Skip to contentSkip to conversation

Implement outgoing MP_JOIN

PATCH Superseded Peter Krystad 3 participants 25 messages Jul 17, 2019 – Jul 26, 2019

Subsystems (3)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Peter Krystad

author
#0 lore raw

cover Implement outgoing MP_JOIN

This is several accumulated fixes that are required for MP_JOIN (secondary subflows) plus the outgoing code.

Both outgoing and incoming MP_JOIN now inter-operate with the multipath.org implementation although there is still a deadlock when incoming is fully established. Since the incoming direction did not work at all before this is an improvement. I have not yet tested running our implmentation against itself.

The final patch in the set is an "interim path manager", this is throw-away code that I include for testing purposes. For an incoming connection it is necessary for the server to advertise a secondary address and for my testing I have been hard-coding the IP address of my secondary interface when I build the kernel. Obviously not the final solution but we need to discuss what approach to use for the self tests. I also think we may not want to publish this patch on our net-next branch.

Peter Krystad (9):
Extend path manager interface
Fix locking in mptcp_finish_join.
Fix parsing MP_JOIN third ACK.
Re-factor and fixes for crypto_hmac_sha1() Rename mptcp_attach_dss()
Fix return value if no DSS option is created. Move setting request_ fields of subflow to MPTCP layer
mptcp: Add handling of outgoing MP_JOIN requests
mptcp: Implement interim path manager

 include/net/mptcp.h  |  12 +--
 net/ipv4/tcp_input.c |   4 +-
 net/mptcp/crypto.c   |  23 +++---
 net/mptcp/options.c  |  77 ++++++++++++++++---
 net/mptcp/pm.c       | 171 +++++++++++++++++++++++++++++++++++++++++--
 net/mptcp/protocol.c |  29 ++++++--
 net/mptcp/protocol.h |  54 +++++++++++---
 net/mptcp/subflow.c  |  72 ++++++++++++++++--
 net/mptcp/token.c    |  82 ++++++++++++++++++---
 9 files changed, 452 insertions(+), 72 deletions(-)
Signature · 1 line
2.17.2
author
#1 lore raw

PATCH 1/9 Extend path manager interface

A little more functionality and some comments clarifying which routines are for a path manager to call and which are callbacks.

squashto: Add path manager interface

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 net/mptcp/options.c  | 19 +++++++++--
 net/mptcp/pm.c       | 77 ++++++++++++++++++++++++++++++++++++++++----
 net/mptcp/protocol.c |  5 +--
 net/mptcp/protocol.h | 35 ++++++++++++++++----
 4 files changed, 117 insertions(+), 19 deletions(-)
net/mptcp/options.c +16 -3
index 58215f19829a..0d0d47f84adb 100644
@@ -412,10 +412,10 @@ static bool mptcp_established_options_addr(struct sock *sk,
 	struct subflow_context *subflow = subflow_ctx(sk);
 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 
-	if (subflow->fourth_ack)
-		return pm_addr_signal(msk, size, remaining, opts);
+	if (!msk)
+		return false;
 
-	return false;
+	return pm_addr_signal(msk, size, remaining, opts);
 }
 
 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
@@ -479,11 +479,20 @@ bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
 void mptcp_attach_dss(struct sock *sk, struct sk_buff *skb,
 		      struct tcp_options_received *opt_rx)
 {
+	struct subflow_context *subflow = subflow_ctx(sk);
+	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 	struct mptcp_options_received *mp_opt;
 	struct mptcp_ext *mpext;
 
 	mp_opt = &opt_rx->mptcp;
 
+	if (mp_opt->add_addr && mp_opt->family == MPTCP_ADDR_IPVERSION_4) {
+		if (msk) {
+			pm_add_addr(msk, &mp_opt->addr, mp_opt->addr_id);
+			mp_opt->add_addr = 0;
+		}
+	}
+
 	if (!mp_opt->dss)
 		return;
 
@@ -510,6 +519,10 @@ void mptcp_attach_dss(struct sock *sk, struct sk_buff *skb,
 	}
 
 	mpext->data_fin = mp_opt->data_fin;
+
+	if (msk)
+		pm_fully_established(msk);
+
 }
 
 void mptcp_write_options(__be32 *ptr, struct mptcp_out_options *opts)
net/mptcp/pm.c +70 -7

110 lines not on this page · raw message

net/mptcp/protocol.c +3 -2
index 774ed25d3b6d..a56085742cf7 100644
@@ -632,7 +632,7 @@ static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
 		token_update_accept(new_sock->sk, new_mptcp_sock);
 		msk->subflow = NULL;
 
-		pm_new_connection(msk);
+		pm_new_connection(msk, 1);
 
 		crypto_key_sha1(msk->remote_key, NULL, &ack_seq);
 		msk->write_seq = subflow->idsn + 1;
@@ -759,7 +759,7 @@ void mptcp_finish_connect(struct sock *sk, int mp_capable)
 		msk->token = subflow->token;
 		pr_debug("msk=%p, token=%u", msk, msk->token);
 
-		pm_new_connection(msk);
+		pm_new_connection(msk, 0);
 
 		crypto_key_sha1(msk->remote_key, NULL, &ack_seq);
 		msk->write_seq = subflow->idsn + 1;
@@ -1008,6 +1008,7 @@ void __init mptcp_init(void)
 	token_init();
 	crypto_init();
 	subflow_init();
+	pm_init();
 
 	if (proto_register(&mptcp_prot, 1) != 0)
 		panic("Failed to register MPTCP proto.\n");
net/mptcp/protocol.h +28 -7
index 7f15f6aab93d..8c04774ea2ec 100644
@@ -77,15 +77,33 @@ static inline u32 mptcp_option(u8 subopt, u8 len, u8 nib, u8 field)
 		     ((nib & 0xF) << 8) | field);
 }
 
-struct pm_data {
-	u8 addr_id;
-	sa_family_t family;
+struct mptcp_pm_data {
+	u8	local_valid;
+	u8	local_id;
+	sa_family_t local_family;
+	union {
+		struct in_addr local_addr;
+#if IS_ENABLED(CONFIG_IPV6)
+		struct in6_addr local_addr6;
+#endif
+	};
+	u8	remote_valid;
+	u8	remote_id;
+	sa_family_t remote_family;
 	union {
-		struct in_addr addr;
+		struct in_addr remote_addr;
 #if IS_ENABLED(CONFIG_IPV6)
-		struct in6_addr addr6;
+		struct in6_addr remote_addr6;
 #endif
 	};
+	u32	remote_token;
+	u8	server_side : 1,
+		fully_established : 1;
+
+	/* for interim path manager */
+	struct	work_struct addr_work;
+	struct	work_struct subflow_work;
+	u32	token;
 };
 
 /* MPTCP connection sock */
@@ -99,7 +117,7 @@ struct mptcp_sock {
 	u32		token;
 	struct list_head conn_list;
 	struct socket	*subflow; /* outgoing connect/listener/!mp_capable */
-	struct pm_data	pm;
+	struct mptcp_pm_data	pm;
 	u8		addr_signal;
 };
 
@@ -222,7 +240,8 @@ void crypto_key_sha1(u64 key, u32 *token, u64 *idsn);
 void crypto_hmac_sha1(u64 key1, u64 key2, u32 *hash_out,
 		      int arg_num, ...);
 
-void pm_new_connection(struct mptcp_sock *msk);
+void pm_init(void);
+void pm_new_connection(struct mptcp_sock *msk, int server_side);
 void pm_fully_established(struct mptcp_sock *msk);
 void pm_connection_closed(struct mptcp_sock *msk);
 void pm_subflow_established(struct mptcp_sock *msk, u8 id);
@@ -232,6 +251,8 @@ void pm_add_addr6(struct mptcp_sock *msk, const struct in6_addr *addr, u8 id);
 void pm_rm_addr(struct mptcp_sock *msk, u8 id);
 bool pm_addr_signal(struct mptcp_sock *msk, unsigned int *size,
 		    unsigned int remaining, struct mptcp_out_options *opts);
+int pm_get_local_id(struct request_sock *req, struct sock *sk,
+		    const struct sk_buff *skb);
 
 static inline struct mptcp_ext *mptcp_get_ext(struct sk_buff *skb)
 {
Signature · 1 line
2.17.2
author
#4 lore raw

PATCH 2/9 Fix locking in mptcp_finish_join.

