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
55 void tcp_mstamp_refresh(struct tcp_sock *tp) in tcp_mstamp_refresh() argument
59 tp->tcp_clock_cache = val; in tcp_mstamp_refresh()
60 tp->tcp_mstamp = div_u64(val, NSEC_PER_USEC); in tcp_mstamp_refresh()
70 struct tcp_sock *tp = tcp_sk(sk); in tcp_event_new_data_sent() local
71 unsigned int prior_packets = tp->packets_out; in tcp_event_new_data_sent()
73 WRITE_ONCE(tp->snd_nxt, TCP_SKB_CB(skb)->end_seq); in tcp_event_new_data_sent()
75 __skb_unlink(skb, &sk->sk_write_queue); in tcp_event_new_data_sent()
76 tcp_rbtree_insert(&sk->tcp_rtx_queue, skb); in tcp_event_new_data_sent()
78 if (tp->highest_sack == NULL) in tcp_event_new_data_sent()
79 tp->highest_sack = skb; in tcp_event_new_data_sent()
81 tp->packets_out += tcp_skb_pcount(skb); in tcp_event_new_data_sent()
82 if (!prior_packets || icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) in tcp_event_new_data_sent()
93 * Using SND.UNA we will fail to open window, SND.NXT is out of window. :-(
99 const struct tcp_sock *tp = tcp_sk(sk); in tcp_acceptable_seq() local
101 if (!before(tcp_wnd_end(tp), tp->snd_nxt) || in tcp_acceptable_seq()
102 (tp->rx_opt.wscale_ok && in tcp_acceptable_seq()
103 ((tp->snd_nxt - tcp_wnd_end(tp)) < (1 << tp->rx_opt.rcv_wscale)))) in tcp_acceptable_seq()
104 return tp->snd_nxt; in tcp_acceptable_seq()
106 return tcp_wnd_end(tp); in tcp_acceptable_seq()
110 * RFC1122, RFC1063, draft-ietf-tcpimpl-pmtud-01 state that:
113 * 2. Ideally, it is maximal possible segment size i.e. 65535-40.
125 struct tcp_sock *tp = tcp_sk(sk); in tcp_advertise_mss() local
127 int mss = tp->advmss; in tcp_advertise_mss()
134 tp->advmss = mss; in tcp_advertise_mss()
146 struct tcp_sock *tp = tcp_sk(sk); in tcp_cwnd_restart() local
147 u32 restart_cwnd = tcp_init_cwnd(tp, __sk_dst_get(sk)); in tcp_cwnd_restart()
148 u32 cwnd = tp->snd_cwnd; in tcp_cwnd_restart()
152 tp->snd_ssthresh = tcp_current_ssthresh(sk); in tcp_cwnd_restart()
155 while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd) in tcp_cwnd_restart()
157 tp->snd_cwnd = max(cwnd, restart_cwnd); in tcp_cwnd_restart()
158 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_cwnd_restart()
159 tp->snd_cwnd_used = 0; in tcp_cwnd_restart()
163 static void tcp_event_data_sent(struct tcp_sock *tp, in tcp_event_data_sent() argument
169 if (tcp_packets_in_flight(tp) == 0) in tcp_event_data_sent()
172 tp->lsndtime = now; in tcp_event_data_sent()
177 if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato) in tcp_event_data_sent()
184 struct tcp_sock *tp = tcp_sk(sk); in tcp_event_ack_sent() local
186 if (unlikely(tp->compressed_ack)) { in tcp_event_ack_sent()
188 tp->compressed_ack); in tcp_event_ack_sent()
189 tp->compressed_ack = 0; in tcp_event_ack_sent()
190 if (hrtimer_try_to_cancel(&tp->compressed_ack_timer) == 1) in tcp_event_ack_sent()
194 if (unlikely(rcv_nxt != tp->rcv_nxt)) in tcp_event_ack_sent()
202 * will be offered. Store the results in the tp structure.
226 * we will truncate our initial window offering to 32K-1 in tcp_select_initial_window()
231 if (sock_net(sk)->ipv4.sysctl_tcp_workaround_signed_windows) in tcp_select_initial_window()
242 space = max_t(u32, space, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2])); in tcp_select_initial_window()
245 *rcv_wscale = clamp_t(int, ilog2(space) - 15, in tcp_select_initial_window()
255 * value can be stuffed directly into th->window for an outgoing
260 struct tcp_sock *tp = tcp_sk(sk); in tcp_select_window() local
261 u32 old_win = tp->rcv_wnd; in tcp_select_window()
262 u32 cur_win = tcp_receive_window(tp); in tcp_select_window()
270 * window in time. --DaveM in tcp_select_window()
277 new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale); in tcp_select_window()
279 tp->rcv_wnd = new_win; in tcp_select_window()
280 tp->rcv_wup = tp->rcv_nxt; in tcp_select_window()
285 if (!tp->rx_opt.rcv_wscale && in tcp_select_window()
286 sock_net(sk)->ipv4.sysctl_tcp_workaround_signed_windows) in tcp_select_window()
289 new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale)); in tcp_select_window()
292 new_win >>= tp->rx_opt.rcv_wscale; in tcp_select_window()
296 tp->pred_flags = 0; in tcp_select_window()
307 /* Packet ECN state for a SYN-ACK */
310 const struct tcp_sock *tp = tcp_sk(sk); in tcp_ecn_send_synack() local
312 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR; in tcp_ecn_send_synack()
313 if (!(tp->ecn_flags & TCP_ECN_OK)) in tcp_ecn_send_synack()
314 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE; in tcp_ecn_send_synack()
323 struct tcp_sock *tp = tcp_sk(sk); in tcp_ecn_send_syn() local
325 bool use_ecn = sock_net(sk)->ipv4.sysctl_tcp_ecn == 1 || in tcp_ecn_send_syn()
335 tp->ecn_flags = 0; in tcp_ecn_send_syn()
338 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR; in tcp_ecn_send_syn()
339 tp->ecn_flags = TCP_ECN_OK; in tcp_ecn_send_syn()
347 if (sock_net(sk)->ipv4.sysctl_tcp_ecn_fallback) in tcp_ecn_clear_syn()
348 /* tp->ecn_flags are cleared at a later point in time when in tcp_ecn_clear_syn()
351 TCP_SKB_CB(skb)->tcp_flags &= ~(TCPHDR_ECE | TCPHDR_CWR); in tcp_ecn_clear_syn()
357 if (inet_rsk(req)->ecn_ok) in tcp_ecn_make_synack()
358 th->ece = 1; in tcp_ecn_make_synack()
367 struct tcp_sock *tp = tcp_sk(sk); in tcp_ecn_send() local
369 if (tp->ecn_flags & TCP_ECN_OK) { in tcp_ecn_send()
370 /* Not-retransmitted data segment: set ECT and inject CWR. */ in tcp_ecn_send()
371 if (skb->len != tcp_header_len && in tcp_ecn_send()
372 !before(TCP_SKB_CB(skb)->seq, tp->snd_nxt)) { in tcp_ecn_send()
374 if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) { in tcp_ecn_send()
375 tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR; in tcp_ecn_send()
376 th->cwr = 1; in tcp_ecn_send()
377 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; in tcp_ecn_send()
383 if (tp->ecn_flags & TCP_ECN_DEMAND_CWR) in tcp_ecn_send()
384 th->ece = 1; in tcp_ecn_send()
388 /* Constructs common control bits of non-data skb. If SYN/FIN is present,
393 skb->ip_summed = CHECKSUM_PARTIAL; in tcp_init_nondata_skb()
395 TCP_SKB_CB(skb)->tcp_flags = flags; in tcp_init_nondata_skb()
396 TCP_SKB_CB(skb)->sacked = 0; in tcp_init_nondata_skb()
400 TCP_SKB_CB(skb)->seq = seq; in tcp_init_nondata_skb()
403 TCP_SKB_CB(skb)->end_seq = seq; in tcp_init_nondata_skb()
406 static inline bool tcp_urg_mode(const struct tcp_sock *tp) in tcp_urg_mode() argument
408 return tp->snd_una != tp->snd_up; in tcp_urg_mode()
450 if (unlikely(OPTION_MPTCP & opts->options)) in mptcp_options_write()
451 mptcp_write_options(ptr, &opts->mptcp); in mptcp_options_write()
497 * Thus, "req" is passed here and the cgroup-bpf-progs in bpf_skops_hdr_opt_len()
502 * consistent between fastopen and non-fastopen on in bpf_skops_hdr_opt_len()
525 opts->bpf_opt_len = *remaining - sock_ops.remaining_opt_len; in bpf_skops_hdr_opt_len()
527 opts->bpf_opt_len = (opts->bpf_opt_len + 3) & ~3; in bpf_skops_hdr_opt_len()
529 *remaining -= opts->bpf_opt_len; in bpf_skops_hdr_opt_len()
538 u8 first_opt_off, nr_written, max_opt_len = opts->bpf_opt_len; in bpf_skops_write_hdr_opt()
561 first_opt_off = tcp_hdrlen(skb) - max_opt_len; in bpf_skops_write_hdr_opt()
569 nr_written = max_opt_len - sock_ops.remaining_opt_len; in bpf_skops_write_hdr_opt()
572 memset(skb->data + first_opt_off + nr_written, TCPOPT_NOP, in bpf_skops_write_hdr_opt()
573 max_opt_len - nr_written); in bpf_skops_write_hdr_opt()
596 * Beware: Something in the Internet is very sensitive to the ordering of
598 * Luckily we can at least blame others for their non-compliance but from
599 * inter-operability perspective it seems that we're somewhat stuck with
607 static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp, in tcp_options_write() argument
610 u16 options = opts->options; /* mungable copy */ in tcp_options_write()
616 opts->hash_location = (__u8 *)ptr; in tcp_options_write()
620 if (unlikely(opts->mss)) { in tcp_options_write()
623 opts->mss); in tcp_options_write()
639 *ptr++ = htonl(opts->tsval); in tcp_options_write()
640 *ptr++ = htonl(opts->tsecr); in tcp_options_write()
654 opts->ws); in tcp_options_write()
657 if (unlikely(opts->num_sack_blocks)) { in tcp_options_write()
658 struct tcp_sack_block *sp = tp->rx_opt.dsack ? in tcp_options_write()
659 tp->duplicate_sack : tp->selective_acks; in tcp_options_write()
665 (TCPOLEN_SACK_BASE + (opts->num_sack_blocks * in tcp_options_write()
668 for (this_sack = 0; this_sack < opts->num_sack_blocks; in tcp_options_write()
674 tp->rx_opt.dsack = 0; in tcp_options_write()
678 struct tcp_fastopen_cookie *foc = opts->fastopen_cookie; in tcp_options_write()
682 if (foc->exp) { in tcp_options_write()
683 len = TCPOLEN_EXP_FASTOPEN_BASE + foc->len; in tcp_options_write()
