aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--net/ipv4/tcp_input.c436
-rw-r--r--net/ipv4/tcp_output.c146
2 files changed, 300 insertions, 282 deletions
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 1e7fd8113663..18e099c6fa62 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -121,8 +121,7 @@ int sysctl_tcp_abc __read_mostly;
121/* Adapt the MSS value used to make delayed ack decision to the 121/* Adapt the MSS value used to make delayed ack decision to the
122 * real world. 122 * real world.
123 */ 123 */
124static void tcp_measure_rcv_mss(struct sock *sk, 124static void tcp_measure_rcv_mss(struct sock *sk, const struct sk_buff *skb)
125 const struct sk_buff *skb)
126{ 125{
127 struct inet_connection_sock *icsk = inet_csk(sk); 126 struct inet_connection_sock *icsk = inet_csk(sk);
128 const unsigned int lss = icsk->icsk_ack.last_seg_size; 127 const unsigned int lss = icsk->icsk_ack.last_seg_size;
@@ -133,7 +132,7 @@ static void tcp_measure_rcv_mss(struct sock *sk,
133 /* skb->len may jitter because of SACKs, even if peer 132 /* skb->len may jitter because of SACKs, even if peer
134 * sends good full-sized frames. 133 * sends good full-sized frames.
135 */ 134 */
136 len = skb_shinfo(skb)->gso_size ?: skb->len; 135 len = skb_shinfo(skb)->gso_size ? : skb->len;
137 if (len >= icsk->icsk_ack.rcv_mss) { 136 if (len >= icsk->icsk_ack.rcv_mss) {
138 icsk->icsk_ack.rcv_mss = len; 137 icsk->icsk_ack.rcv_mss = len;
139 } else { 138 } else {
@@ -173,8 +172,8 @@ static void tcp_incr_quickack(struct sock *sk)
173 struct inet_connection_sock *icsk = inet_csk(sk); 172 struct inet_connection_sock *icsk = inet_csk(sk);
174 unsigned quickacks = tcp_sk(sk)->rcv_wnd / (2 * icsk->icsk_ack.rcv_mss); 173 unsigned quickacks = tcp_sk(sk)->rcv_wnd / (2 * icsk->icsk_ack.rcv_mss);
175 174
176 if (quickacks==0) 175 if (quickacks == 0)
177 quickacks=2; 176 quickacks = 2;
178 if (quickacks > icsk->icsk_ack.quick) 177 if (quickacks > icsk->icsk_ack.quick)
179 icsk->icsk_ack.quick = min(quickacks, TCP_MAX_QUICKACKS); 178 icsk->icsk_ack.quick = min(quickacks, TCP_MAX_QUICKACKS);
180} 179}
@@ -199,7 +198,7 @@ static inline int tcp_in_quickack_mode(const struct sock *sk)
199 198
200static inline void TCP_ECN_queue_cwr(struct tcp_sock *tp) 199static inline void TCP_ECN_queue_cwr(struct tcp_sock *tp)
201{ 200{
202 if (tp->ecn_flags&TCP_ECN_OK) 201 if (tp->ecn_flags & TCP_ECN_OK)
203 tp->ecn_flags |= TCP_ECN_QUEUE_CWR; 202 tp->ecn_flags |= TCP_ECN_QUEUE_CWR;
204} 203}
205 204
@@ -216,7 +215,7 @@ static inline void TCP_ECN_withdraw_cwr(struct tcp_sock *tp)
216 215
217static inline void TCP_ECN_check_ce(struct tcp_sock *tp, struct sk_buff *skb) 216static inline void TCP_ECN_check_ce(struct tcp_sock *tp, struct sk_buff *skb)
218{ 217{
219 if (tp->ecn_flags&TCP_ECN_OK) { 218 if (tp->ecn_flags & TCP_ECN_OK) {
220 if (INET_ECN_is_ce(TCP_SKB_CB(skb)->flags)) 219 if (INET_ECN_is_ce(TCP_SKB_CB(skb)->flags))
221 tp->ecn_flags |= TCP_ECN_DEMAND_CWR; 220 tp->ecn_flags |= TCP_ECN_DEMAND_CWR;
222 /* Funny extension: if ECT is not set on a segment, 221 /* Funny extension: if ECT is not set on a segment,
@@ -229,19 +228,19 @@ static inline void TCP_ECN_check_ce(struct tcp_sock *tp, struct sk_buff *skb)
229 228
230static inline void TCP_ECN_rcv_synack(struct tcp_sock *tp, struct tcphdr *th) 229static inline void TCP_ECN_rcv_synack(struct tcp_sock *tp, struct tcphdr *th)
231{ 230{
232 if ((tp->ecn_flags&TCP_ECN_OK) && (!th->ece || th->cwr)) 231 if ((tp->ecn_flags & TCP_ECN_OK) && (!th->ece || th->cwr))
233 tp->ecn_flags &= ~TCP_ECN_OK; 232 tp->ecn_flags &= ~TCP_ECN_OK;
234} 233}
235 234
236static inline void TCP_ECN_rcv_syn(struct tcp_sock *tp, struct tcphdr *th) 235static inline void TCP_ECN_rcv_syn(struct tcp_sock *tp, struct tcphdr *th)
237{ 236{
238 if ((tp->ecn_flags&TCP_ECN_OK) && (!th->ece || !th->cwr)) 237 if ((tp->ecn_flags & TCP_ECN_OK) && (!th->ece || !th->cwr))
239 tp->ecn_flags &= ~TCP_ECN_OK; 238 tp->ecn_flags &= ~TCP_ECN_OK;
240} 239}
241 240
242static inline int TCP_ECN_rcv_ecn_echo(struct tcp_sock *tp, struct tcphdr *th) 241static inline int TCP_ECN_rcv_ecn_echo(struct tcp_sock *tp, struct tcphdr *th)
243{ 242{
244 if (th->ece && !th->syn && (tp->ecn_flags&TCP_ECN_OK)) 243 if (th->ece && !th->syn && (tp->ecn_flags & TCP_ECN_OK))
245 return 1; 244 return 1;
246 return 0; 245 return 0;
247} 246}
@@ -303,8 +302,7 @@ static int __tcp_grow_window(const struct sock *sk, const struct sk_buff *skb)
303 return 0; 302 return 0;
304} 303}
305 304
306static void tcp_grow_window(struct sock *sk, 305static void tcp_grow_window(struct sock *sk, struct sk_buff *skb)
307 struct sk_buff *skb)
308{ 306{
309 struct tcp_sock *tp = tcp_sk(sk); 307 struct tcp_sock *tp = tcp_sk(sk);
310 308
@@ -318,12 +316,13 @@ static void tcp_grow_window(struct sock *sk,
318 * will fit to rcvbuf in future. 316 * will fit to rcvbuf in future.
319 */ 317 */
320 if (tcp_win_from_space(skb->truesize) <= skb->len) 318 if (tcp_win_from_space(skb->truesize) <= skb->len)
321 incr = 2*tp->advmss; 319 incr = 2 * tp->advmss;
322 else 320 else
323 incr = __tcp_grow_window(sk, skb); 321 incr = __tcp_grow_window(sk, skb);
324 322
325 if (incr) { 323 if (incr) {
326 tp->rcv_ssthresh = min(tp->rcv_ssthresh + incr, tp->window_clamp); 324 tp->rcv_ssthresh = min(tp->rcv_ssthresh + incr,
325 tp->window_clamp);
327 inet_csk(sk)->icsk_ack.quick |= 1; 326 inet_csk(sk)->icsk_ack.quick |= 1;
328 } 327 }
329 } 328 }
@@ -398,10 +397,9 @@ static void tcp_clamp_window(struct sock *sk)
398 sysctl_tcp_rmem[2]); 397 sysctl_tcp_rmem[2]);
399 } 398 }
400 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf) 399 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf)
401 tp->rcv_ssthresh = min(tp->window_clamp, 2U*tp->advmss); 400 tp->rcv_ssthresh = min(tp->window_clamp, 2U * tp->advmss);
402} 401}
403 402
404
405/* Initialize RCV_MSS value. 403/* Initialize RCV_MSS value.
406 * RCV_MSS is an our guess about MSS used by the peer. 404 * RCV_MSS is an our guess about MSS used by the peer.
407 * We haven't any direct information about the MSS. 405 * We haven't any direct information about the MSS.
@@ -414,7 +412,7 @@ void tcp_initialize_rcv_mss(struct sock *sk)
414 struct tcp_sock *tp = tcp_sk(sk); 412 struct tcp_sock *tp = tcp_sk(sk);
415 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache); 413 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
416 414
417 hint = min(hint, tp->rcv_wnd/2); 415 hint = min(hint, tp->rcv_wnd / 2);
418 hint = min(hint, TCP_MIN_RCVMSS); 416 hint = min(hint, TCP_MIN_RCVMSS);
419 hint = max(hint, TCP_MIN_MSS); 417 hint = max(hint, TCP_MIN_MSS);
420 418
@@ -471,16 +469,15 @@ static inline void tcp_rcv_rtt_measure(struct tcp_sock *tp)
471 goto new_measure; 469 goto new_measure;
472 if (before(tp->rcv_nxt, tp->rcv_rtt_est.seq)) 470 if (before(tp->rcv_nxt, tp->rcv_rtt_est.seq))
473 return; 471 return;
474 tcp_rcv_rtt_update(tp, 472 tcp_rcv_rtt_update(tp, jiffies - tp->rcv_rtt_est.time, 1);
475 jiffies - tp->rcv_rtt_est.time,
476 1);
477 473
478new_measure: 474new_measure:
479 tp->rcv_rtt_est.seq = tp->rcv_nxt + tp->rcv_wnd; 475 tp->rcv_rtt_est.seq = tp->rcv_nxt + tp->rcv_wnd;
480 tp->rcv_rtt_est.time = tcp_time_stamp; 476 tp->rcv_rtt_est.time = tcp_time_stamp;
481} 477}
482 478
483static inline void tcp_rcv_rtt_measure_ts(struct sock *sk, const struct sk_buff *skb) 479static inline void tcp_rcv_rtt_measure_ts(struct sock *sk,
480 const struct sk_buff *skb)
484{ 481{
485 struct tcp_sock *tp = tcp_sk(sk); 482 struct tcp_sock *tp = tcp_sk(sk);
486 if (tp->rx_opt.rcv_tsecr && 483 if (tp->rx_opt.rcv_tsecr &&
@@ -503,8 +500,7 @@ void tcp_rcv_space_adjust(struct sock *sk)
503 goto new_measure; 500 goto new_measure;
504 501
505 time = tcp_time_stamp - tp->rcvq_space.time; 502 time = tcp_time_stamp - tp->rcvq_space.time;
506 if (time < (tp->rcv_rtt_est.rtt >> 3) || 503 if (time < (tp->rcv_rtt_est.rtt >> 3) || tp->rcv_rtt_est.rtt == 0)
507 tp->rcv_rtt_est.rtt == 0)
508 return; 504 return;
509 505
510 space = 2 * (tp->copied_seq - tp->rcvq_space.seq); 506 space = 2 * (tp->copied_seq - tp->rcvq_space.seq);
@@ -580,7 +576,7 @@ static void tcp_event_data_recv(struct sock *sk, struct sk_buff *skb)
580 } else { 576 } else {
581 int m = now - icsk->icsk_ack.lrcvtime; 577 int m = now - icsk->icsk_ack.lrcvtime;
582 578
583 if (m <= TCP_ATO_MIN/2) { 579 if (m <= TCP_ATO_MIN / 2) {
584 /* The fastest case is the first. */ 580 /* The fastest case is the first. */
585 icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + TCP_ATO_MIN / 2; 581 icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + TCP_ATO_MIN / 2;
586 } else if (m < icsk->icsk_ack.ato) { 582 } else if (m < icsk->icsk_ack.ato) {
@@ -609,7 +605,7 @@ static u32 tcp_rto_min(struct sock *sk)
609 u32 rto_min = TCP_RTO_MIN; 605 u32 rto_min = TCP_RTO_MIN;
610 606
611 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 607 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
612 rto_min = dst->metrics[RTAX_RTO_MIN-1]; 608 rto_min = dst->metrics[RTAX_RTO_MIN - 1];
613 return rto_min; 609 return rto_min;
614} 610}
615 611
@@ -672,14 +668,14 @@ static void tcp_rtt_estimator(struct sock *sk, const __u32 mrtt)
672 } 668 }
673 if (after(tp->snd_una, tp->rtt_seq)) { 669 if (after(tp->snd_una, tp->rtt_seq)) {
674 if (tp->mdev_max < tp->rttvar) 670 if (tp->mdev_max < tp->rttvar)
675 tp->rttvar -= (tp->rttvar-tp->mdev_max)>>2; 671 tp->rttvar -= (tp->rttvar - tp->mdev_max) >> 2;
676 tp->rtt_seq = tp->snd_nxt; 672 tp->rtt_seq = tp->snd_nxt;
677 tp->mdev_max = tcp_rto_min(sk); 673 tp->mdev_max = tcp_rto_min(sk);
678 } 674 }
679 } else { 675 } else {
680 /* no previous measure. */ 676 /* no previous measure. */
681 tp->srtt = m<<3; /* take the measured time to be rtt */ 677 tp->srtt = m << 3; /* take the measured time to be rtt */
682 tp->mdev = m<<1; /* make sure rto = 3*rtt */ 678 tp->mdev = m << 1; /* make sure rto = 3*rtt */
683 tp->mdev_max = tp->rttvar = max(tp->mdev, tcp_rto_min(sk)); 679 tp->mdev_max = tp->rttvar = max(tp->mdev, tcp_rto_min(sk));
684 tp->rtt_seq = tp->snd_nxt; 680 tp->rtt_seq = tp->snd_nxt;
685 } 681 }
@@ -733,7 +729,7 @@ void tcp_update_metrics(struct sock *sk)
733 729
734 dst_confirm(dst); 730 dst_confirm(dst);
735 731
736 if (dst && (dst->flags&DST_HOST)) { 732 if (dst && (dst->flags & DST_HOST)) {
737 const struct inet_connection_sock *icsk = inet_csk(sk); 733 const struct inet_connection_sock *icsk = inet_csk(sk);
738 int m; 734 int m;
739 735
@@ -743,7 +739,7 @@ void tcp_update_metrics(struct sock *sk)
743 * Reset our results. 739 * Reset our results.