In reply to Peter Krystad

Lock was being taken on the subflow sock, not mptcp_sock.

squashto: Add handling of incoming MP_JOIN requests

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 net/mptcp/protocol.c | 9 +++++----
 net/mptcp/protocol.h | 2 +-
 net/mptcp/subflow.c  | 2 +-
 3 files changed, 7 insertions(+), 6 deletions(-)
net/mptcp/protocol.c +5 -4
index a56085742cf7..1664a69c8e30 100644
@@ -777,18 +777,19 @@ void mptcp_finish_connect(struct sock *sk, int mp_capable)
 	inet_sk_state_store(sk, TCP_ESTABLISHED);
 }
 
-void mptcp_finish_join(struct sock *conn, struct sock *sk)
+void mptcp_finish_join(struct sock *sk)
 {
 	struct subflow_context *subflow = subflow_ctx(sk);
-	struct mptcp_sock *msk = mptcp_sk(conn);
+	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 
 	pr_debug("msk=%p, subflow=%p", msk, subflow);
 
 	local_bh_disable();
-	bh_lock_sock_nested(sk);
+	bh_lock_sock_nested(subflow->conn);
 	list_add_tail(&subflow->node, &msk->conn_list);
-	bh_unlock_sock(sk);
+	bh_unlock_sock(subflow->conn);
 	local_bh_enable();
+	inet_sk_state_store(sk, TCP_ESTABLISHED);
 }
 
 bool mptcp_sk_is_subflow(const struct sock *sk)
net/mptcp/protocol.h +1 -1
index 8c04774ea2ec..9d6b10e5479c 100644
@@ -210,7 +210,7 @@ void mptcp_get_options(const struct sk_buff *skb,
 		       struct tcp_options_received *opt_rx);
 
 void mptcp_finish_connect(struct sock *sk, int mp_capable);
-void mptcp_finish_join(struct sock *conn, struct sock *sk);
+void mptcp_finish_join(struct sock *sk);
 
 void token_init(void);
 void token_new_request(struct request_sock *req, const struct sk_buff *skb);
net/mptcp/subflow.c +1 -1
index fdc5aecab897..32652f7d1497 100644
@@ -171,7 +171,7 @@ static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 			if (token_new_join(child))
 				goto close_child;
 			else
-				mptcp_finish_join(ctx->conn, child);
+				mptcp_finish_join(child);
 		}
 	}
 
Signature · 1 line
2.17.2
author
#5 lore raw

PATCH 3/9 Fix parsing MP_JOIN third ACK.

In reply to Peter Krystad

There is a reserved byte of zero's in there.

squashto: Add handling of incoming MP_JOIN requests

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 net/mptcp/options.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/mptcp/options.c +1 -1
index 0d0d47f84adb..b5a86f47f097 100644
@@ -92,7 +92,7 @@ void mptcp_parse_option(const unsigned char *ptr, int opsize,
 				 mp_opt->backup, mp_opt->join_id,
 				 mp_opt->thmac, mp_opt->nonce);
 		} else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
-			ptr++;
+			ptr += 2;
 			memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
 			pr_debug("MP_JOIN hmac");
 		} else {
Signature · 1 line
2.17.2
author
#6 lore raw

PATCH 4/9 Re-factor and fixes for crypto_hmac_sha1()

In reply to Peter Krystad

Re-factor to use parameters specific to MPTCP use case similar to crypto_key_sha1() and get rid of var args. Also fix endianness issues.

squash to: Add key generation and token tree
and: Add handling of incoming MP_JOIN requests

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 net/mptcp/crypto.c   | 23 +++++++++--------------
 net/mptcp/protocol.h |  4 ++--
 net/mptcp/token.c    | 19 ++++++++-----------
 3 files changed, 19 insertions(+), 27 deletions(-)
net/mptcp/crypto.c +9 -14
index 26a68aa1933a..7ab22b3de2d2 100644
@@ -118,23 +118,23 @@ void crypto_key_sha1(u64 key, u32 *token, u64 *idsn)
 		*idsn = ((u64)mptcp_hashed_key[3] << 32) + mptcp_hashed_key[4];
 }
 
-void crypto_hmac_sha1(u64 key1, u64 key2, u32 *hash_out,
-		      int arg_num, ...)
+void crypto_hmac_sha1(u64 key1, u64 key2, u32 nonce1, u32 nonce2, u32 *hash_out)
 {
 	u32 workspace[SHA_WORKSPACE_WORDS];
 	u8 input[128]; /* 2 512-bit blocks */
 	int i;
 	int index;
-	int length;
-	u8 *msg;
-	va_list list;
 	u8 key_1[8];
 	u8 key_2[8];
+	u8 nonce_1[4];
+	u8 nonce_2[4];
 
 	memset(workspace, 0, sizeof(workspace));
 
 	put_unaligned_be64(key1, key_1);
 	put_unaligned_be64(key2, key_2);
+	put_unaligned_be32(nonce1, nonce_1);
+	put_unaligned_be32(nonce2, nonce_2);
 
 	/* Generate key xored with ipad */
 	memset(input, 0x36, 64);
@@ -143,16 +143,11 @@ void crypto_hmac_sha1(u64 key1, u64 key2, u32 *hash_out,
 	for (i = 0; i < 8; i++)
 		input[i + 8] ^= key_2[i];
 
-	va_start(list, arg_num);
 	index = 64;
-	for (i = 0; i < arg_num; i++) {
-		length = va_arg(list, int);
-		msg = va_arg(list, u8 *);
-		WARN_ON(index + length > 125); /* Message is too long */
-		memcpy(&input[index], msg, length);
-		index += length;
-	}
-	va_end(list);
+	memcpy(&input[index], nonce_1, 4);
+	index = 68;
+	memcpy(&input[index], nonce_2, 4);
+	index = 72;
 
 	input[index] = 0x80; /* Padding: First bit after message = 1 */
 	memset(&input[index + 1], 0, (126 - index));
net/mptcp/protocol.h +2 -2
index 9d6b10e5479c..ff4f38da9276 100644
@@ -237,8 +237,8 @@ u32 crypto_v6_get_nonce(const struct in6_addr *saddr,
 			const struct in6_addr *daddr,
 			__be16 sport, __be16 dport);
 void crypto_key_sha1(u64 key, u32 *token, u64 *idsn);
-void crypto_hmac_sha1(u64 key1, u64 key2, u32 *hash_out,
-		      int arg_num, ...);
+void crypto_hmac_sha1(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
+		      u32 *hash_out);
 
 void pm_init(void);
 void pm_new_connection(struct mptcp_sock *msk, int server_side);
net/mptcp/token.c +8 -11
index c2f4fcb37566..c9160f50f047 100644
@@ -114,12 +114,11 @@ static void new_req_join(struct request_sock *req, struct sock *sk,
 	}
 	subflow_req->local_nonce = nonce;
 
-	crypto_hmac_sha1(msk->local_key,
-			 msk->remote_key,
-			 (u32 *)hmac, 2,
-			 4, (u8 *)&subflow_req->local_nonce,
-			 4, (u8 *)&subflow_req->remote_nonce);
-	subflow_req->thmac = *(u64 *)hmac;
+	crypto_hmac_sha1(msk->local_key, msk->remote_key,
+			 subflow_req->local_nonce, subflow_req->remote_nonce,
+			 (u32 *)hmac);
+
+	subflow_req->thmac = get_unaligned_be64(hmac);
 	pr_debug("local_nonce=%u, thmac=%llu", subflow_req->local_nonce,
 		 subflow_req->thmac);
 }
@@ -131,11 +130,9 @@ static int new_join_valid(struct request_sock *req, struct sock *sk,
 	struct mptcp_sock *msk = mptcp_sk(sk);
 	u8 hmac[MPTCPOPT_HMAC_LEN];
 
