aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--include/net/tcp.h25
-rw-r--r--include/net/tcp_ecn.h11
-rw-r--r--net/ipv4/tcp.c39
-rw-r--r--net/ipv4/tcp_input.c145
-rw-r--r--net/ipv4/tcp_output.c54
5 files changed, 158 insertions, 116 deletions
diff --git a/include/net/tcp.h b/include/net/tcp.h
index e79803353c83..43910fe3c448 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -420,9 +420,9 @@ extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
420 420
421/* tcp_output.c */ 421/* tcp_output.c */
422 422
423extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp, 423extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
424 unsigned int cur_mss, int nonagle); 424 int nonagle);
425extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp); 425extern int tcp_may_send_now(struct sock *sk);
426extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 426extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
427extern void tcp_xmit_retransmit_queue(struct sock *); 427extern void tcp_xmit_retransmit_queue(struct sock *);
428extern void tcp_simple_retransmit(struct sock *); 428extern void tcp_simple_retransmit(struct sock *);
@@ -479,8 +479,10 @@ static inline void tcp_fast_path_on(struct tcp_sock *tp)
479 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 479 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
480} 480}
481 481
482static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp) 482static inline void tcp_fast_path_check(struct sock *sk)
483{ 483{
484 struct tcp_sock *tp = tcp_sk(sk);
485
484 if (skb_queue_empty(&tp->out_of_order_queue) && 486 if (skb_queue_empty(&tp->out_of_order_queue) &&
485 tp->rcv_wnd && 487 tp->rcv_wnd &&
486 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 488 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
@@ -591,10 +593,10 @@ static inline void tcp_dec_pcount_approx(__u32 *count,
591 } 593 }
592} 594}
593 595
594static inline void tcp_packets_out_inc(struct sock *sk, 596static inline void tcp_packets_out_inc(struct sock *sk,
595 struct tcp_sock *tp,
596 const struct sk_buff *skb) 597 const struct sk_buff *skb)
597{ 598{
599 struct tcp_sock *tp = tcp_sk(sk);
598 int orig = tp->packets_out; 600 int orig = tp->packets_out;
599 601
600 tp->packets_out += tcp_skb_pcount(skb); 602 tp->packets_out += tcp_skb_pcount(skb);
@@ -778,18 +780,21 @@ static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
778 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 780 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
779} 781}
780 782
781static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp) 783static inline void tcp_check_probe_timer(struct sock *sk)
782{ 784{
785 struct tcp_sock *tp = tcp_sk(sk);
783 const struct inet_connection_sock *icsk = inet_csk(sk); 786 const struct inet_connection_sock *icsk = inet_csk(sk);
787
784 if (!tp->packets_out && !icsk->icsk_pending) 788 if (!tp->packets_out && !icsk->icsk_pending)
785 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 789 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
786 icsk->icsk_rto, TCP_RTO_MAX); 790 icsk->icsk_rto, TCP_RTO_MAX);
787} 791}
788 792
789static inline void tcp_push_pending_frames(struct sock *sk, 793static inline void tcp_push_pending_frames(struct sock *sk)
790 struct tcp_sock *tp)
791{ 794{
792 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle); 795 struct tcp_sock *tp = tcp_sk(sk);
796
797 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
793} 798}
794 799
795static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq) 800static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
diff --git a/include/net/tcp_ecn.h b/include/net/tcp_ecn.h
index b5f7c6ac0880..89eb3e05116d 100644
--- a/include/net/tcp_ecn.h
+++ b/include/net/tcp_ecn.h
@@ -27,9 +27,10 @@ static inline void TCP_ECN_send_synack(struct tcp_sock *tp,
27 TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_ECE; 27 TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_ECE;
28} 28}
29 29
30static inline void TCP_ECN_send_syn(struct sock *sk, struct tcp_sock *tp, 30static inline void TCP_ECN_send_syn(struct sock *sk, struct sk_buff *skb)
31 struct sk_buff *skb)
32{ 31{
32 struct tcp_sock *tp = tcp_sk(sk);
33
33 tp->ecn_flags = 0; 34 tp->ecn_flags = 0;
34 if (sysctl_tcp_ecn) { 35 if (sysctl_tcp_ecn) {
35 TCP_SKB_CB(skb)->flags |= TCPCB_FLAG_ECE|TCPCB_FLAG_CWR; 36 TCP_SKB_CB(skb)->flags |= TCPCB_FLAG_ECE|TCPCB_FLAG_CWR;
@@ -44,9 +45,11 @@ TCP_ECN_make_synack(struct request_sock *req, struct tcphdr *th)
44 th->ece = 1; 45 th->ece = 1;
45} 46}
46 47
47static inline void TCP_ECN_send(struct sock *sk, struct tcp_sock *tp, 48static inline void TCP_ECN_send(struct sock *sk, struct sk_buff *skb,
48 struct sk_buff *skb, int tcp_header_len) 49 int tcp_header_len)
49{ 50{
51 struct tcp_sock *tp = tcp_sk(sk);
52
50 if (tp->ecn_flags & TCP_ECN_OK) { 53 if (tp->ecn_flags & TCP_ECN_OK) {
51 /* Not-retransmitted data segment: set ECT and inject CWR. */ 54 /* Not-retransmitted data segment: set ECT and inject CWR. */
52 if (skb->len != tcp_header_len && 55 if (skb->len != tcp_header_len &&
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 99ad52c00c96..2cf9a898ce50 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -460,9 +460,9 @@ static inline int forced_push(struct tcp_sock *tp)
460 return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1)); 460 return after(tp->write_seq, tp->pushed_seq + (tp->max_window >> 1));
461} 461}
462 462
463static inline void skb_entail(struct sock *sk, struct tcp_sock *tp, 463static inline void skb_entail(struct sock *sk, struct sk_buff *skb)
464 struct sk_buff *skb)
465{ 464{
465 struct tcp_sock *tp = tcp_sk(sk);
466 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); 466 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
467 467
468 skb->csum = 0; 468 skb->csum = 0;
@@ -486,15 +486,17 @@ static inline void tcp_mark_urg(struct tcp_sock *tp, int flags,
486 } 486 }
487} 487}
488 488
489static inline void tcp_push(struct sock *sk, struct tcp_sock *tp, int flags, 489static inline void tcp_push(struct sock *sk, int flags, int mss_now,
490 int mss_now, int nonagle) 490 int nonagle)
491{ 491{
492 struct tcp_sock *tp = tcp_sk(sk);
493
492 if (tcp_send_head(sk)) { 494 if (tcp_send_head(sk)) {
493 struct sk_buff *skb = tcp_write_queue_tail(sk); 495 struct sk_buff *skb = tcp_write_queue_tail(sk);
494 if (!(flags & MSG_MORE) || forced_push(tp)) 496 if (!(flags & MSG_MORE) || forced_push(tp))
495 tcp_mark_push(tp, skb); 497 tcp_mark_push(tp, skb);
496 tcp_mark_urg(tp, flags, skb); 498 tcp_mark_urg(tp, flags, skb);
497 __tcp_push_pending_frames(sk, tp, mss_now, 499 __tcp_push_pending_frames(sk, mss_now,
498 (flags & MSG_MORE) ? TCP_NAGLE_CORK : nonagle); 500 (flags & MSG_MORE) ? TCP_NAGLE_CORK : nonagle);
499 } 501 }
500} 502}
@@ -540,7 +542,7 @@ new_segment:
540 if (!skb) 542 if (!skb)
541 goto wait_for_memory; 543 goto wait_for_memory;
542 544
