aboutsummaryrefslogtreecommitdiffstats
path: root/include/net/tcp.h
diff options
context:
space:
mode:
Diffstat (limited to 'include/net/tcp.h')
-rw-r--r--include/net/tcp.h246
1 files changed, 31 insertions, 215 deletions
diff --git a/include/net/tcp.h b/include/net/tcp.h
index d78025f9fbea..77f21c65bbca 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -225,53 +225,6 @@ extern atomic_t tcp_sockets_allocated;
225extern int tcp_memory_pressure; 225extern int tcp_memory_pressure;
226 226
227/* 227/*
228 * Pointers to address related TCP functions
229 * (i.e. things that depend on the address family)
230 */
231
232struct tcp_func {
233 int (*queue_xmit) (struct sk_buff *skb,
234 int ipfragok);
235
236 void (*send_check) (struct sock *sk,
237 struct tcphdr *th,
238 int len,
239 struct sk_buff *skb);
240
241 int (*rebuild_header) (struct sock *sk);
242
243 int (*conn_request) (struct sock *sk,
244 struct sk_buff *skb);
245
246 struct sock * (*syn_recv_sock) (struct sock *sk,
247 struct sk_buff *skb,
248 struct request_sock *req,
249 struct dst_entry *dst);
250
251 int (*remember_stamp) (struct sock *sk);
252
253 __u16 net_header_len;
254
255 int (*setsockopt) (struct sock *sk,
256 int level,
257 int optname,
258 char __user *optval,
259 int optlen);
260
261 int (*getsockopt) (struct sock *sk,
262 int level,
263 int optname,
264 char __user *optval,
265 int __user *optlen);
266
267
268 void (*addr2sockaddr) (struct sock *sk,
269 struct sockaddr *);
270
271 int sockaddr_len;
272};
273
274/*
275 * The next routines deal with comparing 32 bit unsigned ints 228 * The next routines deal with comparing 32 bit unsigned ints
276 * and worry about wraparound (automatic with unsigned arithmetic). 229 * and worry about wraparound (automatic with unsigned arithmetic).
277 */ 230 */
@@ -334,6 +287,9 @@ extern int tcp_rcv_established(struct sock *sk,
334 287
335extern void tcp_rcv_space_adjust(struct sock *sk); 288extern void tcp_rcv_space_adjust(struct sock *sk);
336 289
290extern int tcp_twsk_unique(struct sock *sk,
291 struct sock *sktw, void *twp);
292
337static inline void tcp_dec_quickack_mode(struct sock *sk, 293static inline void tcp_dec_quickack_mode(struct sock *sk,
338 const unsigned int pkts) 294 const unsigned int pkts)
339{ 295{
@@ -405,8 +361,7 @@ extern void tcp_parse_options(struct sk_buff *skb,
405 * TCP v4 functions exported for the inet6 API 361 * TCP v4 functions exported for the inet6 API
406 */ 362 */
407 363
408extern void tcp_v4_send_check(struct sock *sk, 364extern void tcp_v4_send_check(struct sock *sk, int len,
409 struct tcphdr *th, int len,
410 struct sk_buff *skb); 365 struct sk_buff *skb);
411 366
412extern int tcp_v4_conn_request(struct sock *sk, 367extern int tcp_v4_conn_request(struct sock *sk,
@@ -490,34 +445,16 @@ typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
490extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 445extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
491 sk_read_actor_t recv_actor); 446 sk_read_actor_t recv_actor);
492 447
493/* Initialize RCV_MSS value. 448extern void tcp_initialize_rcv_mss(struct sock *sk);
494 * RCV_MSS is an our guess about MSS used by the peer.
495 * We haven't any direct information about the MSS.
496 * It's better to underestimate the RCV_MSS rather than overestimate.
497 * Overestimations make us ACKing less frequently than needed.
498 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
499 */
500 449
501static inline void tcp_initialize_rcv_mss(struct sock *sk) 450static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
502{
503 struct tcp_sock *tp = tcp_sk(sk);
504 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
505
506 hint = min(hint, tp->rcv_wnd/2);
507 hint = min(hint, TCP_MIN_RCVMSS);
508 hint = max(hint, TCP_MIN_MSS);
509
510 inet_csk(sk)->icsk_ack.rcv_mss = hint;
511}
512
513static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
514{ 451{
515 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 452 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
516 ntohl(TCP_FLAG_ACK) | 453 ntohl(TCP_FLAG_ACK) |
517 snd_wnd); 454 snd_wnd);
518} 455}
519 456
520static __inline__ void tcp_fast_path_on(struct tcp_sock *tp) 457static inline void tcp_fast_path_on(struct tcp_sock *tp)
521{ 458{
522 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 459 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
523} 460}
@@ -535,7 +472,7 @@ static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
535 * Rcv_nxt can be after the window if our peer push more data 472 * Rcv_nxt can be after the window if our peer push more data
536 * than the offered window. 473 * than the offered window.