744 */ 740 */
745 if (!(dst_metric_locked(dst, RTAX_RTT))) 741 if (!(dst_metric_locked(dst, RTAX_RTT)))
746 dst->metrics[RTAX_RTT-1] = 0; 742 dst->metrics[RTAX_RTT - 1] = 0;
747 return; 743 return;
748 } 744 }
749 745
@@ -755,9 +751,9 @@ void tcp_update_metrics(struct sock *sk)
755 */ 751 */
756 if (!(dst_metric_locked(dst, RTAX_RTT))) { 752 if (!(dst_metric_locked(dst, RTAX_RTT))) {
757 if (m <= 0) 753 if (m <= 0)
758 dst->metrics[RTAX_RTT-1] = tp->srtt; 754 dst->metrics[RTAX_RTT - 1] = tp->srtt;
759 else 755 else
760 dst->metrics[RTAX_RTT-1] -= (m>>3); 756 dst->metrics[RTAX_RTT - 1] -= (m >> 3);
761 } 757 }
762 758
763 if (!(dst_metric_locked(dst, RTAX_RTTVAR))) { 759 if (!(dst_metric_locked(dst, RTAX_RTTVAR))) {
@@ -770,7 +766,7 @@ void tcp_update_metrics(struct sock *sk)
770 m = tp->mdev; 766 m = tp->mdev;
771 767
772 if (m >= dst_metric(dst, RTAX_RTTVAR)) 768 if (m >= dst_metric(dst, RTAX_RTTVAR))
773 dst->metrics[RTAX_RTTVAR-1] = m; 769 dst->metrics[RTAX_RTTVAR - 1] = m;
774 else 770 else
775 dst->metrics[RTAX_RTTVAR-1] -= 771 dst->metrics[RTAX_RTTVAR-1] -=
776 (dst->metrics[RTAX_RTTVAR-1] - m)>>2; 772 (dst->metrics[RTAX_RTTVAR-1] - m)>>2;
@@ -784,7 +780,7 @@ void tcp_update_metrics(struct sock *sk)
784 dst->metrics[RTAX_SSTHRESH-1] = tp->snd_cwnd >> 1; 780 dst->metrics[RTAX_SSTHRESH-1] = tp->snd_cwnd >> 1;
785 if (!dst_metric_locked(dst, RTAX_CWND) && 781 if (!dst_metric_locked(dst, RTAX_CWND) &&
786 tp->snd_cwnd > dst_metric(dst, RTAX_CWND)) 782 tp->snd_cwnd > dst_metric(dst, RTAX_CWND))
787 dst->metrics[RTAX_CWND-1] = tp->snd_cwnd; 783 dst->metrics[RTAX_CWND - 1] = tp->snd_cwnd;
788 } else if (tp->snd_cwnd > tp->snd_ssthresh && 784 } else if (tp->snd_cwnd > tp->snd_ssthresh &&
789 icsk->icsk_ca_state == TCP_CA_Open) { 785 icsk->icsk_ca_state == TCP_CA_Open) {
790 /* Cong. avoidance phase, cwnd is reliable. */ 786 /* Cong. avoidance phase, cwnd is reliable. */
@@ -1353,12 +1349,14 @@ static struct sk_buff *tcp_sacktag_walk(struct sk_buff *skb, struct sock *sk,
1353 } 1349 }
1354 1350
1355 if (in_sack <= 0) 1351 if (in_sack <= 0)
1356 in_sack = tcp_match_skb_to_sack(sk, skb, start_seq, end_seq); 1352 in_sack = tcp_match_skb_to_sack(sk, skb, start_seq,
1353 end_seq);
1357 if (unlikely(in_sack < 0)) 1354 if (unlikely(in_sack < 0))
1358 break; 1355 break;
1359 1356
1360 if (in_sack) 1357 if (in_sack)
1361 *flag |= tcp_sacktag_one(skb, sk, reord, dup_sack, *fack_count); 1358 *flag |= tcp_sacktag_one(skb, sk, reord, dup_sack,
1359 *fack_count);
1362 1360
1363 *fack_count += tcp_skb_pcount(skb); 1361 *fack_count += tcp_skb_pcount(skb);
1364 } 1362 }
@@ -1407,7 +1405,8 @@ static int tcp_sack_cache_ok(struct tcp_sock *tp, struct tcp_sack_block *cache)
1407} 1405}
1408 1406
1409static int 1407static int
1410tcp_sacktag_write_queue(struct sock *sk, struct sk_buff *ack_skb, u32 prior_snd_una) 1408tcp_sacktag_write_queue(struct sock *sk, struct sk_buff *ack_skb,
1409 u32 prior_snd_una)
1411{ 1410{
1412 const struct inet_connection_sock *icsk = inet_csk(sk); 1411 const struct inet_connection_sock *icsk = inet_csk(sk);
1413 struct tcp_sock *tp = tcp_sk(sk); 1412 struct tcp_sock *tp = tcp_sk(sk);
@@ -1417,7 +1416,7 @@ tcp_sacktag_write_queue(struct sock *sk, struct sk_buff *ack_skb, u32 prior_snd_
1417 struct tcp_sack_block sp[4]; 1416 struct tcp_sack_block sp[4];
1418 struct tcp_sack_block *cache; 1417 struct tcp_sack_block *cache;
1419 struct sk_buff *skb; 1418 struct sk_buff *skb;
1420 int num_sacks = (ptr[1] - TCPOLEN_SACK_BASE)>>3; 1419 int num_sacks = (ptr[1] - TCPOLEN_SACK_BASE) >> 3;
1421 int used_sacks; 1420 int used_sacks;
1422 int reord = tp->packets_out; 1421 int reord = tp->packets_out;
1423 int flag = 0; 1422 int flag = 0;
@@ -1484,17 +1483,17 @@ tcp_sacktag_write_queue(struct sock *sk, struct sk_buff *ack_skb, u32 prior_snd_
1484 1483
1485 /* order SACK blocks to allow in order walk of the retrans queue */ 1484 /* order SACK blocks to allow in order walk of the retrans queue */
1486 for (i = used_sacks - 1; i > 0; i--) { 1485 for (i = used_sacks - 1; i > 0; i--) {
1487 for (j = 0; j < i; j++){ 1486 for (j = 0; j < i; j++) {
1488 if (after(sp[j].start_seq, sp[j+1].start_seq)) { 1487 if (after(sp[j].start_seq, sp[j + 1].start_seq)) {
1489 struct tcp_sack_block tmp; 1488 struct tcp_sack_block tmp;
1490 1489
1491 tmp = sp[j]; 1490 tmp = sp[j];
1492 sp[j] = sp[j+1]; 1491 sp[j] = sp[j + 1];
1493 sp[j+1] = tmp; 1492 sp[j + 1] = tmp;
1494 1493
1495 /* Track where the first SACK block goes to */ 1494 /* Track where the first SACK block goes to */
1496 if (j == first_sack_index) 1495 if (j == first_sack_index)
1497 first_sack_index = j+1; 1496 first_sack_index = j + 1;
1498 } 1497 }
1499 } 1498 }
1500 } 1499 }
@@ -1539,17 +1538,21 @@ tcp_sacktag_write_queue(struct sock *sk, struct sk_buff *ack_skb, u32 prior_snd_
1539 /* Head todo? */ 1538 /* Head todo? */
1540 if (before(start_seq, cache->start_seq)) { 1539 if (before(start_seq, cache->start_seq)) {
1541 skb = tcp_sacktag_skip(skb, sk, start_seq); 1540 skb = tcp_sacktag_skip(skb, sk, start_seq);
1542 skb = tcp_sacktag_walk(skb, sk, next_dup, start_seq, 1541 skb = tcp_sacktag_walk(skb, sk, next_dup,
1543 cache->start_seq, dup_sack, 1542 start_seq,
1544 &fack_count, &reord, &flag); 1543 cache->start_seq,
1544 dup_sack, &fack_count,
1545 &reord, &flag);
1545 } 1546 }
1546 1547
1547 /* Rest of the block already fully processed? */ 1548 /* Rest of the block already fully processed? */
1548 if (!after(end_seq, cache->end_seq)) 1549 if (!after(end_seq, cache->end_seq))
1549 goto advance_sp; 1550 goto advance_sp;
1550 1551
1551 skb = tcp_maybe_skipping_dsack(skb, sk, next_dup, cache->end_seq, 1552 skb = tcp_maybe_skipping_dsack(skb, sk, next_dup,
1552 &fack_count, &reord, &flag); 1553 cache->end_seq,
1554 &fack_count, &reord,
1555 &flag);
1553 1556
1554 /* ...tail remains todo... */ 1557 /* ...tail remains todo... */
1555 if (tcp_highest_sack_seq(tp) == cache->end_seq) { 1558 if (tcp_highest_sack_seq(tp) == cache->end_seq) {
@@ -1654,10 +1657,10 @@ static void tcp_remove_reno_sacks(struct sock *sk, int acked)
1654 1657
1655 if (acked > 0) { 1658 if (acked > 0) {
1656 /* One ACK acked hole. The rest eat duplicate ACKs. */ 1659 /* One ACK acked hole. The rest eat duplicate ACKs. */
1657 if (acked-1 >= tp->sacked_out) 1660 if (acked - 1 >= tp->sacked_out)
1658 tp->sacked_out = 0; 1661 tp->sacked_out = 0;
1659 else 1662 else
1660 tp->sacked_out -= acked-1; 1663 tp->sacked_out -= acked - 1;
1661 } 1664 }
1662 tcp_check_reno_reordering(sk, acked); 1665 tcp_check_reno_reordering(sk, acked);
1663 tcp_verify_left_out(tp); 1666 tcp_verify_left_out(tp);
@@ -1691,10 +1694,10 @@ int tcp_use_frto(struct sock *sk)
1691 tcp_for_write_queue_from(skb, sk) { 1694 tcp_for_write_queue_from(skb, sk) {
1692 if (skb == tcp_send_head(sk)) 1695 if (skb == tcp_send_head(sk))
1693 break; 1696 break;
1694 if (TCP_SKB_CB(skb)->sacked&TCPCB_RETRANS) 1697 if (TCP_SKB_CB(skb)->sacked & TCPCB_RETRANS)
1695 return 0; 1698 return 0;
1696 /* Short-circuit when first non-SACKed skb has been checked */ 1699 /* Short-circuit when first non-SACKed skb has been checked */
1697 if (!(TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_ACKED)) 1700 if (!(TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED))
1698 break; 1701 break;
1699 } 1702 }
1700 return 1; 1703 return 1;
@@ -1804,7 +1807,7 @@ static void tcp_enter_frto_loss(struct sock *sk, int allowed_segments, int flag)
1804 * Count the retransmission made on RTO correctly (only when 1807 * Count the retransmission made on RTO correctly (only when
1805 * waiting for the first ACK and did not get it)... 1808 * waiting for the first ACK and did not get it)...
1806 */ 1809 */
1807 if ((tp->frto_counter == 1) && !(flag&FLAG_DATA_ACKED)) { 1810 if ((tp->frto_counter == 1) && !(flag & FLAG_DATA_ACKED)) {
1808 /* For some reason this R-bit might get cleared? */ 1811 /* For some reason this R-bit might get cleared? */
1809 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) 1812 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS)
1810 tp->retrans_out += tcp_skb_pcount(skb); 1813 tp->retrans_out += tcp_skb_pcount(skb);
@@ -1817,7 +1820,7 @@ static void tcp_enter_frto_loss(struct sock *sk, int allowed_segments, int flag)
1817 } 1820 }
1818 1821
1819 /* Don't lost mark skbs that were fwd transmitted after RTO */ 1822 /* Don't lost mark skbs that were fwd transmitted after RTO */
1820 if (!(TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_ACKED) && 1823 if (!(TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) &&
1821 !after(TCP_SKB_CB(skb)->end_seq, tp->frto_highmark)) { 1824 !after(TCP_SKB_CB(skb)->end_seq, tp->frto_highmark)) {
1822 TCP_SKB_CB(skb)->sacked |= TCPCB_LOST; 1825 TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;
1823 tp->lost_out += tcp_skb_pcount(skb); 1826 tp->lost_out += tcp_skb_pcount(skb);
@@ -1832,7 +1835,7 @@ static void tcp_enter_frto_loss(struct sock *sk, int allowed_segments, int flag)
1832 tp->bytes_acked = 0; 1835 tp->bytes_acked = 0;
1833 1836
1834 tp->reordering = min_t(unsigned int, tp->reordering, 1837 tp->reordering = min_t(unsigned int, tp->reordering,
1835 sysctl_tcp_reordering); 1838 sysctl_tcp_reordering);
1836 tcp_set_ca_state(sk, TCP_CA_Loss); 1839 tcp_set_ca_state(sk, TCP_CA_Loss);
1837 tp->high_seq = tp->frto_highmark; 1840 tp->high_seq = tp->frto_highmark;
1838 TCP_ECN_queue_cwr(tp); 1841 TCP_ECN_queue_cwr(tp);
@@ -1899,7 +1902,7 @@ void tcp_enter_loss(struct sock *sk, int how)
1899 if (skb == tcp_send_head(sk)) 1902 if (skb == tcp_send_head(sk))
1900 break; 1903 break;
1901 1904
1902 if (TCP_SKB_CB(skb)->sacked&TCPCB_RETRANS) 1905 if (TCP_SKB_CB(skb)->sacked & TCPCB_RETRANS)
1903 tp->undo_marker = 0; 1906 tp->undo_marker = 0;
1904 TCP_SKB_CB(skb)->sacked &= (~TCPCB_TAGBITS)|TCPCB_SACKED_ACKED; 1907 TCP_SKB_CB(skb)->sacked &= (~TCPCB_TAGBITS)|TCPCB_SACKED_ACKED;
1905 if (!(TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_ACKED) || how) { 1908 if (!(TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_ACKED) || how) {
@@ -1911,7 +1914,7 @@ void tcp_enter_loss(struct sock *sk, int how)
1911 tcp_verify_left_out(tp); 1914 tcp_verify_left_out(tp);
1912 1915
1913 tp->reordering = min_t(unsigned int, tp->reordering, 1916 tp->reordering = min_t(unsigned int, tp->reordering,
1914 sysctl_tcp_reordering); 1917 sysctl_tcp_reordering);
1915 tcp_set_ca_state(sk, TCP_CA_Loss); 1918 tcp_set_ca_state(sk, TCP_CA_Loss);
1916 tp->high_seq = tp->snd_nxt; 1919 tp->high_seq = tp->snd_nxt;
1917 TCP_ECN_queue_cwr(tp); 1920 TCP_ECN_queue_cwr(tp);
@@ -1943,7 +1946,7 @@ static int tcp_check_sack_reneging(struct sock *sk, int flag)
1943 1946
1944static inline int tcp_fackets_out(struct tcp_sock *tp) 1947static inline int tcp_fackets_out(struct tcp_sock *tp)
1945{ 1948{
1946 return tcp_is_reno(tp) ? tp->sacked_out+1 : tp->fackets_out; 1949 return tcp_is_reno(tp) ? tp->sacked_out + 1 : tp->fackets_out;
1947} 1950}
1948 1951
1949/* Heurestics to calculate number of duplicate ACKs. There's no dupACKs 1952/* Heurestics to calculate number of duplicate ACKs. There's no dupACKs
@@ -2116,12 +2119,11 @@ static int tcp_time_to_recover(struct sock *sk)
2116 * retransmitted past LOST markings in the first place? I'm not fully sure 2119 * retransmitted past LOST markings in the first place? I'm not fully sure
2117 * about undo and end of connection cases, which can cause R without L? 2120 * about undo and end of connection cases, which can cause R without L?