688 len = TCPOLEN_FASTOPEN_BASE + foc->len; in tcp_options_write()
693 memcpy(p, foc->val, foc->len); in tcp_options_write()
695 p[foc->len] = TCPOPT_NOP; in tcp_options_write()
696 p[foc->len + 1] = TCPOPT_NOP; in tcp_options_write()
706 static void smc_set_option(const struct tcp_sock *tp, in smc_set_option() argument
712 if (tp->syn_smc) { in smc_set_option()
714 opts->options |= OPTION_SMC; in smc_set_option()
715 *remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED; in smc_set_option()
722 static void smc_set_option_cond(const struct tcp_sock *tp, in smc_set_option_cond() argument
729 if (tp->syn_smc && ireq->smc_ok) { in smc_set_option_cond()
731 opts->options |= OPTION_SMC; in smc_set_option_cond()
732 *remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED; in smc_set_option_cond()
746 if (mptcp_synack_options(req, &size, &opts->mptcp)) { in mptcp_set_option_cond()
748 opts->options |= OPTION_MPTCP; in mptcp_set_option_cond()
749 *remaining -= size; in mptcp_set_option_cond()
762 struct tcp_sock *tp = tcp_sk(sk); in tcp_syn_options() local
764 struct tcp_fastopen_request *fastopen = tp->fastopen_req; in tcp_syn_options()
769 rcu_access_pointer(tp->md5sig_info)) { in tcp_syn_options()
770 *md5 = tp->af_specific->md5_lookup(sk, sk); in tcp_syn_options()
772 opts->options |= OPTION_MD5; in tcp_syn_options()
773 remaining -= TCPOLEN_MD5SIG_ALIGNED; in tcp_syn_options()
780 * advertised. But we subtract them from tp->mss_cache so that in tcp_syn_options()
787 opts->mss = tcp_advertise_mss(sk); in tcp_syn_options()
788 remaining -= TCPOLEN_MSS_ALIGNED; in tcp_syn_options()
790 if (likely(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_timestamps) && !*md5)) { in tcp_syn_options()
791 opts->options |= OPTION_TS; in tcp_syn_options()
792 opts->tsval = tcp_skb_timestamp(skb) + tp->tsoffset; in tcp_syn_options()
793 opts->tsecr = tp->rx_opt.ts_recent; in tcp_syn_options()
794 remaining -= TCPOLEN_TSTAMP_ALIGNED; in tcp_syn_options()
796 if (likely(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_window_scaling))) { in tcp_syn_options()
797 opts->ws = tp->rx_opt.rcv_wscale; in tcp_syn_options()
798 opts->options |= OPTION_WSCALE; in tcp_syn_options()
799 remaining -= TCPOLEN_WSCALE_ALIGNED; in tcp_syn_options()
801 if (likely(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_sack))) { in tcp_syn_options()
802 opts->options |= OPTION_SACK_ADVERTISE; in tcp_syn_options()
803 if (unlikely(!(OPTION_TS & opts->options))) in tcp_syn_options()
804 remaining -= TCPOLEN_SACKPERM_ALIGNED; in tcp_syn_options()
807 if (fastopen && fastopen->cookie.len >= 0) { in tcp_syn_options()
808 u32 need = fastopen->cookie.len; in tcp_syn_options()
810 need += fastopen->cookie.exp ? TCPOLEN_EXP_FASTOPEN_BASE : in tcp_syn_options()
814 opts->options |= OPTION_FAST_OPEN_COOKIE; in tcp_syn_options()
815 opts->fastopen_cookie = &fastopen->cookie; in tcp_syn_options()
816 remaining -= need; in tcp_syn_options()
817 tp->syn_fastopen = 1; in tcp_syn_options()
818 tp->syn_fastopen_exp = fastopen->cookie.exp ? 1 : 0; in tcp_syn_options()
822 smc_set_option(tp, opts, &remaining); in tcp_syn_options()
827 if (mptcp_syn_options(sk, skb, &size, &opts->mptcp)) { in tcp_syn_options()
828 opts->options |= OPTION_MPTCP; in tcp_syn_options()
829 remaining -= size; in tcp_syn_options()
835 return MAX_TCP_OPTION_SPACE - remaining; in tcp_syn_options()
838 /* Set up TCP options for SYN-ACKs. */
853 opts->options |= OPTION_MD5; in tcp_synack_options()
854 remaining -= TCPOLEN_MD5SIG_ALIGNED; in tcp_synack_options()
862 ireq->tstamp_ok &= !ireq->sack_ok; in tcp_synack_options()
867 opts->mss = mss; in tcp_synack_options()
868 remaining -= TCPOLEN_MSS_ALIGNED; in tcp_synack_options()
870 if (likely(ireq->wscale_ok)) { in tcp_synack_options()
871 opts->ws = ireq->rcv_wscale; in tcp_synack_options()
872 opts->options |= OPTION_WSCALE; in tcp_synack_options()
873 remaining -= TCPOLEN_WSCALE_ALIGNED; in tcp_synack_options()
875 if (likely(ireq->tstamp_ok)) { in tcp_synack_options()
876 opts->options |= OPTION_TS; in tcp_synack_options()
877 opts->tsval = tcp_skb_timestamp(skb) + tcp_rsk(req)->ts_off; in tcp_synack_options()
878 opts->tsecr = READ_ONCE(req->ts_recent); in tcp_synack_options()
879 remaining -= TCPOLEN_TSTAMP_ALIGNED; in tcp_synack_options()
881 if (likely(ireq->sack_ok)) { in tcp_synack_options()
882 opts->options |= OPTION_SACK_ADVERTISE; in tcp_synack_options()
883 if (unlikely(!ireq->tstamp_ok)) in tcp_synack_options()
884 remaining -= TCPOLEN_SACKPERM_ALIGNED; in tcp_synack_options()
886 if (foc != NULL && foc->len >= 0) { in tcp_synack_options()
887 u32 need = foc->len; in tcp_synack_options()
889 need += foc->exp ? TCPOLEN_EXP_FASTOPEN_BASE : in tcp_synack_options()
893 opts->options |= OPTION_FAST_OPEN_COOKIE; in tcp_synack_options()
894 opts->fastopen_cookie = foc; in tcp_synack_options()
895 remaining -= need; in tcp_synack_options()
906 return MAX_TCP_OPTION_SPACE - remaining; in tcp_synack_options()
916 struct tcp_sock *tp = tcp_sk(sk); in tcp_established_options() local
920 opts->options = 0; in tcp_established_options()
925 rcu_access_pointer(tp->md5sig_info)) { in tcp_established_options()
926 *md5 = tp->af_specific->md5_lookup(sk, sk); in tcp_established_options()
928 opts->options |= OPTION_MD5; in tcp_established_options()
934 if (likely(tp->rx_opt.tstamp_ok)) { in tcp_established_options()
935 opts->options |= OPTION_TS; in tcp_established_options()
936 opts->tsval = skb ? tcp_skb_timestamp(skb) + tp->tsoffset : 0; in tcp_established_options()
937 opts->tsecr = tp->rx_opt.ts_recent; in tcp_established_options()
948 unsigned int remaining = MAX_TCP_OPTION_SPACE - size; in tcp_established_options()
952 &opts->mptcp)) { in tcp_established_options()
953 opts->options |= OPTION_MPTCP; in tcp_established_options()
958 eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack; in tcp_established_options()
960 const unsigned int remaining = MAX_TCP_OPTION_SPACE - size; in tcp_established_options()
965 opts->num_sack_blocks = in tcp_established_options()
967 (remaining - TCPOLEN_SACK_BASE_ALIGNED) / in tcp_established_options()
971 opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK; in tcp_established_options()
974 if (unlikely(BPF_SOCK_OPS_TEST_FLAG(tp, in tcp_established_options()
976 unsigned int remaining = MAX_TCP_OPTION_SPACE - size; in tcp_established_options()
980 size = MAX_TCP_OPTION_SPACE - remaining; in tcp_established_options()
995 * The invariant being skb->truesize subtracted from sk->sk_wmem_alloc
1009 if ((1 << sk->sk_state) & in tcp_tsq_write()
1012 struct tcp_sock *tp = tcp_sk(sk); in tcp_tsq_write() local
1014 if (tp->lost_out > tp->retrans_out && in tcp_tsq_write()
1015 tp->snd_cwnd > tcp_packets_in_flight(tp)) { in tcp_tsq_write()
1016 tcp_mstamp_refresh(tp); in tcp_tsq_write()
1020 tcp_write_xmit(sk, tcp_current_mss(sk), tp->nonagle, in tcp_tsq_write()
1030 else if (!test_and_set_bit(TCP_TSQ_DEFERRED, &sk->sk_tsq_flags)) in tcp_tsq_handler()
1037 * transferring tsq->head because tcp_wfree() might
1046 struct tcp_sock *tp; in tcp_tasklet_func() local
1050 list_splice_init(&tsq->head, &list); in tcp_tasklet_func()
1054 tp = list_entry(q, struct tcp_sock, tsq_node); in tcp_tasklet_func()
1055 list_del(&tp->tsq_node); in tcp_tasklet_func()
1057 sk = (struct sock *)tp; in tcp_tasklet_func()
1059 clear_bit(TSQ_QUEUED, &sk->sk_tsq_flags); in tcp_tasklet_func()
1071 * tcp_release_cb - tcp release_sock() callback
1083 flags = sk->sk_tsq_flags; in tcp_release_cb()
1087 } while (cmpxchg(&sk->sk_tsq_flags, flags, nflags) != flags); in tcp_release_cb()
1097 * 3) socket owned by us (sk->sk_lock.owned == 1) in tcp_release_cb()
1113 inet_csk(sk)->icsk_af_ops->mtu_reduced(sk); in tcp_release_cb()
1126 INIT_LIST_HEAD(&tsq->head); in tcp_tasklet_init()
1127 tasklet_init(&tsq->tasklet, in tcp_tasklet_init()
1140 struct sock *sk = skb->sk; in tcp_wfree()
1141 struct tcp_sock *tp = tcp_sk(sk); in tcp_wfree() local
1147 WARN_ON(refcount_sub_and_test(skb->truesize - 1, &sk->sk_wmem_alloc)); in tcp_wfree()
1152 * - less callbacks to tcp_write_xmit(), reducing stress (batches) in tcp_wfree()
1153 * - chance for incoming ACK (processed by another cpu maybe) in tcp_wfree()
1154 * to migrate this flow (skb->ooo_okay will be eventually set) in tcp_wfree()
1156 if (refcount_read(&sk->sk_wmem_alloc) >= SKB_TRUESIZE(1) && this_cpu_ksoftirqd() == current) in tcp_wfree()
1159 for (oval = READ_ONCE(sk->sk_tsq_flags);; oval = nval) { in tcp_wfree()
1167 nval = cmpxchg(&sk->sk_tsq_flags, oval, nval); in tcp_wfree()
1174 empty = list_empty(&tsq->head); in tcp_wfree()
1175 list_add(&tp->tsq_node, &tsq->head); in tcp_wfree()
1177 tasklet_schedule(&tsq->tasklet); in tcp_wfree()
