aboutsummaryrefslogtreecommitdiffstats
path: root/net/ipv4/tcp_input.c
diff options
context:
space:
mode:
Diffstat (limited to 'net/ipv4/tcp_input.c')
-rw-r--r--net/ipv4/tcp_input.c82
1 files changed, 46 insertions, 36 deletions
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index f8af018dd224..faba9beb3613 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -111,10 +111,6 @@ int sysctl_tcp_abc __read_mostly;
111#define FLAG_FORWARD_PROGRESS (FLAG_ACKED|FLAG_DATA_SACKED) 111#define FLAG_FORWARD_PROGRESS (FLAG_ACKED|FLAG_DATA_SACKED)
112#define FLAG_ANY_PROGRESS (FLAG_FORWARD_PROGRESS|FLAG_SND_UNA_ADVANCED) 112#define FLAG_ANY_PROGRESS (FLAG_FORWARD_PROGRESS|FLAG_SND_UNA_ADVANCED)
113 113
114#define IsReno(tp) ((tp)->rx_opt.sack_ok == 0)
115#define IsFack(tp) ((tp)->rx_opt.sack_ok & 2)
116#define IsDSack(tp) ((tp)->rx_opt.sack_ok & 4)
117
118#define IsSackFrto() (sysctl_tcp_frto == 0x2) 114#define IsSackFrto() (sysctl_tcp_frto == 0x2)
119 115
120#define TCP_REMNANT (TCP_FLAG_FIN|TCP_FLAG_URG|TCP_FLAG_SYN|TCP_FLAG_PSH) 116#define TCP_REMNANT (TCP_FLAG_FIN|TCP_FLAG_URG|TCP_FLAG_SYN|TCP_FLAG_PSH)
@@ -860,6 +856,21 @@ void tcp_enter_cwr(struct sock *sk, const int set_ssthresh)
860 } 856 }
861} 857}
862 858
859/*
860 * Packet counting of FACK is based on in-order assumptions, therefore TCP
861 * disables it when reordering is detected
862 */
863static void tcp_disable_fack(struct tcp_sock *tp)
864{
865 tp->rx_opt.sack_ok &= ~2;
866}
867
868/* Take a notice that peer is sending DSACKs */
869static void tcp_dsack_seen(struct tcp_sock *tp)
870{
871 tp->rx_opt.sack_ok |= 4;
872}
873
863/* Initialize metrics on socket. */ 874/* Initialize metrics on socket. */
864 875
865static void tcp_init_metrics(struct sock *sk) 876static void tcp_init_metrics(struct sock *sk)
@@ -881,7 +892,7 @@ static void tcp_init_metrics(struct sock *sk)
881 } 892 }
882 if (dst_metric(dst, RTAX_REORDERING) && 893 if (dst_metric(dst, RTAX_REORDERING) &&
883 tp->reordering != dst_metric(dst, RTAX_REORDERING)) { 894 tp->reordering != dst_metric(dst, RTAX_REORDERING)) {
884 tp->rx_opt.sack_ok &= ~2; 895 tcp_disable_fack(tp);
885 tp->reordering = dst_metric(dst, RTAX_REORDERING); 896 tp->reordering = dst_metric(dst, RTAX_REORDERING);
886 } 897 }
887 898
@@ -943,9 +954,9 @@ static void tcp_update_reordering(struct sock *sk, const int metric,
943 /* This exciting event is worth to be remembered. 8) */ 954 /* This exciting event is worth to be remembered. 8) */
944 if (ts) 955 if (ts)
945 NET_INC_STATS_BH(LINUX_MIB_TCPTSREORDER); 956 NET_INC_STATS_BH(LINUX_MIB_TCPTSREORDER);
946 else if (IsReno(tp)) 957 else if (tcp_is_reno(tp))
947 NET_INC_STATS_BH(LINUX_MIB_TCPRENOREORDER); 958 NET_INC_STATS_BH(LINUX_MIB_TCPRENOREORDER);