2118 */ 2121 */
2119static void tcp_verify_retransmit_hint(struct tcp_sock *tp, 2122static void tcp_verify_retransmit_hint(struct tcp_sock *tp, struct sk_buff *skb)
2120 struct sk_buff *skb)
2121{ 2123{
2122 if ((tp->retransmit_skb_hint != NULL) && 2124 if ((tp->retransmit_skb_hint != NULL) &&
2123 before(TCP_SKB_CB(skb)->seq, 2125 before(TCP_SKB_CB(skb)->seq,
2124 TCP_SKB_CB(tp->retransmit_skb_hint)->seq)) 2126 TCP_SKB_CB(tp->retransmit_skb_hint)->seq))
2125 tp->retransmit_skb_hint = NULL; 2127 tp->retransmit_skb_hint = NULL;
2126} 2128}
2127 2129
@@ -2156,7 +2158,7 @@ static void tcp_mark_head_lost(struct sock *sk, int packets, int fast_rexmit)
2156 cnt += tcp_skb_pcount(skb); 2158 cnt += tcp_skb_pcount(skb);
2157 2159
2158 if (((!fast_rexmit || (tp->lost_out > 0)) && (cnt > packets)) || 2160 if (((!fast_rexmit || (tp->lost_out > 0)) && (cnt > packets)) ||
2159 after(TCP_SKB_CB(skb)->end_seq, tp->high_seq)) 2161 after(TCP_SKB_CB(skb)->end_seq, tp->high_seq))
2160 break; 2162 break;
2161 if (!(TCP_SKB_CB(skb)->sacked & (TCPCB_SACKED_ACKED|TCPCB_LOST))) { 2163 if (!(TCP_SKB_CB(skb)->sacked & (TCPCB_SACKED_ACKED|TCPCB_LOST))) {
2162 TCP_SKB_CB(skb)->sacked |= TCPCB_LOST; 2164 TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;
@@ -2223,7 +2225,7 @@ static void tcp_update_scoreboard(struct sock *sk, int fast_rexmit)
2223static inline void tcp_moderate_cwnd(struct tcp_sock *tp) 2225static inline void tcp_moderate_cwnd(struct tcp_sock *tp)
2224{ 2226{
2225 tp->snd_cwnd = min(tp->snd_cwnd, 2227 tp->snd_cwnd = min(tp->snd_cwnd,
2226 tcp_packets_in_flight(tp)+tcp_max_burst(tp)); 2228 tcp_packets_in_flight(tp) + tcp_max_burst(tp));
2227 tp->snd_cwnd_stamp = tcp_time_stamp; 2229 tp->snd_cwnd_stamp = tcp_time_stamp;
2228} 2230}
2229 2231
@@ -2243,15 +2245,15 @@ static void tcp_cwnd_down(struct sock *sk, int flag)
2243 struct tcp_sock *tp = tcp_sk(sk); 2245 struct tcp_sock *tp = tcp_sk(sk);
2244 int decr = tp->snd_cwnd_cnt + 1; 2246 int decr = tp->snd_cwnd_cnt + 1;
2245 2247
2246 if ((flag&(FLAG_ANY_PROGRESS|FLAG_DSACKING_ACK)) || 2248 if ((flag & (FLAG_ANY_PROGRESS | FLAG_DSACKING_ACK)) ||
2247 (tcp_is_reno(tp) && !(flag&FLAG_NOT_DUP))) { 2249 (tcp_is_reno(tp) && !(flag & FLAG_NOT_DUP))) {
2248 tp->snd_cwnd_cnt = decr&1; 2250 tp->snd_cwnd_cnt = decr & 1;
2249 decr >>= 1; 2251 decr >>= 1;
2250 2252
2251 if (decr && tp->snd_cwnd > tcp_cwnd_min(sk)) 2253 if (decr && tp->snd_cwnd > tcp_cwnd_min(sk))
2252 tp->snd_cwnd -= decr; 2254 tp->snd_cwnd -= decr;
2253 2255
2254 tp->snd_cwnd = min(tp->snd_cwnd, tcp_packets_in_flight(tp)+1); 2256 tp->snd_cwnd = min(tp->snd_cwnd, tcp_packets_in_flight(tp) + 1);
2255 tp->snd_cwnd_stamp = tcp_time_stamp; 2257 tp->snd_cwnd_stamp = tcp_time_stamp;
2256 } 2258 }
2257} 2259}
@@ -2295,7 +2297,7 @@ static void tcp_undo_cwr(struct sock *sk, const int undo)
2295 if (icsk->icsk_ca_ops->undo_cwnd) 2297 if (icsk->icsk_ca_ops->undo_cwnd)
2296 tp->snd_cwnd = icsk->icsk_ca_ops->undo_cwnd(sk); 2298 tp->snd_cwnd = icsk->icsk_ca_ops->undo_cwnd(sk);
2297 else 2299 else
2298 tp->snd_cwnd = max(tp->snd_cwnd, tp->snd_ssthresh<<1); 2300 tp->snd_cwnd = max(tp->snd_cwnd, tp->snd_ssthresh << 1);
2299 2301
2300 if (undo && tp->prior_ssthresh > tp->snd_ssthresh) { 2302 if (undo && tp->prior_ssthresh > tp->snd_ssthresh) {
2301 tp->snd_ssthresh = tp->prior_ssthresh; 2303 tp->snd_ssthresh = tp->prior_ssthresh;
@@ -2314,8 +2316,7 @@ static void tcp_undo_cwr(struct sock *sk, const int undo)
2314 2316
2315static inline int tcp_may_undo(struct tcp_sock *tp) 2317static inline int tcp_may_undo(struct tcp_sock *tp)
2316{ 2318{
2317 return tp->undo_marker && 2319 return tp->undo_marker && (!tp->undo_retrans || tcp_packet_delayed(tp));
2318 (!tp->undo_retrans || tcp_packet_delayed(tp));
2319} 2320}
2320 2321
2321/* People celebrate: "We love our President!" */ 2322/* People celebrate: "We love our President!" */
@@ -2434,7 +2435,7 @@ static void tcp_try_to_open(struct sock *sk, int flag)
2434 if (tp->retrans_out == 0) 2435 if (tp->retrans_out == 0)
2435 tp->retrans_stamp = 0; 2436 tp->retrans_stamp = 0;
2436 2437
2437 if (flag&FLAG_ECE) 2438 if (flag & FLAG_ECE)
2438 tcp_enter_cwr(sk, 1); 2439 tcp_enter_cwr(sk, 1);
2439 2440
2440 if (inet_csk(sk)->icsk_ca_state != TCP_CA_CWR) { 2441 if (inet_csk(sk)->icsk_ca_state != TCP_CA_CWR) {
@@ -2480,7 +2481,6 @@ static void tcp_mtup_probe_success(struct sock *sk, struct sk_buff *skb)
2480 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); 2481 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
2481} 2482}
2482 2483
2483
2484/* Process an event, which can update packets-in-flight not trivially. 2484/* Process an event, which can update packets-in-flight not trivially.
2485 * Main goal of this function is to calculate new estimate for left_out, 2485 * Main goal of this function is to calculate new estimate for left_out,
2486 * taking into account both packets sitting in receiver's buffer and 2486 * taking into account both packets sitting in receiver's buffer and
@@ -2492,13 +2492,12 @@ static void tcp_mtup_probe_success(struct sock *sk, struct sk_buff *skb)
2492 * It does _not_ decide what to send, it is made in function 2492 * It does _not_ decide what to send, it is made in function
2493 * tcp_xmit_retransmit_queue(). 2493 * tcp_xmit_retransmit_queue().
2494 */ 2494 */
2495static void 2495static void tcp_fastretrans_alert(struct sock *sk, int pkts_acked, int flag)
2496tcp_fastretrans_alert(struct sock *sk, int pkts_acked, int flag)
2497{ 2496{
2498 struct inet_connection_sock *icsk = inet_csk(sk); 2497 struct inet_connection_sock *icsk = inet_csk(sk);
2499 struct tcp_sock *tp = tcp_sk(sk); 2498 struct tcp_sock *tp = tcp_sk(sk);
2500 int is_dupack = !(flag&(FLAG_SND_UNA_ADVANCED|FLAG_NOT_DUP)); 2499 int is_dupack = !(flag & (FLAG_SND_UNA_ADVANCED | FLAG_NOT_DUP));
2501 int do_lost = is_dupack || ((flag&FLAG_DATA_SACKED) && 2500 int do_lost = is_dupack || ((flag & FLAG_DATA_SACKED) &&
2502 (tcp_fackets_out(tp) > tp->reordering)); 2501 (tcp_fackets_out(tp) > tp->reordering));
2503 int fast_rexmit = 0; 2502 int fast_rexmit = 0;
2504 2503
@@ -2509,7 +2508,7 @@ tcp_fastretrans_alert(struct sock *sk, int pkts_acked, int flag)
2509 2508
2510 /* Now state machine starts. 2509 /* Now state machine starts.
2511 * A. ECE, hence prohibit cwnd undoing, the reduction is required. */ 2510 * A. ECE, hence prohibit cwnd undoing, the reduction is required. */
2512 if (flag&FLAG_ECE) 2511 if (flag & FLAG_ECE)
2513 tp->prior_ssthresh = 0; 2512 tp->prior_ssthresh = 0;
2514 2513
2515 /* B. In all the states check for reneging SACKs. */ 2514 /* B. In all the states check for reneging SACKs. */
@@ -2521,7 +2520,7 @@ tcp_fastretrans_alert(struct sock *sk, int pkts_acked, int flag)
2521 before(tp->snd_una, tp->high_seq) && 2520 before(tp->snd_una, tp->high_seq) &&
2522 icsk->icsk_ca_state != TCP_CA_Open && 2521 icsk->icsk_ca_state != TCP_CA_Open &&
2523 tp->fackets_out > tp->reordering) { 2522 tp->fackets_out > tp->reordering) {
2524 tcp_mark_head_lost(sk, tp->fackets_out-tp->reordering, 0); 2523 tcp_mark_head_lost(sk, tp->fackets_out - tp->reordering, 0);
2525 NET_INC_STATS_BH(LINUX_MIB_TCPLOSS); 2524 NET_INC_STATS_BH(LINUX_MIB_TCPLOSS);
2526 } 2525 }
2527 2526
@@ -2581,7 +2580,7 @@ tcp_fastretrans_alert(struct sock *sk, int pkts_acked, int flag)
2581 do_lost = tcp_try_undo_partial(sk, pkts_acked); 2580 do_lost = tcp_try_undo_partial(sk, pkts_acked);
2582 break; 2581 break;
2583 case TCP_CA_Loss: 2582 case TCP_CA_Loss:
2584 if (flag&FLAG_DATA_ACKED) 2583 if (flag & FLAG_DATA_ACKED)
2585 icsk->icsk_retransmits = 0; 2584 icsk->icsk_retransmits = 0;
2586 if (!tcp_try_undo_loss(sk)) { 2585 if (!tcp_try_undo_loss(sk)) {
2587 tcp_moderate_cwnd(tp); 2586 tcp_moderate_cwnd(tp);
@@ -2631,7 +2630,7 @@ tcp_fastretrans_alert(struct sock *sk, int pkts_acked, int flag)
2631 tp->undo_retrans = tp->retrans_out; 2630 tp->undo_retrans = tp->retrans_out;
2632 2631
2633 if (icsk->icsk_ca_state < TCP_CA_CWR) { 2632 if (icsk->icsk_ca_state < TCP_CA_CWR) {
2634 if (!(flag&FLAG_ECE)) 2633 if (!(flag & FLAG_ECE))
2635 tp->prior_ssthresh = tcp_current_ssthresh(sk); 2634 tp->prior_ssthresh = tcp_current_ssthresh(sk);
2636 tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk); 2635 tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk);
2637 TCP_ECN_queue_cwr(tp); 2636 TCP_ECN_queue_cwr(tp);
@@ -2725,7 +2724,8 @@ static void tcp_rearm_rto(struct sock *sk)
2725 if (!tp->packets_out) { 2724 if (!tp->packets_out) {
2726 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS); 2725 inet_csk_clear_xmit_timer(sk, ICSK_TIME_RETRANS);
2727 } else { 2726 } else {
2728 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, inet_csk(sk)->icsk_rto, TCP_RTO_MAX); 2727 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
2728 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
2729 } 2729 }
2730} 2730}
2731 2731
@@ -2803,8 +2803,7 @@ static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets)
2803 flag |= FLAG_RETRANS_DATA_ACKED; 2803 flag |= FLAG_RETRANS_DATA_ACKED;
2804 ca_seq_rtt = -1; 2804 ca_seq_rtt = -1;
2805 seq_rtt = -1; 2805 seq_rtt = -1;
2806 if ((flag & FLAG_DATA_ACKED) || 2806 if ((flag & FLAG_DATA_ACKED) || (acked_pcount > 1))
2807 (acked_pcount > 1))
2808 flag |= FLAG_NONHEAD_RETRANS_ACKED; 2807 flag |= FLAG_NONHEAD_RETRANS_ACKED;
2809 } else { 2808 } else {
2810 ca_seq_rtt = now - scb->when; 2809 ca_seq_rtt = now - scb->when;
@@ -2950,8 +2949,9 @@ static inline int tcp_may_raise_cwnd(const struct sock *sk, const int flag)
2950/* Check that window update is acceptable. 2949/* Check that window update is acceptable.
2951 * The function assumes that snd_una<=ack<=snd_next. 2950 * The function assumes that snd_una<=ack<=snd_next.
