Multipath TCP prerequisites
Subsystems (6)
v3 cover Multipath TCP prerequisites
The MPTCP upstreaming community has prepared a net-next RFCv3 patch set for review. The scope of this patch set is limited to prerequisite TCP core changes so we can get focused feedback in these areas.
In this patch set we introduce some MPTCP definitions, additional ULP and skb extension features, TCP option space checking, and a few exported symbols.
Our intent is to send this as a non-RFC series when the next phase of changes are ready to post. That second patch set will add CONFIG_MPTCP in Kconfig, introduce the MPTCP socket type, implement the basic features of the protocol, and add self tests.
Clone/fetch:
https://github.com/multipath-tcp/mptcp_net-next.git (tag: netdev-rfcv3)
Browse:
https://github.com/multipath-tcp/mptcp_net-next/tree/netdev-rfcv3
Thank you for your review. You can find us at [email protected] 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 (1):
tcp: clean ext on tx recycle
include/linux/skbuff.h | 3 +++ include/net/mptcp.h | 43 +++++++++++++++++++++++++++++++++ include/net/sock.h | 6 ++--- include/net/tcp.h | 19 +++++++++++++++ include/trace/events/sock.h | 5 ++-- include/uapi/linux/in.h | 2 ++ net/ax25/af_ax25.c | 2 +- net/core/skbuff.c | 7 ++++++ 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 | 5 +++- net/ipv4/tcp_ulp.c | 12 +++++++++ tools/include/uapi/linux/in.h | 2 ++ 16 files changed, 114 insertions(+), 14 deletions(-) create mode 100644 include/net/mptcp.h
Signature · 1 line
2.23.0
RFC PATCH v3 1/10 net: Make sock protocol value checks more specific
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.
--- include/net/sock.h | 1 - net/ax25/af_ax25.c | 2 +- net/decnet/af_decnet.c | 2 +- 3 files changed, 2 insertions(+), 3 deletions(-)
include/net/sock.h +0 -1
@@ -459,7 +459,6 @@ struct sock { sk_userlocks : 4, sk_protocol : 8, sk_type : 16; -#define SK_PROTOCOL_MAX U8_MAXu16 sk_gso_max_segs;u8 sk_pacing_shift;unsigned long sk_lingertime;
net/ax25/af_ax25.c +1 -1
@@ -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))
net/decnet/af_decnet.c +1 -1
@@ -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))
Signature · 1 line
2.23.0
RFC PATCH v3 2/10 sock: Make sk_protocol a 16-bit value
In reply to Mat MartineauMatch the 16-bit width of skbuff->protocol. Fills an 8-bit hole so sizeof(struct sock) does not change.
--- include/net/sock.h | 4 ++-- include/trace/events/sock.h | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-)
include/net/sock.h +2 -2
@@ -457,10 +457,10 @@ struct sock { sk_no_check_tx : 1, sk_no_check_rx : 1, sk_userlocks : 4, - sk_protocol : 8, + sk_pacing_shift : 8, sk_type : 16; +u16 sk_protocol;u16 sk_gso_max_segs; -u8 sk_pacing_shift;unsigned long sk_lingertime;struct proto *sk_prot_creator;rwlock_t sk_callback_lock;
include/trace/events/sock.h +1 -1
@@ -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 )
Signature · 1 line
2.23.0
RFC PATCH v3 3/10 tcp: Define IPPROTO_MPTCP
In reply to Mat MartineauTo 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.
--- 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(-)
include/trace/events/sock.h +2 -1
@@ -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) \
include/uapi/linux/in.h +2 -0
@@ -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_MPTCPIPPROTO_MAX };#endif
tools/include/uapi/linux/in.h +2 -0
@@ -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_MPTCPIPPROTO_MAX };#endif
Signature · 1 line
2.23.0
RFC PATCH v3 4/10 tcp: Add MPTCP option number
In reply to Mat MartineauTCP option 30 is allocated for MPTCP by the IANA.
--- include/net/tcp.h | 1 + 1 file changed, 1 insertion(+)
include/net/tcp.h +1 -0
@@ -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
Signature · 1 line
2.23.0
RFC PATCH v3 5/10 tcp, ulp: Add clone operation to tcp_ulp_ops
In reply to Mat MartineauIf 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.