948 else if (IsFack(tp)) 959 else if (tcp_is_fack(tp))
949 NET_INC_STATS_BH(LINUX_MIB_TCPFACKREORDER); 960 NET_INC_STATS_BH(LINUX_MIB_TCPFACKREORDER);
950 else 961 else
951 NET_INC_STATS_BH(LINUX_MIB_TCPSACKREORDER); 962 NET_INC_STATS_BH(LINUX_MIB_TCPSACKREORDER);
@@ -957,8 +968,7 @@ static void tcp_update_reordering(struct sock *sk, const int metric,
957 tp->sacked_out, 968 tp->sacked_out,
958 tp->undo_marker ? tp->undo_retrans : 0); 969 tp->undo_marker ? tp->undo_retrans : 0);
959#endif 970#endif
960 /* Disable FACK yet. */ 971 tcp_disable_fack(tp);
961 tp->rx_opt.sack_ok &= ~2;
962 } 972 }
963} 973}
964 974
@@ -1020,7 +1030,7 @@ static int tcp_check_dsack(struct tcp_sock *tp, struct sk_buff *ack_skb,
1020 1030
1021 if (before(start_seq_0, TCP_SKB_CB(ack_skb)->ack_seq)) { 1031 if (before(start_seq_0, TCP_SKB_CB(ack_skb)->ack_seq)) {
1022 dup_sack = 1; 1032 dup_sack = 1;
1023 tp->rx_opt.sack_ok |= 4; 1033 tcp_dsack_seen(tp);
1024 NET_INC_STATS_BH(LINUX_MIB_TCPDSACKRECV); 1034 NET_INC_STATS_BH(LINUX_MIB_TCPDSACKRECV);
1025 } else if (num_sacks > 1) { 1035 } else if (num_sacks > 1) {
1026 u32 end_seq_1 = ntohl(get_unaligned(&sp[1].end_seq)); 1036 u32 end_seq_1 = ntohl(get_unaligned(&sp[1].end_seq));
@@ -1029,7 +1039,7 @@ static int tcp_check_dsack(struct tcp_sock *tp, struct sk_buff *ack_skb,
1029 if (!after(end_seq_0, end_seq_1) && 1039 if (!after(end_seq_0, end_seq_1) &&
1030 !before(start_seq_0, start_seq_1)) { 1040 !before(start_seq_0, start_seq_1)) {
1031 dup_sack = 1; 1041 dup_sack = 1;
1032 tp->rx_opt.sack_ok |= 4; 1042 tcp_dsack_seen(tp);
1033 NET_INC_STATS_BH(LINUX_MIB_TCPDSACKOFORECV); 1043 NET_INC_STATS_BH(LINUX_MIB_TCPDSACKOFORECV);
1034 } 1044 }
1035 } 1045 }
@@ -1326,7 +1336,7 @@ tcp_sacktag_write_queue(struct sock *sk, struct sk_buff *ack_skb, u32 prior_snd_
1326 continue; 1336 continue;
1327 if ((TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_RETRANS) && 1337 if ((TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_RETRANS) &&
1328 after(lost_retrans, TCP_SKB_CB(skb)->ack_seq) && 1338 after(lost_retrans, TCP_SKB_CB(skb)->ack_seq) &&
1329 (IsFack(tp) || 1339 (tcp_is_fack(tp) ||
1330 !before(lost_retrans, 1340 !before(lost_retrans,
1331 TCP_SKB_CB(skb)->ack_seq + tp->reordering * 1341 TCP_SKB_CB(skb)->ack_seq + tp->reordering *
1332 tp->mss_cache))) { 1342 tp->mss_cache))) {
@@ -1526,7 +1536,7 @@ static void tcp_enter_frto_loss(struct sock *sk, int allowed_segments, int flag)
1526 1536
1527 tp->lost_out = 0; 1537 tp->lost_out = 0;
1528 tp->retrans_out = 0; 1538 tp->retrans_out = 0;
1529 if (IsReno(tp)) 1539 if (tcp_is_reno(tp))
1530 tcp_reset_reno_sack(tp); 1540 tcp_reset_reno_sack(tp);