2952 */ 2951 */
2953static inline int tcp_may_update_window(const struct tcp_sock *tp, const u32 ack, 2952static inline int tcp_may_update_window(const struct tcp_sock *tp,
2954 const u32 ack_seq, const u32 nwin) 2953 const u32 ack, const u32 ack_seq,
2954 const u32 nwin)
2955{ 2955{
2956 return (after(ack, tp->snd_una) || 2956 return (after(ack, tp->snd_una) ||
2957 after(ack_seq, tp->snd_wl1) || 2957 after(ack_seq, tp->snd_wl1) ||
@@ -3020,7 +3020,7 @@ static void tcp_ratehalving_spur_to_response(struct sock *sk)
3020 3020
3021static void tcp_undo_spur_to_response(struct sock *sk, int flag) 3021static void tcp_undo_spur_to_response(struct sock *sk, int flag)
3022{ 3022{
3023 if (flag&FLAG_ECE) 3023 if (flag & FLAG_ECE)
3024 tcp_ratehalving_spur_to_response(sk); 3024 tcp_ratehalving_spur_to_response(sk);
3025 else 3025 else
3026 tcp_undo_cwr(sk, 1); 3026 tcp_undo_cwr(sk, 1);
@@ -3063,7 +3063,7 @@ static int tcp_process_frto(struct sock *sk, int flag)
3063 tcp_verify_left_out(tp); 3063 tcp_verify_left_out(tp);
3064 3064
3065 /* Duplicate the behavior from Loss state (fastretrans_alert) */ 3065 /* Duplicate the behavior from Loss state (fastretrans_alert) */
3066 if (flag&FLAG_DATA_ACKED) 3066 if (flag & FLAG_DATA_ACKED)
3067 inet_csk(sk)->icsk_retransmits = 0; 3067 inet_csk(sk)->icsk_retransmits = 0;
3068 3068
3069 if ((flag & FLAG_NONHEAD_RETRANS_ACKED) || 3069 if ((flag & FLAG_NONHEAD_RETRANS_ACKED) ||
@@ -3080,16 +3080,16 @@ static int tcp_process_frto(struct sock *sk, int flag)
3080 * ACK isn't duplicate nor advances window, e.g., opposite dir 3080 * ACK isn't duplicate nor advances window, e.g., opposite dir
3081 * data, winupdate 3081 * data, winupdate
3082 */ 3082 */
3083 if (!(flag&FLAG_ANY_PROGRESS) && (flag&FLAG_NOT_DUP)) 3083 if (!(flag & FLAG_ANY_PROGRESS) && (flag & FLAG_NOT_DUP))
3084 return 1; 3084 return 1;
3085 3085
3086 if (!(flag&FLAG_DATA_ACKED)) { 3086 if (!(flag & FLAG_DATA_ACKED)) {
3087 tcp_enter_frto_loss(sk, (tp->frto_counter == 1 ? 0 : 3), 3087 tcp_enter_frto_loss(sk, (tp->frto_counter == 1 ? 0 : 3),
3088 flag); 3088 flag);
3089 return 1; 3089 return 1;
3090 } 3090 }
3091 } else { 3091 } else {
3092 if (!(flag&FLAG_DATA_ACKED) && (tp->frto_counter == 1)) { 3092 if (!(flag & FLAG_DATA_ACKED) && (tp->frto_counter == 1)) {
3093 /* Prevent sending of new data. */ 3093 /* Prevent sending of new data. */
3094 tp->snd_cwnd = min(tp->snd_cwnd, 3094 tp->snd_cwnd = min(tp->snd_cwnd,
3095 tcp_packets_in_flight(tp)); 3095 tcp_packets_in_flight(tp));
@@ -3097,10 +3097,12 @@ static int tcp_process_frto(struct sock *sk, int flag)
3097 } 3097 }
3098 3098
3099 if ((tp->frto_counter >= 2) && 3099 if ((tp->frto_counter >= 2) &&
3100 (!(flag&FLAG_FORWARD_PROGRESS) || 3100 (!(flag & FLAG_FORWARD_PROGRESS) ||
3101 ((flag&FLAG_DATA_SACKED) && !(flag&FLAG_ONLY_ORIG_SACKED)))) { 3101 ((flag & FLAG_DATA_SACKED) &&
3102 !(flag & FLAG_ONLY_ORIG_SACKED)))) {
3102 /* RFC4138 shortcoming (see comment above) */ 3103 /* RFC4138 shortcoming (see comment above) */
3103 if (!(flag&FLAG_FORWARD_PROGRESS) && (flag&FLAG_NOT_DUP)) 3104 if (!(flag & FLAG_FORWARD_PROGRESS) &&
3105 (flag & FLAG_NOT_DUP))
3104 return 1; 3106 return 1;
3105 3107
3106 tcp_enter_frto_loss(sk, 3, flag); 3108 tcp_enter_frto_loss(sk, 3, flag);
@@ -3166,13 +3168,14 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag)
3166 tp->bytes_acked += ack - prior_snd_una; 3168 tp->bytes_acked += ack - prior_snd_una;
3167 else if (icsk->icsk_ca_state == TCP_CA_Loss) 3169 else if (icsk->icsk_ca_state == TCP_CA_Loss)
3168 /* we assume just one segment left network */ 3170 /* we assume just one segment left network */
3169 tp->bytes_acked += min(ack - prior_snd_una, tp->mss_cache); 3171 tp->bytes_acked += min(ack - prior_snd_una,
3172 tp->mss_cache);
3170 } 3173 }
3171 3174
3172 prior_fackets = tp->fackets_out; 3175 prior_fackets = tp->fackets_out;
3173 prior_in_flight = tcp_packets_in_flight(tp); 3176 prior_in_flight = tcp_packets_in_flight(tp);
3174 3177
3175 if (!(flag&FLAG_SLOWPATH) && after(ack, prior_snd_una)) { 3178 if (!(flag & FLAG_SLOWPATH) && after(ack, prior_snd_una)) {
3176 /* Window is constant, pure forward advance. 3179 /* Window is constant, pure forward advance.
3177 * No more checks are required. 3180 * No more checks are required.
3178 * Note, we use the fact that SND.UNA>=SND.WL2. 3181 * Note, we use the fact that SND.UNA>=SND.WL2.
@@ -3224,13 +3227,14 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag)
3224 if ((flag & FLAG_DATA_ACKED) && !frto_cwnd && 3227 if ((flag & FLAG_DATA_ACKED) && !frto_cwnd &&
3225 tcp_may_raise_cwnd(sk, flag)) 3228 tcp_may_raise_cwnd(sk, flag))
3226 tcp_cong_avoid(sk, ack, prior_in_flight); 3229 tcp_cong_avoid(sk, ack, prior_in_flight);
3227 tcp_fastretrans_alert(sk, prior_packets - tp->packets_out, flag); 3230 tcp_fastretrans_alert(sk, prior_packets - tp->packets_out,
3231 flag);
3228 } else { 3232 } else {
3229 if ((flag & FLAG_DATA_ACKED) && !frto_cwnd) 3233 if ((flag & FLAG_DATA_ACKED) && !frto_cwnd)
3230 tcp_cong_avoid(sk, ack, prior_in_flight); 3234 tcp_cong_avoid(sk, ack, prior_in_flight);
3231 } 3235 }
3232 3236
3233 if ((flag & FLAG_FORWARD_PROGRESS) || !(flag&FLAG_NOT_DUP)) 3237 if ((flag & FLAG_FORWARD_PROGRESS) || !(flag & FLAG_NOT_DUP))
3234 dst_confirm(sk->sk_dst_cache); 3238 dst_confirm(sk->sk_dst_cache);
3235 3239
3236 return 1; 3240 return 1;
@@ -3255,22 +3259,22 @@ uninteresting_ack:
3255 return 0; 3259 return 0;
3256} 3260}
3257 3261
3258
3259/* Look for tcp options. Normally only called on SYN and SYNACK packets. 3262/* Look for tcp options. Normally only called on SYN and SYNACK packets.
3260 * But, this can also be called on packets in the established flow when 3263 * But, this can also be called on packets in the established flow when
3261 * the fast version below fails. 3264 * the fast version below fails.
3262 */ 3265 */
3263void tcp_parse_options(struct sk_buff *skb, struct tcp_options_received *opt_rx, int estab) 3266void tcp_parse_options(struct sk_buff *skb, struct tcp_options_received *opt_rx,
3267 int estab)
3264{ 3268{
3265 unsigned char *ptr; 3269 unsigned char *ptr;
3266 struct tcphdr *th = tcp_hdr(skb); 3270 struct tcphdr *th = tcp_hdr(skb);
3267 int length=(th->doff*4)-sizeof(struct tcphdr); 3271 int length = (th->doff * 4) - sizeof(struct tcphdr);
3268 3272
3269 ptr = (unsigned char *)(th + 1); 3273 ptr = (unsigned char *)(th + 1);
3270 opt_rx->saw_tstamp = 0; 3274 opt_rx->saw_tstamp = 0;
3271 3275
3272 while (length > 0) { 3276 while (length > 0) {
3273 int opcode=*ptr++; 3277 int opcode = *ptr++;
3274 int opsize; 3278 int opsize;
3275 3279
3276 switch (opcode) { 3280 switch (opcode) {
@@ -3359,7 +3363,7 @@ void tcp_parse_options(struct sk_buff *skb, struct tcp_options_received *opt_rx,
3359static int tcp_fast_parse_options(struct sk_buff *skb, struct tcphdr *th, 3363static int tcp_fast_parse_options(struct sk_buff *skb, struct tcphdr *th,
3360 struct tcp_sock *tp) 3364 struct tcp_sock *tp)
3361{ 3365{
3362 if (th->doff == sizeof(struct tcphdr)>>2) { 3366 if (th->doff == sizeof(struct tcphdr) >> 2) {
3363 tp->rx_opt.saw_tstamp = 0; 3367 tp->rx_opt.saw_tstamp = 0;
3364 return 0; 3368 return 0;
3365 } else if (tp->rx_opt.tstamp_ok && 3369 } else if (tp->rx_opt.tstamp_ok &&
@@ -3444,7 +3448,8 @@ static int tcp_disordered_ack(const struct sock *sk, const struct sk_buff *skb)
3444 (s32)(tp->rx_opt.ts_recent - tp->rx_opt.rcv_tsval) <= (inet_csk(sk)->icsk_rto * 1024) / HZ); 3448 (s32)(tp->rx_opt.ts_recent - tp->rx_opt.rcv_tsval) <= (inet_csk(sk)->icsk_rto * 1024) / HZ);
3445} 3449}
3446 3450
3447static inline int tcp_paws_discard(const struct sock *sk, const struct sk_buff *skb) 3451static inline int tcp_paws_discard(const struct sock *sk,
3452 const struct sk_buff *skb)
3448{ 3453{
3449 const struct tcp_sock *tp = tcp_sk(sk); 3454 const struct tcp_sock *tp = tcp_sk(sk);
3450 return ((s32)(tp->rx_opt.ts_recent - tp->rx_opt.rcv_tsval) > TCP_PAWS_WINDOW && 3455 return ((s32)(tp->rx_opt.ts_recent - tp->rx_opt.rcv_tsval) > TCP_PAWS_WINDOW &&
@@ -3476,16 +3481,16 @@ static void tcp_reset(struct sock *sk)
3476{ 3481{
3477 /* We want the right error as BSD sees it (and indeed as we do). */ 3482 /* We want the right error as BSD sees it (and indeed as we do). */
3478 switch (sk->sk_state) { 3483 switch (sk->sk_state) {
3479 case TCP_SYN_SENT: 3484 case TCP_SYN_SENT:
3480 sk->sk_err = ECONNREFUSED; 3485 sk->sk_err = ECONNREFUSED;
3481 break; 3486 break;
3482 case TCP_CLOSE_WAIT: 3487 case TCP_CLOSE_WAIT:
3483 sk->sk_err = EPIPE; 3488 sk->sk_err = EPIPE;
3484 break; 3489 break;
3485 case TCP_CLOSE: 3490 case TCP_CLOSE:
3486 return; 3491 return;
3487 default: 3492 default:
3488 sk->sk_err = ECONNRESET; 3493 sk->sk_err = ECONNRESET;
3489 } 3494 }
3490 3495
3491 if (!sock_flag(sk, SOCK_DEAD)) 3496 if (!sock_flag(sk, SOCK_DEAD))
@@ -3518,43 +3523,43 @@ static void tcp_fin(struct sk_buff *skb, struct sock *sk, struct tcphdr *th)
3518 sock_set_flag(sk, SOCK_DONE); 3523 sock_set_flag(sk, SOCK_DONE);
3519 3524
3520 switch (sk->sk_state) { 3525 switch (sk->sk_state) {
3521 case TCP_SYN_RECV: 3526 case TCP_SYN_RECV:
3522 case TCP_ESTABLISHED: 3527 case TCP_ESTABLISHED:
3523 /* Move to CLOSE_WAIT */ 3528 /* Move to CLOSE_WAIT */
3524 tcp_set_state(sk, TCP_CLOSE_WAIT); 3529 tcp_set_state(sk, TCP_CLOSE_WAIT);
3525 inet_csk(sk)->icsk_ack.pingpong = 1; 3530 inet_csk(sk)->icsk_ack.pingpong = 1;
3526 break; 3531 break;
3527 3532
3528 case TCP_CLOSE_WAIT: 3533 case TCP_CLOSE_WAIT:
3529 case TCP_CLOSING: 3534 case TCP_CLOSING:
3530 /* Received a retransmission of the FIN, do 3535 /* Received a retransmission of the FIN, do
3531 * nothing. 3536 * nothing.
3532 */ 3537 */
3533 break; 3538 break;
3534 case TCP_LAST_ACK: 3539 case TCP_LAST_ACK:
3535 /* RFC793: Remain in the LAST-ACK state. */ 3540 /* RFC793: Remain in the LAST-ACK state. */
3536 break; 3541 break;
3537 3542
3538 case TCP_FIN_WAIT1: 3543 case TCP_FIN_WAIT1:
3539 /* This case occurs when a simultaneous close 3544 /* This case occurs when a simultaneous close
3540 * happens, we must ack the received FIN and 3545 * happens, we must ack the received FIN and
3541 * enter the CLOSING state. 3546 * enter the CLOSING state.
3542 */ 3547 */
3543 tcp_send_ack(sk); 3548 tcp_send_ack(sk);
3544 tcp_set_state(sk, TCP_CLOSING); 3549 tcp_set_state(sk, TCP_CLOSING);
3545 break; 3550 break;
3546 case TCP_FIN_WAIT2: 3551 case TCP_FIN_WAIT2:
3547 /* Received a FIN -- send ACK and enter TIME_WAIT. */ 3552 /* Received a FIN -- send ACK and enter TIME_WAIT. */
3548 tcp_send_ack(sk); 3553 tcp_send_ack(sk);
3549 tcp_time_wait(sk, TCP_TIME_WAIT, 0); 3554 tcp_time_wait(sk, TCP_TIME_WAIT, 0);
3550 break; 3555 break;
3551 default: 3556 default:
3552 /* Only TCP_LISTEN and TCP_CLOSE are left, in these 3557 /* Only TCP_LISTEN and TCP_CLOSE are left, in these
3553 * cases we should never reach this piece of code. 3558 * cases we should never reach this piece of code.
3554 */ 3559 */
3555 printk(KERN_ERR "%s: Impossible, sk->sk_state=%d\n", 3560 printk(KERN_ERR "%s: Impossible, sk->sk_state=%d\n",
3556 __FUNCTION__, sk->sk_state); 3561 __FUNCTION__, sk->sk_state);
3557 break; 3562 break;
3558 } 3563 }
3559 3564
3560 /* It _is_ possible, that we have something out-of-order _after_ FIN. 3565 /* It _is_ possible, that we have something out-of-order _after_ FIN.
@@ -3577,7 +3582,8 @@ static void tcp_fin(struct sk_buff *skb, struct sock *sk, struct tcphdr *th)
3577 } 3582 }
3578} 3583}
3579 3584
3580static inline int tcp_sack_extend(struct tcp_sack_block *sp, u32 seq, u32 end_seq) 3585static inline int tcp_sack_extend(struct tcp_sack_block *sp, u32 seq,
3586 u32 end_seq)
3581{ 3587{
3582 if (!after(seq, sp->end_seq) && !after(sp->start_seq, end_seq)) { 3588 if (!after(seq, sp->end_seq) && !after(sp->start_seq, end_seq)) {
3583 if (before(seq, sp->start_seq)) 3589 if (before(seq, sp->start_seq))
@@ -3600,7 +3606,8 @@ static void tcp_dsack_set(struct tcp_sock *tp, u32 seq, u32 end_seq)
3600 tp->rx_opt.dsack = 1; 3606 tp->rx_opt.dsack = 1;
3601 tp->duplicate_sack[0].start_seq = seq; 3607 tp->duplicate_sack[0].start_seq = seq;
3602 tp->duplicate_sack[0].end_seq = end_seq; 3608 tp->duplicate_sack[0].end_seq = end_seq;
3603 tp->rx_opt.eff_sacks = min(tp->rx_opt.num_sacks + 1, 4 - tp->rx_opt.tstamp_ok); 3609 tp->rx_opt.eff_sacks = min(tp->rx_opt.num_sacks + 1,
3610 4 - tp->rx_opt.tstamp_ok);
3604 } 3611 }
3605} 3612}
3606 3613
@@ -3640,12 +3647,12 @@ static void tcp_sack_maybe_coalesce(struct tcp_sock *tp)
3640{ 3647{
3641 int this_sack; 3648 int this_sack;
3642 struct tcp_sack_block *sp = &tp->selective_acks[0]; 3649 struct tcp_sack_block *sp = &tp->selective_acks[0];
3643 struct tcp_sack_block *swalk = sp+1; 3650 struct tcp_sack_block *swalk = sp + 1;
3644 3651
3645 /* See if the recent change to the first SACK eats into 3652 /* See if the recent change to the first SACK eats into
3646 * or hits the sequence space of other SACK blocks, if so coalesce. 3653 * or hits the sequence space of other SACK blocks, if so coalesce.