1190 struct tcp_sock *tp = container_of(timer, struct tcp_sock, pacing_timer); in tcp_pace_kick() local
1191 struct sock *sk = (struct sock *)tp; in tcp_pace_kick()
1202 struct tcp_sock *tp = tcp_sk(sk); in tcp_update_skb_after_send() local
1204 if (sk->sk_pacing_status != SK_PACING_NONE) { in tcp_update_skb_after_send()
1205 unsigned long rate = sk->sk_pacing_rate; in tcp_update_skb_after_send()
1208 * Note that tp->data_segs_out overflows after 2^32 packets, in tcp_update_skb_after_send()
1211 if (rate != ~0UL && rate && tp->data_segs_out >= 10) { in tcp_update_skb_after_send()
1212 u64 len_ns = div64_ul((u64)skb->len * NSEC_PER_SEC, rate); in tcp_update_skb_after_send()
1213 u64 credit = tp->tcp_wstamp_ns - prior_wstamp; in tcp_update_skb_after_send()
1216 len_ns -= min_t(u64, len_ns / 2, credit); in tcp_update_skb_after_send()
1217 tp->tcp_wstamp_ns += len_ns; in tcp_update_skb_after_send()
1220 list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue); in tcp_update_skb_after_send()
1243 struct tcp_sock *tp; in __tcp_transmit_skb() local
1254 tp = tcp_sk(sk); in __tcp_transmit_skb()
1255 prior_wstamp = tp->tcp_wstamp_ns; in __tcp_transmit_skb()
1256 tp->tcp_wstamp_ns = max(tp->tcp_wstamp_ns, tp->tcp_clock_cache); in __tcp_transmit_skb()
1257 skb->skb_mstamp_ns = tp->tcp_wstamp_ns; in __tcp_transmit_skb()
1259 TCP_SKB_CB(skb)->tx.in_flight = TCP_SKB_CB(skb)->end_seq in __tcp_transmit_skb()
1260 - tp->snd_una; in __tcp_transmit_skb()
1271 return -ENOBUFS; in __tcp_transmit_skb()
1272 /* retransmit skbs might have a non zero value in skb->dev in __tcp_transmit_skb()
1273 * because skb->dev is aliased with skb->rbnode.rb_left in __tcp_transmit_skb()
1275 skb->dev = NULL; in __tcp_transmit_skb()
1282 if (unlikely(tcb->tcp_flags & TCPHDR_SYN)) { in __tcp_transmit_skb()
1291 * and in this case it is better to delay the delivery of 1-MSS in __tcp_transmit_skb()
1296 tcb->tcp_flags |= TCPHDR_PSH; in __tcp_transmit_skb()
1304 * TODO: Ideally, in-flight pure ACK packets should not matter here. in __tcp_transmit_skb()
1305 * One way to get this would be to set skb->truesize = 2 on them. in __tcp_transmit_skb()
1307 skb->ooo_okay = sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1); in __tcp_transmit_skb()
1314 skb->pfmemalloc = 0; in __tcp_transmit_skb()
1320 skb->sk = sk; in __tcp_transmit_skb()
1321 skb->destructor = skb_is_tcp_pure_ack(skb) ? __sock_wfree : tcp_wfree; in __tcp_transmit_skb()
1323 refcount_add(skb->truesize, &sk->sk_wmem_alloc); in __tcp_transmit_skb()
1325 skb_set_dst_pending_confirm(skb, READ_ONCE(sk->sk_dst_pending_confirm)); in __tcp_transmit_skb()
1328 th = (struct tcphdr *)skb->data; in __tcp_transmit_skb()
1329 th->source = inet->inet_sport; in __tcp_transmit_skb()
1330 th->dest = inet->inet_dport; in __tcp_transmit_skb()
1331 th->seq = htonl(tcb->seq); in __tcp_transmit_skb()
1332 th->ack_seq = htonl(rcv_nxt); in __tcp_transmit_skb()
1334 tcb->tcp_flags); in __tcp_transmit_skb()
1336 th->check = 0; in __tcp_transmit_skb()
1337 th->urg_ptr = 0; in __tcp_transmit_skb()
1340 if (unlikely(tcp_urg_mode(tp) && before(tcb->seq, tp->snd_up))) { in __tcp_transmit_skb()
1341 if (before(tp->snd_up, tcb->seq + 0x10000)) { in __tcp_transmit_skb()
1342 th->urg_ptr = htons(tp->snd_up - tcb->seq); in __tcp_transmit_skb()
1343 th->urg = 1; in __tcp_transmit_skb()
1344 } else if (after(tcb->seq + 0xFFFF, tp->snd_nxt)) { in __tcp_transmit_skb()
1345 th->urg_ptr = htons(0xFFFF); in __tcp_transmit_skb()
1346 th->urg = 1; in __tcp_transmit_skb()
1350 tcp_options_write((__be32 *)(th + 1), tp, &opts); in __tcp_transmit_skb()
1351 skb_shinfo(skb)->gso_type = sk->sk_gso_type; in __tcp_transmit_skb()
1352 if (likely(!(tcb->tcp_flags & TCPHDR_SYN))) { in __tcp_transmit_skb()
1353 th->window = htons(tcp_select_window(sk)); in __tcp_transmit_skb()
1359 th->window = htons(min(tp->rcv_wnd, 65535U)); in __tcp_transmit_skb()
1365 tp->af_specific->calc_md5_hash(opts.hash_location, in __tcp_transmit_skb()
1373 INDIRECT_CALL_INET(icsk->icsk_af_ops->send_check, in __tcp_transmit_skb()
1377 if (likely(tcb->tcp_flags & TCPHDR_ACK)) in __tcp_transmit_skb()
1380 if (skb->len != tcp_header_size) { in __tcp_transmit_skb()
1381 tcp_event_data_sent(tp, sk); in __tcp_transmit_skb()
1382 tp->data_segs_out += tcp_skb_pcount(skb); in __tcp_transmit_skb()
1383 tp->bytes_sent += skb->len - tcp_header_size; in __tcp_transmit_skb()
1386 if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq) in __tcp_transmit_skb()
1390 tp->segs_out += tcp_skb_pcount(skb); in __tcp_transmit_skb()
1391 /* OK, its time to fill skb_shinfo(skb)->gso_{segs|size} */ in __tcp_transmit_skb()
1392 skb_shinfo(skb)->gso_segs = tcp_skb_pcount(skb); in __tcp_transmit_skb()
1393 skb_shinfo(skb)->gso_size = tcp_skb_mss(skb); in __tcp_transmit_skb()
1395 /* Leave earliest departure time in skb->tstamp (skb->skb_mstamp_ns) */ in __tcp_transmit_skb()
1398 memset(skb->cb, 0, max(sizeof(struct inet_skb_parm), in __tcp_transmit_skb()
1401 tcp_add_tx_delay(skb, tp); in __tcp_transmit_skb()
1403 err = INDIRECT_CALL_INET(icsk->icsk_af_ops->queue_xmit, in __tcp_transmit_skb()
1405 sk, skb, &inet->cork.fl); in __tcp_transmit_skb()
1422 tcp_sk(sk)->rcv_nxt); in tcp_transmit_skb()
1432 struct tcp_sock *tp = tcp_sk(sk); in tcp_queue_skb() local
1435 WRITE_ONCE(tp->write_seq, TCP_SKB_CB(skb)->end_seq); in tcp_queue_skb()
1438 sk_wmem_queued_add(sk, skb->truesize); in tcp_queue_skb()
1439 sk_mem_charge(sk, skb->truesize); in tcp_queue_skb()
1445 if (skb->len <= mss_now) { in tcp_set_skb_tso_segs()
1447 * non-TSO case. in tcp_set_skb_tso_segs()
1450 TCP_SKB_CB(skb)->tcp_gso_size = 0; in tcp_set_skb_tso_segs()
1452 tcp_skb_pcount_set(skb, DIV_ROUND_UP(skb->len, mss_now)); in tcp_set_skb_tso_segs()
1453 TCP_SKB_CB(skb)->tcp_gso_size = mss_now; in tcp_set_skb_tso_segs()
1462 struct tcp_sock *tp = tcp_sk(sk); in tcp_adjust_pcount() local
1464 tp->packets_out -= decr; in tcp_adjust_pcount()
1466 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) in tcp_adjust_pcount()
1467 tp->sacked_out -= decr; in tcp_adjust_pcount()
1468 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) in tcp_adjust_pcount()
1469 tp->retrans_out -= decr; in tcp_adjust_pcount()
1470 if (TCP_SKB_CB(skb)->sacked & TCPCB_LOST) in tcp_adjust_pcount()
1471 tp->lost_out -= decr; in tcp_adjust_pcount()
1474 if (tcp_is_reno(tp) && decr > 0) in tcp_adjust_pcount()
1475 tp->sacked_out -= min_t(u32, tp->sacked_out, decr); in tcp_adjust_pcount()
1477 if (tp->lost_skb_hint && in tcp_adjust_pcount()
1478 before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(tp->lost_skb_hint)->seq) && in tcp_adjust_pcount()
1479 (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) in tcp_adjust_pcount()
1480 tp->lost_cnt_hint -= decr; in tcp_adjust_pcount()
1482 tcp_verify_left_out(tp); in tcp_adjust_pcount()
1487 return TCP_SKB_CB(skb)->txstamp_ack || in tcp_has_tx_tstamp()
1488 (skb_shinfo(skb)->tx_flags & SKBTX_ANY_TSTAMP); in tcp_has_tx_tstamp()
1496 !before(shinfo->tskey, TCP_SKB_CB(skb2)->seq)) { in tcp_fragment_tstamp()
1498 u8 tsflags = shinfo->tx_flags & SKBTX_ANY_TSTAMP; in tcp_fragment_tstamp()
1500 shinfo->tx_flags &= ~tsflags; in tcp_fragment_tstamp()
1501 shinfo2->tx_flags |= tsflags; in tcp_fragment_tstamp()
1502 swap(shinfo->tskey, shinfo2->tskey); in tcp_fragment_tstamp()
1503 TCP_SKB_CB(skb2)->txstamp_ack = TCP_SKB_CB(skb)->txstamp_ack; in tcp_fragment_tstamp()
1504 TCP_SKB_CB(skb)->txstamp_ack = 0; in tcp_fragment_tstamp()
1510 TCP_SKB_CB(skb2)->eor = TCP_SKB_CB(skb)->eor; in tcp_skb_fragment_eor()
1511 TCP_SKB_CB(skb)->eor = 0; in tcp_skb_fragment_eor()
1521 __skb_queue_after(&sk->sk_write_queue, skb, buff); in tcp_insert_write_queue_after()
1523 tcp_rbtree_insert(&sk->tcp_rtx_queue, buff); in tcp_insert_write_queue_after()
1535 struct tcp_sock *tp = tcp_sk(sk); in tcp_fragment() local
1542 if (WARN_ON(len > skb->len)) in tcp_fragment()
1543 return -EINVAL; in tcp_fragment()
1545 nsize = skb_headlen(skb) - len; in tcp_fragment()
1554 limit = sk->sk_sndbuf + 2 * SKB_TRUESIZE(GSO_MAX_SIZE); in tcp_fragment()
1555 if (unlikely((sk->sk_wmem_queued >> 1) > limit && in tcp_fragment()
1560 return -ENOMEM; in tcp_fragment()
1564 return -ENOMEM; in tcp_fragment()
1569 return -ENOMEM; /* We'll just try again later. */ in tcp_fragment()
1572 sk_wmem_queued_add(sk, buff->truesize); in tcp_fragment()
1573 sk_mem_charge(sk, buff->truesize); in tcp_fragment()
1574 nlen = skb->len - len - nsize; in tcp_fragment()
1575 buff->truesize += nlen; in tcp_fragment()
1576 skb->truesize -= nlen; in tcp_fragment()
1579 TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len; in tcp_fragment()
1580 TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq; in tcp_fragment()