543 skb_entail(sk, tp, skb); 545 skb_entail(sk, skb);
544 copy = size_goal; 546 copy = size_goal;
545 } 547 }
546 548
@@ -586,7 +588,7 @@ new_segment:
586 588
587 if (forced_push(tp)) { 589 if (forced_push(tp)) {
588 tcp_mark_push(tp, skb); 590 tcp_mark_push(tp, skb);
589 __tcp_push_pending_frames(sk, tp, mss_now, TCP_NAGLE_PUSH); 591 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
590 } else if (skb == tcp_send_head(sk)) 592 } else if (skb == tcp_send_head(sk))
591 tcp_push_one(sk, mss_now); 593 tcp_push_one(sk, mss_now);
592 continue; 594 continue;
@@ -595,7 +597,7 @@ wait_for_sndbuf:
595 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 597 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
596wait_for_memory: 598wait_for_memory:
597 if (copied) 599 if (copied)
598 tcp_push(sk, tp, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH); 600 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
599 601
600 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0) 602 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
601 goto do_error; 603 goto do_error;
@@ -606,7 +608,7 @@ wait_for_memory:
606 608
607out: 609out:
608 if (copied) 610 if (copied)
609 tcp_push(sk, tp, flags, mss_now, tp->nonagle); 611 tcp_push(sk, flags, mss_now, tp->nonagle);
610 return copied; 612 return copied;
611 613
612do_error: 614do_error:
@@ -637,8 +639,9 @@ ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset,
637#define TCP_PAGE(sk) (sk->sk_sndmsg_page) 639#define TCP_PAGE(sk) (sk->sk_sndmsg_page)
638#define TCP_OFF(sk) (sk->sk_sndmsg_off) 640#define TCP_OFF(sk) (sk->sk_sndmsg_off)
639 641
640static inline int select_size(struct sock *sk, struct tcp_sock *tp) 642static inline int select_size(struct sock *sk)
641{ 643{
644 struct tcp_sock *tp = tcp_sk(sk);
642 int tmp = tp->mss_cache; 645 int tmp = tp->mss_cache;
643 646
644 if (sk->sk_route_caps & NETIF_F_SG) { 647 if (sk->sk_route_caps & NETIF_F_SG) {
@@ -714,7 +717,7 @@ new_segment:
714 if (!sk_stream_memory_free(sk)) 717 if (!sk_stream_memory_free(sk))
715 goto wait_for_sndbuf; 718 goto wait_for_sndbuf;
716 719
717 skb = sk_stream_alloc_pskb(sk, select_size(sk, tp), 720 skb = sk_stream_alloc_pskb(sk, select_size(sk),
718 0, sk->sk_allocation); 721 0, sk->sk_allocation);
719 if (!skb) 722 if (!skb)
720 goto wait_for_memory; 723 goto wait_for_memory;
@@ -725,7 +728,7 @@ new_segment:
725 if (sk->sk_route_caps & NETIF_F_ALL_CSUM) 728 if (sk->sk_route_caps & NETIF_F_ALL_CSUM)
726 skb->ip_summed = CHECKSUM_PARTIAL; 729 skb->ip_summed = CHECKSUM_PARTIAL;
727 730
728 skb_entail(sk, tp, skb); 731 skb_entail(sk, skb);
729 copy = size_goal; 732 copy = size_goal;
730 } 733 }
731 734
@@ -830,7 +833,7 @@ new_segment:
830 833
831 if (forced_push(tp)) { 834 if (forced_push(tp)) {
832 tcp_mark_push(tp, skb); 835 tcp_mark_push(tp, skb);
833 __tcp_push_pending_frames(sk, tp, mss_now, TCP_NAGLE_PUSH); 836 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_PUSH);
834 } else if (skb == tcp_send_head(sk)) 837 } else if (skb == tcp_send_head(sk))
835 tcp_push_one(sk, mss_now); 838 tcp_push_one(sk, mss_now);
836 continue; 839 continue;
@@ -839,7 +842,7 @@ wait_for_sndbuf:
839 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 842 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
840wait_for_memory: 843wait_for_memory:
841 if (copied) 844 if (copied)
842 tcp_push(sk, tp, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH); 845 tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
843 846
844 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0) 847 if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
845 goto do_error; 848 goto do_error;
@@ -851,7 +854,7 @@ wait_for_memory:
851 854
852out: 855out:
853 if (copied) 856 if (copied)
854 tcp_push(sk, tp, flags, mss_now, tp->nonagle); 857 tcp_push(sk, flags, mss_now, tp->nonagle);
855 TCP_CHECK_TIMER(sk); 858 TCP_CHECK_TIMER(sk);
856 release_sock(sk); 859 release_sock(sk);
857 return copied; 860 return copied;
@@ -1389,7 +1392,7 @@ do_prequeue:
1389skip_copy: 1392skip_copy:
1390 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) { 1393 if (tp->urg_data && after(tp->copied_seq, tp->urg_seq)) {
1391 tp->urg_data = 0; 1394 tp->urg_data = 0;
1392 tcp_fast_path_check(sk, tp); 1395 tcp_fast_path_check(sk);
1393 } 1396 }
1394 if (used + offset < skb->len) 1397 if (used + offset < skb->len)
1395 continue; 1398 continue;
@@ -1830,7 +1833,7 @@ static int do_tcp_setsockopt(struct sock *sk, int level,
1830 * for currently queued segments. 1833 * for currently queued segments.
1831 */ 1834 */
1832 tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH; 1835 tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
1833 tcp_push_pending_frames(sk, tp); 1836 tcp_push_pending_frames(sk);
1834 } else { 1837 } else {
1835 tp->nonagle &= ~TCP_NAGLE_OFF; 1838 tp->nonagle &= ~TCP_NAGLE_OFF;
1836 } 1839 }
@@ -1854,7 +1857,7 @@ static int do_tcp_setsockopt(struct sock *sk, int level,
1854 tp->nonagle &= ~TCP_NAGLE_CORK; 1857 tp->nonagle &= ~TCP_NAGLE_CORK;
1855 if (tp->nonagle&TCP_NAGLE_OFF) 1858 if (tp->nonagle&TCP_NAGLE_OFF)
1856 tp->nonagle |= TCP_NAGLE_PUSH; 1859 tp->nonagle |= TCP_NAGLE_PUSH;
1857 tcp_push_pending_frames(sk, tp); 1860 tcp_push_pending_frames(sk);
1858 } 1861 }
1859 break; 1862 break;
1860 1863
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 2fbfc2e4209c..633389390788 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -235,9 +235,9 @@ static void tcp_fixup_sndbuf(struct sock *sk)
235 */ 235 */
236 236
237/* Slow part of check#2. */ 237/* Slow part of check#2. */
238static int __tcp_grow_window(const struct sock *sk, struct tcp_sock *tp, 238static int __tcp_grow_window(const struct sock *sk, const struct sk_buff *skb)
239 const struct sk_buff *skb)
240{ 239{
240 struct tcp_sock *tp = tcp_sk(sk);
241 /* Optimize this! */ 241 /* Optimize this! */
242 int truesize = tcp_win_from_space(skb->truesize)/2; 242 int truesize = tcp_win_from_space(skb->truesize)/2;
243 int window = tcp_win_from_space(sysctl_tcp_rmem[2])/2; 243 int window = tcp_win_from_space(sysctl_tcp_rmem[2])/2;
@@ -252,9 +252,11 @@ static int __tcp_grow_window(const struct sock *sk, struct tcp_sock *tp,
252 return 0; 252 return 0;
253} 253}
254 254
255static void tcp_grow_window(struct sock *sk, struct tcp_sock *tp, 255static void tcp_grow_window(struct sock *sk,
256 struct sk_buff *skb) 256 struct sk_buff *skb)
257{ 257{
258 struct tcp_sock *tp = tcp_sk(sk);
259
258 /* Check #1 */ 260 /* Check #1 */
259 if (tp->rcv_ssthresh < tp->window_clamp && 261 if (tp->rcv_ssthresh < tp->window_clamp &&
260 (int)tp->rcv_ssthresh < tcp_space(sk) && 262 (int)tp->rcv_ssthresh < tcp_space(sk) &&
@@ -267,7 +269,7 @@ static void tcp_grow_window(struct sock *sk, struct tcp_sock *tp,
267 if (tcp_win_from_space(skb->truesize) <= skb->len) 269 if (tcp_win_from_space(skb->truesize) <= skb->len)
268 incr = 2*tp->advmss; 270 incr = 2*tp->advmss;
269 else 271 else
