Lines Matching +full:tp +full:- +full:sensitive +full:- +full:adjust
1 // SPDX-License-Identifier: GPL-2.0-only
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche, <flla@stud.uni-sb.de>
33 * Cacophonix Gaul : draft-minshall-nagle-01
53 void tcp_mstamp_refresh(struct tcp_sock *tp) in tcp_mstamp_refresh() argument
57 tp->tcp_clock_cache = val; in tcp_mstamp_refresh()
58 tp->tcp_mstamp = div_u64(val, NSEC_PER_USEC); in tcp_mstamp_refresh()
68 struct tcp_sock *tp = tcp_sk(sk); in tcp_event_new_data_sent() local
69 unsigned int prior_packets = tp->packets_out; in tcp_event_new_data_sent()
71 WRITE_ONCE(tp->snd_nxt, TCP_SKB_CB(skb)->end_seq); in tcp_event_new_data_sent()
73 __skb_unlink(skb, &sk->sk_write_queue); in tcp_event_new_data_sent()
74 tcp_rbtree_insert(&sk->tcp_rtx_queue, skb); in tcp_event_new_data_sent()
76 if (tp->highest_sack == NULL) in tcp_event_new_data_sent()
77 tp->highest_sack = skb; in tcp_event_new_data_sent()
79 tp->packets_out += tcp_skb_pcount(skb); in tcp_event_new_data_sent()
80 if (!prior_packets || icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) in tcp_event_new_data_sent()
91 * Using SND.UNA we will fail to open window, SND.NXT is out of window. :-(
97 const struct tcp_sock *tp = tcp_sk(sk); in tcp_acceptable_seq() local
99 if (!before(tcp_wnd_end(tp), tp->snd_nxt) || in tcp_acceptable_seq()
100 (tp->rx_opt.wscale_ok && in tcp_acceptable_seq()
101 ((tp->snd_nxt - tcp_wnd_end(tp)) < (1 << tp->rx_opt.rcv_wscale)))) in tcp_acceptable_seq()
102 return tp->snd_nxt; in tcp_acceptable_seq()
104 return tcp_wnd_end(tp); in tcp_acceptable_seq()
108 * RFC1122, RFC1063, draft-ietf-tcpimpl-pmtud-01 state that:
111 * 2. Ideally, it is maximal possible segment size i.e. 65535-40.
123 struct tcp_sock *tp = tcp_sk(sk); in tcp_advertise_mss() local
125 int mss = tp->advmss; in tcp_advertise_mss()
132 tp->advmss = mss; in tcp_advertise_mss()
144 struct tcp_sock *tp = tcp_sk(sk); in tcp_cwnd_restart() local
145 u32 restart_cwnd = tcp_init_cwnd(tp, __sk_dst_get(sk)); in tcp_cwnd_restart()
146 u32 cwnd = tcp_snd_cwnd(tp); in tcp_cwnd_restart()
150 tp->snd_ssthresh = tcp_current_ssthresh(sk); in tcp_cwnd_restart()
153 while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd) in tcp_cwnd_restart()
155 tcp_snd_cwnd_set(tp, max(cwnd, restart_cwnd)); in tcp_cwnd_restart()
156 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_cwnd_restart()
157 tp->snd_cwnd_used = 0; in tcp_cwnd_restart()
161 static void tcp_event_data_sent(struct tcp_sock *tp, in tcp_event_data_sent() argument
167 if (tcp_packets_in_flight(tp) == 0) in tcp_event_data_sent()
170 tp->lsndtime = now; in tcp_event_data_sent()
175 if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato) in tcp_event_data_sent()
182 struct tcp_sock *tp = tcp_sk(sk); in tcp_event_ack_sent() local
184 if (unlikely(tp->compressed_ack)) { in tcp_event_ack_sent()
186 tp->compressed_ack); in tcp_event_ack_sent()
187 tp->compressed_ack = 0; in tcp_event_ack_sent()
188 if (hrtimer_try_to_cancel(&tp->compressed_ack_timer) == 1) in tcp_event_ack_sent()
192 if (unlikely(rcv_nxt != tp->rcv_nxt)) in tcp_event_ack_sent()
200 * will be offered. Store the results in the tp structure.
224 * we will truncate our initial window offering to 32K-1 in tcp_select_initial_window()
229 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_workaround_signed_windows)) in tcp_select_initial_window()
240 space = max_t(u32, space, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2])); in tcp_select_initial_window()
243 *rcv_wscale = clamp_t(int, ilog2(space) - 15, in tcp_select_initial_window()
253 * value can be stuffed directly into th->window for an outgoing
258 struct tcp_sock *tp = tcp_sk(sk); in tcp_select_window() local
260 u32 old_win = tp->rcv_wnd; in tcp_select_window()
267 if (unlikely(inet_csk(sk)->icsk_ack.pending & ICSK_ACK_NOMEM)) in tcp_select_window()
270 cur_win = tcp_receive_window(tp); in tcp_select_window()
276 * window in time. --DaveM in tcp_select_window()
280 if (!READ_ONCE(net->ipv4.sysctl_tcp_shrink_window) || !tp->rx_opt.rcv_wscale) { in tcp_select_window()
284 new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale); in tcp_select_window()
288 tp->rcv_wnd = new_win; in tcp_select_window()
289 tp->rcv_wup = tp->rcv_nxt; in tcp_select_window()
294 if (!tp->rx_opt.rcv_wscale && in tcp_select_window()
295 READ_ONCE(net->ipv4.sysctl_tcp_workaround_signed_windows)) in tcp_select_window()
298 new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale)); in tcp_select_window()
301 new_win >>= tp->rx_opt.rcv_wscale; in tcp_select_window()
305 tp->pred_flags = 0; in tcp_select_window()
315 /* Packet ECN state for a SYN-ACK */
318 const struct tcp_sock *tp = tcp_sk(sk); in tcp_ecn_send_synack() local
320 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR; in tcp_ecn_send_synack()
321 if (!(tp->ecn_flags & TCP_ECN_OK)) in tcp_ecn_send_synack()
322 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE; in tcp_ecn_send_synack()
331 struct tcp_sock *tp = tcp_sk(sk); in tcp_ecn_send_syn() local
333 bool use_ecn = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_ecn) == 1 || in tcp_ecn_send_syn()
343 tp->ecn_flags = 0; in tcp_ecn_send_syn()
346 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR; in tcp_ecn_send_syn()
347 tp->ecn_flags = TCP_ECN_OK; in tcp_ecn_send_syn()
355 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_ecn_fallback)) in tcp_ecn_clear_syn()
356 /* tp->ecn_flags are cleared at a later point in time when in tcp_ecn_clear_syn()
359 TCP_SKB_CB(skb)->tcp_flags &= ~(TCPHDR_ECE | TCPHDR_CWR); in tcp_ecn_clear_syn()
365 if (inet_rsk(req)->ecn_ok) in tcp_ecn_make_synack()
366 th->ece = 1; in tcp_ecn_make_synack()
375 struct tcp_sock *tp = tcp_sk(sk); in tcp_ecn_send() local
377 if (tp->ecn_flags & TCP_ECN_OK) { in tcp_ecn_send()
378 /* Not-retransmitted data segment: set ECT and inject CWR. */ in tcp_ecn_send()
379 if (skb->len != tcp_header_len && in tcp_ecn_send()
380 !before(TCP_SKB_CB(skb)->seq, tp->snd_nxt)) { in tcp_ecn_send()
382 if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) { in tcp_ecn_send()
383 tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR; in tcp_ecn_send()
384 th->cwr = 1; in tcp_ecn_send()
385 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; in tcp_ecn_send()
391 if (tp->ecn_flags & TCP_ECN_DEMAND_CWR) in tcp_ecn_send()
392 th->ece = 1; in tcp_ecn_send()
396 /* Constructs common control bits of non-data skb. If SYN/FIN is present,
401 skb->ip_summed = CHECKSUM_PARTIAL; in tcp_init_nondata_skb()
403 TCP_SKB_CB(skb)->tcp_flags = flags; in tcp_init_nondata_skb()
407 TCP_SKB_CB(skb)->seq = seq; in tcp_init_nondata_skb()
410 TCP_SKB_CB(skb)->end_seq = seq; in tcp_init_nondata_skb()
413 static inline bool tcp_urg_mode(const struct tcp_sock *tp) in tcp_urg_mode() argument
415 return tp->snd_una != tp->snd_up; in tcp_urg_mode()
455 struct tcp_sock *tp, in mptcp_options_write() argument
459 if (unlikely(OPTION_MPTCP & opts->options)) in mptcp_options_write()
460 mptcp_write_options(th, ptr, tp, &opts->mptcp); in mptcp_options_write()
506 * Thus, "req" is passed here and the cgroup-bpf-progs in bpf_skops_hdr_opt_len()
511 * consistent between fastopen and non-fastopen on in bpf_skops_hdr_opt_len()
534 opts->bpf_opt_len = *remaining - sock_ops.remaining_opt_len; in bpf_skops_hdr_opt_len()
536 opts->bpf_opt_len = (opts->bpf_opt_len + 3) & ~3; in bpf_skops_hdr_opt_len()
538 *remaining -= opts->bpf_opt_len; in bpf_skops_hdr_opt_len()
547 u8 first_opt_off, nr_written, max_opt_len = opts->bpf_opt_len; in bpf_skops_write_hdr_opt()
570 first_opt_off = tcp_hdrlen(skb) - max_opt_len; in bpf_skops_write_hdr_opt()
578 nr_written = max_opt_len - sock_ops.remaining_opt_len; in bpf_skops_write_hdr_opt()
581 memset(skb->data + first_opt_off + nr_written, TCPOPT_NOP, in bpf_skops_write_hdr_opt()
582 max_opt_len - nr_written); in bpf_skops_write_hdr_opt()
605 * Beware: Something in the Internet is very sensitive to the ordering of
607 * Luckily we can at least blame others for their non-compliance but from
608 * inter-operability perspective it seems that we're somewhat stuck with
616 static void tcp_options_write(struct tcphdr *th, struct tcp_sock *tp, in tcp_options_write() argument
620 u16 options = opts->options; /* mungable copy */ in tcp_options_write()
626 opts->hash_location = (__u8 *)ptr; in tcp_options_write()
630 if (unlikely(opts->mss)) { in tcp_options_write()
633 opts->mss); in tcp_options_write()
649 *ptr++ = htonl(opts->tsval); in tcp_options_write()
650 *ptr++ = htonl(opts->tsecr); in tcp_options_write()
664 opts->ws); in tcp_options_write()
667 if (unlikely(opts->num_sack_blocks)) { in tcp_options_write()
668 struct tcp_sack_block *sp = tp->rx_opt.dsack ? in tcp_options_write()
669 tp->duplicate_sack : tp->selective_acks; in tcp_options_write()
675 (TCPOLEN_SACK_BASE + (opts->num_sack_blocks * in tcp_options_write()
678 for (this_sack = 0; this_sack < opts->num_sack_blocks; in tcp_options_write()
