Lines Matching +full:wire +full:- +full:sized
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
59 tp->tcp_clock_cache = val; in tcp_mstamp_refresh()
60 tp->tcp_mstamp = div_u64(val, NSEC_PER_USEC); in tcp_mstamp_refresh()
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. :-(
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()
110 * RFC1122, RFC1063, draft-ietf-tcpimpl-pmtud-01 state that:
113 * 2. Ideally, it is maximal possible segment size i.e. 65535-40.
127 int mss = tp->advmss; in tcp_advertise_mss()
134 tp->advmss = mss; in tcp_advertise_mss()
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()
158 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_cwnd_restart()
159 tp->snd_cwnd_used = 0; in tcp_cwnd_restart()
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()
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()
227 * we will truncate our initial window offering to 32K-1 in tcp_select_initial_window()
232 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_workaround_signed_windows)) in tcp_select_initial_window()
243 space = max_t(u32, space, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2])); in tcp_select_initial_window()
246 *rcv_wscale = clamp_t(int, ilog2(space) - 15, in tcp_select_initial_window()
257 * value can be stuffed directly into th->window for an outgoing
264 u32 old_win = tp->rcv_wnd; in tcp_select_window()
270 if (unlikely(inet_csk(sk)->icsk_ack.pending & ICSK_ACK_NOMEM)) { in tcp_select_window()
271 tp->pred_flags = 0; in tcp_select_window()
272 tp->rcv_wnd = 0; in tcp_select_window()
273 tp->rcv_wup = tp->rcv_nxt; in tcp_select_window()
283 * window in time. --DaveM in tcp_select_window()
287 if (!READ_ONCE(net->ipv4.sysctl_tcp_shrink_window) || !tp->rx_opt.rcv_wscale) { in tcp_select_window()
291 new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale); in tcp_select_window()
295 tp->rcv_wnd = new_win; in tcp_select_window()
296 tp->rcv_wup = tp->rcv_nxt; in tcp_select_window()
301 if (!tp->rx_opt.rcv_wscale && in tcp_select_window()
302 READ_ONCE(net->ipv4.sysctl_tcp_workaround_signed_windows)) in tcp_select_window()
305 new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale)); in tcp_select_window()
308 new_win >>= tp->rx_opt.rcv_wscale; in tcp_select_window()
312 tp->pred_flags = 0; in tcp_select_window()
322 /* Packet ECN state for a SYN-ACK */
327 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR; in tcp_ecn_send_synack()
329 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE; in tcp_ecn_send_synack()
340 bool use_ecn = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_ecn) == 1 || in tcp_ecn_send_syn()
350 tp->ecn_flags = 0; in tcp_ecn_send_syn()
353 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR; in tcp_ecn_send_syn()
362 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_ecn_fallback)) in tcp_ecn_clear_syn()
363 /* tp->ecn_flags are cleared at a later point in time when in tcp_ecn_clear_syn()
366 TCP_SKB_CB(skb)->tcp_flags &= ~(TCPHDR_ECE | TCPHDR_CWR); in tcp_ecn_clear_syn()
372 if (inet_rsk(req)->ecn_ok) in tcp_ecn_make_synack()
373 th->ece = 1; in tcp_ecn_make_synack()
385 /* Not-retransmitted data segment: set ECT and inject CWR. */ in tcp_ecn_send()
386 if (skb->len != tcp_header_len && in tcp_ecn_send()
387 !before(TCP_SKB_CB(skb)->seq, tp->snd_nxt)) { in tcp_ecn_send()
389 if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) { in tcp_ecn_send()
390 tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR; in tcp_ecn_send()
391 th->cwr = 1; in tcp_ecn_send()
392 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; in tcp_ecn_send()
398 if (tp->ecn_flags & TCP_ECN_DEMAND_CWR) in tcp_ecn_send()
399 th->ece = 1; in tcp_ecn_send()
403 /* Constructs common control bits of non-data skb. If SYN/FIN is present,
408 skb->ip_summed = CHECKSUM_PARTIAL; in tcp_init_nondata_skb()
410 TCP_SKB_CB(skb)->tcp_flags = flags; in tcp_init_nondata_skb()
414 TCP_SKB_CB(skb)->seq = seq; in tcp_init_nondata_skb()
417 TCP_SKB_CB(skb)->end_seq = seq; in tcp_init_nondata_skb()
422 return tp->snd_una != tp->snd_up; in tcp_urg_mode()
467 if (unlikely(OPTION_MPTCP & opts->options)) in mptcp_options_write()
468 mptcp_write_options(th, ptr, tp, &opts->mptcp); in mptcp_options_write()
514 * Thus, "req" is passed here and the cgroup-bpf-progs in bpf_skops_hdr_opt_len()
519 * consistent between fastopen and non-fastopen on in bpf_skops_hdr_opt_len()
543 opts->bpf_opt_len = *remaining - sock_ops.remaining_opt_len; in bpf_skops_hdr_opt_len()
545 opts->bpf_opt_len = (opts->bpf_opt_len + 3) & ~3; in bpf_skops_hdr_opt_len()
547 *remaining -= opts->bpf_opt_len; in bpf_skops_hdr_opt_len()
556 u8 first_opt_off, nr_written, max_opt_len = opts->bpf_opt_len; in bpf_skops_write_hdr_opt()
580 first_opt_off = tcp_hdrlen(skb) - max_opt_len; in bpf_skops_write_hdr_opt()
588 nr_written = max_opt_len - sock_ops.remaining_opt_len; in bpf_skops_write_hdr_opt()
591 memset(skb->data + first_opt_off + nr_written, TCPOPT_NOP, in bpf_skops_write_hdr_opt()
592 max_opt_len - nr_written); in bpf_skops_write_hdr_opt()
619 u8 maclen = tcp_ao_maclen(key->ao_key); in process_tcp_ao_options()
625 (tcprsk->ao_keyid << 8) | in process_tcp_ao_options()
626 (tcprsk->ao_rcv_next)); in process_tcp_ao_options()
631 ao_info = rcu_dereference_check(tp->ao_info, in process_tcp_ao_options()
632 lockdep_sock_is_held(&tp->inet_conn.icsk_inet.sk)); in process_tcp_ao_options()
633 rnext_key = READ_ONCE(ao_info->rnext_key); in process_tcp_ao_options()
637 (tcp_ao_len(key->ao_key) << 16) | in process_tcp_ao_options()
638 (key->ao_key->sndid << 8) | in process_tcp_ao_options()
639 (rnext_key->rcvid)); in process_tcp_ao_options()
641 opts->hash_location = (__u8 *)ptr; in process_tcp_ao_options()
655 * Luckily we can at least blame others for their non-compliance but from
656 * inter-operability perspective it seems that we're somewhat stuck with
670 u16 options = opts->options; /* mungable copy */ in tcp_options_write()
676 opts->hash_location = (__u8 *)ptr; in tcp_options_write()
681 if (unlikely(opts->mss)) { in tcp_options_write()
684 opts->mss); in tcp_options_write()
700 *ptr++ = htonl(opts->tsval); in tcp_options_write()
701 *ptr++ = htonl(opts->tsecr); in tcp_options_write()
715 opts->ws); in tcp_options_write()
718 if (unlikely(opts->num_sack_blocks)) { in tcp_options_write()
719 struct tcp_sack_block *sp = tp->rx_opt.dsack ? in tcp_options_write()
720 tp->duplicate_sack : tp->selective_acks; in tcp_options_write()
726 (TCPOLEN_SACK_BASE + (opts->num_sack_blocks * in tcp_options_write()
729 for (this_sack = 0; this_sack < opts->num_sack_blocks; in tcp_options_write()
735 tp->rx_opt.dsack = 0; in tcp_options_write()
739 struct tcp_fastopen_cookie *foc = opts->fastopen_cookie; in tcp_options_write()
743 if (foc->exp) { in tcp_options_write()
744 len = TCPOLEN_EXP_FASTOPEN_BASE + foc->len; in tcp_options_write()
749 len = TCPOLEN_FASTOPEN_BASE + foc->len; in tcp_options_write()
754 memcpy(p, foc->val, foc->len); in tcp_options_write()
756 p[foc->len] = TCPOPT_NOP; in tcp_options_write()
757 p[foc->len + 1] = TCPOPT_NOP; in tcp_options_write()
773 if (tp->syn_smc) { in smc_set_option()
775 opts->options |= OPTION_SMC; in smc_set_option()
776 *remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED; in smc_set_option()
790 if (tp->syn_smc && ireq->smc_ok) { in smc_set_option_cond()
792 opts->options |= OPTION_SMC; in smc_set_option_cond()
793 *remaining -= TCPOLEN_EXP_SMC_BASE_ALIGNED; in smc_set_option_cond()
807 if (mptcp_synack_options(req, &size, &opts->mptcp)) { in mptcp_set_option_cond()
809 opts->options |= OPTION_MPTCP; in mptcp_set_option_cond()
810 *remaining -= size; in mptcp_set_option_cond()
817 * network wire format yet.