270 incr = __tcp_grow_window(sk, tp, skb); 272 incr = __tcp_grow_window(sk, skb);
271 273
272 if (incr) { 274 if (incr) {
273 tp->rcv_ssthresh = min(tp->rcv_ssthresh + incr, tp->window_clamp); 275 tp->rcv_ssthresh = min(tp->rcv_ssthresh + incr, tp->window_clamp);
@@ -330,8 +332,9 @@ static void tcp_init_buffer_space(struct sock *sk)
330} 332}
331 333
332/* 5. Recalculate window clamp after socket hit its memory bounds. */ 334/* 5. Recalculate window clamp after socket hit its memory bounds. */
333static void tcp_clamp_window(struct sock *sk, struct tcp_sock *tp) 335static void tcp_clamp_window(struct sock *sk)
334{ 336{
337 struct tcp_sock *tp = tcp_sk(sk);
335 struct inet_connection_sock *icsk = inet_csk(sk); 338 struct inet_connection_sock *icsk = inet_csk(sk);
336 339
337 icsk->icsk_ack.quick = 0; 340 icsk->icsk_ack.quick = 0;
@@ -503,8 +506,9 @@ new_measure:
503 * each ACK we send, he increments snd_cwnd and transmits more of his 506 * each ACK we send, he increments snd_cwnd and transmits more of his
504 * queue. -DaveM 507 * queue. -DaveM
505 */ 508 */
506static void tcp_event_data_recv(struct sock *sk, struct tcp_sock *tp, struct sk_buff *skb) 509static void tcp_event_data_recv(struct sock *sk, struct sk_buff *skb)
507{ 510{
511 struct tcp_sock *tp = tcp_sk(sk);
508 struct inet_connection_sock *icsk = inet_csk(sk); 512 struct inet_connection_sock *icsk = inet_csk(sk);
509 u32 now; 513 u32 now;
510 514
@@ -545,7 +549,7 @@ static void tcp_event_data_recv(struct sock *sk, struct tcp_sock *tp, struct sk_
545 TCP_ECN_check_ce(tp, skb); 549 TCP_ECN_check_ce(tp, skb);
546 550
547 if (skb->len >= 128) 551 if (skb->len >= 128)
548 tcp_grow_window(sk, tp, skb); 552 tcp_grow_window(sk, skb);
549} 553}
550 554
551/* Called to compute a smoothed rtt estimate. The data fed to this 555/* Called to compute a smoothed rtt estimate. The data fed to this
@@ -1541,8 +1545,10 @@ static inline int tcp_skb_timedout(struct sock *sk, struct sk_buff *skb)
1541 return (tcp_time_stamp - TCP_SKB_CB(skb)->when > inet_csk(sk)->icsk_rto); 1545 return (tcp_time_stamp - TCP_SKB_CB(skb)->when > inet_csk(sk)->icsk_rto);
1542} 1546}
1543 1547
1544static inline int tcp_head_timedout(struct sock *sk, struct tcp_sock *tp) 1548static inline int tcp_head_timedout(struct sock *sk)
1545{ 1549{
1550 struct tcp_sock *tp = tcp_sk(sk);
1551
1546 return tp->packets_out && 1552 return tp->packets_out &&
1547 tcp_skb_timedout(sk, tcp_write_queue_head(sk)); 1553 tcp_skb_timedout(sk, tcp_write_queue_head(sk));
1548} 1554}
@@ -1640,8 +1646,9 @@ static inline int tcp_head_timedout(struct sock *sk, struct tcp_sock *tp)
1640 * Main question: may we further continue forward transmission 1646 * Main question: may we further continue forward transmission
1641 * with the same cwnd? 1647 * with the same cwnd?
1642 */ 1648 */
1643static int tcp_time_to_recover(struct sock *sk, struct tcp_sock *tp) 1649static int tcp_time_to_recover(struct sock *sk)
1644{ 1650{
1651 struct tcp_sock *tp = tcp_sk(sk);
1645 __u32 packets_out; 1652 __u32 packets_out;
1646 1653
1647 /* Do not perform any recovery during FRTO algorithm */ 1654 /* Do not perform any recovery during FRTO algorithm */
@@ -1659,7 +1666,7 @@ static int tcp_time_to_recover(struct sock *sk, struct tcp_sock *tp)
1659 /* Trick#3 : when we use RFC2988 timer restart, fast 1666 /* Trick#3 : when we use RFC2988 timer restart, fast
1660 * retransmit can be triggered by timeout of queue head. 1667 * retransmit can be triggered by timeout of queue head.
1661 */ 1668 */
1662 if (tcp_head_timedout(sk, tp)) 1669 if (tcp_head_timedout(sk))
1663 return 1; 1670 return 1;
1664 1671
1665 /* Trick#4: It is still not OK... But will it be useful to delay 1672 /* Trick#4: It is still not OK... But will it be useful to delay
@@ -1668,7 +1675,7 @@ static int tcp_time_to_recover(struct sock *sk, struct tcp_sock *tp)
1668 packets_out = tp->packets_out; 1675 packets_out = tp->packets_out;
1669 if (packets_out <= tp->reordering && 1676 if (packets_out <= tp->reordering &&
1670 tp->sacked_out >= max_t(__u32, packets_out/2, sysctl_tcp_reordering) && 1677 tp->sacked_out >= max_t(__u32, packets_out/2, sysctl_tcp_reordering) &&
1671 !tcp_may_send_now(sk, tp)) { 1678 !tcp_may_send_now(sk)) {
1672 /* We have nothing to send. This connection is limited 1679 /* We have nothing to send. This connection is limited
1673 * either by receiver window or by application. 1680 * either by receiver window or by application.
1674 */ 1681 */
@@ -1708,8 +1715,10 @@ static void tcp_add_reno_sack(struct sock *sk)
1708 1715
1709/* Account for ACK, ACKing some data in Reno Recovery phase. */ 1716/* Account for ACK, ACKing some data in Reno Recovery phase. */
1710 1717
1711static void tcp_remove_reno_sacks(struct sock *sk, struct tcp_sock *tp, int acked) 1718static void tcp_remove_reno_sacks(struct sock *sk, int acked)
1712{ 1719{
1720 struct tcp_sock *tp = tcp_sk(sk);
1721
1713 if (acked > 0) { 1722 if (acked > 0) {
1714 /* One ACK acked hole. The rest eat duplicate ACKs. */ 1723 /* One ACK acked hole. The rest eat duplicate ACKs. */
1715 if (acked-1 >= tp->sacked_out) 1724 if (acked-1 >= tp->sacked_out)
@@ -1728,9 +1737,10 @@ static inline void tcp_reset_reno_sack(struct tcp_sock *tp)
1728} 1737}
1729 1738
1730/* Mark head of queue up as lost. */ 1739/* Mark head of queue up as lost. */
1731static void tcp_mark_head_lost(struct sock *sk, struct tcp_sock *tp, 1740static void tcp_mark_head_lost(struct sock *sk,
1732 int packets, u32 high_seq) 1741 int packets, u32 high_seq)
1733{ 1742{
1743 struct tcp_sock *tp = tcp_sk(sk);
1734 struct sk_buff *skb; 1744 struct sk_buff *skb;
1735 int cnt; 1745 int cnt;
1736 1746
@@ -1771,15 +1781,17 @@ static void tcp_mark_head_lost(struct sock *sk, struct tcp_sock *tp,
1771 1781
1772/* Account newly detected lost packet(s) */ 1782/* Account newly detected lost packet(s) */
1773 1783
1774static void tcp_update_scoreboard(struct sock *sk, struct tcp_sock *tp) 1784static void tcp_update_scoreboard(struct sock *sk)
1775{ 1785{
1786 struct tcp_sock *tp = tcp_sk(sk);
1787
1776 if (IsFack(tp)) { 1788 if (IsFack(tp)) {
1777 int lost = tp->fackets_out - tp->reordering; 1789 int lost = tp->fackets_out - tp->reordering;
1778 if (lost <= 0) 1790 if (lost <= 0)
1779 lost = 1; 1791 lost = 1;
1780 tcp_mark_head_lost(sk, tp, lost, tp->high_seq); 1792 tcp_mark_head_lost(sk, lost, tp->high_seq);
1781 } else { 1793 } else {
1782 tcp_mark_head_lost(sk, tp, 1, tp->high_seq); 1794 tcp_mark_head_lost(sk, 1, tp->high_seq);
1783 } 1795 }
1784 1796
1785 /* New heuristics: it is possible only after we switched 1797 /* New heuristics: it is possible only after we switched
@@ -1787,7 +1799,7 @@ static void tcp_update_scoreboard(struct sock *sk, struct tcp_sock *tp)
1787 * Hence, we can detect timed out packets during fast 1799 * Hence, we can detect timed out packets during fast
1788 * retransmit without falling to slow start. 1800 * retransmit without falling to slow start.