3647 */ 3654 */
3648 for (this_sack = 1; this_sack < tp->rx_opt.num_sacks; ) { 3655 for (this_sack = 1; this_sack < tp->rx_opt.num_sacks;) {
3649 if (tcp_sack_extend(sp, swalk->start_seq, swalk->end_seq)) { 3656 if (tcp_sack_extend(sp, swalk->start_seq, swalk->end_seq)) {
3650 int i; 3657 int i;
3651 3658
@@ -3653,16 +3660,19 @@ static void tcp_sack_maybe_coalesce(struct tcp_sock *tp)
3653 * Decrease num_sacks. 3660 * Decrease num_sacks.
3654 */ 3661 */
3655 tp->rx_opt.num_sacks--; 3662 tp->rx_opt.num_sacks--;
3656 tp->rx_opt.eff_sacks = min(tp->rx_opt.num_sacks + tp->rx_opt.dsack, 4 - tp->rx_opt.tstamp_ok); 3663 tp->rx_opt.eff_sacks = min(tp->rx_opt.num_sacks +
3657 for (i=this_sack; i < tp->rx_opt.num_sacks; i++) 3664 tp->rx_opt.dsack,
3658 sp[i] = sp[i+1]; 3665 4 - tp->rx_opt.tstamp_ok);
3666 for (i = this_sack; i < tp->rx_opt.num_sacks; i++)
3667 sp[i] = sp[i + 1];
3659 continue; 3668 continue;
3660 } 3669 }
3661 this_sack++, swalk++; 3670 this_sack++, swalk++;
3662 } 3671 }
3663} 3672}
3664 3673
3665static inline void tcp_sack_swap(struct tcp_sack_block *sack1, struct tcp_sack_block *sack2) 3674static inline void tcp_sack_swap(struct tcp_sack_block *sack1,
3675 struct tcp_sack_block *sack2)
3666{ 3676{
3667 __u32 tmp; 3677 __u32 tmp;
3668 3678
@@ -3685,11 +3695,11 @@ static void tcp_sack_new_ofo_skb(struct sock *sk, u32 seq, u32 end_seq)
3685 if (!cur_sacks) 3695 if (!cur_sacks)
3686 goto new_sack; 3696 goto new_sack;
3687 3697
3688 for (this_sack=0; this_sack<cur_sacks; this_sack++, sp++) { 3698 for (this_sack = 0; this_sack < cur_sacks; this_sack++, sp++) {
3689 if (tcp_sack_extend(sp, seq, end_seq)) { 3699 if (tcp_sack_extend(sp, seq, end_seq)) {
3690 /* Rotate this_sack to the first one. */ 3700 /* Rotate this_sack to the first one. */
3691 for (; this_sack>0; this_sack--, sp--) 3701 for (; this_sack > 0; this_sack--, sp--)
3692 tcp_sack_swap(sp, sp-1); 3702 tcp_sack_swap(sp, sp - 1);
3693 if (cur_sacks > 1) 3703 if (cur_sacks > 1)
3694 tcp_sack_maybe_coalesce(tp); 3704 tcp_sack_maybe_coalesce(tp);
3695 return; 3705 return;
@@ -3708,14 +3718,15 @@ static void tcp_sack_new_ofo_skb(struct sock *sk, u32 seq, u32 end_seq)
3708 sp--; 3718 sp--;
3709 } 3719 }
3710 for (; this_sack > 0; this_sack--, sp--) 3720 for (; this_sack > 0; this_sack--, sp--)
3711 *sp = *(sp-1); 3721 *sp = *(sp - 1);
3712 3722
3713new_sack: 3723new_sack:
3714 /* Build the new head SACK, and we're done. */ 3724 /* Build the new head SACK, and we're done. */
3715 sp->start_seq = seq; 3725 sp->start_seq = seq;
3716 sp->end_seq = end_seq; 3726 sp->end_seq = end_seq;
3717 tp->rx_opt.num_sacks++; 3727 tp->rx_opt.num_sacks++;
3718 tp->rx_opt.eff_sacks = min(tp->rx_opt.num_sacks + tp->rx_opt.dsack, 4 - tp->rx_opt.tstamp_ok); 3728 tp->rx_opt.eff_sacks = min(tp->rx_opt.num_sacks + tp->rx_opt.dsack,
3729 4 - tp->rx_opt.tstamp_ok);
3719} 3730}
3720 3731
3721/* RCV.NXT advances, some SACKs should be eaten. */ 3732/* RCV.NXT advances, some SACKs should be eaten. */
@@ -3733,7 +3744,7 @@ static void tcp_sack_remove(struct tcp_sock *tp)
3733 return; 3744 return;
3734 } 3745 }
3735 3746
3736 for (this_sack = 0; this_sack < num_sacks; ) { 3747 for (this_sack = 0; this_sack < num_sacks;) {
3737 /* Check if the start of the sack is covered by RCV.NXT. */ 3748 /* Check if the start of the sack is covered by RCV.NXT. */
3738 if (!before(tp->rcv_nxt, sp->start_seq)) { 3749 if (!before(tp->rcv_nxt, sp->start_seq)) {
3739 int i; 3750 int i;
@@ -3752,7 +3763,9 @@ static void tcp_sack_remove(struct tcp_sock *tp)
3752 } 3763 }
3753 if (num_sacks != tp->rx_opt.num_sacks) { 3764 if (num_sacks != tp->rx_opt.num_sacks) {
3754 tp->rx_opt.num_sacks = num_sacks; 3765 tp->rx_opt.num_sacks = num_sacks;
3755 tp->rx_opt.eff_sacks = min(tp->rx_opt.num_sacks + tp->rx_opt.dsack, 4 - tp->rx_opt.tstamp_ok); 3766 tp->rx_opt.eff_sacks = min(tp->rx_opt.num_sacks +
3767 tp->rx_opt.dsack,
3768 4 - tp->rx_opt.tstamp_ok);
3756 } 3769 }
3757} 3770}
3758 3771
@@ -3805,14 +3818,14 @@ static void tcp_data_queue(struct sock *sk, struct sk_buff *skb)
3805 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) 3818 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq)
3806 goto drop; 3819 goto drop;
3807 3820
3808 __skb_pull(skb, th->doff*4); 3821 __skb_pull(skb, th->doff * 4);
3809 3822
3810 TCP_ECN_accept_cwr(tp, skb); 3823 TCP_ECN_accept_cwr(tp, skb);
3811 3824
3812 if (tp->rx_opt.dsack) { 3825 if (tp->rx_opt.dsack) {
3813 tp->rx_opt.dsack = 0; 3826 tp->rx_opt.dsack = 0;
3814 tp->rx_opt.eff_sacks = min_t(unsigned int, tp->rx_opt.num_sacks, 3827 tp->rx_opt.eff_sacks = min_t(unsigned int, tp->rx_opt.num_sacks,
3815 4 - tp->rx_opt.tstamp_ok); 3828 4 - tp->rx_opt.tstamp_ok);
3816 } 3829 }
3817 3830
3818 /* Queue data for delivery to the user. 3831 /* Queue data for delivery to the user.
@@ -3828,7 +3841,7 @@ static void tcp_data_queue(struct sock *sk, struct sk_buff *skb)
3828 tp->copied_seq == tp->rcv_nxt && tp->ucopy.len && 3841 tp->copied_seq == tp->rcv_nxt && tp->ucopy.len &&
3829 sock_owned_by_user(sk) && !tp->urg_data) { 3842 sock_owned_by_user(sk) && !tp->urg_data) {
3830 int chunk = min_t(unsigned int, skb->len, 3843 int chunk = min_t(unsigned int, skb->len,
3831 tp->ucopy.len); 3844 tp->ucopy.len);
3832 3845
3833 __set_current_state(TASK_RUNNING); 3846 __set_current_state(TASK_RUNNING);
3834 3847
@@ -3945,7 +3958,7 @@ drop:
3945 tp->selective_acks[0].end_seq = 3958 tp->selective_acks[0].end_seq =
3946 TCP_SKB_CB(skb)->end_seq; 3959 TCP_SKB_CB(skb)->end_seq;
3947 } 3960 }
3948 __skb_queue_head(&tp->out_of_order_queue,skb); 3961 __skb_queue_head(&tp->out_of_order_queue, skb);
3949 } else { 3962 } else {
3950 struct sk_buff *skb1 = tp->out_of_order_queue.prev; 3963 struct sk_buff *skb1 = tp->out_of_order_queue.prev;
3951 u32 seq = TCP_SKB_CB(skb)->seq; 3964 u32 seq = TCP_SKB_CB(skb)->seq;
@@ -3968,10 +3981,10 @@ drop:
3968 if (!after(TCP_SKB_CB(skb1)->seq, seq)) 3981 if (!after(TCP_SKB_CB(skb1)->seq, seq))
3969 break; 3982 break;
3970 } while ((skb1 = skb1->prev) != 3983 } while ((skb1 = skb1->prev) !=
3971 (struct sk_buff*)&tp->out_of_order_queue); 3984 (struct sk_buff *)&tp->out_of_order_queue);
3972 3985
3973 /* Do skb overlap to previous one? */ 3986 /* Do skb overlap to previous one? */
3974 if (skb1 != (struct sk_buff*)&tp->out_of_order_queue && 3987 if (skb1 != (struct sk_buff *)&tp->out_of_order_queue &&
3975 before(seq, TCP_SKB_CB(skb1)->end_seq)) { 3988 before(seq, TCP_SKB_CB(skb1)->end_seq)) {
3976 if (!after(end_seq, TCP_SKB_CB(skb1)->end_seq)) { 3989 if (!after(end_seq, TCP_SKB_CB(skb1)->end_seq)) {
3977 /* All the bits are present. Drop. */ 3990 /* All the bits are present. Drop. */
@@ -3981,7 +3994,8 @@ drop:
3981 } 3994 }
3982 if (after(seq, TCP_SKB_CB(skb1)->seq)) { 3995 if (after(seq, TCP_SKB_CB(skb1)->seq)) {
3983 /* Partial overlap. */ 3996 /* Partial overlap. */
3984 tcp_dsack_set(tp, seq, TCP_SKB_CB(skb1)->end_seq); 3997 tcp_dsack_set(tp, seq,
3998 TCP_SKB_CB(skb1)->end_seq);
3985 } else { 3999 } else {
3986 skb1 = skb1->prev; 4000 skb1 = skb1->prev;
3987 } 4001 }
@@ -3990,15 +4004,17 @@ drop:
3990 4004
3991 /* And clean segments covered by new one as whole. */ 4005 /* And clean segments covered by new one as whole. */
3992 while ((skb1 = skb->next) != 4006 while ((skb1 = skb->next) !=
3993 (struct sk_buff*)&tp->out_of_order_queue && 4007 (struct sk_buff *)&tp->out_of_order_queue &&
3994 after(end_seq, TCP_SKB_CB(skb1)->seq)) { 4008 after(end_seq, TCP_SKB_CB(skb1)->seq)) {
3995 if (before(end_seq, TCP_SKB_CB(skb1)->end_seq)) { 4009 if (before(end_seq, TCP_SKB_CB(skb1)->end_seq)) {
3996 tcp_dsack_extend(tp, TCP_SKB_CB(skb1)->seq, end_seq); 4010 tcp_dsack_extend(tp, TCP_SKB_CB(skb1)->seq,
3997 break; 4011 end_seq);
3998 } 4012 break;
3999 __skb_unlink(skb1, &tp->out_of_order_queue); 4013 }
4000 tcp_dsack_extend(tp, TCP_SKB_CB(skb1)->seq, TCP_SKB_CB(skb1)->end_seq); 4014 __skb_unlink(skb1, &tp->out_of_order_queue);
4001 __kfree_skb(skb1); 4015 tcp_dsack_extend(tp, TCP_SKB_CB(skb1)->seq,
4016 TCP_SKB_CB(skb1)->end_seq);
4017 __kfree_skb(skb1);
4002 } 4018 }
4003 4019
4004add_sack: 4020add_sack:
@@ -4021,7 +4037,7 @@ tcp_collapse(struct sock *sk, struct sk_buff_head *list,
4021 4037
4022 /* First, check that queue is collapsible and find 4038 /* First, check that queue is collapsible and find
4023 * the point where collapsing can be useful. */ 4039 * the point where collapsing can be useful. */
4024 for (skb = head; skb != tail; ) { 4040 for (skb = head; skb != tail;) {
4025 /* No new bits? It is possible on ofo queue. */ 4041 /* No new bits? It is possible on ofo queue. */
4026 if (!before(start, TCP_SKB_CB(skb)->end_seq)) { 4042 if (!before(start, TCP_SKB_CB(skb)->end_seq)) {
4027 struct sk_buff *next = skb->next; 4043 struct sk_buff *next = skb->next;
@@ -4059,9 +4075,9 @@ tcp_collapse(struct sock *sk, struct sk_buff_head *list,
4059 /* Too big header? This can happen with IPv6. */ 4075 /* Too big header? This can happen with IPv6. */
4060 if (copy < 0) 4076 if (copy < 0)
4061 return; 4077 return;
4062 if (end-start < copy) 4078 if (end - start < copy)
4063 copy = end-start; 4079 copy = end - start;
4064 nskb = alloc_skb(copy+header, GFP_ATOMIC); 4080 nskb = alloc_skb(copy + header, GFP_ATOMIC);
4065 if (!nskb) 4081 if (!nskb)
4066 return; 4082 return;
4067 4083
@@ -4171,7 +4187,7 @@ static int tcp_prune_queue(struct sock *sk)
4171 tcp_collapse_ofo_queue(sk); 4187 tcp_collapse_ofo_queue(sk);
4172 tcp_collapse(sk, &sk->sk_receive_queue, 4188 tcp_collapse(sk, &sk->sk_receive_queue,
4173 sk->sk_receive_queue.next, 4189 sk->sk_receive_queue.next,
4174 (struct sk_buff*)&sk->sk_receive_queue, 4190 (struct sk_buff *)&sk->sk_receive_queue,
4175 tp->copied_seq, tp->rcv_nxt); 4191 tp->copied_seq, tp->rcv_nxt);
4176 sk_mem_reclaim(sk); 4192 sk_mem_reclaim(sk);
4177 4193
@@ -4210,7 +4226,6 @@ static int tcp_prune_queue(struct sock *sk)
4210 return -1; 4226 return -1;
4211} 4227}
4212 4228
4213
4214/* RFC2861, slow part. Adjust cwnd, after it was not full during one rto. 4229/* RFC2861, slow part. Adjust cwnd, after it was not full during one rto.
4215 * As additional protections, we do not touch cwnd in retransmission phases, 4230 * As additional protections, we do not touch cwnd in retransmission phases,
4216 * and if application hit its sndbuf limit recently. 4231 * and if application hit its sndbuf limit recently.
