Blamed commit moved the TIME_WAIT-derived ISN from the skb control block to a per-CPU variable, assuming the value would always be consumed by tcp_conn_request() for the same packet that wrote it. That assumption is violated by multiple drop paths between the producer (__this_cpu_write(tcp_tw_isn, isn) in tcp_v{4,6}_rcv()) and the consumer (tcp_conn_request()): - min_ttl / min_hopcount check - xfrm policy check - tcp_inbound_hash() MD5/AO mismatch - tcp_filter() eBPF/SO_ATTACH_FILTER drop - th->syn && th->fin discard in tcp_rcv_state_process() TCP_LISTEN - psp_sk_rx_policy_check() in tcp_v{4,6}_do_rcv() - tcp_checksum_complete() in tcp_v{4,6}_do_rcv() - tcp_v{4,6}_cookie_check() returning NULL When a packet is dropped on any of these paths, tcp_tw_isn is left set. The next SYN processed on the same CPU then consumes the non zero value in tcp_conn_request(), receiving a potentially predictable ISN. This patch moves back tcp_tw_isn to skb->cb[], getting rid of the per-cpu variable. Note that tcp_v{4,6}_fill_cb() do not set it. Very litle impact on overall code size/complexity: $ scripts/bloat-o-meter -t vmlinux.old vmlinux.new add/remove: 0/0 grow/shrink: 2/1 up/down: 8/-15 (-7) Function old new delta tcp_v6_rcv 3038 3042 +4 tcp_v4_rcv 3035 3039 +4 tcp_conn_request 2938 2923 -15 Total: Before=24436060, After=24436053, chg -0.00% Fixes: 41eecbd712b7 ("tcp: replace TCP_SKB_CB(skb)->tcp_tw_isn with a per-cpu field") Reported-by: Chris Mason Signed-off-by: Eric Dumazet --- v4: okay, move back tcp_tw_isn to skb->cb[] include/net/tcp.h | 7 ++++--- net/ipv4/tcp.c | 3 --- net/ipv4/tcp_input.c | 15 ++++++--------- net/ipv4/tcp_ipv4.c | 3 ++- net/ipv6/tcp_ipv6.c | 3 ++- 5 files changed, 14 insertions(+), 17 deletions(-) diff --git a/include/net/tcp.h b/include/net/tcp.h index ecbadcb3a7446cb18c245e670ba49ff574dfaff7..98848db62894aa4453efa9db7ea425d0aab263da 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -65,8 +65,6 @@ static inline void tcp_orphan_count_dec(void) this_cpu_dec(tcp_orphan_count); } -DECLARE_PER_CPU(u32, tcp_tw_isn); - void tcp_time_wait(struct sock *sk, int state, int timeo); #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) @@ -1102,10 +1100,13 @@ struct tcp_skb_cb { __u32 seq; /* Starting sequence number */ __u32 end_seq; /* SEQ + FIN + SYN + datalen */ union { - /* Note : + /* Notes : + * tcp_tw_isn is used in input path only + * (isn chosen by tcp_timewait_state_process()) * tcp_gso_segs/size are used in write queue only, * cf tcp_skb_pcount()/tcp_skb_mss() */ + u32 tcp_tw_isn; struct { u16 tcp_gso_segs; u16 tcp_gso_size; diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index 432fa28e47d4c8ef5d50339bfdf7da0ea8772b94..389a7cc17110daa5b3b490b3c339e53e212969f8 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -299,9 +299,6 @@ enum { DEFINE_PER_CPU(unsigned int, tcp_orphan_count); EXPORT_PER_CPU_SYMBOL_GPL(tcp_orphan_count); -DEFINE_PER_CPU(u32, tcp_tw_isn); -EXPORT_PER_CPU_SYMBOL_GPL(tcp_tw_isn); - long sysctl_tcp_mem[3] __read_mostly; DEFINE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc); diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index d5c9e65d97606d8eb57aba8ebc2373adf1bed62b..de9f68a9c0cf04109101b0d1bca20440376d4b05 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -7589,6 +7589,7 @@ int tcp_conn_request(struct request_sock_ops *rsk_ops, struct sock *sk, struct sk_buff *skb) { struct tcp_fastopen_cookie foc = { .len = -1 }; + u32 isn = TCP_SKB_CB(skb)->tcp_tw_isn; struct tcp_options_received tmp_opt; const struct tcp_sock *tp = tcp_sk(sk); struct net *net = sock_net(sk); @@ -7599,20 +7600,16 @@ int tcp_conn_request(struct request_sock_ops *rsk_ops, struct dst_entry *dst; struct flowi fl; u8 syncookies; - u32 isn; #ifdef CONFIG_TCP_AO const struct tcp_ao_hdr *aoh; #endif - isn = __this_cpu_read(tcp_tw_isn); - if (isn) { - /* TW buckets are converted to open requests without - * limitations, they conserve resources and peer is - * evidently real one. - */ - __this_cpu_write(tcp_tw_isn, 0); - } else { + /* If isn is non-zero, this SYN originally matched a TIME_WAIT socket. + * TW sockets are converted to open requests without limitations, + * we skip the queue limits and syncookie checks in the block below. + */ + if (!isn) { syncookies = READ_ONCE(net->ipv4.sysctl_tcp_syncookies); if (syncookies == 2 || inet_csk_reqsk_queue_is_full(sk)) { diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c index c0526cc0398049fb34b5de20a1175d54942e80cd..fdc81150ff6cf938b1971c33b2b997e5d0d8fcaa 100644 --- a/net/ipv4/tcp_ipv4.c +++ b/net/ipv4/tcp_ipv4.c @@ -2198,6 +2198,7 @@ int tcp_v4_rcv(struct sk_buff *skb) } } + isn = 0; process: if (static_branch_unlikely(&ip4_min_ttl)) { /* min_ttl can be changed concurrently from do_ip_setsockopt() */ @@ -2227,6 +2228,7 @@ int tcp_v4_rcv(struct sk_buff *skb) th = (const struct tcphdr *)skb->data; iph = ip_hdr(skb); tcp_v4_fill_cb(skb, iph, th); + TCP_SKB_CB(skb)->tcp_tw_isn = isn; skb->dev = NULL; @@ -2313,7 +2315,6 @@ int tcp_v4_rcv(struct sk_buff *skb) sk = sk2; tcp_v4_restore_cb(skb); refcounted = false; - __this_cpu_write(tcp_tw_isn, isn); goto process; } diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c index d13d49bfef19457cc5902cb556605a80f4c0ab2c..36d75fb50a70b728fedb7c316e023758bd61d62c 100644 --- a/net/ipv6/tcp_ipv6.c +++ b/net/ipv6/tcp_ipv6.c @@ -1839,6 +1839,7 @@ INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb) } } + isn = 0; process: if (static_branch_unlikely(&ip6_min_hopcount)) { /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */ @@ -1868,6 +1869,7 @@ INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb) th = (const struct tcphdr *)skb->data; hdr = ipv6_hdr(skb); tcp_v6_fill_cb(skb, hdr, th); + TCP_SKB_CB(skb)->tcp_tw_isn = isn; skb->dev = NULL; @@ -1956,7 +1958,6 @@ INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb) sk = sk2; tcp_v6_restore_cb(skb); refcounted = false; - __this_cpu_write(tcp_tw_isn, isn); goto process; } -- 2.54.0.563.g4f69b47b94-goog