1531 1541
1532 tcp_for_write_queue(skb, sk) { 1542 tcp_for_write_queue(skb, sk) {
@@ -1668,7 +1678,7 @@ static int tcp_check_sack_reneging(struct sock *sk)
1668 1678
1669static inline int tcp_fackets_out(struct tcp_sock *tp) 1679static inline int tcp_fackets_out(struct tcp_sock *tp)
1670{ 1680{
1671 return IsReno(tp) ? tp->sacked_out+1 : tp->fackets_out; 1681 return tcp_is_reno(tp) ? tp->sacked_out+1 : tp->fackets_out;
1672} 1682}
1673 1683
1674static inline int tcp_skb_timedout(struct sock *sk, struct sk_buff *skb) 1684static inline int tcp_skb_timedout(struct sock *sk, struct sk_buff *skb)
@@ -1872,7 +1882,7 @@ static void tcp_update_scoreboard(struct sock *sk)
1872{ 1882{
1873 struct tcp_sock *tp = tcp_sk(sk); 1883 struct tcp_sock *tp = tcp_sk(sk);
1874 1884
1875 if (IsFack(tp)) { 1885 if (tcp_is_fack(tp)) {
1876 int lost = tp->fackets_out - tp->reordering; 1886 int lost = tp->fackets_out - tp->reordering;
1877 if (lost <= 0) 1887 if (lost <= 0)
1878 lost = 1; 1888 lost = 1;
@@ -1886,7 +1896,7 @@ static void tcp_update_scoreboard(struct sock *sk)
1886 * Hence, we can detect timed out packets during fast 1896 * Hence, we can detect timed out packets during fast
1887 * retransmit without falling to slow start. 1897 * retransmit without falling to slow start.
1888 */ 1898 */
1889 if (!IsReno(tp) && tcp_head_timedout(sk)) { 1899 if (!tcp_is_reno(tp) && tcp_head_timedout(sk)) {
1890 struct sk_buff *skb; 1900 struct sk_buff *skb;
1891 1901
1892 skb = tp->scoreboard_skb_hint ? tp->scoreboard_skb_hint 1902 skb = tp->scoreboard_skb_hint ? tp->scoreboard_skb_hint
@@ -1938,7 +1948,7 @@ static void tcp_cwnd_down(struct sock *sk, int flag)
1938 int decr = tp->snd_cwnd_cnt + 1; 1948 int decr = tp->snd_cwnd_cnt + 1;
1939 1949
1940 if ((flag&(FLAG_ANY_PROGRESS|FLAG_DSACKING_ACK)) || 1950 if ((flag&(FLAG_ANY_PROGRESS|FLAG_DSACKING_ACK)) ||
1941 (IsReno(tp) && !(flag&FLAG_NOT_DUP))) { 1951 (tcp_is_reno(tp) && !(flag&FLAG_NOT_DUP))) {
1942 tp->snd_cwnd_cnt = decr&1; 1952 tp->snd_cwnd_cnt = decr&1;
1943 decr >>= 1; 1953 decr >>= 1;
1944 1954
@@ -2029,7 +2039,7 @@ static int tcp_try_undo_recovery(struct sock *sk)
2029 NET_INC_STATS_BH(LINUX_MIB_TCPFULLUNDO); 2039 NET_INC_STATS_BH(LINUX_MIB_TCPFULLUNDO);
2030 tp->undo_marker = 0; 2040 tp->undo_marker = 0;
2031 } 2041 }
2032 if (tp->snd_una == tp->high_seq && IsReno(tp)) { 2042 if (tp->snd_una == tp->high_seq && tcp_is_reno(tp)) {
2033 /* Hold old state until something *above* high_seq 2043 /* Hold old state until something *above* high_seq
2034 * is ACKed. For Reno it is MUST to prevent false 2044 * is ACKed. For Reno it is MUST to prevent false
2035 * fast retransmits (RFC2582). SACK TCP is safe. */ 2045 * fast retransmits (RFC2582). SACK TCP is safe. */