@@ -4272,8 +4287,8 @@ static void tcp_new_space(struct sock *sk)
4272 int sndmem = max_t(u32, tp->rx_opt.mss_clamp, tp->mss_cache) + 4287 int sndmem = max_t(u32, tp->rx_opt.mss_clamp, tp->mss_cache) +
4273 MAX_TCP_HEADER + 16 + sizeof(struct sk_buff), 4288 MAX_TCP_HEADER + 16 + sizeof(struct sk_buff),
4274 demanded = max_t(unsigned int, tp->snd_cwnd, 4289 demanded = max_t(unsigned int, tp->snd_cwnd,
4275 tp->reordering + 1); 4290 tp->reordering + 1);
4276 sndmem *= 2*demanded; 4291 sndmem *= 2 * demanded;
4277 if (sndmem > sk->sk_sndbuf) 4292 if (sndmem > sk->sk_sndbuf)
4278 sk->sk_sndbuf = min(sndmem, sysctl_tcp_wmem[2]); 4293 sk->sk_sndbuf = min(sndmem, sysctl_tcp_wmem[2]);
4279 tp->snd_cwnd_stamp = tcp_time_stamp; 4294 tp->snd_cwnd_stamp = tcp_time_stamp;
@@ -4314,8 +4329,7 @@ static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible)
4314 /* We ACK each frame or... */ 4329 /* We ACK each frame or... */
4315 tcp_in_quickack_mode(sk) || 4330 tcp_in_quickack_mode(sk) ||
4316 /* We have out of order data. */ 4331 /* We have out of order data. */
4317 (ofo_possible && 4332 (ofo_possible && skb_peek(&tp->out_of_order_queue))) {
4318 skb_peek(&tp->out_of_order_queue))) {
4319 /* Then ack it now */ 4333 /* Then ack it now */
4320 tcp_send_ack(sk); 4334 tcp_send_ack(sk);
4321 } else { 4335 } else {
@@ -4343,7 +4357,7 @@ static inline void tcp_ack_snd_check(struct sock *sk)
4343 * either form (or just set the sysctl tcp_stdurg). 4357 * either form (or just set the sysctl tcp_stdurg).
4344 */ 4358 */
4345 4359
4346static void tcp_check_urg(struct sock * sk, struct tcphdr * th) 4360static void tcp_check_urg(struct sock *sk, struct tcphdr *th)
4347{ 4361{
4348 struct tcp_sock *tp = tcp_sk(sk); 4362 struct tcp_sock *tp = tcp_sk(sk);
4349 u32 ptr = ntohs(th->urg_ptr); 4363 u32 ptr = ntohs(th->urg_ptr);
@@ -4392,8 +4406,7 @@ static void tcp_check_urg(struct sock * sk, struct tcphdr * th)
4392 * buggy users. 4406 * buggy users.
4393 */ 4407 */
4394 if (tp->urg_seq == tp->copied_seq && tp->urg_data && 4408 if (tp->urg_seq == tp->copied_seq && tp->urg_data &&
4395 !sock_flag(sk, SOCK_URGINLINE) && 4409 !sock_flag(sk, SOCK_URGINLINE) && tp->copied_seq != tp->rcv_nxt) {
4396 tp->copied_seq != tp->rcv_nxt) {
4397 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); 4410 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
4398 tp->copied_seq++; 4411 tp->copied_seq++;
4399 if (skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq)) { 4412 if (skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq)) {
@@ -4402,8 +4415,8 @@ static void tcp_check_urg(struct sock * sk, struct tcphdr * th)
4402 } 4415 }
4403 } 4416 }
4404 4417
4405 tp->urg_data = TCP_URG_NOTYET; 4418 tp->urg_data = TCP_URG_NOTYET;
4406 tp->urg_seq = ptr; 4419 tp->urg_seq = ptr;
4407 4420
4408 /* Disable header prediction. */ 4421 /* Disable header prediction. */
4409 tp->pred_flags = 0; 4422 tp->pred_flags = 0;
@@ -4416,7 +4429,7 @@ static void tcp_urg(struct sock *sk, struct sk_buff *skb, struct tcphdr *th)
4416 4429
4417 /* Check if we get a new urgent pointer - normally not. */ 4430 /* Check if we get a new urgent pointer - normally not. */
4418 if (th->urg) 4431 if (th->urg)
4419 tcp_check_urg(sk,th); 4432 tcp_check_urg(sk, th);
4420 4433
4421 /* Do we wait for any urgent data? - normally not... */ 4434 /* Do we wait for any urgent data? - normally not... */
4422 if (tp->urg_data == TCP_URG_NOTYET) { 4435 if (tp->urg_data == TCP_URG_NOTYET) {
@@ -4458,7 +4471,8 @@ static int tcp_copy_to_iovec(struct sock *sk, struct sk_buff *skb, int hlen)
4458 return err; 4471 return err;
4459} 4472}
4460 4473
4461static __sum16 __tcp_checksum_complete_user(struct sock *sk, struct sk_buff *skb) 4474static __sum16 __tcp_checksum_complete_user(struct sock *sk,
4475 struct sk_buff *skb)
4462{ 4476{
4463 __sum16 result; 4477 __sum16 result;
4464 4478
@@ -4472,14 +4486,16 @@ static __sum16 __tcp_checksum_complete_user(struct sock *sk, struct sk_buff *skb
4472 return result; 4486 return result;
4473} 4487}
4474 4488
4475static inline int tcp_checksum_complete_user(struct sock *sk, struct sk_buff *skb) 4489static inline int tcp_checksum_complete_user(struct sock *sk,
4490 struct sk_buff *skb)
4476{ 4491{
4477 return !skb_csum_unnecessary(skb) && 4492 return !skb_csum_unnecessary(skb) &&
4478 __tcp_checksum_complete_user(sk, skb); 4493 __tcp_checksum_complete_user(sk, skb);
4479} 4494}
4480 4495
4481#ifdef CONFIG_NET_DMA 4496#ifdef CONFIG_NET_DMA
4482static int tcp_dma_try_early_copy(struct sock *sk, struct sk_buff *skb, int hlen) 4497static int tcp_dma_try_early_copy(struct sock *sk, struct sk_buff *skb,
4498 int hlen)
4483{ 4499{
4484 struct tcp_sock *tp = tcp_sk(sk); 4500 struct tcp_sock *tp = tcp_sk(sk);
4485 int chunk = skb->len - hlen; 4501 int chunk = skb->len - hlen;
@@ -4495,7 +4511,9 @@ static int tcp_dma_try_early_copy(struct sock *sk, struct sk_buff *skb, int hlen
4495 if (tp->ucopy.dma_chan && skb_csum_unnecessary(skb)) { 4511 if (tp->ucopy.dma_chan && skb_csum_unnecessary(skb)) {
4496 4512
4497 dma_cookie = dma_skb_copy_datagram_iovec(tp->ucopy.dma_chan, 4513 dma_cookie = dma_skb_copy_datagram_iovec(tp->ucopy.dma_chan,
4498 skb, hlen, tp->ucopy.iov, chunk, tp->ucopy.pinned_list); 4514 skb, hlen,
4515 tp->ucopy.iov, chunk,
4516 tp->ucopy.pinned_list);
4499 4517
4500 if (dma_cookie < 0) 4518 if (dma_cookie < 0)
4501 goto out; 4519 goto out;
@@ -4577,7 +4595,7 @@ int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
4577 */ 4595 */
4578 4596
4579 if ((tcp_flag_word(th) & TCP_HP_BITS) == tp->pred_flags && 4597 if ((tcp_flag_word(th) & TCP_HP_BITS) == tp->pred_flags &&
4580 TCP_SKB_CB(skb)->seq == tp->rcv_nxt) { 4598 TCP_SKB_CB(skb)->seq == tp->rcv_nxt) {
4581 int tcp_header_len = tp->tcp_header_len; 4599 int tcp_header_len = tp->tcp_header_len;
4582 4600
4583 /* Timestamp header prediction: tcp_header_len 4601 /* Timestamp header prediction: tcp_header_len
@@ -4646,7 +4664,8 @@ int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
4646 eaten = 1; 4664 eaten = 1;
4647 } 4665 }
4648#endif 4666#endif
4649 if (tp->ucopy.task == current && sock_owned_by_user(sk) && !copied_early) { 4667 if (tp->ucopy.task == current &&
4668 sock_owned_by_user(sk) && !copied_early) {
4650 __set_current_state(TASK_RUNNING); 4669 __set_current_state(TASK_RUNNING);
4651 4670
4652 if (!tcp_copy_to_iovec(sk, skb, tcp_header_len)) 4671 if (!tcp_copy_to_iovec(sk, skb, tcp_header_len))
@@ -4693,7 +4712,7 @@ int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
4693 NET_INC_STATS_BH(LINUX_MIB_TCPHPHITS); 4712 NET_INC_STATS_BH(LINUX_MIB_TCPHPHITS);
4694 4713
4695 /* Bulk data transfer: receiver */ 4714 /* Bulk data transfer: receiver */
4696 __skb_pull(skb,tcp_header_len); 4715 __skb_pull(skb, tcp_header_len);
4697 __skb_queue_tail(&sk->sk_receive_queue, skb); 4716 __skb_queue_tail(&sk->sk_receive_queue, skb);
4698 skb_set_owner_r(skb, sk); 4717 skb_set_owner_r(skb, sk);
4699 tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq; 4718 tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
@@ -4725,7 +4744,7 @@ no_ack:
4725 } 4744 }
4726 4745
4727slow_path: 4746slow_path:
4728 if (len < (th->doff<<2) || tcp_checksum_complete_user(sk, skb)) 4747 if (len < (th->doff << 2) || tcp_checksum_complete_user(sk, skb))
4729 goto csum_error; 4748 goto csum_error;
4730 4749
4731 /* 4750 /*
@@ -4975,7 +4994,8 @@ discard:
4975 } 4994 }
4976 4995
4977 /* PAWS check. */ 4996 /* PAWS check. */
4978 if (tp->rx_opt.ts_recent_stamp && tp->rx_opt.saw_tstamp && tcp_paws_check(&tp->rx_opt, 0)) 4997 if (tp->rx_opt.ts_recent_stamp && tp->rx_opt.saw_tstamp &&
4998 tcp_paws_check(&tp->rx_opt, 0))
4979 goto discard_and_undo; 4999 goto discard_and_undo;
4980 5000
4981 if (th->syn) { 5001 if (th->syn) {
@@ -5010,7 +5030,6 @@ discard:
5010 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); 5030 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
5011 tcp_initialize_rcv_mss(sk); 5031 tcp_initialize_rcv_mss(sk);
5012 5032
5013
5014 tcp_send_synack(sk); 5033 tcp_send_synack(sk);
5015#if 0 5034#if 0
5016 /* Note, we could accept data and URG from this segment. 5035 /* Note, we could accept data and URG from this segment.
@@ -5042,7 +5061,6 @@ reset_and_undo:
5042 return 1; 5061 return 1;
5043} 5062}
5044 5063
5045
5046/* 5064/*
5047 * This function implements the receiving procedure of RFC 793 for 5065 * This function implements the receiving procedure of RFC 793 for
5048 * all states except ESTABLISHED and TIME_WAIT. 5066 * all states except ESTABLISHED and TIME_WAIT.
@@ -5164,7 +5182,7 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
5164 */ 5182 */
5165 if (sk->sk_socket) 5183 if (sk->sk_socket)
5166 sk_wake_async(sk, 5184 sk_wake_async(sk,
5167 SOCK_WAKE_IO, POLL_OUT); 5185 SOCK_WAKE_IO, POLL_OUT);
5168 5186
5169 tp->snd_una = TCP_SKB_CB(skb)->ack_seq; 5187 tp->snd_una = TCP_SKB_CB(skb)->ack_seq;
5170 tp->snd_wnd = ntohs(th->window) << 5188 tp->snd_wnd = ntohs(th->window) <<
@@ -5176,8 +5194,8 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
5176 * and does not calculate rtt. 5194 * and does not calculate rtt.
5177 * Fix it at least with timestamps. 5195 * Fix it at least with timestamps.
5178 */ 5196 */
5179 if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr && 5197 if (tp->rx_opt.saw_tstamp &&
5180 !tp->srtt) 5198 tp->rx_opt.rcv_tsecr && !tp->srtt)
5181 tcp_ack_saw_tstamp(sk, 0); 5199 tcp_ack_saw_tstamp(sk, 0);
5182 5200
5183 if (tp->rx_opt.tstamp_ok) 5201 if (tp->rx_opt.tstamp_ok)
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 454cf84b6154..bb7e80a284e7 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -221,14 +221,14 @@ void tcp_select_initial_window(int __space, __u32 mss,
221 * following RFC2414. Senders, not following this RFC, 221 * following RFC2414. Senders, not following this RFC,
222 * will be satisfied with 2. 222 * will be satisfied with 2.
223 */ 223 */
224 if (mss > (1<<*rcv_wscale)) { 224 if (mss > (1 << *rcv_wscale)) {
225 int init_cwnd = 4; 225 int init_cwnd = 4;
226 if (mss > 1460*3) 226 if (mss > 1460 * 3)
227 init_cwnd = 2; 227 init_cwnd = 2;
228 else if (mss > 1460) 228 else if (mss > 1460)
229 init_cwnd = 3; 229 init_cwnd = 3;
230 if (*rcv_wnd > init_cwnd*mss) 230 if (*rcv_wnd > init_cwnd * mss)
231 *rcv_wnd = init_cwnd*mss; 231 *rcv_wnd = init_cwnd * mss;
232 } 232 }
233 233
234 /* Set the clamp no higher than max representable value */ 234 /* Set the clamp no higher than max representable value */
@@ -278,11 +278,10 @@ static u16 tcp_select_window(struct sock *sk)
278 return new_win; 278 return new_win;
279} 279}
280 280
281static inline void TCP_ECN_send_synack(struct tcp_sock *tp, 281static inline void TCP_ECN_send_synack(struct tcp_sock *tp, struct sk_buff *skb)
282 struct sk_buff *skb)
283{ 282{
284 TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_CWR; 283 TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_CWR;
285 if (!(tp->ecn_flags&TCP_ECN_OK)) 284 if (!(tp->ecn_flags & TCP_ECN_OK))
286 TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_ECE; 285 TCP_SKB_CB(skb)->flags &= ~TCPCB_FLAG_ECE;
287} 286}
288 287
@@ -292,7 +291,7 @@ static inline void TCP_ECN_send_syn(struct sock *sk, struct sk_buff *skb)
292 291
293 tp->ecn_flags = 0; 292 tp->ecn_flags = 0;
294 if (sysctl_tcp_ecn) { 293 if (sysctl_tcp_ecn) {
295 TCP_SKB_CB(skb)->flags |= TCPCB_FLAG_ECE|TCPCB_FLAG_CWR; 294 TCP_SKB_CB(skb)->flags |= TCPCB_FLAG_ECE | TCPCB_FLAG_CWR;
296 tp->ecn_flags = TCP_ECN_OK; 295 tp->ecn_flags = TCP_ECN_OK;
297 } 296 }
298} 297}
@@ -314,7 +313,7 @@ static inline void TCP_ECN_send(struct sock *sk, struct sk_buff *skb,
314 if (skb->len != tcp_header_len && 313 if (skb->len != tcp_header_len &&
315 !before(TCP_SKB_CB(skb)->seq, tp->snd_nxt)) { 314 !before(TCP_SKB_CB(skb)->seq, tp->snd_nxt)) {
316 INET_ECN_xmit(sk); 315 INET_ECN_xmit(sk);
317 if (tp->ecn_flags&TCP_ECN_QUEUE_CWR) { 316 if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) {
318 tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR; 317 tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR;
319 tcp_hdr(skb)->cwr = 1; 318 tcp_hdr(skb)->cwr = 1;
320 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; 319 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
@@ -431,7 +430,7 @@ static void tcp_syn_build_options(__be32 *ptr, int mss, int ts, int sack,
431 (TCPOPT_NOP << 16) | 430 (TCPOPT_NOP << 16) |
432 (TCPOPT_MD5SIG << 8) | 431 (TCPOPT_MD5SIG << 8) |
433 TCPOLEN_MD5SIG); 432 TCPOLEN_MD5SIG);
434 *md5_hash = (__u8 *) ptr; 433 *md5_hash = (__u8 *)ptr;
435 } 434 }
436#endif 435#endif
437} 436}
@@ -447,7 +446,8 @@ static void tcp_syn_build_options(__be32 *ptr, int mss, int ts, int sack,
447 * We are working here with either a clone of the original 446 * We are working here with either a clone of the original
448 * SKB, or a fresh unique copy made by the retransmit engine. 447 * SKB, or a fresh unique copy made by the retransmit engine.