1581 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq; in tcp_fragment()
1584 flags = TCP_SKB_CB(skb)->tcp_flags; in tcp_fragment()
1585 TCP_SKB_CB(skb)->tcp_flags = flags & ~(TCPHDR_FIN | TCPHDR_PSH); in tcp_fragment()
1586 TCP_SKB_CB(buff)->tcp_flags = flags; in tcp_fragment()
1587 TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked; in tcp_fragment()
1592 buff->ip_summed = CHECKSUM_PARTIAL; in tcp_fragment()
1594 buff->tstamp = skb->tstamp; in tcp_fragment()
1604 TCP_SKB_CB(buff)->tx = TCP_SKB_CB(skb)->tx; in tcp_fragment()
1607 * adjust the various packet counters. in tcp_fragment()
1609 if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) { in tcp_fragment()
1610 int diff = old_factor - tcp_skb_pcount(skb) - in tcp_fragment()
1621 list_add(&buff->tcp_tsorted_anchor, &skb->tcp_tsorted_anchor); in tcp_fragment()
1637 len -= eat; in __pskb_trim_head()
1644 for (i = 0; i < shinfo->nr_frags; i++) { in __pskb_trim_head()
1645 int size = skb_frag_size(&shinfo->frags[i]); in __pskb_trim_head()
1649 eat -= size; in __pskb_trim_head()
1651 shinfo->frags[k] = shinfo->frags[i]; in __pskb_trim_head()
1653 skb_frag_off_add(&shinfo->frags[k], eat); in __pskb_trim_head()
1654 skb_frag_size_sub(&shinfo->frags[k], eat); in __pskb_trim_head()
1660 shinfo->nr_frags = k; in __pskb_trim_head()
1662 skb->data_len -= len; in __pskb_trim_head()
1663 skb->len = skb->data_len; in __pskb_trim_head()
1673 return -ENOMEM; in tcp_trim_head()
1677 TCP_SKB_CB(skb)->seq += len; in tcp_trim_head()
1678 skb->ip_summed = CHECKSUM_PARTIAL; in tcp_trim_head()
1681 skb->truesize -= delta_truesize; in tcp_trim_head()
1682 sk_wmem_queued_add(sk, -delta_truesize); in tcp_trim_head()
1686 /* Any change of skb->len requires recalculation of tso factor. */ in tcp_trim_head()
1696 const struct tcp_sock *tp = tcp_sk(sk); in __tcp_mtu_to_mss() local
1701 It is MMS_S - sizeof(tcphdr) of rfc1122 in __tcp_mtu_to_mss()
1703 mss_now = pmtu - icsk->icsk_af_ops->net_header_len - sizeof(struct tcphdr); in __tcp_mtu_to_mss()
1706 if (icsk->icsk_af_ops->net_frag_header_len) { in __tcp_mtu_to_mss()
1710 mss_now -= icsk->icsk_af_ops->net_frag_header_len; in __tcp_mtu_to_mss()
1714 if (mss_now > tp->rx_opt.mss_clamp) in __tcp_mtu_to_mss()
1715 mss_now = tp->rx_opt.mss_clamp; in __tcp_mtu_to_mss()
1718 mss_now -= icsk->icsk_ext_hdr_len; in __tcp_mtu_to_mss()
1722 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_snd_mss)); in __tcp_mtu_to_mss()
1730 return __tcp_mtu_to_mss(sk, pmtu) - in tcp_mtu_to_mss()
1731 (tcp_sk(sk)->tcp_header_len - sizeof(struct tcphdr)); in tcp_mtu_to_mss()
1738 const struct tcp_sock *tp = tcp_sk(sk); in tcp_mss_to_mtu() local
1743 tp->tcp_header_len + in tcp_mss_to_mtu()
1744 icsk->icsk_ext_hdr_len + in tcp_mss_to_mtu()
1745 icsk->icsk_af_ops->net_header_len; in tcp_mss_to_mtu()
1748 if (icsk->icsk_af_ops->net_frag_header_len) { in tcp_mss_to_mtu()
1752 mtu += icsk->icsk_af_ops->net_frag_header_len; in tcp_mss_to_mtu()
1761 struct tcp_sock *tp = tcp_sk(sk); in tcp_mtup_init() local
1765 icsk->icsk_mtup.enabled = READ_ONCE(net->ipv4.sysctl_tcp_mtu_probing) > 1; in tcp_mtup_init()
1766 icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) + in tcp_mtup_init()
1767 icsk->icsk_af_ops->net_header_len; in tcp_mtup_init()
1768 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, READ_ONCE(net->ipv4.sysctl_tcp_base_mss)); in tcp_mtup_init()
1769 icsk->icsk_mtup.probe_size = 0; in tcp_mtup_init()
1770 if (icsk->icsk_mtup.enabled) in tcp_mtup_init()
1771 icsk->icsk_mtup.probe_timestamp = tcp_jiffies32; in tcp_mtup_init()
1777 tp->rx_opt.user_mss is mss set by user by TCP_MAXSEG. It does NOT counts
1780 tp->rx_opt.mss_clamp is mss negotiated at connection setup.
1784 inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
1786 tp->mss_cache is current effective sending mss, including
1789 tp->rx_opt.mss_clamp.
1794 NOTE2. inet_csk(sk)->icsk_pmtu_cookie and tp->mss_cache
1795 are READ ONLY outside this function. --ANK (980731)
1799 struct tcp_sock *tp = tcp_sk(sk); in tcp_sync_mss() local
1803 if (icsk->icsk_mtup.search_high > pmtu) in tcp_sync_mss()
1804 icsk->icsk_mtup.search_high = pmtu; in tcp_sync_mss()
1807 mss_now = tcp_bound_to_half_wnd(tp, mss_now); in tcp_sync_mss()
1810 icsk->icsk_pmtu_cookie = pmtu; in tcp_sync_mss()
1811 if (icsk->icsk_mtup.enabled) in tcp_sync_mss()
1812 mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low)); in tcp_sync_mss()
1813 tp->mss_cache = mss_now; in tcp_sync_mss()
1824 const struct tcp_sock *tp = tcp_sk(sk); in tcp_current_mss() local
1831 mss_now = tp->mss_cache; in tcp_current_mss()
1835 if (mtu != inet_csk(sk)->icsk_pmtu_cookie) in tcp_current_mss()
1841 /* The mss_cache is sized based on tp->tcp_header_len, which assumes in tcp_current_mss()
1844 * we have to adjust mss_now correspondingly */ in tcp_current_mss()
1845 if (header_len != tp->tcp_header_len) { in tcp_current_mss()
1846 int delta = (int) header_len - tp->tcp_header_len; in tcp_current_mss()
1847 mss_now -= delta; in tcp_current_mss()
1853 /* RFC2861, slow part. Adjust cwnd, after it was not full during one rto.
1859 struct tcp_sock *tp = tcp_sk(sk); in tcp_cwnd_application_limited() local
1861 if (inet_csk(sk)->icsk_ca_state == TCP_CA_Open && in tcp_cwnd_application_limited()
1862 sk->sk_socket && !test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { in tcp_cwnd_application_limited()
1864 u32 init_win = tcp_init_cwnd(tp, __sk_dst_get(sk)); in tcp_cwnd_application_limited()
1865 u32 win_used = max(tp->snd_cwnd_used, init_win); in tcp_cwnd_application_limited()
1866 if (win_used < tp->snd_cwnd) { in tcp_cwnd_application_limited()
1867 tp->snd_ssthresh = tcp_current_ssthresh(sk); in tcp_cwnd_application_limited()
1868 tp->snd_cwnd = (tp->snd_cwnd + win_used) >> 1; in tcp_cwnd_application_limited()
1870 tp->snd_cwnd_used = 0; in tcp_cwnd_application_limited()
1872 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_cwnd_application_limited()
1877 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; in tcp_cwnd_validate()
1878 struct tcp_sock *tp = tcp_sk(sk); in tcp_cwnd_validate() local
1881 * is fully utilized. If cwnd-limited then remember that fact for the in tcp_cwnd_validate()
1882 * current window. If not cwnd-limited then track the maximum number of in tcp_cwnd_validate()
1883 * outstanding packets in the current window. (If cwnd-limited then we in tcp_cwnd_validate()
1884 * chose to not update tp->max_packets_out to avoid an extra else in tcp_cwnd_validate()
1887 if (!before(tp->snd_una, tp->cwnd_usage_seq) || in tcp_cwnd_validate()
1889 (!tp->is_cwnd_limited && in tcp_cwnd_validate()
1890 tp->packets_out > tp->max_packets_out)) { in tcp_cwnd_validate()
1891 tp->is_cwnd_limited = is_cwnd_limited; in tcp_cwnd_validate()
1892 tp->max_packets_out = tp->packets_out; in tcp_cwnd_validate()
1893 tp->cwnd_usage_seq = tp->snd_nxt; in tcp_cwnd_validate()
1898 tp->snd_cwnd_used = 0; in tcp_cwnd_validate()
1899 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_cwnd_validate()
1902 if (tp->packets_out > tp->snd_cwnd_used) in tcp_cwnd_validate()
1903 tp->snd_cwnd_used = tp->packets_out; in tcp_cwnd_validate()
1905 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) && in tcp_cwnd_validate()
1906 (s32)(tcp_jiffies32 - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto && in tcp_cwnd_validate()
1907 !ca_ops->cong_control) in tcp_cwnd_validate()
1917 if (tcp_write_queue_empty(sk) && sk->sk_socket && in tcp_cwnd_validate()
1918 test_bit(SOCK_NOSPACE, &sk->sk_socket->flags) && in tcp_cwnd_validate()
1919 (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) in tcp_cwnd_validate()
1925 static bool tcp_minshall_check(const struct tcp_sock *tp) in tcp_minshall_check() argument
1927 return after(tp->snd_sml, tp->snd_una) && in tcp_minshall_check()
1928 !after(tp->snd_sml, tp->snd_nxt); in tcp_minshall_check()
1932 * Note that a TSO packet might end with a sub-mss segment
1934 * if ((skb->len % mss) != 0)
1935 * tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1937 * skb_pcount = skb->len / mss_now
1939 static void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss_now, in tcp_minshall_update() argument
1942 if (skb->len < tcp_skb_pcount(skb) * mss_now) in tcp_minshall_update()
1943 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; in tcp_minshall_update()
1953 static bool tcp_nagle_check(bool partial, const struct tcp_sock *tp, in tcp_nagle_check() argument
1958 (!nonagle && tp->packets_out && tcp_minshall_check(tp))); in tcp_nagle_check()
1970 sk->sk_pacing_rate >> READ_ONCE(sk->sk_pacing_shift), in tcp_tso_autosize()
1971 sk->sk_gso_max_size - 1 - MAX_TCP_HEADER); in tcp_tso_autosize()
1988 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; in tcp_tso_segs()