537 */ 474 */
538static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp) 475static inline u32 tcp_receive_window(const struct tcp_sock *tp)
539{ 476{
540 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 477 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
541 478
@@ -707,6 +644,7 @@ extern void tcp_cleanup_congestion_control(struct sock *sk);
707extern int tcp_set_default_congestion_control(const char *name); 644extern int tcp_set_default_congestion_control(const char *name);
708extern void tcp_get_default_congestion_control(char *name); 645extern void tcp_get_default_congestion_control(char *name);
709extern int tcp_set_congestion_control(struct sock *sk, const char *name); 646extern int tcp_set_congestion_control(struct sock *sk, const char *name);
647extern void tcp_slow_start(struct tcp_sock *tp);
710 648
711extern struct tcp_congestion_ops tcp_init_congestion_ops; 649extern struct tcp_congestion_ops tcp_init_congestion_ops;
712extern u32 tcp_reno_ssthresh(struct sock *sk); 650extern u32 tcp_reno_ssthresh(struct sock *sk);
@@ -746,7 +684,7 @@ static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
746 * "Packets left network, but not honestly ACKed yet" PLUS 684 * "Packets left network, but not honestly ACKed yet" PLUS
747 * "Packets fast retransmitted" 685 * "Packets fast retransmitted"
748 */ 686 */
749static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 687static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
750{ 688{
751 return (tp->packets_out - tp->left_out + tp->retrans_out); 689 return (tp->packets_out - tp->left_out + tp->retrans_out);
752} 690}
@@ -766,33 +704,6 @@ static inline __u32 tcp_current_ssthresh(const struct sock *sk)
766 (tp->snd_cwnd >> 2))); 704 (tp->snd_cwnd >> 2)));
767} 705}
768 706
769/*
770 * Linear increase during slow start
771 */
772static inline void tcp_slow_start(struct tcp_sock *tp)
773{
774 if (sysctl_tcp_abc) {
775 /* RFC3465: Slow Start
776 * TCP sender SHOULD increase cwnd by the number of
777 * previously unacknowledged bytes ACKed by each incoming
778 * acknowledgment, provided the increase is not more than L
779 */
780 if (tp->bytes_acked < tp->mss_cache)
781 return;
782
783 /* We MAY increase by 2 if discovered delayed ack */
784 if (sysctl_tcp_abc > 1 && tp->bytes_acked > 2*tp->mss_cache) {
785 if (tp->snd_cwnd < tp->snd_cwnd_clamp)
786 tp->snd_cwnd++;
787 }
788 }
789 tp->bytes_acked = 0;
790
791 if (tp->snd_cwnd < tp->snd_cwnd_clamp)
792 tp->snd_cwnd++;
793}
794
795
796static inline void tcp_sync_left_out(struct tcp_sock *tp) 707static inline void tcp_sync_left_out(struct tcp_sock *tp)
797{ 708{
798 if (tp->rx_opt.sack_ok && 709 if (tp->rx_opt.sack_ok &&
@@ -801,34 +712,7 @@ static inline void tcp_sync_left_out(struct tcp_sock *tp)
801 tp->left_out = tp->sacked_out + tp->lost_out; 712 tp->left_out = tp->sacked_out + tp->lost_out;
802} 713}
803 714
804/* Set slow start threshold and cwnd not falling to slow start */ 715extern void tcp_enter_cwr(struct sock *sk);
805static inline void __tcp_enter_cwr(struct sock *sk)
806{
807 const struct inet_connection_sock *icsk = inet_csk(sk);
808 struct tcp_sock *tp = tcp_sk(sk);
809
810 tp->undo_marker = 0;
811 tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk);
812 tp->snd_cwnd = min(tp->snd_cwnd,
813 tcp_packets_in_flight(tp) + 1U);
814 tp->snd_cwnd_cnt = 0;
815 tp->high_seq = tp->snd_nxt;
816 tp->snd_cwnd_stamp = tcp_time_stamp;
817 TCP_ECN_queue_cwr(tp);
818}
819
820static inline void tcp_enter_cwr(struct sock *sk)
821{
822 struct tcp_sock *tp = tcp_sk(sk);
823
824 tp->prior_ssthresh = 0;
825 tp->bytes_acked = 0;
826 if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) {
827 __tcp_enter_cwr(sk);
828 tcp_set_ca_state(sk, TCP_CA_CWR);
829 }
830}
831
832extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); 716extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