@@ -2059,7 +2069,7 @@ static int tcp_try_undo_partial(struct sock *sk, int acked)
2059{ 2069{
2060 struct tcp_sock *tp = tcp_sk(sk); 2070 struct tcp_sock *tp = tcp_sk(sk);
2061 /* Partial ACK arrived. Force Hoe's retransmit. */ 2071 /* Partial ACK arrived. Force Hoe's retransmit. */
2062 int failed = IsReno(tp) || tp->fackets_out>tp->reordering; 2072 int failed = tcp_is_reno(tp) || tp->fackets_out>tp->reordering;
2063 2073
2064 if (tcp_may_undo(tp)) { 2074 if (tcp_may_undo(tp)) {
2065 /* Plain luck! Hole if filled with delayed 2075 /* Plain luck! Hole if filled with delayed
@@ -2104,7 +2114,7 @@ static int tcp_try_undo_loss(struct sock *sk)
2104 NET_INC_STATS_BH(LINUX_MIB_TCPLOSSUNDO); 2114 NET_INC_STATS_BH(LINUX_MIB_TCPLOSSUNDO);
2105 inet_csk(sk)->icsk_retransmits = 0; 2115 inet_csk(sk)->icsk_retransmits = 0;
2106 tp->undo_marker = 0; 2116 tp->undo_marker = 0;
2107 if (!IsReno(tp)) 2117 if (tcp_is_sack(tp))
2108 tcp_set_ca_state(sk, TCP_CA_Open); 2118 tcp_set_ca_state(sk, TCP_CA_Open);
2109 return 1; 2119 return 1;
2110 } 2120 }
@@ -2251,14 +2261,14 @@ tcp_fastretrans_alert(struct sock *sk, int pkts_acked, int flag)
2251 if (!tp->undo_marker || 2261 if (!tp->undo_marker ||
2252 /* For SACK case do not Open to allow to undo 2262 /* For SACK case do not Open to allow to undo
2253 * catching for all duplicate ACKs. */ 2263 * catching for all duplicate ACKs. */
2254 IsReno(tp) || tp->snd_una != tp->high_seq) { 2264 tcp_is_reno(tp) || tp->snd_una != tp->high_seq) {
2255 tp->undo_marker = 0; 2265 tp->undo_marker = 0;
2256 tcp_set_ca_state(sk, TCP_CA_Open); 2266 tcp_set_ca_state(sk, TCP_CA_Open);
2257 } 2267 }
2258 break; 2268 break;
2259 2269
2260 case TCP_CA_Recovery: 2270 case TCP_CA_Recovery:
2261 if (IsReno(tp)) 2271 if (tcp_is_reno(tp))
2262 tcp_reset_reno_sack(tp); 2272 tcp_reset_reno_sack(tp);
2263 if (tcp_try_undo_recovery(sk)) 2273 if (tcp_try_undo_recovery(sk))
2264 return; 2274 return;
@@ -2271,7 +2281,7 @@ tcp_fastretrans_alert(struct sock *sk, int pkts_acked, int flag)
2271 switch (icsk->icsk_ca_state) { 2281 switch (icsk->icsk_ca_state) {
2272 case TCP_CA_Recovery: 2282 case TCP_CA_Recovery:
2273 if (!(flag & FLAG_SND_UNA_ADVANCED)) { 2283 if (!(flag & FLAG_SND_UNA_ADVANCED)) {
2274 if (IsReno(tp) && is_dupack) 2284 if (tcp_is_reno(tp) && is_dupack)
2275 tcp_add_reno_sack(sk); 2285 tcp_add_reno_sack(sk);
2276 } else 2286 } else
2277 do_lost = tcp_try_undo_partial(sk, pkts_acked); 2287 do_lost = tcp_try_undo_partial(sk, pkts_acked);
@@ -2288,7 +2298,7 @@ tcp_fastretrans_alert(struct sock *sk, int pkts_acked, int flag)
2288 return; 2298 return;