1991 min_tso = ca_ops->min_tso_segs ? in tcp_tso_segs()
1992 ca_ops->min_tso_segs(sk) : in tcp_tso_segs()
1993 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs); in tcp_tso_segs()
1996 return min_t(u32, tso_segs, sk->sk_gso_max_segs); in tcp_tso_segs()
2006 const struct tcp_sock *tp = tcp_sk(sk); in tcp_mss_split_point() local
2009 window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in tcp_mss_split_point()
2015 needed = min(skb->len, window); in tcp_mss_split_point()
2025 if (tcp_nagle_check(partial != 0, tp, nonagle)) in tcp_mss_split_point()
2026 return needed - partial; in tcp_mss_split_point()
2034 static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp, in tcp_cwnd_test() argument
2040 if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) && in tcp_cwnd_test()
2044 in_flight = tcp_packets_in_flight(tp); in tcp_cwnd_test()
2045 cwnd = tp->snd_cwnd; in tcp_cwnd_test()
2053 return min(halfcwnd, cwnd - in_flight); in tcp_cwnd_test()
2075 static inline bool tcp_nagle_test(const struct tcp_sock *tp, const struct sk_buff *skb, in tcp_nagle_test() argument
2088 if (tcp_urg_mode(tp) || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) in tcp_nagle_test()
2091 if (!tcp_nagle_check(skb->len < cur_mss, tp, nonagle)) in tcp_nagle_test()
2098 static bool tcp_snd_wnd_test(const struct tcp_sock *tp, in tcp_snd_wnd_test() argument
2102 u32 end_seq = TCP_SKB_CB(skb)->end_seq; in tcp_snd_wnd_test()
2104 if (skb->len > cur_mss) in tcp_snd_wnd_test()
2105 end_seq = TCP_SKB_CB(skb)->seq + cur_mss; in tcp_snd_wnd_test()
2107 return !after(end_seq, tcp_wnd_end(tp)); in tcp_snd_wnd_test()
2114 * know that all the data is in scatter-gather pages, and that the
2120 int nlen = skb->len - len; in tso_fragment()
2125 if (skb->len != skb->data_len) in tso_fragment()
2131 return -ENOMEM; in tso_fragment()
2134 sk_wmem_queued_add(sk, buff->truesize); in tso_fragment()
2135 sk_mem_charge(sk, buff->truesize); in tso_fragment()
2136 buff->truesize += nlen; in tso_fragment()
2137 skb->truesize -= nlen; in tso_fragment()
2140 TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len; in tso_fragment()
2141 TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq; in tso_fragment()
2142 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq; in tso_fragment()
2145 flags = TCP_SKB_CB(skb)->tcp_flags; in tso_fragment()
2146 TCP_SKB_CB(skb)->tcp_flags = flags & ~(TCPHDR_FIN | TCPHDR_PSH); in tso_fragment()
2147 TCP_SKB_CB(buff)->tcp_flags = flags; in tso_fragment()
2150 TCP_SKB_CB(buff)->sacked = 0; in tso_fragment()
2154 buff->ip_summed = CHECKSUM_PARTIAL; in tso_fragment()
2181 struct tcp_sock *tp = tcp_sk(sk); in tcp_tso_should_defer() local
2186 if (icsk->icsk_ca_state >= TCP_CA_Recovery) in tcp_tso_should_defer()
2191 * Note that tp->tcp_wstamp_ns can be in the future if we have in tcp_tso_should_defer()
2194 delta = tp->tcp_clock_cache - tp->tcp_wstamp_ns - NSEC_PER_MSEC; in tcp_tso_should_defer()
2198 in_flight = tcp_packets_in_flight(tp); in tcp_tso_should_defer()
2201 BUG_ON(tp->snd_cwnd <= in_flight); in tcp_tso_should_defer()
2203 send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in tcp_tso_should_defer()
2206 cong_win = (tp->snd_cwnd - in_flight) * tp->mss_cache; in tcp_tso_should_defer()
2210 /* If a full-sized TSO skb can be sent, do it. */ in tcp_tso_should_defer()
2211 if (limit >= max_segs * tp->mss_cache) in tcp_tso_should_defer()
2215 if ((skb != tcp_write_queue_tail(sk)) && (limit >= skb->len)) in tcp_tso_should_defer()
2218 win_divisor = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tso_win_divisor); in tcp_tso_should_defer()
2220 u32 chunk = min(tp->snd_wnd, tp->snd_cwnd * tp->mss_cache); in tcp_tso_should_defer()
2234 if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache) in tcp_tso_should_defer()
2242 delta = tp->tcp_clock_cache - head->tstamp; in tcp_tso_should_defer()
2244 if ((s64)(delta - (u64)NSEC_PER_USEC * (tp->srtt_us >> 4)) < 0) in tcp_tso_should_defer()
2249 * 1) We are cwnd-limited in tcp_tso_should_defer()
2250 * 2) We are rwnd-limited in tcp_tso_should_defer()
2254 if (cong_win <= skb->len) { in tcp_tso_should_defer()
2259 if (send_win <= skb->len) { in tcp_tso_should_defer()
2266 if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) || in tcp_tso_should_defer()
2267 TCP_SKB_CB(skb)->eor) in tcp_tso_should_defer()
2279 struct tcp_sock *tp = tcp_sk(sk); in tcp_mtu_check_reprobe() local
2284 interval = READ_ONCE(net->ipv4.sysctl_tcp_probe_interval); in tcp_mtu_check_reprobe()
2285 delta = tcp_jiffies32 - icsk->icsk_mtup.probe_timestamp; in tcp_mtu_check_reprobe()
2290 icsk->icsk_mtup.probe_size = 0; in tcp_mtu_check_reprobe()
2291 icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + in tcp_mtu_check_reprobe()
2293 icsk->icsk_af_ops->net_header_len; in tcp_mtu_check_reprobe()
2294 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss); in tcp_mtu_check_reprobe()
2297 icsk->icsk_mtup.probe_timestamp = tcp_jiffies32; in tcp_mtu_check_reprobe()
2307 if (len <= skb->len) in tcp_can_coalesce_send_queue_head()
2310 if (unlikely(TCP_SKB_CB(skb)->eor) || tcp_has_tx_tstamp(skb)) in tcp_can_coalesce_send_queue_head()
2313 len -= skb->len; in tcp_can_coalesce_send_queue_head()
2326 * -1 otherwise
2331 struct tcp_sock *tp = tcp_sk(sk); in tcp_mtu_probe() local
2345 if (likely(!icsk->icsk_mtup.enabled || in tcp_mtu_probe()
2346 icsk->icsk_mtup.probe_size || in tcp_mtu_probe()
2347 inet_csk(sk)->icsk_ca_state != TCP_CA_Open || in tcp_mtu_probe()
2348 tp->snd_cwnd < 11 || in tcp_mtu_probe()
2349 tp->rx_opt.num_sacks || tp->rx_opt.dsack)) in tcp_mtu_probe()
2350 return -1; in tcp_mtu_probe()
2353 * and current mss_clamp. if (search_high - search_low) in tcp_mtu_probe()
2357 probe_size = tcp_mtu_to_mss(sk, (icsk->icsk_mtup.search_high + in tcp_mtu_probe()
2358 icsk->icsk_mtup.search_low) >> 1); in tcp_mtu_probe()
2359 size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache; in tcp_mtu_probe()
2360 interval = icsk->icsk_mtup.search_high - icsk->icsk_mtup.search_low; in tcp_mtu_probe()
2365 if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high) || in tcp_mtu_probe()
2366 interval < READ_ONCE(net->ipv4.sysctl_tcp_probe_threshold)) { in tcp_mtu_probe()
2371 return -1; in tcp_mtu_probe()
2375 if (tp->write_seq - tp->snd_nxt < size_needed) in tcp_mtu_probe()
2376 return -1; in tcp_mtu_probe()
2378 if (tp->snd_wnd < size_needed) in tcp_mtu_probe()
2379 return -1; in tcp_mtu_probe()
2380 if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp))) in tcp_mtu_probe()
2384 if (tcp_packets_in_flight(tp) + 2 > tp->snd_cwnd) { in tcp_mtu_probe()
2385 if (!tcp_packets_in_flight(tp)) in tcp_mtu_probe()
2386 return -1; in tcp_mtu_probe()
2392 return -1; in tcp_mtu_probe()
2397 return -1; in tcp_mtu_probe()
2398 sk_wmem_queued_add(sk, nskb->truesize); in tcp_mtu_probe()
2399 sk_mem_charge(sk, nskb->truesize); in tcp_mtu_probe()
2404 TCP_SKB_CB(nskb)->seq = TCP_SKB_CB(skb)->seq; in tcp_mtu_probe()
2405 TCP_SKB_CB(nskb)->end_seq = TCP_SKB_CB(skb)->seq + probe_size; in tcp_mtu_probe()
2406 TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK; in tcp_mtu_probe()
2407 TCP_SKB_CB(nskb)->sacked = 0; in tcp_mtu_probe()
2408 nskb->csum = 0; in tcp_mtu_probe()
2409 nskb->ip_summed = CHECKSUM_PARTIAL; in tcp_mtu_probe()
2416 copy = min_t(int, skb->len, probe_size - len); in tcp_mtu_probe()
2419 if (skb->len <= copy) { in tcp_mtu_probe()
2422 TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags; in tcp_mtu_probe()
2426 TCP_SKB_CB(nskb)->eor = TCP_SKB_CB(skb)->eor; in tcp_mtu_probe()
2431 TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags & in tcp_mtu_probe()
2433 if (!skb_shinfo(skb)->nr_frags) { in tcp_mtu_probe()
2439 TCP_SKB_CB(skb)->seq += copy; in tcp_mtu_probe()
2447 tcp_init_tso_segs(nskb, nskb->len); in tcp_mtu_probe()
2450 * be resegmented into mss-sized pieces by tcp_write_xmit(). in tcp_mtu_probe()
2455 tp->snd_cwnd--; in tcp_mtu_probe()
2458 icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len); in tcp_mtu_probe()
2459 tp->mtu_probe.probe_seq_start = TCP_SKB_CB(nskb)->seq; in tcp_mtu_probe()
2460 tp->mtu_probe.probe_seq_end = TCP_SKB_CB(nskb)->end_seq; in tcp_mtu_probe()
2465 return -1; in tcp_mtu_probe()
2470 struct tcp_sock *tp = tcp_sk(sk); in tcp_pacing_check() local
2475 if (tp->tcp_wstamp_ns <= tp->tcp_clock_cache) in tcp_pacing_check()
2478 if (!hrtimer_is_queued(&tp->pacing_timer)) { in tcp_pacing_check()
2479 hrtimer_start(&tp->pacing_timer, in tcp_pacing_check()
2480 ns_to_ktime(tp->tcp_wstamp_ns), in tcp_pacing_check()
2489 const struct rb_node *node = sk->tcp_rtx_queue.rb_node; in tcp_rtx_queue_empty_or_single_skb()
2496 return !node->rb_left && !node->rb_right; in tcp_rtx_queue_empty_or_single_skb()