825 struct tcp_fastopen_request *fastopen = tp->fastopen_req; in tcp_syn_options()
831 opts->options |= OPTION_MD5; in tcp_syn_options()
832 remaining -= TCPOLEN_MD5SIG_ALIGNED; in tcp_syn_options()
834 timestamps = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_timestamps); in tcp_syn_options()
836 opts->options |= OPTION_AO; in tcp_syn_options()
837 remaining -= tcp_ao_len_aligned(key->ao_key); in tcp_syn_options()
843 * advertised. But we subtract them from tp->mss_cache so that in tcp_syn_options()
846 * receiver we won't recognize data packets as being full sized when we in tcp_syn_options()
850 opts->mss = tcp_advertise_mss(sk); in tcp_syn_options()
851 remaining -= TCPOLEN_MSS_ALIGNED; in tcp_syn_options()
854 opts->options |= OPTION_TS; in tcp_syn_options()
855 opts->tsval = tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb) + tp->tsoffset; in tcp_syn_options()
856 opts->tsecr = tp->rx_opt.ts_recent; in tcp_syn_options()
857 remaining -= TCPOLEN_TSTAMP_ALIGNED; in tcp_syn_options()
859 if (likely(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_window_scaling))) { in tcp_syn_options()
860 opts->ws = tp->rx_opt.rcv_wscale; in tcp_syn_options()
861 opts->options |= OPTION_WSCALE; in tcp_syn_options()
862 remaining -= TCPOLEN_WSCALE_ALIGNED; in tcp_syn_options()
864 if (likely(READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_sack))) { in tcp_syn_options()
865 opts->options |= OPTION_SACK_ADVERTISE; in tcp_syn_options()
866 if (unlikely(!(OPTION_TS & opts->options))) in tcp_syn_options()
867 remaining -= TCPOLEN_SACKPERM_ALIGNED; in tcp_syn_options()
870 if (fastopen && fastopen->cookie.len >= 0) { in tcp_syn_options()
871 u32 need = fastopen->cookie.len; in tcp_syn_options()
873 need += fastopen->cookie.exp ? TCPOLEN_EXP_FASTOPEN_BASE : in tcp_syn_options()
877 opts->options |= OPTION_FAST_OPEN_COOKIE; in tcp_syn_options()
878 opts->fastopen_cookie = &fastopen->cookie; in tcp_syn_options()
879 remaining -= need; in tcp_syn_options()
880 tp->syn_fastopen = 1; in tcp_syn_options()
881 tp->syn_fastopen_exp = fastopen->cookie.exp ? 1 : 0; in tcp_syn_options()
890 if (mptcp_syn_options(sk, skb, &size, &opts->mptcp)) { in tcp_syn_options()
892 opts->options |= OPTION_MPTCP; in tcp_syn_options()
893 remaining -= size; in tcp_syn_options()
900 return MAX_TCP_OPTION_SPACE - remaining; in tcp_syn_options()
903 /* Set up TCP options for SYN-ACKs. */
917 opts->options |= OPTION_MD5; in tcp_synack_options()
918 remaining -= TCPOLEN_MD5SIG_ALIGNED; in tcp_synack_options()
926 ireq->tstamp_ok &= !ireq->sack_ok; in tcp_synack_options()
928 opts->options |= OPTION_AO; in tcp_synack_options()
929 remaining -= tcp_ao_len_aligned(key->ao_key); in tcp_synack_options()
930 ireq->tstamp_ok &= !ireq->sack_ok; in tcp_synack_options()
934 opts->mss = mss; in tcp_synack_options()
935 remaining -= TCPOLEN_MSS_ALIGNED; in tcp_synack_options()
937 if (likely(ireq->wscale_ok)) { in tcp_synack_options()
938 opts->ws = ireq->rcv_wscale; in tcp_synack_options()
939 opts->options |= OPTION_WSCALE; in tcp_synack_options()
940 remaining -= TCPOLEN_WSCALE_ALIGNED; in tcp_synack_options()
942 if (likely(ireq->tstamp_ok)) { in tcp_synack_options()
943 opts->options |= OPTION_TS; in tcp_synack_options()
944 opts->tsval = tcp_skb_timestamp_ts(tcp_rsk(req)->req_usec_ts, skb) + in tcp_synack_options()
945 tcp_rsk(req)->ts_off; in tcp_synack_options()
946 if (!tcp_rsk(req)->snt_tsval_first) { in tcp_synack_options()
947 if (!opts->tsval) in tcp_synack_options()
948 opts->tsval = ~0U; in tcp_synack_options()
949 tcp_rsk(req)->snt_tsval_first = opts->tsval; in tcp_synack_options()
951 WRITE_ONCE(tcp_rsk(req)->snt_tsval_last, opts->tsval); in tcp_synack_options()
952 opts->tsecr = req->ts_recent; in tcp_synack_options()
953 remaining -= TCPOLEN_TSTAMP_ALIGNED; in tcp_synack_options()
955 if (likely(ireq->sack_ok)) { in tcp_synack_options()
956 opts->options |= OPTION_SACK_ADVERTISE; in tcp_synack_options()
957 if (unlikely(!ireq->tstamp_ok)) in tcp_synack_options()
958 remaining -= TCPOLEN_SACKPERM_ALIGNED; in tcp_synack_options()
960 if (foc != NULL && foc->len >= 0) { in tcp_synack_options()
961 u32 need = foc->len; in tcp_synack_options()
963 need += foc->exp ? TCPOLEN_EXP_FASTOPEN_BASE : in tcp_synack_options()
967 opts->options |= OPTION_FAST_OPEN_COOKIE; in tcp_synack_options()
968 opts->fastopen_cookie = foc; in tcp_synack_options()
969 remaining -= need; in tcp_synack_options()
980 return MAX_TCP_OPTION_SPACE - remaining; in tcp_synack_options()
984 * final wire format yet.
994 opts->options = 0; in tcp_established_options()
998 opts->options |= OPTION_MD5; in tcp_established_options()
1001 opts->options |= OPTION_AO; in tcp_established_options()
1002 size += tcp_ao_len_aligned(key->ao_key); in tcp_established_options()
1005 if (likely(tp->rx_opt.tstamp_ok)) { in tcp_established_options()
1006 opts->options |= OPTION_TS; in tcp_established_options()
1007 opts->tsval = skb ? tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb) + in tcp_established_options()
1008 tp->tsoffset : 0; in tcp_established_options()
1009 opts->tsecr = tp->rx_opt.ts_recent; in tcp_established_options()
1020 unsigned int remaining = MAX_TCP_OPTION_SPACE - size; in tcp_established_options()
1024 &opts->mptcp)) { in tcp_established_options()
1025 opts->options |= OPTION_MPTCP; in tcp_established_options()
1030 eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack; in tcp_established_options()
1032 const unsigned int remaining = MAX_TCP_OPTION_SPACE - size; in tcp_established_options()
1037 opts->num_sack_blocks = in tcp_established_options()
1039 (remaining - TCPOLEN_SACK_BASE_ALIGNED) / in tcp_established_options()
1043 opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK; in tcp_established_options()
1048 unsigned int remaining = MAX_TCP_OPTION_SPACE - size; in tcp_established_options()
1052 size = MAX_TCP_OPTION_SPACE - remaining; in tcp_established_options()
1067 * The invariant being skb->truesize subtracted from sk->sk_wmem_alloc
1081 if ((1 << sk->sk_state) & in tcp_tsq_write()
1086 if (tp->lost_out > tp->retrans_out && in tcp_tsq_write()
1092 tcp_write_xmit(sk, tcp_current_mss(sk), tp->nonagle, in tcp_tsq_write()
1102 else if (!test_and_set_bit(TCP_TSQ_DEFERRED, &sk->sk_tsq_flags)) in tcp_tsq_handler()
1109 * transferring tsq->head because tcp_wfree() might
1122 list_splice_init(&tsq->head, &list); in tcp_tasklet_func()
1127 list_del(&tp->tsq_node); in tcp_tasklet_func()
1131 clear_bit(TSQ_QUEUED, &sk->sk_tsq_flags); in tcp_tasklet_func()
1144 * tcp_release_cb - tcp release_sock() callback
1152 unsigned long flags = smp_load_acquire(&sk->sk_tsq_flags); in tcp_release_cb()
1160 } while (!try_cmpxchg(&sk->sk_tsq_flags, &flags, nflags)); in tcp_release_cb()
1176 inet_csk(sk)->icsk_af_ops->mtu_reduced(sk); in tcp_release_cb()
1191 INIT_LIST_HEAD(&tsq->head); in tcp_tasklet_init()
1192 tasklet_setup(&tsq->tasklet, tcp_tasklet_func); in tcp_tasklet_init()
1203 struct sock *sk = skb->sk; in tcp_wfree()
1212 WARN_ON(refcount_sub_and_test(skb->truesize - 1, &sk->sk_wmem_alloc)); in tcp_wfree()
1217 * - less callbacks to tcp_write_xmit(), reducing stress (batches) in tcp_wfree()
1218 * - chance for incoming ACK (processed by another cpu maybe) in tcp_wfree()
1219 * to migrate this flow (skb->ooo_okay will be eventually set) in tcp_wfree()
1221 if (refcount_read(&sk->sk_wmem_alloc) >= SKB_TRUESIZE(1) && this_cpu_ksoftirqd() == current) in tcp_wfree()
1224 oval = smp_load_acquire(&sk->sk_tsq_flags); in tcp_wfree()
1230 } while (!try_cmpxchg(&sk->sk_tsq_flags, &oval, nval)); in tcp_wfree()
1235 empty = list_empty(&tsq->head); in tcp_wfree()
1236 list_add(&tp->tsq_node, &tsq->head); in tcp_wfree()
1238 tasklet_schedule(&tsq->tasklet); in tcp_wfree()
1264 if (sk->sk_pacing_status != SK_PACING_NONE) { in tcp_update_skb_after_send()
1265 unsigned long rate = READ_ONCE(sk->sk_pacing_rate); in tcp_update_skb_after_send()
1268 * Note that tp->data_segs_out overflows after 2^32 packets, in tcp_update_skb_after_send()
1271 if (rate != ~0UL && rate && tp->data_segs_out >= 10) { in tcp_update_skb_after_send()
1272 u64 len_ns = div64_ul((u64)skb->len * NSEC_PER_SEC, rate); in tcp_update_skb_after_send()
1273 u64 credit = tp->tcp_wstamp_ns - prior_wstamp; in tcp_update_skb_after_send()
1276 len_ns -= min_t(u64, len_ns / 2, credit); in tcp_update_skb_after_send()
1277 tp->tcp_wstamp_ns += len_ns; in tcp_update_skb_after_send()
1280 list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue); in tcp_update_skb_after_send()
1315 prior_wstamp = tp->tcp_wstamp_ns; in __tcp_transmit_skb()
1316 tp->tcp_wstamp_ns = max(tp->tcp_wstamp_ns, tp->tcp_clock_cache); in __tcp_transmit_skb()
1317 skb_set_delivery_time(skb, tp->tcp_wstamp_ns, SKB_CLOCK_MONOTONIC); in __tcp_transmit_skb()
1329 return -ENOBUFS; in __tcp_transmit_skb()
1330 /* retransmit skbs might have a non zero value in skb->dev in __tcp_transmit_skb()
1331 * because skb->dev is aliased with skb->rbnode.rb_left in __tcp_transmit_skb()
1333 skb->dev = NULL; in __tcp_transmit_skb()
1341 if (unlikely(tcb->tcp_flags & TCPHDR_SYN)) { in __tcp_transmit_skb()
1349 * and in this case it is better to delay the delivery of 1-MSS in __tcp_transmit_skb()
1354 tcb->tcp_flags |= TCPHDR_PSH; in __tcp_transmit_skb()