2289 /* Loss is undone; fall through to processing in Open state. */ 2299 /* Loss is undone; fall through to processing in Open state. */
2290 default: 2300 default:
2291 if (IsReno(tp)) { 2301 if (tcp_is_reno(tp)) {
2292 if (flag & FLAG_SND_UNA_ADVANCED) 2302 if (flag & FLAG_SND_UNA_ADVANCED)
2293 tcp_reset_reno_sack(tp); 2303 tcp_reset_reno_sack(tp);
2294 if (is_dupack) 2304 if (is_dupack)
@@ -2316,7 +2326,7 @@ tcp_fastretrans_alert(struct sock *sk, int pkts_acked, int flag)
2316 2326
2317 /* Otherwise enter Recovery state */ 2327 /* Otherwise enter Recovery state */
2318 2328
2319 if (IsReno(tp)) 2329 if (tcp_is_reno(tp))
2320 NET_INC_STATS_BH(LINUX_MIB_TCPRENORECOVERY); 2330 NET_INC_STATS_BH(LINUX_MIB_TCPRENORECOVERY);
2321 else 2331 else
2322 NET_INC_STATS_BH(LINUX_MIB_TCPSACKRECOVERY); 2332 NET_INC_STATS_BH(LINUX_MIB_TCPSACKRECOVERY);
@@ -2573,7 +2583,7 @@ static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p)
2573 tcp_ack_update_rtt(sk, acked, seq_rtt); 2583 tcp_ack_update_rtt(sk, acked, seq_rtt);
2574 tcp_ack_packets_out(sk); 2584 tcp_ack_packets_out(sk);
2575 2585
2576 if (IsReno(tp)) 2586 if (tcp_is_reno(tp))
2577 tcp_remove_reno_sacks(sk, pkts_acked); 2587 tcp_remove_reno_sacks(sk, pkts_acked);
2578 2588
2579 if (ca_ops->pkts_acked) { 2589 if (ca_ops->pkts_acked) {
@@ -2599,7 +2609,7 @@ static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p)
2599 BUG_TRAP((int)tp->sacked_out >= 0); 2609 BUG_TRAP((int)tp->sacked_out >= 0);
2600 BUG_TRAP((int)tp->lost_out >= 0); 2610 BUG_TRAP((int)tp->lost_out >= 0);
2601 BUG_TRAP((int)tp->retrans_out >= 0); 2611 BUG_TRAP((int)tp->retrans_out >= 0);
2602 if (!tp->packets_out && tp->rx_opt.sack_ok) { 2612 if (!tp->packets_out && tcp_is_sack(tp)) {
2603 const struct inet_connection_sock *icsk = inet_csk(sk); 2613 const struct inet_connection_sock *icsk = inet_csk(sk);
2604 if (tp->lost_out) { 2614 if (tp->lost_out) {
2605 printk(KERN_DEBUG "Leak l=%u %d\n", 2615 printk(KERN_DEBUG "Leak l=%u %d\n",
@@ -2779,7 +2789,7 @@ static int tcp_process_frto(struct sock *sk, int flag)
2779 return 1; 2789 return 1;
2780 } 2790 }
2781 2791
2782 if (!IsSackFrto() || IsReno(tp)) { 2792 if (!IsSackFrto() || tcp_is_reno(tp)) {
2783 /* RFC4138 shortcoming in step 2; should also have case c): 2793 /* RFC4138 shortcoming in step 2; should also have case c):
2784 * ACK isn't duplicate nor advances window, e.g., opposite dir 2794 * ACK isn't duplicate nor advances window, e.g., opposite dir
2785 * data, winupdate 2795 * data, winupdate
@@ -3263,7 +3273,7 @@ static void tcp_fin(struct sk_buff *skb, struct sock *sk, struct tcphdr *th)
3263 * Probably, we should reset in this case. For now drop them. 3273 * Probably, we should reset in this case. For now drop them.