2503 * - better RTT estimation and ACK scheduling
2504 * - faster recovery
2505 * - high rates
2516 2 * skb->truesize, in tcp_small_queue_check()
2517 sk->sk_pacing_rate >> READ_ONCE(sk->sk_pacing_shift)); in tcp_small_queue_check()
2518 if (sk->sk_pacing_status == SK_PACING_NONE) in tcp_small_queue_check()
2520 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_limit_output_bytes)); in tcp_small_queue_check()
2524 tcp_sk(sk)->tcp_tx_delay) { in tcp_small_queue_check()
2525 u64 extra_bytes = (u64)sk->sk_pacing_rate * tcp_sk(sk)->tcp_tx_delay; in tcp_small_queue_check()
2528 * approximate our needs assuming an ~100% skb->truesize overhead. in tcp_small_queue_check()
2532 extra_bytes >>= (20 - 1); in tcp_small_queue_check()
2535 if (refcount_read(&sk->sk_wmem_alloc) > limit) { in tcp_small_queue_check()
2544 set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags); in tcp_small_queue_check()
2550 if (refcount_read(&sk->sk_wmem_alloc) > limit) in tcp_small_queue_check()
2556 static void tcp_chrono_set(struct tcp_sock *tp, const enum tcp_chrono new) in tcp_chrono_set() argument
2559 enum tcp_chrono old = tp->chrono_type; in tcp_chrono_set()
2562 tp->chrono_stat[old - 1] += now - tp->chrono_start; in tcp_chrono_set()
2563 tp->chrono_start = now; in tcp_chrono_set()
2564 tp->chrono_type = new; in tcp_chrono_set()
2569 struct tcp_sock *tp = tcp_sk(sk); in tcp_chrono_start() local
2576 if (type > tp->chrono_type) in tcp_chrono_start()
2577 tcp_chrono_set(tp, type); in tcp_chrono_start()
2582 struct tcp_sock *tp = tcp_sk(sk); in tcp_chrono_stop() local
2593 tcp_chrono_set(tp, TCP_CHRONO_UNSPEC); in tcp_chrono_stop()
2594 else if (type == tp->chrono_type) in tcp_chrono_stop()
2595 tcp_chrono_set(tp, TCP_CHRONO_BUSY); in tcp_chrono_stop()
2603 * snd_up-64k-mss .. snd_up cannot be large. However, taking into
2615 struct tcp_sock *tp = tcp_sk(sk); in tcp_write_xmit() local
2625 tcp_mstamp_refresh(tp); in tcp_write_xmit()
2640 if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE) { in tcp_write_xmit()
2642 skb->skb_mstamp_ns = tp->tcp_wstamp_ns = tp->tcp_clock_cache; in tcp_write_xmit()
2643 list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue); in tcp_write_xmit()
2654 cwnd_quota = tcp_cwnd_test(tp, skb); in tcp_write_xmit()
2663 if (unlikely(!tcp_snd_wnd_test(tp, skb, mss_now))) { in tcp_write_xmit()
2669 if (unlikely(!tcp_nagle_test(tp, skb, mss_now, in tcp_write_xmit()
2681 if (tso_segs > 1 && !tcp_urg_mode(tp)) in tcp_write_xmit()
2688 if (skb->len > limit && in tcp_write_xmit()
2697 * We do not want to send a pure-ack packet and have in tcp_write_xmit()
2700 if (TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) in tcp_write_xmit()
2712 tcp_minshall_update(tp, mss_now, skb); in tcp_write_xmit()
2724 is_cwnd_limited |= (tcp_packets_in_flight(tp) >= tp->snd_cwnd); in tcp_write_xmit()
2730 tp->prr_out += sent_pkts; in tcp_write_xmit()
2737 return !tp->packets_out && !tcp_write_queue_empty(sk); in tcp_write_xmit()
2743 struct tcp_sock *tp = tcp_sk(sk); in tcp_schedule_loss_probe() local
2750 if (rcu_access_pointer(tp->fastopen_rsk)) in tcp_schedule_loss_probe()
2753 early_retrans = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_early_retrans); in tcp_schedule_loss_probe()
2758 !tp->packets_out || !tcp_is_sack(tp) || in tcp_schedule_loss_probe()
2759 (icsk->icsk_ca_state != TCP_CA_Open && in tcp_schedule_loss_probe()
2760 icsk->icsk_ca_state != TCP_CA_CWR)) in tcp_schedule_loss_probe()
2767 if (tp->srtt_us) { in tcp_schedule_loss_probe()
2768 timeout_us = tp->srtt_us >> 2; in tcp_schedule_loss_probe()
2769 if (tp->packets_out == 1) in tcp_schedule_loss_probe()
2780 jiffies_to_usecs(inet_csk(sk)->icsk_rto) : in tcp_schedule_loss_probe()
2809 struct tcp_sock *tp = tcp_sk(sk); in tcp_send_loss_probe() local
2815 if (tp->tlp_high_seq) in tcp_send_loss_probe()
2818 tp->tlp_retrans = 0; in tcp_send_loss_probe()
2820 if (skb && tcp_snd_wnd_test(tp, skb, mss)) { in tcp_send_loss_probe()
2821 pcount = tp->packets_out; in tcp_send_loss_probe()
2823 if (tp->packets_out > pcount) in tcp_send_loss_probe()
2827 skb = skb_rb_last(&sk->tcp_rtx_queue); in tcp_send_loss_probe()
2829 WARN_ONCE(tp->packets_out, in tcp_send_loss_probe()
2831 tp->packets_out, sk->sk_state, tp->snd_cwnd, mss); in tcp_send_loss_probe()
2832 inet_csk(sk)->icsk_pending = 0; in tcp_send_loss_probe()
2843 if ((pcount > 1) && (skb->len > (pcount - 1) * mss)) { in tcp_send_loss_probe()
2845 (pcount - 1) * mss, mss, in tcp_send_loss_probe()
2857 tp->tlp_retrans = 1; in tcp_send_loss_probe()
2861 tp->tlp_high_seq = tp->snd_nxt; in tcp_send_loss_probe()
2865 inet_csk(sk)->icsk_pending = 0; in tcp_send_loss_probe()
2881 if (unlikely(sk->sk_state == TCP_CLOSE)) in __tcp_push_pending_frames()
2896 BUG_ON(!skb || skb->len < mss_now); in tcp_push_one()
2898 tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation); in tcp_push_one()
2910 * RCV.BUFF - RCV.USER - RCV.WINDOW >= min(1/2 RCV.BUFF, MSS)"
2916 * since header prediction assumes th->window stays fixed.
2918 * Strictly speaking, keeping th->window fixed violates the receiver
2950 * Note, we don't "adjust" for TIMESTAMP or SACK option bytes.
2956 struct tcp_sock *tp = tcp_sk(sk); in __tcp_select_window() local
2961 * fluctuations. --SAW 1998/11/1 in __tcp_select_window()
2963 int mss = icsk->icsk_ack.rcv_mss; in __tcp_select_window()
2971 full_space = min_t(int, tp->window_clamp, allowed_space); in __tcp_select_window()
2979 icsk->icsk_ack.quick = 0; in __tcp_select_window()
2982 tp->rcv_ssthresh = min(tp->rcv_ssthresh, in __tcp_select_window()
2983 4U * tp->advmss); in __tcp_select_window()
2988 free_space = round_down(free_space, 1 << tp->rx_opt.rcv_wscale); in __tcp_select_window()
2991 * of the maximum allowed, try to move to zero-window, else in __tcp_select_window()
3001 if (free_space > tp->rcv_ssthresh) in __tcp_select_window()
3002 free_space = tp->rcv_ssthresh; in __tcp_select_window()
3007 if (tp->rx_opt.rcv_wscale) { in __tcp_select_window()
3014 window = ALIGN(window, (1 << tp->rx_opt.rcv_wscale)); in __tcp_select_window()
3016 window = tp->rcv_wnd; in __tcp_select_window()
3025 if (window <= free_space - mss || window > free_space) in __tcp_select_window()
3043 shinfo->tx_flags |= next_shinfo->tx_flags & SKBTX_ANY_TSTAMP; in tcp_skb_collapse_tstamp()
3044 shinfo->tskey = next_shinfo->tskey; in tcp_skb_collapse_tstamp()
3045 TCP_SKB_CB(skb)->txstamp_ack |= in tcp_skb_collapse_tstamp()
3046 TCP_SKB_CB(next_skb)->txstamp_ack; in tcp_skb_collapse_tstamp()
3053 struct tcp_sock *tp = tcp_sk(sk); in tcp_collapse_retrans() local
3057 next_skb_size = next_skb->len; in tcp_collapse_retrans()
3071 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(next_skb)->end_seq; in tcp_collapse_retrans()
3074 TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags; in tcp_collapse_retrans()
3079 TCP_SKB_CB(skb)->sacked |= TCP_SKB_CB(next_skb)->sacked & TCPCB_EVER_RETRANS; in tcp_collapse_retrans()
3080 TCP_SKB_CB(skb)->eor = TCP_SKB_CB(next_skb)->eor; in tcp_collapse_retrans()
3083 tcp_clear_retrans_hints_partial(tp); in tcp_collapse_retrans()
3084 if (next_skb == tp->retransmit_skb_hint) in tcp_collapse_retrans()
3085 tp->retransmit_skb_hint = skb; in tcp_collapse_retrans()
3103 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) in tcp_can_collapse()
3115 struct tcp_sock *tp = tcp_sk(sk); in tcp_retrans_try_collapse() local
3119 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_retrans_collapse)) in tcp_retrans_try_collapse()
3121 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN) in tcp_retrans_try_collapse()
3131 space -= skb->len; in tcp_retrans_try_collapse()
3141 if (after(TCP_SKB_CB(skb)->end_seq, tcp_wnd_end(tp))) in tcp_retrans_try_collapse()
3150 * state updates are done by the caller. Returns non-zero if an
3156 struct tcp_sock *tp = tcp_sk(sk); in __tcp_retransmit_skb() local
3162 if (icsk->icsk_mtup.probe_size) in __tcp_retransmit_skb()
3163 icsk->icsk_mtup.probe_size = 0; in __tcp_retransmit_skb()
3168 if (refcount_read(&sk->sk_wmem_alloc) > in __tcp_retransmit_skb()
3169 min_t(u32, sk->sk_wmem_queued + (sk->sk_wmem_queued >> 2), in __tcp_retransmit_skb()
3170 sk->sk_sndbuf)) in __tcp_retransmit_skb()
3171 return -EAGAIN; in __tcp_retransmit_skb()
3174 return -EBUSY; in __tcp_retransmit_skb()
3177 if (before(TCP_SKB_CB(skb)->seq, tp->snd_una)) { in __tcp_retransmit_skb()
3178 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { in __tcp_retransmit_skb()
3179 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_SYN; in __tcp_retransmit_skb()
3180 TCP_SKB_CB(skb)->seq++; in __tcp_retransmit_skb()
3183 if (unlikely(before(TCP_SKB_CB(skb)->end_seq, tp->snd_una))) { in __tcp_retransmit_skb()
3185 return -EINVAL; in __tcp_retransmit_skb()
