diff options
Diffstat (limited to 'net/ipv4/tcp_input.c')
-rw-r--r-- | net/ipv4/tcp_input.c | 82 |
1 files changed, 61 insertions, 21 deletions
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 981d1203b152..0a461232329f 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c | |||
@@ -115,8 +115,8 @@ int sysctl_tcp_abc = 1; | |||
115 | /* Adapt the MSS value used to make delayed ack decision to the | 115 | /* Adapt the MSS value used to make delayed ack decision to the |
116 | * real world. | 116 | * real world. |
117 | */ | 117 | */ |
118 | static inline void tcp_measure_rcv_mss(struct sock *sk, | 118 | static void tcp_measure_rcv_mss(struct sock *sk, |
119 | const struct sk_buff *skb) | 119 | const struct sk_buff *skb) |
120 | { | 120 | { |
121 | struct inet_connection_sock *icsk = inet_csk(sk); | 121 | struct inet_connection_sock *icsk = inet_csk(sk); |
122 | const unsigned int lss = icsk->icsk_ack.last_seg_size; | 122 | const unsigned int lss = icsk->icsk_ack.last_seg_size; |
@@ -246,8 +246,8 @@ static int __tcp_grow_window(const struct sock *sk, struct tcp_sock *tp, | |||
246 | return 0; | 246 | return 0; |
247 | } | 247 | } |
248 | 248 | ||
249 | static inline void tcp_grow_window(struct sock *sk, struct tcp_sock *tp, | 249 | static void tcp_grow_window(struct sock *sk, struct tcp_sock *tp, |
250 | struct sk_buff *skb) | 250 | struct sk_buff *skb) |
251 | { | 251 | { |
252 | /* Check #1 */ | 252 | /* Check #1 */ |
253 | if (tp->rcv_ssthresh < tp->window_clamp && | 253 | if (tp->rcv_ssthresh < tp->window_clamp && |
@@ -341,6 +341,26 @@ static void tcp_clamp_window(struct sock *sk, struct tcp_sock *tp) | |||
341 | tp->rcv_ssthresh = min(tp->window_clamp, 2U*tp->advmss); | 341 | tp->rcv_ssthresh = min(tp->window_clamp, 2U*tp->advmss); |
342 | } | 342 | } |
343 | 343 | ||
344 | |||
345 | /* Initialize RCV_MSS value. | ||
346 | * RCV_MSS is an our guess about MSS used by the peer. | ||
347 | * We haven't any direct information about the MSS. | ||
348 | * It's better to underestimate the RCV_MSS rather than overestimate. | ||
349 | * Overestimations make us ACKing less frequently than needed. | ||
350 | * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss(). | ||
351 | */ | ||
352 | void tcp_initialize_rcv_mss(struct sock *sk) | ||
353 | { | ||
354 | struct tcp_sock *tp = tcp_sk(sk); | ||
355 | unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache); | ||
356 | |||
357 | hint = min(hint, tp->rcv_wnd/2); | ||
358 | hint = min(hint, TCP_MIN_RCVMSS); | ||
359 | hint = max(hint, TCP_MIN_MSS); | ||
360 | |||
361 | inet_csk(sk)->icsk_ack.rcv_mss = hint; | ||
362 | } | ||
363 | |||
344 | /* Receiver "autotuning" code. | 364 | /* Receiver "autotuning" code. |
345 | * | 365 | * |
346 | * The algorithm for RTT estimation w/o timestamps is based on | 366 | * The algorithm for RTT estimation w/o timestamps is based on |
@@ -735,6 +755,27 @@ __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst) | |||
