I'll confirm GRO behavior and work on maintaining 
coalesce/collapse with identical MPTCP extension data.

--
Mat Martineau
Intel

Re: [PATCH net-next v3 07/11] tcp: Prevent coalesce/collapse when skb has MPTCP extensions

[subashab]

> Ok, understood. Not every packet has this MPTCP extension data so
> coalescing was not always turned off, but given the importance of
> avoiding
> 
> this memory waste I'll confirm GRO behavior and work on maintaining
> coalesce/collapse with identical MPTCP extension data.
> 

Hi Mat

Are identical MPTCP extensions a common case?
AFAIK the data sequence number and the subflow sequence number change
per packet even in a single stream scenario.

Re: [PATCH net-next v3 07/11] tcp: Prevent coalesce/collapse when skb has MPTCP extensions

[Paolo Abeni]

On Wed, 2019-12-18 at 14:36 -0700, subashab@codeaurora.org wrote:
> > Ok, understood. Not every packet has this MPTCP extension data so
> > coalescing was not always turned off, but given the importance of
> > avoiding
> > 
> > this memory waste I'll confirm GRO behavior and work on maintaining
> > coalesce/collapse with identical MPTCP extension data.
> > 
> 
> Hi Mat
> 
> Are identical MPTCP extensions a common case?

Yes, they are.

> AFAIK the data sequence number and the subflow sequence number change
> per packet even in a single stream scenario.

What actually change on per packet basis is the TCP sequence number.
The DSS mapping can spawn on multiple packets, and will have constand
(base) sequence number and length.

Cheers,

Paolo

Re: [MPTCP] Re: [PATCH net-next v3 07/11] tcp: Prevent coalesce/collapse when skb has MPTCP extensions

[Paolo Abeni]

On Wed, 2019-12-18 at 12:45 -0800, David Miller wrote:
> From: Eric Dumazet <eric.dumazet@gmail.com>
> Date: Wed, 18 Dec 2019 11:50:24 -0800
> 
> > On 12/17/19 12:38 PM, Mat Martineau wrote:
> >> The MPTCP extension data needs to be preserved as it passes through the
> >> TCP stack. Make sure that these skbs are not appended to others during
> >> coalesce or collapse, so the data remains associated with the payload of
> >> the given skb.
> > 
> > This seems a very pessimistic change to me.
> > 
> > Are you planing later to refine this limitation ?
> > 
> > Surely if a sender sends TSO packet, we allow all the segments
> > being aggregated at receive side either by GRO or TCP coalescing.
> 
> This turns off absolutely crucial functional elements of our TCP
> stack, and will avoid all of the machinery that avoids wastage in TCP
> packets sitting in the various queues.  skb->truesize management, etc.

Thank you for the feedback!

Just to clarify, with the code we have currently posted TSO trains of
MPTCP packets can be aggregated by the GRO engine almost exactly as
currently happens for plain TCP packets.

We still have chances to aggregate packets belonging to a MPTCP stream,
as not all of them carry a DSS option.

We opted to not coalesce at the TCP level for the moment to avoid
adding additional hook code inside the coalescing code.

If you are ok without such hooks in the initial version, we can handle
MPTCP coalescing, too. The real work will likely land in part 2.

Would that fit you?

Thanks,

Paolo

Re: [MPTCP] Re: [PATCH net-next v3 07/11] tcp: Prevent coalesce/collapse when skb has MPTCP extensions

[David Miller]

From: Paolo Abeni <pabeni@redhat.com>
Date: Wed, 18 Dec 2019 23:15:46 +0100

> Just to clarify, with the code we have currently posted TSO trains of
> MPTCP packets can be aggregated by the GRO engine almost exactly as
> currently happens for plain TCP packets.
> 
> We still have chances to aggregate packets belonging to a MPTCP stream,
> as not all of them carry a DSS option.
> 
> We opted to not coalesce at the TCP level for the moment to avoid
> adding additional hook code inside the coalescing code.
> 
> If you are ok without such hooks in the initial version, we can handle
> MPTCP coalescing, too. The real work will likely land in part 2.
> 
> Would that fit you?

Can't you someone encode the MPTCP metadata so that it will only apply
to a range or something like that?  Would that help?

[PATCH net-next v3 08/11] tcp: Export TCP functions and ops struct

[Mat Martineau]

MPTCP will make use of tcp_send_mss() and tcp_push() when sending
data to specific TCP subflows.

tcp_request_sock_ipvX_ops and ipvX_specific will be referenced
during TCP subflow creation.

