tcp: provide correct skb->priority
Subsystems (5)
cover tcp: provide correct skb->priority
SO_PRIORITY socket option requests TCP egress packets to contain a user provided value.
TCP manages to send most packets with the requested values, notably for TCP_ESTABLISHED state, but fails to do so for few packets.
These packets are control packets sent on behalf of SYN_RECV or TIME_WAIT states.
Note that to test this with packetdrill, it is a bit of a hassle, since packetdrill can not verify priority of egress packets, other than indirect observations, using for example sch_prio on its tunnel device.
The bad skb priorities cause problems for GCP,
as this field is one of the keys used in routing.
Eric Dumazet (3):
ipv6: add priority parameter to ip6_xmit()
ipv6: tcp: provide sk->sk_priority to ctl packets
tcp: honor SO_PRIORITY in TIME_WAIT state
include/net/inet_timewait_sock.h | 1 + include/net/ipv6.h | 2 +- net/dccp/ipv6.c | 5 +++-- net/ipv4/ip_output.c | 1 - net/ipv4/tcp_ipv4.c | 4 ++++ net/ipv4/tcp_minisocks.c | 1 + net/ipv6/inet6_connection_sock.c | 2 +- net/ipv6/ip6_output.c | 4 ++-- net/ipv6/tcp_ipv6.c | 24 +++++++++++++++--------- net/sctp/ipv6.c | 2 +- 10 files changed, 29 insertions(+), 17 deletions(-)
Signature · 1 line
2.23.0.351.gc4317032e6-goog
PATCH 1/3 ipv6: add priority parameter to ip6_xmit()
Currently, ip6_xmit() sets skb->priority based on sk->sk_priority
This is not desirable for TCP since TCP shares the same ctl socket for a given netns. We want to be able to send RST or ACK packets with a non zero skb->priority.
This patch has no functional change.
--- include/net/ipv6.h | 2 +- net/dccp/ipv6.c | 5 +++-- net/ipv6/inet6_connection_sock.c | 2 +- net/ipv6/ip6_output.c | 4 ++-- net/ipv6/tcp_ipv6.c | 6 ++++-- net/sctp/ipv6.c | 2 +- 6 files changed, 12 insertions(+), 9 deletions(-)
include/net/ipv6.h +1 -1
@@ -981,7 +981,7 @@ int ip6_rcv_finish(struct net *net, struct sock *sk, struct sk_buff *skb);* upper-layer output functions */ int ip6_xmit (const struct sock *sk,struct sk_buff *skb,struct flowi6 *fl6, -__u32 mark,struct ipv6_txoptions *opt,int tclass); +__u32 mark,struct ipv6_txoptions *opt,int tclass,u32 priority);int ip6_find_1stfragopt (struct sk_buff *skb,u8 **nexthdr);
net/dccp/ipv6.c +3 -2
@@ -230,7 +230,8 @@ static int dccp_v6_send_response(const struct sock *sk, struct request_sock *req opt = ireq->ipv6_opt;if (!opt) opt =rcu_dereference (np->opt); - err =ip6_xmit (sk, skb, &fl6, sk->sk_mark, opt, np->tclass); + err =ip6_xmit (sk, skb, &fl6, sk->sk_mark, opt, np->tclass, + sk->sk_priority);rcu_read_unlock (); err =net_xmit_eval (err); }
@@ -284,7 +285,7 @@ static void dccp_v6_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb) dst =ip6_dst_lookup_flow (ctl_sk, &fl6,NULL );if (!IS_ERR (dst)) {skb_dst_set (skb, dst); -ip6_xmit (ctl_sk, skb, &fl6,0 ,NULL ,0 ); +ip6_xmit (ctl_sk, skb, &fl6,0 ,NULL ,0 ,0 );DCCP_INC_STATS (DCCP_MIB_OUTSEGS );DCCP_INC_STATS (DCCP_MIB_OUTRSTS );return ;
net/ipv6/inet6_connection_sock.c +1 -1
@@ -133,7 +133,7 @@ int inet6_csk_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl_unused fl6.daddr = sk->sk_v6_daddr; res =ip6_xmit (sk, skb, &fl6, sk->sk_mark,rcu_dereference (np->opt), - np->tclass); + np->tclass, sk->sk_priority);rcu_read_unlock ();return res; }
net/ipv6/ip6_output.c +2 -2
@@ -193,7 +193,7 @@ bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)* which are using proper atomic operations or spinlocks. */ int ip6_xmit (const struct sock *sk,struct sk_buff *skb,struct flowi6 *fl6, -__u32 mark,struct ipv6_txoptions *opt,int tclass) +__u32 mark,struct ipv6_txoptions *opt,int tclass,u32 priority) {struct net *net =sock_net (sk);const struct ipv6_pinfo *np =inet6_sk (sk);
@@ -258,7 +258,7 @@ int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6, hdr->daddr = *first_hop; skb->protocol =htons (ETH_P_IPV6 ); - skb->priority = sk->sk_priority; + skb->priority = priority; skb->mark = mark; mtu =dst_mtu (dst);
net/ipv6/tcp_ipv6.c +4 -2