1789 */ 1801 */
1790 if (!IsReno(tp) && tcp_head_timedout(sk, tp)) { 1802 if (!IsReno(tp) && tcp_head_timedout(sk)) {
1791 struct sk_buff *skb; 1803 struct sk_buff *skb;
1792 1804
1793 skb = tp->scoreboard_skb_hint ? tp->scoreboard_skb_hint 1805 skb = tp->scoreboard_skb_hint ? tp->scoreboard_skb_hint
@@ -1867,9 +1879,11 @@ static inline int tcp_packet_delayed(struct tcp_sock *tp)
1867/* Undo procedures. */ 1879/* Undo procedures. */
1868 1880
1869#if FASTRETRANS_DEBUG > 1 1881#if FASTRETRANS_DEBUG > 1
1870static void DBGUNDO(struct sock *sk, struct tcp_sock *tp, const char *msg) 1882static void DBGUNDO(struct sock *sk, const char *msg)
1871{ 1883{
1884 struct tcp_sock *tp = tcp_sk(sk);
1872 struct inet_sock *inet = inet_sk(sk); 1885 struct inet_sock *inet = inet_sk(sk);
1886
1873 printk(KERN_DEBUG "Undo %s %u.%u.%u.%u/%u c%u l%u ss%u/%u p%u\n", 1887 printk(KERN_DEBUG "Undo %s %u.%u.%u.%u/%u c%u l%u ss%u/%u p%u\n",
1874 msg, 1888 msg,
1875 NIPQUAD(inet->daddr), ntohs(inet->dport), 1889 NIPQUAD(inet->daddr), ntohs(inet->dport),
@@ -1915,13 +1929,15 @@ static inline int tcp_may_undo(struct tcp_sock *tp)
1915} 1929}
1916 1930
1917/* People celebrate: "We love our President!" */ 1931/* People celebrate: "We love our President!" */
1918static int tcp_try_undo_recovery(struct sock *sk, struct tcp_sock *tp) 1932static int tcp_try_undo_recovery(struct sock *sk)
1919{ 1933{
1934 struct tcp_sock *tp = tcp_sk(sk);
1935
1920 if (tcp_may_undo(tp)) { 1936 if (tcp_may_undo(tp)) {
1921 /* Happy end! We did not retransmit anything 1937 /* Happy end! We did not retransmit anything
1922 * or our original transmission succeeded. 1938 * or our original transmission succeeded.
1923 */ 1939 */
1924 DBGUNDO(sk, tp, inet_csk(sk)->icsk_ca_state == TCP_CA_Loss ? "loss" : "retrans"); 1940 DBGUNDO(sk, inet_csk(sk)->icsk_ca_state == TCP_CA_Loss ? "loss" : "retrans");
1925 tcp_undo_cwr(sk, 1); 1941 tcp_undo_cwr(sk, 1);
1926 if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss) 1942 if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss)
1927 NET_INC_STATS_BH(LINUX_MIB_TCPLOSSUNDO); 1943 NET_INC_STATS_BH(LINUX_MIB_TCPLOSSUNDO);
@@ -1941,10 +1957,12 @@ static int tcp_try_undo_recovery(struct sock *sk, struct tcp_sock *tp)
1941} 1957}
1942 1958
1943/* Try to undo cwnd reduction, because D-SACKs acked all retransmitted data */ 1959/* Try to undo cwnd reduction, because D-SACKs acked all retransmitted data */
1944static void tcp_try_undo_dsack(struct sock *sk, struct tcp_sock *tp) 1960static void tcp_try_undo_dsack(struct sock *sk)
1945{ 1961{
1962 struct tcp_sock *tp = tcp_sk(sk);
1963
1946 if (tp->undo_marker && !tp->undo_retrans) { 1964 if (tp->undo_marker && !tp->undo_retrans) {
1947 DBGUNDO(sk, tp, "D-SACK"); 1965 DBGUNDO(sk, "D-SACK");
1948 tcp_undo_cwr(sk, 1); 1966 tcp_undo_cwr(sk, 1);
1949 tp->undo_marker = 0; 1967 tp->undo_marker = 0;
1950 NET_INC_STATS_BH(LINUX_MIB_TCPDSACKUNDO); 1968 NET_INC_STATS_BH(LINUX_MIB_TCPDSACKUNDO);
@@ -1953,9 +1971,9 @@ static void tcp_try_undo_dsack(struct sock *sk, struct tcp_sock *tp)
1953 1971
1954/* Undo during fast recovery after partial ACK. */ 1972/* Undo during fast recovery after partial ACK. */
1955 1973
1956static int tcp_try_undo_partial(struct sock *sk, struct tcp_sock *tp, 1974static int tcp_try_undo_partial(struct sock *sk, int acked)
1957 int acked)
1958{ 1975{
1976 struct tcp_sock *tp = tcp_sk(sk);
1959 /* Partial ACK arrived. Force Hoe's retransmit. */ 1977 /* Partial ACK arrived. Force Hoe's retransmit. */
1960 int failed = IsReno(tp) || tp->fackets_out>tp->reordering; 1978 int failed = IsReno(tp) || tp->fackets_out>tp->reordering;
1961 1979
@@ -1968,7 +1986,7 @@ static int tcp_try_undo_partial(struct sock *sk, struct tcp_sock *tp,
1968 1986
1969 tcp_update_reordering(sk, tcp_fackets_out(tp) + acked, 1); 1987 tcp_update_reordering(sk, tcp_fackets_out(tp) + acked, 1);
1970 1988
1971 DBGUNDO(sk, tp, "Hoe"); 1989 DBGUNDO(sk, "Hoe");
1972 tcp_undo_cwr(sk, 0); 1990 tcp_undo_cwr(sk, 0);
1973 NET_INC_STATS_BH(LINUX_MIB_TCPPARTIALUNDO); 1991 NET_INC_STATS_BH(LINUX_MIB_TCPPARTIALUNDO);
1974 1992
@@ -1982,8 +2000,10 @@ static int tcp_try_undo_partial(struct sock *sk, struct tcp_sock *tp,
1982} 2000}
1983 2001
1984/* Undo during loss recovery after partial ACK. */ 2002/* Undo during loss recovery after partial ACK. */
1985static int tcp_try_undo_loss(struct sock *sk, struct tcp_sock *tp) 2003static int tcp_try_undo_loss(struct sock *sk)
1986{ 2004{
2005 struct tcp_sock *tp = tcp_sk(sk);
2006
1987 if (tcp_may_undo(tp)) { 2007 if (tcp_may_undo(tp)) {
1988 struct sk_buff *skb; 2008 struct sk_buff *skb;
1989 tcp_for_write_queue(skb, sk) { 2009 tcp_for_write_queue(skb, sk) {
@@ -1994,7 +2014,7 @@ static int tcp_try_undo_loss(struct sock *sk, struct tcp_sock *tp)
1994 2014