-	crypto_hmac_sha1(msk->remote_key,
-			 msk->local_key,
-			 (u32 *)hmac, 2,
-			 4, (u8 *)&subflow_req->remote_nonce,
-			 4, (u8 *)&subflow_req->local_nonce);
+	crypto_hmac_sha1(msk->remote_key, msk->local_key,
+			 subflow_req->remote_nonce, subflow_req->local_nonce,
+			 (u32 *)hmac);
 
 	return memcmp(hmac, (char *)rx_opt->mptcp.hmac, MPTCPOPT_HMAC_LEN);
 }
Signature · 1 line
2.17.2
author
#9 lore raw

PATCH 5/9 Rename mptcp_attach_dss()

In reply to Peter Krystad

Other incoming MPTCP options will be handled in this routine.

squashto: Implement MPTCP receive path

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 include/net/mptcp.h  | 10 +++++-----
 net/ipv4/tcp_input.c |  4 ++--
 net/mptcp/options.c  |  4 ++--
 3 files changed, 9 insertions(+), 9 deletions(-)
include/net/mptcp.h +5 -5
index 6cd650838ddf..bb2dd193c0c5 100644
@@ -68,9 +68,8 @@ bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
 			       unsigned int *size, unsigned int remaining,
 			       struct mptcp_out_options *opts);
-
-void mptcp_attach_dss(struct sock *sk, struct sk_buff *skb,
-		      struct tcp_options_received *opt_rx);
+void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb,
+			    struct tcp_options_received *opt_rx);
 
 static inline bool mptcp_skb_ext_exist(const struct sk_buff *skb)
 {
@@ -128,8 +127,9 @@ static inline bool mptcp_established_options(struct sock *sk,
 	return false;
 }
 
-static inline void mptcp_attach_dss(struct sock *sk, struct sk_buff *skb,
-				    struct tcp_options_received *opt_rx)
+static inline void mptcp_incoming_options(struct sock *sk,
+					  struct sk_buff *skb,
+					  struct tcp_options_received *opt_rx)
 {
 }
 
net/ipv4/tcp_input.c +2 -2
index b92985b4762c..ca2fea882281 100644
@@ -5668,9 +5668,9 @@ void tcp_rcv_established(struct sock *sk, struct sk_buff *skb)
 	/* Process urgent data. */
 	tcp_urg(sk, skb, th);
 
-	/* Prepare MPTCP sequence data */
+	/* Process MPTCP options */
 	if (sk_is_mptcp(sk))
-		mptcp_attach_dss(sk, skb, &tp->rx_opt);
+		mptcp_incoming_options(sk, skb, &tp->rx_opt);
 
 	/* step 7: process the segment text */
 	tcp_data_queue(sk, skb);
net/mptcp/options.c +2 -2
index b5a86f47f097..1a46cfb1f664 100644
@@ -476,8 +476,8 @@ bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
 	return false;
 }
 
-void mptcp_attach_dss(struct sock *sk, struct sk_buff *skb,
-		      struct tcp_options_received *opt_rx)
+void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb,
+			    struct tcp_options_received *opt_rx)
 {
 	struct subflow_context *subflow = subflow_ctx(sk);
 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
Signature · 1 line
2.17.2
author
#10 lore raw

PATCH 6/9 Fix return value if no DSS option is created.

In reply to Peter Krystad

Strictly speaking mptcp_established_options_dss() should return false if no DSS option is created.

squashto: Write MPTCP DSS headers to outgoing data packets

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 net/mptcp/options.c | 3 +++
 1 file changed, 3 insertions(+)
net/mptcp/options.c +3 -0
index 1a46cfb1f664..b0e3a1edc0ff 100644
@@ -400,6 +400,9 @@ static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
 		WARN(1, "MPTCP: Ack dropped");
 	}
 
+	if (!dss_size)
+		return false;
+
 	*size = ALIGN(dss_size, 4);
 	return true;
 }
Signature · 1 line
2.17.2
author
#11 lore raw

PATCH 7/9 Move setting request_ fields of subflow to MPTCP layer

In reply to Peter Krystad

For secondary subflows a different set of request_ fields will be added that will be set differently based on subflow type (initial vs secondary).

squashto: Associate MPTCP context with TCP socket

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 net/mptcp/protocol.c | 13 +++++++++++++
 net/mptcp/protocol.h |  2 +-
 net/mptcp/subflow.c  |  7 ++-----
 3 files changed, 16 insertions(+), 6 deletions(-)
net/mptcp/protocol.c +13 -0
index 1664a69c8e30..fe73379a81df 100644
@@ -840,6 +840,7 @@ static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 				int addr_len, int flags)
 {
 	struct mptcp_sock *msk = mptcp_sk(sock->sk);
+	struct subflow_context *subflow;
 	int err = -ENOTSUPP;
 
 	pr_debug("msk=%p", msk);
@@ -853,6 +854,11 @@ static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 			return err;
 	}
 
+	subflow = subflow_ctx(msk->subflow->sk);
+	subflow->request_mptcp = 1; /* @@ if MPTCP enabled */
+	subflow->request_cksum = 1; /* @@ if checksum enabled */
+	subflow->request_version = 0; /* only v0 supported */
+
 	return inet_stream_connect(msk->subflow, uaddr, addr_len, flags);
 }
 
@@ -902,6 +908,7 @@ static int mptcp_getname(struct socket *sock, struct sockaddr *uaddr,
 static int mptcp_listen(struct socket *sock, int backlog)
 {
 	struct mptcp_sock *msk = mptcp_sk(sock->sk);
+	struct subflow_context *subflow;
 	int err;
 
 	pr_debug("msk=%p", msk);
@@ -911,6 +918,12 @@ static int mptcp_listen(struct socket *sock, int backlog)
 		if (err)
 			return err;
 	}
+
+	subflow = subflow_ctx(msk->subflow->sk);
+	subflow->request_mptcp = 1; /* @@ if MPTCP enabled */
+	subflow->request_cksum = 1; /* @@ if checksum enabled */
+	subflow->request_version = 0; /* only v0 supported */
+
 	return inet_listen(msk->subflow, backlog);
 }
 
net/mptcp/protocol.h +1 -1
index ff4f38da9276..1ce9130dee66 100644
@@ -168,10 +168,10 @@ struct subflow_context {
 	u16	map_data_len;
 	u16	request_mptcp : 1,  /* send MP_CAPABLE */
 		request_cksum : 1,
+		request_version : 4,
 		mp_capable : 1,     /* remote is MPTCP capable */
 		mp_join : 1,        /* remote is JOINing */
 		fourth_ack : 1,     /* send initial DSS */
-		version : 4,
 		conn_finished : 1,
 		use_checksum : 1,
 		map_valid : 1,
net/mptcp/subflow.c +2 -5
index 32652f7d1497..b9a40cdb1d4d 100644
@@ -62,8 +62,8 @@ static void subflow_v4_init_req(struct request_sock *req,
 
 	if (rx_opt.mptcp.mp_capable && listener->request_mptcp) {
 		subflow_req->mp_capable = 1;
-		if (rx_opt.mptcp.version >= listener->version)
-			subflow_req->version = listener->version;
+		if (rx_opt.mptcp.version >= listener->request_version)
+			subflow_req->version = listener->request_version;
 		else
 			subflow_req->version = rx_opt.mptcp.version;
 		if ((rx_opt.mptcp.flags & MPTCP_CAP_CHECKSUM_REQD) ||
@@ -224,9 +224,6 @@ int subflow_create_socket(struct sock *sk, struct socket **new_sock)
 
 	*new_sock = sf;
 	subflow->conn = sk;
-	subflow->request_mptcp = 1; // @@ if MPTCP enabled
-	subflow->request_cksum = 1; // @@ if checksum enabled
-	subflow->version = 0;
 
 	return 0;
 }
Signature · 1 line
2.17.2
author
#14 lore raw

PATCH 8/9 mptcp: Add handling of outgoing MP_JOIN requests

In reply to Peter Krystad

Subflow creation may be initiated by the path manager when the primary connection is fully established and a remote address has been received via ADD_ADDR.

