diff options
Diffstat (limited to 'net/ipv4')
-rw-r--r-- | net/ipv4/tcp_input.c | 62 |
1 files changed, 30 insertions, 32 deletions
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 3e98b57578dc..e43065654930 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c | |||
@@ -548,10 +548,9 @@ static void tcp_event_data_recv(struct sock *sk, struct tcp_sock *tp, struct sk_ | |||
548 | * To save cycles in the RFC 1323 implementation it was better to break | 548 | * To save cycles in the RFC 1323 implementation it was better to break |
549 | * it up into three procedures. -- erics | 549 | * it up into three procedures. -- erics |
550 | */ | 550 | */ |
551 | static void tcp_rtt_estimator(struct sock *sk, const __u32 mrtt, u32 *usrtt) | 551 | static void tcp_rtt_estimator(struct sock *sk, const __u32 mrtt) |
552 | { | 552 | { |
553 | struct tcp_sock *tp = tcp_sk(sk); | 553 | struct tcp_sock *tp = tcp_sk(sk); |
554 | const struct inet_connection_sock *icsk = inet_csk(sk); | ||
555 | long m = mrtt; /* RTT */ | 554 | long m = mrtt; /* RTT */ |
556 | 555 | ||
557 | /* The following amusing code comes from Jacobson's | 556 | /* The following amusing code comes from Jacobson's |
@@ -610,9 +609,6 @@ static void tcp_rtt_estimator(struct sock *sk, const __u32 mrtt, u32 *usrtt) | |||
610 | tp->mdev_max = tp->rttvar = max(tp->mdev, TCP_RTO_MIN); | 609 | tp->mdev_max = tp->rttvar = max(tp->mdev, TCP_RTO_MIN); |
611 | tp->rtt_seq = tp->snd_nxt; | 610 | tp->rtt_seq = tp->snd_nxt; |
612 | } | 611 | } |
613 | |||
614 | if (icsk->icsk_ca_ops->rtt_sample) | ||
615 | icsk->icsk_ca_ops->rtt_sample(sk, *usrtt); | ||
616 | } | 612 | } |
617 | 613 | ||
618 | /* Calculate rto without backoff. This is the second half of Van Jacobson's | 614 | /* Calculate rto without backoff. This is the second half of Van Jacobson's |
@@ -1921,7 +1917,7 @@ tcp_fastretrans_alert(struct sock *sk, u32 prior_snd_una, | |||
1921 | /* Read draft-ietf-tcplw-high-performance before mucking | 1917 | /* Read draft-ietf-tcplw-high-performance before mucking |
1922 | * with this code. (Superceeds RFC1323) | 1918 | * with this code. (Superceeds RFC1323) |
1923 | */ | 1919 | */ |
1924 | static void tcp_ack_saw_tstamp(struct sock *sk, u32 *usrtt, int flag) | 1920 | static void tcp_ack_saw_tstamp(struct sock *sk, int flag) |
1925 | { | 1921 | { |
1926 | /* RTTM Rule: A TSecr value received in a segment is used to | 1922 | /* RTTM Rule: A TSecr value received in a segment is used to |
1927 | * update the averaged RTT measurement only if the segment | 1923 | * update the averaged RTT measurement only if the segment |
@@ -1940,13 +1936,13 @@ static void tcp_ack_saw_tstamp(struct sock *sk, u32 *usrtt, int flag) | |||
1940 | */ | 1936 | */ |
1941 | struct tcp_sock *tp = tcp_sk(sk); | 1937 | struct tcp_sock *tp = tcp_sk(sk); |
1942 | const __u32 seq_rtt = tcp_time_stamp - tp->rx_opt.rcv_tsecr; | 1938 | const __u32 seq_rtt = tcp_time_stamp - tp->rx_opt.rcv_tsecr; |
1943 | tcp_rtt_estimator(sk, seq_rtt, usrtt); | 1939 | tcp_rtt_estimator(sk, seq_rtt); |
1944 | tcp_set_rto(sk); | 1940 | tcp_set_rto(sk); |
1945 | inet_csk(sk)->icsk_backoff = 0; | 1941 | inet_csk(sk)->icsk_backoff = 0; |
1946 | tcp_bound_rto(sk); | 1942 | tcp_bound_rto(sk); |
1947 | } | 1943 | } |
1948 | 1944 | ||
1949 | static void tcp_ack_no_tstamp(struct sock *sk, u32 seq_rtt, u32 *usrtt, int flag) | 1945 | static void tcp_ack_no_tstamp(struct sock *sk, u32 seq_rtt, int flag) |
1950 | { | 1946 | { |
1951 | /* We don't have a timestamp. Can only use | 1947 | /* We don't have a timestamp. Can only use |
1952 | * packets that are not retransmitted to determine | 1948 | * packets that are not retransmitted to determine |