684 tp->rx_opt.dsack = 0; in tcp_options_write()
688 struct tcp_fastopen_cookie *foc = opts->fastopen_cookie; in tcp_options_write()
692 if (foc->exp) { in tcp_options_write()
693 len = TCPOLEN_EXP_FASTOPEN_BASE + foc->len; in tcp_options_write()
698 len = TCPOLEN_FASTOPEN_BASE + foc->len; in tcp_options_write()
703 memcpy(p, foc->val, foc->len); in tcp_options_write()
705 p[foc->len] = TCPOPT_NOP; in tcp_options_write()
706 p[foc->len + 1] = TCPOPT_NOP; in tcp_options_write()
713 mptcp_options_write(th, ptr, tp, opts); in tcp_options_write()
716 static void smc_set_option(const struct tcp_sock *tp, in smc_set_option() argument
722 if (tp->syn_smc) { in smc_set_option()
724 opts->options |= OPTION_SMC; in smc_set_option()
725 *remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED; in smc_set_option()
732 static void smc_set_option_cond(const struct tcp_sock *tp, in smc_set_option_cond() argument
739 if (tp->syn_smc && ireq->smc_ok) { in smc_set_option_cond()
741 opts->options |= OPTION_SMC; in smc_set_option_cond()
742 *remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED; in smc_set_option_cond()
756 if (mptcp_synack_options(req, &size, &opts->mptcp)) { in mptcp_set_option_cond()
758 opts->options |= OPTION_MPTCP; in mptcp_set_option_cond()
759 *remaining -= size; in mptcp_set_option_cond()
772 struct tcp_sock *tp = tcp_sk(sk); in tcp_syn_options() local
774 struct tcp_fastopen_request *fastopen = tp->fastopen_req; in tcp_syn_options()
779 rcu_access_pointer(tp->md5sig_info)) { in tcp_syn_options()
780 *md5 = tp->af_specific->md5_lookup(sk, sk); in tcp_syn_options()
782 opts->options |= OPTION_MD5; in tcp_syn_options()
783 remaining -= TCPOLEN_MD5SIG_ALIGNED; in tcp_syn_options()
790 * advertised. But we subtract them from tp->mss_cache so that in tcp_syn_options()
797 opts->mss = tcp_advertise_mss(sk); in tcp_syn_options()
798 remaining -= TCPOLEN_MSS_ALIGNED; in tcp_syn_options()
800 if (likely(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_timestamps) && !*md5)) { in tcp_syn_options()
801 opts->options |= OPTION_TS; in tcp_syn_options()
802 opts->tsval = tcp_skb_timestamp(skb) + tp->tsoffset; in tcp_syn_options()
803 opts->tsecr = tp->rx_opt.ts_recent; in tcp_syn_options()
804 remaining -= TCPOLEN_TSTAMP_ALIGNED; in tcp_syn_options()
806 if (likely(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_window_scaling))) { in tcp_syn_options()
807 opts->ws = tp->rx_opt.rcv_wscale; in tcp_syn_options()
808 opts->options |= OPTION_WSCALE; in tcp_syn_options()
809 remaining -= TCPOLEN_WSCALE_ALIGNED; in tcp_syn_options()
811 if (likely(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_sack))) { in tcp_syn_options()
812 opts->options |= OPTION_SACK_ADVERTISE; in tcp_syn_options()
813 if (unlikely(!(OPTION_TS & opts->options))) in tcp_syn_options()
814 remaining -= TCPOLEN_SACKPERM_ALIGNED; in tcp_syn_options()
817 if (fastopen && fastopen->cookie.len >= 0) { in tcp_syn_options()
818 u32 need = fastopen->cookie.len; in tcp_syn_options()
820 need += fastopen->cookie.exp ? TCPOLEN_EXP_FASTOPEN_BASE : in tcp_syn_options()
824 opts->options |= OPTION_FAST_OPEN_COOKIE; in tcp_syn_options()
825 opts->fastopen_cookie = &fastopen->cookie; in tcp_syn_options()
826 remaining -= need; in tcp_syn_options()
827 tp->syn_fastopen = 1; in tcp_syn_options()
828 tp->syn_fastopen_exp = fastopen->cookie.exp ? 1 : 0; in tcp_syn_options()
832 smc_set_option(tp, opts, &remaining); in tcp_syn_options()
837 if (mptcp_syn_options(sk, skb, &size, &opts->mptcp)) { in tcp_syn_options()
838 opts->options |= OPTION_MPTCP; in tcp_syn_options()
839 remaining -= size; in tcp_syn_options()
845 return MAX_TCP_OPTION_SPACE - remaining; in tcp_syn_options()
848 /* Set up TCP options for SYN-ACKs. */
863 opts->options |= OPTION_MD5; in tcp_synack_options()
864 remaining -= TCPOLEN_MD5SIG_ALIGNED; in tcp_synack_options()
872 ireq->tstamp_ok &= !ireq->sack_ok; in tcp_synack_options()
877 opts->mss = mss; in tcp_synack_options()
878 remaining -= TCPOLEN_MSS_ALIGNED; in tcp_synack_options()
880 if (likely(ireq->wscale_ok)) { in tcp_synack_options()
881 opts->ws = ireq->rcv_wscale; in tcp_synack_options()
882 opts->options |= OPTION_WSCALE; in tcp_synack_options()
883 remaining -= TCPOLEN_WSCALE_ALIGNED; in tcp_synack_options()
885 if (likely(ireq->tstamp_ok)) { in tcp_synack_options()
886 opts->options |= OPTION_TS; in tcp_synack_options()
887 opts->tsval = tcp_skb_timestamp(skb) + tcp_rsk(req)->ts_off; in tcp_synack_options()
888 opts->tsecr = READ_ONCE(req->ts_recent); in tcp_synack_options()
889 remaining -= TCPOLEN_TSTAMP_ALIGNED; in tcp_synack_options()
891 if (likely(ireq->sack_ok)) { in tcp_synack_options()
892 opts->options |= OPTION_SACK_ADVERTISE; in tcp_synack_options()
893 if (unlikely(!ireq->tstamp_ok)) in tcp_synack_options()
894 remaining -= TCPOLEN_SACKPERM_ALIGNED; in tcp_synack_options()
896 if (foc != NULL && foc->len >= 0) { in tcp_synack_options()
897 u32 need = foc->len; in tcp_synack_options()
899 need += foc->exp ? TCPOLEN_EXP_FASTOPEN_BASE : in tcp_synack_options()
903 opts->options |= OPTION_FAST_OPEN_COOKIE; in tcp_synack_options()
904 opts->fastopen_cookie = foc; in tcp_synack_options()
905 remaining -= need; in tcp_synack_options()
916 return MAX_TCP_OPTION_SPACE - remaining; in tcp_synack_options()
926 struct tcp_sock *tp = tcp_sk(sk); in tcp_established_options() local
930 opts->options = 0; in tcp_established_options()
935 rcu_access_pointer(tp->md5sig_info)) { in tcp_established_options()
936 *md5 = tp->af_specific->md5_lookup(sk, sk); in tcp_established_options()
938 opts->options |= OPTION_MD5; in tcp_established_options()
944 if (likely(tp->rx_opt.tstamp_ok)) { in tcp_established_options()
945 opts->options |= OPTION_TS; in tcp_established_options()
946 opts->tsval = skb ? tcp_skb_timestamp(skb) + tp->tsoffset : 0; in tcp_established_options()
947 opts->tsecr = tp->rx_opt.ts_recent; in tcp_established_options()
958 unsigned int remaining = MAX_TCP_OPTION_SPACE - size; in tcp_established_options()
962 &opts->mptcp)) { in tcp_established_options()
963 opts->options |= OPTION_MPTCP; in tcp_established_options()
968 eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack; in tcp_established_options()
970 const unsigned int remaining = MAX_TCP_OPTION_SPACE - size; in tcp_established_options()
975 opts->num_sack_blocks = in tcp_established_options()
977 (remaining - TCPOLEN_SACK_BASE_ALIGNED) / in tcp_established_options()
981 opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK; in tcp_established_options()
984 if (unlikely(BPF_SOCK_OPS_TEST_FLAG(tp, in tcp_established_options()
986 unsigned int remaining = MAX_TCP_OPTION_SPACE - size; in tcp_established_options()
990 size = MAX_TCP_OPTION_SPACE - remaining; in tcp_established_options()
1005 * The invariant being skb->truesize subtracted from sk->sk_wmem_alloc
1019 if ((1 << sk->sk_state) & in tcp_tsq_write()
1022 struct tcp_sock *tp = tcp_sk(sk); in tcp_tsq_write() local
1024 if (tp->lost_out > tp->retrans_out && in tcp_tsq_write()
1025 tcp_snd_cwnd(tp) > tcp_packets_in_flight(tp)) { in tcp_tsq_write()
1026 tcp_mstamp_refresh(tp); in tcp_tsq_write()
1030 tcp_write_xmit(sk, tcp_current_mss(sk), tp->nonagle, in tcp_tsq_write()
1040 else if (!test_and_set_bit(TCP_TSQ_DEFERRED, &sk->sk_tsq_flags)) in tcp_tsq_handler()
1047 * transferring tsq->head because tcp_wfree() might
1056 struct tcp_sock *tp; in tcp_tasklet_func() local
1060 list_splice_init(&tsq->head, &list); in tcp_tasklet_func()
1064 tp = list_entry(q, struct tcp_sock, tsq_node); in tcp_tasklet_func()
1065 list_del(&tp->tsq_node); in tcp_tasklet_func()
1067 sk = (struct sock *)tp; in tcp_tasklet_func()
1069 clear_bit(TSQ_QUEUED, &sk->sk_tsq_flags); in tcp_tasklet_func()
1081 * tcp_release_cb - tcp release_sock() callback
1089 unsigned long flags = smp_load_acquire(&sk->sk_tsq_flags); in tcp_release_cb()
1097 } while (!try_cmpxchg(&sk->sk_tsq_flags, &flags, nflags)); in tcp_release_cb()
1107 * 3) socket owned by us (sk->sk_lock.owned == 1) in tcp_release_cb()
1123 inet_csk(sk)->icsk_af_ops->mtu_reduced(sk); in tcp_release_cb()
1136 INIT_LIST_HEAD(&tsq->head); in tcp_tasklet_init()
1137 tasklet_setup(&tsq->tasklet, tcp_tasklet_func); in tcp_tasklet_init()
1148 struct sock *sk = skb->sk; in tcp_wfree()
1149 struct tcp_sock *tp = tcp_sk(sk); in tcp_wfree() local
1157 WARN_ON(refcount_sub_and_test(skb->truesize - 1, &sk->sk_wmem_alloc)); in tcp_wfree()
1162 * - less callbacks to tcp_write_xmit(), reducing stress (batches) in tcp_wfree()
1163 * - chance for incoming ACK (processed by another cpu maybe) in tcp_wfree()
1164 * to migrate this flow (skb->ooo_okay will be eventually set) in tcp_wfree()
1166 if (refcount_read(&sk->sk_wmem_alloc) >= SKB_TRUESIZE(1) && this_cpu_ksoftirqd() == current) in tcp_wfree()
1169 oval = smp_load_acquire(&sk->sk_tsq_flags); in tcp_wfree()
1175 } while (!try_cmpxchg(&sk->sk_tsq_flags, &oval, nval)); in tcp_wfree()
1180 empty = list_empty(&tsq->head); in tcp_wfree()