@@ -512,7 +512,8 @@ static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst, opt = ireq->ipv6_opt;if (!opt) opt =rcu_dereference (np->opt); - err =ip6_xmit (sk, skb, fl6, sk->sk_mark, opt, np->tclass); + err =ip6_xmit (sk, skb, fl6, sk->sk_mark, opt, np->tclass, + sk->sk_priority);rcu_read_unlock (); err =net_xmit_eval (err); }
@@ -907,7 +908,8 @@ static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 dst =ip6_dst_lookup_flow (ctl_sk, &fl6,NULL );if (!IS_ERR (dst)) {skb_dst_set (buff, dst); -ip6_xmit (ctl_sk, buff, &fl6, fl6.flowi6_mark,NULL , tclass); +ip6_xmit (ctl_sk, buff, &fl6, fl6.flowi6_mark,NULL , tclass, +0 );TCP_INC_STATS (net,TCP_MIB_OUTSEGS );if (rst)TCP_INC_STATS (net,TCP_MIB_OUTRSTS );
net/sctp/ipv6.c +1 -1
@@ -215,7 +215,7 @@ static int sctp_v6_xmit(struct sk_buff *skb, struct sctp_transport *transport)rcu_read_lock (); res =ip6_xmit (sk, skb, fl6, sk->sk_mark,rcu_dereference (np->opt), - tclass); + tclass, sk->sk_priority);rcu_read_unlock ();return res; }
Signature · 1 line
2.23.0.351.gc4317032e6-goog
PATCH 2/3 ipv6: tcp: provide sk->sk_priority to ctl packets
In reply to Eric DumazetWe can populate skb->priority for some ctl packets instead of always using zero.
--- net/ipv6/tcp_ipv6.c | 16 +++++++++------- 1 file changed, 9 insertions(+), 7 deletions(-)
net/ipv6/tcp_ipv6.c +9 -7
@@ -804,7 +804,7 @@ static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {static void tcp_v6_send_response (const struct sock *sk,struct sk_buff *skb,u32 seq,u32 ack,u32 win,u32 tsval,u32 tsecr,int oif,struct tcp_md5sig_key *key,int rst, -u8 tclass,__be32 label) +u8 tclass,__be32 label,u32 priority) {const struct tcphdr *th =tcp_hdr (skb);struct tcphdr *t1;
@@ -909,7 +909,7 @@ static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32if (!IS_ERR (dst)) {skb_dst_set (buff, dst);ip6_xmit (ctl_sk, buff, &fl6, fl6.flowi6_mark,NULL , tclass, -0 ); + priority);TCP_INC_STATS (net,TCP_MIB_OUTSEGS );if (rst)TCP_INC_STATS (net,TCP_MIB_OUTRSTS );
@@ -932,6 +932,7 @@ static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)struct sock *sk1 =NULL ;#endif __be32 label =0 ; +u32 priority =0 ;struct net *net;int oif =0 ;
@@ -992,6 +993,7 @@ static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)trace_tcp_send_reset (sk, skb);if (np->repflow) label =ip6_flowlabel (ipv6h); + priority = sk->sk_priority; }if (sk->sk_state ==TCP_TIME_WAIT ) label =cpu_to_be32 (inet_twsk (sk)->tw_flowlabel);
@@ -1001,7 +1003,7 @@ static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb) }tcp_v6_send_response (sk, skb, seq, ack_seq,0 ,0 ,0 , oif, key,1 ,0 , - label); + label, priority);#ifdef CONFIG_TCP_MD5SIG out:
@@ -1012,10 +1014,10 @@ static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)static void tcp_v6_send_ack (const struct sock *sk,struct sk_buff *skb,u32 seq,u32 ack,u32 win,u32 tsval,u32 tsecr,int oif,struct tcp_md5sig_key *key,u8 tclass, -__be32 label) +__be32 label,u32 priority) {tcp_v6_send_response (sk, skb, seq, ack, win, tsval, tsecr, oif, key,0 , - tclass, label); + tclass, label, priority); }static void tcp_v6_timewait_ack (struct sock *sk,struct sk_buff *skb)
@@ -1027,7 +1029,7 @@ static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb) tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,tcp_time_stamp_raw () + tcptw->tw_ts_offset, tcptw->tw_ts_recent, tw->tw_bound_dev_if,tcp_twsk_md5_key (tcptw), - tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel)); + tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), 0);inet_twsk_put (tw); }
@@ -1050,7 +1052,7 @@ static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,tcp_time_stamp_raw () +tcp_rsk (req)->ts_off, req->ts_recent, sk->sk_bound_dev_if,tcp_v6_md5_do_lookup (sk, &ipv6_hdr (skb)->saddr), - 0, 0); + 0, 0, sk->sk_priority); }
Signature · 1 line
2.23.0.351.gc4317032e6-goog
PATCH 3/3 tcp: honor SO_PRIORITY in TIME_WAIT state
In reply to Eric Dumazetctl packets sent on behalf of TIME_WAIT sockets currently have a zero skb->priority, which can cause various problems.