1358 /* We set skb->ooo_okay to one if this packet can select in __tcp_transmit_skb()
1362 * if XPS is enabled, or sk->sk_txhash otherwise. in __tcp_transmit_skb()
1371 skb->ooo_okay = sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1) || in __tcp_transmit_skb()
1379 skb->pfmemalloc = 0; in __tcp_transmit_skb()
1385 skb->sk = sk; in __tcp_transmit_skb()
1386 skb->destructor = skb_is_tcp_pure_ack(skb) ? __sock_wfree : tcp_wfree; in __tcp_transmit_skb()
1387 refcount_add(skb->truesize, &sk->sk_wmem_alloc); in __tcp_transmit_skb()
1389 skb_set_dst_pending_confirm(skb, READ_ONCE(sk->sk_dst_pending_confirm)); in __tcp_transmit_skb()
1392 th = (struct tcphdr *)skb->data; in __tcp_transmit_skb()
1393 th->source = inet->inet_sport; in __tcp_transmit_skb()
1394 th->dest = inet->inet_dport; in __tcp_transmit_skb()
1395 th->seq = htonl(tcb->seq); in __tcp_transmit_skb()
1396 th->ack_seq = htonl(rcv_nxt); in __tcp_transmit_skb()
1398 (tcb->tcp_flags & TCPHDR_FLAGS_MASK)); in __tcp_transmit_skb()
1400 th->check = 0; in __tcp_transmit_skb()
1401 th->urg_ptr = 0; in __tcp_transmit_skb()
1404 if (unlikely(tcp_urg_mode(tp) && before(tcb->seq, tp->snd_up))) { in __tcp_transmit_skb()
1405 if (before(tp->snd_up, tcb->seq + 0x10000)) { in __tcp_transmit_skb()
1406 th->urg_ptr = htons(tp->snd_up - tcb->seq); in __tcp_transmit_skb()
1407 th->urg = 1; in __tcp_transmit_skb()
1408 } else if (after(tcb->seq + 0xFFFF, tp->snd_nxt)) { in __tcp_transmit_skb()
1409 th->urg_ptr = htons(0xFFFF); in __tcp_transmit_skb()
1410 th->urg = 1; in __tcp_transmit_skb()
1414 skb_shinfo(skb)->gso_type = sk->sk_gso_type; in __tcp_transmit_skb()
1415 if (likely(!(tcb->tcp_flags & TCPHDR_SYN))) { in __tcp_transmit_skb()
1416 th->window = htons(tcp_select_window(sk)); in __tcp_transmit_skb()
1422 th->window = htons(min(tp->rcv_wnd, 65535U)); in __tcp_transmit_skb()
1431 tp->af_specific->calc_md5_hash(opts.hash_location, in __tcp_transmit_skb()
1441 return -ENOMEM; in __tcp_transmit_skb()
1448 INDIRECT_CALL_INET(icsk->icsk_af_ops->send_check, in __tcp_transmit_skb()
1452 if (likely(tcb->tcp_flags & TCPHDR_ACK)) in __tcp_transmit_skb()
1455 if (skb->len != tcp_header_size) { in __tcp_transmit_skb()
1457 tp->data_segs_out += tcp_skb_pcount(skb); in __tcp_transmit_skb()
1458 tp->bytes_sent += skb->len - tcp_header_size; in __tcp_transmit_skb()
1461 if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq) in __tcp_transmit_skb()
1465 tp->segs_out += tcp_skb_pcount(skb); in __tcp_transmit_skb()
1467 /* OK, its time to fill skb_shinfo(skb)->gso_{segs|size} */ in __tcp_transmit_skb()
1468 skb_shinfo(skb)->gso_segs = tcp_skb_pcount(skb); in __tcp_transmit_skb()
1469 skb_shinfo(skb)->gso_size = tcp_skb_mss(skb); in __tcp_transmit_skb()
1471 /* Leave earliest departure time in skb->tstamp (skb->skb_mstamp_ns) */ in __tcp_transmit_skb()
1474 memset(skb->cb, 0, max(sizeof(struct inet_skb_parm), in __tcp_transmit_skb()
1479 err = INDIRECT_CALL_INET(icsk->icsk_af_ops->queue_xmit, in __tcp_transmit_skb()
1481 sk, skb, &inet->cork.fl); in __tcp_transmit_skb()
1498 tcp_sk(sk)->rcv_nxt); in tcp_transmit_skb()
1511 WRITE_ONCE(tp->write_seq, TCP_SKB_CB(skb)->end_seq); in tcp_queue_skb()
1514 sk_wmem_queued_add(sk, skb->truesize); in tcp_queue_skb()
1515 sk_mem_charge(sk, skb->truesize); in tcp_queue_skb()
1523 if (skb->len <= mss_now) { in tcp_set_skb_tso_segs()
1525 * non-TSO case. in tcp_set_skb_tso_segs()
1527 TCP_SKB_CB(skb)->tcp_gso_size = 0; in tcp_set_skb_tso_segs()
1531 TCP_SKB_CB(skb)->tcp_gso_size = mss_now; in tcp_set_skb_tso_segs()
1532 tso_segs = DIV_ROUND_UP(skb->len, mss_now); in tcp_set_skb_tso_segs()
1544 tp->packets_out -= decr; in tcp_adjust_pcount()
1546 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) in tcp_adjust_pcount()
1547 tp->sacked_out -= decr; in tcp_adjust_pcount()
1548 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) in tcp_adjust_pcount()
1549 tp->retrans_out -= decr; in tcp_adjust_pcount()
1550 if (TCP_SKB_CB(skb)->sacked & TCPCB_LOST) in tcp_adjust_pcount()
1551 tp->lost_out -= decr; in tcp_adjust_pcount()
1555 tp->sacked_out -= min_t(u32, tp->sacked_out, decr); in tcp_adjust_pcount()
1557 if (tp->lost_skb_hint && in tcp_adjust_pcount()
1558 before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(tp->lost_skb_hint)->seq) && in tcp_adjust_pcount()
1559 (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) in tcp_adjust_pcount()
1560 tp->lost_cnt_hint -= decr; in tcp_adjust_pcount()
1567 return TCP_SKB_CB(skb)->txstamp_ack || in tcp_has_tx_tstamp()
1568 (skb_shinfo(skb)->tx_flags & SKBTX_ANY_TSTAMP); in tcp_has_tx_tstamp()
1576 !before(shinfo->tskey, TCP_SKB_CB(skb2)->seq)) { in tcp_fragment_tstamp()
1578 u8 tsflags = shinfo->tx_flags & SKBTX_ANY_TSTAMP; in tcp_fragment_tstamp()
1580 shinfo->tx_flags &= ~tsflags; in tcp_fragment_tstamp()
1581 shinfo2->tx_flags |= tsflags; in tcp_fragment_tstamp()
1582 swap(shinfo->tskey, shinfo2->tskey); in tcp_fragment_tstamp()
1583 TCP_SKB_CB(skb2)->txstamp_ack = TCP_SKB_CB(skb)->txstamp_ack; in tcp_fragment_tstamp()
1584 TCP_SKB_CB(skb)->txstamp_ack = 0; in tcp_fragment_tstamp()
1590 TCP_SKB_CB(skb2)->eor = TCP_SKB_CB(skb)->eor; in tcp_skb_fragment_eor()
1591 TCP_SKB_CB(skb)->eor = 0; in tcp_skb_fragment_eor()
1601 __skb_queue_after(&sk->sk_write_queue, skb, buff); in tcp_insert_write_queue_after()
1603 tcp_rbtree_insert(&sk->tcp_rtx_queue, buff); in tcp_insert_write_queue_after()
1622 if (WARN_ON(len > skb->len)) in tcp_fragment()
1623 return -EINVAL; in tcp_fragment()
1632 limit = sk->sk_sndbuf + 2 * SKB_TRUESIZE(GSO_LEGACY_MAX_SIZE); in tcp_fragment()
1633 if (unlikely((sk->sk_wmem_queued >> 1) > limit && in tcp_fragment()
1638 return -ENOMEM; in tcp_fragment()
1642 return -ENOMEM; in tcp_fragment()
1647 return -ENOMEM; /* We'll just try again later. */ in tcp_fragment()
1651 sk_wmem_queued_add(sk, buff->truesize); in tcp_fragment()
1652 sk_mem_charge(sk, buff->truesize); in tcp_fragment()
1653 nlen = skb->len - len; in tcp_fragment()
1654 buff->truesize += nlen; in tcp_fragment()
1655 skb->truesize -= nlen; in tcp_fragment()
1658 TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len; in tcp_fragment()
1659 TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq; in tcp_fragment()
1660 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq; in tcp_fragment()
1663 flags = TCP_SKB_CB(skb)->tcp_flags; in tcp_fragment()
1664 TCP_SKB_CB(skb)->tcp_flags = flags & ~(TCPHDR_FIN | TCPHDR_PSH); in tcp_fragment()
1665 TCP_SKB_CB(buff)->tcp_flags = flags; in tcp_fragment()
1666 TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked; in tcp_fragment()
1671 skb_set_delivery_time(buff, skb->tstamp, SKB_CLOCK_MONOTONIC); in tcp_fragment()
1681 TCP_SKB_CB(buff)->tx = TCP_SKB_CB(skb)->tx; in tcp_fragment()
1686 if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) { in tcp_fragment()
1687 int diff = old_factor - tcp_skb_pcount(skb) - in tcp_fragment()
1698 list_add(&buff->tcp_tsorted_anchor, &skb->tcp_tsorted_anchor); in tcp_fragment()
1715 for (i = 0; i < shinfo->nr_frags; i++) { in __pskb_trim_head()
1716 int size = skb_frag_size(&shinfo->frags[i]); in __pskb_trim_head()
1720 eat -= size; in __pskb_trim_head()
1722 shinfo->frags[k] = shinfo->frags[i]; in __pskb_trim_head()
1724 skb_frag_off_add(&shinfo->frags[k], eat); in __pskb_trim_head()
1725 skb_frag_size_sub(&shinfo->frags[k], eat); in __pskb_trim_head()
1731 shinfo->nr_frags = k; in __pskb_trim_head()
1733 skb->data_len -= len; in __pskb_trim_head()
1734 skb->len = skb->data_len; in __pskb_trim_head()
1744 return -ENOMEM; in tcp_trim_head()
1748 TCP_SKB_CB(skb)->seq += len; in tcp_trim_head()
1750 skb->truesize -= delta_truesize; in tcp_trim_head()
1751 sk_wmem_queued_add(sk, -delta_truesize); in tcp_trim_head()
1755 /* Any change of skb->len requires recalculation of tso factor. */ in tcp_trim_head()
1770 It is MMS_S - sizeof(tcphdr) of rfc1122 in __tcp_mtu_to_mss()
1772 mss_now = pmtu - icsk->icsk_af_ops->net_header_len - sizeof(struct tcphdr); in __tcp_mtu_to_mss()
1775 if (mss_now > tp->rx_opt.mss_clamp) in __tcp_mtu_to_mss()
1776 mss_now = tp->rx_opt.mss_clamp; in __tcp_mtu_to_mss()
1779 mss_now -= icsk->icsk_ext_hdr_len; in __tcp_mtu_to_mss()
1783 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_snd_mss)); in __tcp_mtu_to_mss()
1791 return __tcp_mtu_to_mss(sk, pmtu) - in tcp_mtu_to_mss()
1792 (tcp_sk(sk)->tcp_header_len - sizeof(struct tcphdr)); in tcp_mtu_to_mss()
1803 tp->tcp_header_len + in tcp_mss_to_mtu()
1804 icsk->icsk_ext_hdr_len + in tcp_mss_to_mtu()
1805 icsk->icsk_af_ops->net_header_len; in tcp_mss_to_mtu()
1816 icsk->icsk_mtup.enabled = READ_ONCE(net->ipv4.sysctl_tcp_mtu_probing) > 1; in tcp_mtup_init()
1817 icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) + in tcp_mtup_init()
1818 icsk->icsk_af_ops->net_header_len; in tcp_mtup_init()
1819 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, READ_ONCE(net->ipv4.sysctl_tcp_base_mss)); in tcp_mtup_init()
1820 icsk->icsk_mtup.probe_size = 0; in tcp_mtup_init()
1821 if (icsk->icsk_mtup.enabled) in tcp_mtup_init()
1822 icsk->icsk_mtup.probe_timestamp = tcp_jiffies32; in tcp_mtup_init()
1827 tp->rx_opt.user_mss is mss set by user by TCP_MAXSEG. It does NOT counts
1830 tp->rx_opt.mss_clamp is mss negotiated at connection setup.