Create an in-kernel sock and use kernel_connect() to initiate connection. When a valid SYN-ACK is received the new sock is added to the tail of the mptcp sock conn_list where it will not interfere with data flow on the original connection.

Data flow and connection failover not addressed by this commit.

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 include/net/mptcp.h  |  2 ++
 net/mptcp/options.c  | 51 ++++++++++++++++++++++++++++++++---
 net/mptcp/protocol.c |  2 ++
 net/mptcp/protocol.h | 11 ++++++++
 net/mptcp/subflow.c  | 63 +++++++++++++++++++++++++++++++++++++++++++-
 net/mptcp/token.c    | 63 ++++++++++++++++++++++++++++++++++++++++++++
 6 files changed, 187 insertions(+), 5 deletions(-)
include/net/mptcp.h +2 -0
index bb2dd193c0c5..50cd1b31ebdd 100644
@@ -40,6 +40,8 @@ struct mptcp_out_options {
 	u8 backup;
 	u32 nonce;
 	u64 thmac;
+	u32 token;
+	u8 hmac[20];
 	struct mptcp_ext ext_copy;
 #endif
 };
net/mptcp/options.c +47 -4

105 lines not on this page · raw message

net/mptcp/protocol.c +2 -0
index fe73379a81df..26047d681dd3 100644
@@ -758,6 +758,8 @@ void mptcp_finish_connect(struct sock *sk, int mp_capable)
 		msk->local_key = subflow->local_key;
 		msk->token = subflow->token;
 		pr_debug("msk=%p, token=%u", msk, msk->token);
+		msk->dport = ntohs(inet_sk(msk->subflow->sk)->inet_dport);
+		pr_debug("dport=%d", msk->dport);
 
 		pm_new_connection(msk, 0);
 
net/mptcp/protocol.h +11 -0
index 1ce9130dee66..0e265759c575 100644
@@ -48,8 +48,10 @@
 #define TCPOLEN_MPTCP_ADD_ADDR6		20
 #define TCPOLEN_MPTCP_RM_ADDR		4
 
+/* MPTCP MP_JOIN flags */
 #define MPTCPOPT_BACKUP		BIT(0)
 #define MPTCPOPT_HMAC_LEN	20
+#define MPTCPOPT_THMAC_LEN	8
 
 /* MPTCP MP_CAPABLE flags */
 #define MPTCP_VERSION_MASK	(0x0F)
@@ -115,6 +117,7 @@ struct mptcp_sock {
 	u64		write_seq;
 	u64		ack_seq;
 	u32		token;
+	u16		dport;
 	struct list_head conn_list;
 	struct socket	*subflow; /* outgoing connect/listener/!mp_capable */
 	struct mptcp_pm_data	pm;
@@ -167,7 +170,9 @@ struct subflow_context {
 	u32	ssn_offset;
 	u16	map_data_len;
 	u16	request_mptcp : 1,  /* send MP_CAPABLE */
+		request_join : 1,   /* send MP_JOIN */
 		request_cksum : 1,
+		request_bkup : 1,
 		request_version : 4,
 		mp_capable : 1,     /* remote is MPTCP capable */
 		mp_join : 1,        /* remote is JOINing */
@@ -179,6 +184,7 @@ struct subflow_context {
 	u32	remote_nonce;
 	u64	thmac;
 	u32	local_nonce;
+	u8	hmac[MPTCPOPT_HMAC_LEN];
 	u8	local_id;
 	u8	remote_id;
 
@@ -202,6 +208,8 @@ mptcp_subflow_tcp_socket(const struct subflow_context *subflow)
 }
 
 void subflow_init(void);
+int subflow_connect(struct sock *sk, struct sockaddr_in *local,
+		    struct sockaddr_in *remote, u8 remote_id);
 int subflow_create_socket(struct sock *sk, struct socket **new_sock);
 
 extern const struct inet_connection_sock_af_ops ipv4_specific;
@@ -215,13 +223,16 @@ void mptcp_finish_join(struct sock *sk);
 void token_init(void);
 void token_new_request(struct request_sock *req, const struct sk_buff *skb);
 int token_join_request(struct request_sock *req, const struct sk_buff *skb);
+int token_join_response(struct sock *sk);
 int token_join_valid(struct request_sock *req,
 		     struct tcp_options_received *rx_opt);
 void token_destroy_request(u32 token);
 void token_new_connect(struct sock *sk);
+void token_new_subflow(struct sock *sk);
 void token_new_accept(struct sock *sk);
 int token_new_join(struct sock *sk);
 void token_update_accept(struct sock *sk, struct sock *conn);
+struct sock *token_lookup(u32 token);
 void token_release(u32 token);
 void token_destroy(u32 token);
 
net/mptcp/subflow.c +62 -1

87 lines not on this page · raw message

net/mptcp/token.c +63 -0

87 lines not on this page · raw message

Signature · 1 line
2.17.2
author
#19 lore raw

PATCH 9/9 mptcp: Implement interim path manager

In reply to Peter Krystad

Two features: 1) When a new connection occurs announce a local address and 2) When an outgoing connection is fully established (and a remote address was received) initiate a secondary subflow.

The second local address must be hard-coded for now.

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 net/mptcp/pm.c | 100 +++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 97 insertions(+), 3 deletions(-)
net/mptcp/pm.c +97 -3
index 0d53a0f23d2d..69df6332b4d0 100644
@@ -8,21 +8,84 @@
 #include <net/mptcp.h>
 #include "protocol.h"
 
+struct workqueue_struct *mptcp_wq;
+static void announce_addr_worker(struct work_struct *work);
+static void create_subflow_worker(struct work_struct *work);
+
 /* path manager command handlers */
 
 int pm_announce_addr(u32 token, sa_family_t family, u8 id, struct in_addr *addr)
 {
-	return -ENOTSUPP;
+	struct mptcp_sock *msk = mptcp_sk(token_lookup(token));
+	int err = 0;
+
+	if (!msk)
+		return -EINVAL;
+
+	if (msk->pm.local_valid) {
+		err = -EBADR;
+		goto announce_put;
+	}
+
+	pr_debug("msk=%p, id=%d, addr=%x", msk, id, addr->s_addr);
+	msk->pm.local_valid = 1;
+	msk->pm.local_id = id;
+	msk->pm.local_family = family;
+	msk->pm.local_addr.s_addr = addr->s_addr;
+	msk->addr_signal = 1;
+
+announce_put:
+	sock_put((struct sock *)msk);
+	return err;
 }
 
 int pm_remove_addr(u32 token, u8 id)
 {
-	return -ENOTSUPP;
+	struct mptcp_sock *msk = mptcp_sk(token_lookup(token));
+
+	if (!msk)
+		return -EINVAL;
+
+	pr_debug("msk=%p", msk);
+	msk->pm.local_valid = 0;
+
+	sock_put((struct sock *)msk);
+	return 0;
 }
 
 int pm_create_subflow(u32 token, u8 local_id, u8 remote_id)
 {
-	return -ENOTSUPP;
+	struct mptcp_sock *msk = mptcp_sk(token_lookup(token));
+	struct sockaddr_in remote;
+	struct sockaddr_in local;
+	int err;
+
+	if (!msk)
+		return -EINVAL;
+
+	pr_debug("msk=%p", msk);
+	if (!msk->pm.local_valid || local_id != msk->pm.local_id) {
+		err = -EBADR;
+		goto create_put;
+	}
+	local.sin_family = AF_INET;
+	local.sin_port = 0;
+	local.sin_addr.s_addr = INADDR_ANY;
+
+	if (!msk->pm.remote_valid || remote_id != msk->pm.remote_id) {
+		err = -EBADR;
+		goto create_put;
+	}
+
+	remote.sin_family = msk->pm.remote_family;
+	remote.sin_port = htons(msk->dport);
+	remote.sin_addr.s_addr = htonl(msk->pm.remote_addr.s_addr);
+
+	err = subflow_connect((struct sock *)msk, &local, &remote, remote_id);
+
+create_put:
+	sock_put((struct sock *)msk);
+	return err;
 }
 
 int pm_remove_subflow(u32 token, u8 local_id, u8 remote_id)
@@ -37,12 +100,22 @@ void pm_new_connection(struct mptcp_sock *msk, int server_side)
 	pr_debug("msk=%p", msk);
 
 	msk->pm.server_side = server_side;
+
+	/* trigger announce address in interim local path manager */
+	INIT_WORK(&msk->pm.addr_work, announce_addr_worker);
+	msk->pm.token = msk->token;
+	queue_work(mptcp_wq, &msk->pm.addr_work);
 }
 
 void pm_fully_established(struct mptcp_sock *msk)
 {
 	pr_debug("msk=%p", msk);
 