1181 list_add(&tp->tsq_node, &tsq->head); in tcp_wfree()
1183 tasklet_schedule(&tsq->tasklet); in tcp_wfree()
1195 struct tcp_sock *tp = container_of(timer, struct tcp_sock, pacing_timer); in tcp_pace_kick() local
1196 struct sock *sk = (struct sock *)tp; in tcp_pace_kick()
1207 struct tcp_sock *tp = tcp_sk(sk); in tcp_update_skb_after_send() local
1209 if (sk->sk_pacing_status != SK_PACING_NONE) { in tcp_update_skb_after_send()
1210 unsigned long rate = sk->sk_pacing_rate; in tcp_update_skb_after_send()
1213 * Note that tp->data_segs_out overflows after 2^32 packets, in tcp_update_skb_after_send()
1216 if (rate != ~0UL && rate && tp->data_segs_out >= 10) { in tcp_update_skb_after_send()
1217 u64 len_ns = div64_ul((u64)skb->len * NSEC_PER_SEC, rate); in tcp_update_skb_after_send()
1218 u64 credit = tp->tcp_wstamp_ns - prior_wstamp; in tcp_update_skb_after_send()
1221 len_ns -= min_t(u64, len_ns / 2, credit); in tcp_update_skb_after_send()
1222 tp->tcp_wstamp_ns += len_ns; in tcp_update_skb_after_send()
1225 list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue); in tcp_update_skb_after_send()
1248 struct tcp_sock *tp; in __tcp_transmit_skb() local
1259 tp = tcp_sk(sk); in __tcp_transmit_skb()
1260 prior_wstamp = tp->tcp_wstamp_ns; in __tcp_transmit_skb()
1261 tp->tcp_wstamp_ns = max(tp->tcp_wstamp_ns, tp->tcp_clock_cache); in __tcp_transmit_skb()
1262 skb_set_delivery_time(skb, tp->tcp_wstamp_ns, true); in __tcp_transmit_skb()
1274 return -ENOBUFS; in __tcp_transmit_skb()
1275 /* retransmit skbs might have a non zero value in skb->dev in __tcp_transmit_skb()
1276 * because skb->dev is aliased with skb->rbnode.rb_left in __tcp_transmit_skb()
1278 skb->dev = NULL; in __tcp_transmit_skb()
1285 if (unlikely(tcb->tcp_flags & TCPHDR_SYN)) { in __tcp_transmit_skb()
1294 * and in this case it is better to delay the delivery of 1-MSS in __tcp_transmit_skb()
1299 tcb->tcp_flags |= TCPHDR_PSH; in __tcp_transmit_skb()
1303 /* We set skb->ooo_okay to one if this packet can select in __tcp_transmit_skb()
1307 * if XPS is enabled, or sk->sk_txhash otherwise. in __tcp_transmit_skb()
1316 skb->ooo_okay = sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1) || in __tcp_transmit_skb()
1324 skb->pfmemalloc = 0; in __tcp_transmit_skb()
1330 skb->sk = sk; in __tcp_transmit_skb()
1331 skb->destructor = skb_is_tcp_pure_ack(skb) ? __sock_wfree : tcp_wfree; in __tcp_transmit_skb()
1332 refcount_add(skb->truesize, &sk->sk_wmem_alloc); in __tcp_transmit_skb()
1334 skb_set_dst_pending_confirm(skb, READ_ONCE(sk->sk_dst_pending_confirm)); in __tcp_transmit_skb()
1337 th = (struct tcphdr *)skb->data; in __tcp_transmit_skb()
1338 th->source = inet->inet_sport; in __tcp_transmit_skb()
1339 th->dest = inet->inet_dport; in __tcp_transmit_skb()
1340 th->seq = htonl(tcb->seq); in __tcp_transmit_skb()
1341 th->ack_seq = htonl(rcv_nxt); in __tcp_transmit_skb()
1343 tcb->tcp_flags); in __tcp_transmit_skb()
1345 th->check = 0; in __tcp_transmit_skb()
1346 th->urg_ptr = 0; in __tcp_transmit_skb()
1349 if (unlikely(tcp_urg_mode(tp) && before(tcb->seq, tp->snd_up))) { in __tcp_transmit_skb()
1350 if (before(tp->snd_up, tcb->seq + 0x10000)) { in __tcp_transmit_skb()
1351 th->urg_ptr = htons(tp->snd_up - tcb->seq); in __tcp_transmit_skb()
1352 th->urg = 1; in __tcp_transmit_skb()
1353 } else if (after(tcb->seq + 0xFFFF, tp->snd_nxt)) { in __tcp_transmit_skb()
1354 th->urg_ptr = htons(0xFFFF); in __tcp_transmit_skb()
1355 th->urg = 1; in __tcp_transmit_skb()
1359 skb_shinfo(skb)->gso_type = sk->sk_gso_type; in __tcp_transmit_skb()
1360 if (likely(!(tcb->tcp_flags & TCPHDR_SYN))) { in __tcp_transmit_skb()
1361 th->window = htons(tcp_select_window(sk)); in __tcp_transmit_skb()
1367 th->window = htons(min(tp->rcv_wnd, 65535U)); in __tcp_transmit_skb()
1370 tcp_options_write(th, tp, &opts); in __tcp_transmit_skb()
1376 tp->af_specific->calc_md5_hash(opts.hash_location, in __tcp_transmit_skb()
1384 INDIRECT_CALL_INET(icsk->icsk_af_ops->send_check, in __tcp_transmit_skb()
1388 if (likely(tcb->tcp_flags & TCPHDR_ACK)) in __tcp_transmit_skb()
1391 if (skb->len != tcp_header_size) { in __tcp_transmit_skb()
1392 tcp_event_data_sent(tp, sk); in __tcp_transmit_skb()
1393 tp->data_segs_out += tcp_skb_pcount(skb); in __tcp_transmit_skb()
1394 tp->bytes_sent += skb->len - tcp_header_size; in __tcp_transmit_skb()
1397 if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq) in __tcp_transmit_skb()
1401 tp->segs_out += tcp_skb_pcount(skb); in __tcp_transmit_skb()
1403 /* OK, its time to fill skb_shinfo(skb)->gso_{segs|size} */ in __tcp_transmit_skb()
1404 skb_shinfo(skb)->gso_segs = tcp_skb_pcount(skb); in __tcp_transmit_skb()
1405 skb_shinfo(skb)->gso_size = tcp_skb_mss(skb); in __tcp_transmit_skb()
1407 /* Leave earliest departure time in skb->tstamp (skb->skb_mstamp_ns) */ in __tcp_transmit_skb()
1410 memset(skb->cb, 0, max(sizeof(struct inet_skb_parm), in __tcp_transmit_skb()
1413 tcp_add_tx_delay(skb, tp); in __tcp_transmit_skb()
1415 err = INDIRECT_CALL_INET(icsk->icsk_af_ops->queue_xmit, in __tcp_transmit_skb()
1417 sk, skb, &inet->cork.fl); in __tcp_transmit_skb()
1434 tcp_sk(sk)->rcv_nxt); in tcp_transmit_skb()
1444 struct tcp_sock *tp = tcp_sk(sk); in tcp_queue_skb() local
1447 WRITE_ONCE(tp->write_seq, TCP_SKB_CB(skb)->end_seq); in tcp_queue_skb()
1450 sk_wmem_queued_add(sk, skb->truesize); in tcp_queue_skb()
1451 sk_mem_charge(sk, skb->truesize); in tcp_queue_skb()
1457 if (skb->len <= mss_now) { in tcp_set_skb_tso_segs()
1459 * non-TSO case. in tcp_set_skb_tso_segs()
1462 TCP_SKB_CB(skb)->tcp_gso_size = 0; in tcp_set_skb_tso_segs()
1464 tcp_skb_pcount_set(skb, DIV_ROUND_UP(skb->len, mss_now)); in tcp_set_skb_tso_segs()
1465 TCP_SKB_CB(skb)->tcp_gso_size = mss_now; in tcp_set_skb_tso_segs()
1474 struct tcp_sock *tp = tcp_sk(sk); in tcp_adjust_pcount() local
1476 tp->packets_out -= decr; in tcp_adjust_pcount()
1478 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) in tcp_adjust_pcount()
1479 tp->sacked_out -= decr; in tcp_adjust_pcount()
1480 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) in tcp_adjust_pcount()
1481 tp->retrans_out -= decr; in tcp_adjust_pcount()
1482 if (TCP_SKB_CB(skb)->sacked & TCPCB_LOST) in tcp_adjust_pcount()
1483 tp->lost_out -= decr; in tcp_adjust_pcount()
1486 if (tcp_is_reno(tp) && decr > 0) in tcp_adjust_pcount()
1487 tp->sacked_out -= min_t(u32, tp->sacked_out, decr); in tcp_adjust_pcount()
1489 if (tp->lost_skb_hint && in tcp_adjust_pcount()
1490 before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(tp->lost_skb_hint)->seq) && in tcp_adjust_pcount()
1491 (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) in tcp_adjust_pcount()
1492 tp->lost_cnt_hint -= decr; in tcp_adjust_pcount()
1494 tcp_verify_left_out(tp); in tcp_adjust_pcount()
1499 return TCP_SKB_CB(skb)->txstamp_ack || in tcp_has_tx_tstamp()
1500 (skb_shinfo(skb)->tx_flags & SKBTX_ANY_TSTAMP); in tcp_has_tx_tstamp()
1508 !before(shinfo->tskey, TCP_SKB_CB(skb2)->seq)) { in tcp_fragment_tstamp()
1510 u8 tsflags = shinfo->tx_flags & SKBTX_ANY_TSTAMP; in tcp_fragment_tstamp()
1512 shinfo->tx_flags &= ~tsflags; in tcp_fragment_tstamp()
1513 shinfo2->tx_flags |= tsflags; in tcp_fragment_tstamp()
1514 swap(shinfo->tskey, shinfo2->tskey); in tcp_fragment_tstamp()
1515 TCP_SKB_CB(skb2)->txstamp_ack = TCP_SKB_CB(skb)->txstamp_ack; in tcp_fragment_tstamp()
1516 TCP_SKB_CB(skb)->txstamp_ack = 0; in tcp_fragment_tstamp()
1522 TCP_SKB_CB(skb2)->eor = TCP_SKB_CB(skb)->eor; in tcp_skb_fragment_eor()
1523 TCP_SKB_CB(skb)->eor = 0; in tcp_skb_fragment_eor()
1533 __skb_queue_after(&sk->sk_write_queue, skb, buff); in tcp_insert_write_queue_after()
1535 tcp_rbtree_insert(&sk->tcp_rtx_queue, buff); in tcp_insert_write_queue_after()
1547 struct tcp_sock *tp = tcp_sk(sk); in tcp_fragment() local
1554 if (WARN_ON(len > skb->len)) in tcp_fragment()
1555 return -EINVAL; in tcp_fragment()
1564 limit = sk->sk_sndbuf + 2 * SKB_TRUESIZE(GSO_LEGACY_MAX_SIZE); in tcp_fragment()
1565 if (unlikely((sk->sk_wmem_queued >> 1) > limit && in tcp_fragment()
1570 return -ENOMEM; in tcp_fragment()
1574 return -ENOMEM; in tcp_fragment()
1579 return -ENOMEM; /* We'll just try again later. */ in tcp_fragment()
1583 sk_wmem_queued_add(sk, buff->truesize); in tcp_fragment()
1584 sk_mem_charge(sk, buff->truesize); in tcp_fragment()
1585 nlen = skb->len - len; in tcp_fragment()
1586 buff->truesize += nlen; in tcp_fragment()
1587 skb->truesize -= nlen; in tcp_fragment()
1590 TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len; in tcp_fragment()
1591 TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq; in tcp_fragment()
1592 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq; in tcp_fragment()
1595 flags = TCP_SKB_CB(skb)->tcp_flags; in tcp_fragment()
1596 TCP_SKB_CB(skb)->tcp_flags = flags & ~(TCPHDR_FIN | TCPHDR_PSH); in tcp_fragment()
1597 TCP_SKB_CB(buff)->tcp_flags = flags; in tcp_fragment()
1598 TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked; in tcp_fragment()