@@ -1960,21 +1956,21 @@ static void tcp_ack_no_tstamp(struct sock *sk, u32 seq_rtt, u32 *usrtt, int flag | |||
1960 | if (flag & FLAG_RETRANS_DATA_ACKED) | 1956 | if (flag & FLAG_RETRANS_DATA_ACKED) |
1961 | return; | 1957 | return; |
1962 | 1958 | ||
1963 | tcp_rtt_estimator(sk, seq_rtt, usrtt); | 1959 | tcp_rtt_estimator(sk, seq_rtt); |
1964 | tcp_set_rto(sk); | 1960 | tcp_set_rto(sk); |
1965 | inet_csk(sk)->icsk_backoff = 0; | 1961 | inet_csk(sk)->icsk_backoff = 0; |
1966 | tcp_bound_rto(sk); | 1962 | tcp_bound_rto(sk); |
1967 | } | 1963 | } |
1968 | 1964 | ||
1969 | static inline void tcp_ack_update_rtt(struct sock *sk, const int flag, | 1965 | static inline void tcp_ack_update_rtt(struct sock *sk, const int flag, |
1970 | const s32 seq_rtt, u32 *usrtt) | 1966 | const s32 seq_rtt) |
1971 | { | 1967 | { |
1972 | const struct tcp_sock *tp = tcp_sk(sk); | 1968 | const struct tcp_sock *tp = tcp_sk(sk); |
1973 | /* Note that peer MAY send zero echo. In this case it is ignored. (rfc1323) */ | 1969 | /* Note that peer MAY send zero echo. In this case it is ignored. (rfc1323) */ |
1974 | if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) | 1970 | if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) |
1975 | tcp_ack_saw_tstamp(sk, usrtt, flag); | 1971 | tcp_ack_saw_tstamp(sk, flag); |
1976 | else if (seq_rtt >= 0) | 1972 | else if (seq_rtt >= 0) |
1977 | tcp_ack_no_tstamp(sk, seq_rtt, usrtt, flag); | 1973 | tcp_ack_no_tstamp(sk, seq_rtt, flag); |
1978 | } | 1974 | } |
1979 | 1975 | ||
1980 | static inline void tcp_cong_avoid(struct sock *sk, u32 ack, u32 rtt, | 1976 | static inline void tcp_cong_avoid(struct sock *sk, u32 ack, u32 rtt, |
@@ -2054,20 +2050,27 @@ static int tcp_tso_acked(struct sock *sk, struct sk_buff *skb, | |||
2054 | return acked; | 2050 | return acked; |
2055 | } | 2051 | } |
2056 | 2052 | ||
2053 | static inline u32 tcp_usrtt(const struct sk_buff *skb) | ||
2054 | { | ||
2055 | struct timeval tv, now; | ||
2056 | |||
2057 | do_gettimeofday(&now); | ||
2058 | skb_get_timestamp(skb, &tv); | ||
2059 | return (now.tv_sec - tv.tv_sec) * 1000000 + (now.tv_usec - tv.tv_usec); | ||
2060 | } | ||
2057 | 2061 | ||
2058 | /* Remove acknowledged frames from the retransmission queue. */ | 2062 | /* Remove acknowledged frames from the retransmission queue. */ |
2059 | static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p, s32 *seq_usrtt) | 2063 | static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p) |
2060 | { | 2064 | { |
2061 | struct tcp_sock *tp = tcp_sk(sk); | 2065 | struct tcp_sock *tp = tcp_sk(sk); |
2066 | const struct inet_connection_sock *icsk = inet_csk(sk); | ||
2062 | struct sk_buff *skb; | 2067 | struct sk_buff *skb; |
2063 | __u32 now = tcp_time_stamp; | 2068 | __u32 now = tcp_time_stamp; |
2064 | int acked = 0; | 2069 | int acked = 0; |
2065 | __s32 seq_rtt = -1; | 2070 | __s32 seq_rtt = -1; |
2066 | struct timeval usnow; | ||
2067 | u32 pkts_acked = 0; | 2071 | u32 pkts_acked = 0; |
2068 | 2072 | void (*rtt_sample)(struct sock *sk, u32 usrtt) | |
2069 | if (seq_usrtt) | 2073 | = icsk->icsk_ca_ops->rtt_sample; |
2070 | do_gettimeofday(&usnow); | ||
2071 | 2074 | ||
2072 | while ((skb = skb_peek(&sk->sk_write_queue)) && | 2075 | while ((skb = skb_peek(&sk->sk_write_queue)) && |
2073 | skb != sk->sk_send_head) { | 2076 | skb != sk->sk_send_head) { |
@@ -2107,16 +2110,11 @@ static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p, s32 *seq_usrtt | |||
2107 | tp->retrans_out -= tcp_skb_pcount(skb); | 2110 | tp->retrans_out -= tcp_skb_pcount(skb); |
2108 | acked |= FLAG_RETRANS_DATA_ACKED; | 2111 | acked |= FLAG_RETRANS_DATA_ACKED; |
2109 | seq_rtt = -1; | 2112 | seq_rtt = -1; |
2110 | } else if (seq_rtt < 0) | 2113 | } else if (seq_rtt < 0) { |