+	/* trigger create subflow in interim local local path manager */
+	if (!msk->pm.server_side && !msk->pm.fully_established) {
+		INIT_WORK(&msk->pm.subflow_work, create_subflow_worker);
+		queue_work(mptcp_wq, &msk->pm.subflow_work);
+	}
 	msk->pm.fully_established = 1;
 }
 
@@ -126,4 +199,25 @@ int pm_get_local_id(struct request_sock *req, struct sock *sk,
 
 void pm_init(void)
 {
+	mptcp_wq = alloc_workqueue("mptcp_wq", WQ_UNBOUND | WQ_MEM_RECLAIM, 8);
+	if (!mptcp_wq)
+		panic("Failed to allocate workqueue");
+}
+
+static void announce_addr_worker(struct work_struct *work)
+{
+	struct mptcp_pm_data *pm = container_of(work, struct mptcp_pm_data,
+						addr_work);
+	struct in_addr addr;
+
+	/* @@ hard-code address to announce here... */
+	pm_announce_addr(pm->token, AF_INET, 1, &addr);
+}
+
+static void create_subflow_worker(struct work_struct *work)
+{
+	struct mptcp_pm_data *pm = container_of(work, struct mptcp_pm_data,
+						subflow_work);
+
+	pm_create_subflow(pm->token, pm->local_id, pm->remote_id);
 }
Signature · 1 line
2.17.2
#7 lore raw

Re: [MPTCP] [PATCH 4/9] Re-factor and fixes for crypto_hmac_sha1()

In reply to Peter Krystad

Hi,

Quoted 12 lines · Peter Krystad

Re-factor to use parameters specific to MPTCP use case similar to

net/mptcp/token.c +0 -3
index c2f4fcb37566..c9160f50f047 100644
@@ -114,12 +114,11 @@ static void new_req_join(struct request_sock *req, struct sock *sk,
 	}
 	subflow_req->local_nonce = nonce;
 >local_key,
-			 msk->remote_key,
-			 (u32 *)hmac, 2,
-			 4, (u8 *)&subflow_req->local_nonce,

net/mptcp/token.c static void new_req_join(struct request_sock *req, struct sock *sk, · line 124

Quoted 10 lines
net/mptcp/token.c +5 -2
-			 4, (u8 *)&subflow_req->remote_nonce);
-	subflow_req->thmac = *(u64 *)hmac;
+	crypto_hmac_sha1(msk->local_key, msk->remote_key,
+			 subflow_req->local_nonce, subflow_req->remote_nonce,
+			 (u32 *)hmac);
+
+	subflow_req->thmac = get_unaligned_be64(hmac);
 	pr_debug("local_nonce=%u, thmac=%llu", subflow_req->local_nonce,
 		 subflow_req->thmac);
 }

It looks like this is going to conflict with Davide's pending patch "net: mptcp: randomness improvements for crypto.c"

Other than this, LGTM!

Paolo

#2 lore raw

Re: [MPTCP] [PATCH 1/9] Extend path manager interface

In reply to Peter Krystad

Hi,

I have some questions out of sheer ignorance on my side, please see below ;)

Quoted 20 lines · Peter Krystad
net/mptcp/pm.c +6 -3
@@ -48,19 +82,48 @@ void pm_rm_addr(struct mptcp_sock *msk, u8 id)
 	pr_debug("msk=%p", msk);
 }
 
+/* path manager helpers */
+
 bool pm_addr_signal(struct mptcp_sock *msk, unsigned int *size,
 		    unsigned int remaining, struct mptcp_out_options *opts)
 {
-	if (!msk || !msk->addr_signal)
+	if (!msk->pm.fully_established || !msk->addr_signal)
 		return false;
 
-	if (msk->pm.family == AF_INET && remaining < TCPOLEN_MPTCP_ADD_ADDR)
+	if (!msk->pm.local_valid)
 		return false;
 
-	pr_debug("msk=%p", msk);
+	if (msk->pm.local_family == AF_INET &&
+	    remaining < TCPOLEN_MPTCP_ADD_ADDR)

net/mptcp/pm.c void pm_rm_addr(struct mptcp_sock *msk, u8 id) · line 105

Quoted 10 lines
net/mptcp/pm.c +7 -2
+		return false;
+
+	pr_debug("msk=%p, addr_id=%d", msk, msk->pm.local_id);
 	opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
-	opts->addr_id = msk->pm.addr_id;
-	opts->addr.s_addr = msk->pm.addr.s_addr;
+	opts->addr_id = msk->pm.local_id;
+	opts->addr.s_addr = msk->pm.local_addr.s_addr;
+	*size = TCPOLEN_MPTCP_ADD_ADDR;
+	msk->addr_signal = 0;

Strictly speaking not related to this patch, but more to the existing code... perhaps 'opts' updating belong to options.c/mptcp_established_options_addr() more than the path manager?

Quoted 21 lines

index 7f15f6aab93d..8c04774ea2ec 100644

net/mptcp/protocol.h +11 -3
@@ -77,15 +77,33 @@ static inline u32 mptcp_option(u8 subopt, u8 len, u8 nib, u8 field)
 		     ((nib & 0xF) << 8) | field);
 }
 
-struct pm_data {
-	u8 addr_id;
-	sa_family_t family;
+struct mptcp_pm_data {
+	u8	local_valid;
+	u8	local_id;
+	sa_family_t local_family;
+	union {
+		struct in_addr local_addr;
+#if IS_ENABLED(CONFIG_IPV6)
+		struct in6_addr local_addr6;
+#endif
+	};
+	u8	remote_valid;

net/mptcp/protocol.h static inline u32 mptcp_option(u8 subopt, u8 len, u8 nib, u8 field) · line 98

Quoted 10 lines
net/mptcp/protocol.h +4 -2
+	u8	remote_id;
+	sa_family_t remote_family;
 	union {
-		struct in_addr addr;
+		struct in_addr remote_addr;
 #if IS_ENABLED(CONFIG_IPV6)
-		struct in6_addr addr6;
+		struct in6_addr remote_addr6;
 #endif
 	};

Are we going to need a list/container here? (to handle multiples addrs) I see is likely too early to cope with the extra complexity, just to have a clearer picture of the road ahead ;)

Cheers,

Paolo

#12 lore raw

Re: [MPTCP] [PATCH 7/9] Move setting request_ fields of subflow to MPTCP layer

In reply to Peter Krystad

net/mptcp/protocol.c static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr, · line 861

Quoted 13 lines · Peter Krystad
net/mptcp/protocol.c +5 -0
index 1664a69c8e30..fe73379a81df 100644
@@ -853,6 +854,11 @@ static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 			return err;
 	}
 
+	subflow = subflow_ctx(msk->subflow->sk);
+	subflow->request_mptcp = 1; /* @@ if MPTCP enabled */
+	subflow->request_cksum = 1; /* @@ if checksum enabled */
+	subflow->request_version = 0; /* only v0 supported */
+

Since this chunck is replicated a few lines below, and possibly some more code is going to be added to this section, with time, what about adding an helper for that?