1995 clear_all_retrans_hints(tp); 2015 clear_all_retrans_hints(tp);
1996 2016
1997 DBGUNDO(sk, tp, "partial loss"); 2017 DBGUNDO(sk, "partial loss");
1998 tp->lost_out = 0; 2018 tp->lost_out = 0;
1999 tp->left_out = tp->sacked_out; 2019 tp->left_out = tp->sacked_out;
2000 tcp_undo_cwr(sk, 1); 2020 tcp_undo_cwr(sk, 1);
@@ -2016,8 +2036,10 @@ static inline void tcp_complete_cwr(struct sock *sk)
2016 tcp_ca_event(sk, CA_EVENT_COMPLETE_CWR); 2036 tcp_ca_event(sk, CA_EVENT_COMPLETE_CWR);
2017} 2037}
2018 2038
2019static void tcp_try_to_open(struct sock *sk, struct tcp_sock *tp, int flag) 2039static void tcp_try_to_open(struct sock *sk, int flag)
2020{ 2040{
2041 struct tcp_sock *tp = tcp_sk(sk);
2042
2021 tp->left_out = tp->sacked_out; 2043 tp->left_out = tp->sacked_out;
2022 2044
2023 if (tp->retrans_out == 0) 2045 if (tp->retrans_out == 0)
@@ -2111,7 +2133,7 @@ tcp_fastretrans_alert(struct sock *sk, u32 prior_snd_una,
2111 before(tp->snd_una, tp->high_seq) && 2133 before(tp->snd_una, tp->high_seq) &&
2112 icsk->icsk_ca_state != TCP_CA_Open && 2134 icsk->icsk_ca_state != TCP_CA_Open &&
2113 tp->fackets_out > tp->reordering) { 2135 tp->fackets_out > tp->reordering) {
2114 tcp_mark_head_lost(sk, tp, tp->fackets_out-tp->reordering, tp->high_seq); 2136 tcp_mark_head_lost(sk, tp->fackets_out-tp->reordering, tp->high_seq);
2115 NET_INC_STATS_BH(LINUX_MIB_TCPLOSS); 2137 NET_INC_STATS_BH(LINUX_MIB_TCPLOSS);
2116 } 2138 }
2117 2139
@@ -2127,7 +2149,7 @@ tcp_fastretrans_alert(struct sock *sk, u32 prior_snd_una,
2127 switch (icsk->icsk_ca_state) { 2149 switch (icsk->icsk_ca_state) {
2128 case TCP_CA_Loss: 2150 case TCP_CA_Loss:
2129 icsk->icsk_retransmits = 0; 2151 icsk->icsk_retransmits = 0;
2130 if (tcp_try_undo_recovery(sk, tp)) 2152 if (tcp_try_undo_recovery(sk))
2131 return; 2153 return;
2132 break; 2154 break;
2133 2155
@@ -2141,7 +2163,7 @@ tcp_fastretrans_alert(struct sock *sk, u32 prior_snd_una,
2141 break; 2163 break;
2142 2164
2143 case TCP_CA_Disorder: 2165 case TCP_CA_Disorder:
2144 tcp_try_undo_dsack(sk, tp); 2166 tcp_try_undo_dsack(sk);
2145 if (!tp->undo_marker || 2167 if (!tp->undo_marker ||
2146 /* For SACK case do not Open to allow to undo 2168 /* For SACK case do not Open to allow to undo
2147 * catching for all duplicate ACKs. */ 2169 * catching for all duplicate ACKs. */
@@ -2154,7 +2176,7 @@ tcp_fastretrans_alert(struct sock *sk, u32 prior_snd_una,
2154 case TCP_CA_Recovery: 2176 case TCP_CA_Recovery:
2155 if (IsReno(tp)) 2177 if (IsReno(tp))
2156 tcp_reset_reno_sack(tp); 2178 tcp_reset_reno_sack(tp);
2157 if (tcp_try_undo_recovery(sk, tp)) 2179 if (tcp_try_undo_recovery(sk))
2158 return; 2180 return;
2159 tcp_complete_cwr(sk); 2181 tcp_complete_cwr(sk);
2160 break; 2182 break;
@@ -2170,14 +2192,14 @@ tcp_fastretrans_alert(struct sock *sk, u32 prior_snd_una,
2170 } else { 2192 } else {
2171 int acked = prior_packets - tp->packets_out; 2193 int acked = prior_packets - tp->packets_out;
2172 if (IsReno(tp)) 2194 if (IsReno(tp))
2173 tcp_remove_reno_sacks(sk, tp, acked); 2195 tcp_remove_reno_sacks(sk, acked);
2174 is_dupack = tcp_try_undo_partial(sk, tp, acked); 2196 is_dupack = tcp_try_undo_partial(sk, acked);
2175 } 2197 }
2176 break; 2198 break;
2177 case TCP_CA_Loss: 2199 case TCP_CA_Loss:
2178 if (flag&FLAG_DATA_ACKED) 2200 if (flag&FLAG_DATA_ACKED)
2179 icsk->icsk_retransmits = 0; 2201 icsk->icsk_retransmits = 0;
2180 if (!tcp_try_undo_loss(sk, tp)) { 2202 if (!tcp_try_undo_loss(sk)) {
2181 tcp_moderate_cwnd(tp); 2203 tcp_moderate_cwnd(tp);
2182 tcp_xmit_retransmit_queue(sk); 2204 tcp_xmit_retransmit_queue(sk);
2183 return; 2205 return;
@@ -2194,10 +2216,10 @@ tcp_fastretrans_alert(struct sock *sk, u32 prior_snd_una,
2194 } 2216 }
2195 2217
2196 if (icsk->icsk_ca_state == TCP_CA_Disorder) 2218 if (icsk->icsk_ca_state == TCP_CA_Disorder)
2197 tcp_try_undo_dsack(sk, tp); 2219 tcp_try_undo_dsack(sk);
2198 2220
2199 if (!tcp_time_to_recover(sk, tp)) { 2221 if (!tcp_time_to_recover(sk)) {
2200 tcp_try_to_open(sk, tp, flag); 2222 tcp_try_to_open(sk, flag);
2201 return; 2223 return;
2202 } 2224 }
2203 2225
@@ -2236,8 +2258,8 @@ tcp_fastretrans_alert(struct sock *sk, u32 prior_snd_una,
2236 tcp_set_ca_state(sk, TCP_CA_Recovery); 2258 tcp_set_ca_state(sk, TCP_CA_Recovery);
2237 } 2259 }
2238 2260
2239 if (is_dupack || tcp_head_timedout(sk, tp)) 2261 if (is_dupack || tcp_head_timedout(sk))
2240 tcp_update_scoreboard(sk, tp); 2262 tcp_update_scoreboard(sk);
2241 tcp_cwnd_down(sk); 2263 tcp_cwnd_down(sk);
2242 tcp_xmit_retransmit_queue(sk); 2264 tcp_xmit_retransmit_queue(sk);
2243} 2265}
@@ -2313,8 +2335,10 @@ static void tcp_cong_avoid(struct sock *sk, u32 ack, u32 rtt,
2313 * RFC2988 recommends to restart timer to now+rto. 2335 * RFC2988 recommends to restart timer to now+rto.