Co-developed-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Peter Krystad <peter.krystad@linux.intel.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
---
 include/net/tcp.h   | 8 ++++++++
 net/ipv4/tcp.c      | 6 +++---
 net/ipv4/tcp_ipv4.c | 2 +-
 net/ipv6/tcp_ipv6.c | 6 +++---
 4 files changed, 15 insertions(+), 7 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index c483c73b8d41..921288afa11b 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -330,6 +330,9 @@ int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
 			size_t size, int flags);
 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
 		 size_t size, int flags);
+int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
+void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
+	      int size_goal);
 void tcp_release_cb(struct sock *sk);
 void tcp_wfree(struct sk_buff *skb);
 void tcp_write_timer_handler(struct sock *sk);
@@ -2002,6 +2005,11 @@ struct tcp_request_sock_ops {
 			   enum tcp_synack_type synack_type);
 };
 
+extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
+#if IS_ENABLED(CONFIG_IPV6)
+extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
+#endif
+
 #ifdef CONFIG_SYN_COOKIES
 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
 					 const struct sock *sk, struct sk_buff *skb,
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 93fe77c5b02d..dc9bd27f4222 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -690,8 +690,8 @@ static bool tcp_should_autocork(struct sock *sk, struct sk_buff *skb,
 	       refcount_read(&sk->sk_wmem_alloc) > skb->truesize;
 }
 
-static void tcp_push(struct sock *sk, int flags, int mss_now,
-		     int nonagle, int size_goal)
+void tcp_push(struct sock *sk, int flags, int mss_now,
+	      int nonagle, int size_goal)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct sk_buff *skb;
@@ -925,7 +925,7 @@ static unsigned int tcp_xmit_size_goal(struct sock *sk, u32 mss_now,
 	return max(size_goal, mss_now);
 }
 
-static int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
+int tcp_send_mss(struct sock *sk, int *size_goal, int flags)
 {
 	int mss_now;
 
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 26637fce324d..c3b91789de65 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -1372,7 +1372,7 @@ struct request_sock_ops tcp_request_sock_ops __read_mostly = {
 	.syn_ack_timeout =	tcp_syn_ack_timeout,
 };
 
-static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
+const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
 	.mss_clamp	=	TCP_MSS_DEFAULT,
 #ifdef CONFIG_TCP_MD5SIG
 	.req_md5_lookup	=	tcp_v4_md5_lookup,
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index df5fd9109696..30dceac8a608 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -75,7 +75,7 @@ static void	tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
 static int	tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
 
 static const struct inet_connection_sock_af_ops ipv6_mapped;
-static const struct inet_connection_sock_af_ops ipv6_specific;
+const struct inet_connection_sock_af_ops ipv6_specific;
 #ifdef CONFIG_TCP_MD5SIG
 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
@@ -785,7 +785,7 @@ struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
 	.syn_ack_timeout =	tcp_syn_ack_timeout,
 };
 
-static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
+const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 	.mss_clamp	=	IPV6_MIN_MTU - sizeof(struct tcphdr) -
 				sizeof(struct ipv6hdr),
 #ifdef CONFIG_TCP_MD5SIG
@@ -1739,7 +1739,7 @@ static struct timewait_sock_ops tcp6_timewait_sock_ops = {
 	.twsk_destructor = tcp_twsk_destructor,
 };
 
-static const struct inet_connection_sock_af_ops ipv6_specific = {
+const struct inet_connection_sock_af_ops ipv6_specific = {
 	.queue_xmit	   = inet6_csk_xmit,
 	.send_check	   = tcp_v6_send_check,
 	.rebuild_header	   = inet6_sk_rebuild_header,
-- 
2.24.1

[PATCH net-next v3 09/11] tcp: Check for filled TCP option space before SACK

[Mat Martineau]

Update the SACK check to work with zero option space available, a case
that's possible with MPTCP but not MD5+TS. Maintained only one
conditional branch for insufficient SACK space.

v1 -> v2:
- Moves the check inside the SACK branch by taking recent SACK fix:

    9424e2e7ad93 (tcp: md5: fix potential overestimation of TCP option space)

  in to account, but modifies it to work in MPTCP scenarios beyond the
  MD5+TS corner case.

Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
---
 net/ipv4/tcp_output.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 9e04d45bc0e4..e797ca6c6d7d 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -751,13 +751,17 @@ static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb
 	eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack;
 	if (unlikely(eff_sacks)) {
 		const unsigned int remaining = MAX_TCP_OPTION_SPACE - size;
+		if (unlikely(remaining < TCPOLEN_SACK_BASE_ALIGNED +
+					 TCPOLEN_SACK_PERBLOCK))
+			return size;
+
 		opts->num_sack_blocks =
 			min_t(unsigned int, eff_sacks,
 			      (remaining - TCPOLEN_SACK_BASE_ALIGNED) /
 			      TCPOLEN_SACK_PERBLOCK);
-		if (likely(opts->num_sack_blocks))
-			size += TCPOLEN_SACK_BASE_ALIGNED +
-				opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK;
+
+		size += TCPOLEN_SACK_BASE_ALIGNED +
+			opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK;
 	}
 
 	return size;
-- 
2.24.1

[PATCH net-next v3 10/11] tcp: clean ext on tx recycle

[Mat Martineau]

From: Paolo Abeni <pabeni@redhat.com>

Otherwise we will find stray/unexpected/old extensions value on next
iteration.

On tcp_write_xmit() we can end-up splitting an already queued skb in two
parts, via tso_fragment(). The newly created skb can be allocated via
the tx cache and an upper layer will not be aware of it, so that upper
layer cannot set the ext properly.

Resetting the ext on recycle ensures that stale data is not propagated
in to packet headers or elsewhere.

An alternative would be add an additional hook in tso_fragment() or in
sk_stream_alloc_skb() to init the ext for upper layers that need it.

Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
---
 include/net/sock.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/include/net/sock.h b/include/net/sock.h
index b93cadba1a3b..23efed7f4e70 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -1474,6 +1474,7 @@ static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
 	sk_mem_uncharge(sk, skb->truesize);
 	if (static_branch_unlikely(&tcp_tx_skb_cache_key) &&
 	    !sk->sk_tx_skb_cache && !skb_cloned(skb)) {
+		skb_ext_reset(skb);
 		skb_zcopy_clear(skb, true);
 		sk->sk_tx_skb_cache = skb;
 		return;
-- 
2.24.1

[PATCH net-next v3 11/11] skb: add helpers to allocate ext independently from sk_buff

[Mat Martineau]

From: Paolo Abeni <pabeni@redhat.com>

Currently we can allocate the extension only after the skb,
this change allows the user to do the opposite, will simplify
allocation failure handling from MPTCP.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
---
 include/linux/skbuff.h |  3 +++
 net/core/skbuff.c      | 35 +++++++++++++++++++++++++++++++++--
 2 files changed, 36 insertions(+), 2 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 1a261c3ee074..af9b6cf79a65 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -4115,6 +4115,9 @@ struct skb_ext {
 	char data[0] __aligned(8);
 };
 
+struct skb_ext *__skb_ext_alloc(void);
+void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
+		    struct skb_ext *ext);
 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
 void __skb_ext_put(struct skb_ext *ext);
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index fa67036dd928..504e9bd5ebce 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -5983,7 +5983,14 @@ static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
 	return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
 }
 
-static struct skb_ext *skb_ext_alloc(void)
+/**
+ * __skb_ext_alloc - allocate a new skb extensions storage
+ *
+ * Returns the newly allocated pointer. The pointer can later attached to a
+ * skb via __skb_ext_set().
+ * Note: caller must handle the skb_ext as an opaque data.
+ */
+struct skb_ext *__skb_ext_alloc(void)
 {
 	struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
 
@@ -6023,6 +6030,30 @@ static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
 	return new;
 }
 
+/**
+ * __skb_ext_set - attach the specified extension storage to this skb
+ * @skb: buffer
+ * @id: extension id
+ * @ext: extension storage previously allocated via __skb_ext_alloc()
+ *
+ * Existing extensions, if any, are cleared.
+ *
+ * Returns the pointer to the extension.
+ */
+void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
+		    struct skb_ext *ext)
+{
+	unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
+
+	skb_ext_put(skb);
+	newlen = newoff + skb_ext_type_len[id];
+	ext->chunks = newlen;
+	ext->offset[id] = newoff;
+	skb->extensions = ext;
+	skb->active_extensions = 1 << id;
+	return skb_ext_get_ptr(ext, id);
+}
+
 /**
  * skb_ext_add - allocate space for given extension, COW if needed
  * @skb: buffer
@@ -6056,7 +6087,7 @@ void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
 	} else {
 		newoff = SKB_EXT_CHUNKSIZEOF(*new);
 
-		new = skb_ext_alloc();
+		new = __skb_ext_alloc();
 		if (!new)
 			return NULL;
 	}
-- 
2.24.1