1834 inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
1836 tp->mss_cache is current effective sending mss, including
1839 tp->rx_opt.mss_clamp.
1844 NOTE2. inet_csk(sk)->icsk_pmtu_cookie and tp->mss_cache
1845 are READ ONLY outside this function. --ANK (980731)
1853 if (icsk->icsk_mtup.search_high > pmtu) in tcp_sync_mss()
1854 icsk->icsk_mtup.search_high = pmtu; in tcp_sync_mss()
1860 icsk->icsk_pmtu_cookie = pmtu; in tcp_sync_mss()
1861 if (icsk->icsk_mtup.enabled) in tcp_sync_mss()
1862 mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low)); in tcp_sync_mss()
1863 tp->mss_cache = mss_now; in tcp_sync_mss()
1881 mss_now = tp->mss_cache; in tcp_current_mss()
1885 if (mtu != inet_csk(sk)->icsk_pmtu_cookie) in tcp_current_mss()
1891 /* The mss_cache is sized based on tp->tcp_header_len, which assumes in tcp_current_mss()
1895 if (header_len != tp->tcp_header_len) { in tcp_current_mss()
1896 int delta = (int) header_len - tp->tcp_header_len; in tcp_current_mss()
1897 mss_now -= delta; in tcp_current_mss()
1911 if (inet_csk(sk)->icsk_ca_state == TCP_CA_Open && in tcp_cwnd_application_limited()
1912 sk->sk_socket && !test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { in tcp_cwnd_application_limited()
1915 u32 win_used = max(tp->snd_cwnd_used, init_win); in tcp_cwnd_application_limited()
1917 tp->snd_ssthresh = tcp_current_ssthresh(sk); in tcp_cwnd_application_limited()
1920 tp->snd_cwnd_used = 0; in tcp_cwnd_application_limited()
1922 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_cwnd_application_limited()
1927 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; in tcp_cwnd_validate()
1931 * is fully utilized. If cwnd-limited then remember that fact for the in tcp_cwnd_validate()
1932 * current window. If not cwnd-limited then track the maximum number of in tcp_cwnd_validate()
1933 * outstanding packets in the current window. (If cwnd-limited then we in tcp_cwnd_validate()
1934 * chose to not update tp->max_packets_out to avoid an extra else in tcp_cwnd_validate()
1937 if (!before(tp->snd_una, tp->cwnd_usage_seq) || in tcp_cwnd_validate()
1939 (!tp->is_cwnd_limited && in tcp_cwnd_validate()
1940 tp->packets_out > tp->max_packets_out)) { in tcp_cwnd_validate()
1941 tp->is_cwnd_limited = is_cwnd_limited; in tcp_cwnd_validate()
1942 tp->max_packets_out = tp->packets_out; in tcp_cwnd_validate()
1943 tp->cwnd_usage_seq = tp->snd_nxt; in tcp_cwnd_validate()
1948 tp->snd_cwnd_used = 0; in tcp_cwnd_validate()
1949 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_cwnd_validate()
1952 if (tp->packets_out > tp->snd_cwnd_used) in tcp_cwnd_validate()
1953 tp->snd_cwnd_used = tp->packets_out; in tcp_cwnd_validate()
1955 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) && in tcp_cwnd_validate()
1956 (s32)(tcp_jiffies32 - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto && in tcp_cwnd_validate()
1957 !ca_ops->cong_control) in tcp_cwnd_validate()
1967 if (tcp_write_queue_empty(sk) && sk->sk_socket && in tcp_cwnd_validate()
1968 test_bit(SOCK_NOSPACE, &sk->sk_socket->flags) && in tcp_cwnd_validate()
1969 (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) in tcp_cwnd_validate()
1977 return after(tp->snd_sml, tp->snd_una) && in tcp_minshall_check()
1978 !after(tp->snd_sml, tp->snd_nxt); in tcp_minshall_check()
1982 * Note that a TSO packet might end with a sub-mss segment
1984 * if ((skb->len % mss) != 0)
1985 * tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
1987 * skb_pcount = skb->len / mss_now
1992 if (skb->len < tcp_skb_pcount(skb) * mss_now) in tcp_minshall_update()
1993 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; in tcp_minshall_update()
1997 * 1. It is full sized. (provided by caller in %partial bool)
2008 (!nonagle && tp->packets_out && tcp_minshall_check(tp))); in tcp_nagle_check()
2015 * - For close peers, we rather send bigger packets to reduce
2017 * - For long distance/rtt flows, we would like to get ACK clocking
2021 * in bigger TSO bursts. We we cut the RTT-based allowance in half
2022 * for every 2^9 usec (aka 512 us) of RTT, so that the RTT-based allowance
2031 bytes = READ_ONCE(sk->sk_pacing_rate) >> READ_ONCE(sk->sk_pacing_shift); in tcp_tso_autosize()
2033 r = tcp_min_rtt(tcp_sk(sk)) >> READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tso_rtt_log); in tcp_tso_autosize()
2034 if (r < BITS_PER_TYPE(sk->sk_gso_max_size)) in tcp_tso_autosize()
2035 bytes += sk->sk_gso_max_size >> r; in tcp_tso_autosize()
2037 bytes = min_t(unsigned long, bytes, sk->sk_gso_max_size); in tcp_tso_autosize()
2047 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; in tcp_tso_segs()
2050 min_tso = ca_ops->min_tso_segs ? in tcp_tso_segs()
2051 ca_ops->min_tso_segs(sk) : in tcp_tso_segs()
2052 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs); in tcp_tso_segs()
2055 return min_t(u32, tso_segs, sk->sk_gso_max_segs); in tcp_tso_segs()
2068 window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in tcp_mss_split_point()
2074 needed = min(skb->len, window); in tcp_mss_split_point()
2085 return needed - partial; in tcp_mss_split_point()
2106 return min(halfcwnd, cwnd - in_flight); in tcp_cwnd_test()
2111 * SKB onto the wire.