735 | return min_t(__u32, cwnd, tp->snd_cwnd_clamp); | 755 | return min_t(__u32, cwnd, tp->snd_cwnd_clamp); |
736 | } | 756 | } |
737 | 757 | ||
758 | /* Set slow start threshold and cwnd not falling to slow start */ | ||
759 | void tcp_enter_cwr(struct sock *sk) | ||
760 | { | ||
761 | struct tcp_sock *tp = tcp_sk(sk); | ||
762 | |||
763 | tp->prior_ssthresh = 0; | ||
764 | tp->bytes_acked = 0; | ||
765 | if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) { | ||
766 | tp->undo_marker = 0; | ||
767 | tp->snd_ssthresh = inet_csk(sk)->icsk_ca_ops->ssthresh(sk); | ||
768 | tp->snd_cwnd = min(tp->snd_cwnd, | ||
769 | tcp_packets_in_flight(tp) + 1U); | ||
770 | tp->snd_cwnd_cnt = 0; | ||
771 | tp->high_seq = tp->snd_nxt; | ||
772 | tp->snd_cwnd_stamp = tcp_time_stamp; | ||
773 | TCP_ECN_queue_cwr(tp); | ||
774 | |||
775 | tcp_set_ca_state(sk, TCP_CA_CWR); | ||
776 | } | ||
777 | } | ||
778 | |||
738 | /* Initialize metrics on socket. */ | 779 | /* Initialize metrics on socket. */ |
739 | 780 | ||
740 | static void tcp_init_metrics(struct sock *sk) | 781 | static void tcp_init_metrics(struct sock *sk) |
@@ -2070,8 +2111,8 @@ static inline void tcp_ack_update_rtt(struct sock *sk, const int flag, | |||
2070 | tcp_ack_no_tstamp(sk, seq_rtt, flag); | 2111 | tcp_ack_no_tstamp(sk, seq_rtt, flag); |
2071 | } | 2112 | } |
2072 | 2113 | ||
2073 | static inline void tcp_cong_avoid(struct sock *sk, u32 ack, u32 rtt, | 2114 | static void tcp_cong_avoid(struct sock *sk, u32 ack, u32 rtt, |
2074 | u32 in_flight, int good) | 2115 | u32 in_flight, int good) |
2075 | { | 2116 | { |
2076 | const struct inet_connection_sock *icsk = inet_csk(sk); | 2117 | const struct inet_connection_sock *icsk = inet_csk(sk); |
2077 | icsk->icsk_ca_ops->cong_avoid(sk, ack, rtt, in_flight, good); | 2118 | icsk->icsk_ca_ops->cong_avoid(sk, ack, rtt, in_flight, good); |
@@ -2082,7 +2123,7 @@ static inline void tcp_cong_avoid(struct sock *sk, u32 ack, u32 rtt, | |||
2082 | * RFC2988 recommends to restart timer to now+rto. | 2123 | * RFC2988 recommends to restart timer to now+rto. |
2083 | */ | 2124 | */ |
2084 | 2125 | ||
2085 | static inline void tcp_ack_packets_out(struct sock *sk, struct tcp_sock *tp) | 2126 | static void tcp_ack_packets_out(struct sock *sk, struct tcp_sock *tp) |
2086 | { | 2127 | { |
2087 | if (!tp->packets_out) { | 2128 | if (!tp->packets_out) { |
2088 | inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS); | 2129 | inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS); |
@@ -2147,7 +2188,7 @@ static int tcp_tso_acked(struct sock *sk, struct sk_buff *skb, | |||
2147 | return acked; | 2188 | return acked; |
2148 | } | 2189 | } |
2149 | 2190 | ||
2150 | static inline u32 tcp_usrtt(const struct sk_buff *skb) | 2191 | static u32 tcp_usrtt(const struct sk_buff *skb) |
2151 | { | 2192 | { |
2152 | struct timeval tv, now; | 2193 | struct timeval tv, now; |
2153 | 2194 | ||
@@ -2583,8 +2624,8 @@ void tcp_parse_options(struct sk_buff *skb, struct tcp_options_received *opt_rx, | |||