--- include/net/tcp.h | 5 +++++ net/ipv4/inet_connection_sock.c | 2 ++ net/ipv4/tcp_ulp.c | 12 ++++++++++++ 3 files changed, 19 insertions(+)
include/net/tcp.h +5 -0
@@ -2126,6 +2126,9 @@ 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;
@@ -2136,6 +2139,8 @@ 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); \
net/ipv4/inet_connection_sock.c +2 -0
@@ -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;
net/ipv4/tcp_ulp.c +12 -0
@@ -127,6 +127,18 @@ 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);
Signature · 1 line
2.23.0
RFC PATCH v3 6/10 mptcp: Add MPTCP to skb extensions
In reply to Mat MartineauAdd enum value for MPTCP and update config dependencies
--- include/linux/skbuff.h | 3 +++ include/net/mptcp.h | 27 +++++++++++++++++++++++++++ net/core/skbuff.c | 7 +++++++ 3 files changed, 37 insertions(+) create mode 100644 include/net/mptcp.h
include/linux/skbuff.h +3 -0
@@ -4068,6 +4068,9 @@ 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 */ };
include/net/mptcp.h new file +27 -0
@@ -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 */
net/core/skbuff.c +7 -0
@@ -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 ; }
Signature · 1 line
2.23.0
RFC PATCH v3 7/10 tcp: Prevent coalesce/collapse when skb has MPTCP extensions
In reply to Mat MartineauThe 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.
--- 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(-)
include/net/mptcp.h +16 -0
@@ -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 */
include/net/tcp.h +8 -0
@@ -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>
@@ -961,6 +962,13 @@ 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 */
net/ipv4/tcp_input.c +8 -2
@@ -1420,7 +1420,7 @@ 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) &&
@@ -4418,6 +4418,9 @@ 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;
@@ -4926,10 +4929,12 @@ 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;
@@ -4945,7 +4950,7 @@ 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 ;
@@ -4988,6 +4993,7 @@ 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
net/ipv4/tcp_output.c +1 -1
@@ -2853,7 +2853,7 @@ 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;
Signature · 1 line
2.23.0
RFC PATCH v3 8/10 tcp: Export TCP functions and ops struct
In reply to Mat MartineauMPTCP will make use of tcp_send_mss() and tcp_push() when sending data to specific TCP subflows.
tcp_request_sock_ipv4_ops will be referenced during TCP subflow creation.
--- include/net/tcp.h | 5 +++++ net/ipv4/tcp.c | 6 +++--- net/ipv4/tcp_ipv4.c | 2 +- 3 files changed, 9 insertions(+), 4 deletions(-)
include/net/tcp.h +5 -0
@@ -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);
@@ -1983,6 +1986,8 @@ struct tcp_request_sock_ops { enum tcp_synack_type synack_type); }; +extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops; +#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,
net/ipv4/tcp.c +3 -3
@@ -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;
net/ipv4/tcp_ipv4.c +1 -1
@@ -1369,7 +1369,7 @@ 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,
Signature · 1 line
2.23.0
RFC PATCH v3 9/10 tcp: Check for filled TCP option space before SACK
In reply to Mat MartineauThe SACK code would potentially add four bytes to the expected TCP option size even if all option space was already used.
--- net/ipv4/tcp_output.c | 3 +++ 1 file changed, 3 insertions(+)
net/ipv4/tcp_output.c +3 -0
@@ -748,6 +748,9 @@ static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb size +=TCPOLEN_TSTAMP_ALIGNED ; } +if (size +TCPOLEN_SACK_BASE_ALIGNED >=MAX_TCP_OPTION_SPACE ) +return size; + eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack;if (unlikely (eff_sacks)) {const unsigned int remaining =MAX_TCP_OPTION_SPACE - size;
Signature · 1 line
2.23.0
RFC PATCH v3 10/10 tcp: clean ext on tx recycle
In reply to Mat MartineauFrom: Paolo Abeni <[email protected]>
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 the mptcp stack will not be aware of it, so nobody set properly the MPTCP ext.
End result, we transmit the skb using an hold MPTCP DSS map and that confuses the rx side/corrupt the stream. It requires some concurrent conditions, so it's not deterministic.
Resetting the ext on recycle fixes all the current mptcp self tests issues.
Apparently only MPTCP has issues with this kind of stray ext, so an alternative would be add an additional mptcp hook in tso_fragment() or in sk_stream_alloc_skb() to always init the ext.
--- include/net/sock.h | 1 + 1 file changed, 1 insertion(+)
include/net/sock.h +1 -0
@@ -1470,6 +1470,7 @@ 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 ;
Signature · 1 line
2.23.0