3264 */ 3274 */
3265 __skb_queue_purge(&tp->out_of_order_queue); 3275 __skb_queue_purge(&tp->out_of_order_queue);
3266 if (tp->rx_opt.sack_ok) 3276 if (tcp_is_sack(tp))
3267 tcp_sack_reset(&tp->rx_opt); 3277 tcp_sack_reset(&tp->rx_opt);
3268 sk_stream_mem_reclaim(sk); 3278 sk_stream_mem_reclaim(sk);
3269 3279
@@ -3293,7 +3303,7 @@ static inline int tcp_sack_extend(struct tcp_sack_block *sp, u32 seq, u32 end_se
3293 3303
3294static void tcp_dsack_set(struct tcp_sock *tp, u32 seq, u32 end_seq) 3304static void tcp_dsack_set(struct tcp_sock *tp, u32 seq, u32 end_seq)
3295{ 3305{
3296 if (tp->rx_opt.sack_ok && sysctl_tcp_dsack) { 3306 if (tcp_is_sack(tp) && sysctl_tcp_dsack) {
3297 if (before(seq, tp->rcv_nxt)) 3307 if (before(seq, tp->rcv_nxt))
3298 NET_INC_STATS_BH(LINUX_MIB_TCPDSACKOLDSENT); 3308 NET_INC_STATS_BH(LINUX_MIB_TCPDSACKOLDSENT);
3299 else 3309 else
@@ -3323,7 +3333,7 @@ static void tcp_send_dupack(struct sock *sk, struct sk_buff *skb)
3323 NET_INC_STATS_BH(LINUX_MIB_DELAYEDACKLOST); 3333 NET_INC_STATS_BH(LINUX_MIB_DELAYEDACKLOST);
3324 tcp_enter_quickack_mode(sk); 3334 tcp_enter_quickack_mode(sk);
3325 3335
3326 if (tp->rx_opt.sack_ok && sysctl_tcp_dsack) { 3336 if (tcp_is_sack(tp) && sysctl_tcp_dsack) {
3327 u32 end_seq = TCP_SKB_CB(skb)->end_seq; 3337 u32 end_seq = TCP_SKB_CB(skb)->end_seq;
3328 3338
3329 if (after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) 3339 if (after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt))
@@ -3639,7 +3649,7 @@ drop:
3639 3649
3640 if (!skb_peek(&tp->out_of_order_queue)) { 3650 if (!skb_peek(&tp->out_of_order_queue)) {
3641 /* Initial out of order segment, build 1 SACK. */ 3651 /* Initial out of order segment, build 1 SACK. */
3642 if (tp->rx_opt.sack_ok) { 3652 if (tcp_is_sack(tp)) {
3643 tp->rx_opt.num_sacks = 1; 3653 tp->rx_opt.num_sacks = 1;
3644 tp->rx_opt.dsack = 0; 3654 tp->rx_opt.dsack = 0;
3645 tp->rx_opt.eff_sacks = 1; 3655 tp->rx_opt.eff_sacks = 1;
@@ -3704,7 +3714,7 @@ drop:
3704 } 3714 }
3705 3715
3706add_sack: 3716add_sack:
3707 if (tp->rx_opt.sack_ok) 3717 if (tcp_is_sack(tp))
3708 tcp_sack_new_ofo_skb(sk, seq, end_seq); 3718 tcp_sack_new_ofo_skb(sk, seq, end_seq);
3709 } 3719 }
3710} 3720}
@@ -3893,7 +3903,7 @@ static int tcp_prune_queue(struct sock *sk)
3893 * is in a sad state like this, we care only about integrity 3903 * is in a sad state like this, we care only about integrity
3894 * of the connection not performance. 3904 * of the connection not performance.
3895 */ 3905 */
3896 if (tp->rx_opt.sack_ok) 3906 if (tcp_is_sack(tp))
3897 tcp_sack_reset(&tp->rx_opt); 3907 tcp_sack_reset(&tp->rx_opt);
3898 sk_stream_mem_reclaim(sk); 3908 sk_stream_mem_reclaim(sk);
3899 } 3909 }
@@ -4594,8 +4604,8 @@ static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb,
4594 tp->tcp_header_len = sizeof(struct tcphdr); 4604 tp->tcp_header_len = sizeof(struct tcphdr);
4595 } 4605 }
4596 4606
4597 if (tp->rx_opt.sack_ok && sysctl_tcp_fack) 4607 if (tcp_is_sack(tp) && sysctl_tcp_fack)
4598 tp->rx_opt.sack_ok |= 2; 4608 tcp_enable_fack(tp);
4599 4609
4600 tcp_mtup_init(sk); 4610 tcp_mtup_init(sk);
4601 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); 4611 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);