2314 */ 2336 */
2315 2337
2316static void tcp_ack_packets_out(struct sock *sk, struct tcp_sock *tp) 2338static void tcp_ack_packets_out(struct sock *sk)
2317{ 2339{
2340 struct tcp_sock *tp = tcp_sk(sk);
2341
2318 if (!tp->packets_out) { 2342 if (!tp->packets_out) {
2319 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS); 2343 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
2320 } else { 2344 } else {
@@ -2471,7 +2495,7 @@ static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p)
2471 2495
2472 if (acked&FLAG_ACKED) { 2496 if (acked&FLAG_ACKED) {
2473 tcp_ack_update_rtt(sk, acked, seq_rtt); 2497 tcp_ack_update_rtt(sk, acked, seq_rtt);
2474 tcp_ack_packets_out(sk, tp); 2498 tcp_ack_packets_out(sk);
2475 if (rtt_sample && !(acked & FLAG_RETRANS_DATA_ACKED)) 2499 if (rtt_sample && !(acked & FLAG_RETRANS_DATA_ACKED))
2476 (*rtt_sample)(sk, tcp_usrtt(&tv)); 2500 (*rtt_sample)(sk, tcp_usrtt(&tv));
2477 2501
@@ -2556,9 +2580,10 @@ static inline int tcp_may_update_window(const struct tcp_sock *tp, const u32 ack
2556 * Window update algorithm, described in RFC793/RFC1122 (used in linux-2.2 2580 * Window update algorithm, described in RFC793/RFC1122 (used in linux-2.2
2557 * and in FreeBSD. NetBSD's one is even worse.) is wrong. 2581 * and in FreeBSD. NetBSD's one is even worse.) is wrong.
2558 */ 2582 */
2559static int tcp_ack_update_window(struct sock *sk, struct tcp_sock *tp, 2583static int tcp_ack_update_window(struct sock *sk, struct sk_buff *skb, u32 ack,
2560 struct sk_buff *skb, u32 ack, u32 ack_seq) 2584 u32 ack_seq)
2561{ 2585{
2586 struct tcp_sock *tp = tcp_sk(sk);
2562 int flag = 0; 2587 int flag = 0;
2563 u32 nwin = ntohs(tcp_hdr(skb)->window); 2588 u32 nwin = ntohs(tcp_hdr(skb)->window);
2564 2589
@@ -2576,7 +2601,7 @@ static int tcp_ack_update_window(struct sock *sk, struct tcp_sock *tp,
2576 * fast path is recovered for sending TCP. 2601 * fast path is recovered for sending TCP.
2577 */ 2602 */
2578 tp->pred_flags = 0; 2603 tp->pred_flags = 0;
2579 tcp_fast_path_check(sk, tp); 2604 tcp_fast_path_check(sk);
2580 2605
2581 if (nwin > tp->max_window) { 2606 if (nwin > tp->max_window) {
2582 tp->max_window = nwin; 2607 tp->max_window = nwin;
@@ -2762,7 +2787,7 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag)
2762 else 2787 else
2763 NET_INC_STATS_BH(LINUX_MIB_TCPPUREACKS); 2788 NET_INC_STATS_BH(LINUX_MIB_TCPPUREACKS);
2764 2789
2765 flag |= tcp_ack_update_window(sk, tp, skb, ack, ack_seq); 2790 flag |= tcp_ack_update_window(sk, skb, ack, ack_seq);
2766 2791
2767 if (TCP_SKB_CB(skb)->sacked) 2792 if (TCP_SKB_CB(skb)->sacked)
2768 flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una); 2793 flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una);
@@ -3426,7 +3451,7 @@ queue_and_out:
3426 } 3451 }
3427 tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq; 3452 tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
3428 if (skb->len) 3453 if (skb->len)
3429 tcp_event_data_recv(sk, tp, skb); 3454 tcp_event_data_recv(sk, skb);
3430 if (th->fin) 3455 if (th->fin)
3431 tcp_fin(skb, sk, th); 3456 tcp_fin(skb, sk, th);
3432 3457
@@ -3443,7 +3468,7 @@ queue_and_out:
3443 if (tp->rx_opt.num_sacks) 3468 if (tp->rx_opt.num_sacks)
3444 tcp_sack_remove(tp); 3469 tcp_sack_remove(tp);
3445 3470
3446 tcp_fast_path_check(sk, tp); 3471 tcp_fast_path_check(sk);
3447 3472
3448 if (eaten > 0) 3473 if (eaten > 0)
3449 __kfree_skb(skb); 3474 __kfree_skb(skb);
@@ -3734,7 +3759,7 @@ static int tcp_prune_queue(struct sock *sk)
3734 NET_INC_STATS_BH(LINUX_MIB_PRUNECALLED); 3759 NET_INC_STATS_BH(LINUX_MIB_PRUNECALLED);
3735 3760
3736 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) 3761 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
3737 tcp_clamp_window(sk, tp); 3762 tcp_clamp_window(sk);
3738 else if (tcp_memory_pressure) 3763 else if (tcp_memory_pressure)
3739 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss); 3764 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss);
3740 3765
@@ -3803,8 +3828,10 @@ void tcp_cwnd_application_limited(struct sock *sk)
3803 tp->snd_cwnd_stamp = tcp_time_stamp; 3828 tp->snd_cwnd_stamp = tcp_time_stamp;
3804} 3829}
3805 3830
3806static int tcp_should_expand_sndbuf(struct sock *sk, struct tcp_sock *tp) 3831static int tcp_should_expand_sndbuf(struct sock *sk)
3807{ 3832{
3833 struct tcp_sock *tp = tcp_sk(sk);
3834
3808 /* If the user specified a specific send buffer setting, do 3835 /* If the user specified a specific send buffer setting, do
3809 * not modify it. 3836 * not modify it.
3810 */ 3837 */
@@ -3836,7 +3863,7 @@ static void tcp_new_space(struct sock *sk)
3836{ 3863{
3837 struct tcp_sock *tp = tcp_sk(sk); 3864 struct tcp_sock *tp = tcp_sk(sk);
3838 3865
3839 if (tcp_should_expand_sndbuf(sk, tp)) { 3866 if (tcp_should_expand_sndbuf(sk)) {
3840 int sndmem = max_t(u32, tp->rx_opt.mss_clamp, tp->mss_cache) + 3867 int sndmem = max_t(u32, tp->rx_opt.mss_clamp, tp->mss_cache) +
3841 MAX_TCP_HEADER + 16 + sizeof(struct sk_buff), 3868 MAX_TCP_HEADER + 16 + sizeof(struct sk_buff),
3842 demanded = max_t(unsigned int, tp->snd_cwnd, 3869 demanded = max_t(unsigned int, tp->snd_cwnd,
@@ -3860,9 +3887,9 @@ static void tcp_check_space(struct sock *sk)
3860 } 3887 }
3861} 3888}
3862 3889
3863static inline void tcp_data_snd_check(struct sock *sk, struct tcp_sock *tp) 3890static inline void tcp_data_snd_check(struct sock *sk)
3864{ 3891{
3865 tcp_push_pending_frames(sk, tp); 3892 tcp_push_pending_frames(sk);
3866 tcp_check_space(sk); 3893 tcp_check_space(sk);
3867} 3894}
3868 3895
@@ -4196,7 +4223,7 @@ int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
4196 */ 4223 */
4197 tcp_ack(sk, skb, 0); 4224 tcp_ack(sk, skb, 0);
4198 __kfree_skb(skb); 4225 __kfree_skb(skb);
4199 tcp_data_snd_check(sk, tp); 4226 tcp_data_snd_check(sk);
4200 return 0; 4227 return 0;
4201 } else { /* Header too small */ 4228 } else { /* Header too small */
4202 TCP_INC_STATS_BH(TCP_MIB_INERRS); 4229 TCP_INC_STATS_BH(TCP_MIB_INERRS);
@@ -4267,12 +4294,12 @@ int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
4267 tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq; 4294 tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
4268 } 4295 }
4269 4296
4270 tcp_event_data_recv(sk, tp, skb); 4297 tcp_event_data_recv(sk, skb);
4271 4298
4272 if (TCP_SKB_CB(skb)->ack_seq != tp->snd_una) { 4299 if (TCP_SKB_CB(skb)->ack_seq != tp->snd_una) {