833 717
834/* Slow start with delack produces 3 packets of burst, so that 718/* Slow start with delack produces 3 packets of burst, so that
@@ -860,14 +744,14 @@ static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
860 return left <= tcp_max_burst(tp); 744 return left <= tcp_max_burst(tp);
861} 745}
862 746
863static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss, 747static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
864 const struct sk_buff *skb) 748 const struct sk_buff *skb)
865{ 749{
866 if (skb->len < mss) 750 if (skb->len < mss)
867 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 751 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
868} 752}
869 753
870static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp) 754static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
871{ 755{
872 const struct inet_connection_sock *icsk = inet_csk(sk); 756 const struct inet_connection_sock *icsk = inet_csk(sk);
873 if (!tp->packets_out && !icsk->icsk_pending) 757 if (!tp->packets_out && !icsk->icsk_pending)
@@ -875,18 +759,18 @@ static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *t
875 icsk->icsk_rto, TCP_RTO_MAX); 759 icsk->icsk_rto, TCP_RTO_MAX);
876} 760}
877 761
878static __inline__ void tcp_push_pending_frames(struct sock *sk, 762static inline void tcp_push_pending_frames(struct sock *sk,
879 struct tcp_sock *tp) 763 struct tcp_sock *tp)
880{ 764{
881 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle); 765 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
882} 766}
883 767
884static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq) 768static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
885{ 769{
886 tp->snd_wl1 = seq; 770 tp->snd_wl1 = seq;
887} 771}
888 772
889static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq) 773static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
890{ 774{
891 tp->snd_wl1 = seq; 775 tp->snd_wl1 = seq;
892} 776}
@@ -894,19 +778,19 @@ static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
894/* 778/*
895 * Calculate(/check) TCP checksum 779 * Calculate(/check) TCP checksum
896 */ 780 */
897static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len, 781static inline u16 tcp_v4_check(struct tcphdr *th, int len,
898 unsigned long saddr, unsigned long daddr, 782 unsigned long saddr, unsigned long daddr,
899 unsigned long base) 783 unsigned long base)
900{ 784{
901 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 785 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
902} 786}
903 787
904static __inline__ int __tcp_checksum_complete(struct sk_buff *skb) 788static inline int __tcp_checksum_complete(struct sk_buff *skb)
905{ 789{
906 return __skb_checksum_complete(skb); 790 return __skb_checksum_complete(skb);
907} 791}
908 792
909static __inline__ int tcp_checksum_complete(struct sk_buff *skb) 793static inline int tcp_checksum_complete(struct sk_buff *skb)
910{ 794{
911 return skb->ip_summed != CHECKSUM_UNNECESSARY && 795 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
912 __tcp_checksum_complete(skb); 796 __tcp_checksum_complete(skb);
@@ -914,7 +798,7 @@ static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
914 798
915/* Prequeue for VJ style copy to user, combined with checksumming. */ 799/* Prequeue for VJ style copy to user, combined with checksumming. */
916 800
917static __inline__ void tcp_prequeue_init(struct tcp_sock *tp) 801static inline void tcp_prequeue_init(struct tcp_sock *tp)
918{ 802{
919 tp->ucopy.task = NULL; 803 tp->ucopy.task = NULL;
920 tp->ucopy.len = 0; 804 tp->ucopy.len = 0;
@@ -930,7 +814,7 @@ static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
930 * 814 *
931 * NOTE: is this not too big to inline? 815 * NOTE: is this not too big to inline?