1603 skb_set_delivery_time(buff, skb->tstamp, true); in tcp_fragment()
1613 TCP_SKB_CB(buff)->tx = TCP_SKB_CB(skb)->tx; in tcp_fragment()
1616 * adjust the various packet counters. in tcp_fragment()
1618 if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) { in tcp_fragment()
1619 int diff = old_factor - tcp_skb_pcount(skb) - in tcp_fragment()
1630 list_add(&buff->tcp_tsorted_anchor, &skb->tcp_tsorted_anchor); in tcp_fragment()
1647 for (i = 0; i < shinfo->nr_frags; i++) { in __pskb_trim_head()
1648 int size = skb_frag_size(&shinfo->frags[i]); in __pskb_trim_head()
1652 eat -= size; in __pskb_trim_head()
1654 shinfo->frags[k] = shinfo->frags[i]; in __pskb_trim_head()
1656 skb_frag_off_add(&shinfo->frags[k], eat); in __pskb_trim_head()
1657 skb_frag_size_sub(&shinfo->frags[k], eat); in __pskb_trim_head()
1663 shinfo->nr_frags = k; in __pskb_trim_head()
1665 skb->data_len -= len; in __pskb_trim_head()
1666 skb->len = skb->data_len; in __pskb_trim_head()
1676 return -ENOMEM; in tcp_trim_head()
1680 TCP_SKB_CB(skb)->seq += len; in tcp_trim_head()
1682 skb->truesize -= delta_truesize; in tcp_trim_head()
1683 sk_wmem_queued_add(sk, -delta_truesize); in tcp_trim_head()
1687 /* Any change of skb->len requires recalculation of tso factor. */ in tcp_trim_head()
1697 const struct tcp_sock *tp = tcp_sk(sk); in __tcp_mtu_to_mss() local
1702 It is MMS_S - sizeof(tcphdr) of rfc1122 in __tcp_mtu_to_mss()
1704 mss_now = pmtu - icsk->icsk_af_ops->net_header_len - sizeof(struct tcphdr); in __tcp_mtu_to_mss()
1707 if (icsk->icsk_af_ops->net_frag_header_len) { in __tcp_mtu_to_mss()
1711 mss_now -= icsk->icsk_af_ops->net_frag_header_len; in __tcp_mtu_to_mss()
1715 if (mss_now > tp->rx_opt.mss_clamp) in __tcp_mtu_to_mss()
1716 mss_now = tp->rx_opt.mss_clamp; in __tcp_mtu_to_mss()
1719 mss_now -= icsk->icsk_ext_hdr_len; in __tcp_mtu_to_mss()
1723 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_snd_mss)); in __tcp_mtu_to_mss()
1731 return __tcp_mtu_to_mss(sk, pmtu) - in tcp_mtu_to_mss()
1732 (tcp_sk(sk)->tcp_header_len - sizeof(struct tcphdr)); in tcp_mtu_to_mss()
1739 const struct tcp_sock *tp = tcp_sk(sk); in tcp_mss_to_mtu() local
1744 tp->tcp_header_len + in tcp_mss_to_mtu()
1745 icsk->icsk_ext_hdr_len + in tcp_mss_to_mtu()
1746 icsk->icsk_af_ops->net_header_len; in tcp_mss_to_mtu()
1749 if (icsk->icsk_af_ops->net_frag_header_len) { in tcp_mss_to_mtu()
1753 mtu += icsk->icsk_af_ops->net_frag_header_len; in tcp_mss_to_mtu()
1762 struct tcp_sock *tp = tcp_sk(sk); in tcp_mtup_init() local
1766 icsk->icsk_mtup.enabled = READ_ONCE(net->ipv4.sysctl_tcp_mtu_probing) > 1; in tcp_mtup_init()
1767 icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) + in tcp_mtup_init()
1768 icsk->icsk_af_ops->net_header_len; in tcp_mtup_init()
1769 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, READ_ONCE(net->ipv4.sysctl_tcp_base_mss)); in tcp_mtup_init()
1770 icsk->icsk_mtup.probe_size = 0; in tcp_mtup_init()
1771 if (icsk->icsk_mtup.enabled) in tcp_mtup_init()
1772 icsk->icsk_mtup.probe_timestamp = tcp_jiffies32; in tcp_mtup_init()
1778 tp->rx_opt.user_mss is mss set by user by TCP_MAXSEG. It does NOT counts
1781 tp->rx_opt.mss_clamp is mss negotiated at connection setup.
1785 inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
1787 tp->mss_cache is current effective sending mss, including
1790 tp->rx_opt.mss_clamp.
1795 NOTE2. inet_csk(sk)->icsk_pmtu_cookie and tp->mss_cache
1796 are READ ONLY outside this function. --ANK (980731)
1800 struct tcp_sock *tp = tcp_sk(sk); in tcp_sync_mss() local
1804 if (icsk->icsk_mtup.search_high > pmtu) in tcp_sync_mss()
1805 icsk->icsk_mtup.search_high = pmtu; in tcp_sync_mss()
1808 mss_now = tcp_bound_to_half_wnd(tp, mss_now); in tcp_sync_mss()
1811 icsk->icsk_pmtu_cookie = pmtu; in tcp_sync_mss()
1812 if (icsk->icsk_mtup.enabled) in tcp_sync_mss()
1813 mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low)); in tcp_sync_mss()
1814 tp->mss_cache = mss_now; in tcp_sync_mss()
1825 const struct tcp_sock *tp = tcp_sk(sk); in tcp_current_mss() local
1832 mss_now = tp->mss_cache; in tcp_current_mss()
1836 if (mtu != inet_csk(sk)->icsk_pmtu_cookie) in tcp_current_mss()
1842 /* The mss_cache is sized based on tp->tcp_header_len, which assumes in tcp_current_mss()
1845 * we have to adjust mss_now correspondingly */ in tcp_current_mss()
1846 if (header_len != tp->tcp_header_len) { in tcp_current_mss()
1847 int delta = (int) header_len - tp->tcp_header_len; in tcp_current_mss()
1848 mss_now -= delta; in tcp_current_mss()
1854 /* RFC2861, slow part. Adjust cwnd, after it was not full during one rto.
1860 struct tcp_sock *tp = tcp_sk(sk); in tcp_cwnd_application_limited() local
1862 if (inet_csk(sk)->icsk_ca_state == TCP_CA_Open && in tcp_cwnd_application_limited()
1863 sk->sk_socket && !test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { in tcp_cwnd_application_limited()
1865 u32 init_win = tcp_init_cwnd(tp, __sk_dst_get(sk)); in tcp_cwnd_application_limited()
1866 u32 win_used = max(tp->snd_cwnd_used, init_win); in tcp_cwnd_application_limited()
1867 if (win_used < tcp_snd_cwnd(tp)) { in tcp_cwnd_application_limited()
1868 tp->snd_ssthresh = tcp_current_ssthresh(sk); in tcp_cwnd_application_limited()
1869 tcp_snd_cwnd_set(tp, (tcp_snd_cwnd(tp) + win_used) >> 1); in tcp_cwnd_application_limited()
1871 tp->snd_cwnd_used = 0; in tcp_cwnd_application_limited()
1873 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_cwnd_application_limited()
1878 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; in tcp_cwnd_validate()
1879 struct tcp_sock *tp = tcp_sk(sk); in tcp_cwnd_validate() local
1882 * is fully utilized. If cwnd-limited then remember that fact for the in tcp_cwnd_validate()
1883 * current window. If not cwnd-limited then track the maximum number of in tcp_cwnd_validate()
1884 * outstanding packets in the current window. (If cwnd-limited then we in tcp_cwnd_validate()
1885 * chose to not update tp->max_packets_out to avoid an extra else in tcp_cwnd_validate()
1888 if (!before(tp->snd_una, tp->cwnd_usage_seq) || in tcp_cwnd_validate()
1890 (!tp->is_cwnd_limited && in tcp_cwnd_validate()
1891 tp->packets_out > tp->max_packets_out)) { in tcp_cwnd_validate()
1892 tp->is_cwnd_limited = is_cwnd_limited; in tcp_cwnd_validate()
1893 tp->max_packets_out = tp->packets_out; in tcp_cwnd_validate()
1894 tp->cwnd_usage_seq = tp->snd_nxt; in tcp_cwnd_validate()
1899 tp->snd_cwnd_used = 0; in tcp_cwnd_validate()
1900 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_cwnd_validate()
1903 if (tp->packets_out > tp->snd_cwnd_used) in tcp_cwnd_validate()
1904 tp->snd_cwnd_used = tp->packets_out; in tcp_cwnd_validate()
1906 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) && in tcp_cwnd_validate()
1907 (s32)(tcp_jiffies32 - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto && in tcp_cwnd_validate()
1908 !ca_ops->cong_control) in tcp_cwnd_validate()
1918 if (tcp_write_queue_empty(sk) && sk->sk_socket && in tcp_cwnd_validate()
1919 test_bit(SOCK_NOSPACE, &sk->sk_socket->flags) && in tcp_cwnd_validate()
1920 (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) in tcp_cwnd_validate()
1926 static bool tcp_minshall_check(const struct tcp_sock *tp) in tcp_minshall_check() argument
1928 return after(tp->snd_sml, tp->snd_una) && in tcp_minshall_check()
1929 !after(tp->snd_sml, tp->snd_nxt); in tcp_minshall_check()
1933 * Note that a TSO packet might end with a sub-mss segment
1935 * if ((skb->len % mss) != 0)
1936 * tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1938 * skb_pcount = skb->len / mss_now
1940 static void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss_now, in tcp_minshall_update() argument
1943 if (skb->len < tcp_skb_pcount(skb) * mss_now) in tcp_minshall_update()
1944 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; in tcp_minshall_update()
1954 static bool tcp_nagle_check(bool partial, const struct tcp_sock *tp, in tcp_nagle_check() argument
1959 (!nonagle && tp->packets_out && tcp_minshall_check(tp))); in tcp_nagle_check()
1966 * - For close peers, we rather send bigger packets to reduce
1968 * - For long distance/rtt flows, we would like to get ACK clocking
1972 * in bigger TSO bursts. We we cut the RTT-based allowance in half
1973 * for every 2^9 usec (aka 512 us) of RTT, so that the RTT-based allowance
1982 bytes = sk->sk_pacing_rate >> READ_ONCE(sk->sk_pacing_shift); in tcp_tso_autosize()
1984 r = tcp_min_rtt(tcp_sk(sk)) >> READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tso_rtt_log); in tcp_tso_autosize()
1985 if (r < BITS_PER_TYPE(sk->sk_gso_max_size)) in tcp_tso_autosize()
1986 bytes += sk->sk_gso_max_size >> r; in tcp_tso_autosize()
1988 bytes = min_t(unsigned long, bytes, sk->sk_gso_max_size); in tcp_tso_autosize()
1998 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; in tcp_tso_segs()
2001 min_tso = ca_ops->min_tso_segs ? in tcp_tso_segs()
2002 ca_ops->min_tso_segs(sk) : in tcp_tso_segs()
2003 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs); in tcp_tso_segs()