2111 | seq_rtt = now - scb->when; | 2114 | seq_rtt = now - scb->when; |
2112 | if (seq_usrtt) { | 2115 | if (rtt_sample) |
2113 | struct timeval tv; | 2116 | (*rtt_sample)(sk, tcp_usrtt(skb)); |
2114 | |||
2115 | skb_get_timestamp(skb, &tv); | ||
2116 | *seq_usrtt = (usnow.tv_sec - tv.tv_sec) * 1000000 | ||
2117 | + (usnow.tv_usec - tv.tv_usec); | ||
2118 | } | 2117 | } |
2119 | |||
2120 | if (sacked & TCPCB_SACKED_ACKED) | 2118 | if (sacked & TCPCB_SACKED_ACKED) |
2121 | tp->sacked_out -= tcp_skb_pcount(skb); | 2119 | tp->sacked_out -= tcp_skb_pcount(skb); |
2122 | if (sacked & TCPCB_LOST) | 2120 | if (sacked & TCPCB_LOST) |
@@ -2126,8 +2124,11 @@ static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p, s32 *seq_usrtt | |||
2126 | !before(scb->end_seq, tp->snd_up)) | 2124 | !before(scb->end_seq, tp->snd_up)) |
2127 | tp->urg_mode = 0; | 2125 | tp->urg_mode = 0; |
2128 | } | 2126 | } |
2129 | } else if (seq_rtt < 0) | 2127 | } else if (seq_rtt < 0) { |
2130 | seq_rtt = now - scb->when; | 2128 | seq_rtt = now - scb->when; |
2129 | if (rtt_sample) | ||
2130 | (*rtt_sample)(sk, tcp_usrtt(skb)); | ||
2131 | } | ||
2131 | tcp_dec_pcount_approx(&tp->fackets_out, skb); | 2132 | tcp_dec_pcount_approx(&tp->fackets_out, skb); |
2132 | tcp_packets_out_dec(tp, skb); | 2133 | tcp_packets_out_dec(tp, skb); |
2133 | __skb_unlink(skb, &sk->sk_write_queue); | 2134 | __skb_unlink(skb, &sk->sk_write_queue); |
@@ -2135,8 +2136,7 @@ static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p, s32 *seq_usrtt | |||
2135 | } | 2136 | } |
2136 | 2137 | ||
2137 | if (acked&FLAG_ACKED) { | 2138 | if (acked&FLAG_ACKED) { |
2138 | const struct inet_connection_sock *icsk = inet_csk(sk); | 2139 | tcp_ack_update_rtt(sk, acked, seq_rtt); |
2139 | tcp_ack_update_rtt(sk, acked, seq_rtt, seq_usrtt); | ||
2140 | tcp_ack_packets_out(sk, tp); | 2140 | tcp_ack_packets_out(sk, tp); |
2141 | 2141 | ||
2142 | if (icsk->icsk_ca_ops->pkts_acked) | 2142 | if (icsk->icsk_ca_ops->pkts_acked) |
@@ -2299,7 +2299,6 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag) | |||
2299 | u32 ack = TCP_SKB_CB(skb)->ack_seq; | 2299 | u32 ack = TCP_SKB_CB(skb)->ack_seq; |
2300 | u32 prior_in_flight; | 2300 | u32 prior_in_flight; |
2301 | s32 seq_rtt; | 2301 | s32 seq_rtt; |
2302 | s32 seq_usrtt = 0; | ||
2303 | int prior_packets; | 2302 | int prior_packets; |
2304 | 2303 | ||
2305 | /* If the ack is newer than sent or older than previous acks | 2304 | /* If the ack is newer than sent or older than previous acks |
@@ -2352,8 +2351,7 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag) | |||
2352 | prior_in_flight = tcp_packets_in_flight(tp); | 2351 | prior_in_flight = tcp_packets_in_flight(tp); |
2353 | 2352 | ||
2354 | /* See if we can take anything off of the retransmit queue. */ | 2353 | /* See if we can take anything off of the retransmit queue. */ |
2355 | flag |= tcp_clean_rtx_queue(sk, &seq_rtt, | 2354 | flag |= tcp_clean_rtx_queue(sk, &seq_rtt); |
2356 | icsk->icsk_ca_ops->rtt_sample ? &seq_usrtt : NULL); | ||
2357 | 2355 | ||
2358 | if (tp->frto_counter) | 2356 | if (tp->frto_counter) |
2359 | tcp_process_frto(sk, prior_snd_una); | 2357 | tcp_process_frto(sk, prior_snd_una); |
@@ -4242,7 +4240,7 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, | |||
4242 | */ | 4240 | */ |
4243 | if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr && | 4241 | if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr && |
4244 | !tp->srtt) | 4242 | !tp->srtt) |
4245 | tcp_ack_saw_tstamp(sk, NULL, 0); | 4243 | tcp_ack_saw_tstamp(sk, 0); |
4246 | 4244 | ||
4247 | if (tp->rx_opt.tstamp_ok) | 4245 | if (tp->rx_opt.tstamp_ok) |
4248 | tp->advmss -= TCPOLEN_TSTAMP_ALIGNED; | 4246 | tp->advmss -= TCPOLEN_TSTAMP_ALIGNED; |