2140 if (tcp_urg_mode(tp) || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) in tcp_nagle_test()
2143 if (!tcp_nagle_check(skb->len < cur_mss, tp, nonagle)) in tcp_nagle_test()
2154 u32 end_seq = TCP_SKB_CB(skb)->end_seq; in tcp_snd_wnd_test()
2156 if (skb->len > cur_mss) in tcp_snd_wnd_test()
2157 end_seq = TCP_SKB_CB(skb)->seq + cur_mss; in tcp_snd_wnd_test()
2166 * know that all the data is in scatter-gather pages, and that the
2172 int nlen = skb->len - len; in tso_fragment()
2177 DEBUG_NET_WARN_ON_ONCE(skb->len != skb->data_len); in tso_fragment()
2181 return -ENOMEM; in tso_fragment()
2185 sk_wmem_queued_add(sk, buff->truesize); in tso_fragment()
2186 sk_mem_charge(sk, buff->truesize); in tso_fragment()
2187 buff->truesize += nlen; in tso_fragment()
2188 skb->truesize -= nlen; in tso_fragment()
2191 TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len; in tso_fragment()
2192 TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq; in tso_fragment()
2193 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq; in tso_fragment()
2196 flags = TCP_SKB_CB(skb)->tcp_flags; in tso_fragment()
2197 TCP_SKB_CB(skb)->tcp_flags = flags & ~(TCPHDR_FIN | TCPHDR_PSH); in tso_fragment()
2198 TCP_SKB_CB(buff)->tcp_flags = flags; in tso_fragment()
2233 if (icsk->icsk_ca_state >= TCP_CA_Recovery) in tcp_tso_should_defer()
2238 * Note that tp->tcp_wstamp_ns can be in the future if we have in tcp_tso_should_defer()
2241 delta = tp->tcp_clock_cache - tp->tcp_wstamp_ns - NSEC_PER_MSEC; in tcp_tso_should_defer()
2250 send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in tcp_tso_should_defer()
2253 cong_win = (tcp_snd_cwnd(tp) - in_flight) * tp->mss_cache; in tcp_tso_should_defer()
2257 /* If a full-sized TSO skb can be sent, do it. */ in tcp_tso_should_defer()
2258 if (limit >= max_segs * tp->mss_cache) in tcp_tso_should_defer()
2262 if ((skb != tcp_write_queue_tail(sk)) && (limit >= skb->len)) in tcp_tso_should_defer()
2265 win_divisor = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tso_win_divisor); in tcp_tso_should_defer()
2267 u32 chunk = min(tp->snd_wnd, tcp_snd_cwnd(tp) * tp->mss_cache); in tcp_tso_should_defer()
2277 * ACK. Receiver should ACK every other full sized in tcp_tso_should_defer()
2281 if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache) in tcp_tso_should_defer()
2289 delta = tp->tcp_clock_cache - head->tstamp; in tcp_tso_should_defer()
2291 if ((s64)(delta - (u64)NSEC_PER_USEC * (tp->srtt_us >> 4)) < 0) in tcp_tso_should_defer()
2296 * 1) We are cwnd-limited in tcp_tso_should_defer()
2297 * 2) We are rwnd-limited in tcp_tso_should_defer()
2301 if (cong_win <= skb->len) { in tcp_tso_should_defer()
2306 if (send_win <= skb->len) { in tcp_tso_should_defer()
2313 if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) || in tcp_tso_should_defer()
2314 TCP_SKB_CB(skb)->eor) in tcp_tso_should_defer()
2331 interval = READ_ONCE(net->ipv4.sysctl_tcp_probe_interval); in tcp_mtu_check_reprobe()
2332 delta = tcp_jiffies32 - icsk->icsk_mtup.probe_timestamp; in tcp_mtu_check_reprobe()
2337 icsk->icsk_mtup.probe_size = 0; in tcp_mtu_check_reprobe()
2338 icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + in tcp_mtu_check_reprobe()
2340 icsk->icsk_af_ops->net_header_len; in tcp_mtu_check_reprobe()
2341 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss); in tcp_mtu_check_reprobe()
2344 icsk->icsk_mtup.probe_timestamp = tcp_jiffies32; in tcp_mtu_check_reprobe()
2354 if (len <= skb->len) in tcp_can_coalesce_send_queue_head()
2360 len -= skb->len; in tcp_can_coalesce_send_queue_head()
2369 skb_frag_t *lastfrag = NULL, *fragto = skb_shinfo(to)->frags; in tcp_clone_payload()
2373 if (!sk_wmem_schedule(sk, to->truesize + probe_size)) in tcp_clone_payload()
2374 return -ENOMEM; in tcp_clone_payload()
2376 skb_queue_walk(&sk->sk_write_queue, skb) { in tcp_clone_payload()
2377 const skb_frag_t *fragfrom = skb_shinfo(skb)->frags; in tcp_clone_payload()
2380 return -EINVAL; in tcp_clone_payload()
2382 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++, fragfrom++) { in tcp_clone_payload()
2386 probe_size - len); in tcp_clone_payload()
2396 return -E2BIG; in tcp_clone_payload()
2409 skb_shinfo(to)->nr_frags = nr_frags; in tcp_clone_payload()
2410 to->truesize += probe_size; in tcp_clone_payload()
2411 to->len += probe_size; in tcp_clone_payload()
2412 to->data_len += probe_size; in tcp_clone_payload()
2425 TCP_SKB_CB(dst)->tcp_flags |= TCP_SKB_CB(src)->tcp_flags; in tcp_eat_one_skb()
2426 TCP_SKB_CB(dst)->eor = TCP_SKB_CB(src)->eor; in tcp_eat_one_skb()
2439 * -1 otherwise
2458 if (likely(!icsk->icsk_mtup.enabled || in tcp_mtu_probe()
2459 icsk->icsk_mtup.probe_size || in tcp_mtu_probe()
2460 inet_csk(sk)->icsk_ca_state != TCP_CA_Open || in tcp_mtu_probe()
2462 tp->rx_opt.num_sacks || tp->rx_opt.dsack)) in tcp_mtu_probe()
2463 return -1; in tcp_mtu_probe()
2466 * and current mss_clamp. if (search_high - search_low) in tcp_mtu_probe()
2470 probe_size = tcp_mtu_to_mss(sk, (icsk->icsk_mtup.search_high + in tcp_mtu_probe()
2471 icsk->icsk_mtup.search_low) >> 1); in tcp_mtu_probe()
2472 size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache; in tcp_mtu_probe()
2473 interval = icsk->icsk_mtup.search_high - icsk->icsk_mtup.search_low; in tcp_mtu_probe()
2478 if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high) || in tcp_mtu_probe()
2479 interval < READ_ONCE(net->ipv4.sysctl_tcp_probe_threshold)) { in tcp_mtu_probe()
2484 return -1; in tcp_mtu_probe()
2488 if (tp->write_seq - tp->snd_nxt < size_needed) in tcp_mtu_probe()
2489 return -1; in tcp_mtu_probe()
2491 if (tp->snd_wnd < size_needed) in tcp_mtu_probe()
2492 return -1; in tcp_mtu_probe()
2493 if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp))) in tcp_mtu_probe()
2499 return -1; in tcp_mtu_probe()
2505 return -1; in tcp_mtu_probe()
2510 return -1; in tcp_mtu_probe()
2516 return -1; in tcp_mtu_probe()
2518 sk_wmem_queued_add(sk, nskb->truesize); in tcp_mtu_probe()
2519 sk_mem_charge(sk, nskb->truesize); in tcp_mtu_probe()
2525 TCP_SKB_CB(nskb)->seq = TCP_SKB_CB(skb)->seq; in tcp_mtu_probe()
2526 TCP_SKB_CB(nskb)->end_seq = TCP_SKB_CB(skb)->seq + probe_size; in tcp_mtu_probe()
2527 TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK; in tcp_mtu_probe()
2534 copy = min_t(int, skb->len, probe_size - len); in tcp_mtu_probe()
2536 if (skb->len <= copy) { in tcp_mtu_probe()
2539 TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags & in tcp_mtu_probe()
2543 TCP_SKB_CB(skb)->seq += copy; in tcp_mtu_probe()
2551 tcp_init_tso_segs(nskb, nskb->len); in tcp_mtu_probe()
2554 * be resegmented into mss-sized pieces by tcp_write_xmit(). in tcp_mtu_probe()
2559 tcp_snd_cwnd_set(tp, tcp_snd_cwnd(tp) - 1); in tcp_mtu_probe()
2562 icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len); in tcp_mtu_probe()
2563 tp->mtu_probe.probe_seq_start = TCP_SKB_CB(nskb)->seq; in tcp_mtu_probe()
2564 tp->mtu_probe.probe_seq_end = TCP_SKB_CB(nskb)->end_seq; in tcp_mtu_probe()
2569 return -1; in tcp_mtu_probe()
2579 if (tp->tcp_wstamp_ns <= tp->tcp_clock_cache) in tcp_pacing_check()
2582 if (!hrtimer_is_queued(&tp->pacing_timer)) { in tcp_pacing_check()
2583 hrtimer_start(&tp->pacing_timer, in tcp_pacing_check()
2584 ns_to_ktime(tp->tcp_wstamp_ns), in tcp_pacing_check()
2593 const struct rb_node *node = sk->tcp_rtx_queue.rb_node; in tcp_rtx_queue_empty_or_single_skb()
2600 return !node->rb_left && !node->rb_right; in tcp_rtx_queue_empty_or_single_skb()
2607 * - better RTT estimation and ACK scheduling
2608 * - faster recovery
2609 * - high rates
2620 2 * skb->truesize, in tcp_small_queue_check()
2621 READ_ONCE(sk->sk_pacing_rate) >> READ_ONCE(sk->sk_pacing_shift)); in tcp_small_queue_check()
2622 if (sk->sk_pacing_status == SK_PACING_NONE) in tcp_small_queue_check()
2624 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_limit_output_bytes)); in tcp_small_queue_check()
2628 tcp_sk(sk)->tcp_tx_delay) { in tcp_small_queue_check()
2629 u64 extra_bytes = (u64)READ_ONCE(sk->sk_pacing_rate) * in tcp_small_queue_check()
2630 tcp_sk(sk)->tcp_tx_delay; in tcp_small_queue_check()
2633 * approximate our needs assuming an ~100% skb->truesize overhead. in tcp_small_queue_check()
2637 extra_bytes >>= (20 - 1); in tcp_small_queue_check()
2640 if (refcount_read(&sk->sk_wmem_alloc) > limit) { in tcp_small_queue_check()
2649 set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags); in tcp_small_queue_check()
2655 if (refcount_read(&sk->sk_wmem_alloc) > limit) in tcp_small_queue_check()
2664 enum tcp_chrono old = tp->chrono_type; in tcp_chrono_set()
2667 tp->chrono_stat[old - 1] += now - tp->chrono_start; in tcp_chrono_set()
2668 tp->chrono_start = now; in tcp_chrono_set()
2669 tp->chrono_type = new; in tcp_chrono_set()
2681 if (type > tp->chrono_type) in tcp_chrono_start()
2699 else if (type == tp->chrono_type) in tcp_chrono_stop()
2708 struct sk_buff *next_skb = skb->next; in tcp_grow_skb()
2717 nlen = min_t(u32, amount, next_skb->len); in tcp_grow_skb()
2721 TCP_SKB_CB(skb)->end_seq += nlen; in tcp_grow_skb()
2722 TCP_SKB_CB(next_skb)->seq += nlen; in tcp_grow_skb()
2724 if (!next_skb->len) { in tcp_grow_skb()
2726 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(next_skb)->end_seq; in tcp_grow_skb()
2737 * snd_up-64k-mss .. snd_up cannot be large. However, taking into
2774 if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE) { in tcp_write_xmit()
2776 tp->tcp_wstamp_ns = tp->tcp_clock_cache; in tcp_write_xmit()
2777 skb_set_delivery_time(skb, tp->tcp_wstamp_ns, SKB_CLOCK_MONOTONIC); in tcp_write_xmit()
2778 list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue); in tcp_write_xmit()
2795 missing_bytes = cwnd_quota * mss_now - skb->len; in tcp_write_xmit()
2824 if (skb->len > limit && in tcp_write_xmit()
2833 * We do not want to send a pure-ack packet and have in tcp_write_xmit()