Cheers,

Paolo

#15 lore raw

Re: [MPTCP] [PATCH 8/9] mptcp: Add handling of outgoing MP_JOIN requests

In reply to Peter Krystad
Quoted 5 lines · Peter Krystad
net/mptcp/token.c +1 -0
@@ -329,6 +381,17 @@ int token_new_join(struct sock *sk)
 	return -1;
 }
 
+struct sock *token_lookup(u32 token)

net/mptcp/token.c int token_new_join(struct sock *sk) · line 394

Quoted 10 lines
net/mptcp/token.c +10 -0
+{
+	struct sock *conn;
+
+	conn = lookup_token(token);
+	if (conn)
+		sock_hold(conn);
+
+	return conn;
+}
+

Really not a big deal, but this is small, unused in this patch and looks unrelated to the code added here. As it will be used by the next one, possibly it's cleaner move definition and declaration there.

Paolo

#17 lore raw

Re: [MPTCP] [PATCH 8/9] mptcp: Add handling of outgoing MP_JOIN requests

In reply to Peter Krystad

net/mptcp/token.c int token_new_join(struct sock *sk) · line 387

Quoted 8 lines · Peter Krystad
net/mptcp/token.c +4 -0
@@ -329,6 +381,17 @@ int token_new_join(struct sock *sk)
 	return -1;
 }
 
+struct sock *token_lookup(u32 token)
+{
+	struct sock *conn;
+

Possibly some locking is missing ??? like:

	spin_lock_bh(&token_tree_lock);

net/mptcp/token.c int token_new_join(struct sock *sk) · line 391

Quoted 4 lines
net/mptcp/token.c +4 -0
+	conn = lookup_token(token);
+	if (conn)
+		sock_hold(conn);
+
	spin_unlock_bh(&token_tree_lock);

Sorry for not noticing earlier,

Paolo

#20 lore raw

Re: [MPTCP] [PATCH 9/9] mptcp: Implement interim path manager

In reply to Peter Krystad
Quoted 5 lines · Peter Krystad
net/mptcp/pm.c +0 -0
index 0d53a0f23d2d..69df6332b4d0 100644
@@ -8,21 +8,84 @@

[...]

Quoted 9 lines

int pm_create_subflow(u32 token, u8 local_id, u8 remote_id) {

net/mptcp/pm.c +6 -1
-	return -ENOTSUPP;
+	struct mptcp_sock *msk = mptcp_sk(token_lookup(token));
+	struct sockaddr_in remote;
+	struct sockaddr_in local;
+	int err;
+
+	if (!msk)

net/mptcp/pm.c line 73

Quoted 10 lines
net/mptcp/pm.c +10 -0
+		return -EINVAL;
+
+	pr_debug("msk=%p", msk);
+	if (!msk->pm.local_valid || local_id != msk->pm.local_id) {
+		err = -EBADR;
+		goto create_put;
+	}
+	local.sin_family = AF_INET;
+	local.sin_port = 0;
+	local.sin_addr.s_addr = INADDR_ANY;

It's unclear to me the 'local_id' usage here. I thought 'local_id == msk->pm.local_id' would imply 'local.sin_addr.s_addr == msk-

Quoted 1 line

pm.local_addr.s_addr' ???

Is the PM supposed to validate the local address vs. the known one? or should it allow binding to arbitrary local address? I thought the local address should be usually unspecified - otherwise the PM should have a deep knowledge of the routing tables/policies to get the correct local address for a given destination, likely resorting to a route lookup before pm_create_subflow() ???

net/mptcp/pm.c void pm_new_connection(struct mptcp_sock *msk, int server_side) · line 107

Quoted 9 lines
net/mptcp/pm.c +5 -0
@@ -37,12 +100,22 @@ void pm_new_connection(struct mptcp_sock *msk, int server_side)
 	pr_debug("msk=%p", msk);
 
 	msk->pm.server_side = server_side;
+
+	/* trigger announce address in interim local path manager */
+	INIT_WORK(&msk->pm.addr_work, announce_addr_worker);
+	msk->pm.token = msk->token;
+	queue_work(mptcp_wq, &msk->pm.addr_work);

IIRC we not going to support subflow opening from the server towards the client - at least initially. Can we restrict announce to server side only?

Quoted 8 lines
net/mptcp/pm.c +4 -0
@@ -126,4 +199,25 @@ int pm_get_local_id(struct request_sock *req, struct sock *sk,
 
 void pm_init(void)
 {
+	mptcp_wq = alloc_workqueue("mptcp_wq", WQ_UNBOUND | WQ_MEM_RECLAIM, 8);
+	if (!mptcp_wq)
+		panic("Failed to allocate workqueue");
+}

net/mptcp/pm.c int pm_get_local_id(struct request_sock *req, struct sock *sk, · line 215

Quoted 10 lines
net/mptcp/pm.c +10 -0
+
+static void announce_addr_worker(struct work_struct *work)
+{
+	struct mptcp_pm_data *pm = container_of(work, struct mptcp_pm_data,
+						addr_work);
+	struct in_addr addr;
+
+	/* @@ hard-code address to announce here... */
+	pm_announce_addr(pm->token, AF_INET, 1, &addr);
+}

What about a temporary (to be dropped with a later patch when a saner interface will be available) boot parameter/sysctl/other nasty and quick hack to select/disable/enable the announce address, so that we can start self-testing this code early?

Thanks!

Paolo

#22 lore raw
In reply to Peter Krystad

Hi,

Quoted 2 lines · Peter Krystad

This is several accumulated fixes that are required for MP_JOIN (secondary subflows) plus the outgoing code.

This is great work! Thank you!

I had a few question on individual patches, but I think the only functionally relevant is about locking on 8/9.

Quoted 6 lines

Both outgoing and incoming MP_JOIN now inter-operate with the multipath.org implementation although there is still a deadlock when incoming is fully established. Since the incoming direction did not work at all before this is an improvement. I have not yet tested running our implmentation against itself.

I think we can possibly have to revisit part of the current locking schema - specifically I think we could/should substiture acquiring the msk lock in *_finish_connect() with a new spinlock.

Quoted 8 lines

The final patch in the set is an "interim path manager", this is throw-away code that I include for testing purposes. For an incoming connection it is necessary for the server to advertise a secondary address and for my testing I have been hard-coding the IP address of my secondary interface when I build the kernel. Obviously not the final solution but we need to discuss what approach to use for the self tests. I also think we may not want to publish this patch on our net-next branch.

I think that patch is actually good/needed - at least 'transiently'. I think we can expand it a bit to allow self-testing on different hosts/setups without re-compiling the kernel.

Perhpas it can be split in 2 parts, the first one containing the code we intend to keep in the git repo (e.g. pm_create_subflow(), pm_remove_addr() and pm_create_subflow() looks goot to me!) and the 2nd half contaning only the hacks to enable self-tests - to be reverted dropped later.

WDYT?

Paolo

author
#3 lore raw

Re: [MPTCP] [PATCH 1/9] Extend path manager interface

In reply to Paolo Abeni
Quoted 32 lines · Paolo Abeni

Hi,

I have some questions out of sheer ignorance on my side, please see below ;)

Quoted 26 lines · Peter Krystad
net/mptcp/pm.c +9 -5
@@ -48,19 +82,48 @@ void pm_rm_addr(struct mptcp_sock *msk, u8 id)
 	pr_debug("msk=%p", msk);
 }
 
+/* path manager helpers */
+
 bool pm_addr_signal(struct mptcp_sock *msk, unsigned int *size,
 		    unsigned int remaining, struct mptcp_out_options *opts)
 {
-	if (!msk || !msk->addr_signal)
+	if (!msk->pm.fully_established || !msk->addr_signal)
 		return false;
 
-	if (msk->pm.family == AF_INET && remaining < TCPOLEN_MPTCP_ADD_ADDR)
+	if (!msk->pm.local_valid)
 		return false;
 
-	pr_debug("msk=%p", msk);
+	if (msk->pm.local_family == AF_INET &&
+	    remaining < TCPOLEN_MPTCP_ADD_ADDR)
+		return false;
+
+	pr_debug("msk=%p, addr_id=%d", msk, msk->pm.local_id);
 	opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
-	opts->addr_id = msk->pm.addr_id;
-	opts->addr.s_addr = msk->pm.addr.s_addr;
Quoted 8 lines

net/mptcp/pm.c void pm_rm_addr(struct mptcp_sock *msk, u8 id) · line 105

Quoted 4 lines
net/mptcp/pm.c +4 -0
+	opts->addr_id = msk->pm.local_id;
+	opts->addr.s_addr = msk->pm.local_addr.s_addr;
+	*size = TCPOLEN_MPTCP_ADD_ADDR;
+	msk->addr_signal = 0;

Strictly speaking not related to this patch, but more to the existing code... perhaps 'opts' updating belong to options.c/mptcp_established_options_addr() more than the path manager?