4273 /* Well, only one small jumplet in fast path... */ 4300 /* Well, only one small jumplet in fast path... */
4274 tcp_ack(sk, skb, FLAG_DATA); 4301 tcp_ack(sk, skb, FLAG_DATA);
4275 tcp_data_snd_check(sk, tp); 4302 tcp_data_snd_check(sk);
4276 if (!inet_csk_ack_scheduled(sk)) 4303 if (!inet_csk_ack_scheduled(sk))
4277 goto no_ack; 4304 goto no_ack;
4278 } 4305 }
@@ -4355,7 +4382,7 @@ step5:
4355 /* step 7: process the segment text */ 4382 /* step 7: process the segment text */
4356 tcp_data_queue(sk, skb); 4383 tcp_data_queue(sk, skb);
4357 4384
4358 tcp_data_snd_check(sk, tp); 4385 tcp_data_snd_check(sk);
4359 tcp_ack_snd_check(sk); 4386 tcp_ack_snd_check(sk);
4360 return 0; 4387 return 0;
4361 4388
@@ -4672,7 +4699,7 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
4672 /* Do step6 onward by hand. */ 4699 /* Do step6 onward by hand. */
4673 tcp_urg(sk, skb, th); 4700 tcp_urg(sk, skb, th);
4674 __kfree_skb(skb); 4701 __kfree_skb(skb);
4675 tcp_data_snd_check(sk, tp); 4702 tcp_data_snd_check(sk);
4676 return 0; 4703 return 0;
4677 } 4704 }
4678 4705
@@ -4864,7 +4891,7 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
4864 4891
4865 /* tcp_data could move socket to TIME-WAIT */ 4892 /* tcp_data could move socket to TIME-WAIT */
4866 if (sk->sk_state != TCP_CLOSE) { 4893 if (sk->sk_state != TCP_CLOSE) {
4867 tcp_data_snd_check(sk, tp); 4894 tcp_data_snd_check(sk);
4868 tcp_ack_snd_check(sk); 4895 tcp_ack_snd_check(sk);
4869 } 4896 }
4870 4897
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 94d9f0c63682..3a60aea744ae 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -62,12 +62,13 @@ int sysctl_tcp_base_mss __read_mostly = 512;
62/* By default, RFC2861 behavior. */ 62/* By default, RFC2861 behavior. */
63int sysctl_tcp_slow_start_after_idle __read_mostly = 1; 63int sysctl_tcp_slow_start_after_idle __read_mostly = 1;
64 64
65static void update_send_head(struct sock *sk, struct tcp_sock *tp, 65static void update_send_head(struct sock *sk, struct sk_buff *skb)
66 struct sk_buff *skb)
67{ 66{
67 struct tcp_sock *tp = tcp_sk(sk);
68
68 tcp_advance_send_head(sk, skb); 69 tcp_advance_send_head(sk, skb);
69 tp->snd_nxt = TCP_SKB_CB(skb)->end_seq; 70 tp->snd_nxt = TCP_SKB_CB(skb)->end_seq;
70 tcp_packets_out_inc(sk, tp, skb); 71 tcp_packets_out_inc(sk, skb);
71} 72}
72 73
73/* SND.NXT, if window was not shrunk. 74/* SND.NXT, if window was not shrunk.
@@ -76,8 +77,10 @@ static void update_send_head(struct sock *sk, struct tcp_sock *tp,
76 * Anything in between SND.UNA...SND.UNA+SND.WND also can be already 77 * Anything in between SND.UNA...SND.UNA+SND.WND also can be already
77 * invalid. OK, let's make this for now: 78 * invalid. OK, let's make this for now:
78 */ 79 */
79static inline __u32 tcp_acceptable_seq(struct sock *sk, struct tcp_sock *tp) 80static inline __u32 tcp_acceptable_seq(struct sock *sk)
80{ 81{
82 struct tcp_sock *tp = tcp_sk(sk);
83
81 if (!before(tp->snd_una+tp->snd_wnd, tp->snd_nxt)) 84 if (!before(tp->snd_una+tp->snd_wnd, tp->snd_nxt))
82 return tp->snd_nxt; 85 return tp->snd_nxt;
83 else 86 else
@@ -516,7 +519,7 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
516 md5 ? &md5_hash_location : 519 md5 ? &md5_hash_location :
517#endif 520#endif
518 NULL); 521 NULL);
519 TCP_ECN_send(sk, tp, skb, tcp_header_size); 522 TCP_ECN_send(sk, skb, tcp_header_size);
520 } 523 }
521 524
522#ifdef CONFIG_TCP_MD5SIG 525#ifdef CONFIG_TCP_MD5SIG
@@ -927,8 +930,9 @@ unsigned int tcp_current_mss(struct sock *sk, int large_allowed)
927 930
928/* Congestion window validation. (RFC2861) */ 931/* Congestion window validation. (RFC2861) */
929 932
930static void tcp_cwnd_validate(struct sock *sk, struct tcp_sock *tp) 933static void tcp_cwnd_validate(struct sock *sk)
931{ 934{
935 struct tcp_sock *tp = tcp_sk(sk);
932 __u32 packets_out = tp->packets_out; 936 __u32 packets_out = tp->packets_out;
933 937
934 if (packets_out >= tp->snd_cwnd) { 938 if (packets_out >= tp->snd_cwnd) {
@@ -1076,8 +1080,9 @@ static unsigned int tcp_snd_test(struct sock *sk, struct sk_buff *skb,
1076 return cwnd_quota; 1080 return cwnd_quota;
1077} 1081}
1078 1082
1079int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp) 1083int tcp_may_send_now(struct sock *sk)
1080{ 1084{
1085 struct tcp_sock *tp = tcp_sk(sk);
1081 struct sk_buff *skb = tcp_send_head(sk); 1086 struct sk_buff *skb = tcp_send_head(sk);
1082 1087
1083 return (skb && 1088 return (skb &&
@@ -1144,8 +1149,9 @@ static int tso_fragment(struct sock *sk, struct sk_buff *skb, unsigned int len,
1144 * 1149 *
1145 * This algorithm is from John Heffner. 1150 * This algorithm is from John Heffner.
1146 */ 1151 */
1147static int tcp_tso_should_defer(struct sock *sk, struct tcp_sock *tp, struct sk_buff *skb) 1152static int tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb)
1148{ 1153{
1154 struct tcp_sock *tp = tcp_sk(sk);
1149 const struct inet_connection_sock *icsk = inet_csk(sk); 1155 const struct inet_connection_sock *icsk = inet_csk(sk);
1150 u32 send_win, cong_win, limit, in_flight; 1156 u32 send_win, cong_win, limit, in_flight;
1151 1157
@@ -1324,7 +1330,7 @@ static int tcp_mtu_probe(struct sock *sk)
1324 /* Decrement cwnd here because we are sending 1330 /* Decrement cwnd here because we are sending
1325 * effectively two packets. */ 1331 * effectively two packets. */
1326 tp->snd_cwnd--; 1332 tp->snd_cwnd--;
1327 update_send_head(sk, tp, nskb); 1333 update_send_head(sk, nskb);
1328 1334
1329 icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len); 1335 icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len);
1330 tp->mtu_probe.probe_seq_start = TCP_SKB_CB(nskb)->seq; 1336 tp->mtu_probe.probe_seq_start = TCP_SKB_CB(nskb)->seq;
@@ -1387,7 +1393,7 @@ static int tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle)
1387 nonagle : TCP_NAGLE_PUSH)))) 1393 nonagle : TCP_NAGLE_PUSH))))
1388 break; 1394 break;
1389 } else { 1395 } else {
1390 if (tcp_tso_should_defer(sk, tp, skb)) 1396 if (tcp_tso_should_defer(sk, skb))
1391 break; 1397 break;
1392 } 1398 }
1393 1399
@@ -1416,14 +1422,14 @@ static int tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle)
1416 /* Advance the send_head. This one is sent out. 1422 /* Advance the send_head. This one is sent out.
1417 * This call will increment packets_out. 1423 * This call will increment packets_out.