2583 | /* Fast parse options. This hopes to only see timestamps. | 2624 | /* Fast parse options. This hopes to only see timestamps. |
2584 | * If it is wrong it falls back on tcp_parse_options(). | 2625 | * If it is wrong it falls back on tcp_parse_options(). |
2585 | */ | 2626 | */ |
2586 | static inline int tcp_fast_parse_options(struct sk_buff *skb, struct tcphdr *th, | 2627 | static int tcp_fast_parse_options(struct sk_buff *skb, struct tcphdr *th, |
2587 | struct tcp_sock *tp) | 2628 | struct tcp_sock *tp) |
2588 | { | 2629 | { |
2589 | if (th->doff == sizeof(struct tcphdr)>>2) { | 2630 | if (th->doff == sizeof(struct tcphdr)>>2) { |
2590 | tp->rx_opt.saw_tstamp = 0; | 2631 | tp->rx_opt.saw_tstamp = 0; |
@@ -2804,8 +2845,7 @@ static void tcp_fin(struct sk_buff *skb, struct sock *sk, struct tcphdr *th) | |||
2804 | } | 2845 | } |
2805 | } | 2846 | } |
2806 | 2847 | ||
2807 | static __inline__ int | 2848 | static inline int tcp_sack_extend(struct tcp_sack_block *sp, u32 seq, u32 end_seq) |
2808 | tcp_sack_extend(struct tcp_sack_block *sp, u32 seq, u32 end_seq) | ||
2809 | { | 2849 | { |
2810 | if (!after(seq, sp->end_seq) && !after(sp->start_seq, end_seq)) { | 2850 | if (!after(seq, sp->end_seq) && !after(sp->start_seq, end_seq)) { |
2811 | if (before(seq, sp->start_seq)) | 2851 | if (before(seq, sp->start_seq)) |
@@ -2817,7 +2857,7 @@ tcp_sack_extend(struct tcp_sack_block *sp, u32 seq, u32 end_seq) | |||
2817 | return 0; | 2857 | return 0; |
2818 | } | 2858 | } |
2819 | 2859 | ||
2820 | static inline void tcp_dsack_set(struct tcp_sock *tp, u32 seq, u32 end_seq) | 2860 | static void tcp_dsack_set(struct tcp_sock *tp, u32 seq, u32 end_seq) |
2821 | { | 2861 | { |
2822 | if (tp->rx_opt.sack_ok && sysctl_tcp_dsack) { | 2862 | if (tp->rx_opt.sack_ok && sysctl_tcp_dsack) { |
2823 | if (before(seq, tp->rcv_nxt)) | 2863 | if (before(seq, tp->rcv_nxt)) |
@@ -2832,7 +2872,7 @@ static inline void tcp_dsack_set(struct tcp_sock *tp, u32 seq, u32 end_seq) | |||
2832 | } | 2872 | } |
2833 | } | 2873 | } |
2834 | 2874 | ||
2835 | static inline void tcp_dsack_extend(struct tcp_sock *tp, u32 seq, u32 end_seq) | 2875 | static void tcp_dsack_extend(struct tcp_sock *tp, u32 seq, u32 end_seq) |
2836 | { | 2876 | { |
2837 | if (!tp->rx_opt.dsack) | 2877 | if (!tp->rx_opt.dsack) |
2838 | tcp_dsack_set(tp, seq, end_seq); | 2878 | tcp_dsack_set(tp, seq, end_seq); |
@@ -2890,7 +2930,7 @@ static void tcp_sack_maybe_coalesce(struct tcp_sock *tp) | |||
2890 | } | 2930 | } |
2891 | } | 2931 | } |
2892 | 2932 | ||
2893 | static __inline__ void tcp_sack_swap(struct tcp_sack_block *sack1, struct tcp_sack_block *sack2) | 2933 | static inline void tcp_sack_swap(struct tcp_sack_block *sack1, struct tcp_sack_block *sack2) |
2894 | { | 2934 | { |
2895 | __u32 tmp; | 2935 | __u32 tmp; |
2896 | 2936 | ||
@@ -3455,7 +3495,7 @@ void tcp_cwnd_application_limited(struct sock *sk) | |||