2836 if (TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) in tcp_write_xmit()
2866 tp->prr_out += sent_pkts; in tcp_write_xmit()
2873 return !tp->packets_out && !tcp_write_queue_empty(sk); in tcp_write_xmit()
2886 if (rcu_access_pointer(tp->fastopen_rsk)) in tcp_schedule_loss_probe()
2889 early_retrans = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_early_retrans); in tcp_schedule_loss_probe()
2894 !tp->packets_out || !tcp_is_sack(tp) || in tcp_schedule_loss_probe()
2895 (icsk->icsk_ca_state != TCP_CA_Open && in tcp_schedule_loss_probe()
2896 icsk->icsk_ca_state != TCP_CA_CWR)) in tcp_schedule_loss_probe()
2903 if (tp->srtt_us) { in tcp_schedule_loss_probe()
2904 timeout_us = tp->srtt_us >> 2; in tcp_schedule_loss_probe()
2905 if (tp->packets_out == 1) in tcp_schedule_loss_probe()
2916 jiffies_to_usecs(inet_csk(sk)->icsk_rto) : in tcp_schedule_loss_probe()
2933 set_bit(TSQ_THROTTLED, &sk->sk_tsq_flags); in skb_still_in_host_queue()
2955 if (tp->tlp_high_seq) in tcp_send_loss_probe()
2958 tp->tlp_retrans = 0; in tcp_send_loss_probe()
2961 pcount = tp->packets_out; in tcp_send_loss_probe()
2963 if (tp->packets_out > pcount) in tcp_send_loss_probe()
2967 skb = skb_rb_last(&sk->tcp_rtx_queue); in tcp_send_loss_probe()
2969 tcp_warn_once(sk, tp->packets_out, "invalid inflight: "); in tcp_send_loss_probe()
2970 smp_store_release(&inet_csk(sk)->icsk_pending, 0); in tcp_send_loss_probe()
2981 if ((pcount > 1) && (skb->len > (pcount - 1) * mss)) { in tcp_send_loss_probe()
2983 (pcount - 1) * mss, mss, in tcp_send_loss_probe()
2995 tp->tlp_retrans = 1; in tcp_send_loss_probe()
2999 tp->tlp_high_seq = tp->snd_nxt; in tcp_send_loss_probe()
3003 smp_store_release(&inet_csk(sk)->icsk_pending, 0); in tcp_send_loss_probe()
3019 if (unlikely(sk->sk_state == TCP_CLOSE)) in __tcp_push_pending_frames()
3034 BUG_ON(!skb || skb->len < mss_now); in tcp_push_one()
3036 tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation); in tcp_push_one()
3048 * RCV.BUFF - RCV.USER - RCV.WINDOW >= min(1/2 RCV.BUFF, MSS)"
3054 * since header prediction assumes th->window stays fixed.
3056 * Strictly speaking, keeping th->window fixed violates the receiver
3100 * fluctuations. --SAW 1998/11/1 in __tcp_select_window()
3102 int mss = icsk->icsk_ack.rcv_mss; in __tcp_select_window()
3110 full_space = min_t(int, tp->window_clamp, allowed_space); in __tcp_select_window()
3119 * a non-zero scaling factor in effect. in __tcp_select_window()
3121 if (READ_ONCE(net->ipv4.sysctl_tcp_shrink_window) && tp->rx_opt.rcv_wscale) in __tcp_select_window()
3127 icsk->icsk_ack.quick = 0; in __tcp_select_window()
3135 free_space = round_down(free_space, 1 << tp->rx_opt.rcv_wscale); in __tcp_select_window()
3138 * of the maximum allowed, try to move to zero-window, else in __tcp_select_window()
3148 if (free_space > tp->rcv_ssthresh) in __tcp_select_window()
3149 free_space = tp->rcv_ssthresh; in __tcp_select_window()
3154 if (tp->rx_opt.rcv_wscale) { in __tcp_select_window()
3161 window = ALIGN(window, (1 << tp->rx_opt.rcv_wscale)); in __tcp_select_window()
3163 window = tp->rcv_wnd; in __tcp_select_window()
3172 if (window <= free_space - mss || window > free_space) in __tcp_select_window()
3183 free_space = round_down(free_space, 1 << tp->rx_opt.rcv_wscale); in __tcp_select_window()
3186 icsk->icsk_ack.quick = 0; in __tcp_select_window()
3193 free_space < (1 << tp->rx_opt.rcv_wscale)) in __tcp_select_window()
3197 if (free_space > tp->rcv_ssthresh) { in __tcp_select_window()
3198 free_space = tp->rcv_ssthresh; in __tcp_select_window()
3203 * the memory-based limit, and rcv_ssthresh is not a hard limit in __tcp_select_window()
3206 free_space = ALIGN(free_space, (1 << tp->rx_opt.rcv_wscale)); in __tcp_select_window()
3220 shinfo->tx_flags |= next_shinfo->tx_flags & SKBTX_ANY_TSTAMP; in tcp_skb_collapse_tstamp()
3221 shinfo->tskey = next_shinfo->tskey; in tcp_skb_collapse_tstamp()
3222 TCP_SKB_CB(skb)->txstamp_ack |= in tcp_skb_collapse_tstamp()
3223 TCP_SKB_CB(next_skb)->txstamp_ack; in tcp_skb_collapse_tstamp()
3234 next_skb_size = next_skb->len; in tcp_collapse_retrans()
3244 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(next_skb)->end_seq; in tcp_collapse_retrans()
3247 TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags; in tcp_collapse_retrans()
3252 TCP_SKB_CB(skb)->sacked |= TCP_SKB_CB(next_skb)->sacked & TCPCB_EVER_RETRANS; in tcp_collapse_retrans()
3253 TCP_SKB_CB(skb)->eor = TCP_SKB_CB(next_skb)->eor; in tcp_collapse_retrans()
3257 if (next_skb == tp->retransmit_skb_hint) in tcp_collapse_retrans()
3258 tp->retransmit_skb_hint = skb; in tcp_collapse_retrans()
3278 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) in tcp_can_collapse()
3285 * less packets on the wire. This is only done on retransmission.
3294 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_retrans_collapse)) in tcp_retrans_try_collapse()
3296 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN) in tcp_retrans_try_collapse()
3306 space -= skb->len; in tcp_retrans_try_collapse()
3316 if (after(TCP_SKB_CB(skb)->end_seq, tcp_wnd_end(tp))) in tcp_retrans_try_collapse()
3325 * state updates are done by the caller. Returns non-zero if an
3337 if (icsk->icsk_mtup.probe_size) in __tcp_retransmit_skb()
3338 icsk->icsk_mtup.probe_size = 0; in __tcp_retransmit_skb()
3341 return -EBUSY; in __tcp_retransmit_skb()
3344 if (before(TCP_SKB_CB(skb)->seq, tp->snd_una)) { in __tcp_retransmit_skb()
3345 if (unlikely(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { in __tcp_retransmit_skb()
3346 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_SYN; in __tcp_retransmit_skb()
3347 TCP_SKB_CB(skb)->seq++; in __tcp_retransmit_skb()
3350 if (unlikely(before(TCP_SKB_CB(skb)->end_seq, tp->snd_una))) { in __tcp_retransmit_skb()
3352 return -EINVAL; in __tcp_retransmit_skb()
3354 if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq)) in __tcp_retransmit_skb()
3355 return -ENOMEM; in __tcp_retransmit_skb()
3358 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk)) in __tcp_retransmit_skb()
3359 return -EHOSTUNREACH; /* Routing failure or similar. */ in __tcp_retransmit_skb()
3362 avail_wnd = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in __tcp_retransmit_skb()
3370 if (TCP_SKB_CB(skb)->seq != tp->snd_una) in __tcp_retransmit_skb()
3371 return -EAGAIN; in __tcp_retransmit_skb()
3381 if (skb->len > len) { in __tcp_retransmit_skb()
3384 return -ENOMEM; /* We'll try again later. */ in __tcp_retransmit_skb()
3387 return -ENOMEM; in __tcp_retransmit_skb()
3391 diff -= tcp_skb_pcount(skb); in __tcp_retransmit_skb()
3395 if (skb->len < avail_wnd) in __tcp_retransmit_skb()
3400 if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN_ECN) == TCPHDR_SYN_ECN) in __tcp_retransmit_skb()
3406 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN) in __tcp_retransmit_skb()
3408 tp->total_retrans += segs; in __tcp_retransmit_skb()
3409 tp->bytes_retrans += skb->len; in __tcp_retransmit_skb()
3411 /* make sure skb->data is aligned on arches that require it in __tcp_retransmit_skb()
3412 * and check if ack-trimming & collapsing extended the headroom in __tcp_retransmit_skb()
3415 if (unlikely((NET_IP_ALIGN && ((unsigned long)skb->data & 3)) || in __tcp_retransmit_skb()
3422 nskb->dev = NULL; in __tcp_retransmit_skb()
3425 err = -ENOBUFS; in __tcp_retransmit_skb()
3430 tcp_update_skb_after_send(sk, skb, tp->tcp_wstamp_ns); in __tcp_retransmit_skb()
3439 TCP_SKB_CB(skb)->seq, segs, err); in __tcp_retransmit_skb()
3443 } else if (err != -EBUSY) { in __tcp_retransmit_skb()
3450 TCP_SKB_CB(skb)->sacked |= TCPCB_EVER_RETRANS; in __tcp_retransmit_skb()
3462 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) { in tcp_retransmit_skb()
3466 TCP_SKB_CB(skb)->sacked |= TCPCB_RETRANS; in tcp_retransmit_skb()
3467 tp->retrans_out += tcp_skb_pcount(skb); in tcp_retransmit_skb()
3471 if (!tp->retrans_stamp) in tcp_retransmit_skb()
3472 tp->retrans_stamp = tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb); in tcp_retransmit_skb()
3474 if (tp->undo_retrans < 0) in tcp_retransmit_skb()
3475 tp->undo_retrans = 0; in tcp_retransmit_skb()
3476 tp->undo_retrans += tcp_skb_pcount(skb); in tcp_retransmit_skb()
3494 if (!tp->packets_out) in tcp_xmit_retransmit_queue()
3498 skb = tp->retransmit_skb_hint ?: rtx_head; in tcp_xmit_retransmit_queue()
3509 tp->retransmit_skb_hint = skb; in tcp_xmit_retransmit_queue()
3511 segs = tcp_snd_cwnd(tp) - tcp_packets_in_flight(tp); in tcp_xmit_retransmit_queue()
3514 sacked = TCP_SKB_CB(skb)->sacked; in tcp_xmit_retransmit_queue()
3520 if (tp->retrans_out >= tp->lost_out) { in tcp_xmit_retransmit_queue()
3528 if (icsk->icsk_ca_state != TCP_CA_Loss) in tcp_xmit_retransmit_queue()
3546 tp->prr_out += tcp_skb_pcount(skb); in tcp_xmit_retransmit_queue()
3549 icsk->icsk_pending != ICSK_TIME_REO_TIMEOUT) in tcp_xmit_retransmit_queue()
3555 inet_csk(sk)->icsk_rto, true); in tcp_xmit_retransmit_queue()
3569 delta = size - sk->sk_forward_alloc; in sk_forced_mem_schedule()
3576 if (mem_cgroup_sockets_enabled && sk->sk_memcg) in sk_forced_mem_schedule()
3577 mem_cgroup_charge_skmem(sk->sk_memcg, amt, in sk_forced_mem_schedule()
3596 tskb = skb_rb_last(&sk->tcp_rtx_queue); in tcp_send_fin()
3599 TCP_SKB_CB(tskb)->tcp_flags |= TCPHDR_FIN; in tcp_send_fin()
3600 TCP_SKB_CB(tskb)->end_seq++; in tcp_send_fin()
3601 tp->write_seq++; in tcp_send_fin()
3605 * We need to set tp->snd_nxt to the value it would have in tcp_send_fin()
3607 * does not change tp->snd_nxt. in tcp_send_fin()
3609 WRITE_ONCE(tp->snd_nxt, tp->snd_nxt + 1); in tcp_send_fin()
3619 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor); in tcp_send_fin()
3621 sk_forced_mem_schedule(sk, skb->truesize); in tcp_send_fin()
3623 tcp_init_nondata_skb(skb, tp->write_seq, in tcp_send_fin()
3633 * by RFC 2525, section 2.17. -DaveM
3664 /* Send a crossed SYN-ACK during socket establishment.