Agreed, I'll revise.

Quoted 28 lines
Quoted 28 lines

index 7f15f6aab93d..8c04774ea2ec 100644

net/mptcp/protocol.h +14 -4
@@ -77,15 +77,33 @@ static inline u32 mptcp_option(u8 subopt, u8 len, u8 nib, u8 field)
 		     ((nib & 0xF) << 8) | field);
 }
 
-struct pm_data {
-	u8 addr_id;
-	sa_family_t family;
+struct mptcp_pm_data {
+	u8	local_valid;
+	u8	local_id;
+	sa_family_t local_family;
+	union {
+		struct in_addr local_addr;
+#if IS_ENABLED(CONFIG_IPV6)
+		struct in6_addr local_addr6;
+#endif
+	};
+	u8	remote_valid;
+	u8	remote_id;
+	sa_family_t remote_family;
 	union {
-		struct in_addr addr;
+		struct in_addr remote_addr;
 #if IS_ENABLED(CONFIG_IPV6)
Quoted 9 lines

net/mptcp/protocol.h static inline u32 mptcp_option(u8 subopt, u8 len, u8 nib, u8 field) · line 98

Quoted 4 lines
net/mptcp/protocol.h +1 -1
-		struct in6_addr addr6;
+		struct in6_addr remote_addr6;
 #endif
 	};

Are we going to need a list/container here? (to handle multiples addrs) I see is likely too early to cope with the extra complexity, just to have a clearer picture of the road ahead ;)

Indeed. For now this is a list of length one. :)

Although you night argue for some of our primary use cases this is probably enough...

Peter.

Quoted 4 lines

Cheers,

Paolo

author
#8 lore raw

Re: [MPTCP] [PATCH 4/9] Re-factor and fixes for crypto_hmac_sha1()

In reply to Paolo Abeni
Quoted 20 lines · Paolo Abeni

Hi,

Quoted 17 lines · Peter Krystad

Re-factor to use parameters specific to MPTCP use case similar to

net/mptcp/token.c +3 -5
index c2f4fcb37566..c9160f50f047 100644
@@ -114,12 +114,11 @@ static void new_req_join(struct request_sock *req, struct sock *sk,
 	}
 	subflow_req->local_nonce = nonce;
 >local_key,
-			 msk->remote_key,
-			 (u32 *)hmac, 2,
-			 4, (u8 *)&subflow_req->local_nonce,
-			 4, (u8 *)&subflow_req->remote_nonce);
-	subflow_req->thmac = *(u64 *)hmac;
+	crypto_hmac_sha1(msk->local_key, msk->remote_key,
+			 subflow_req->local_nonce, subflow_req->remote_nonce,
+			 (u32 *)hmac);
Quoted 8 lines

net/mptcp/token.c static void new_req_join(struct request_sock *req, struct sock *sk, · line 124

Quoted 5 lines
net/mptcp/token.c +2 -0
+
+	subflow_req->thmac = get_unaligned_be64(hmac);
 	pr_debug("local_nonce=%u, thmac=%llu", subflow_req->local_nonce,
 		 subflow_req->thmac);
 }

It looks like this is going to conflict with Davide's pending patch "net: mptcp: randomness improvements for crypto.c"

I will revise this once his is merged.

Peter.

Quoted 5 lines

Other than this, LGTM!

Paolo

author
#13 lore raw

Re: [MPTCP] [PATCH 7/9] Move setting request_ fields of subflow to MPTCP layer

In reply to Paolo Abeni
Quoted 9 lines · Paolo Abeni
Quoted 8 lines · Peter Krystad
net/mptcp/protocol.c +0 -0
index 1664a69c8e30..fe73379a81df 100644
@@ -853,6 +854,11 @@ static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 			return err;
 	}
 
Quoted 9 lines

net/mptcp/protocol.c static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr, · line 861

Quoted 5 lines
net/mptcp/protocol.c +5 -0
+	subflow = subflow_ctx(msk->subflow->sk);
+	subflow->request_mptcp = 1; /* @@ if MPTCP enabled */
+	subflow->request_cksum = 1; /* @@ if checksum enabled */
+	subflow->request_version = 0; /* only v0 supported */
+

Since this chunck is replicated a few lines below, and possibly some more code is going to be added to this section, with time, what about adding an helper for that?

OK, will do.

Quoted 5 lines

Cheers,

Paolo

author
#16 lore raw

Re: [MPTCP] [PATCH 8/9] mptcp: Add handling of outgoing MP_JOIN requests

In reply to Paolo Abeni
Quoted 5 lines · Paolo Abeni
Quoted 4 lines · Peter Krystad
net/mptcp/token.c +0 -0
@@ -329,6 +381,17 @@ int token_new_join(struct sock *sk)
 	return -1;
 }
 
Quoted 15 lines

net/mptcp/token.c int token_new_join(struct sock *sk) · line 394

Quoted 11 lines
net/mptcp/token.c +11 -0
+struct sock *token_lookup(u32 token)
+{
+	struct sock *conn;
+
+	conn = lookup_token(token);
+	if (conn)
+		sock_hold(conn);
+
+	return conn;
+}
+

Really not a big deal, but this is small, unused in this patch and looks unrelated to the code added here. As it will be used by the next one, possibly it's cleaner move definition and declaration there.

Yes, if we split the next patch as you suggest then I can move this to the "keeper" of those two patches.

Peter.

Quoted 5 lines

Paolo

author
#18 lore raw

Re: [MPTCP] [PATCH 8/9] mptcp: Add handling of outgoing MP_JOIN requests

In reply to Paolo Abeni
Quoted 12 lines · Paolo Abeni

net/mptcp/token.c int token_new_join(struct sock *sk) · line 387

Quoted 8 lines · Peter Krystad
net/mptcp/token.c +4 -0
@@ -329,6 +381,17 @@ int token_new_join(struct sock *sk)
 	return -1;
 }
 
+struct sock *token_lookup(u32 token)
+{
+	struct sock *conn;
+

Possibly some locking is missing ??? like:

Quoted 7 lines
	spin_lock_bh(&token_tree_lock);

net/mptcp/token.c int token_new_join(struct sock *sk) · line 391

Quoted 4 lines
net/mptcp/token.c +4 -0
+	conn = lookup_token(token);
+	if (conn)
+		sock_hold(conn);
+
	spin_unlock_bh(&token_tree_lock);

Yes, I think you are right. Will clean up.

Peter.

Quoted 4 lines

Sorry for not noticing earlier,

Paolo

author
#21 lore raw

Re: [MPTCP] [PATCH 9/9] mptcp: Implement interim path manager

In reply to Paolo Abeni

Sorry for not responding to these last few comments Paolo, here we are:

Quoted 29 lines · Paolo Abeni
Quoted 5 lines · Peter Krystad
net/mptcp/pm.c +0 -0
index 0d53a0f23d2d..69df6332b4d0 100644
@@ -8,21 +8,84 @@

[...]