932 */ 816 */
933static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb) 817static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
934{ 818{
935 struct tcp_sock *tp = tcp_sk(sk); 819 struct tcp_sock *tp = tcp_sk(sk);
936 820
@@ -971,7 +855,7 @@ static const char *statename[]={
971}; 855};
972#endif 856#endif
973 857
974static __inline__ void tcp_set_state(struct sock *sk, int state) 858static inline void tcp_set_state(struct sock *sk, int state)
975{ 859{
976 int oldstate = sk->sk_state; 860 int oldstate = sk->sk_state;
977 861
@@ -1005,7 +889,7 @@ static __inline__ void tcp_set_state(struct sock *sk, int state)
1005#endif 889#endif
1006} 890}
1007 891
1008static __inline__ void tcp_done(struct sock *sk) 892static inline void tcp_done(struct sock *sk)
1009{ 893{
1010 tcp_set_state(sk, TCP_CLOSE); 894 tcp_set_state(sk, TCP_CLOSE);
1011 tcp_clear_xmit_timers(sk); 895 tcp_clear_xmit_timers(sk);
@@ -1018,81 +902,13 @@ static __inline__ void tcp_done(struct sock *sk)
1018 inet_csk_destroy_sock(sk); 902 inet_csk_destroy_sock(sk);
1019} 903}
1020 904
1021static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt) 905static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1022{ 906{
1023 rx_opt->dsack = 0; 907 rx_opt->dsack = 0;
1024 rx_opt->eff_sacks = 0; 908 rx_opt->eff_sacks = 0;
1025 rx_opt->num_sacks = 0; 909 rx_opt->num_sacks = 0;
1026} 910}
1027 911
1028static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
1029{
1030 if (tp->rx_opt.tstamp_ok) {
1031 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1032 (TCPOPT_NOP << 16) |
1033 (TCPOPT_TIMESTAMP << 8) |
1034 TCPOLEN_TIMESTAMP);
1035 *ptr++ = htonl(tstamp);
1036 *ptr++ = htonl(tp->rx_opt.ts_recent);
1037 }
1038 if (tp->rx_opt.eff_sacks) {
1039 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
1040 int this_sack;
1041
1042 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1043 (TCPOPT_NOP << 16) |
1044 (TCPOPT_SACK << 8) |
1045 (TCPOLEN_SACK_BASE +
1046 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1047 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
1048 *ptr++ = htonl(sp[this_sack].start_seq);
1049 *ptr++ = htonl(sp[this_sack].end_seq);
1050 }
1051 if (tp->rx_opt.dsack) {
1052 tp->rx_opt.dsack = 0;
1053 tp->rx_opt.eff_sacks--;
1054 }
1055 }
1056}
1057
1058/* Construct a tcp options header for a SYN or SYN_ACK packet.
1059 * If this is every changed make sure to change the definition of
1060 * MAX_SYN_SIZE to match the new maximum number of options that you
1061 * can generate.
1062 */
1063static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1064 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1065{
1066 /* We always get an MSS option.
1067 * The option bytes which will be seen in normal data
1068 * packets should timestamps be used, must be in the MSS
1069 * advertised. But we subtract them from tp->mss_cache so
1070 * that calculations in tcp_sendmsg are simpler etc.
1071 * So account for this fact here if necessary. If we
1072 * don't do this correctly, as a receiver we won't
1073 * recognize data packets as being full sized when we
1074 * should, and thus we won't abide by the delayed ACK
1075 * rules correctly.
1076 * SACKs don't matter, we never delay an ACK when we
1077 * have any of those going out.
1078 */
1079 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1080 if (ts) {
1081 if(sack)
1082 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1083 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1084 else
1085 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1086 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1087 *ptr++ = htonl(tstamp); /* TSVAL */
1088 *ptr++ = htonl(ts_recent); /* TSECR */
1089 } else if(sack)
1090 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1091 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1092 if (offer_wscale)
1093 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1094}
1095
1096/* Determine a window scaling and initial window to offer. */ 912/* Determine a window scaling and initial window to offer. */
1097extern void tcp_select_initial_window(int __space, __u32 mss, 913extern void tcp_select_initial_window(int __space, __u32 mss,
1098 __u32 *rcv_wnd, __u32 *window_clamp, 914 __u32 *rcv_wnd, __u32 *window_clamp,
@@ -1117,9 +933,9 @@ static inline int tcp_full_space(const struct sock *sk)
1117 return tcp_win_from_space(sk->sk_rcvbuf); 933 return tcp_win_from_space(sk->sk_rcvbuf);
1118} 934}
1119 935
1120static __inline__ void tcp_openreq_init(struct request_sock *req, 936static inline void tcp_openreq_init(struct request_sock *req,
1121 struct tcp_options_received *rx_opt, 937 struct tcp_options_received *rx_opt,
1122 struct sk_buff *skb) 938 struct sk_buff *skb)
1123{ 939{
1124 struct inet_request_sock *ireq = inet_rsk(req); 940 struct inet_request_sock *ireq = inet_rsk(req);
1125 941