2006 return min_t(u32, tso_segs, sk->sk_gso_max_segs); in tcp_tso_segs()
2016 const struct tcp_sock *tp = tcp_sk(sk); in tcp_mss_split_point() local
2019 window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in tcp_mss_split_point()
2025 needed = min(skb->len, window); in tcp_mss_split_point()
2035 if (tcp_nagle_check(partial != 0, tp, nonagle)) in tcp_mss_split_point()
2036 return needed - partial; in tcp_mss_split_point()
2044 static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp, in tcp_cwnd_test() argument
2050 if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) && in tcp_cwnd_test()
2054 in_flight = tcp_packets_in_flight(tp); in tcp_cwnd_test()
2055 cwnd = tcp_snd_cwnd(tp); in tcp_cwnd_test()
2063 return min(halfcwnd, cwnd - in_flight); in tcp_cwnd_test()
2085 static inline bool tcp_nagle_test(const struct tcp_sock *tp, const struct sk_buff *skb, in tcp_nagle_test() argument
2098 if (tcp_urg_mode(tp) || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) in tcp_nagle_test()
2101 if (!tcp_nagle_check(skb->len < cur_mss, tp, nonagle)) in tcp_nagle_test()
2108 static bool tcp_snd_wnd_test(const struct tcp_sock *tp, in tcp_snd_wnd_test() argument
2112 u32 end_seq = TCP_SKB_CB(skb)->end_seq; in tcp_snd_wnd_test()
2114 if (skb->len > cur_mss) in tcp_snd_wnd_test()
2115 end_seq = TCP_SKB_CB(skb)->seq + cur_mss; in tcp_snd_wnd_test()
2117 return !after(end_seq, tcp_wnd_end(tp)); in tcp_snd_wnd_test()
2124 * know that all the data is in scatter-gather pages, and that the
2130 int nlen = skb->len - len; in tso_fragment()
2135 DEBUG_NET_WARN_ON_ONCE(skb->len != skb->data_len); in tso_fragment()
2139 return -ENOMEM; in tso_fragment()
2143 sk_wmem_queued_add(sk, buff->truesize); in tso_fragment()
2144 sk_mem_charge(sk, buff->truesize); in tso_fragment()
2145 buff->truesize += nlen; in tso_fragment()
2146 skb->truesize -= nlen; in tso_fragment()
2149 TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len; in tso_fragment()
2150 TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq; in tso_fragment()
2151 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq; in tso_fragment()
2154 flags = TCP_SKB_CB(skb)->tcp_flags; in tso_fragment()
2155 TCP_SKB_CB(skb)->tcp_flags = flags & ~(TCPHDR_FIN | TCPHDR_PSH); in tso_fragment()
2156 TCP_SKB_CB(buff)->tcp_flags = flags; in tso_fragment()
2186 struct tcp_sock *tp = tcp_sk(sk); in tcp_tso_should_defer() local
2191 if (icsk->icsk_ca_state >= TCP_CA_Recovery) in tcp_tso_should_defer()
2196 * Note that tp->tcp_wstamp_ns can be in the future if we have in tcp_tso_should_defer()
2199 delta = tp->tcp_clock_cache - tp->tcp_wstamp_ns - NSEC_PER_MSEC; in tcp_tso_should_defer()
2203 in_flight = tcp_packets_in_flight(tp); in tcp_tso_should_defer()
2206 BUG_ON(tcp_snd_cwnd(tp) <= in_flight); in tcp_tso_should_defer()
2208 send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in tcp_tso_should_defer()
2211 cong_win = (tcp_snd_cwnd(tp) - in_flight) * tp->mss_cache; in tcp_tso_should_defer()
2215 /* If a full-sized TSO skb can be sent, do it. */ in tcp_tso_should_defer()
2216 if (limit >= max_segs * tp->mss_cache) in tcp_tso_should_defer()
2220 if ((skb != tcp_write_queue_tail(sk)) && (limit >= skb->len)) in tcp_tso_should_defer()
2223 win_divisor = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tso_win_divisor); in tcp_tso_should_defer()
2225 u32 chunk = min(tp->snd_wnd, tcp_snd_cwnd(tp) * tp->mss_cache); in tcp_tso_should_defer()
2239 if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache) in tcp_tso_should_defer()
2247 delta = tp->tcp_clock_cache - head->tstamp; in tcp_tso_should_defer()
2249 if ((s64)(delta - (u64)NSEC_PER_USEC * (tp->srtt_us >> 4)) < 0) in tcp_tso_should_defer()
2254 * 1) We are cwnd-limited in tcp_tso_should_defer()
2255 * 2) We are rwnd-limited in tcp_tso_should_defer()
2259 if (cong_win <= skb->len) { in tcp_tso_should_defer()
2264 if (send_win <= skb->len) { in tcp_tso_should_defer()
2271 if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) || in tcp_tso_should_defer()
2272 TCP_SKB_CB(skb)->eor) in tcp_tso_should_defer()
2284 struct tcp_sock *tp = tcp_sk(sk); in tcp_mtu_check_reprobe() local
2289 interval = READ_ONCE(net->ipv4.sysctl_tcp_probe_interval); in tcp_mtu_check_reprobe()
2290 delta = tcp_jiffies32 - icsk->icsk_mtup.probe_timestamp; in tcp_mtu_check_reprobe()
2295 icsk->icsk_mtup.probe_size = 0; in tcp_mtu_check_reprobe()
2296 icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + in tcp_mtu_check_reprobe()
2298 icsk->icsk_af_ops->net_header_len; in tcp_mtu_check_reprobe()
2299 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss); in tcp_mtu_check_reprobe()
2302 icsk->icsk_mtup.probe_timestamp = tcp_jiffies32; in tcp_mtu_check_reprobe()
2312 if (len <= skb->len) in tcp_can_coalesce_send_queue_head()
2315 if (unlikely(TCP_SKB_CB(skb)->eor) || in tcp_can_coalesce_send_queue_head()
2320 len -= skb->len; in tcp_can_coalesce_send_queue_head()
2329 skb_frag_t *lastfrag = NULL, *fragto = skb_shinfo(to)->frags; in tcp_clone_payload()
2333 if (!sk_wmem_schedule(sk, to->truesize + probe_size)) in tcp_clone_payload()
2334 return -ENOMEM; in tcp_clone_payload()
2336 skb_queue_walk(&sk->sk_write_queue, skb) { in tcp_clone_payload()
2337 const skb_frag_t *fragfrom = skb_shinfo(skb)->frags; in tcp_clone_payload()
2340 return -EINVAL; in tcp_clone_payload()
2342 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++, fragfrom++) { in tcp_clone_payload()
2346 probe_size - len); in tcp_clone_payload()
2356 return -E2BIG; in tcp_clone_payload()
2369 skb_shinfo(to)->nr_frags = nr_frags; in tcp_clone_payload()
2370 to->truesize += probe_size; in tcp_clone_payload()
2371 to->len += probe_size; in tcp_clone_payload()
2372 to->data_len += probe_size; in tcp_clone_payload()
2384 * -1 otherwise
2389 struct tcp_sock *tp = tcp_sk(sk); in tcp_mtu_probe() local
2403 if (likely(!icsk->icsk_mtup.enabled || in tcp_mtu_probe()
2404 icsk->icsk_mtup.probe_size || in tcp_mtu_probe()
2405 inet_csk(sk)->icsk_ca_state != TCP_CA_Open || in tcp_mtu_probe()
2406 tcp_snd_cwnd(tp) < 11 || in tcp_mtu_probe()
2407 tp->rx_opt.num_sacks || tp->rx_opt.dsack)) in tcp_mtu_probe()
2408 return -1; in tcp_mtu_probe()
2411 * and current mss_clamp. if (search_high - search_low) in tcp_mtu_probe()
2415 probe_size = tcp_mtu_to_mss(sk, (icsk->icsk_mtup.search_high + in tcp_mtu_probe()
2416 icsk->icsk_mtup.search_low) >> 1); in tcp_mtu_probe()
2417 size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache; in tcp_mtu_probe()
2418 interval = icsk->icsk_mtup.search_high - icsk->icsk_mtup.search_low; in tcp_mtu_probe()
2423 if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high) || in tcp_mtu_probe()
2424 interval < READ_ONCE(net->ipv4.sysctl_tcp_probe_threshold)) { in tcp_mtu_probe()
2429 return -1; in tcp_mtu_probe()
2433 if (tp->write_seq - tp->snd_nxt < size_needed) in tcp_mtu_probe()
2434 return -1; in tcp_mtu_probe()
2436 if (tp->snd_wnd < size_needed) in tcp_mtu_probe()
2437 return -1; in tcp_mtu_probe()
2438 if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp))) in tcp_mtu_probe()
2442 if (tcp_packets_in_flight(tp) + 2 > tcp_snd_cwnd(tp)) { in tcp_mtu_probe()
2443 if (!tcp_packets_in_flight(tp)) in tcp_mtu_probe()
2444 return -1; in tcp_mtu_probe()
2450 return -1; in tcp_mtu_probe()
2455 return -1; in tcp_mtu_probe()
2461 return -1; in tcp_mtu_probe()
2463 sk_wmem_queued_add(sk, nskb->truesize); in tcp_mtu_probe()
2464 sk_mem_charge(sk, nskb->truesize); in tcp_mtu_probe()
2470 TCP_SKB_CB(nskb)->seq = TCP_SKB_CB(skb)->seq; in tcp_mtu_probe()
2471 TCP_SKB_CB(nskb)->end_seq = TCP_SKB_CB(skb)->seq + probe_size; in tcp_mtu_probe()
2472 TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK; in tcp_mtu_probe()
2479 copy = min_t(int, skb->len, probe_size - len); in tcp_mtu_probe()
2481 if (skb->len <= copy) { in tcp_mtu_probe()
2484 TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags; in tcp_mtu_probe()
2488 TCP_SKB_CB(nskb)->eor = TCP_SKB_CB(skb)->eor; in tcp_mtu_probe()
2493 TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags & in tcp_mtu_probe()
2497 TCP_SKB_CB(skb)->seq += copy; in tcp_mtu_probe()
2505 tcp_init_tso_segs(nskb, nskb->len); in tcp_mtu_probe()
2508 * be resegmented into mss-sized pieces by tcp_write_xmit(). in tcp_mtu_probe()
2513 tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) - 1); in tcp_mtu_probe()
2516 icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len); in tcp_mtu_probe()
2517 tp->mtu_probe.probe_seq_start = TCP_SKB_CB(nskb)->seq; in tcp_mtu_probe()
2518 tp->mtu_probe.probe_seq_end = TCP_SKB_CB(nskb)->end_seq; in tcp_mtu_probe()
2523 return -1; in tcp_mtu_probe()
2528 struct tcp_sock *tp = tcp_sk(sk); in tcp_pacing_check() local
2533 if (tp->tcp_wstamp_ns <= tp->tcp_clock_cache) in tcp_pacing_check()
2536 if (!hrtimer_is_queued(&tp->pacing_timer)) { in tcp_pacing_check()
2537 hrtimer_start(&tp->pacing_timer, in tcp_pacing_check()
2538 ns_to_ktime(tp->tcp_wstamp_ns), in tcp_pacing_check()