In this patch we :
- add a tw_priority field in struct inet_timewait_sock.
- populate it from sk->sk_priority when a TIME_WAIT is created.
- For IPv4, change ip_send_unicast_reply() and its two callers to propagate tw_priority correctly. ip_send_unicast_reply() no longer changes sk->sk_priority.
- For IPv6, make sure TIME_WAIT sockets pass their tw_priority field to tcp_v6_send_response() and tcp_v6_send_ack().
--- include/net/inet_timewait_sock.h | 1 + net/ipv4/ip_output.c | 1 - net/ipv4/tcp_ipv4.c | 4 ++++ net/ipv4/tcp_minisocks.c | 1 + net/ipv6/tcp_ipv6.c | 6 ++++-- 5 files changed, 10 insertions(+), 3 deletions(-)
include/net/inet_timewait_sock.h +1 -0
@@ -71,6 +71,7 @@ struct inet_timewait_sock { tw_pad :2 ,/* 2 bits hole */ tw_tos : 8;u32 tw_txhash; +u32 tw_priority;struct timer_list tw_timer;struct inet_bind_bucket *tw_tb; };
net/ipv4/ip_output.c +0 -1
@@ -1694,7 +1694,6 @@ void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,inet_sk (sk)->tos = arg->tos; - sk->sk_priority = skb->priority; sk->sk_protocol =ip_hdr (skb)->protocol; sk->sk_bound_dev_if = arg->bound_dev_if; sk->sk_sndbuf = sysctl_wmem_default;
net/ipv4/tcp_ipv4.c +4 -0
@@ -771,6 +771,8 @@ static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)if (sk) { ctl_sk->sk_mark = (sk->sk_state ==TCP_TIME_WAIT ) ?inet_twsk (sk)->tw_mark : sk->sk_mark; + ctl_sk->sk_priority = (sk->sk_state ==TCP_TIME_WAIT ) ? +inet_twsk (sk)->tw_priority : sk->sk_priority; transmit_time =tcp_transmit_time (sk); } ip_send_unicast_reply(ctl_sk,
@@ -866,6 +868,8 @@ static void tcp_v4_send_ack(const struct sock *sk, ctl_sk =this_cpu_read (*net->ipv4.tcp_sk); ctl_sk->sk_mark = (sk->sk_state ==TCP_TIME_WAIT ) ?inet_twsk (sk)->tw_mark : sk->sk_mark; + ctl_sk->sk_priority = (sk->sk_state ==TCP_TIME_WAIT ) ? +inet_twsk (sk)->tw_priority : sk->sk_priority; transmit_time =tcp_transmit_time (sk); ip_send_unicast_reply(ctl_sk, skb, &TCP_SKB_CB(skb)->header.h4.opt,
net/ipv4/tcp_minisocks.c +1 -0
@@ -266,6 +266,7 @@ void tcp_time_wait(struct sock *sk, int state, int timeo) tw->tw_transparent = inet->transparent; tw->tw_mark = sk->sk_mark; + tw->tw_priority = sk->sk_priority; tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale; tcptw->tw_rcv_nxt = tp->rcv_nxt; tcptw->tw_snd_nxt = tp->snd_nxt;
net/ipv6/tcp_ipv6.c +4 -2
@@ -995,8 +995,10 @@ static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb) label =ip6_flowlabel (ipv6h); priority = sk->sk_priority; } -if (sk->sk_state ==TCP_TIME_WAIT ) +if (sk->sk_state ==TCP_TIME_WAIT ) { label =cpu_to_be32 (inet_twsk (sk)->tw_flowlabel); + priority =inet_twsk (sk)->tw_priority; + } }else {if (net->ipv6.sysctl.flowlabel_reflect &FLOWLABEL_REFLECT_TCP_RESET ) label =ip6_flowlabel (ipv6h);
@@ -1029,7 +1031,7 @@ static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb) tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,tcp_time_stamp_raw () + tcptw->tw_ts_offset, tcptw->tw_ts_recent, tw->tw_bound_dev_if,tcp_twsk_md5_key (tcptw), - tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), 0); + tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority);inet_twsk_put (tw); }
Signature · 1 line
2.23.0.351.gc4317032e6-goog
David Miller
maintainerFrom: Eric Dumazet <[email protected]>
Date: Tue, 24 Sep 2019 08:01:13 -0700
Quoted 10 lines
SO_PRIORITYsocket option requests TCP egress packets to contain a user provided value.TCP manages to send most packets with the requested values, notably for
TCP_ESTABLISHEDstate, but fails to do so for few packets.These packets are control packets sent on behalf of
SYN_RECVorTIME_WAITstates.
Quoted 7 lines
Note that to test this with packetdrill, it is a bit of a hassle, since packetdrill can not verify priority of egress packets, other than indirect observations, using for example
sch_prioon its tunnel device.The bad skb priorities cause problems for GCP,
as this field is one of the keys used in routing.
Series applied, thanks Eric.