3675 if (!skb || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { in tcp_send_synack()
3677 return -EFAULT; in tcp_send_synack()
3679 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) { in tcp_send_synack()
3687 return -ENOMEM; in tcp_send_synack()
3688 INIT_LIST_HEAD(&nskb->tcp_tsorted_anchor); in tcp_send_synack()
3692 tcp_rbtree_insert(&sk->tcp_rtx_queue, nskb); in tcp_send_synack()
3693 sk_wmem_queued_add(sk, nskb->truesize); in tcp_send_synack()
3694 sk_mem_charge(sk, nskb->truesize); in tcp_send_synack()
3698 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK; in tcp_send_synack()
3705 * tcp_make_synack - Allocate one skb and build a SYNACK packet.
3750 * sk->sk_wmem_alloc in an atomic, we can promote to rw. in tcp_make_synack()
3762 if (unlikely(synack_type == TCP_SYNACK_COOKIE && ireq->tstamp_ok)) in tcp_make_synack()
3769 if (!tcp_rsk(req)->snt_synack) /* Timestamp first SYNACK */ in tcp_make_synack()
3770 tcp_rsk(req)->snt_synack = tcp_skb_timestamp_us(skb); in tcp_make_synack()
3779 u8 keyid = tcp_rsk(req)->ao_keyid; in tcp_make_synack()
3780 u8 rnext = tcp_rsk(req)->ao_rcv_next; in tcp_make_synack()
3782 ao_key = tcp_sk(sk)->af_specific->ao_lookup(sk, req_to_sk(req), in tcp_make_synack()
3783 keyid, -1); in tcp_make_synack()
3784 /* If there is no matching key - avoid sending anything, in tcp_make_synack()
3786 * for another peer-matching key, but the peer has requested in tcp_make_synack()
3793 … net_warn_ratelimited("TCP-AO: the keyid %u from SYN packet is not present - not sending SYNACK\n", in tcp_make_synack()
3802 key.md5_key = tcp_rsk(req)->af_specific->req_md5_lookup(sk, in tcp_make_synack()
3808 skb_set_hash(skb, READ_ONCE(tcp_rsk(req)->txhash), PKT_HASH_TYPE_L4); in tcp_make_synack()
3810 TCP_SKB_CB(skb)->tcp_flags = TCPHDR_SYN | TCPHDR_ACK; in tcp_make_synack()
3818 th = (struct tcphdr *)skb->data; in tcp_make_synack()
3820 th->syn = 1; in tcp_make_synack()
3821 th->ack = 1; in tcp_make_synack()
3823 th->source = htons(ireq->ir_num); in tcp_make_synack()
3824 th->dest = ireq->ir_rmt_port; in tcp_make_synack()
3825 skb->mark = ireq->ir_mark; in tcp_make_synack()
3826 skb->ip_summed = CHECKSUM_PARTIAL; in tcp_make_synack()
3827 th->seq = htonl(tcp_rsk(req)->snt_isn); in tcp_make_synack()
3829 th->ack_seq = htonl(tcp_rsk(req)->rcv_nxt); in tcp_make_synack()
3832 th->window = htons(min(req->rsk_rcv_wnd, 65535U)); in tcp_make_synack()
3834 th->doff = (tcp_header_size >> 2); in tcp_make_synack()
3837 /* Okay, we have all we need - do the md5 hash if needed */ in tcp_make_synack()
3840 tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location, in tcp_make_synack()
3845 tcp_rsk(req)->af_specific->ao_synack_hash(opts.hash_location, in tcp_make_synack()
3847 opts.hash_location - (u8 *)th, 0); in tcp_make_synack()
3875 if (likely(ca && bpf_try_module_get(ca, ca->owner))) { in tcp_ca_dst_init()
3876 bpf_module_put(icsk->icsk_ca_ops, icsk->icsk_ca_ops->owner); in tcp_ca_dst_init()
3877 icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst); in tcp_ca_dst_init()
3878 icsk->icsk_ca_ops = ca; in tcp_ca_dst_init()
3894 tp->tcp_header_len = sizeof(struct tcphdr); in tcp_connect_init()
3895 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_timestamps)) in tcp_connect_init()
3896 tp->tcp_header_len += TCPOLEN_TSTAMP_ALIGNED; in tcp_connect_init()
3901 if (tp->rx_opt.user_mss) in tcp_connect_init()
3902 tp->rx_opt.mss_clamp = tp->rx_opt.user_mss; in tcp_connect_init()
3903 tp->max_window = 0; in tcp_connect_init()
3909 if (!tp->window_clamp) in tcp_connect_init()
3910 WRITE_ONCE(tp->window_clamp, dst_metric(dst, RTAX_WINDOW)); in tcp_connect_init()
3911 tp->advmss = tcp_mss_clamp(tp, dst_metric_advmss(dst)); in tcp_connect_init()
3916 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK && in tcp_connect_init()
3917 (tp->window_clamp > tcp_full_space(sk) || tp->window_clamp == 0)) in tcp_connect_init()
3918 WRITE_ONCE(tp->window_clamp, tcp_full_space(sk)); in tcp_connect_init()
3925 tp->advmss - (tp->rx_opt.ts_recent_stamp ? tp->tcp_header_len - sizeof(struct tcphdr) : 0), in tcp_connect_init()
3926 &tp->rcv_wnd, in tcp_connect_init()
3927 &tp->window_clamp, in tcp_connect_init()
3928 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_window_scaling), in tcp_connect_init()
3932 tp->rx_opt.rcv_wscale = rcv_wscale; in tcp_connect_init()
3933 tp->rcv_ssthresh = tp->rcv_wnd; in tcp_connect_init()
3935 WRITE_ONCE(sk->sk_err, 0); in tcp_connect_init()
3937 tp->snd_wnd = 0; in tcp_connect_init()
3940 tp->snd_una = tp->write_seq; in tcp_connect_init()
3941 tp->snd_sml = tp->write_seq; in tcp_connect_init()
3942 tp->snd_up = tp->write_seq; in tcp_connect_init()
3943 WRITE_ONCE(tp->snd_nxt, tp->write_seq); in tcp_connect_init()
3945 if (likely(!tp->repair)) in tcp_connect_init()
3946 tp->rcv_nxt = 0; in tcp_connect_init()
3948 tp->rcv_tstamp = tcp_jiffies32; in tcp_connect_init()
3949 tp->rcv_wup = tp->rcv_nxt; in tcp_connect_init()
3950 WRITE_ONCE(tp->copied_seq, tp->rcv_nxt); in tcp_connect_init()
3952 inet_csk(sk)->icsk_rto = tcp_timeout_init(sk); in tcp_connect_init()
3953 inet_csk(sk)->icsk_retransmits = 0; in tcp_connect_init()
3962 tcb->end_seq += skb->len; in tcp_connect_queue_skb()
3964 sk_wmem_queued_add(sk, skb->truesize); in tcp_connect_queue_skb()
3965 sk_mem_charge(sk, skb->truesize); in tcp_connect_queue_skb()
3966 WRITE_ONCE(tp->write_seq, tcb->end_seq); in tcp_connect_queue_skb()
3967 tp->packets_out += tcp_skb_pcount(skb); in tcp_connect_queue_skb()
3971 * queue a data-only packet after the regular SYN, such that regular SYNs
3972 * are retransmitted on timeouts. Also if the remote SYN-ACK acknowledges
3981 struct tcp_fastopen_request *fo = tp->fastopen_req; in tcp_send_syn_data()
3986 tp->rx_opt.mss_clamp = tp->advmss; /* If MSS is not cached */ in tcp_send_syn_data()
3987 if (!tcp_fastopen_cookie_check(sk, &tp->rx_opt.mss_clamp, &fo->cookie)) in tcp_send_syn_data()
3990 /* MSS for SYN-data is based on cached MSS and bounded by PMTU and in tcp_send_syn_data()
3991 * user-MSS. Reserve maximum option space for middleboxes that add in tcp_send_syn_data()
3994 tp->rx_opt.mss_clamp = tcp_mss_clamp(tp, tp->rx_opt.mss_clamp); in tcp_send_syn_data()
3996 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); in tcp_send_syn_data()
3998 space = __tcp_mtu_to_mss(sk, icsk->icsk_pmtu_cookie) - in tcp_send_syn_data()
4001 space = min_t(size_t, space, fo->size); in tcp_send_syn_data()
4005 pfrag, sk->sk_allocation)) in tcp_send_syn_data()
4007 syn_data = tcp_stream_alloc_skb(sk, sk->sk_allocation, false); in tcp_send_syn_data()
4010 memcpy(syn_data->cb, syn->cb, sizeof(syn->cb)); in tcp_send_syn_data()
4012 space = min_t(size_t, space, pfrag->size - pfrag->offset); in tcp_send_syn_data()
4016 space = copy_page_from_iter(pfrag->page, pfrag->offset, in tcp_send_syn_data()