3455 | tp->snd_cwnd_stamp = tcp_time_stamp; | 3495 | tp->snd_cwnd_stamp = tcp_time_stamp; |
3456 | } | 3496 | } |
3457 | 3497 | ||
3458 | static inline int tcp_should_expand_sndbuf(struct sock *sk, struct tcp_sock *tp) | 3498 | static int tcp_should_expand_sndbuf(struct sock *sk, struct tcp_sock *tp) |
3459 | { | 3499 | { |
3460 | /* If the user specified a specific send buffer setting, do | 3500 | /* If the user specified a specific send buffer setting, do |
3461 | * not modify it. | 3501 | * not modify it. |
@@ -3502,7 +3542,7 @@ static void tcp_new_space(struct sock *sk) | |||
3502 | sk->sk_write_space(sk); | 3542 | sk->sk_write_space(sk); |
3503 | } | 3543 | } |
3504 | 3544 | ||
3505 | static inline void tcp_check_space(struct sock *sk) | 3545 | static void tcp_check_space(struct sock *sk) |
3506 | { | 3546 | { |
3507 | if (sock_flag(sk, SOCK_QUEUE_SHRUNK)) { | 3547 | if (sock_flag(sk, SOCK_QUEUE_SHRUNK)) { |
3508 | sock_reset_flag(sk, SOCK_QUEUE_SHRUNK); | 3548 | sock_reset_flag(sk, SOCK_QUEUE_SHRUNK); |
@@ -3512,7 +3552,7 @@ static inline void tcp_check_space(struct sock *sk) | |||
3512 | } | 3552 | } |
3513 | } | 3553 | } |
3514 | 3554 | ||
3515 | static __inline__ void tcp_data_snd_check(struct sock *sk, struct tcp_sock *tp) | 3555 | static inline void tcp_data_snd_check(struct sock *sk, struct tcp_sock *tp) |
3516 | { | 3556 | { |
3517 | tcp_push_pending_frames(sk, tp); | 3557 | tcp_push_pending_frames(sk, tp); |
3518 | tcp_check_space(sk); | 3558 | tcp_check_space(sk); |
@@ -3544,7 +3584,7 @@ static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible) | |||
3544 | } | 3584 | } |
3545 | } | 3585 | } |
3546 | 3586 | ||
3547 | static __inline__ void tcp_ack_snd_check(struct sock *sk) | 3587 | static inline void tcp_ack_snd_check(struct sock *sk) |
3548 | { | 3588 | { |
3549 | if (!inet_csk_ack_scheduled(sk)) { | 3589 | if (!inet_csk_ack_scheduled(sk)) { |
3550 | /* We sent a data segment already. */ | 3590 | /* We sent a data segment already. */ |
@@ -3692,8 +3732,7 @@ static int __tcp_checksum_complete_user(struct sock *sk, struct sk_buff *skb) | |||
3692 | return result; | 3732 | return result; |
3693 | } | 3733 | } |
3694 | 3734 | ||
3695 | static __inline__ int | 3735 | static inline int tcp_checksum_complete_user(struct sock *sk, struct sk_buff *skb) |
3696 | tcp_checksum_complete_user(struct sock *sk, struct sk_buff *skb) | ||
3697 | { | 3736 | { |
3698 | return skb->ip_summed != CHECKSUM_UNNECESSARY && | 3737 | return skb->ip_summed != CHECKSUM_UNNECESSARY && |
3699 | __tcp_checksum_complete_user(sk, skb); | 3738 | __tcp_checksum_complete_user(sk, skb); |
@@ -4474,3 +4513,4 @@ EXPORT_SYMBOL(sysctl_tcp_abc); | |||
4474 | EXPORT_SYMBOL(tcp_parse_options); | 4513 | EXPORT_SYMBOL(tcp_parse_options); |
4475 | EXPORT_SYMBOL(tcp_rcv_established); | 4514 | EXPORT_SYMBOL(tcp_rcv_established); |
4476 | EXPORT_SYMBOL(tcp_rcv_state_process); | 4515 | EXPORT_SYMBOL(tcp_rcv_state_process); |
4516 | EXPORT_SYMBOL(tcp_initialize_rcv_mss); | ||