2547 const struct rb_node *node = sk->tcp_rtx_queue.rb_node; in tcp_rtx_queue_empty_or_single_skb()
2554 return !node->rb_left && !node->rb_right; in tcp_rtx_queue_empty_or_single_skb()
2561 * - better RTT estimation and ACK scheduling
2562 * - faster recovery
2563 * - high rates
2574 2 * skb->truesize, in tcp_small_queue_check()
2575 sk->sk_pacing_rate >> READ_ONCE(sk->sk_pacing_shift)); in tcp_small_queue_check()
2576 if (sk->sk_pacing_status == SK_PACING_NONE) in tcp_small_queue_check()
2578 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_limit_output_bytes)); in tcp_small_queue_check()
2582 tcp_sk(sk)->tcp_tx_delay) { in tcp_small_queue_check()
2583 u64 extra_bytes = (u64)sk->sk_pacing_rate * tcp_sk(sk)->tcp_tx_delay; in tcp_small_queue_check()
2586 * approximate our needs assuming an ~100% skb->truesize overhead. in tcp_small_queue_check()
2590 extra_bytes >>= (20 - 1); in tcp_small_queue_check()
2593 if (refcount_read(&sk->sk_wmem_alloc) > limit) { in tcp_small_queue_check()
2602 set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags); in tcp_small_queue_check()
2608 if (refcount_read(&sk->sk_wmem_alloc) > limit) in tcp_small_queue_check()
2614 static void tcp_chrono_set(struct tcp_sock *tp, const enum tcp_chrono new) in tcp_chrono_set() argument
2617 enum tcp_chrono old = tp->chrono_type; in tcp_chrono_set()
2620 tp->chrono_stat[old - 1] += now - tp->chrono_start; in tcp_chrono_set()
2621 tp->chrono_start = now; in tcp_chrono_set()
2622 tp->chrono_type = new; in tcp_chrono_set()
2627 struct tcp_sock *tp = tcp_sk(sk); in tcp_chrono_start() local
2634 if (type > tp->chrono_type) in tcp_chrono_start()
2635 tcp_chrono_set(tp, type); in tcp_chrono_start()
2640 struct tcp_sock *tp = tcp_sk(sk); in tcp_chrono_stop() local
2651 tcp_chrono_set(tp, TCP_CHRONO_UNSPEC); in tcp_chrono_stop()
2652 else if (type == tp->chrono_type) in tcp_chrono_stop()
2653 tcp_chrono_set(tp, TCP_CHRONO_BUSY); in tcp_chrono_stop()
2661 * snd_up-64k-mss .. snd_up cannot be large. However, taking into
2673 struct tcp_sock *tp = tcp_sk(sk); in tcp_write_xmit() local
2683 tcp_mstamp_refresh(tp); in tcp_write_xmit()
2698 if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE) { in tcp_write_xmit()
2700 tp->tcp_wstamp_ns = tp->tcp_clock_cache; in tcp_write_xmit()
2701 skb_set_delivery_time(skb, tp->tcp_wstamp_ns, true); in tcp_write_xmit()
2702 list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue); in tcp_write_xmit()
2713 cwnd_quota = tcp_cwnd_test(tp, skb); in tcp_write_xmit()
2722 if (unlikely(!tcp_snd_wnd_test(tp, skb, mss_now))) { in tcp_write_xmit()
2728 if (unlikely(!tcp_nagle_test(tp, skb, mss_now, in tcp_write_xmit()
2740 if (tso_segs > 1 && !tcp_urg_mode(tp)) in tcp_write_xmit()
2747 if (skb->len > limit && in tcp_write_xmit()
2756 * We do not want to send a pure-ack packet and have in tcp_write_xmit()
2759 if (TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) in tcp_write_xmit()
2771 tcp_minshall_update(tp, mss_now, skb); in tcp_write_xmit()
2783 is_cwnd_limited |= (tcp_packets_in_flight(tp) >= tcp_snd_cwnd(tp)); in tcp_write_xmit()
2789 tp->prr_out += sent_pkts; in tcp_write_xmit()
2796 return !tp->packets_out && !tcp_write_queue_empty(sk); in tcp_write_xmit()
2802 struct tcp_sock *tp = tcp_sk(sk); in tcp_schedule_loss_probe() local
2809 if (rcu_access_pointer(tp->fastopen_rsk)) in tcp_schedule_loss_probe()
2812 early_retrans = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_early_retrans); in tcp_schedule_loss_probe()
2817 !tp->packets_out || !tcp_is_sack(tp) || in tcp_schedule_loss_probe()
2818 (icsk->icsk_ca_state != TCP_CA_Open && in tcp_schedule_loss_probe()
2819 icsk->icsk_ca_state != TCP_CA_CWR)) in tcp_schedule_loss_probe()
2826 if (tp->srtt_us) { in tcp_schedule_loss_probe()
2827 timeout_us = tp->srtt_us >> 2; in tcp_schedule_loss_probe()
2828 if (tp->packets_out == 1) in tcp_schedule_loss_probe()
2839 jiffies_to_usecs(inet_csk(sk)->icsk_rto) : in tcp_schedule_loss_probe()
2856 set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags); in skb_still_in_host_queue()
2872 struct tcp_sock *tp = tcp_sk(sk); in tcp_send_loss_probe() local
2878 if (tp->tlp_high_seq) in tcp_send_loss_probe()
2881 tp->tlp_retrans = 0; in tcp_send_loss_probe()
2883 if (skb && tcp_snd_wnd_test(tp, skb, mss)) { in tcp_send_loss_probe()
2884 pcount = tp->packets_out; in tcp_send_loss_probe()
2886 if (tp->packets_out > pcount) in tcp_send_loss_probe()
2890 skb = skb_rb_last(&sk->tcp_rtx_queue); in tcp_send_loss_probe()
2892 WARN_ONCE(tp->packets_out, in tcp_send_loss_probe()
2894 tp->packets_out, sk->sk_state, tcp_snd_cwnd(tp), mss); in tcp_send_loss_probe()
2895 inet_csk(sk)->icsk_pending = 0; in tcp_send_loss_probe()
2906 if ((pcount > 1) && (skb->len > (pcount - 1) * mss)) { in tcp_send_loss_probe()
2908 (pcount - 1) * mss, mss, in tcp_send_loss_probe()
2920 tp->tlp_retrans = 1; in tcp_send_loss_probe()
2924 tp->tlp_high_seq = tp->snd_nxt; in tcp_send_loss_probe()
2928 inet_csk(sk)->icsk_pending = 0; in tcp_send_loss_probe()
2944 if (unlikely(sk->sk_state == TCP_CLOSE)) in __tcp_push_pending_frames()
2959 BUG_ON(!skb || skb->len < mss_now); in tcp_push_one()
2961 tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation); in tcp_push_one()
2973 * RCV.BUFF - RCV.USER - RCV.WINDOW >= min(1/2 RCV.BUFF, MSS)"
2979 * since header prediction assumes th->window stays fixed.
2981 * Strictly speaking, keeping th->window fixed violates the receiver
3013 * Note, we don't "adjust" for TIMESTAMP or SACK option bytes.
3019 struct tcp_sock *tp = tcp_sk(sk); in __tcp_select_window() local
3025 * fluctuations. --SAW 1998/11/1 in __tcp_select_window()
3027 int mss = icsk->icsk_ack.rcv_mss; in __tcp_select_window()
3035 full_space = min_t(int, tp->window_clamp, allowed_space); in __tcp_select_window()
3044 * a non-zero scaling factor in effect. in __tcp_select_window()
3046 if (READ_ONCE(net->ipv4.sysctl_tcp_shrink_window) && tp->rx_opt.rcv_wscale) in __tcp_select_window()
3052 icsk->icsk_ack.quick = 0; in __tcp_select_window()
3060 free_space = round_down(free_space, 1 << tp->rx_opt.rcv_wscale); in __tcp_select_window()
3063 * of the maximum allowed, try to move to zero-window, else in __tcp_select_window()
3073 if (free_space > tp->rcv_ssthresh) in __tcp_select_window()
3074 free_space = tp->rcv_ssthresh; in __tcp_select_window()
3079 if (tp->rx_opt.rcv_wscale) { in __tcp_select_window()
3086 window = ALIGN(window, (1 << tp->rx_opt.rcv_wscale)); in __tcp_select_window()
3088 window = tp->rcv_wnd; in __tcp_select_window()
3097 if (window <= free_space - mss || window > free_space) in __tcp_select_window()
3108 free_space = round_down(free_space, 1 << tp->rx_opt.rcv_wscale); in __tcp_select_window()
3111 icsk->icsk_ack.quick = 0; in __tcp_select_window()
3118 free_space < (1 << tp->rx_opt.rcv_wscale)) in __tcp_select_window()
3122 if (free_space > tp->rcv_ssthresh) { in __tcp_select_window()
3123 free_space = tp->rcv_ssthresh; in __tcp_select_window()
3128 * the memory-based limit, and rcv_ssthresh is not a hard limit in __tcp_select_window()
3131 free_space = ALIGN(free_space, (1 << tp->rx_opt.rcv_wscale)); in __tcp_select_window()
3145 shinfo->tx_flags |= next_shinfo->tx_flags & SKBTX_ANY_TSTAMP; in tcp_skb_collapse_tstamp()
3146 shinfo->tskey = next_shinfo->tskey; in tcp_skb_collapse_tstamp()
3147 TCP_SKB_CB(skb)->txstamp_ack |= in tcp_skb_collapse_tstamp()
3148 TCP_SKB_CB(next_skb)->txstamp_ack; in tcp_skb_collapse_tstamp()
3155 struct tcp_sock *tp = tcp_sk(sk); in tcp_collapse_retrans() local
3159 next_skb_size = next_skb->len; in tcp_collapse_retrans()
3169 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(next_skb)->end_seq; in tcp_collapse_retrans()
3172 TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags; in tcp_collapse_retrans()
3177 TCP_SKB_CB(skb)->sacked |= TCP_SKB_CB(next_skb)->sacked & TCPCB_EVER_RETRANS; in tcp_collapse_retrans()
3178 TCP_SKB_CB(skb)->eor = TCP_SKB_CB(next_skb)->eor; in tcp_collapse_retrans()
3181 tcp_clear_retrans_hints_partial(tp); in tcp_collapse_retrans()
3182 if (next_skb == tp->retransmit_skb_hint) in tcp_collapse_retrans()
3183 tp->retransmit_skb_hint = skb; in tcp_collapse_retrans()
3201 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) in tcp_can_collapse()
3213 struct tcp_sock *tp = tcp_sk(sk); in tcp_retrans_try_collapse() local
3217 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_retrans_collapse)) in tcp_retrans_try_collapse()
3219 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN) in tcp_retrans_try_collapse()
3229 space -= skb->len; in tcp_retrans_try_collapse()
3239 if (after(TCP_SKB_CB(skb)->end_seq, tcp_wnd_end(tp))) in tcp_retrans_try_collapse()
3248 * state updates are done by the caller. Returns non-zero if an
3254 struct tcp_sock *tp = tcp_sk(sk); in __tcp_retransmit_skb() local
3260 if (icsk->icsk_mtup.probe_size) in __tcp_retransmit_skb()
3261 icsk->icsk_mtup.probe_size = 0; in __tcp_retransmit_skb()
3264 return -EBUSY; in __tcp_retransmit_skb()