449 */ 448 */
450static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it, gfp_t gfp_mask) 449static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
450 gfp_t gfp_mask)
451{ 451{
452 const struct inet_connection_sock *icsk = inet_csk(sk); 452 const struct inet_connection_sock *icsk = inet_csk(sk);
453 struct inet_sock *inet; 453 struct inet_sock *inet;
@@ -551,8 +551,8 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
551 th->urg_ptr = 0; 551 th->urg_ptr = 0;
552 552
553 if (unlikely(tp->urg_mode && 553 if (unlikely(tp->urg_mode &&
554 between(tp->snd_up, tcb->seq+1, tcb->seq+0xFFFF))) { 554 between(tp->snd_up, tcb->seq + 1, tcb->seq + 0xFFFF))) {
555 th->urg_ptr = htons(tp->snd_up-tcb->seq); 555 th->urg_ptr = htons(tp->snd_up - tcb->seq);
556 th->urg = 1; 556 th->urg = 1;
557 } 557 }
558 558
@@ -616,7 +616,6 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
616#undef SYSCTL_FLAG_SACK 616#undef SYSCTL_FLAG_SACK
617} 617}
618 618
619
620/* This routine just queue's the buffer 619/* This routine just queue's the buffer
621 * 620 *
622 * NOTE: probe0 timer is not checked, do not forget tcp_push_pending_frames, 621 * NOTE: probe0 timer is not checked, do not forget tcp_push_pending_frames,
@@ -634,7 +633,8 @@ static void tcp_queue_skb(struct sock *sk, struct sk_buff *skb)
634 sk_mem_charge(sk, skb->truesize); 633 sk_mem_charge(sk, skb->truesize);
635} 634}
636 635
637static void tcp_set_skb_tso_segs(struct sock *sk, struct sk_buff *skb, unsigned int mss_now) 636static void tcp_set_skb_tso_segs(struct sock *sk, struct sk_buff *skb,
637 unsigned int mss_now)
638{ 638{
639 if (skb->len <= mss_now || !sk_can_gso(sk)) { 639 if (skb->len <= mss_now || !sk_can_gso(sk)) {
640 /* Avoid the costly divide in the normal 640 /* Avoid the costly divide in the normal
@@ -670,7 +670,8 @@ static void tcp_adjust_fackets_out(struct sock *sk, struct sk_buff *skb,
670 * packet to the list. This won't be called frequently, I hope. 670 * packet to the list. This won't be called frequently, I hope.
671 * Remember, these are still headerless SKBs at this point. 671 * Remember, these are still headerless SKBs at this point.
672 */ 672 */
673int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len, unsigned int mss_now) 673int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len,
674 unsigned int mss_now)
674{ 675{
675 struct tcp_sock *tp = tcp_sk(sk); 676 struct tcp_sock *tp = tcp_sk(sk);
676 struct sk_buff *buff; 677 struct sk_buff *buff;
@@ -708,13 +709,14 @@ int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len, unsigned int mss
708 709
709 /* PSH and FIN should only be set in the second packet. */ 710 /* PSH and FIN should only be set in the second packet. */
710 flags = TCP_SKB_CB(skb)->flags; 711 flags = TCP_SKB_CB(skb)->flags;
711 TCP_SKB_CB(skb)->flags = flags & ~(TCPCB_FLAG_FIN|TCPCB_FLAG_PSH); 712 TCP_SKB_CB(skb)->flags = flags & ~(TCPCB_FLAG_FIN | TCPCB_FLAG_PSH);
712 TCP_SKB_CB(buff)->flags = flags; 713 TCP_SKB_CB(buff)->flags = flags;
713 TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked; 714 TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked;
714 715
715 if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) { 716 if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) {
716 /* Copy and checksum data tail into the new buffer. */ 717 /* Copy and checksum data tail into the new buffer. */
717 buff->csum = csum_partial_copy_nocheck(skb->data + len, skb_put(buff, nsize), 718 buff->csum = csum_partial_copy_nocheck(skb->data + len,
719 skb_put(buff, nsize),
718 nsize, 0); 720 nsize, 0);
719 721
720 skb_trim(skb, len); 722 skb_trim(skb, len);
@@ -781,7 +783,7 @@ static void __pskb_trim_head(struct sk_buff *skb, int len)
781 783
782 eat = len; 784 eat = len;
783 k = 0; 785 k = 0;
784 for (i=0; i<skb_shinfo(skb)->nr_frags; i++) { 786 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
785 if (skb_shinfo(skb)->frags[i].size <= eat) { 787 if (skb_shinfo(skb)->frags[i].size <= eat) {
786 put_page(skb_shinfo(skb)->frags[i].page); 788 put_page(skb_shinfo(skb)->frags[i].page);
787 eat -= skb_shinfo(skb)->frags[i].size; 789 eat -= skb_shinfo(skb)->frags[i].size;
@@ -804,8 +806,7 @@ static void __pskb_trim_head(struct sk_buff *skb, int len)
804 806
805int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len) 807int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len)
806{ 808{
807 if (skb_cloned(skb) && 809 if (skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
808 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
809 return -ENOMEM; 810 return -ENOMEM;
810 811
811 /* If len == headlen, we avoid __skb_pull to preserve alignment. */ 812 /* If len == headlen, we avoid __skb_pull to preserve alignment. */
@@ -909,7 +910,6 @@ void tcp_mtup_init(struct sock *sk)
909 NOTE2. inet_csk(sk)->icsk_pmtu_cookie and tp->mss_cache 910 NOTE2. inet_csk(sk)->icsk_pmtu_cookie and tp->mss_cache
910 are READ ONLY outside this function. --ANK (980731) 911 are READ ONLY outside this function. --ANK (980731)
911 */ 912 */
912
913unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu) 913unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu)
914{ 914{
915 struct tcp_sock *tp = tcp_sk(sk); 915 struct tcp_sock *tp = tcp_sk(sk);
@@ -922,8 +922,8 @@ unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu)
922 mss_now = tcp_mtu_to_mss(sk, pmtu); 922 mss_now = tcp_mtu_to_mss(sk, pmtu);
923 923
924 /* Bound mss with half of window */ 924 /* Bound mss with half of window */
925 if (tp->max_window && mss_now > (tp->max_window>>1)) 925 if (tp->max_window && mss_now > (tp->max_window >> 1))
926 mss_now = max((tp->max_window>>1), 68U - tp->tcp_header_len); 926 mss_now = max((tp->max_window >> 1), 68U - tp->tcp_header_len);
927 927
928 /* And store cached results */ 928 /* And store cached results */
929 icsk->icsk_pmtu_cookie = pmtu; 929 icsk->icsk_pmtu_cookie = pmtu;
@@ -977,8 +977,7 @@ unsigned int tcp_current_mss(struct sock *sk, int large_allowed)
977 inet_csk(sk)->icsk_ext_hdr_len - 977 inet_csk(sk)->icsk_ext_hdr_len -
978 tp->tcp_header_len); 978 tp->tcp_header_len);
979 979
980 if (tp->max_window && 980 if (tp->max_window && (xmit_size_goal > (tp->max_window >> 1)))
981 (xmit_size_goal > (tp->max_window >> 1)))
982 xmit_size_goal = max((tp->max_window >> 1), 981 xmit_size_goal = max((tp->max_window >> 1),
983 68U - tp->tcp_header_len); 982 68U - tp->tcp_header_len);
984 983
@@ -990,7 +989,6 @@ unsigned int tcp_current_mss(struct sock *sk, int large_allowed)
990} 989}
991 990
992/* Congestion window validation. (RFC2861) */ 991/* Congestion window validation. (RFC2861) */
993
994static void tcp_cwnd_validate(struct sock *sk) 992static void tcp_cwnd_validate(struct sock *sk)
995{ 993{
996 struct tcp_sock *tp = tcp_sk(sk); 994 struct tcp_sock *tp = tcp_sk(sk);
@@ -1017,8 +1015,7 @@ static void tcp_cwnd_validate(struct sock *sk)
1017 * per input skb which could be mostly avoided here (if desired). 1015 * per input skb which could be mostly avoided here (if desired).
1018 */ 1016 */
1019static unsigned int tcp_mss_split_point(struct sock *sk, struct sk_buff *skb, 1017static unsigned int tcp_mss_split_point(struct sock *sk, struct sk_buff *skb,
1020 unsigned int mss_now, 1018 unsigned int mss_now, unsigned int cwnd)
1021 unsigned int cwnd)
1022{ 1019{
1023 struct tcp_sock *tp = tcp_sk(sk); 1020 struct tcp_sock *tp = tcp_sk(sk);
1024 u32 needed, window, cwnd_len; 1021 u32 needed, window, cwnd_len;
@@ -1039,7 +1036,8 @@ static unsigned int tcp_mss_split_point(struct sock *sk, struct sk_buff *skb,
1039/* Can at least one segment of SKB be sent right now, according to the 1036/* Can at least one segment of SKB be sent right now, according to the
1040 * congestion window rules? If so, return how many segments are allowed. 1037 * congestion window rules? If so, return how many segments are allowed.
1041 */ 1038 */
1042static inline unsigned int tcp_cwnd_test(struct tcp_sock *tp, struct sk_buff *skb) 1039static inline unsigned int tcp_cwnd_test(struct tcp_sock *tp,
1040 struct sk_buff *skb)
1043{ 1041{
1044 u32 in_flight, cwnd; 1042 u32 in_flight, cwnd;
1045 1043
@@ -1059,13 +1057,12 @@ static inline unsigned int tcp_cwnd_test(struct tcp_sock *tp, struct sk_buff *sk
1059/* This must be invoked the first time we consider transmitting 1057/* This must be invoked the first time we consider transmitting
1060 * SKB onto the wire. 1058 * SKB onto the wire.
1061 */ 1059 */
1062static int tcp_init_tso_segs(struct sock *sk, struct sk_buff *skb, unsigned int mss_now) 1060static int tcp_init_tso_segs(struct sock *sk, struct sk_buff *skb,
1061 unsigned int mss_now)
1063{ 1062{
1064 int tso_segs = tcp_skb_pcount(skb); 1063 int tso_segs = tcp_skb_pcount(skb);
1065 1064
1066 if (!tso_segs || 1065 if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) {
1067 (tso_segs > 1 &&
1068 tcp_skb_mss(skb) != mss_now)) {
1069 tcp_set_skb_tso_segs(sk, skb, mss_now); 1066 tcp_set_skb_tso_segs(sk, skb, mss_now);
1070 tso_segs = tcp_skb_pcount(skb); 1067 tso_segs = tcp_skb_pcount(skb);
1071 } 1068 }
@@ -1085,16 +1082,13 @@ static inline int tcp_minshall_check(const struct tcp_sock *tp)
1085 * 4. Or TCP_CORK is not set, and all sent packets are ACKed. 1082 * 4. Or TCP_CORK is not set, and all sent packets are ACKed.
1086 * With Minshall's modification: all sent small packets are ACKed. 1083 * With Minshall's modification: all sent small packets are ACKed.
1087 */ 1084 */
1088
1089static inline int tcp_nagle_check(const struct tcp_sock *tp, 1085static inline int tcp_nagle_check(const struct tcp_sock *tp,
1090 const struct sk_buff *skb, 1086 const struct sk_buff *skb,
1091 unsigned mss_now, int nonagle) 1087 unsigned mss_now, int nonagle)
1092{ 1088{
1093 return (skb->len < mss_now && 1089 return (skb->len < mss_now &&
1094 ((nonagle&TCP_NAGLE_CORK) || 1090 ((nonagle & TCP_NAGLE_CORK) ||
1095 (!nonagle && 1091 (!nonagle && tp->packets_out && tcp_minshall_check(tp))));
1096 tp->packets_out &&
1097 tcp_minshall_check(tp))));
1098} 1092}
1099 1093
1100/* Return non-zero if the Nagle test allows this packet to be 1094/* Return non-zero if the Nagle test allows this packet to be
@@ -1126,7 +1120,8 @@ static inline int tcp_nagle_test(struct tcp_sock *tp, struct sk_buff *skb,
1126} 1120}
1127 1121
1128/* Does at least the first segment of SKB fit into the send window? */ 1122/* Does at least the first segment of SKB fit into the send window? */
1129static inline int tcp_snd_wnd_test(struct tcp_sock *tp, struct sk_buff *skb, unsigned int cur_mss) 1123static inline int tcp_snd_wnd_test(struct tcp_sock *tp, struct sk_buff *skb,
1124 unsigned int cur_mss)
1130{ 1125{
1131 u32 end_seq = TCP_SKB_CB(skb)->end_seq; 1126 u32 end_seq = TCP_SKB_CB(skb)->end_seq;
1132 1127
@@ -1152,8 +1147,7 @@ static unsigned int tcp_snd_test(struct sock *sk, struct sk_buff *skb,
1152 return 0; 1147 return 0;
1153 1148
1154 cwnd_quota = tcp_cwnd_test(tp, skb); 1149 cwnd_quota = tcp_cwnd_test(tp, skb);
1155 if (cwnd_quota && 1150 if (cwnd_quota && !tcp_snd_wnd_test(tp, skb, cur_mss))
1156 !tcp_snd_wnd_test(tp, skb, cur_mss))
1157 cwnd_quota = 0; 1151 cwnd_quota = 0;
1158 1152
1159 return cwnd_quota; 1153 return cwnd_quota;
@@ -1177,7 +1171,8 @@ int tcp_may_send_now(struct sock *sk)
1177 * know that all the data is in scatter-gather pages, and that the 1171 * know that all the data is in scatter-gather pages, and that the
1178 * packet has never been sent out before (and thus is not cloned). 1172 * packet has never been sent out before (and thus is not cloned).
1179 */ 1173 */
1180static int tso_fragment(struct sock *sk, struct sk_buff *skb, unsigned int len, unsigned int mss_now) 1174static int tso_fragment(struct sock *sk, struct sk_buff *skb, unsigned int len,
1175 unsigned int mss_now)
1181{ 1176{
1182 struct sk_buff *buff; 1177 struct sk_buff *buff;
1183 int nlen = skb->len - len; 1178 int nlen = skb->len - len;
@@ -1203,7 +1198,7 @@ static int tso_fragment(struct sock *sk, struct sk_buff *skb, unsigned int len,
1203 1198
1204 /* PSH and FIN should only be set in the second packet. */ 1199 /* PSH and FIN should only be set in the second packet. */
1205 flags = TCP_SKB_CB(skb)->flags; 1200 flags = TCP_SKB_CB(skb)->flags;
1206 TCP_SKB_CB(skb)->flags = flags & ~(TCPCB_FLAG_FIN|TCPCB_FLAG_PSH); 1201 TCP_SKB_CB(skb)->flags = flags & ~(TCPCB_FLAG_FIN | TCPCB_FLAG_PSH);
1207 TCP_SKB_CB(buff)->flags = flags; 1202 TCP_SKB_CB(buff)->flags = flags;
1208 1203
1209 /* This packet was never sent out yet, so no SACK bits. */ 1204 /* This packet was never sent out yet, so no SACK bits. */
@@ -1247,8 +1242,7 @@ static int tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb)
1247 1242
1248 in_flight = tcp_packets_in_flight(tp); 1243 in_flight = tcp_packets_in_flight(tp);
1249 1244
1250 BUG_ON(tcp_skb_pcount(skb) <= 1 || 1245 BUG_ON(tcp_skb_pcount(skb) <= 1 || (tp->snd_cwnd <= in_flight));
1251 (tp->snd_cwnd <= in_flight));
1252 1246
1253 send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; 1247 send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
1254 1248
@@ -1281,7 +1275,7 @@ static int tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb)
1281 } 1275 }
1282 1276
1283 /* Ok, it looks like it is advisable to defer. */ 1277 /* Ok, it looks like it is advisable to defer. */
1284 tp->tso_deferred = 1 | (jiffies<<1); 1278 tp->tso_deferred = 1 | (jiffies << 1);
1285 1279
1286 return 1; 1280 return 1;
1287 1281
@@ -1293,7 +1287,8 @@ send_now:
1293/* Create a new MTU probe if we are ready. 1287/* Create a new MTU probe if we are ready.