1418 */ 1424 */
1419 update_send_head(sk, tp, skb); 1425 update_send_head(sk, skb);
1420 1426
1421 tcp_minshall_update(tp, mss_now, skb); 1427 tcp_minshall_update(tp, mss_now, skb);
1422 sent_pkts++; 1428 sent_pkts++;
1423 } 1429 }
1424 1430
1425 if (likely(sent_pkts)) { 1431 if (likely(sent_pkts)) {
1426 tcp_cwnd_validate(sk, tp); 1432 tcp_cwnd_validate(sk);
1427 return 0; 1433 return 0;
1428 } 1434 }
1429 return !tp->packets_out && tcp_send_head(sk); 1435 return !tp->packets_out && tcp_send_head(sk);
@@ -1433,14 +1439,14 @@ static int tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle)
1433 * TCP_CORK or attempt at coalescing tiny packets. 1439 * TCP_CORK or attempt at coalescing tiny packets.
1434 * The socket must be locked by the caller. 1440 * The socket must be locked by the caller.
1435 */ 1441 */
1436void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp, 1442void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
1437 unsigned int cur_mss, int nonagle) 1443 int nonagle)
1438{ 1444{
1439 struct sk_buff *skb = tcp_send_head(sk); 1445 struct sk_buff *skb = tcp_send_head(sk);
1440 1446
1441 if (skb) { 1447 if (skb) {
1442 if (tcp_write_xmit(sk, cur_mss, nonagle)) 1448 if (tcp_write_xmit(sk, cur_mss, nonagle))
1443 tcp_check_probe_timer(sk, tp); 1449 tcp_check_probe_timer(sk);
1444 } 1450 }
1445} 1451}
1446 1452
@@ -1484,8 +1490,8 @@ void tcp_push_one(struct sock *sk, unsigned int mss_now)
1484 TCP_SKB_CB(skb)->when = tcp_time_stamp; 1490 TCP_SKB_CB(skb)->when = tcp_time_stamp;
1485 1491
1486 if (likely(!tcp_transmit_skb(sk, skb, 1, sk->sk_allocation))) { 1492 if (likely(!tcp_transmit_skb(sk, skb, 1, sk->sk_allocation))) {
1487 update_send_head(sk, tp, skb); 1493 update_send_head(sk, skb);
1488 tcp_cwnd_validate(sk, tp); 1494 tcp_cwnd_validate(sk);
1489 return; 1495 return;
1490 } 1496 }
1491 } 1497 }
@@ -1933,7 +1939,7 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
1933 * segments to send. 1939 * segments to send.
1934 */ 1940 */
1935 1941
1936 if (tcp_may_send_now(sk, tp)) 1942 if (tcp_may_send_now(sk))
1937 return; 1943 return;
1938 1944
1939 if (tp->forward_skb_hint) { 1945 if (tp->forward_skb_hint) {
@@ -2023,7 +2029,7 @@ void tcp_send_fin(struct sock *sk)
2023 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(skb)->seq + 1; 2029 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(skb)->seq + 1;
2024 tcp_queue_skb(sk, skb); 2030 tcp_queue_skb(sk, skb);
2025 } 2031 }
2026 __tcp_push_pending_frames(sk, tp, mss_now, TCP_NAGLE_OFF); 2032 __tcp_push_pending_frames(sk, mss_now, TCP_NAGLE_OFF);
2027} 2033}
2028 2034
2029/* We get here when a process closes a file descriptor (either due to 2035/* We get here when a process closes a file descriptor (either due to
@@ -2033,7 +2039,6 @@ void tcp_send_fin(struct sock *sk)
2033 */ 2039 */
2034void tcp_send_active_reset(struct sock *sk, gfp_t priority) 2040void tcp_send_active_reset(struct sock *sk, gfp_t priority)
2035{ 2041{
2036 struct tcp_sock *tp = tcp_sk(sk);
2037 struct sk_buff *skb; 2042 struct sk_buff *skb;
2038 2043
2039 /* NOTE: No TCP options attached and we never retransmit this. */ 2044 /* NOTE: No TCP options attached and we never retransmit this. */
@@ -2053,7 +2058,7 @@ void tcp_send_active_reset(struct sock *sk, gfp_t priority)
2053 skb_shinfo(skb)->gso_type = 0; 2058 skb_shinfo(skb)->gso_type = 0;
2054 2059
2055 /* Send it off. */ 2060 /* Send it off. */
2056 TCP_SKB_CB(skb)->seq = tcp_acceptable_seq(sk, tp); 2061 TCP_SKB_CB(skb)->seq = tcp_acceptable_seq(sk);
2057 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(skb)->seq; 2062 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(skb)->seq;
2058 TCP_SKB_CB(skb)->when = tcp_time_stamp; 2063 TCP_SKB_CB(skb)->when = tcp_time_stamp;
2059 if (tcp_transmit_skb(sk, skb, 0, priority)) 2064 if (tcp_transmit_skb(sk, skb, 0, priority))
@@ -2271,7 +2276,7 @@ int tcp_connect(struct sock *sk)
2271 skb_reserve(buff, MAX_TCP_HEADER); 2276 skb_reserve(buff, MAX_TCP_HEADER);
2272 2277
2273 TCP_SKB_CB(buff)->flags = TCPCB_FLAG_SYN; 2278 TCP_SKB_CB(buff)->flags = TCPCB_FLAG_SYN;
2274 TCP_ECN_send_syn(sk, tp, buff); 2279 TCP_ECN_send_syn(sk, buff);
2275 TCP_SKB_CB(buff)->sacked = 0; 2280 TCP_SKB_CB(buff)->sacked = 0;
2276 skb_shinfo(buff)->gso_segs = 1; 2281 skb_shinfo(buff)->gso_segs = 1;
2277 skb_shinfo(buff)->gso_size = 0; 2282 skb_shinfo(buff)->gso_size = 0;
@@ -2363,7 +2368,6 @@ void tcp_send_ack(struct sock *sk)
2363{ 2368{
2364 /* If we have been reset, we may not send again. */ 2369 /* If we have been reset, we may not send again. */
2365 if (sk->sk_state != TCP_CLOSE) { 2370 if (sk->sk_state != TCP_CLOSE) {
2366 struct tcp_sock *tp = tcp_sk(sk);
2367 struct sk_buff *buff; 2371 struct sk_buff *buff;
2368 2372
2369 /* We are not putting this on the write queue, so 2373 /* We are not putting this on the write queue, so
@@ -2389,7 +2393,7 @@ void tcp_send_ack(struct sock *sk)
2389 skb_shinfo(buff)->gso_type = 0; 2393 skb_shinfo(buff)->gso_type = 0;
2390 2394
2391 /* Send it off, this clears delayed acks for us. */ 2395 /* Send it off, this clears delayed acks for us. */
2392 TCP_SKB_CB(buff)->seq = TCP_SKB_CB(buff)->end_seq = tcp_acceptable_seq(sk, tp); 2396 TCP_SKB_CB(buff)->seq = TCP_SKB_CB(buff)->end_seq = tcp_acceptable_seq(sk);
2393 TCP_SKB_CB(buff)->when = tcp_time_stamp; 2397 TCP_SKB_CB(buff)->when = tcp_time_stamp;
2394 tcp_transmit_skb(sk, buff, 0, GFP_ATOMIC); 2398 tcp_transmit_skb(sk, buff, 0, GFP_ATOMIC);
2395 } 2399 }
@@ -2467,7 +2471,7 @@ int tcp_write_wakeup(struct sock *sk)
2467 TCP_SKB_CB(skb)->when = tcp_time_stamp; 2471 TCP_SKB_CB(skb)->when = tcp_time_stamp;
2468 err = tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC); 2472 err = tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
2469 if (!err) { 2473 if (!err) {
2470 update_send_head(sk, tp, skb); 2474 update_send_head(sk, skb);
2471 } 2475 }
2472 return err; 2476 return err;
2473 } else { 2477 } else {