3187 if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq)) in __tcp_retransmit_skb()
3188 return -ENOMEM; in __tcp_retransmit_skb()
3191 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk)) in __tcp_retransmit_skb()
3192 return -EHOSTUNREACH; /* Routing failure or similar. */ in __tcp_retransmit_skb()
3195 avail_wnd = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in __tcp_retransmit_skb()
3203 if (TCP_SKB_CB(skb)->seq != tp->snd_una) in __tcp_retransmit_skb()
3204 return -EAGAIN; in __tcp_retransmit_skb()
3214 if (skb->len > len) { in __tcp_retransmit_skb()
3217 return -ENOMEM; /* We'll try again later. */ in __tcp_retransmit_skb()
3220 return -ENOMEM; in __tcp_retransmit_skb()
3224 diff -= tcp_skb_pcount(skb); in __tcp_retransmit_skb()
3228 if (skb->len < avail_wnd) in __tcp_retransmit_skb()
3233 if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN_ECN) == TCPHDR_SYN_ECN) in __tcp_retransmit_skb()
3239 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN) in __tcp_retransmit_skb()
3241 tp->total_retrans += segs; in __tcp_retransmit_skb()
3242 tp->bytes_retrans += skb->len; in __tcp_retransmit_skb()
3244 /* make sure skb->data is aligned on arches that require it in __tcp_retransmit_skb()
3245 * and check if ack-trimming & collapsing extended the headroom in __tcp_retransmit_skb()
3248 if (unlikely((NET_IP_ALIGN && ((unsigned long)skb->data & 3)) || in __tcp_retransmit_skb()
3255 nskb->dev = NULL; in __tcp_retransmit_skb()
3258 err = -ENOBUFS; in __tcp_retransmit_skb()
3263 tcp_update_skb_after_send(sk, skb, tp->tcp_wstamp_ns); in __tcp_retransmit_skb()
3273 TCP_SKB_CB(skb)->sacked |= TCPCB_EVER_RETRANS; in __tcp_retransmit_skb()
3275 if (BPF_SOCK_OPS_TEST_FLAG(tp, BPF_SOCK_OPS_RETRANS_CB_FLAG)) in __tcp_retransmit_skb()
3277 TCP_SKB_CB(skb)->seq, segs, err); in __tcp_retransmit_skb()
3281 } else if (err != -EBUSY) { in __tcp_retransmit_skb()
3289 struct tcp_sock *tp = tcp_sk(sk); in tcp_retransmit_skb() local
3294 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) { in tcp_retransmit_skb()
3298 TCP_SKB_CB(skb)->sacked |= TCPCB_RETRANS; in tcp_retransmit_skb()
3299 tp->retrans_out += tcp_skb_pcount(skb); in tcp_retransmit_skb()
3303 if (!tp->retrans_stamp) in tcp_retransmit_skb()
3304 tp->retrans_stamp = tcp_skb_timestamp(skb); in tcp_retransmit_skb()
3306 if (tp->undo_retrans < 0) in tcp_retransmit_skb()
3307 tp->undo_retrans = 0; in tcp_retransmit_skb()
3308 tp->undo_retrans += tcp_skb_pcount(skb); in tcp_retransmit_skb()
3321 struct tcp_sock *tp = tcp_sk(sk); in tcp_xmit_retransmit_queue() local
3326 if (!tp->packets_out) in tcp_xmit_retransmit_queue()
3330 skb = tp->retransmit_skb_hint ?: rtx_head; in tcp_xmit_retransmit_queue()
3341 tp->retransmit_skb_hint = skb; in tcp_xmit_retransmit_queue()
3343 segs = tp->snd_cwnd - tcp_packets_in_flight(tp); in tcp_xmit_retransmit_queue()
3346 sacked = TCP_SKB_CB(skb)->sacked; in tcp_xmit_retransmit_queue()
3352 if (tp->retrans_out >= tp->lost_out) { in tcp_xmit_retransmit_queue()
3360 if (icsk->icsk_ca_state != TCP_CA_Loss) in tcp_xmit_retransmit_queue()
3378 tp->prr_out += tcp_skb_pcount(skb); in tcp_xmit_retransmit_queue()
3381 icsk->icsk_pending != ICSK_TIME_REO_TIMEOUT) in tcp_xmit_retransmit_queue()
3387 inet_csk(sk)->icsk_rto, in tcp_xmit_retransmit_queue()
3402 delta = size - sk->sk_forward_alloc; in sk_forced_mem_schedule()
3406 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM; in sk_forced_mem_schedule()
3409 if (mem_cgroup_sockets_enabled && sk->sk_memcg) in sk_forced_mem_schedule()
3410 mem_cgroup_charge_skmem(sk->sk_memcg, amt); in sk_forced_mem_schedule()
3419 struct tcp_sock *tp = tcp_sk(sk); in tcp_send_fin() local
3428 tskb = skb_rb_last(&sk->tcp_rtx_queue); in tcp_send_fin()
3431 TCP_SKB_CB(tskb)->tcp_flags |= TCPHDR_FIN; in tcp_send_fin()
3432 TCP_SKB_CB(tskb)->end_seq++; in tcp_send_fin()
3433 tp->write_seq++; in tcp_send_fin()
3437 * We need to set tp->snd_nxt to the value it would have in tcp_send_fin()
3439 * does not change tp->snd_nxt. in tcp_send_fin()
3441 WRITE_ONCE(tp->snd_nxt, tp->snd_nxt + 1); in tcp_send_fin()
3451 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor); in tcp_send_fin()
3453 sk_forced_mem_schedule(sk, skb->truesize); in tcp_send_fin()
3455 tcp_init_nondata_skb(skb, tp->write_seq, in tcp_send_fin()
3465 * by RFC 2525, section 2.17. -DaveM
3495 /* Send a crossed SYN-ACK during socket establishment.
3506 if (!skb || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { in tcp_send_synack()
3508 return -EFAULT; in tcp_send_synack()
3510 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) { in tcp_send_synack()
3518 return -ENOMEM; in tcp_send_synack()
3519 INIT_LIST_HEAD(&nskb->tcp_tsorted_anchor); in tcp_send_synack()
3523 tcp_rbtree_insert(&sk->tcp_rtx_queue, nskb); in tcp_send_synack()
3524 sk_wmem_queued_add(sk, nskb->truesize); in tcp_send_synack()
3525 sk_mem_charge(sk, nskb->truesize); in tcp_send_synack()
3529 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK; in tcp_send_synack()
3536 * tcp_make_synack - Allocate one skb and build a SYNACK packet.
3552 const struct tcp_sock *tp = tcp_sk(sk); in tcp_make_synack() local
3581 * sk->sk_wmem_alloc in an atomic, we can promote to rw. in tcp_make_synack()
3588 mss = tcp_mss_clamp(tp, dst_metric_advmss(dst)); in tcp_make_synack()
3593 if (unlikely(synack_type == TCP_SYNACK_COOKIE && ireq->tstamp_ok)) in tcp_make_synack()
3594 skb->skb_mstamp_ns = cookie_init_timestamp(req, now); in tcp_make_synack()
3598 skb->skb_mstamp_ns = now; in tcp_make_synack()
3599 if (!tcp_rsk(req)->snt_synack) /* Timestamp first SYNACK */ in tcp_make_synack()
3600 tcp_rsk(req)->snt_synack = tcp_skb_timestamp_us(skb); in tcp_make_synack()
3605 md5 = tcp_rsk(req)->af_specific->req_md5_lookup(sk, req_to_sk(req)); in tcp_make_synack()
3607 skb_set_hash(skb, tcp_rsk(req)->txhash, PKT_HASH_TYPE_L4); in tcp_make_synack()
3609 TCP_SKB_CB(skb)->tcp_flags = TCPHDR_SYN | TCPHDR_ACK; in tcp_make_synack()
3617 th = (struct tcphdr *)skb->data; in tcp_make_synack()
3619 th->syn = 1; in tcp_make_synack()
3620 th->ack = 1; in tcp_make_synack()
3622 th->source = htons(ireq->ir_num); in tcp_make_synack()
3623 th->dest = ireq->ir_rmt_port; in tcp_make_synack()
3624 skb->mark = ireq->ir_mark; in tcp_make_synack()
3625 skb->ip_summed = CHECKSUM_PARTIAL; in tcp_make_synack()
3626 th->seq = htonl(tcp_rsk(req)->snt_isn); in tcp_make_synack()
3628 th->ack_seq = htonl(tcp_rsk(req)->rcv_nxt); in tcp_make_synack()
3631 th->window = htons(min(req->rsk_rcv_wnd, 65535U)); in tcp_make_synack()
3633 th->doff = (tcp_header_size >> 2); in tcp_make_synack()
3637 /* Okay, we have all we need - do the md5 hash if needed */ in tcp_make_synack()
3639 tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location, in tcp_make_synack()
3647 skb->skb_mstamp_ns = now; in tcp_make_synack()
3648 tcp_add_tx_delay(skb, tp); in tcp_make_synack()
3665 if (likely(ca && bpf_try_module_get(ca, ca->owner))) { in tcp_ca_dst_init()
3666 bpf_module_put(icsk->icsk_ca_ops, icsk->icsk_ca_ops->owner); in tcp_ca_dst_init()
3667 icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst); in tcp_ca_dst_init()
3668 icsk->icsk_ca_ops = ca; in tcp_ca_dst_init()
3677 struct tcp_sock *tp = tcp_sk(sk); in tcp_connect_init() local
3684 tp->tcp_header_len = sizeof(struct tcphdr); in tcp_connect_init()
3685 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_timestamps)) in tcp_connect_init()
3686 tp->tcp_header_len += TCPOLEN_TSTAMP_ALIGNED; in tcp_connect_init()
3689 if (tp->af_specific->md5_lookup(sk, sk)) in tcp_connect_init()
3690 tp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED; in tcp_connect_init()
3694 if (tp->rx_opt.user_mss) in tcp_connect_init()
3695 tp->rx_opt.mss_clamp = tp->rx_opt.user_mss; in tcp_connect_init()
3696 tp->max_window = 0; in tcp_connect_init()
3702 if (!tp->window_clamp) in tcp_connect_init()
3703 tp->window_clamp = dst_metric(dst, RTAX_WINDOW); in tcp_connect_init()
3704 tp->advmss = tcp_mss_clamp(tp, dst_metric_advmss(dst)); in tcp_connect_init()
3709 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK && in tcp_connect_init()
3710 (tp->window_clamp > tcp_full_space(sk) || tp->window_clamp == 0)) in tcp_connect_init()
3711 tp->window_clamp = tcp_full_space(sk); in tcp_connect_init()
3718 tp->advmss - (tp->rx_opt.ts_recent_stamp ? tp->tcp_header_len - sizeof(struct tcphdr) : 0), in tcp_connect_init()
3719 &tp->rcv_wnd, in tcp_connect_init()
3720 &tp->window_clamp, in tcp_connect_init()
3721 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_window_scaling), in tcp_connect_init()
3726 tp->rcv_wnd = TCP_RCV_WND_INIT; in tcp_connect_init()
3728 tp->rx_opt.rcv_wscale = rcv_wscale; in tcp_connect_init()
3729 tp->rcv_ssthresh = tp->rcv_wnd; in tcp_connect_init()