1294 * Returns 0 if we should wait to probe (no cwnd available), 1288 * Returns 0 if we should wait to probe (no cwnd available),
1295 * 1 if a probe was sent, 1289 * 1 if a probe was sent,
1296 * -1 otherwise */ 1290 * -1 otherwise
1291 */
1297static int tcp_mtu_probe(struct sock *sk) 1292static int tcp_mtu_probe(struct sock *sk)
1298{ 1293{
1299 struct tcp_sock *tp = tcp_sk(sk); 1294 struct tcp_sock *tp = tcp_sk(sk);
@@ -1318,7 +1313,7 @@ static int tcp_mtu_probe(struct sock *sk)
1318 1313
1319 /* Very simple search strategy: just double the MSS. */ 1314 /* Very simple search strategy: just double the MSS. */
1320 mss_now = tcp_current_mss(sk, 0); 1315 mss_now = tcp_current_mss(sk, 0);
1321 probe_size = 2*tp->mss_cache; 1316 probe_size = 2 * tp->mss_cache;
1322 size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache; 1317 size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache;
1323 if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high)) { 1318 if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high)) {
1324 /* TODO: set timer for probe_converge_event */ 1319 /* TODO: set timer for probe_converge_event */
@@ -1366,7 +1361,8 @@ static int tcp_mtu_probe(struct sock *sk)
1366 skb_copy_bits(skb, 0, skb_put(nskb, copy), copy); 1361 skb_copy_bits(skb, 0, skb_put(nskb, copy), copy);
1367 else 1362 else
1368 nskb->csum = skb_copy_and_csum_bits(skb, 0, 1363 nskb->csum = skb_copy_and_csum_bits(skb, 0,
1369 skb_put(nskb, copy), copy, nskb->csum); 1364 skb_put(nskb, copy),
1365 copy, nskb->csum);
1370 1366
1371 if (skb->len <= copy) { 1367 if (skb->len <= copy) {
1372 /* We've eaten all the data from this skb. 1368 /* We've eaten all the data from this skb.
@@ -1380,7 +1376,8 @@ static int tcp_mtu_probe(struct sock *sk)
1380 if (!skb_shinfo(skb)->nr_frags) { 1376 if (!skb_shinfo(skb)->nr_frags) {
1381 skb_pull(skb, copy); 1377 skb_pull(skb, copy);
1382 if (skb->ip_summed != CHECKSUM_PARTIAL) 1378 if (skb->ip_summed != CHECKSUM_PARTIAL)
1383 skb->csum = csum_partial(skb->data, skb->len, 0); 1379 skb->csum = csum_partial(skb->data,
1380 skb->len, 0);
1384 } else { 1381 } else {
1385 __pskb_trim_head(skb, copy); 1382 __pskb_trim_head(skb, copy);
1386 tcp_set_skb_tso_segs(sk, skb, mss_now); 1383 tcp_set_skb_tso_segs(sk, skb, mss_now);
@@ -1400,7 +1397,7 @@ static int tcp_mtu_probe(struct sock *sk)
1400 TCP_SKB_CB(nskb)->when = tcp_time_stamp; 1397 TCP_SKB_CB(nskb)->when = tcp_time_stamp;
1401 if (!tcp_transmit_skb(sk, nskb, 1, GFP_ATOMIC)) { 1398 if (!tcp_transmit_skb(sk, nskb, 1, GFP_ATOMIC)) {
1402 /* Decrement cwnd here because we are sending 1399 /* Decrement cwnd here because we are sending
1403 * effectively two packets. */ 1400 * effectively two packets. */
1404 tp->snd_cwnd--; 1401 tp->snd_cwnd--;
1405 tcp_event_new_data_sent(sk, nskb); 1402 tcp_event_new_data_sent(sk, nskb);
1406 1403
@@ -1414,7 +1411,6 @@ static int tcp_mtu_probe(struct sock *sk)
1414 return -1; 1411 return -1;
1415} 1412}
1416 1413
1417
1418/* This routine writes packets to the network. It advances the 1414/* This routine writes packets to the network. It advances the
1419 * send_head. This happens as incoming acks open up the remote 1415 * send_head. This happens as incoming acks open up the remote
1420 * window for us. 1416 * window for us.
@@ -1626,7 +1622,8 @@ u32 __tcp_select_window(struct sock *sk)
1626 icsk->icsk_ack.quick = 0; 1622 icsk->icsk_ack.quick = 0;
1627 1623
1628 if (tcp_memory_pressure) 1624 if (tcp_memory_pressure)
1629 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U*tp->advmss); 1625 tp->rcv_ssthresh = min(tp->rcv_ssthresh,
1626 4U * tp->advmss);
1630 1627
1631 if (free_space < mss) 1628 if (free_space < mss)
1632 return 0; 1629 return 0;
@@ -1659,7 +1656,7 @@ u32 __tcp_select_window(struct sock *sk)
1659 * is too small. 1656 * is too small.
1660 */ 1657 */
1661 if (window <= free_space - mss || window > free_space) 1658 if (window <= free_space - mss || window > free_space)
1662 window = (free_space/mss)*mss; 1659 window = (free_space / mss) * mss;
1663 else if (mss == full_space && 1660 else if (mss == full_space &&
1664 free_space > window + (full_space >> 1)) 1661 free_space > window + (full_space >> 1))
1665 window = free_space; 1662 window = free_space;
@@ -1669,7 +1666,8 @@ u32 __tcp_select_window(struct sock *sk)
1669} 1666}
1670 1667
1671/* Attempt to collapse two adjacent SKB's during retransmission. */ 1668/* Attempt to collapse two adjacent SKB's during retransmission. */
1672static void tcp_retrans_try_collapse(struct sock *sk, struct sk_buff *skb, int mss_now) 1669static void tcp_retrans_try_collapse(struct sock *sk, struct sk_buff *skb,
1670 int mss_now)
1673{ 1671{
1674 struct tcp_sock *tp = tcp_sk(sk); 1672 struct tcp_sock *tp = tcp_sk(sk);
1675 struct sk_buff *next_skb = tcp_write_queue_next(sk, skb); 1673 struct sk_buff *next_skb = tcp_write_queue_next(sk, skb);
@@ -1762,12 +1760,12 @@ void tcp_simple_retransmit(struct sock *sk)
1762 if (skb == tcp_send_head(sk)) 1760 if (skb == tcp_send_head(sk))
1763 break; 1761 break;
1764 if (skb->len > mss && 1762 if (skb->len > mss &&
1765 !(TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_ACKED)) { 1763 !(TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) {
1766 if (TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_RETRANS) { 1764 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
1767 TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS; 1765 TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS;
1768 tp->retrans_out -= tcp_skb_pcount(skb); 1766 tp->retrans_out -= tcp_skb_pcount(skb);
1769 } 1767 }
1770 if (!(TCP_SKB_CB(skb)->sacked&TCPCB_LOST)) { 1768 if (!(TCP_SKB_CB(skb)->sacked & TCPCB_LOST)) {
1771 TCP_SKB_CB(skb)->sacked |= TCPCB_LOST; 1769 TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;
1772 tp->lost_out += tcp_skb_pcount(skb); 1770 tp->lost_out += tcp_skb_pcount(skb);
1773 lost = 1; 1771 lost = 1;
@@ -1846,8 +1844,10 @@ int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
1846 (skb->len < (cur_mss >> 1)) && 1844 (skb->len < (cur_mss >> 1)) &&
1847 (tcp_write_queue_next(sk, skb) != tcp_send_head(sk)) && 1845 (tcp_write_queue_next(sk, skb) != tcp_send_head(sk)) &&
1848 (!tcp_skb_is_last(sk, skb)) && 1846 (!tcp_skb_is_last(sk, skb)) &&
1849 (skb_shinfo(skb)->nr_frags == 0 && skb_shinfo(tcp_write_queue_next(sk, skb))->nr_frags == 0) && 1847 (skb_shinfo(skb)->nr_frags == 0 &&
1850 (tcp_skb_pcount(skb) == 1 && tcp_skb_pcount(tcp_write_queue_next(sk, skb)) == 1) && 1848 skb_shinfo(tcp_write_queue_next(sk, skb))->nr_frags == 0) &&
1849 (tcp_skb_pcount(skb) == 1 &&
1850 tcp_skb_pcount(tcp_write_queue_next(sk, skb)) == 1) &&
1851 (sysctl_tcp_retrans_collapse != 0)) 1851 (sysctl_tcp_retrans_collapse != 0))
1852 tcp_retrans_try_collapse(sk, skb, cur_mss); 1852 tcp_retrans_try_collapse(sk, skb, cur_mss);
1853 1853
@@ -1885,7 +1885,7 @@ int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
1885 tp->total_retrans++; 1885 tp->total_retrans++;
1886 1886
1887#if FASTRETRANS_DEBUG > 0 1887#if FASTRETRANS_DEBUG > 0
1888 if (TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_RETRANS) { 1888 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
1889 if (net_ratelimit()) 1889 if (net_ratelimit())
1890 printk(KERN_DEBUG "retrans_out leaked.\n"); 1890 printk(KERN_DEBUG "retrans_out leaked.\n");
1891 } 1891 }
@@ -1927,7 +1927,7 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
1927 if (tp->retransmit_skb_hint) { 1927 if (tp->retransmit_skb_hint) {
1928 skb = tp->retransmit_skb_hint; 1928 skb = tp->retransmit_skb_hint;
1929 packet_cnt = tp->retransmit_cnt_hint; 1929 packet_cnt = tp->retransmit_cnt_hint;
1930 }else{ 1930 } else {
1931 skb = tcp_write_queue_head(sk); 1931 skb = tcp_write_queue_head(sk);
1932 packet_cnt = 0; 1932 packet_cnt = 0;
1933 } 1933 }
@@ -1954,7 +1954,7 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
1954 return; 1954 return;
1955 1955
1956 if (sacked & TCPCB_LOST) { 1956 if (sacked & TCPCB_LOST) {
1957 if (!(sacked&(TCPCB_SACKED_ACKED|TCPCB_SACKED_RETRANS))) { 1957 if (!(sacked & (TCPCB_SACKED_ACKED|TCPCB_SACKED_RETRANS))) {
1958 if (tcp_retransmit_skb(sk, skb)) { 1958 if (tcp_retransmit_skb(sk, skb)) {
1959 tp->retransmit_skb_hint = NULL; 1959 tp->retransmit_skb_hint = NULL;
1960 return; 1960 return;
@@ -2036,7 +2036,6 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
2036 } 2036 }
2037} 2037}
2038 2038
2039
2040/* Send a fin. The caller locks the socket for us. This cannot be 2039/* Send a fin. The caller locks the socket for us. This cannot be
2041 * allowed to fail queueing a FIN frame under any circumstances. 2040 * allowed to fail queueing a FIN frame under any circumstances.
2042 */ 2041 */
@@ -2122,14 +2121,14 @@ void tcp_send_active_reset(struct sock *sk, gfp_t priority)
2122 */ 2121 */
2123int tcp_send_synack(struct sock *sk) 2122int tcp_send_synack(struct sock *sk)
2124{ 2123{
2125 struct sk_buff* skb; 2124 struct sk_buff *skb;
2126 2125
2127 skb = tcp_write_queue_head(sk); 2126 skb = tcp_write_queue_head(sk);
2128 if (skb == NULL || !(TCP_SKB_CB(skb)->flags&TCPCB_FLAG_SYN)) { 2127 if (skb == NULL || !(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_SYN)) {
2129 printk(KERN_DEBUG "tcp_send_synack: wrong queue state\n"); 2128 printk(KERN_DEBUG "tcp_send_synack: wrong queue state\n");
2130 return -EFAULT; 2129 return -EFAULT;
2131 } 2130 }
2132 if (!(TCP_SKB_CB(skb)->flags&TCPCB_FLAG_ACK)) { 2131 if (!(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_ACK)) {
2133 if (skb_cloned(skb)) { 2132 if (skb_cloned(skb)) {
2134 struct sk_buff *nskb = skb_copy(skb, GFP_ATOMIC); 2133 struct sk_buff *nskb = skb_copy(skb, GFP_ATOMIC);
2135 if (nskb == NULL) 2134 if (nskb == NULL)
@@ -2153,8 +2152,8 @@ int tcp_send_synack(struct sock *sk)
2153/* 2152/*
2154 * Prepare a SYN-ACK. 2153 * Prepare a SYN-ACK.
2155 */ 2154 */
2156struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 2155struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
2157 struct request_sock *req) 2156 struct request_sock *req)
2158{ 2157{
2159 struct inet_request_sock *ireq = inet_rsk(req); 2158 struct inet_request_sock *ireq = inet_rsk(req);
2160 struct tcp_sock *tp = tcp_sk(sk); 2159 struct tcp_sock *tp = tcp_sk(sk);
@@ -2372,9 +2371,10 @@ void tcp_send_delayed_ack(struct sock *sk)
2372 2371
2373 if (ato > TCP_DELACK_MIN) { 2372 if (ato > TCP_DELACK_MIN) {
2374 const struct tcp_sock *tp = tcp_sk(sk); 2373 const struct tcp_sock *tp = tcp_sk(sk);
2375 int max_ato = HZ/2; 2374 int max_ato = HZ / 2;
2376 2375
2377 if (icsk->icsk_ack.pingpong || (icsk->icsk_ack.pending & ICSK_ACK_PUSHED)) 2376 if (icsk->icsk_ack.pingpong ||
2377 (icsk->icsk_ack.pending & ICSK_ACK_PUSHED))
2378 max_ato = TCP_DELACK_MAX; 2378 max_ato = TCP_DELACK_MAX;
2379 2379
2380 /* Slow path, intersegment interval is "high". */ 2380 /* Slow path, intersegment interval is "high". */
@@ -2384,7 +2384,7 @@ void tcp_send_delayed_ack(struct sock *sk)
2384 * directly. 2384 * directly.
2385 */ 2385 */
2386 if (tp->srtt) { 2386 if (tp->srtt) {
2387 int rtt = max(tp->srtt>>3, TCP_DELACK_MIN); 2387 int rtt = max(tp->srtt >> 3, TCP_DELACK_MIN);
2388 2388
2389 if (rtt < max_ato) 2389 if (rtt < max_ato)
2390 max_ato = rtt; 2390 max_ato = rtt;
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-04 19:47:01 -0500