3267 if (before(TCP_SKB_CB(skb)->seq, tp->snd_una)) { in __tcp_retransmit_skb()
3268 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { in __tcp_retransmit_skb()
3269 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_SYN; in __tcp_retransmit_skb()
3270 TCP_SKB_CB(skb)->seq++; in __tcp_retransmit_skb()
3273 if (unlikely(before(TCP_SKB_CB(skb)->end_seq, tp->snd_una))) { in __tcp_retransmit_skb()
3275 return -EINVAL; in __tcp_retransmit_skb()
3277 if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq)) in __tcp_retransmit_skb()
3278 return -ENOMEM; in __tcp_retransmit_skb()
3281 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk)) in __tcp_retransmit_skb()
3282 return -EHOSTUNREACH; /* Routing failure or similar. */ in __tcp_retransmit_skb()
3285 avail_wnd = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in __tcp_retransmit_skb()
3293 if (TCP_SKB_CB(skb)->seq != tp->snd_una) in __tcp_retransmit_skb()
3294 return -EAGAIN; in __tcp_retransmit_skb()
3304 if (skb->len > len) { in __tcp_retransmit_skb()
3307 return -ENOMEM; /* We'll try again later. */ in __tcp_retransmit_skb()
3310 return -ENOMEM; in __tcp_retransmit_skb()
3314 diff -= tcp_skb_pcount(skb); in __tcp_retransmit_skb()
3318 if (skb->len < avail_wnd) in __tcp_retransmit_skb()
3323 if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN_ECN) == TCPHDR_SYN_ECN) in __tcp_retransmit_skb()
3329 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN) in __tcp_retransmit_skb()
3331 tp->total_retrans += segs; in __tcp_retransmit_skb()
3332 tp->bytes_retrans += skb->len; in __tcp_retransmit_skb()
3334 /* make sure skb->data is aligned on arches that require it in __tcp_retransmit_skb()
3335 * and check if ack-trimming & collapsing extended the headroom in __tcp_retransmit_skb()
3338 if (unlikely((NET_IP_ALIGN && ((unsigned long)skb->data & 3)) || in __tcp_retransmit_skb()
3345 nskb->dev = NULL; in __tcp_retransmit_skb()
3348 err = -ENOBUFS; in __tcp_retransmit_skb()
3353 tcp_update_skb_after_send(sk, skb, tp->tcp_wstamp_ns); in __tcp_retransmit_skb()
3363 TCP_SKB_CB(skb)->sacked |= TCPCB_EVER_RETRANS; in __tcp_retransmit_skb()
3365 if (BPF_SOCK_OPS_TEST_FLAG(tp, BPF_SOCK_OPS_RETRANS_CB_FLAG)) in __tcp_retransmit_skb()
3367 TCP_SKB_CB(skb)->seq, segs, err); in __tcp_retransmit_skb()
3371 } else if (err != -EBUSY) { in __tcp_retransmit_skb()
3379 struct tcp_sock *tp = tcp_sk(sk); in tcp_retransmit_skb() local
3384 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) { in tcp_retransmit_skb()
3388 TCP_SKB_CB(skb)->sacked |= TCPCB_RETRANS; in tcp_retransmit_skb()
3389 tp->retrans_out += tcp_skb_pcount(skb); in tcp_retransmit_skb()
3393 if (!tp->retrans_stamp) in tcp_retransmit_skb()
3394 tp->retrans_stamp = tcp_skb_timestamp(skb); in tcp_retransmit_skb()
3396 if (tp->undo_retrans < 0) in tcp_retransmit_skb()
3397 tp->undo_retrans = 0; in tcp_retransmit_skb()
3398 tp->undo_retrans += tcp_skb_pcount(skb); in tcp_retransmit_skb()
3411 struct tcp_sock *tp = tcp_sk(sk); in tcp_xmit_retransmit_queue() local
3416 if (!tp->packets_out) in tcp_xmit_retransmit_queue()
3420 skb = tp->retransmit_skb_hint ?: rtx_head; in tcp_xmit_retransmit_queue()
3431 tp->retransmit_skb_hint = skb; in tcp_xmit_retransmit_queue()
3433 segs = tcp_snd_cwnd(tp) - tcp_packets_in_flight(tp); in tcp_xmit_retransmit_queue()
3436 sacked = TCP_SKB_CB(skb)->sacked; in tcp_xmit_retransmit_queue()
3442 if (tp->retrans_out >= tp->lost_out) { in tcp_xmit_retransmit_queue()
3450 if (icsk->icsk_ca_state != TCP_CA_Loss) in tcp_xmit_retransmit_queue()
3468 tp->prr_out += tcp_skb_pcount(skb); in tcp_xmit_retransmit_queue()
3471 icsk->icsk_pending != ICSK_TIME_REO_TIMEOUT) in tcp_xmit_retransmit_queue()
3477 inet_csk(sk)->icsk_rto, in tcp_xmit_retransmit_queue()
3492 delta = size - sk->sk_forward_alloc; in sk_forced_mem_schedule()
3499 if (mem_cgroup_sockets_enabled && sk->sk_memcg) in sk_forced_mem_schedule()
3500 mem_cgroup_charge_skmem(sk->sk_memcg, amt, in sk_forced_mem_schedule()
3510 struct tcp_sock *tp = tcp_sk(sk); in tcp_send_fin() local
3519 tskb = skb_rb_last(&sk->tcp_rtx_queue); in tcp_send_fin()
3522 TCP_SKB_CB(tskb)->tcp_flags |= TCPHDR_FIN; in tcp_send_fin()
3523 TCP_SKB_CB(tskb)->end_seq++; in tcp_send_fin()
3524 tp->write_seq++; in tcp_send_fin()
3528 * We need to set tp->snd_nxt to the value it would have in tcp_send_fin()
3530 * does not change tp->snd_nxt. in tcp_send_fin()
3532 WRITE_ONCE(tp->snd_nxt, tp->snd_nxt + 1); in tcp_send_fin()
3536 skb = alloc_skb_fclone(MAX_TCP_HEADER, sk->sk_allocation); in tcp_send_fin()
3540 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor); in tcp_send_fin()
3542 sk_forced_mem_schedule(sk, skb->truesize); in tcp_send_fin()
3544 tcp_init_nondata_skb(skb, tp->write_seq, in tcp_send_fin()
3554 * by RFC 2525, section 2.17. -DaveM
3584 /* Send a crossed SYN-ACK during socket establishment.
3595 if (!skb || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { in tcp_send_synack()
3597 return -EFAULT; in tcp_send_synack()
3599 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) { in tcp_send_synack()
3607 return -ENOMEM; in tcp_send_synack()
3608 INIT_LIST_HEAD(&nskb->tcp_tsorted_anchor); in tcp_send_synack()
3612 tcp_rbtree_insert(&sk->tcp_rtx_queue, nskb); in tcp_send_synack()
3613 sk_wmem_queued_add(sk, nskb->truesize); in tcp_send_synack()
3614 sk_mem_charge(sk, nskb->truesize); in tcp_send_synack()
3618 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK; in tcp_send_synack()
3625 * tcp_make_synack - Allocate one skb and build a SYNACK packet.
3641 const struct tcp_sock *tp = tcp_sk(sk); in tcp_make_synack() local
3670 * sk->sk_wmem_alloc in an atomic, we can promote to rw. in tcp_make_synack()
3677 mss = tcp_mss_clamp(tp, dst_metric_advmss(dst)); in tcp_make_synack()
3682 if (unlikely(synack_type == TCP_SYNACK_COOKIE && ireq->tstamp_ok)) in tcp_make_synack()
3689 if (!tcp_rsk(req)->snt_synack) /* Timestamp first SYNACK */ in tcp_make_synack()
3690 tcp_rsk(req)->snt_synack = tcp_skb_timestamp_us(skb); in tcp_make_synack()
3695 md5 = tcp_rsk(req)->af_specific->req_md5_lookup(sk, req_to_sk(req)); in tcp_make_synack()
3697 skb_set_hash(skb, READ_ONCE(tcp_rsk(req)->txhash), PKT_HASH_TYPE_L4); in tcp_make_synack()
3699 TCP_SKB_CB(skb)->tcp_flags = TCPHDR_SYN | TCPHDR_ACK; in tcp_make_synack()
3707 th = (struct tcphdr *)skb->data; in tcp_make_synack()
3709 th->syn = 1; in tcp_make_synack()
3710 th->ack = 1; in tcp_make_synack()
3712 th->source = htons(ireq->ir_num); in tcp_make_synack()
3713 th->dest = ireq->ir_rmt_port; in tcp_make_synack()
3714 skb->mark = ireq->ir_mark; in tcp_make_synack()
3715 skb->ip_summed = CHECKSUM_PARTIAL; in tcp_make_synack()
3716 th->seq = htonl(tcp_rsk(req)->snt_isn); in tcp_make_synack()
3718 th->ack_seq = htonl(tcp_rsk(req)->rcv_nxt); in tcp_make_synack()
3721 th->window = htons(min(req->rsk_rcv_wnd, 65535U)); in tcp_make_synack()
3723 th->doff = (tcp_header_size >> 2); in tcp_make_synack()
3727 /* Okay, we have all we need - do the md5 hash if needed */ in tcp_make_synack()
3729 tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location, in tcp_make_synack()
3738 tcp_add_tx_delay(skb, tp); in tcp_make_synack()
3755 if (likely(ca && bpf_try_module_get(ca, ca->owner))) { in tcp_ca_dst_init()
3756 bpf_module_put(icsk->icsk_ca_ops, icsk->icsk_ca_ops->owner); in tcp_ca_dst_init()
3757 icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst); in tcp_ca_dst_init()
3758 icsk->icsk_ca_ops = ca; in tcp_ca_dst_init()
3767 struct tcp_sock *tp = tcp_sk(sk); in tcp_connect_init() local
3774 tp->tcp_header_len = sizeof(struct tcphdr); in tcp_connect_init()
3775 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_timestamps)) in tcp_connect_init()
3776 tp->tcp_header_len += TCPOLEN_TSTAMP_ALIGNED; in tcp_connect_init()
3779 if (tp->rx_opt.user_mss) in tcp_connect_init()
3780 tp->rx_opt.mss_clamp = tp->rx_opt.user_mss; in tcp_connect_init()
3781 tp->max_window = 0; in tcp_connect_init()
3787 if (!tp->window_clamp) in tcp_connect_init()
3788 tp->window_clamp = dst_metric(dst, RTAX_WINDOW); in tcp_connect_init()
3789 tp->advmss = tcp_mss_clamp(tp, dst_metric_advmss(dst)); in tcp_connect_init()
3794 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK && in tcp_connect_init()
3795 (tp->window_clamp > tcp_full_space(sk) || tp->window_clamp == 0)) in tcp_connect_init()
3796 tp->window_clamp = tcp_full_space(sk); in tcp_connect_init()
3803 tp->advmss - (tp->rx_opt.ts_recent_stamp ? tp->tcp_header_len - sizeof(struct tcphdr) : 0), in tcp_connect_init()
3804 &tp->rcv_wnd, in tcp_connect_init()
3805 &tp->window_clamp, in tcp_connect_init()
3806 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_window_scaling), in tcp_connect_init()
3810 tp->rx_opt.rcv_wscale = rcv_wscale; in tcp_connect_init()
3811 tp->rcv_ssthresh = tp->rcv_wnd; in tcp_connect_init()
3813 WRITE_ONCE(sk->sk_err, 0); in tcp_connect_init()
3815 tp->snd_wnd = 0; in tcp_connect_init()
3816 tcp_init_wl(tp, 0); in tcp_connect_init()
3818 tp->snd_una = tp->write_seq; in tcp_connect_init()