Quoted 9 lines

int pm_create_subflow(u32 token, u8 local_id, u8 remote_id) {

net/mptcp/pm.c +6 -1
-	return -ENOTSUPP;
+	struct mptcp_sock *msk = mptcp_sk(token_lookup(token));
+	struct sockaddr_in remote;
+	struct sockaddr_in local;
+	int err;
+
+	if (!msk)

net/mptcp/pm.c line 73

Quoted 10 lines
net/mptcp/pm.c +10 -0
+		return -EINVAL;
+
+	pr_debug("msk=%p", msk);
+	if (!msk->pm.local_valid || local_id != msk->pm.local_id) {
+		err = -EBADR;
+		goto create_put;
+	}
+	local.sin_family = AF_INET;
+	local.sin_port = 0;
+	local.sin_addr.s_addr = INADDR_ANY;

It's unclear to me the 'local_id' usage here. I thought 'local_id == msk->pm.local_id' would imply 'local.sin_addr.s_addr == msk-

Quoted 8 lines
Quoted 1 line

pm.local_addr.s_addr' ???

Is the PM supposed to validate the local address vs. the known one? or should it allow binding to arbitrary local address? I thought the local address should be usually unspecified - otherwise the PM should have a deep knowledge of the routing tables/policies to get the correct local address for a given destination, likely resorting to a route lookup before pm_create_subflow() ???

Yes, for our purposes I agree the bind address should always be unspecified. I will remove the local_id parameter and the check against it.

Quoted 4 lines
Quoted 4 lines
net/mptcp/pm.c +0 -0
@@ -37,12 +100,22 @@ void pm_new_connection(struct mptcp_sock *msk, int server_side)
 	pr_debug("msk=%p", msk);
 
Quoted 10 lines

net/mptcp/pm.c void pm_new_connection(struct mptcp_sock *msk, int server_side) · line 107

Quoted 6 lines
net/mptcp/pm.c +5 -0
 	msk->pm.server_side = server_side;
+
+	/* trigger announce address in interim local path manager */
+	INIT_WORK(&msk->pm.addr_work, announce_addr_worker);
+	msk->pm.token = msk->token;
+	queue_work(mptcp_wq, &msk->pm.addr_work);

IIRC we not going to support subflow opening from the server towards the client - at least initially. Can we restrict announce to server side only?

Good point, I'll add this restriction.

Quoted 16 lines
Quoted 16 lines
net/mptcp/pm.c +11 -0
@@ -126,4 +199,25 @@ int pm_get_local_id(struct request_sock *req, struct sock *sk,
 
 void pm_init(void)
 {
+	mptcp_wq = alloc_workqueue("mptcp_wq", WQ_UNBOUND | WQ_MEM_RECLAIM, 8);
+	if (!mptcp_wq)
+		panic("Failed to allocate workqueue");
+}
+
+static void announce_addr_worker(struct work_struct *work)
+{
+	struct mptcp_pm_data *pm = container_of(work, struct mptcp_pm_data,
+						addr_work);
+	struct in_addr addr;
+
Quoted 8 lines

net/mptcp/pm.c int pm_get_local_id(struct request_sock *req, struct sock *sk, · line 215

Quoted 3 lines
net/mptcp/pm.c +3 -0
+	/* @@ hard-code address to announce here... */
+	pm_announce_addr(pm->token, AF_INET, 1, &addr);
+}

What about a temporary (to be dropped with a later patch when a saner interface will be available) boot parameter/sysctl/other nasty and quick hack to select/disable/enable the announce address, so that we can start self-testing this code early?

How about once Matthieu's sysctl change is merged I will add such fields (a value field and a enable/disable field) to the interim path manager patch? I plan to separate that patch from this patchset so it stands alone as the "throwaway" patch.

Peter.

Quoted 5 lines

Thanks!

Paolo

author
#23 lore raw
In reply to Paolo Abeni
Quoted 18 lines · Paolo Abeni

Hi,

Quoted 2 lines · Peter Krystad

This is several accumulated fixes that are required for MP_JOIN (secondary subflows) plus the outgoing code.

This is great work! Thank you!

I had a few question on individual patches, but I think the only functionally relevant is about locking on 8/9.

Quoted 6 lines

Both outgoing and incoming MP_JOIN now inter-operate with the multipath.org implementation although there is still a deadlock when incoming is fully established. Since the incoming direction did not work at all before this is an improvement. I have not yet tested running our implmentation against itself.

Quoted 3 lines

I think we can possibly have to revisit part of the current locking schema - specifically I think we could/should substiture acquiring the msk lock in *_finish_connect() with a new spinlock.

I suspect this may be true unless I resolve above mention deadlock....

Quoted 14 lines
Quoted 9 lines

The final patch in the set is an "interim path manager", this is throw-away code that I include for testing purposes. For an incoming connection it is necessary for the server to advertise a secondary address and for my testing I have been hard-coding the IP address of my secondary interface when I build the kernel. Obviously not the final solution but we need to discuss what approach to use for the self tests. I also think we may not want to publish this patch on our net-next branch.

I think that patch is actually good/needed - at least 'transiently'. I think we can expand it a bit to allow self-testing on different hosts/setups without re-compiling the kernel.

Quoted 7 lines

Perhpas it can be split in 2 parts, the first one containing the code we intend to keep in the git repo (e.g. pm_create_subflow(), pm_remove_addr() and pm_create_subflow() looks goot to me!) and the 2nd half contaning only the hacks to enable self-tests - to be reverted dropped later.

WDYT?

Good idea. For v2 of the MP_JOIN patchset I will split it like this, with the keeper code in this patchset and the throwaway code in a patch that I will post separately (with sysctl for announce address).

Peter.

Quoted 5 lines

Paolo

#24 lore raw
In reply to Peter Krystad
Quoted 14 lines · Peter Krystad

This is several accumulated fixes that are required for MP_JOIN (secondary subflows) plus the outgoing code.

Both outgoing and incoming MP_JOIN now inter-operate with the multipath.org implementation although there is still a deadlock when incoming is fully established. Since the incoming direction did not work at all before this is an improvement. I have not yet tested running our implmentation against itself.

Quoted 4 lines

The final patch in the set is an "interim path manager", this is throw-away code that I include for testing purposes. For an incoming connection it is necessary for the server to advertise a secondary address

Just as an FYI:

if you are testing the upstreaming-MPTCP on the server-side and multipath-tcp.org <http://multipath-tcp.org/> on the client-side, you can configure on the client the ndiffports path-manager. That makes the client establish subflows even if there are not multiple IPs on the host, simply among the same IP-addresses by using different port-numbers.

That way you don't need to force the ADD_ADDR from the server.

Cheers,
Christoph

Quoted 35 lines

and for my testing I have been hard-coding the IP address of my secondary interface when I build the kernel. Obviously not the final solution but we need to discuss what approach to use for the self tests. I also think we may not want to publish this patch on our net-next branch.

Peter Krystad (9):
Extend path manager interface
Fix locking in mptcp_finish_join.
Fix parsing MP_JOIN third ACK.
Re-factor and fixes for crypto_hmac_sha1() Rename mptcp_attach_dss()
Fix return value if no DSS option is created. Move setting request_ fields of subflow to MPTCP layer
mptcp: Add handling of outgoing MP_JOIN requests
mptcp: Implement interim path manager

include/net/mptcp.h  |  12 +--
net/ipv4/tcp_input.c |   4 +-
net/mptcp/crypto.c   |  23 +++---
net/mptcp/options.c  |  77 ++++++++++++++++---
net/mptcp/pm.c       | 171 +++++++++++++++++++++++++++++++++++++++++--
net/mptcp/protocol.c |  29 ++++++--
net/mptcp/protocol.h |  54 +++++++++++---
net/mptcp/subflow.c  |  72 ++++++++++++++++--
net/mptcp/token.c    |  82 ++++++++++++++++++---
9 files changed, 452 insertions(+), 72 deletions(-)
Signature · 6 lines
2.17.2

_______________________________________________
mptcp mailing list
mptcp(a)lists.01.org
https://lists.01.org/mailman/listinfo/mptcp