3731 sk->sk_err = 0; in tcp_connect_init()
3733 tp->snd_wnd = 0; in tcp_connect_init()
3734 tcp_init_wl(tp, 0); in tcp_connect_init()
3736 tp->snd_una = tp->write_seq; in tcp_connect_init()
3737 tp->snd_sml = tp->write_seq; in tcp_connect_init()
3738 tp->snd_up = tp->write_seq; in tcp_connect_init()
3739 WRITE_ONCE(tp->snd_nxt, tp->write_seq); in tcp_connect_init()
3741 if (likely(!tp->repair)) in tcp_connect_init()
3742 tp->rcv_nxt = 0; in tcp_connect_init()
3744 tp->rcv_tstamp = tcp_jiffies32; in tcp_connect_init()
3745 tp->rcv_wup = tp->rcv_nxt; in tcp_connect_init()
3746 WRITE_ONCE(tp->copied_seq, tp->rcv_nxt); in tcp_connect_init()
3748 inet_csk(sk)->icsk_rto = tcp_timeout_init(sk); in tcp_connect_init()
3749 inet_csk(sk)->icsk_retransmits = 0; in tcp_connect_init()
3750 tcp_clear_retrans(tp); in tcp_connect_init()
3755 struct tcp_sock *tp = tcp_sk(sk); in tcp_connect_queue_skb() local
3758 tcb->end_seq += skb->len; in tcp_connect_queue_skb()
3760 sk_wmem_queued_add(sk, skb->truesize); in tcp_connect_queue_skb()
3761 sk_mem_charge(sk, skb->truesize); in tcp_connect_queue_skb()
3762 WRITE_ONCE(tp->write_seq, tcb->end_seq); in tcp_connect_queue_skb()
3763 tp->packets_out += tcp_skb_pcount(skb); in tcp_connect_queue_skb()
3767 * queue a data-only packet after the regular SYN, such that regular SYNs
3768 * are retransmitted on timeouts. Also if the remote SYN-ACK acknowledges
3776 struct tcp_sock *tp = tcp_sk(sk); in tcp_send_syn_data() local
3777 struct tcp_fastopen_request *fo = tp->fastopen_req; in tcp_send_syn_data()
3781 tp->rx_opt.mss_clamp = tp->advmss; /* If MSS is not cached */ in tcp_send_syn_data()
3782 if (!tcp_fastopen_cookie_check(sk, &tp->rx_opt.mss_clamp, &fo->cookie)) in tcp_send_syn_data()
3785 /* MSS for SYN-data is based on cached MSS and bounded by PMTU and in tcp_send_syn_data()
3786 * user-MSS. Reserve maximum option space for middleboxes that add in tcp_send_syn_data()
3789 tp->rx_opt.mss_clamp = tcp_mss_clamp(tp, tp->rx_opt.mss_clamp); in tcp_send_syn_data()
3791 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); in tcp_send_syn_data()
3793 space = __tcp_mtu_to_mss(sk, icsk->icsk_pmtu_cookie) - in tcp_send_syn_data()
3796 space = min_t(size_t, space, fo->size); in tcp_send_syn_data()
3798 /* limit to order-0 allocations */ in tcp_send_syn_data()
3801 syn_data = sk_stream_alloc_skb(sk, space, sk->sk_allocation, false); in tcp_send_syn_data()
3804 syn_data->ip_summed = CHECKSUM_PARTIAL; in tcp_send_syn_data()
3805 memcpy(syn_data->cb, syn->cb, sizeof(syn->cb)); in tcp_send_syn_data()
3808 &fo->data->msg_iter); in tcp_send_syn_data()
3818 skb_zcopy_set(syn_data, fo->uarg, NULL); in tcp_send_syn_data()
3821 if (space == fo->size) in tcp_send_syn_data()
3822 fo->data = NULL; in tcp_send_syn_data()
3823 fo->copied = space; in tcp_send_syn_data()
3826 if (syn_data->len) in tcp_send_syn_data()
3829 err = tcp_transmit_skb(sk, syn_data, 1, sk->sk_allocation); in tcp_send_syn_data()
3831 syn->skb_mstamp_ns = syn_data->skb_mstamp_ns; in tcp_send_syn_data()
3838 TCP_SKB_CB(syn_data)->seq++; in tcp_send_syn_data()
3839 TCP_SKB_CB(syn_data)->tcp_flags = TCPHDR_ACK | TCPHDR_PSH; in tcp_send_syn_data()
3841 tp->syn_data = (fo->copied > 0); in tcp_send_syn_data()
3842 tcp_rbtree_insert(&sk->tcp_rtx_queue, syn_data); in tcp_send_syn_data()
3848 __skb_queue_tail(&sk->sk_write_queue, syn_data); in tcp_send_syn_data()
3849 tp->packets_out -= tcp_skb_pcount(syn_data); in tcp_send_syn_data()
3853 if (fo->cookie.len > 0) in tcp_send_syn_data()
3854 fo->cookie.len = 0; in tcp_send_syn_data()
3855 err = tcp_transmit_skb(sk, syn, 1, sk->sk_allocation); in tcp_send_syn_data()
3857 tp->syn_fastopen = 0; in tcp_send_syn_data()
3859 fo->cookie.len = -1; /* Exclude Fast Open option for SYN retries */ in tcp_send_syn_data()
3866 struct tcp_sock *tp = tcp_sk(sk); in tcp_connect() local
3872 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk)) in tcp_connect()
3873 return -EHOSTUNREACH; /* Routing failure or similar. */ in tcp_connect()
3877 if (unlikely(tp->repair)) { in tcp_connect()
3882 buff = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, true); in tcp_connect()
3884 return -ENOBUFS; in tcp_connect()
3886 tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN); in tcp_connect()
3887 tcp_mstamp_refresh(tp); in tcp_connect()
3888 tp->retrans_stamp = tcp_time_stamp(tp); in tcp_connect()
3891 tcp_rbtree_insert(&sk->tcp_rtx_queue, buff); in tcp_connect()
3894 err = tp->fastopen_req ? tcp_send_syn_data(sk, buff) : in tcp_connect()
3895 tcp_transmit_skb(sk, buff, 1, sk->sk_allocation); in tcp_connect()
3896 if (err == -ECONNREFUSED) in tcp_connect()
3899 /* We change tp->snd_nxt after the tcp_transmit_skb() call in tcp_connect()
3902 WRITE_ONCE(tp->snd_nxt, tp->write_seq); in tcp_connect()
3903 tp->pushed_seq = tp->write_seq; in tcp_connect()
3906 WRITE_ONCE(tp->snd_nxt, TCP_SKB_CB(buff)->seq); in tcp_connect()
3907 tp->pushed_seq = TCP_SKB_CB(buff)->seq; in tcp_connect()
3913 inet_csk(sk)->icsk_rto, TCP_RTO_MAX); in tcp_connect()
3925 int ato = icsk->icsk_ack.ato; in tcp_send_delayed_ack()
3929 const struct tcp_sock *tp = tcp_sk(sk); in tcp_send_delayed_ack() local
3933 (icsk->icsk_ack.pending & ICSK_ACK_PUSHED)) in tcp_send_delayed_ack()
3939 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements in tcp_send_delayed_ack()
3942 if (tp->srtt_us) { in tcp_send_delayed_ack()
3943 int rtt = max_t(int, usecs_to_jiffies(tp->srtt_us >> 3), in tcp_send_delayed_ack()
3953 ato = min_t(u32, ato, inet_csk(sk)->icsk_delack_max); in tcp_send_delayed_ack()
3959 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) { in tcp_send_delayed_ack()
3961 if (time_before_eq(icsk->icsk_ack.timeout, jiffies + (ato >> 2))) { in tcp_send_delayed_ack()
3966 if (!time_before(timeout, icsk->icsk_ack.timeout)) in tcp_send_delayed_ack()
3967 timeout = icsk->icsk_ack.timeout; in tcp_send_delayed_ack()
3969 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER; in tcp_send_delayed_ack()
3970 icsk->icsk_ack.timeout = timeout; in tcp_send_delayed_ack()
3971 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout); in tcp_send_delayed_ack()
3980 if (sk->sk_state == TCP_CLOSE) in __tcp_send_ack()
3993 delay = TCP_DELACK_MAX << icsk->icsk_ack.retry; in __tcp_send_ack()
3995 icsk->icsk_ack.retry++; in __tcp_send_ack()
3997 icsk->icsk_ack.ato = TCP_ATO_MIN; in __tcp_send_ack()
4019 __tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt); in tcp_send_ack()
4029 * Current solution: to send TWO zero-length segments in urgent mode:
4031 * out-of-date with SND.UNA-1 to probe window.
4035 struct tcp_sock *tp = tcp_sk(sk); in tcp_xmit_probe_skb() local
4042 return -1; in tcp_xmit_probe_skb()
4050 tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK); in tcp_xmit_probe_skb()
4058 if (sk->sk_state == TCP_ESTABLISHED) { in tcp_send_window_probe()
4059 tcp_sk(sk)->snd_wl1 = tcp_sk(sk)->rcv_nxt - 1; in tcp_send_window_probe()
4068 struct tcp_sock *tp = tcp_sk(sk); in tcp_write_wakeup() local
4071 if (sk->sk_state == TCP_CLOSE) in tcp_write_wakeup()
4072 return -1; in tcp_write_wakeup()
4075 if (skb && before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp))) { in tcp_write_wakeup()
4078 unsigned int seg_size = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in tcp_write_wakeup()
4080 if (before(tp->pushed_seq, TCP_SKB_CB(skb)->end_seq)) in tcp_write_wakeup()
4081 tp->pushed_seq = TCP_SKB_CB(skb)->end_seq; in tcp_write_wakeup()
4087 if (seg_size < TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq || in tcp_write_wakeup()
4088 skb->len > mss) { in tcp_write_wakeup()
4090 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; in tcp_write_wakeup()
4093 return -1; in tcp_write_wakeup()
4097 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; in tcp_write_wakeup()
4103 if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF)) in tcp_write_wakeup()
4115 struct tcp_sock *tp = tcp_sk(sk); in tcp_send_probe0() local
4121 if (tp->packets_out || tcp_write_queue_empty(sk)) { in tcp_send_probe0()
4123 icsk->icsk_probes_out = 0; in tcp_send_probe0()
4124 icsk->icsk_backoff = 0; in tcp_send_probe0()
4125 icsk->icsk_probes_tstamp = 0; in tcp_send_probe0()
4129 icsk->icsk_probes_out++; in tcp_send_probe0()
4131 if (icsk->icsk_backoff < READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_retries2)) in tcp_send_probe0()
4132 icsk->icsk_backoff++; in tcp_send_probe0()
4147 const struct tcp_request_sock_ops *af_ops = tcp_rsk(req)->af_specific; in tcp_rtx_synack()
4151 tcp_rsk(req)->txhash = net_tx_rndhash(); in tcp_rtx_synack()
4152 res = af_ops->send_synack(sk, NULL, &fl, req, NULL, TCP_SYNACK_NORMAL, in tcp_rtx_synack()
4158 tcp_sk(sk)->total_retrans++; in tcp_rtx_synack()