3819 tp->snd_sml = tp->write_seq; in tcp_connect_init()
3820 tp->snd_up = tp->write_seq; in tcp_connect_init()
3821 WRITE_ONCE(tp->snd_nxt, tp->write_seq); in tcp_connect_init()
3823 if (likely(!tp->repair)) in tcp_connect_init()
3824 tp->rcv_nxt = 0; in tcp_connect_init()
3826 tp->rcv_tstamp = tcp_jiffies32; in tcp_connect_init()
3827 tp->rcv_wup = tp->rcv_nxt; in tcp_connect_init()
3828 WRITE_ONCE(tp->copied_seq, tp->rcv_nxt); in tcp_connect_init()
3830 inet_csk(sk)->icsk_rto = tcp_timeout_init(sk); in tcp_connect_init()
3831 inet_csk(sk)->icsk_retransmits = 0; in tcp_connect_init()
3832 tcp_clear_retrans(tp); in tcp_connect_init()
3837 struct tcp_sock *tp = tcp_sk(sk); in tcp_connect_queue_skb() local
3840 tcb->end_seq += skb->len; in tcp_connect_queue_skb()
3842 sk_wmem_queued_add(sk, skb->truesize); in tcp_connect_queue_skb()
3843 sk_mem_charge(sk, skb->truesize); in tcp_connect_queue_skb()
3844 WRITE_ONCE(tp->write_seq, tcb->end_seq); in tcp_connect_queue_skb()
3845 tp->packets_out += tcp_skb_pcount(skb); in tcp_connect_queue_skb()
3849 * queue a data-only packet after the regular SYN, such that regular SYNs
3850 * are retransmitted on timeouts. Also if the remote SYN-ACK acknowledges
3858 struct tcp_sock *tp = tcp_sk(sk); in tcp_send_syn_data() local
3859 struct tcp_fastopen_request *fo = tp->fastopen_req; in tcp_send_syn_data()
3864 tp->rx_opt.mss_clamp = tp->advmss; /* If MSS is not cached */ in tcp_send_syn_data()
3865 if (!tcp_fastopen_cookie_check(sk, &tp->rx_opt.mss_clamp, &fo->cookie)) in tcp_send_syn_data()
3868 /* MSS for SYN-data is based on cached MSS and bounded by PMTU and in tcp_send_syn_data()
3869 * user-MSS. Reserve maximum option space for middleboxes that add in tcp_send_syn_data()
3872 tp->rx_opt.mss_clamp = tcp_mss_clamp(tp, tp->rx_opt.mss_clamp); in tcp_send_syn_data()
3874 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); in tcp_send_syn_data()
3876 space = __tcp_mtu_to_mss(sk, icsk->icsk_pmtu_cookie) - in tcp_send_syn_data()
3879 space = min_t(size_t, space, fo->size); in tcp_send_syn_data()
3883 pfrag, sk->sk_allocation)) in tcp_send_syn_data()
3885 syn_data = tcp_stream_alloc_skb(sk, sk->sk_allocation, false); in tcp_send_syn_data()
3888 memcpy(syn_data->cb, syn->cb, sizeof(syn->cb)); in tcp_send_syn_data()
3890 space = min_t(size_t, space, pfrag->size - pfrag->offset); in tcp_send_syn_data()
3894 space = copy_page_from_iter(pfrag->page, pfrag->offset, in tcp_send_syn_data()
3895 space, &fo->data->msg_iter); in tcp_send_syn_data()
3901 skb_fill_page_desc(syn_data, 0, pfrag->page, in tcp_send_syn_data()
3902 pfrag->offset, space); in tcp_send_syn_data()
3903 page_ref_inc(pfrag->page); in tcp_send_syn_data()
3904 pfrag->offset += space; in tcp_send_syn_data()
3906 skb_zcopy_set(syn_data, fo->uarg, NULL); in tcp_send_syn_data()
3909 if (space == fo->size) in tcp_send_syn_data()
3910 fo->data = NULL; in tcp_send_syn_data()
3911 fo->copied = space; in tcp_send_syn_data()
3914 if (syn_data->len) in tcp_send_syn_data()
3917 err = tcp_transmit_skb(sk, syn_data, 1, sk->sk_allocation); in tcp_send_syn_data()
3919 skb_set_delivery_time(syn, syn_data->skb_mstamp_ns, true); in tcp_send_syn_data()
3926 TCP_SKB_CB(syn_data)->seq++; in tcp_send_syn_data()
3927 TCP_SKB_CB(syn_data)->tcp_flags = TCPHDR_ACK | TCPHDR_PSH; in tcp_send_syn_data()
3929 tp->syn_data = (fo->copied > 0); in tcp_send_syn_data()
3930 tcp_rbtree_insert(&sk->tcp_rtx_queue, syn_data); in tcp_send_syn_data()
3936 __skb_queue_tail(&sk->sk_write_queue, syn_data); in tcp_send_syn_data()
3937 tp->packets_out -= tcp_skb_pcount(syn_data); in tcp_send_syn_data()
3941 if (fo->cookie.len > 0) in tcp_send_syn_data()
3942 fo->cookie.len = 0; in tcp_send_syn_data()
3943 err = tcp_transmit_skb(sk, syn, 1, sk->sk_allocation); in tcp_send_syn_data()
3945 tp->syn_fastopen = 0; in tcp_send_syn_data()
3947 fo->cookie.len = -1; /* Exclude Fast Open option for SYN retries */ in tcp_send_syn_data()
3954 struct tcp_sock *tp = tcp_sk(sk); in tcp_connect() local
3960 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk)) in tcp_connect()
3961 return -EHOSTUNREACH; /* Routing failure or similar. */ in tcp_connect()
3965 if (unlikely(tp->repair)) { in tcp_connect()
3970 buff = tcp_stream_alloc_skb(sk, sk->sk_allocation, true); in tcp_connect()
3972 return -ENOBUFS; in tcp_connect()
3974 tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN); in tcp_connect()
3975 tcp_mstamp_refresh(tp); in tcp_connect()
3976 tp->retrans_stamp = tcp_time_stamp(tp); in tcp_connect()
3979 tcp_rbtree_insert(&sk->tcp_rtx_queue, buff); in tcp_connect()
3982 err = tp->fastopen_req ? tcp_send_syn_data(sk, buff) : in tcp_connect()
3983 tcp_transmit_skb(sk, buff, 1, sk->sk_allocation); in tcp_connect()
3984 if (err == -ECONNREFUSED) in tcp_connect()
3987 /* We change tp->snd_nxt after the tcp_transmit_skb() call in tcp_connect()
3990 WRITE_ONCE(tp->snd_nxt, tp->write_seq); in tcp_connect()
3991 tp->pushed_seq = tp->write_seq; in tcp_connect()
3994 WRITE_ONCE(tp->snd_nxt, TCP_SKB_CB(buff)->seq); in tcp_connect()
3995 tp->pushed_seq = TCP_SKB_CB(buff)->seq; in tcp_connect()
4001 inet_csk(sk)->icsk_rto, TCP_RTO_MAX); in tcp_connect()
4009 u32 delack_max = inet_csk(sk)->icsk_delack_max; in tcp_delack_max()
4013 u32 delack_from_rto_min = max_t(int, 1, rto_min - 1); in tcp_delack_max()
4027 int ato = icsk->icsk_ack.ato; in tcp_send_delayed_ack()
4031 const struct tcp_sock *tp = tcp_sk(sk); in tcp_send_delayed_ack() local
4035 (icsk->icsk_ack.pending & ICSK_ACK_PUSHED)) in tcp_send_delayed_ack()
4041 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements in tcp_send_delayed_ack()
4044 if (tp->srtt_us) { in tcp_send_delayed_ack()
4045 int rtt = max_t(int, usecs_to_jiffies(tp->srtt_us >> 3), in tcp_send_delayed_ack()
4061 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) { in tcp_send_delayed_ack()
4063 if (time_before_eq(icsk->icsk_ack.timeout, jiffies + (ato >> 2))) { in tcp_send_delayed_ack()
4068 if (!time_before(timeout, icsk->icsk_ack.timeout)) in tcp_send_delayed_ack()
4069 timeout = icsk->icsk_ack.timeout; in tcp_send_delayed_ack()
4071 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER; in tcp_send_delayed_ack()
4072 icsk->icsk_ack.timeout = timeout; in tcp_send_delayed_ack()
4073 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout); in tcp_send_delayed_ack()
4082 if (sk->sk_state == TCP_CLOSE) in __tcp_send_ack()
4095 delay = TCP_DELACK_MAX << icsk->icsk_ack.retry; in __tcp_send_ack()
4097 icsk->icsk_ack.retry++; in __tcp_send_ack()
4099 icsk->icsk_ack.ato = TCP_ATO_MIN; in __tcp_send_ack()
4121 __tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt); in tcp_send_ack()
4131 * Current solution: to send TWO zero-length segments in urgent mode:
4133 * out-of-date with SND.UNA-1 to probe window.
4137 struct tcp_sock *tp = tcp_sk(sk); in tcp_xmit_probe_skb() local
4144 return -1; in tcp_xmit_probe_skb()
4152 tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK); in tcp_xmit_probe_skb()
4160 if (sk->sk_state == TCP_ESTABLISHED) { in tcp_send_window_probe()
4161 tcp_sk(sk)->snd_wl1 = tcp_sk(sk)->rcv_nxt - 1; in tcp_send_window_probe()
4170 struct tcp_sock *tp = tcp_sk(sk); in tcp_write_wakeup() local
4173 if (sk->sk_state == TCP_CLOSE) in tcp_write_wakeup()
4174 return -1; in tcp_write_wakeup()
4177 if (skb && before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp))) { in tcp_write_wakeup()
4180 unsigned int seg_size = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in tcp_write_wakeup()
4182 if (before(tp->pushed_seq, TCP_SKB_CB(skb)->end_seq)) in tcp_write_wakeup()
4183 tp->pushed_seq = TCP_SKB_CB(skb)->end_seq; in tcp_write_wakeup()
4189 if (seg_size < TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq || in tcp_write_wakeup()
4190 skb->len > mss) { in tcp_write_wakeup()
4192 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; in tcp_write_wakeup()
4195 return -1; in tcp_write_wakeup()
4199 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; in tcp_write_wakeup()
4205 if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF)) in tcp_write_wakeup()
4217 struct tcp_sock *tp = tcp_sk(sk); in tcp_send_probe0() local
4224 if (tp->packets_out || tcp_write_queue_empty(sk)) { in tcp_send_probe0()
4226 icsk->icsk_probes_out = 0; in tcp_send_probe0()
4227 icsk->icsk_backoff = 0; in tcp_send_probe0()
4228 icsk->icsk_probes_tstamp = 0; in tcp_send_probe0()
4232 icsk->icsk_probes_out++; in tcp_send_probe0()
4234 if (icsk->icsk_backoff < READ_ONCE(net->ipv4.sysctl_tcp_retries2)) in tcp_send_probe0()
4235 icsk->icsk_backoff++; in tcp_send_probe0()
4250 const struct tcp_request_sock_ops *af_ops = tcp_rsk(req)->af_specific; in tcp_rtx_synack()
4255 if (READ_ONCE(sk->sk_txrehash) == SOCK_TXREHASH_ENABLED) in tcp_rtx_synack()
4256 WRITE_ONCE(tcp_rsk(req)->txhash, net_tx_rndhash()); in tcp_rtx_synack()
4257 res = af_ops->send_synack(sk, NULL, &fl, req, NULL, TCP_SYNACK_NORMAL, in tcp_rtx_synack()
4267 tcp_sk_rw(sk)->total_retrans++; in tcp_rtx_synack()