4017 space, &fo->data->msg_iter); in tcp_send_syn_data()
4023 skb_fill_page_desc(syn_data, 0, pfrag->page, in tcp_send_syn_data()
4024 pfrag->offset, space); in tcp_send_syn_data()
4025 page_ref_inc(pfrag->page); in tcp_send_syn_data()
4026 pfrag->offset += space; in tcp_send_syn_data()
4028 skb_zcopy_set(syn_data, fo->uarg, NULL); in tcp_send_syn_data()
4031 if (space == fo->size) in tcp_send_syn_data()
4032 fo->data = NULL; in tcp_send_syn_data()
4033 fo->copied = space; in tcp_send_syn_data()
4036 if (syn_data->len) in tcp_send_syn_data()
4039 err = tcp_transmit_skb(sk, syn_data, 1, sk->sk_allocation); in tcp_send_syn_data()
4041 skb_set_delivery_time(syn, syn_data->skb_mstamp_ns, SKB_CLOCK_MONOTONIC); in tcp_send_syn_data()
4048 TCP_SKB_CB(syn_data)->seq++; in tcp_send_syn_data()
4049 TCP_SKB_CB(syn_data)->tcp_flags = TCPHDR_ACK | TCPHDR_PSH; in tcp_send_syn_data()
4051 tp->syn_data = (fo->copied > 0); in tcp_send_syn_data()
4052 tcp_rbtree_insert(&sk->tcp_rtx_queue, syn_data); in tcp_send_syn_data()
4058 __skb_queue_tail(&sk->sk_write_queue, syn_data); in tcp_send_syn_data()
4059 tp->packets_out -= tcp_skb_pcount(syn_data); in tcp_send_syn_data()
4063 if (fo->cookie.len > 0) in tcp_send_syn_data()
4064 fo->cookie.len = 0; in tcp_send_syn_data()
4065 err = tcp_transmit_skb(sk, syn, 1, sk->sk_allocation); in tcp_send_syn_data()
4067 tp->syn_fastopen = 0; in tcp_send_syn_data()
4069 fo->cookie.len = -1; /* Exclude Fast Open option for SYN retries */ in tcp_send_syn_data()
4084 * Return error if the peer has both a md5 and a tcp-ao key in tcp_connect()
4087 if (unlikely(rcu_dereference_protected(tp->md5sig_info, in tcp_connect()
4089 bool needs_ao = !!tp->af_specific->ao_lookup(sk, sk, -1, -1); in tcp_connect()
4090 bool needs_md5 = !!tp->af_specific->md5_lookup(sk, sk); in tcp_connect()
4093 ao_info = rcu_dereference_check(tp->ao_info, in tcp_connect()
4100 needs_ao |= ao_info->ao_required; in tcp_connect()
4101 WARN_ON_ONCE(ao_info->ao_required && needs_md5); in tcp_connect()
4104 return -EKEYREJECTED; in tcp_connect()
4106 /* If we have a matching md5 key and no matching tcp-ao key in tcp_connect()
4113 kfree(rcu_replace_pointer(tp->md5sig_info, NULL, in tcp_connect()
4119 if (unlikely(rcu_dereference_protected(tp->ao_info, in tcp_connect()
4124 if (!tp->af_specific->ao_lookup(sk, sk, -1, -1)) in tcp_connect()
4125 return -EKEYREJECTED; in tcp_connect()
4129 if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk)) in tcp_connect()
4130 return -EHOSTUNREACH; /* Routing failure or similar. */ in tcp_connect()
4134 if (unlikely(tp->repair)) { in tcp_connect()
4139 buff = tcp_stream_alloc_skb(sk, sk->sk_allocation, true); in tcp_connect()
4141 return -ENOBUFS; in tcp_connect()
4146 tcp_init_nondata_skb(buff, tp->write_seq, TCPHDR_SYN); in tcp_connect()
4148 tp->retrans_stamp = tcp_time_stamp_ts(tp); in tcp_connect()
4151 tcp_rbtree_insert(&sk->tcp_rtx_queue, buff); in tcp_connect()
4154 err = tp->fastopen_req ? tcp_send_syn_data(sk, buff) : in tcp_connect()
4155 tcp_transmit_skb(sk, buff, 1, sk->sk_allocation); in tcp_connect()
4156 if (err == -ECONNREFUSED) in tcp_connect()
4159 /* We change tp->snd_nxt after the tcp_transmit_skb() call in tcp_connect()
4162 WRITE_ONCE(tp->snd_nxt, tp->write_seq); in tcp_connect()
4163 tp->pushed_seq = tp->write_seq; in tcp_connect()
4166 WRITE_ONCE(tp->snd_nxt, TCP_SKB_CB(buff)->seq); in tcp_connect()
4167 tp->pushed_seq = TCP_SKB_CB(buff)->seq; in tcp_connect()
4173 inet_csk(sk)->icsk_rto, false); in tcp_connect()
4180 u32 delack_from_rto_min = max(tcp_rto_min(sk), 2) - 1; in tcp_delack_max()
4182 return min(READ_ONCE(inet_csk(sk)->icsk_delack_max), delack_from_rto_min); in tcp_delack_max()
4192 int ato = icsk->icsk_ack.ato; in tcp_send_delayed_ack()
4200 (icsk->icsk_ack.pending & ICSK_ACK_PUSHED)) in tcp_send_delayed_ack()
4206 * Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements in tcp_send_delayed_ack()
4209 if (tp->srtt_us) { in tcp_send_delayed_ack()
4210 int rtt = max_t(int, usecs_to_jiffies(tp->srtt_us >> 3), in tcp_send_delayed_ack()
4226 if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) { in tcp_send_delayed_ack()
4236 smp_store_release(&icsk->icsk_ack.pending, in tcp_send_delayed_ack()
4237 icsk->icsk_ack.pending | ICSK_ACK_SCHED | ICSK_ACK_TIMER); in tcp_send_delayed_ack()
4238 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout); in tcp_send_delayed_ack()
4247 if (sk->sk_state == TCP_CLOSE) in __tcp_send_ack()
4260 delay = TCP_DELACK_MAX << icsk->icsk_ack.retry; in __tcp_send_ack()
4262 icsk->icsk_ack.retry++; in __tcp_send_ack()
4264 icsk->icsk_ack.ato = TCP_ATO_MIN; in __tcp_send_ack()
4286 __tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt, 0); in tcp_send_ack()
4296 * Current solution: to send TWO zero-length segments in urgent mode:
4298 * out-of-date with SND.UNA-1 to probe window.
4309 return -1; in tcp_xmit_probe_skb()
4317 tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK); in tcp_xmit_probe_skb()
4325 if (sk->sk_state == TCP_ESTABLISHED) { in tcp_send_window_probe()
4326 tcp_sk(sk)->snd_wl1 = tcp_sk(sk)->rcv_nxt - 1; in tcp_send_window_probe()
4338 if (sk->sk_state == TCP_CLOSE) in tcp_write_wakeup()
4339 return -1; in tcp_write_wakeup()
4342 if (skb && before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp))) { in tcp_write_wakeup()
4345 unsigned int seg_size = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq; in tcp_write_wakeup()
4347 if (before(tp->pushed_seq, TCP_SKB_CB(skb)->end_seq)) in tcp_write_wakeup()
4348 tp->pushed_seq = TCP_SKB_CB(skb)->end_seq; in tcp_write_wakeup()
4354 if (seg_size < TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq || in tcp_write_wakeup()
4355 skb->len > mss) { in tcp_write_wakeup()
4357 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; in tcp_write_wakeup()
4360 return -1; in tcp_write_wakeup()
4364 TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH; in tcp_write_wakeup()
4370 if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF)) in tcp_write_wakeup()
4389 if (tp->packets_out || tcp_write_queue_empty(sk)) { in tcp_send_probe0()
4391 icsk->icsk_probes_out = 0; in tcp_send_probe0()
4392 icsk->icsk_backoff = 0; in tcp_send_probe0()
4393 icsk->icsk_probes_tstamp = 0; in tcp_send_probe0()
4397 icsk->icsk_probes_out++; in tcp_send_probe0()
4399 if (icsk->icsk_backoff < READ_ONCE(net->ipv4.sysctl_tcp_retries2)) in tcp_send_probe0()
4400 icsk->icsk_backoff++; in tcp_send_probe0()
4415 const struct tcp_request_sock_ops *af_ops = tcp_rsk(req)->af_specific; in tcp_rtx_synack()
4420 if (READ_ONCE(sk->sk_txrehash) == SOCK_TXREHASH_ENABLED) in tcp_rtx_synack()
4421 WRITE_ONCE(tcp_rsk(req)->txhash, net_tx_rndhash()); in tcp_rtx_synack()
4422 res = af_ops->send_synack(sk, NULL, &fl, req, NULL, TCP_SYNACK_NORMAL, in tcp_rtx_synack()
4432 tcp_sk_rw(sk)->total_retrans++; in tcp_rtx_synack()